NPG Biotechnology Volume 23 Issue 5 May

www.nature.com/naturebiotechnology

EDITORIAL OFFICE [email protected] Park Avenue South, New York, NY 10010-1707Tel: (212) 726 9200, Fax: (212) 696 9635Editor: Andrew MarshallSenior Editor: Laura DeFrancesco (News & Features), Kathy Aschheim (Research)Associate Editors: , Michael Francisco (Resources and Special Projects), Nadia Cervoni (Research), Gaspar Taroncher-Oldenburg (Research)News Editors: Stephan Herrera, Sabine LouëtEditor-at-Large: John HodgsonContributing Writers: Jeffrey L. Fox, Stephan Herrera, Ken Howard WilanCopy Editor: Teresa MooganSenior Production Editors: Renee Lucas, Ingrid McNamaraCover Design: Erin BoyleEditorial Assistant: Mark Zipkin

MANAGEMENT OFFICESNPG New York345 Park Avenue South, New York, NY 10010-1707Tel: (212) 726 9200, Fax: (212) 696 9006Publishing Director: Peter CollinsPublisher, Biopharma and Biobusiness: Adam SmithExecutive Editor: Linda MillerChief Technology Officer: Howard RatnerGlobal Head of Advertising: Fabien SavenayHead of Nature Research & Reviews Marketing: Sara GirardSenior Marketing Manager: Sandra PetragliaProduction Coordinator: Diane TempranoAssociate Director, New Technology: Timo HannayAssociate Director, Content Systems: Joe LandolfiSenior Web Production Editor: Sarada Callison

NPG LondonThe Macmillan Building, 4 Crinan Street, London N1 9XWTel: 44 207 833 4000, Fax: 44 207 843 4996Managing Director: Annette ThomasPublishing Director: David SwinbanksEditor-in-Chief, Nature Publications: Philip CampbellMarketing Director: Della Sar

NPG TokyoMG Ichigaya Bldg. 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Asia-Pacific Publisher: Antoine E. BocquetManager: Koichi NakamuraSenior Marketing Manager: Peter YoshiharaAsia-Pacific Sales Director: Kate YoneyamaAsia-Pacific Sales Manager: Rinoko Asami

DISPLAY ADVERTISING [email protected] (US/Canada) [email protected] (Europe) [email protected] (Japan)US Head of Display Advertising: Stephen Schwartz, Tel: (212) 726 9256, Fax: (212) 696 9481Global Head of Display Advertising Sales: John Michael, Tel: 44 207 843 4960, Fax: 44 207 843 4996Head of Display Advertising—Europe: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4996Business Development Manager: Claire Hines, Tel: 44 207 843 4960, Fax: 44 207 843 4996Asia-Pacific Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Western Region Sales Manager: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805Sales Executives:New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717New York, Mid-Atlantic, Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481Northwest: Mathieu DesJardins, Tel: (415) 781 6422, Fax: (415) 781 3805Eastern England/Scotland/Italy/Spain/Israel: Matthew Wilkinson, Tel: 44 207 843 4960, Fax: 44 207 843 4749South/West United Kingdom/Scandinavia/Holland: Marianne Boulakas, Tel: 44 207 843 4969, Fax: 44 207 843 4749Northern Germany: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4749Southern Germany/Austria/Switzerland/France/Belgium: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419

NATUREJOBS [email protected] (US/Canada) [email protected] (Europe) [email protected] (Japan)Publisher: Ben Crowe, Tel: (212) 726 9245, Fax: (212) 696 9482US Sales Manager: Peter Bless, Tel: (212) 726 9248, Fax: (212) 696 9482Japan Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746naturejobs Sales Director: Nevin Bayoumi, Tel: 44 207 843 4961, Fax: 44 207 843 4996

CUSTOMER SERVICE www.nature.com/helpSenior Global Customer Service Manager: Gerald CoppinFor all print and online assistance, please visit www.nature.com/helpPurchase subscriptions:Americas: Nature Biotechnology, Subscription Dept., 303 Park Avenue South #1280, New York, NY 10010-3601. Tel: (866) 363 7860, Fax: (212) 689 9108Europe/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 358Japan: Nature Biotechnology, Nature Japan K.K., MG Ichigaya Bldg. 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841. Tel: 81 3 3267 8751, Fax: 81 3 3267 8746India: Nature Biotechnology, Harpal Singh Gill, Macmillan Magazines Ltd, 5A/12 Ansari Road, Darya Ganj, New Delhi, 110 002 India. Tel: 00 91 11 324 4186, Tel/Fax: 00 91 11 327 2010

REPRINTS [email protected] Biotechnology, Reprint Department, Nature Publishing Group, 345 Park Avenue South, New York, NY 10010-1707, USA.For commercial reprint orders of 600 or more, please contact:UK Reprints Sales Executive: Christine Fothergill, Tel: 44 207 843 4967, Fax: 44 207 843 4749US Reprints Sales Executive: Sharda Tulsie, Tel: (212) 726 9631, Fax: (212) 679 0843

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy

NATURE BIOTECHNOLOGY VOLUME 23 NUMBER 5 MAY 2005 507

Memo to FDA head: IT upgrade required

May 1, 2005

Dear Lester (or if there’s a nomination hitch, To Whomever It May Concern),

As Spring is the time of new beginnings and rebirths, we would like to congratulate you on becoming commissioner of the one of the world’s largest and most important government bureaucracies. You are now responsible for regulation and oversight of products that account for over a trillion dollars annually in the United States as well as the health and well-being of over 295 million citizens.

The agency that you head is currently recuperating from a series of health scares in which concerns from internal advisors about toxicities associated with certain antidepressants and widely prescribed analgesics approved by the fast-track process appeared to fall on deaf ears. Criticism of your handling of aseptic standards at a UK facility of a major flu vaccine manufacturer hasn’t helped. Neither has US Republican congressmen, who cite a “crisis of confidence” in an agency that has developed an overly cozy relationship with industry. Many appear to suffer from collective amnesia, conveniently forgetting previous carping about risk aversion at the FDA and unacceptable delays in the approval of lifesaving medicines.

You and we appreciate, however, that the FDA is not in crisis. Rather, it has suffered some blunt trauma and you must now orchestrate the heal-ing. That is why it is critical that in your role as commissioner, you now take steps to ensure that the agency further streamlines the drug approval process and addresses problems in monitoring serious adverse events.

In the next few years, it will not be pharmaceutical companies, but increasingly biotech that will be producing new experimental medicines and technologies. This means that unlike medicines of the past, which have taken decades of development, an increasing number of new drug applications at your agency will have their origins wholly or in large part in biological discoveries reported just a few years before.

As a consequence, more and more of the products under review will be unfamiliar experimental treatments, requiring new knowledge and new reference points. The increasing use of pharmacogenomic information in defining indications, as laid out in your recent Guideline (see p. 510), will only add to this knowledge-based information torrent streaming in from applicants and trial sponsors.

And yet it seems that under your leadership, the FDA is destined to move in the opposite direction. Your recent budget proposal was heavy on protecting Americans “from risky products and potential terrorist threats” but palpably light on mobilizing the increasing knowledge base of biology in the cause of better treatments. The one area you highlighted for savings was IT. Specifically, as acting FDA commissioner you commended “savings of $5,116,000 through continued consolidation and/or postponement of information technology expenditures” in order to “fully embrace the President’s Management Agenda and the Secretarial priorities.”

But the FDA’s IT systems are woefully outdated, even by the standards of the average office computer. In December, according to Scott Gottlieb,

a former senior adviser for medical technology to the FDA commissioner, your agency requested Neurocrine Biosciences to refile its new drug appli-cation for a sleep aid called Indiplon (a pyrazolopyrimidine). This was not because the company had omitted data or because the agency had identified safety concerns. It was because the FDA could not ‘navigate’ the company’s electronic application (rather like grandma struggling with the video remote). And then there are reports that the FDA cannot accept portable document formats (PDFs) submitted via e-mail: perhaps some of the $20 million worth of increased user fees to be collected under PDUFA (the Prescription Drug User Fee Act) could tackle the PDF deficit?

Overhaul of your antiquated computer system might also provide a solution to the process of adverse event reporting. The current system is paper based and consequently severely hobbled because it is passive and depends on the diligence of overworked physicians, who fill out and send lengthy reports, via drug companies, to the FDA. The result is your agency becomes aware of life-threatening toxicities often only months or even years after a critical mass of fatalities appear on the radar.

If you were to seriously upgrade IT, your agency could take advantage of the increasing amounts of medical data that are finding their way onto electronic medical records and into electronic prescriptions. That’s why your ‘to-do’ list should prioritize the development of computerized systems for proactively gathering toxicity information and the refinement of algorithms for spotting potential toxic events and establishing causal links between drugs and side effects. These tools would provide a power-ful complement to existing passive reporting. Indeed, you already have pilot programs underway with healthcare networks, such as the Veteran Administration hospitals and New York-Presbyterian Hospital.

It may still be unfashionable to commend European initiatives, espe-cially among Republican congressmen, but it is becoming obvious that, at least on the question of electronic submission, the European Medicines Agency (EMEA) now leads the FDA. The FDA needs to keep pace with technological change and to foster a framework for evaluating drugs that are the fruits of secured knowledge and not mere hopeful endeavor.

It is your job to fix the FDA’s inefficient and antiquated computer sys-tem. In doing so, you’ll go a long way to fixing drug safety monitoring and overwork within the agency. Lester (or Whomever), it’s your legacy we are concerned about. Oh yes, and other people’s lives.

Content reshuffleSharp-eyed readers will notice that the Patents and Bioentrepreneur sections have been moved toward the front of the magazine. The Computational Biology section has also been relocated before the Research section. These changes have been implemented to bring news, opinion and feature content together, and to group technical content toward the back. Any feedback concerning these changes is welcome ([email protected]).

E D I T O R I A L©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Earlier stage biotechs attract partnersEven as the environment for going public remains hostile to most bio-tech companies and their investors, some are discovering that favorable financing alternatives still exist. A spate of recent acquisitions and a continuation of early-stage deal making signify that there are still appealing ways for many compa-nies to sustain their drug develop-ment programs, especially given the needs of potential acquirers and licensees.

For biotechs whose lead programs are at or near the phase 2 clinical trial inflection point that validates proof of concept, potential partners increasingly are considering acqui-sition. “People are far more willing to consider acquisition than they used to be,” notes Roger Longman, managing partner at healthcare business information specialist Windhover Information. Plus, he says, “bio-techs today may be willing to cut their upside for a present-day payout.”

Indeed, most companies don’t have that kind of product visibility with which to appeal to public investors. Five firms have gone public so far in 2005, to a lukewarm response in terms of the initial public offer-ing (IPO) price and after-market reaction. Moreover, an IPO, unlike an acquisition, is not necessarily an exit. “It’s a financing opportunity,” Longman points out. The level of financing an IPO often provides is especially important for companies that need capital to take a product to the end stages of commercialization, leading to an increase in valuation without further fund-raising. Among the recent class of IPOs, Eyetech Pharmaceuticals, Pharmion, Nitromed, and Corgentech (had its product worked—it recently failed in phase 3) are all companies that went public for that reason, more so than to give private investors an exit.

The vagaries of going public are one part of the reason more and more biotechs are considering a big pharma buyout—an option several have already chosen in 2005, including San Diego-based Idun Pharmaceutical and

Syrrx Pharmaceutical (by Pfizer and Takeda, respectively), TransForm Pharmaceutical and NeoGenesis, both located outside of Boston (Johnson & Johnson and Schering-Plough), and in 2004, metropolitan New York area-based Aton Pharma (Merck).

Acquisition is also an alternative to the old biotech standby—the big corporate deal. “The question of whether to buy or ally is always a trade-off,” notes Steve Sands, an investment banker with Lazard Freres in New York City. In the past, the trend in pharma and biotech was always to form an alliance. But a broad alliance is basically like giv-ing away the company anyway, he suggests, and many alliances that went south—in some cases because long-term social issues arise when working together for extended periods—should have been acquisitions. Moreover, at the board level, potential licen-sor/acquirees are now thinking about acqui-sition earlier and in a more organized way, he says.

With most late-stage product candidates—those that have been validated in phase 2 proof-of-concept clinical trials—either already spoken for or very pricey, pharma-ceutical companies have also become more interested in earlier stage alliances. “Late-stage deals, while expensive, sometimes do not provide the optimal ratio between

price and the probability of suc-cess,” contends venture capitalist Jürgen Drews, former president of R&D at Hoffmann-LaRoche. Big pharma also now has better tools in-house to make fast go/no-go decisions on whether to take prod-ucts forward, in turn enabling them to take more ‘shots on goal’ before investing in costly last-stage drug development—an argument in favor of early-stage alliances as well as acquisitions of companies with early-stage programs.

Thus at Pfizer, according to Martin Mackay, senior vice presi-dent worldwide research & technol-ogy, “we tend to go earlier because we have good predictive tools, to

be able to get to a decision point as quickly as possible.” This was a large part of the rationale behind its acquisition of Idun, a developer of caspase modulators, which are proteins involved in apoptosis but that remain unproven as a drug target. “Idun has built up a know-how that we’ll seek to understand,” he adds. With scale, Pfizer believes it can quickly ascertain the poten-tial for caspase-targeting compounds, which Idun is developing for a broad range of dis-orders including liver diseases, cancer and inflammation.

Given the uncertainties around availabil-ity of future financing and the overall costs of drug development, “conversations for potentially important products often get to the obvious point of considering an acqui-sition,” summarizes Anthony Evnin of ven-ture capital firm Venrock Associates. “With IPO valuations and timing uncertain now, it requires you to think more profoundly about acquisition. I do think there are elements that begin to feel like a trend.”

Mark Ratner, Cambridge, Massachusetts

Pfizer, who recently acquired Idun Pharmaceutical, has, like many other big pharma, been focusing on earlier stage biotech companies.

Spen

cer P

latt/

Get

ty Im

ages

For more news and analysis go to

www.nature.com/news

ALSO IN THIS SECTIONCautious welcome for FDA pharmacogenomics guidance p510

EPO neem patent revocation revives biopiracy debate p511US court case to define EST patentability p513

Syngenta’s gaff embarrasses industry and White House p514PROFILE: Julian Thurston p517

N E W S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

510 VOLUME 23 NUMBER 5 MAY 2005 NATURE BIOTECHNOLOGY

Cautious welcome for FDA pharmacogenomics guidance

The US Food and Drug Administration (FDA) on March 22 issued the long-awaited final version of its Guidance for Industry on submission of pharmacogenomics data, giving industry a strong political signal that it is open to applications concerning therapeutic products ‘personalized’ to patients’ genetic blueprints. The main challenge for the regulator is to encour-age industry to submit voluntary data to help refine the pharma-cogenomics approval process and speed up the arrival of the next wave of such products to the market.

Industry experts have enthusi-astically embraced the new docu-ment, calling it an “important first step” for bringing pharmacoge-nomics-based drugs to market. “I applaud the FDA for taking the step,” says Donald Halbert, executive vice president of R&D of Iconix Pharmaceuticals in Mountain View, California. “They have maybe intention-ally put pressure on themselves to develop the infrastructure, tools and training to work with these kinds of data sets.”

The guideline outlines the circumstances under which companies are required to submit pharmacogenomics data and the procedures for submitting them. Data must be submit-ted when they are based on valid biomarkers that have been rigorously tested by the scien-tific community and explicitly affect how tri-als for a product are designed. Using several examples, the document also defines a second category of data as those obtained under the auspices of exploratory research. Data in this latter category do not have to be submitted; instead, the agency encourages companies to submit it voluntarily and in turn promises not to use it to make regulatory decisions.

The document’s main impact, says Brian Spear, director of pharmacogenomics at Abbott Laboratories in Abott Park, Illinois, is on reassuring companies that conducting early-stage pharmacogenomic experiments will not bring negative regulatory conse-quences. “This has been a real worry,” says Spear. “In the past companies have avoided generating this data. But I’m guessing that the risk has been so removed by this guidance that this is not going to be an issue.”

Pharmacogenomics involves identifying biomarkers, genes that determine a patient’s

drug metabolism risk for a specific disease, drug metabolism or potential to respond to a specific therapy. The guidance document defines what the agency considers a valid bio-marker, but what remains to be fleshed out is the process by which biomarkers come to be validated by the agency and the scientific community. If a company is working with an experimental marker for which it has not been required to submit data, says Donna Mendrick, vice president of toxicogenomics at Gene Logic in Gaithersburg, Maryland, “the concern is that next year the FDA will come to you and say it’s valid. Companies want to know, is there a list of valid biomarkers? How do we get our hands on it?”

Although there are a few drugs to be approved on the basis of pharmacogenom-ics data—such as Herceptin (trastuzumab) first approved in 1998 in the US and since followed by Gleevec (imatinib) and Erbitux (cetuximab)—there hasn’t been that many more since. Because the science is still so new, establishing the guidelines will be an evolving process. The document incorporates many suggestions from the Washington-based Biotechnology Industry Organization (BIO).

And further refinement is expected. The FDA on March 28 released a draft concept paper on codeveloping gene-based diagnostics and therapeutics and are planning to publish another one on the role of DNA in microar-rays. In October, an agency committee, of which Mendrick is a member, is scheduled to discuss biomarker validation.

To build the knowledge base necessary to scientifically validate new biomarkers and create the needed regulatory infrastructure, industry will have to take up the FDA’s request of submitting vol-untary genomics data. But how quickly this trend will catch on remains to be seen, experts say. In the past year, the agency has received only about five to ten vol-untary data submissions—surpris-ingly few, says Spear.

To some extent, notes Mendrick, the low number of submissions may reflect a time lag—provisions for submitting voluntary data were only announced a year and a half ago in the draft version of the guidance document, published in November 2003, and completing the studies necessary to submit an Investigational New Drug applica-

tion often takes longer than that. But companies may simply not yet be will-

ing to embrace voluntary submissions, experts say. “There’s still a belief [among companies] that if this is voluntary, why jump if we don’t have to?” says Spear. “The FDA has really tried to come up with reasons, but those reasons are more relevant for people generating the science” than those on the commercial end, he says. “I think different companies will have different levels of eagerness.”

What could entice companies to submit their data, Spear suggests, is evidence that they are benefiting from the process—for example, by gaining credibility with the FDA, or by edu-cating the agency early on about techniques and issues relevant to their product. According to Carol Reed, vice president of medical affairs at New Haven-based Genaissance, one of the few companies who have submitted volun-tary genomics data, sitting down to a detailed informal discussion with the agency did in fact have very real benefits. “They had very good questions, they’d done their homework, and they were very helpful to us scientifically,” she said.

All recognize that managing the expected volume of data will be a challenge as data acquisition and analysis could bog the FDA officials down, perhaps even draining resources from regulatory review. “Part of the problem for the FDA is, how much data do they want?” says Mendrick. “It’s going to add quite a bit of work to their plate.”

Alla Katsnelson, New York

Dav

id E

ulitt

/KR

T

Breast cancer was the first condition for which a so-called personalized medicine called Herceptin was approved. With the new Pharmacogenomics guidelines, the FDA has given a clear political signal that it is ready to open the door to many more personalized drugs.

NEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


EPO neem patent revocation revives biopiracy debate

Although the revocation on March 8 by the European Patent Office (EPO) of a patent based on the fungicidal properties of the neem tree was hailed as an historic victory over biopiracy, the decision may have mainly a symbolic value. Yet, it could send a signal to biotech companies operating in developing countries that they can-not equate the traditional knowledge of indig-enous communities with freely available public domain information.

The ruling by the EPO’s technical appeals board not to allow an appeal brought by the current patent holders, Certis, of Columbia, Maryland—part of Japan’s largest general trad-ing company, Tokyo-based Mitsui—and the US Secretary of Agriculture, finally brings to a close a battle that has lasted for more than a decade. The EPO has yet to publish the technical appeals board’s full decision as the revocation was only announced orally at the Munich hearing in March.

The EPO originally awarded the patent, number 436,257, to the agrochemical company WR Grace, of Columbia, Maryland, and the US Secretary of Agriculture on September 14, 1994. The patent disclosure described a method for controlling fungi on plants using a hydropho-bic oil extracted from the seeds of the neem tree (Azadirachta indica). Neem is a subtropical mem-ber of the mahogany family, which is native to the Indian subcontinent, and now widely cultivated elsewhere. The disputed patent is one of several dozen based on the properties of the same plant, which are held by Indian and overseas companies (see Table 1).

In 1995, a trio of activists launched their opposition. The group, which was represented throughout by Fritz Dolder, professor of intel-lectual property at the University of Basel, included the veteran Indian anti-biotechnol-ogy activist Vandana Shiva, Magda Aevolet, then president of the Green grouping in the European Parliament, and Linda Bullard, then vice president of the International Federation of Organic Agriculture Movements (IFOAM), a Bonn, Germany-based umbrella organization for organic farmers. An oral hearing of the EPO’s opposition division, held on May 9–10, 2000, found in their favor, ruling that the patent lacked novelty because its opponents had successfully demonstrated prior public use in India.

Central to the opponents’ case was that the neem tree’s fungicidal properties were known about and used in India for centuries. Crucially, however, they backed this claim up with docu-mentary and oral evidence of field trials of simi-lar products conducted in India in the 1980s, as well as references in the scientific literature

that predated the original 1990 filing. “They have recognized that traditional knowledge is a potential prior public use but you have to pro-vide evidence of it,” says Dolder, adding, “It all depends on the precision of the evidence.”

“In a sense, the decision says to the EPO ‘widen your search in relation to novelty,’ and that’s a good thing,” says Julian Kinderlerer, assistant director of the Sheffield Institute of Biotechnological Law and Ethics in the UK. In practice, he says, patent examiners tend to check whether a claimed invention has been patented previously instead of establishing whether it is genuinely novel.

Coincidentally, while the EPO was hearing the neem patent appeal in Munich, delegates from eight developing countries, including India, tabled a fresh proposal at a World Trade Organization (WTO) Council meeting on Trade-Related Intellectual Property Rights (TRIPS) during March 8–9 in Geneva. The proposal would impose additional rules regard-ing the sharing of benefits resulting from pat-ents that involved the use of genetic resources. Biotech companies filing new patent applica-tions would have to include evidence that such benefit-sharing agreements were in place. This is part of a wider, albeit unsuccessful, effort to mandate the inclusion of additional disclosures in patent applications, such as the country of origin of genetic materials and information on the extent to which an invention relies on tradi-tional knowledge.

However, Tony Taubman, director of the traditional knowledge division at the World Intellectual Property Organization (WIPO), the Geneva-based UN agency for intellectual prop-erty (IP) protection, says there has been a trans-formation in the level of recognition accorded to traditional knowledge in debates on IP over the past three years. Individual countries are implementing their own requirements on dis-closure, while at the international level, practical measures, such as including traditional knowl-edge repositories in the patent examination process and amending the International Patent Classification system to include categories of traditional knowledge, are gaining ground. WIPO is also exploring deeper questions, such as “how to frame a true international norm” for preventing the misappropriation of tradi-tional knowledge. The agency was, as Nature Biotechnology went to press, due to publish a new text on the issue.

Several observers have pointed to the egre-giousness of the original WR Grace patent. That it has been overturned now may well point to the weakness of the patent examination process

of the early 1990s and to the glacially slow nature of the EPO’s appeals process rather than to any significant implications for international mul-tilateral agreements on IP or for biotechnology companies with legitimate bioprospecting activ-ities. “It is possible to exaggerate the importance of this,” says Graham Dutfield, senior research fellow at the Queen Mary Intellectual Property Research Institute at the University of London.

Yet, Calestous Juma, professor of the prac-tice of international development at Harvard University, in Cambridge, Massachusetts, and former executive secretary of the United Nations’ Convention on Biodiversity (CBD) points out, “It energizes those who have been questioning for a long time the patenting of products based on traditional knowledge.”

According to IFOAM’s Linda Bullard, the EPO decision may give some impetus to stalled discussions on the Doha Mandate. The mandate requires members of the WTO to consider the relationship between the CBD and the TRIPS agreement on internationally accepted IP pro-tection standards, and to examine the protection of traditional knowledge, in their review of the TRIPS article on the patentability of life forms. “The next obvious step is to take this jurispru-dence into other legal regimes and international patent agreements,” she says.

Cormac Sheridan, Dublin

Caption: Neem oil, cultivated from the eponymous tree’s seeds, is used widely in India as home remedy thought to cure a wide variety of ailments

Cra

ig F

ujii/

AP

Wid

e W

orld

NEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Table 1 Neem patents currently examined by EPO that could be affected by the revocationTitle Applicant Proprietor Relation to neem Patent number

Extraction of substrates from components of the neem tree

Flavex Naturextrakte(Rehlingen, Germany)

Process for extracting substances from neem seeds and nuts

EP 874550 A1 19981104 WO 9725867

Azadirachtin compounds extracted from Azadirachta indica, compositions and use thereof as insecticidal

Fortune Biotech (Secunderabad, India)

Stable azadirachtin prepared from neem kernel extract

WO 0024731 A1 20000504 EP 1124818

Biological wood treatment agent based on extracts of herbs

Jobeck (Hausham, Germany) Treatment agent including neem oil EP 1149670 A1 20011031

Method for acaricidal and microbiocidal treatment of textile materials

Greenworker (Cyprus) Textile comprising neem oil WO 03002807 A2 20030109 EP 1432866

Topical cosmetic composition with skin rejuvena-tion benefits

Avon Products (New York) Cosmetic composition comprising a blend of neem seed

WO 03041636 A2 20030522 EP 1441686

Extraction method Neem Biotech (Cardiff, UK) Extraction of azadirachtin EP 1326870 A1 20030716 WO 02032907

Substrate, especially textile substrate, and method for producing the same

Terra Nostra Produkte mit Naturextrakten (Geisenfeld, Germany)

Substrate comprising extracts of neem tree

WO 03071871 A1 20030904 EP 1478230

Compositions and delivery methods for the treat-ment of wrinkles, fine lines and hyperhidrosis

Avon Products (New York) Limonoids for treating skin; limonoids include the plant alka-loids toosendanin

WO 04060326 A1 20040722 EP 1471874

Pediculicidal compound Natural Science.com (Powys, UK) Composition comprising extract from Melia azadirachta for treating head lice infestation in humans.

EP 1465648 A1 20041013 WO 03057231

Source: Kein Patent auf Leben initiative (No patent on life), Munich Germany. CS

NEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


On May 3, the first case appealed to the US Court of Appeals for the Federal Circuit concerning the pat-entability of expressed sequence tags (ESTs) is due to start, follow-ing a dispute between St. Louis, Missouri-based Monsanto and the US Patent and Trademark Office (USPTO). The case is likely to clearly define how much knowl-edge a patentee needs to have of how newly discovered ESTs work and whether the defined use meets the USPTO requirement of specific and substantial utility. If favorable to Monsanto, the judgment could annihilate or support longstand-ing efforts by the biotech research community to work with a stricter definition of ‘utility’ in patenting.

The issues related to DNA pat-enting have been debated for more than a decade. The May court case, In re Fisher, is the latest develop-ment in a long series of discussions since the early nineties between the USPTO, the National Institutes of Health, the National Academy of Sciences, and other government agencies and academic institutions, and the biotechnology and phar-maceutical industry attempting to define the appropriate scope of DNA-related patent claims and the invention’s utility.

In Fisher, the specific question is ‘how much does a patentee have to know about the func-tion and role of ESTs to actually patent them’? The patent application under scrutiny was initially submitted by Monsanto and then rejected by the USPTO on the grounds that its claims in the written description of the invention were too broad and that it didn’t sufficiently demonstrate “specific and sub-stantial” use of the invention.

This decision was appealed by Monsanto within the USPTO and the rejection upheld, although the issue of overly broad claims was dismissed whereas the lack of utility was affirmed by the USPTO’s appeals board. The federal circuit court is expected to rule on the matter sometime this summer; then if either party appeals the decision, the case could make its way to the US Supreme Court.

ESTs are a sequence or fragment of DNA, which may or may not code for a particu-lar protein. These proteins can enable a researcher to trace back and determine both the proteins’ RNA and DNA, thereby poten-tially linking a disease or biological process

Although, Monsanto could not comment due to the pending liti-gation, the court case could indeed serve to codify the most recent guidelines issued by the USPTO in 2001 the utility of an invention intended to address patenting of genetic and related information.

Along with several other par-ties including Eli Lilly and Dow AgroSciences, both of Indianapolis, Indiana, the National Academy of Sciences and the AAMC are part of an amicus brief in support of the USPTO filed against Monsanto in the federal case. The basis of their opposition is straightforward: by many in the research community, ESTs are seen as research tools that should be part of the public domain unless they are clearly defined in relation to specific bio-logical processes.

Some are concerned that if EST patent applications with overly broad claims—many are worded so as to potentially annex any adja-

cent nucleic acids which could theoretically expand it out to the entire chromosome—or nonspecific utility are approved, it could have implications for other homologous, genetic processes. Although ESTs have largely faded from the current forefront of research, this would affect areas of biotech research still in vogue such as kinases, which are also based on homologous processes.

If the court favored Monsanto, it would be going against the utility requirements of the USPTO. Such a scenario would potentially open the door for increasingly less specific patents thus threatening a decade of work by the USPTO and the biotechnology research community to define utility strictly.

“It’s akin to the old Spanish, English and Portuguese explorers,” concludes Korn. “They would take their boats until they found some edge of land, then they would go up and plant the flag of their king or queen. They didn’t know what they’d discovered; how big it is, where it goes to—but they would claim it anyway.”

Stacy Lawrence, San Francisco

US court case to define EST patentability

to the DNA associated with it. This ‘linking’ would qualify as “specific and substantial utility,” if disclosed. The vexing intellectual property issue in Fisher is to determine when researchers have enough knowledge of this process to lay claim to particular ESTs. In this court case, a general, nonspecific utility is defined in the patent. Monsanto’s plant ESTs described in the patent application aren’t linked to any particular biological or disease process.

This can be a problem, notes the Washington, DC-based Biotechnology Industry Organization’s director of intellectual property Lila Feisee. “It’s not very complicated to find ESTs. It’s not like you’ve done invention of any sort, basically anyone can do it.” BIO does not have an official position on EST patents.

Monsanto, even though it is a voracious patent acquirer, may not be clinging fiercely to this patent case out of the desire to further expand its patent portfolio. Some industry observers have speculated that Monsanto floated this as a test case. “It’s a very impor-tant test,” noted the senior vice president of the division of biomedical and health sci-ences research of the Washington, DC-based Association of American Medical Colleges (AAMC) David Korn. “I wouldn’t be amazed if somebody from Monsanto said they were doing this deliberately to test the guidelines.”


www.nature.com/news

Companies patenting express sequence tags “are akin to Spanish conquistadors” finding a new territory, who did not always know what they had discovered, but would claim it anyway.

Bria

n G

able

NEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Syngenta’s gaff embarrasses industry and White House

Late in March it was widely reported that the Swiss agribusi-ness group Syngenta had inadver-tently mislabeled and sold Bt-10, an unapproved genetically modi-fied (GM) corn seed, as Bt-11, which is approved, to US farmers between 2001 and 2004. Although the US government deemed the matter to be a legal rather than health or environmental matter, this latest industry public rela-tions mishap may strengthen calls to tighten legislation on geneti-cally modified (GM) products in the US. Meanwhile, the incident provided new ammunition for an old gripe over trade between the US and the EU, which launched its own investigation.

After Nature broke the story on March 24, Syngenta disclosed that its Bt-10 test line somehow found its way into five produc-tion lines of Bt-11. Bt-10, which like Bt-11 contains a toxin gene from the soil bacte-rium Bacillus thuringiensis (Bt), was kept around purely for research purposes as it did not prosper quite as well in fields as Bt-11. Syngenta says the amount of Bt-10 corn that was sold as Bt-11 would cover an estimated 37,000 acres. For a sense of scale, during that same time 113 million acres of GM corn were planted in the US.

We may never know exactly how or when the comingling occurred, to what extent the global food system was contaminated, or how Syngenta calculated its acreage proclamation. But, all agree that the fact that it did occur suggests that there was some sloppy handling of materials that should have been treated with the utmost of care at all times for any number of reasons—some scientific, others purely political.

Quite predictably, the incident triggered a chain reaction of high-voltage commentary from some European regulators and biotech critics who all but likened the event to the release of a plume of radioactive particles into the atmosphere and chided the com-pany and regulators for putting the public and the environment at risk. “Incidents like the one with Starlink and Syngenta,” says Steve Strauss, professor of Forest Science at Oregon State University in Corvalis, “unfortunately strengthen the case for tightening regulations not loosening them at a time when regula-tions for [biotech versus nonbiotech crops] are already totally out of sync with actual and

relative risk factors.” (Nat. Biotechnol. 19, 11, 2001).

Although activists have charged that the biotech industry is not to be trusted, Syngenta did, in fact, report this incident immediately upon discovering it in December to the US Department of Agriculture (USDA), the US Food and Drug Administration and the US Environmental Protection Agency (EPA) just as regulations require. US regulators quickly confirmed that Bt-10 posed no human or environmental threat.

Because Bt-10 is not US government approved, planting, selling, distributing, comingling or shipping it without a special government permit is a violation of the Plant Protection Act. Thus far, the USDA has deter-mined that Syngenta was guilty of breaking laws on GM plants and levied a $375,000 fine. The EU, on 15 April, announced its intention to require imports of corn-based feed to be certified as free of Bt-10. Meanwhile, the EPA and EU have launched their own investiga-tions, which could result in more fines for the company—and, industry insiders fear, per-haps new regulations for the whole of the GM crop industry.

Syngenta made much of the fact that the Bt-10 corn is identical to Bt-11, which is approved for human consumption in the US, the EU and Japan. In fact, they are similar but not identical. Bt-10 differs from Bt-11 is that it contains an inactive marker gene which originally conferred resistance to ampicil-lin, a commonly used antibiotic. The inac-tive gene is a relic from the process used to select transgenic corn cells during strain con-struction. The release of such genes into the environment has been contested in the past

because of the small chance that functional versions could transfer from crops to microorganisms and spread problems of antibiotic resistance. “But for the purposes of the government’s investigation,” says Jim Rogers a spokesman for the USDA, “this is not a question about exactly how similar or dif-ferent they are, or about public safety. Both are nearly identical and both are safe. But, only one of them is approved.”

Still, Syngenta did them-selves, newly installed US Trade Representative Rob Portman and the biotech industry no favors—not just by letting the Bt-10 seeds slip off radar in the first place,

but also by taking nearly 4 months to pub-licly release information that something was amiss.

Friends of the Earth Europe was prompt to issue a statement saying, “This is an industry out of control … This [Syngenta] case makes a complete mockery of the US regulatory system for GM crops.” Even the relatively mild-mannered Council for Responsible Genetics in Cambridge, Massachusetts, called it a “massive failure of the US regulatory system … This is cer-tainly going to be a big problem for the United States.”

Indeed, the White House has been dragged into this affair because of the potential of Bt-10 to further complicate trade negotia-tions with Europe. The incident has, however, not alarmed skittish agbiotech investors, who surely would have fled Syngenta shares by now if they believed, as they did with Monsanto shares in 2000 in the wake of a laboratory study on the impact of Bt pollen on Monarch butterfly larvae (Nat. Biotechnol. 17, 627, 1999), that political or regulatory trouble was in the offing. In fact, Syngenta’s share price, after an initial drop when news broke of the Bt-10 inci-dent, remains close to its 52-week high.

Syngenta is not in the clear yet, how-ever. Regardless of what regulators decide to do about Syngenta’s “unintended event,” Margaret Mellon of the advocacy group Union of Concerned Scientists says the damage has been done to both the company and the industry. “Environmentalists and the media might have overreacted to this incident,” she says. “But it was Syngenta that mishandled things from beginning to end.”

Stephan Herrera, New York

Syngenta recently admitted that unapproved genetically modified corn seed, known as Bt10, found its way to US farmers between 2001 and 2004

Teak

Phi

lips/

St. L

ouis

Pos

t-D

ispa

tch

NEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


EMEA fee hike

On March 31, the European Commission (EC) proposed to raise fees for the authori-zation of pharmaceuticals at the European Medicines Agency (EMEA). Although the agency received 51 new applications in 2004—a sharp rise from the 39 received in 2003—the increase was not introduced spe-cifically to deal with higher numbers of appli-cations as they are expected to slow. Instead, the fee hike would mainly cover additional tasks such as more intensive post-authoriza-tion monitoring. These will be undertaken starting in November 2005 as the new legisla-tion comes into force. The annual fees will go up by 10% as of 2006 and a new and variable fee for yet unspecified ‘scientific services’ will be introduced. In addition, biotech compa-nies falling in the small and medium-sized enterprises category would get deferments and discounts. The industry’s total contribu-tion to EMEA’s budget would thus increase slightly from 68.1% this year to 68.4% in 2006. Another measure would introduce new fees that would range between those for new products and traditional generics for follow-on biologics, known as ‘biosimi-lars.’ The European bioindustry association EuropaBio in Brussels welcomes the new fee as an acknowledgment that biosimilars need more review than traditional generics. PV

Israel tightens IP lawThe Israeli Knesset in March approved a data exclusivity bill that will give foreign compa-nies who register drugs in Israel five years’ protection from copying by the country’s strong generics industry. Unlike the US and Europe, Israeli patent law has, until now, included neither market exclusivity, which prevents generic companies from marketing generics for a number of years, nor data exclu-sivity, considered more stringent because it blocks generic companies from accessing the data from the drug’s regulatory file. Both the EU and the US have put pressure on Israel to apply stronger intellectual property pro-tection. Last May, the issue became heated when the US threatened to put Israel back on its priority watch list of countries violat-ing intellectual property rights, after having removed it in 2003. How the legislation will affect drug companies’ willingness to register

drugs in Israel remains to be seen, say Tamar Morag-Sela and Ilan Cohn of the Tel Aviv law firm Reinhold Cohn and Partners. Some industry leaders have insisted that the lack of protection has kept foreign drug compa-nies from building research and manufac-turing facilities in Israel, resulting in a loss for the economy. But Chaim Hurvitz, vice president of generics pharmaceutical com-pany Teva has said that the new bill will cost the national health system ILS300 ($68.5)million a year. AK

Tysabri down and out?The retrospective reclassification of a patient who died during a phase 3 clinical trial of Tysabri (natalizumab) as monotherapy for Crohn’s disease as having developed progressive multifocal leukoencephalopa-thy (PML) appeared to eliminate one big imponderable that had clouded discussions of the drug’s future. The two previous PML cases—which prompted its voluntary with-drawal on February 28—had arisen during the Sentinel phase 3 multiple sclerosis trial comparing a combination of Tysabri plus Avonex (interferon β-1a) with Avonex alone. The possibility remained that those adverse

events had arisen because of an additiveeffect of the two drugs, which enabled Elan, of Dublin, Ireland, and Biogen Idec, of Cambridge, Massachusetts, to hold out some hope that the monoclonal antibody could return to the market, albeit under restricted conditions. Such a return is now a “signifi-cant challenge,” says Richard Parks, analyst at ING Financial Markets in London. The association between Tysabri and PML, “is not just limited to combination therapy, and even shorter doses of Tysabri can result in a predisposition to this potentially fatal side effect,” he says. The patient in the Crohn’s disease trial had received the drug for just eight months, whereas those in the Sentinel trial had each been on Tysbari for more than two years. However, there were additional confounding factors attached to the third PML case, says Jack Gorman, analyst at Davy Stockbrokers in Dublin, as that patient had also received Remicade (infliximab) and the immunosuppressant azathioprine, which has also been linked to PML. CS

News in Brief written by Ichiko Fuyuno, K.S. Jayaraman, Alla Katsnelson, Stacy Lawrence, Sabine Louët, Cormac Sheridan & Peter Vermij.

India’s approves fast track for GM crops

The approval process for genetically modified (GM) crops and recombinant medicines in India is to be put on the fast track under a new policy, which is part of a ‘National Biotechnology Development Strategy’ drafted by the Indian Department of Biotechnology (DBT), and is aimed at accelerating the pace of biotech product development. The policy, announced on April 2, by science minister Kapil Sibal calls for strong support for indigenous discovery of new genes and promoters in agbiotech, but authorizes Indian institutions to license them from multinationals if it’s in the national interest (most of the 40 genes currently used in India are imports).The policy calls for the creation of an independent single regulatory authority with separate divisions to handle applications for commercializing genetically modified crops, recombinant drugs and food products and genetically engineered industrial products. Biosafety will be tested by DBT but exemptions will be made for biosafety tests or large-scale trials of a transgenic crop variety if the same transgene has already been released in another variety. The industry has also welcomed the curtailment of the role of the existing Genetic Engineering Approval Committee (GEAC)—the apex regulatory body under the ministry of environment. Its role has now been limited to ensuring that the released ‘event’ has no harmful impact on the environment. “The policy has rightly defined GEAC’s role and this will help speed up introduction of GM products,” comments Varaprasada Reddy, CEO of Shantha Biotecnics Private in Hyderabad. KSJ


www.nature.com/news

EPA

/STR

NEWS IN BR IEF©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Report flags US pricing

Future revenues of many biotech companies are at significant risk because of new US government powers to lower the costs of drugs administered by physicians and clinicians, the UK-based industry research firm Wood Mackenzie in Edinburgh warned in Medicare Insight, a report reviewing 520 drugs from 60 manufacturers, which was published on April 5. The study high-lights the Medicare Modernization Act (MMA), affecting 40 million senior Americans and signed into law in 2003, which has given administrators at the US Centers for Medicare and Medicaid Services (CMS) authority to control prices in a market heavily targeted by biotech firms. This makes “20% per annum price increases in this class a thing of the past,” the report says. One new option for CMS is to link reimbursement levels to average market prices, and the practice is already “wreaking havoc in oncology,” says Keith Redpath, Wood Mackenzie’s vice president of life sciences. Many oncologists report they can no longer afford to supply patients because of lower reimbursements, forcing producers to market to wholesalers instead. Anticipating steep rises in overall Medicare costs, Redpath expects the US government to use all the new tools in its box. He adds, “I would not be sur-prised if CMS is going to press the [US Food and Drug Administration] to more quickly approve generic versions of existing biotech drugs.” PV

Japan investors shunbiotechOn March 29, Effector Cell Institute, a Tokyo-based university spin-off that aims to devel-op cancer and anti-allergy drugs, was floated on the Nagoya Stock Exchange’s Centrex market. But it failed to trade because investors consid-ered it overpriced. The next day, the stock first traded about 40% lower than the initial pub-lic offering price, and has subsequently fallen further. Many analysts say Effector’s dramatic downturn results from investors’ concern that the company’s outlook is too optimistic given that it has not even started clinical studies. That represents a shift among investors—most of which are individuals—towards a more pru-dent attitude. When a raft of Japanese biotech companies started to trade around 2002 and 2003, investors were bullish on them, leading the biotech sector to reach its peak in the spring of 2004 (Nat. Biotechnol. 21, 1256–1257, 2003). “The market was new, so investors didn’t have clear judgment,” says Kenji Tsujimoto, manager at Nomura Research & Advisory, the research arm of brokerage giant Nomura Securities in Tokyo. But the overall sector plunged by roughy 50% late last year to regain value only this year. Now, Tsujimoto says, investors are looking more carefully to an individual company’s business conditions such as the status of clinical trials, current sales value and business partnership. IF

Selected research collaborationsPartner 1 Partner 2 $

(million)Details

Genentech(S. San Francisco, California)

Curis(Cambridge, Massachusetts)

9 A two-year deal to discover small molecule modulators of an undisclosed pathway that regulatestissue formation and repair and its abnormal activation that is associated with certain cancers. Curiswill use its technology based on use of proteins or small molecules to modulate the pathways of interestto Genentech, which will pay both licensing and research fees. Curis retains rights to resultingcompounds for ex vivo cell therapy in areas outside of cancer and hematopoiesis.

GlaxoSmithKline(GSK; London)

Global Alliance for TB Drug Development (New York)

* A partnership to discover compounds to treat tuberculosis (TB) that would cut two to four months offthe current six- to nine-month treatment as well as to find agents that have fewer drug interactions withantiretrovirals, as patients are often infected with both TB and HIV. The deal covers four projectsincluding the development of pleuromutilins, a new class of antibiotics, the study of two TB targets(isocitrate lyase and InhA) and the screening of GSK’s antimicrobial libraries for novel anti-TB agents.

* Financial details not disclosed. SL

Basic PCR patents expire

Eight patents held by Basel-based F. Hoffmann-La Roche, covering PCR expired in the US on March 28. Otherwise known as the Mullis patents, they cover the fundamental processes behind PCR, the first technique to allow rapid amplification of DNA sequences. The patents were originally acquired from Cetus, now known as Chiron, of Emeryville, California, in 1991, for $300 million. The European and Japanese versions of these patents are set to expire in about a year. According to the company, Roche has more than 800 licensing deals for PCR. For com-panies needing the most current PCR tech-nology, the PCR patent expiration may be of little consequence other than opening the door to performing basic PCR more freely, according to life sciences patent lawyer Simona Levi-Minzi, partner at McDermott Will & Emery in Chicago, Illinois. “But more recent methods are still under patent,” she adds. And Roche still holds hundreds of more current PCR patents. “Naturally, we will experience some reduction in royalty revenue over the short-term,” acknowledges Roche spokesperson Paula Evangelista. “However, we also expect to see increased growth in licensing revenues resulting from the adoption of the real-time PCR methods offsetting some of the losses.” StL

New product tableCompany Details

Symlin (pramlintide acetate)(Amylin Pharmaceuticals,San Diego, California)

On March 16, the US Food and Drug Administration (FDA) announced the approval of the first-in-class antihyperglycemicdrug Symlin, to be used in conjunction with insulin to treat type 2 and type 1 diabetes. Symlin is a synthetic analog ofhuman amylin, a naturally occurring hormone that is made in the beta cells of the pancreas. Diabetes affects over18 million Americans.

Mycamine (micafungin sodium(Fujisawa Healthcare,now Astellas Pharma US,Deerfield, Illinois)

The FDA approved, on March 17, the use of micafungin sodium, the antifungal product for prophylaxis of Candida albicans infections in patients undergoing hematopoietic stem cell transplantation and the treatment of esophageal candidiasis. Mycamine belongs to a new class of antifungal agents, the echinocandins, which inhibit the synthesis of fungus’ cell-wall. Invasive candidiasis kills 10–40% of infected immunocompromised patients. SL

NEWS IN BR IEF©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


or take a license.” Leon Bushara, senior executive vice president of business development at Serono in Geneva concurs: “Larger, better capitalized companies [can] identify projects in [smaller] biotech companies in Europe [and] form partnerships with or without equity investment.”

As an alternative to mature biotech, mid-sized pharmaceutical companies could also be partners, says Thurston. This might be par-ticularly attractive because European family-owned pharma compa-nies are not often quoted on the stock market and frequently “escape the radar screen of US-based analysts.” “It’s the Ipsens, the Pierre Fabres, the Esteves, the Menarinis and the Schwartz Pharmas—all that lot” that biotech should be focusing on, Thurston suggests.

So why aren’t there more deals between mid-sized pharma and biotechs? Thurston cites several reasons: first, mid-sized pharmas are family run and are often unaware of which biotech company is doing what in Europe; second, they are generally risk averse, com-

paratively smaller and have constraints on resources; and third, they “only look at a small number of projects each year.” Warburg Pincus’ Turton points out that the parochial nature of many typical fam-ily-run pharmas is the real deal killer. “The majority are regional companies who are strong in their territory,” he says. But they are often not the partner of choice for biotechs seeking to reach the pan-European market.

With funding in Europe at such a premium, many firms are now turning to the United States for partnering. In March, one of the flagships of French biotech, Immuno-Designed Molecules of Paris (which failed to float in Europe last June), underwent a reverse merger with Californian Nasdaq-quoted company Epimmune of San Diego. According to Thurston, the sad truth is “more science in the US that’s average gets funded.” In contrast, for early-stage European biotech to really succeed, he believes “science has to be exceptional.”

But the major stumbling block for Europe is parochial attitudes. European resources are “far too splintered and too nationalistic,” says Thurston. In Europe, old habits die hard and cultural differences make consensus difficult. Each country tries to grow its own biotech industry, regardless of the fact that European biotech can only thrive by being a borderless deal-making activity.

“It is going to take time to be as good as San Francisco, San Diego or Boston,” he adds. “We should be focusing on just two or three clusters and cross-fertilize the science Europe-wide,” he adds. The UK and Switzerland would be good places to start.

Sabine Louët, Dublin

Julian Thurston started his career as general counsel for what was once the flagship of UK biotech—Celltech, now part of Belgian mid-sized pharma company UCB. For the past 26 years, he has been providing legal expertise on commercial exploitation, intellectual property (IP) protection and licensing, partnering and strategic plan-ning to European biotech companies. Thurston wears many hats—he deals with technology transfer as a nonexecutive director for the UK Cancer Research Campaign and currently is a partner at London law firm Morrison and Foerster. He is now regarded as one of the top life sciences transactional attorneys in Europe.

If nothing else, a quarter century of experience has taught Thurston that biotech is cyclical—its fortunes waxing and waning with inves-tor whim. In Europe, the latest trend is the desertion of early-stage biotech by the venture capital (VC) community. “I think that the classic series A, B and C IPO [and initial public offering] model is not going to work in Europe in the future,” he remarks. To get around this, he believes the European biotech sector needs to rethink its business strategy.

One option would be to promote consolidation. Thurston’s vision is that the European sector should “sell a package of several startups” with complementary technology or products to partners of choice. “There needs to be mechanisms across universities where a German university would link up with a UK university and with an Italian university,” he explains. By packaging several university spinouts together, companies would become more substantial and sustain-able. Peter Heinrich, CEO of Medigene, of Martinsried, Germany, agrees, adding “those focused on niche markets would stand a better chance of being competitive.”

Heinrich is, however, skeptical about whether European investors have the necessary vision to accomplish this. “VCs should drive this [type of] merger,” he says, but whether they will is another matter. Many mergers have failed and consolidation has not been as extensive as it could have been because of what he calls “the ego problem.” This affects both CEOs and venture capitalists: the former bickering over who should get top job at the new company, the latter obsessively focusing on their reputation and the relative valuation of the com-pany they are backing. Simon Turton, an investor at private equity firm Warburg Pincus in London recognizes that ego has a lot to do with the process but points out that “mergers are so difficult to do!” and they may not be the only driver of growth of the sector.

Once consolidation has taken place, Thurston believes that the aggregated companies could be sold to bigger players. And as far as suitable partners go, he thinks that the most likely acquiring candi-dates for this packaged technology would be mid-sized biotech com-panies, such as UK firms Cambridge Antibody Technology (CAT) in Cambridge or Vernalis in Winnersh, which are always on the lookout for partnering or acquisition opportunities.

Where acquisition is not possible, Thurston thinks companies could enter partnership deals. This is exactly the kind of strategy that CAT has adopted. As CAT CEO Peter Chambré explains, “If we’re convinced of the quality of the opportunity, we will ideally want to take full rights to it. If they are not available, we will partner

Julian ThurstonA commercial lawyer in the UK who has been around thetrack more than most, Julian Thurston thinks shortsighted investment strategies in Europe are shackling early biotech venture development.

…for early-stage European biotech to really succeed, he believes “science has to be exceptional,” says Thurston

PROF ILE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Mixed results in Q1Stacy Lawrence

Anyone hoping for a resurgence of biotech stocks last quarter was disap-pointed. Even so, initial public offerings (IPOs) and follow-on offerings did make something of a recovery, up 55% and ~25%, respectively over the fourth quarter. IPOs on European and Japanese markets had three

out of the five largest IPOs of the quarter. Biotech venture capital held its ground, with most going to European companies. But total biotech funds raised from partnering and debt are down from the same quarter in 2004 by 41% and 74%, respectively.

Notable first quarter biotech dealsCompany (lead underwriter) Amount raised

($ millions)Percent change in stock price since opening1

Date launched

IPOs MediciNova (Daiwa Securities SMBC) 121 N/A 8-Feb

Aspreva Pharmaceuticals (Merrill Lynch, Pierce, Fenner & Smith)

79 11% 3-Mar

Intercell (Goldman Sachs, Lehman Brothers)

69 N/A 25-Feb

ViaCell (Credit Suisse First Boston) 60 -51% 21-Jan

Paion 52 N/A 11-Feb

80.0

85.0

90.0

95.0

100.0

105.0

110.0

115.0

120.0

1/2

/20

04

3/2

/20

04

5/2

/20

04

7/2

/20

04

9/2

/20

04

11

/2/2

00

4

1/2

/20

05

3/2

/20

05

Biocentury Biotechnology

Nasdaq Biotechnology

NasdaqSwiss exchange

Dow Jones

Inde

x

$0.0

$100.0

$200.0

$300.0

$400.0

$500.0

$600.0

$700.0

$800.0

$900.0

$1,000.0

1Q04 2Q04 3Q04 4Q04 5Q04

North America

Europe

Asia-Pacific

580.7265.570.0

396.9161.0

0.0

661.736.817.9

264.378.816.4

298.091.6

104.0

Number of IPOsNorth americaEuropeAsia-Pacific

821

1412

631

651

730

0

200

400

600

800

1000

1200

1400

1600

180061.6

220.21,343.2

0302.0968.7

0228.3876.5

1.5345.7891.9

2.5446.8750.2

1Q04

2Q04

3Q04

4Q04

1Q05

North America

Europe

Asia-Pacific

Biotechnology stock market performanceBiotechnology indices have performed worse than the broader market, actually losing value since the beginning of 2004.

Source: Multex, BioCentury

Global biotech initial public offeringsAfter decreasing through 2004, there was an upturn in IPO money raised last quarter.

Source: BioCentury

Global biotech venture capital investmentVC funding totals have been virtually unchanged for the last four quarters, but Europe has started claiming an increasingly large share.

Source: BioCentury

Researcher Investor Value ($ millions) Deal type

Licensing /collaboration

Coley Pharmaceutical Group

Pfizer 515 Collaboration, development, license

Basilia Pharmaceutica

Cilag AG 300 Co-promotion, development, license, manufacturing, marketing

ANDRx First Horizon Pharmaceutical

85 Asset purchase

Savient Pharmaceuticals

Ferring 80 Asset purchase

Vertex Avalon Pharmaceuticals

73 Co-promotion, license

Avecia Merck 65 Asset purchase

ICICI Ventures Dr Reddy’s Laboratories

56 Collaboration, development, license

Durect Endo Pharmaceuticals

45 Development, license

Target Acquirer Value ($ millions)

Date announced

Mergers and acquisitions

Hexal Novartis 6,000 21-Feb

Warner Chilcott Bain Capital Partners, DLJ Merchant Banking, J.P. Morgan Partners, Thomas H. Lee Partners

3,100 19-Jan

Medicis Pharmaceutical INAMED 2,800 21-Mar

Eon Labs Novartis 1,700 21-Feb

CTI Molecular Imaging Siemens 1,000 18-Mar

Kendro Laboratory Products

Thermo Electron 834 20-Jan

Angiosyn Pfizer 527 20-Jan

ESP Pharma Protein Design Labs 475 25-Jan

0 5000 10000 15000

1Q04

2Q04

3Q04

4Q04

1Q05

2,010, 1,006, 1,625, 1,242, 916, 1,387

2,262, 553, 1,271, 1,874, 716, 409

3,017, 792, 1,105, 949, 494, 430

3,644, 1,064, 1,239, 4,476, 360, 578

2,151, 1,305, 1,200, 1,177, 558, 511

Partnering

Follow-on financing

Venture Capital

Debt and other financing

IPO financing

PIPEs

Global biotechnology industry financingAfter a big boost in the fourth quarter, mainly from closure of debt and partnership deals, the first quarter is down substantially.

Source: BioCentury, Burrill & Company

Company (lead investor) Amount invested ($ millions)

Round number

Date closed

Venture capital

FibroGen (Adage Capital Management) 100 N/A 15-Feb

Perlegen Sciences (CSK Venture Capital) 74 4 28-Feb

Alexza Molecular Delivery (NGN Capital) 52 4 6-Jan

Five Prime Therapeutics (Domain Associates) 45 N/A 11-Feb

Predix Pharmaceuticals (Forward Ventures, Boston Millennia Partners, and CMEA Ventures)

43 3 25-Jan

Neuro3d (Gilde Investment Management) 43 3 3-Jan1Prices as of March 30, 2005. Source: BioCentury, Hoover’s, Recombinant Capital

DATA PAGE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Gene therapy: cursed or inching towards credibility?

On February 9 the US Department of Justice announced a civil settlement in the govern-ment’s case related to Jesse Gelsinger’s death during a gene therapy trial at the University of Pennsylvania in 1999 (see News in Brief, p 515). Gelsinger, just 18 when he died, was not the first person to die while receiving experi-mental treatment, but both the circumstances of his death and the anxiety many people feel about genetic manipulation in general brought the case extraordinary attention. “There was a definite negative effect on the entire field [after Gelsinger’s death], as we all paused to reconsider what we were doing,” says Richard Gregory, head of research at Cambridge, Massachusetts’ Genzyme.

This, plus the latest round of interruptions of some gene therapy trials, less than a year after a high profile French trial was resumed (Box 1), has put this once promising field back on the hot seat, just when researchers and companies felt they were finally making some progress. Will this latest round of negative publicity put gene

therapy back in the doldrums or can an approval, believed by some to be coming in the next 2–3 years, finally bring some commercial success to the sector?

Heavy baggageNowhere in biotechnology has the promise been more tantalizing and the failures more devastating than in gene therapy. The idea that scientists could treat the errors in our very DNA initially gave new hope to many, particularly the desperate parents of children with fatal inherited disorders such as cystic fibrosis or muscular dystrophy. Those dreams have been dashed, as repeated set backs have forced investigators to completely rethink the best ways to first move their brave new thera-pies into medical practice. Now that one of the bleakest episodes in gene therapy’s history has finally ended, at least officially, have we entered the dawn of better gene-based therapies? The answer seems to be yes, if the latest wave of products fulfill expectations.

Most agree a pause for reflection was war-ranted, and that the result has been a much better understanding of the underlying sci-ence. But some investigators feel the new regulatory restrictions go overboard, unduly shackling those already using appropriate caution, by adding steps to an already diffi-cult process.

Ironically, although Paul Gelsinger, Jesse’s father, does believe the technology was hyped prematurely, he doesn’t blame gene therapy for his son’s death. “The problem wasn’t gene therapy,” he says. Rather he faults certain indi-viduals and the system (Box 2). Others saw it differently, however, and gene therapy was branded as an unusually risky field within the already volatile biotechnology sector. Some companies shifted their focus to new fields, or recast their work.

“Because gene therapy has such a nasty reputation, people tried to rename it or call it ‘new and improved’ to free themselves of the stigma,” says Michael Zasloff, an analyst with Ferris, Baker Watts of Washington, DC. “That may fool the public, but the market sees through it,” he says. Wherever genetic manipu-lation is involved, no matter for how long or where in the body, investors typically treat it like gene therapy, despite what the product’s developers may say.

Safety and efficacyIt is not just safety problems that have dogged gene therapy, efficacy has been much harder to achieve than expected.

Katherine High, of Children’s Hospital of Philadelphia, has experienced the field’s ups and downs from the front row. Along with Mark Kay’s group at Stanford University and scientists at Alameda, California–based Avigen, she has spent years moving a gene therapy for hemophilia through to human studies. Mouse studies seemed promising and the dog studies were remarkable—with ani-mals producing adequate levels of Factor IX for more than five years. Then, in the human trials, the therapy seemed to be working well in at least one patient. But that quickly proved to be an illusion. Five weeks later, the patient’s Factor IX levels started dropping and quickly reached baseline.

“I have to admit, at first I was devastated. I couldn’t believe it didn’t last,” High says. Someone from Avigen reminded her to “think how the patient feels. For four weeks, he touched the rainbow.” The bottom line is that it has been extraordinarily difficult to get sustained deliv-ery of any gene. Many diseases that originally seemed likely targets have also turned out to be devilishly tough. Cystic fibrosis, for example, was one of the first stops for gene therapy, and

Can gene therapy ever live down its setbacks and live up to its initial promise? A chastened but determined group of pioneers believes it can, and they are pointing to a new generation of products to back up that claim. Malorye A. Branca investigates.

Future gene therapy candidate? Nine-year-old child with Crigler-Najjar syndrome, a genetic disease that causes elevated bilirubin levels, sleeps under UV lights every night.

Kar

en K

asm

ausk

i/Cor

bis

NEWS FEATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


many trials have been carried out by compa-nies including Genzyme. “The CF lung is full of mucus and things like proteases that are hostile to vectors,” Gregory says.

As a result, candidate gene therapies for genetic diseases are dropping like flies. Avigen recently announced it is refocusing entirely, moving into small molecule therapeutics and exploring options for keeping alive its gene ther-apy programs, which include the hemophilia treatment and one for Parkinson disease. “It has just been very tough on the business side,” says Avigen’s Glenn Pierce. “The timeline is long, and the hurdles are bigger than expected.” Strasbourg, France’s Transgene also completely abandoned ‘real’ gene therapy this year, and will now concentrate on its vaccine business. The company’s Duchenne/Becker’s muscular dys-trophy program will continue through support from the French Association against Muscular Dystrophy. Finally, Targeted Genetics recently announced it is ceasing work in cystic fibrosis. Early results from the company’s latest phase 2 trial—the largest gene therapy trial ever con-ducted in this indication—did not confirm earlier encouraging results.

The challenge still is getting enough gene expressed for a sufficient span of time. “The other sad or amusing thing was the gradual discovery of how much the immune system matters here,” says Doug Jolly, a gene ther-apy pioneer and now president and COO of Advantagene of Encinitas, California. An immune response is probably what derailed Avigen’s hemophilia treatment, and it is almost certainly what led to Jesse Gelsinger’s death.

Miracle cures for genetic diseases are hard to deliver. Experts are still optimistic they will come, but the work on these conditions is now largely confined to academia and other research sponsored by nonprofits. Genzyme

may be the one exception here, because it has a long history in the field and a good reason to stay in it. The company launched its first gene therapy trial (for cystic fibrosis) in 1992. It has yet to see a payoff from that work, but in the meantime Genzyme has built a major franchise around protein replacement treat-ment for the major form of Gaucher disease. That treatment, Cerezyme, is remarkably effectively, but patients with rarer forms of this condition are still incurable. “We are com-mitted to serving all of these patients,” says Gregory. If a technology exists that will help more Gaucher patients, or improve upon what

Genzyme already has, the company wants to be the first to get it working. As a result, the company is investigating gene therapies for these alternative forms of Gaucher.

Soberly optimisticThose who’ve been in the field a while are phi-losophical about the problems. “We’ve finally been doing this long enough that bad things are cropping up,” says Gregory. Jolly, mean-while, likes to point out that although trials began in 1990, it took longer for industry to get really involved. “We’ve only been doing seri-ous drug development in this field for about

Table 1 Selected company-sponsored gene therapy trialsCompany/location Indication/treatment site Product/gene Vector Clinical trial phase

Corautus Genetics Atlanta, Georgia

Severe angina/heart muscle Vascular endothelial growth factor (VEGF) 2

Naked plasmid DNA 2b

Genzyme Cambridge, Massachusetts

Peripheral arterial disease/legs HIF-1α (an engineered form of the hypoxia-inducible factor 1 gene)

Adenovirus type 2 2b

GenVecGaithersburg, Maryland

Severe coronary artery disease, angina/ coronary arteries

BIOBYPASS/VEGF121 (proprietary form of VEGF)

Adenovirus 2b

Pancreatic cancer/tumor TNFerade/tumor necrosis factor-α (TNFα) Adenovirus 2

Age-related macular degeneration/eye Pigment epithelium-derived factor (PEDF) Adenovirus 1

IntrogenAustin, Texas

Solid tumors/head and neck, lung, breast, esophagus, prostate, brain, pelvis

ADVEXIN/p53 Adenovirus 1–3

Solid tumors/various INGN 241/mda7 (encodes IL24) Adenovirus 1–2

Solid tumors/lung INGN 401/FUS1 (a tumor suppressor gene)

Nanoparticle 1

Targeted GeneticsSeattle, Washington

Rheumatoid arthritis/joints tgAAC94/TNFα Adeno-associated virus (AAV)

1

Box 1 On the SCIDS

Gene therapy’s big worry has always been that a genetic payload could integrate into the host genome in a trouble spot, where it would cause other diseases or even alter a patient’s germline. That concern now seems to have been validated by a series of events in France that have put gene therapy’s one successful treatment at risk.

The work of Alain Fisher and Marina Cavazzanna-Calvo’s group at Necker Hospital was initially hailed as a stunning achievement, when in 2000 they reported successfully treating X-linked severe combined immunodeficiency (SCID) in infants using retrovirus-based gene therapy. It was called the first real validation of the field. But that victory was marred just a couple of years later, when one of the boys developed T-cell leukemia. Soon after the first case was discovered, another child developed cancer, and then another. One of the three boys died from leukemia last year.

As these developments unfolded, SCID trials around the world were first stopped, then restarted, then stopped again. The FDA also held a special meeting in mid-March of this year to review the problem.

Because X-SCID leads to early death if untreated, the consensus so far is that gene therapy should be considered only in children for whom there are no other treatment options. “It’s still a wonderful success, but with very nasty possible side effects,” says Advantagene’s Douglas Jolly.

In the first two cases, the cancer was triggered in the SCID kids after the retrovirus inserted near the LMO2 oncogene promoter. Something similar occurred in the third case, although a different oncogene was involved. Having reviewed the SCID cases, experts believe this side effect is caused by the very gene being treated in the French trial, because this effect has not been seen in other trials.

NEWS FEATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


12 years,” he says, pointing to the fabled 20 years it took monoclonal antibody developers to get it right.

Important lessons have been learned, not just about side effects, but also about efficacy.

“One rule is ‘Use vectors locally, before you try them systemically’” says High. The realiza-tion that different organs have different side-effect profiles to the same vector has also been critical, reinforcing the importance of vector choice and route of administration. Finally, the explosion in knowledge about certain gene targets has been an unexpected boon. New research on angiogenesis and cancer, in particular, has revealed many attractive gene therapy targets. The result is a nice little wave of progress, albeit in more complicated dis-eases (Table 1). Here the challenge will be find-ing strong end points to measure the therapies against.

As Stephen Dunn of Boca Raton, Florida, securities firm Dawson James sums it up, the issue now is “Right vector? Right gene? Right target?” That’s a much different proposition than curing all genetic diseases with one magic vector, but it’s also a plan that is much more attractive to investors.

“Do I like the new crop of gene therapies more than the old? Absolutely,” says Zasloff. Treating something locally and transiently with a gene therapy is much less risky than permanently altering gene expression. Even better, from the investor’s standpoint, these therapies are targeting more typical markets—conditions such as cancer, cardiovascular dis-ease and rheumatoid arthritis.

No one expects the floodgates to burst open soon, but there is widespread confidence that within the next 2–3 years, a gene therapy will be approved in the US. Others should follow, one by one. “Gene therapy’s first successes will be in something localized like solid tumors or eye diseases,” says Dunn. Work on new vectors will continue. Already, more research is going into the adeno-associated virus, lentivirus and nonviral vectors. High expect major progress on the vector front within the next ten years. Others agree. “In about 15 years, I like to think there will be several gene therapies hitting the market at once,” says Gregory. Although he cautions that if there is another major set back like the one at the University of Pennsylvania, “I can’t predict what the effect will be.”

The first gene therapy approved in the US will most likely be Austin, Texas–based Introgen’s Advexin, which delivers normally functioning p53 to cells. Currently in late phase trials for head and neck, non-small-cell lung

and breast cancer, Advexin is also in develop-ment as a mouthwash to treat precancerous lesions. The most intriguing thing about this treatment is how safe it appears to be. “These are highly specific, minimally toxic and very targeted products,” says Introgen’s president and CEO David G. Nance, who points out that the US Food and Drug Administration has approved trials of Advexin even in precan-cers. He adds that the company has tested its gene therapies in over 700 patients, and never had a trial stopped or put on hold. Introgen, which has several products in development, thus epitomizes the new gene therapy com-pany. Its products are used transiently, tested in combination with other treatments, and locally delivered—Advexin can even be applied directly to a tumor during surgery.

Once one or more gene therapies reach approval in the US, experts believe the big pharmaceutical companies will again start setting up partnerships in the sector. Major players such as Schering Plough of Kenilworth, New Jersey, and Novartis of Basel showed intense interest in gene therapy early on, but have all but abandoned the field over the last few years.

Chinese checkmateA wild card here is China’s bold move into this field. In 2003 the first gene therapy product was approved in that country,

much to many people’s surprise. Probably most surprised were the management at Introgen, whose lead product is quite similar to the Chinese product—Shenzhen SiBiono GenTech’s recombinant Ad-p53 for head and neck squamous cell carcinoma (see Nat. Biotechnol. 22, 3, 2004).

Dunn sees the Chinese connection as impor-tant. China is apparently positioning itself as a leader in the field, and hopes to attract medical tourists from afar with breakthrough thera-pies not available at home. Shenzhen SiBiono claims that about 400 Westerners have already visited China to receive the company’s treat-ment. “Find a Chinese partner,” is Dunn’s advice to gene therapy companies. Everyone in the field is already watching the develop-ments in China closely. Introgen has chosen to file patents there judiciously, and try to work the political scene. It helps that some of the company’s Chinese patents have already been issued. “But we saw what happened to Pfizer,” says Nance. “They marched in with a strong patent for Viagra, and got nothing.”

Advexin is so similar to Shenzhen SiBiono’s product that if the Chinese seek to commer-cialize their product in the US, “It would be an issue,” says an Introgen spokesperson. All the more reason for them to hope they get their US approvals soon and to keep an eye on the East.

Malorye A. Branca, Boston, Massachusetts

Box 2 A final reckoning in Gelsinger case

Just over five years ago Jesse Gelsinger went to the University of Pennsylvania’s Institute of Human Gene Therapy in Philadelphia to take part in a trial aimed at treating inborn ornithine transcarbamylase deficiency. But four days after he received the therapy, Gelsinger died of massive organ failure, apparently sparked by an immune reaction to the adenoviral vector used.

In February of this year, the Department of Justice (DOJ) announced the final civil settlement in the case. The University of Pennsylvania and Children’s National Medical Center each agreed to pay more than $500,000 to the government. Both institutions have also been obliged to shore up patient safety procedures. Clinical research restrictions were placed on the three investigators involved—Penn’s James Wilson and Steven Raper, and Mark Batshaw, a former Penn doctor and now chief academic officer at Children’s National Medical Center. Neither the scientists nor the institutions named admit to any of the government’s allegations.

Jesse’s father, Paul Gelsinger, is now vice president of Citizens for Responsible Care and Research. He says the settlement does not go far enough. Gelsinger maintains that Wilson’s industry ties played a role, that the Gelsingers were repeatedly misled throughout the ordeal and that the FDA should have stopped the trial earlier. “This judgment lets everyone off the hook,” he says.

David Hoffman, the US attorney general who prosecuted the case, says the DOJ, FDA, and NIH all approved the final settlement, and that it should serve as a lesson to investigators everywhere, “To have the sense to always view subjects as people, not just ‘participants.’”

NEWS FEATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


B I O E N E W S

India’s strategy to bridge the public-private divide

Public-private partnership is the cornerstone of India’s new draft National Biotechnology Development Strategy that aims at creating one million jobs and an annual turnover of $5 billion in the biotech industry by 2010.

A draft of the policy for the National Biotechnology Development Strategy released on April 2 by science minister Kapil Sibal is due to be implemented later this year. The policy, which is geared towards encour-aging public-private partnership, has already triggered a debate within the biotech industry because it would allow 100% foreign direct investment, an option that could threaten the local industry, according to opponents.

The new strategy is designed to catapult India into the “global biotech league” says Maharaj Kishan Bhan, secretary to the New Dehli-based Department of Biotechnology (DBT) that drafted the policy. Bhan says the new strategy is built on the conviction that research for the “public good” and research “for profit” should become mutually rein-forcing to help foster the development of innovation in biotech.

The draft policy envisages that by 2010, biopharmaceuticals—mostly vaccines and bio-generics—will be contributing to $2 bil-lion of the annual turnover of the sector in India. Clinical development services is fore-casted to reach $1.5 billion and outsourced research services are estimated to reach $1 billion. The balance of $500 million is attrib-uted to agricultural and industrial biotech-nology. This is an ambitious target but given the growth of the industry in the previous year, the government has strong growth data to support its optimism. The Indian biotech industry grew by 39% between 2003 and 2004 to reach a value of $705 million. Total investment in the sector also increased by 26% during the same period to reach $137 million.

Under the new policy, public funds can be spent on industrial projects while scien-tists who are employees of a public research institution can be seconded to private firms without losing any benefits. Furthermore, the policy states that “at least 30% of

government-funded programs must have a commercial partner who will be respon-sible for directing research and development (R&D) towards commercialization.” “This kind of partnership is very welcome,” says Varaprasada Reddy, managing director of Shantha Biotechnics in Hyderabad, a com-pany that pioneered recombinant vaccines in India in the 1990s.

But one component of the strategy that allows “100% foreign direct investment” is causing resentment in the industry. “The policy should have also insisted on partner-ship between local and foreign companies in the ratio of at least 25:75 [rather] than allowing foreigners total ownership,” points out Reddy.

Indeed, the new policy dispenses with the need for government approval for equity investment in the biotech sector unlike in other sectors like telecommunications or energy. “What this means is multinational companies can come with suitcases full of money, buy up plots, build plants, hire our scientists at low salaries and create wealth for themselves,” says Reddy, adding, “That is a prescription for killing the local industry. We all will be dead.”

Others like Bhimsen Bajaj, president of the southern chapter of the All India Biotechnology Association in Hyderabad, disagree. In his experience, foreign com-panies are not so interested in forming

joint ventures, but he recognizes that with this new policy, they will bring new technol-ogy and generate jobs, thus enabling the country to develop. “There is nothing wrong in this,” he says, adding, “we have to be open minded in these days of globalization.”

The DBT says it will not finalize its strat-egy until it receives all opinions on the draft biotech strategy that is open for comments until May 15. After that, the strategy will be submitted to the cabinet for approval before implementation.

Despite the opposition to this aspect of the policy, India’s priority in supporting public-private partnership could well benefit the industry. As proposed, public-private part-nership can take several forms. In one form government institutes will partner with small and medium-sized companies such as biotech that have bright ideas but lack qualified staff. Once the project passes the proof-of-concept stage the company would become eligible for soft loans for product development and commercialization. The R&D expenditure of public sector institutes, while working on the projects of their private sector partners, will be met by grants from DBT.

In another form, the strategy aims at creat-ing several “Technology Transfer Cells” that will promote the transfer of knowledge gen-erated within publicly funded research insti-tutions to the private sector. Biotech parks promoted by private industry will also get 30% equity funding from government. And the DBT will promote and support at least ten biotech parks by 2010. “This will really give a boost to the biotech infrastructure in the country,” says Krishna Ella, managing director of vaccine and biologics company Bharat Biotech International in Hyderabad.

K.S. Jayaraman, Hyderabad

This story was reprinted with some modification from the BioE News section of the Bioentrepreneur web portal (http://www.nature.com/bioent), 7 April 2005, doi:10.1038/bioent857.

ALSO ON WWW.NATURE.COM/BIOENTUK on a quest for early-stage financing models

Brazil to foster public-private innovative venturesSTART-UP PROFILE: Serenex

The Indian biotech industry has grown significantly over the past few years as it has successfully developed biotech drugs and vaccines

Indranil Mukherjee/Agence France Presse


www.nature.com/news

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy


B U I L D I N G A B U S I N E S S

The role of competitive intelligence in biotech startupsSalvador Carlucci, Anthony Page & David Finegold

Competitive intelligence (CI) gathering is essential to developing a biotech firm’s business strategy, but few startups have sufficient support systems in place to do CI effectively. This article, the first of two on the topic, addresses why it is important and how to do it.

All biotech startups gather competitive intel-ligence (CI), although often they are not aware they are doing so. When a scientist attends a professional conference to learn about emerg-ing technologies and who is working on them, when an employee stops at a rival’s booth at a trade show to pick up information about their products, when a business development expert reads a market report, or even when an execu-tive chats with a friend about trends in their industry, they are gathering CI. The problem is that in most biotech startups no one manages this flow of information, nor is it organized in a systematic way1. As a consequence, the firm does not maximize the value of these critical resources.

In this article, the first of a two-part series, we will describe the types of competitive intel-ligence and the benefits for a small startup, and provide some examples of CI both working well and not. In the next article, we will pro-vide guidelines on structuring a framework for collecting and providing CI to the people in one’s company who need it, and we will examine the legal and ethical considerations of gathering CI.

Competitive intelligence definedCompetitive intelligence is often mistakenly thought of in a narrow way, as a means of gathering ‘secret’ information that can be used to gain advantage over competitors2. We adopt a much broader view of CI, defining it as the analytical process that transforms

disaggregated market and competitor data into relevant strategic knowledge that can be readily put to use by all relevant members of the company. From this broader perspective, CI is closely related to other core manage-ment concepts such as strategic planning, business intelligence, market analysis and knowledge management.

Competitive intelligence is an ongo-ing process that is useful at all levels of an organization (see Box 1). It allows forward-thinking business leaders to clearly define the marketplace, to ask disciplined ques-tions, and to receive timely and reliable answers to them2. It also can help scientists learn about new technologies in the market that could greatly benefit the company by improving discovery platforms or by reduc-ing manufacturing costs. In addition, CI can be used to keep an organization functioning well when key employees leave, by ensuring that they do not take all of their knowledge with them. And it can be used to train new employees so that they more quickly under-stand the firm’s strategy and competitive

Salvador Carlucci and Anthony Page are at HealthIQ, 770 The City Drive South, Suite 7400, Orange, California 92868, USA and David Finegold is at the Keck Graduate Institute of Applied Life Sciences, 535 Watson Drive, Claremont, California 91711, USA.e-mail: [email protected]

Box 1 Potential benefits of competitive intelligence2,4

• Identify opportunities and potential customers• Gain competitive advantage by reducing reaction time• Improve short- and long-term strategic planning• Create company benchmarks and reveal how competitors do business• Provide guidance on pricing, delivery, product development, outsourcing and clinical

research decisions• Anticipate changes in the regulatory and reimbursement environment that may

profoundly affect the firm or its industry• Identify emerging technologies and their potential impact on the competitive

environment• Assess merger and acquisitions candidates, joint-venture, academic and alliance

partners• Provide warning when key strategic assumptions are changing and prevent surprises

marketplace and therefore become fully productive faster3.

For all these reasons, CI done well can enhance a company’s probability of success in the highly risky biotech environment by reducing uncertainty and improving invest-ment decisions, whereas a failure to obtain CI can threaten a firm’s survival (see Box 2).

What types of intelligence are needed?To begin creating a CI system, a company must determine what knowledge it needs to set its strategy and operate its business. The specific types of knowledge needed and the priority placed on them will vary according to the company’s market, but companies gen-erally should have knowledge in at least nine different areas:

1. Intellectual property. Depending on the company’s resources, one should do a comprehensive patent literature search at least once a year. When searching for patents, it is useful to start with the European Patent Office, which publishes

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy


BUILD ING A BUS INESS

all patent applications within 18 months of when they are filed, usually after one year, whereas the USPTO will not publish pat-ents until 18 months after they are filed. This is valuable information, because most countries do not make a patent available to the public until it has been reviewed by the patent office.

2. Market need and size. Identifying target market segments will allow the company to know what markets competitors are plan-ning to move into or are ignoring. During the long development periods required for most biotech products, the market needs

and size will change. It is therefore vital to keep CI up to date by regularly consult-ing with key people, such as members of a carefully chosen Scientific Advisory Board and a broad sample of experts in the rel-evant fields. Networking at professional or industry conferences is a good way to do this.

3. Partnerships. By monitoring new technol-ogies entering the market and in develop-ment, the company can identify possible partnerships with other companies and academic institutions. Scientific journal and patent literature searches along with pro-

fessional conferences can all be potentially fruitful sources of new partners.

4. Competitive environment. It is important to continuously monitor the competi-tion. Some players will drop out, while new, potentially disruptive technologies developed by small firms may enter the market that may not be readily appar-ent as competitors. The company has to be very expansive in thinking about the possible kinds of competitors. By attend-ing conferences and examining relevant ads, the company can assess competitors’ product strategies.

Example 1: Bad. Company A, a small biotech firm, had its strategy consulting firm do a limited CI study of the market for migraine medications to support projections that would be presented to potential investors. To conserve resources, the company focused on immediate competitors with similar compounds. As they were preparing for presentations, it was learned that a competitor product already on the market would be approved for this indication and would offer comparable efficacy at a fraction of the cost. The company’s market had disappeared. Retrospective analysis revealed that this competitive development was the result of a large clinical trial program (with over 6,000 patients) that had been going on for the past two years and that could have been easily detected by a more thorough CI program. The company subsequently abandoned the indication, laid off large numbers of employees, and saw its stock price plunge and its financial support evaporate.

Example 2: Better. Company B commissioned an in-depth CI study while its drug was in phase 2 clinical trials. The study confirmed

that a major pharma company had a compound that was far ahead. This company already had a dominant position in the market, which would be very difficult for Company B to overcome without clearly superior efficacy. When interim phase 2 data showed that this was not the case, the company terminated development and avoiding sinking additional investment into a dead-end indication. The company estimates it saved at least $30 million.

Example 3: Best. Company C directed CI efforts at their competitor’s product development program. Systematic discussions with dozens of researchers on the likely safety and efficacy benefits of products in development revealed that the competitor was including novel safety endpoints in their clinical trial. The company concluded that the competitor intended to use the resulting data to make an enhanced safety claim. The company refined its own protocol to one-up the competitor and was able to make even stronger safety claims that effectively differentiated its product and created a substantial competitive advantage.

Box 2 The benefits of CI and the costs of not doing it

Internal information. Company Y hired a CI provider to determine why they were losing major accounts to a competitor. As a first step, the CI firm reviewed the company’s internal data, including e-mail correspondence from its own sales force reporting rumors they’d heard from customers. This correspondence included references to a sequence of activities that clearly indicated what had happened. Over a four-month period, different company sales reps reported: (i) competitor sales reps visited the customer, (ii) the customer requested information on Company Y’s discounting policy, (iii) the competitor was using “a cost-effectiveness outcomes trial” as a marketing tool, and (iv) the competitor had initiated an “outcomes trial” at the account’s facility. One month later the account switched to the competitor. These pieces were not assembled and analyzed because the company did not have any internal resources dedicated to systematically keeping an eye on competitor activity.

External information. A large pharmaceutical company conducted a CI assessment of a competitor’s product marketing strategy for a synthetically produced compound in development. An analysis of financial information revealed that the competitor had purchased a large volume of agricultural futures in a particular flower. A

review of the company’s foreign marketing materials revealed that their product was being test-marketed in a European country using an “all-natural” direct-to-consumer (DTC) message. Finally, a personality profile of the senior marketing executive showed a history of innovative DTC campaigns. Based on this information, it was determined that the competitor intended to produce the compound using the more expensive natural production method in order to use the “all-natural” marketing position in the US market. The product’s key market positioning strategy was thus identified 12 months before its scheduled US launch.

Creative thinking. While attending a symposium to collect intelligence, a CI analyst heard one of his client’s trial investigators comment during his presentation that a particular treatment pathway (not the client’s) was likely to be an area of “future development.” On returning home, the analyst researched the speaker and found that he had an association with the president of a startup company focused on this pathway. This startup company was funded by a major rival and was working on a next-generation product to leapfrog the company’s own product. Additional research revealed that this researcher also had an undisclosed financial relationship with the competing startup company.

Box 3 Creatively using internal and external information

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy


BUILD ING A BUS INESS

5. Marketing and distribution. By talking with distributors’ and competitors’ sales forces, the company can determine how competi-tors are getting their products to market. This information can help the company develop its own more efficient and targeted strategy for product marketing and distri-bution. The company can, for example, look at how much competitors spend on adver-tising or how big competitors’ sales forces are to create benchmarks for its own goals and performance. Although most people will decline to talk to a “competitor,” many will talk to their “peers” in other companies if the questions are asked in the right way.

6. Technology opportunities and risks. By read-ing the publications of competitors’ scien-tists and their academic partners and talking with them at conferences, the company can identify the bottlenecks that competitors have encountered when developing similar technologies.

7. Regulatory and reimbursement issues. Surveying the regulatory agencies is one way to determine the current regulatory requirements and identify new issues that might affect the approval of a product or the way it is labeled and marketed. With a CI process one can examine the various factors in the regulatory environment and antici-pate changes that may profoundly affect the enterprise.

8. Financing options. One of the most vital tasks for the leader of any startup is ensuring

the resources that the firm needs to operate are available. CI can help determine which venture capitalists are investing in the firm’s technology area and what organizations might be interested in acquiring those tech-nologies. This data can better establish the value of a company during financing and can potentially strengthen a firm’s negoti-ating position. In addition, CI can be used in examining merger and acquisition can-didates, government grants and joint-ven-ture partners that could provide alternative sources of funding, thereby increasing the firm’s negotiation leverage.

9. Human capital. Salary surveys and analyses of job ads can provide important insights into competitors’ staffing strategies. Likewise, recruiting agencies, while keeping their client information confidential, may be good sources for industry skill trends and the strategies of non-client firms. This kind of information can allow the company to determine the type of people it needs to succeed in a market niche and what it will take to attract and retain them.

ConclusionCompetitive intelligence is a vital part of cre-ating a sound business strategy, and obtaining it effectively can bring multiple benefits to an organization. But small companies often do not have the expertise or systems in place to get the full value from CI. They must carefully identify their top CI priorities and the resources required to meet these specific intelligence needs, then track and control the allocation of

resources to achieve these tasks. As the com-pany grows, the CI function should also grow to maintain its competitive advantage and exploit the opportunities that CI provides.

Although broad organizational involvement is important for gaining the full value of CI, there should always be one person accountable for clearly establishing and communicating the CI development objectives. This person should be responsible for ensuring the CI pro-cess is gathering the necessary information and then distributing it to the right people. Those employees directly involved in CI data-gath-ering and analysis need good industry and technical knowledge, especially at the tactical level, and a solid understanding of second-ary research. In addition, CI staff should have strong interpersonal, problem-solving, written and oral communication skills, because they will be collating information from both internal and external sources (see Box 3). They should also have a thorough knowledge of ethical and legal implications of their activities, which will be discussed in the next article in this series.

This story was reprinted with some modification from the Building a Business section of the Bioentrepreneur web portal (http://www.nature.com/bioent), 21 March 2005, doi:10.1038/bioent850.

1. Hodgson, J. The headache of knowledge management. Nat. Biotechnol. 19, BE44–BE46 (2001).

2. Nolan J. Confidential (HarperBusiness, New York, 1999).

3. Stewart, T.A. Intellectual Capital: The New Wealth Of Organizations (Doubleday/Currency, New York, 1997).

4. Ashton, W.B. & Klavans, R.A. Keeping Abreast of Science and Technology: Technical Intelligence for Businesses (Battelle Press, Columbus, Ohio, USA, 1997).©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


The origins of new drugs

To the editor:There is some debate as to the relative contribution of publicly funded research (universities, government research institutes and academic medical centers), biotech companies and pharmaceutical companies to the discovery of new medicines. To gain a clearer understanding of the origin of newly marketed drugs, I have analyzed data from the US Food and Drug Administration (FDA, Rockville, MD, USA), US Securities and Exchange Commission (SEC, Washington, DC) and the US Patent and Trademark Office (PTO, Washington, DC) to determine the origin of most of the new molecular entities (NMEs) and new biological entities (NBEs) approved by the FDA from 1998 to 2003.

To carry out this analysis, I obtained lists of NMEs and NBEs approved each year from 1998 to 2003 from the FDA website (http://www.fda.gov/), which provided each drug’s sponsor (that is, the company seeking drug approval that usually owns the drug or holds an exclusive license to the patents covering the drug). In the case of NMEs, the sponsor must identify the patents (if any) describing

the chemical compounds that constitute the NMEs (if such compounds are patentable), methods of NME manufacture or uses of the NME. I excluded from the analysis nine NMEs that are imaging agents and one chemical warfare protective paste developed by the US Army. I found patents covering all the other NMEs (some expired but still relevant as to origin) except Vioxx (rofecoxib; 1999), which Merck (Rahway, NJ) has recently withdrawn from the market, and nine other NMEs for which the FDA Orange Book states “no unexpired patents”. (SEC documents showed that one of these nine, Valstar (valrubicin; 1998, originated in Dana Farber.) In addition, a few NMEs only have recently filed use or method-of-delivery patents that do not provide clues as to origin. Nevertheless, the patent records combined with SEC documents and occasional internet searches give a fairly good picture of the main loci of early stage and preclinical development in the case of all but 14 of the total 145 NMEs. In the case of NBEs, I reviewed Recombinant Capital’s Signals Magazine (http://www.signalsmag.com), which periodically publishes analyses of

licensing data from its rDNA database (http://www.recap.com/rdna.nsf). I also reviewed 10-K reports filed annually to the SEC by the companies that sought FDA approval for the NBEs. Small and mid-sized biotech companies often mention the existence of in-licenses covering their NBEs that have just received FDA approval, although pharmaceutical companies and large biotechs rarely mention such in-licenses. It is possible that I have not identified the principle origin of some of the NBEs submitted for approval by pharmaceutical companies and large biotechs. The results of the analysis are summarized in Table 1.

The data reveal that at least 39% of all (171) drugs (both NMEs and NBEs) approved by the FDA from 1998 to 2003 originated from outside pharmaceutical companies: ~24% came from biotech companies and at least 15% came from public research. Of the drugs that originated from public research, 19% were licensed to pharmaceutical companies and 81% were licensed to biotech companies. In cases when a public research institution’s patents had expired, the drug

Table 1 The origin of FDA-approved medicinesCategory Year(s) approved by FDA

1998 1999 2000 2001 2002 2003 1998–2003

FDA drug approvals

Total 34 34 28 26 22 27 171

No. originating from biotech R&D 14 11 9 8 7 13 62

No. based on university invention 4 8 4 3 2 5 26

University inventions licenseddirectly to pharma company

1 2 2 0 0 0 5

New molecular entities (NMEs)

Total 29 33 26 21 15 21 145



University inventions licensed directly to pharma company

1 2 2 0 0 0 5

New biological entities (NBEs)

Total 5 1 2 5 7 6 26



C O R R E S P O N D E N C E©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


was simply developed by a pharmaceutical or biotechnology company. Thus, biotech companies either discovered or played a major role in developing 36% of all new drugs (NMEs and NBEs).

As expected, biotech companies have dominated the development of NBEs, discovering or playing a key role in the development of 22 (85%) of the 26 new NBEs. Of the 22 biotech NTBs, six (27%) were licensed to pharmaceutical companies, which then applied for FDA marketing approval. Biotech companies themselves applied for marketing approval for the remaining 16. In 4 of these 16 cases, the biotech that applied for marketing approval had in-licensed the NBE from another biotech. At least 6 (27%) of the 22 biotech-developed NBEs were based upon inventions made in public research. There appear to be no cases of a university directly licensing an invention covering an NBE to a pharmaceutical company.

Biotech companies and public research also contributed to a significant but lesser degree to the discovery of new NMEs, at least 45 (31%) of which were discovered outside of pharmaceutical companies. Forty (27%) were discovered or developed in biotech companies, and in most of these cases, a biotech company pursued development all the way to obtaining marketing approval. In the case of 30 of the 40 biotech company-developed NMEs, the biotech company was also the applicant for FDA marketing approval. Nine of these 30 were licensed from one biotech company to another, which subsequently assumed responsibility for obtaining FDA approval. Twenty (14%) of the NMEs are covered by university patents. Five of the drugs of university origin were licensed directly to pharmaceutical companies rather than to biotechs. Fifteen (38%) of the 40 NMEs developed by biotech companies originated in public research institutions.

Others have described the importance of linkages between universities, biotech companies and pharmaceutical companies for the discovery and development of new drugs1–6. The analysis described here provides an objective estimate of the contribution in drug discovery not only of biotech companies but also of public research (to the extent that university involvement is reflected in patents covering the new drugs).

In the case of NBEs, the data indicating the contribution of public research or biotech companies to drug discovery are lower-bound estimates because the FDA does not publish information about the patents covering NBEs. Thus, it is difficult to know whether a pharmaceutical company that has received

permission to market an NBE might have in-licensed the NBE from a biotech company or public research institution. It is also difficult to know whether a biotech company that has received marketing approval for an NBE might have in-licensed key discoveries from a public research institution, although the SEC filings often provide this information.

In addition, patents reflect only a portion of the total contribution to drug discovery and development. Cockburn1 has shown that even before university patenting of biomedical discoveries became commonplace, the vast majority of the most therapeutically important drugs approved in the 1960s and 1970s owed their discovery in large part to public research. On the other hand, even though patented discoveries in a university or biotech laboratory may have been important in the discovery or development of a new drug, subsequent R&D in the pharmaceutical or biotech company that ultimately applies for approval also reflects considerable scientific and innovative effort. Thus, these findings do not suggest a diminished contribution of pharmaceutical companies but rather confirm the integrated nature of drug discovery and development and the substantial contributions of biotechnology companies and universities.

Compared with Cockburn’s earlier analysis, the data presented here also suggest that a larger proportion of university discoveries directly relevant to drug discovery are now being transferred as formal patent licenses to new small companies. These formal (and presumably exclusive) licenses undoubtedly

help biotech companies to obtain private investment and thereby continue drug development. These findings also indicate that biotech companies which are the original discoverers of drugs ultimately approved (whether NMEs or NBEs) more often than not pursue development of these drugs all the way through approval. One interpretation of this finding is that, when a biotech company discovers a drug that turns out to be a winner, it usually manages to obtain resources to pursue development all the way to marketing approval (that is, biotech companies and their investors do a pretty good job of picking and holding onto winners). However, size does matter. Small biotechs are more likely to out-license their winning drugs than large biotechs. Finally, although I show data for each year, clear time trends are not apparent.

Robert Kneller

University of Tokyo, RCAST, 4-6-1 Komaba, Meguro-ku, Tokyo 153-8904, Japan. e-mail: [email protected]

1. Cockburn, I. & Henderson, R. Public-Private Interaction and the Productivity of Pharmaceutical Research. National Bureau of Economic Research (NBER) Working Paper 6018 (NBER, Cambridge, MA, 1997).

2. Powell, W.W., Korput, K.W. & Smith-Doerr, L. Administr. Sci. Quart. 41, 116–145 (1996).

3. Murray, F. Res. Policy 31, 1389–1403 (2002).4. Henderson, R., Orsenigo, L. & Pisano, G.P. in Sources

of Industrial Leadership, Studies of Seven Industries (eds. Mowery, D. & Nelson, R.) 267–311 (Cambridge University Press, Cambridge, UK, 1999).

5. McKelvey, M. Evolutionary Innovations, the Business of Biotechnology (Oxford University Press, Oxford, 1996).

6. Zucker, L.G. & Darby, M.R. Proc. Natl. Acad. Sci. USA 93, 12709–12716 (1996).

Framing the issues ontransgenic forestsTo the editor:Your News Feature in the February issue (Nat. Biotechnol. 23, 165–167, 2005) highlighted rapid advances being made in forest molecu-lar domestication. Counter to Herrera’s asser-tion that “most of the global funding for forest biotech is being funneled to universities,” the pursuit of genetic engineering in forest research is principally corporate, shaped by the impera-tives of private investment, market forces and government regulatory institutions. Novel for-est tree phenotypes are thus created as a means to increase shareholder value of investor com-panies. And although potential benefits will accrue to shareholders, it is clear that ecological risks of certain transgenic traits engineered into trees are likely to be shared by all. Indeed, as the

forest-products companies driving adoption of transgenic technology hold less than 11% of US forest acreage, it is the remaining majority—public landowners and private small woodlot owners—that stands to lose the most.

Herrera indicates in his article that for forest biotech, “investors are virtually nonexistent.” Even so, private investment in forest biotechnology is still sufficient to be fueling the creation of novel transgenic phenotypes in trees at a rate that is outstripping public policy deliberation and scientific assessment of environmental concerns specific to trees. For example, trees disperse their seed and pollen over unprecedented distances compared with crops. The sheer scale of gene flow dynamics

CORRESPONDENCE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


was simply developed by a pharmaceutical or biotechnology company. Thus, biotech companies either discovered or played a major role in developing 36% of all new drugs (NMEs and NBEs).

As expected, biotech companies have dominated the development of NBEs, discovering or playing a key role in the development of 22 (85%) of the 26 new NBEs. Of the 22 biotech NTBs, six (27%) were licensed to pharmaceutical companies, which then applied for FDA marketing approval. Biotech companies themselves applied for marketing approval for the remaining 16. In 4 of these 16 cases, the biotech that applied for marketing approval had in-licensed the NBE from another biotech. At least 6 (27%) of the 22 biotech-developed NBEs were based upon inventions made in public research. There appear to be no cases of a university directly licensing an invention covering an NBE to a pharmaceutical company.

Biotech companies and public research also contributed to a significant but lesser degree to the discovery of new NMEs, at least 45 (31%) of which were discovered outside of pharmaceutical companies. Forty (27%) were discovered or developed in biotech companies, and in most of these cases, a biotech company pursued development all the way to obtaining marketing approval. In the case of 30 of the 40 biotech company-developed NMEs, the biotech company was also the applicant for FDA marketing approval. Nine of these 30 were licensed from one biotech company to another, which subsequently assumed responsibility for obtaining FDA approval. Twenty (14%) of the NMEs are covered by university patents. Five of the drugs of university origin were licensed directly to pharmaceutical companies rather than to biotechs. Fifteen (38%) of the 40 NMEs developed by biotech companies originated in public research institutions.

Others have described the importance of linkages between universities, biotech companies and pharmaceutical companies for the discovery and development of new drugs1–6. The analysis described here provides an objective estimate of the contribution in drug discovery not only of biotech companies but also of public research (to the extent that university involvement is reflected in patents covering the new drugs).

In the case of NBEs, the data indicating the contribution of public research or biotech companies to drug discovery are lower-bound estimates because the FDA does not publish information about the patents covering NBEs. Thus, it is difficult to know whether a pharmaceutical company that has received

permission to market an NBE might have in-licensed the NBE from a biotech company or public research institution. It is also difficult to know whether a biotech company that has received marketing approval for an NBE might have in-licensed key discoveries from a public research institution, although the SEC filings often provide this information.

In addition, patents reflect only a portion of the total contribution to drug discovery and development. Cockburn1 has shown that even before university patenting of biomedical discoveries became commonplace, the vast majority of the most therapeutically important drugs approved in the 1960s and 1970s owed their discovery in large part to public research. On the other hand, even though patented discoveries in a university or biotech laboratory may have been important in the discovery or development of a new drug, subsequent R&D in the pharmaceutical or biotech company that ultimately applies for approval also reflects considerable scientific and innovative effort. Thus, these findings do not suggest a diminished contribution of pharmaceutical companies but rather confirm the integrated nature of drug discovery and development and the substantial contributions of biotechnology companies and universities.

Compared with Cockburn’s earlier analysis, the data presented here also suggest that a larger proportion of university discoveries directly relevant to drug discovery are now being transferred as formal patent licenses to new small companies. These formal (and presumably exclusive) licenses undoubtedly

help biotech companies to obtain private investment and thereby continue drug development. These findings also indicate that biotech companies which are the original discoverers of drugs ultimately approved (whether NMEs or NBEs) more often than not pursue development of these drugs all the way through approval. One interpretation of this finding is that, when a biotech company discovers a drug that turns out to be a winner, it usually manages to obtain resources to pursue development all the way to marketing approval (that is, biotech companies and their investors do a pretty good job of picking and holding onto winners). However, size does matter. Small biotechs are more likely to out-license their winning drugs than large biotechs. Finally, although I show data for each year, clear time trends are not apparent.

Robert Kneller

University of Tokyo, RCAST, 4-6-1 Komaba, Meguro-ku, Tokyo 153-8904, Japan. e-mail: [email protected]

1. Cockburn, I. & Henderson, R. Public-Private Interaction and the Productivity of Pharmaceutical Research. National Bureau of Economic Research (NBER) Working Paper 6018 (NBER, Cambridge, MA, 1997).

2. Powell, W.W., Korput, K.W. & Smith-Doerr, L. Administr. Sci. Quart. 41, 116–145 (1996).

3. Murray, F. Res. Policy 31, 1389–1403 (2002).4. Henderson, R., Orsenigo, L. & Pisano, G.P. in Sources

of Industrial Leadership, Studies of Seven Industries (eds. Mowery, D. & Nelson, R.) 267–311 (Cambridge University Press, Cambridge, UK, 1999).

5. McKelvey, M. Evolutionary Innovations, the Business of Biotechnology (Oxford University Press, Oxford, 1996).

6. Zucker, L.G. & Darby, M.R. Proc. Natl. Acad. Sci. USA 93, 12709–12716 (1996).

Framing the issues ontransgenic forestsTo the editor:Your News Feature in the February issue (Nat. Biotechnol. 23, 165–167, 2005) highlighted rapid advances being made in forest molecu-lar domestication. Counter to Herrera’s asser-tion that “most of the global funding for forest biotech is being funneled to universities,” the pursuit of genetic engineering in forest research is principally corporate, shaped by the impera-tives of private investment, market forces and government regulatory institutions. Novel for-est tree phenotypes are thus created as a means to increase shareholder value of investor com-panies. And although potential benefits will accrue to shareholders, it is clear that ecological risks of certain transgenic traits engineered into trees are likely to be shared by all. Indeed, as the

forest-products companies driving adoption of transgenic technology hold less than 11% of US forest acreage, it is the remaining majority—public landowners and private small woodlot owners—that stands to lose the most.

Herrera indicates in his article that for forest biotech, “investors are virtually nonexistent.” Even so, private investment in forest biotechnology is still sufficient to be fueling the creation of novel transgenic phenotypes in trees at a rate that is outstripping public policy deliberation and scientific assessment of environmental concerns specific to trees. For example, trees disperse their seed and pollen over unprecedented distances compared with crops. The sheer scale of gene flow dynamics

CORRESPONDENCE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


for trees presents a daunting challenge in assessing the environmental impact of a transgenic trait (Fig. 1).

Second, trees produce an abundance of seed and pollen for many years before they are ready for timber harvesting. Thus, in contrast to seasonally harvested crops, pollen and seeds from trees disperse without hindrance into their surroundings for many years. As seed and pollen production increase with the age and height of a tree, each year more seed and pollen travel progressively farther by a process known as long-distance dispersal.

And third, most commercially cultivated tree species have many wild relatives that grow in similar locations; thus there is a high potential for mating. In contrast, in the US at least, such crops as corn, cotton and soybeans have no wild or weedy relatives in the vicinity, making gene spread from transgenic varieties more unlikely.

It is instructive to discuss these concerns in the context of loblolly pine (Pinus taeda), a tree indigenous to the southeastern United States and our major timber commodity. Pinus taeda grows as natural or plantation forests on nearly 58 million acres in the American South, providing 16% of the world’s annual timber supply. Annual planting demand is roughly one billion seedlings per year. Harvest age for P. taeda is between 25 and 35 years, so standing timber may be bought and sold several times before harvesting.

Pinus taeda and other pines have been domesticated since the mid-20th century—relatively recent compared with food crops. Thus, it is likely that a conifer expressing a transgenic trait would thrive without human intervention after escape into an unmanaged ecosystem. Because traditional breeding is managed at a population level to conserve genetic diversity, neither inbred lines nor even breed structure exists for domesticated P. taeda trees. The cost and effort of traditional breeding has been borne by the private sector with the notable exception of a few state agencies. No gene conservation program has been formalized by the federal government for P. taeda.

Certain biotechnology firms—Arborgen in Summerville, South Carolina, and CellFor in Vancouver, Canada—cited by Herrara in his article now offer clonal P. taeda trees to timber companies via somatic embryogenesis (the culture of undifferentiated cells from immature embryos to yield unlimited quantities of a single genotype). The availability of somatic embryogenesis for P. taeda makes genetic engineering of this species feasible on a commercial scale for the first time.

As yet, no extensive analysis of the environmental impact of a P. taeda transgene has been undertaken. What is clear is that these trees would outcross and produce abundant windborne pollen and seeds each year. Consider that a fraction of seeds uplifted above the forest canopy will move by the long-distance dispersal process as far as 11.9 to 33.7 km. Out of 105 seeds produced per ha–1 yr–1 in a 16-year old plantation, roughly 70 seeds ha–1 will reach distances in excess of 1 km from the source, a distance too great to serve as a biocontainment zone. Pollen dispersal distances are even greater. The probability of long-distance dispersal of transgenic conifer seeds and pollen at distances exceeding 1 km approaches 100%. Although 99.9% of P. taeda seeds and pollen fall near the source tree, via a dispersal process known as local neighborhood diffusion, it is the remaining 0.01% that pose the greatest ecological concerns. Long-distance dispersal provides the biological mechanism for establishment of remote satellite colonies from transgenic P. taeda seeds and pollen, even though it is not the most common process of dispersal.

To date, the benefits of specific transgenic traits in P. taeda have not been fully gauged because technology innovation is recent and transgenic wood products have not reached timber harvest age. But what will happen as these tests get older? At present, transgenic P. taeda test stands must be cut down at onset

of reproduction, whereas the species reaches peak merchantable value only after the age of 25 years. This constitutes a regulatory impasse for collecting data on benefits, especially given prospects of transient expression or ‘gene-silencing’ through harvest age.

Risk analysis is similarly incomplete. Mathematical models suggest that movement of escaped transgenic seed and pollen on the scale of kilometers from the source is a certainty. Movement of transgenic pollen and seeds is problematic only if there is potential harm associated with a specific transgenic trait, but potential harm has not been tested. To be harmful, a tree must express a transgenic trait that exhibits enhanced invasiveness properties compared with a wild type. Increased invasiveness is harmful if it translates into displacement of local endemic species or even long-term forest maladaptation. No experimental evidence, pro or con, yet exists to show whether specific transgenic traits in the context of forests are harmful. The take-home message is that no experimental results for either benefits or risk associated with transgenic P. taeda are available.

Commercial exploitation of transgenic trees, particularly indigenous coniferP. taeda, is technically imminent; putting this into practice will, however, be stymied by concerns over the environmental impact of gene flow and the unique pattern of ownership of forest lands in the United States.

≤0.1 km ≤10 km ≤1000 km

• Experiments• Observations• Paleo-evidence of past invasions• Deterministic and stochastic modeling

Flux of DNA ParcelsCanopy air flowCanopy heightParticle propertiesAmount of pollen and seed

PollinationTopographyPollen viabilityOpportunity for pollination

Figure 1 Gene flow from transgenic conifers is more complex than gene flow from annual row crops. Source: R. Oren, Duke University

CORRESPONDENCE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Lost in the woods

The certainty of gene flow from transgenic forests is problematic because neighboring lands are often less intensively managed public and private forest lands. At present, the scale and staggering expense of regulatory oversight alone could drive the political outcome in the absence of risk-benefit analyses. Ecological consequences of investment decisions on private lands deserve closer scrutiny at a national level.

Calls for public deliberation are coming late in the life of the forest product life cycle. I advocate that transgenic conifers be considered separately from agricultural biosafety policy due to the sheer scale and complexity of forest tree gene flow. Biocontainment zones suited to transgenic food crops cannot deter escape of seeds or pollen from transgenic P. taeda. Reproductive sterility research for conifers, a complex problem, remains in its infancy and has not received serious consideration as a national research priority.

There is thus an urgent need for policy makers to move on two fronts. First, a gene conservation program should be formalized through the National Forest System. In Region 8 of the southern United States, for example, indigenous P. taeda forests need to be protected from the potential impact of transgenic varieties. Widespread use of clonal forests with or without genetic engineering will likely rapidly narrow the numbers of P. taeda genotypes, opening the question of

how to protect undomesticated germ plasm and close relatives, which remain largely undomesticated.

Second, forestry-specific research programs that address key issues specific to the implementation of transgenic technology in forestry need to be promoted within the existing cadre of national competitive funding programs. We are in dire need of funding for research to gauge the environmental impact of gene flow from trees. At present, we remain ignorant on numerous aspects of tree biology and ecology that affect whether or not we should proceed. Can pine pollen move in the jet stream and, if so, will it remain viable? How does gene flow from transgenic P. taeda affect indigenous pine forests or small woodlot or public forest ownership patterns?

A singular priority for forest research is determining the scale of regulatory oversight for transgenic forest trees. Responsible biotechnology governance is indeed questionable for transgenic conifer plantations located within less intensively managed forest ecosystems in the American South. The genetic composition of our nation’s indigenous forests is at issue.

Claire G. Williams

Claire G. Williams is at Duke University, Duke University, Department of Biology, Biological Sciences Building, Box 90338, Durham, North Carolina 27708, USA. e-mail: [email protected]

To the editor:In “Struggling to see the for-est through the trees” (Nat. Biotechnol. 23, 165–167, 2005), Herrera cites many of the important issues sur-rounding the state of forest biotechnology, yet at the same time fails to give an accurate impression of the extremely difficult state of the industry worldwide.

First, there are serious technical problems that stand in the way of this industry maturing. Although it is abundantly clear that simple traits like herbicide resistance and insect resistance, when encoded by single genes as in transgenic agricultural crops, can provide major benefits in some species

and geographies with responsible use1, it is not clear that these traits are valuable enough in forestry, given the costs of transformation, integration into breeding programs and associated field testing. For transformation, this is partly a result of the expected need to use new markers in place of antibiotic resistance genes to get broad international

regulatory approvals2, even though the commercially authorized (USA) nptII gene for kanamycin resistance used in transgenic agricultural crops has never been shown to be a significant health or environmental risk. In addition, transformation methods

must be robust enough to work in the high diversity of germplasm used in most industrial forestry programs—which can include several species and dozens of genotypes. We know of no transformation systems up to this task.

Were there to be a number of companies and/or public sector institutions seriously investing in technological solutions to these problems, we are certain they could be solved. But the reality, in contrast to the impression Herrera gave, is that there is a very low level of industrial activity worldwide. Of the companies listed in Table 1 of his article, only Arborgen in Summerville, South Carolina, is seriously pursuing transgenic breeding science. CellFor in Vancouver, Canada, has ended all transgenic and molecular biology research; SweTree of Umeå, Sweden, works primarily on basic genomics and has never had an applied breeding-related program, and the transgenic breeding research programs in Chile and New Zealand have all been dramatically cut back in recent years. Large, technologically advanced companies like Weyerhaeuser, Federal Way, Washington, have never had their own transgenic research, though they have supported some basic transgenic-related studies in universities, primarily for biosafety and wood quality. Most of the major forestry companies in Chile are effectively turning away from transgenic research because of concerns about activist boycotts and their European markets. Finally, with the high regulatory risks (discussed below), few forestry breeding programs would wish to encumber their efficient programs with transgenic-level regulatory costs and potential liabilities.

Second, and most important, the thorny regulatory environment, designed without regard to the years of scientific consensus from national academies and ecological societies (e.g., see a position paper from the Ecological Society of America3), treats genetic engineering itself as dangerous by choosing to regulate every transgenic product in virtually the same way (the so-called ‘case-by-case’ approach). This extreme ‘precautionary’ system effectively precludes the use of trial-and-error, empirical methods that characterize all tree breeding programs. It is hard to imagine that changes to growth, wood chemistry or structure that are of significant economic benefit, but that do not also impair tree physiology and adaptation so important in all perennial crops, can be identified mainly in glasshouses and laboratories.

CORRESPONDENCE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Lost in the woods

The certainty of gene flow from transgenic forests is problematic because neighboring lands are often less intensively managed public and private forest lands. At present, the scale and staggering expense of regulatory oversight alone could drive the political outcome in the absence of risk-benefit analyses. Ecological consequences of investment decisions on private lands deserve closer scrutiny at a national level.

Calls for public deliberation are coming late in the life of the forest product life cycle. I advocate that transgenic conifers be considered separately from agricultural biosafety policy due to the sheer scale and complexity of forest tree gene flow. Biocontainment zones suited to transgenic food crops cannot deter escape of seeds or pollen from transgenic P. taeda. Reproductive sterility research for conifers, a complex problem, remains in its infancy and has not received serious consideration as a national research priority.

There is thus an urgent need for policy makers to move on two fronts. First, a gene conservation program should be formalized through the National Forest System. In Region 8 of the southern United States, for example, indigenous P. taeda forests need to be protected from the potential impact of transgenic varieties. Widespread use of clonal forests with or without genetic engineering will likely rapidly narrow the numbers of P. taeda genotypes, opening the question of

how to protect undomesticated germ plasm and close relatives, which remain largely undomesticated.

Second, forestry-specific research programs that address key issues specific to the implementation of transgenic technology in forestry need to be promoted within the existing cadre of national competitive funding programs. We are in dire need of funding for research to gauge the environmental impact of gene flow from trees. At present, we remain ignorant on numerous aspects of tree biology and ecology that affect whether or not we should proceed. Can pine pollen move in the jet stream and, if so, will it remain viable? How does gene flow from transgenic P. taeda affect indigenous pine forests or small woodlot or public forest ownership patterns?

A singular priority for forest research is determining the scale of regulatory oversight for transgenic forest trees. Responsible biotechnology governance is indeed questionable for transgenic conifer plantations located within less intensively managed forest ecosystems in the American South. The genetic composition of our nation’s indigenous forests is at issue.

Claire G. Williams

Claire G. Williams is at Duke University, Duke University, Department of Biology, Biological Sciences Building, Box 90338, Durham, North Carolina 27708, USA. e-mail: [email protected]

To the editor:In “Struggling to see the for-est through the trees” (Nat. Biotechnol. 23, 165–167, 2005), Herrera cites many of the important issues sur-rounding the state of forest biotechnology, yet at the same time fails to give an accurate impression of the extremely difficult state of the industry worldwide.

First, there are serious technical problems that stand in the way of this industry maturing. Although it is abundantly clear that simple traits like herbicide resistance and insect resistance, when encoded by single genes as in transgenic agricultural crops, can provide major benefits in some species

and geographies with responsible use1, it is not clear that these traits are valuable enough in forestry, given the costs of transformation, integration into breeding programs and associated field testing. For transformation, this is partly a result of the expected need to use new markers in place of antibiotic resistance genes to get broad international

regulatory approvals2, even though the commercially authorized (USA) nptII gene for kanamycin resistance used in transgenic agricultural crops has never been shown to be a significant health or environmental risk. In addition, transformation methods

must be robust enough to work in the high diversity of germplasm used in most industrial forestry programs—which can include several species and dozens of genotypes. We know of no transformation systems up to this task.

Were there to be a number of companies and/or public sector institutions seriously investing in technological solutions to these problems, we are certain they could be solved. But the reality, in contrast to the impression Herrera gave, is that there is a very low level of industrial activity worldwide. Of the companies listed in Table 1 of his article, only Arborgen in Summerville, South Carolina, is seriously pursuing transgenic breeding science. CellFor in Vancouver, Canada, has ended all transgenic and molecular biology research; SweTree of Umeå, Sweden, works primarily on basic genomics and has never had an applied breeding-related program, and the transgenic breeding research programs in Chile and New Zealand have all been dramatically cut back in recent years. Large, technologically advanced companies like Weyerhaeuser, Federal Way, Washington, have never had their own transgenic research, though they have supported some basic transgenic-related studies in universities, primarily for biosafety and wood quality. Most of the major forestry companies in Chile are effectively turning away from transgenic research because of concerns about activist boycotts and their European markets. Finally, with the high regulatory risks (discussed below), few forestry breeding programs would wish to encumber their efficient programs with transgenic-level regulatory costs and potential liabilities.

Second, and most important, the thorny regulatory environment, designed without regard to the years of scientific consensus from national academies and ecological societies (e.g., see a position paper from the Ecological Society of America3), treats genetic engineering itself as dangerous by choosing to regulate every transgenic product in virtually the same way (the so-called ‘case-by-case’ approach). This extreme ‘precautionary’ system effectively precludes the use of trial-and-error, empirical methods that characterize all tree breeding programs. It is hard to imagine that changes to growth, wood chemistry or structure that are of significant economic benefit, but that do not also impair tree physiology and adaptation so important in all perennial crops, can be identified mainly in glasshouses and laboratories.

CORRESPONDENCE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Yet, costly requirements for containment of pollen and seed from trees of commercially relevant sizes, when grown in a representative diversity of environments, make such essential adaptive research virtually impossible to carry out. This is in spite of highly promising small-scale field results from Europe and elsewhere4, started in the optimistic 1990s. Finally, vandalism has led to local decisions in places such as British Columbia, Canada, to ban all transgenic field research with forest tree species, despite any scientific rationale to do so.

Of course, as Herrera hints, such draconian regulations are in place owing largely to the scare tactics and pressure on government officials from anti-genetically modified organism (GMO) activist organizations, which hope to see all transgenic trees regulated based on imagined worst case scenarios—not based on the increasing interest in modified expression of functionally native genes and pathways enabled by tree genomics. These regulations also ignore the reality that conventional breeding and silviculture, not just genetic engineering, also bring about substantial changes in wood structure, lignin, flowering, growth rate and many other attributes. Yet there is little call for their stringent regulation. It is time that the absurd, anti-scientific (that is, process not product) claims that all Agrobacterium tumefaciens or biolistics-delivered genes are somehow capable of causing ‘destruction and contamination’ of wild forests be identified as the scare-mongering that it is. Instead, lawyers and bureaucrats who have a limited understanding of breeding science or practice are working to insert language into local and national regulations, and into international treaties5, whose effect will be to completely or effectively (due to cost and liability risk) ban all genetic engineering from forestry and agriculture.

Finally, these same groups, primarily by threat of boycott of retailers and corporations, rather than on advice from the leading scientific societies, continue to pressure companies for adoption of ‘green’ certification programs, such as the Forest Stewardship Council’s (FSC), that ban all field use of transgenic trees, even for contained research. For FSC, any use of transgenic trees is considered a major violation of their ‘principles,’ even where

it involves completely contained field research and is intended to solve a major environmental problem (e.g., to reduce chemical use during pulping, increase the rate of bioremediation or reduce the risk of invasiveness of forest trees when they are exotics6). As these programs slowly proliferate under the myth that avoidance of all genetic engineering is somehow an environmental good, companies’ willingness to engage in transgenic research understandably dissipates.

These unwieldy social problems (for a review, see ref. 7), combined with the growing anti-commons caused by the fragmented patent estate of technologies important to forest biotechnology, make it a place where most companies understandably fear to tread. It will take strong political leaders and highly engaged scientists empowered by public funds for outreach, to stand-up and prevent green fundamentalist religion from trumping what could be a highly green new tool for breeding practice. Instead of genetic engineering helping to produce more efficient forms of plantation forestry that generate cost-efficient renewable energy and biobased products, we are instead being forced to continue planting more tree farms and harvesting more wild trees than necessary. How green is that?

Sofia Valenzuela

Forest Science Faculty, Universidad de Concepción, Concepción, Chile. e-mail: [email protected]

Steven H. Strauss

Forest Science, Oregon State University, Corvallis, OR 97331-5752, USA.e-mail: [email protected]

1. Sedjo, R.A. in The BioEngineered Forest: Challenges to Science and Society, (eds. Strauss, S.H. & Bradshaw, H.D.) 23–35 (Resources for the Future, Washington, DC, 2004).

2. König, A.A Nat. Biotechnol. 21, 1274–1279 (2003).3. Snow, A.A. et al. Ecological Society of America Position

Paper. Genetically Engineered Organisms and the Environment: Current Status and Recommendations (ESA, Washington, DC, 2004). http://www.esa.org/pao/esaPositions/Papers/geo_position.htm

4. Pilate, G. et al. Nat. Biotechnol. 20, 607–612 (2002).

5. DeGreef, W. Nat. Biotechnol. 222, 811–812 (2004).

6. Strauss, S.H. et al. Int. Forestry Rev. 3, 85–102 (2001).

7. Strauss, S.H. & Bradshaw, H.D. (eds.) The BioEngineered Forest: Challenges to Science and Society (Resources for the Future, Washington, DC, 2004).


CORRESPONDENCE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


The Human Cancer Genome Project—one more misstep in the war on cancerGeorge L Gabor Miklos

Strap yourself in and get ready for some seri-ous ‘more of the same.’ A recent proposal to sequence cancer genomes holds out the pro-mise of personalized cures for each of 50 dif-ferent cancers. The cost? A mere $12 billion at today’s prices1. This human cancer genome megaproject is the equivalent of 12,500 human genome projects and already has the backing of several prominent scientists. Harold Varmus believes that the project could “completely change how we view cancer”1; Eric Lander argues that “knowing the defects of the cancer cell points you to the Achilles’ heel of tumors”1; and Francis Collins predicts that he “can con-fidently tell you that something will happen here”1. More pragmatically, Craig Venter points out that “...it’s not clear what answer we’d get; there might be better ways to move cancer research forward”1.

In a nutshell, the megaproject aims to cata-log all somatic mutations from primary tumors as the basis for designer drugs to cure most cancers. Success is predicated on the assump-tion that drugs can be targeted to very specific mutated regions of gene products. However, most patients with a localized primary tumor are cured by surgery and local radiation. It is not the primary tumor, but the metastatic spread of a small population of deadly cells that ultimately compromises a normal tissue or organ, that kills in cancer2 (for an excel-lent popular account, see ref. 3). Are primary tumors therefore the appropriate focus for such a massive project and are their bulk mutational spectra therapeutically useful?

The clinical track recordCancer research has consumed hundreds of billions of dollars to date3 and yet the main killers—breast, prostate, lung and colorectal cancer—are essentially as deadly as ever4 (see Fig. 1). Despite the glacial progress in treatment and the advent of ‘molecularly targeted’ therapy, cancer research continues to focus myopically on individual oncogenes, tumor suppressors and repair genes5, with little effort devoted to alternative mechanisms and targets6–8.

Although conventional chemotherapeu-tic agents remain the first-line treatment of choice, newer molecularly targeted therapies are now reaching the market. Thus far, how-ever, these therapies have had very limited success against solid tumors, which after all make up 90% of all cancers. Success has been largely restricted to rare leukemias; for example, imatinib mesylate (Gleevec; Novartis, Basel) has initially proven effec-tive in patients with chronic myelogenous leukemia (CML).

Whereas the initial clinical success of ima-tinib in CML was spectacular, this has not been repeated in most of the succeeding can-cer therapies against solid tumors. Gefitinib (Iressa; AstraZeneca, London)-based treatment shrinks tumors in only about 10% of advanced non-small cell lung cancer patients9. A recent study in one very small (35) patient group indi-cated that trastuzumab (Herceptin; Genentech, S. San Francisco) induces a partial response in only 23% of individuals with advanced HER-2/neu-overexpressing breast cancers10; early indications are that bevacizumab (Avastin; Genentech, S. San Francisco) is not much bet-ter in colon cancer. All of these agents have serious associated toxicities11–13, most extend patient survival only by a matter of months and there is a variable period of remission before a resistant form of the cancer returns, even in the case of imatinib14.

In the light of these findings, the concept of intervening in cancer networks at a single ‘oncoprotein’ or ‘tumor suppressor protein’

George L. Gabor Miklos is at Secure Genetics Pty Limited, 19 Bungan Head Road, Newport Beach, Sydney, New South Wales, Australia 2106. He was an advisor to the Berkeley Drosophila Genome Project and to the human and mouse genome projects at Celera. e-mail: [email protected]

Figure 1 Glacial progress. National Cancer Institute data of 5 year relative survival for patients with distant metastases of colorectal, lung, breast and prostate cancer.

0

5

10

15

20

25

30

35

40

Prostate cancer

Female breast cancer

Lung and bronchal cancer

Colon and rectal cancer

1993–19971988–19921983–19871978–19821973–1977

Year

Su

rviv

al (

%)

C O M M E N TA RY©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


has thus far had decidedly mixed results when translated to the clinic. This should not be surprising as the majority of solid tumors are characterized not by single gene-based events, but by multiple genomic alter-ations, which are specific to each tumor in an individual.

Correlations and mutational dataAnalysis of somatic mutations in dissemi-nated single tumor cells from the bone mar-row of breast cancer patients reveals that contrary to dogma, mutation of TP53 (also known as p53) is not an early event in sys-temic breast cancer, but may occur in some individuals later during metastatic progres-sion2,15,16. Likewise, whereas mutations in the phosphatidylinositol 3-kinase gene have been documented in 74 out of 199 colorectal tumors and 4 out of 15 glioblastomas, they are found in only 1 out of 12 breast cancers, 1 out of 24 lung cancers, none out of 11 pan-creatic cancers and none out of 12 medul-loblastomas17. In the case of the epidermal growth factor receptor (EGFR), somatic mutations are found in 1 out of 61 non-small cell lung cancer tumors from US patients, but in 15 out of 58 Japanese patients with the same cancer18. Furthermore, although somatic mutations in CDC4 are claimed to be a chief cause of chromosomal instability in cancers generally19, only 22 out of 190 colon tumor samples have these mutations. It is not known which of the above mutations existed in a homo-, hetero- or hemizygous condition within particular cells in these samples—a prerequisite for functional and causal interpretations. Thus, correlations remain weak and the above data are equally consistent with the majority of these muta-tions being innocent bystanders in the pro-cesses of tumor progression.

There is enormous phenotypic variation in the extent of human cancer phenotypes, even among family members inheriting the same mutation in the adenomatous polypo-sis coli (APC) gene believed to be causal for colon cancer. In the experimental mouse knockout of the catalytic gamma subunit of the phosphatidyl-3-OH kinase, there can be a high incidence of colorectal carcinomas or no cancers at all, depending on the mouse strain in which the knockout is created, or into which the knockout is crossed20. Finally, the experimental centerpiece on which theo-ries of human oncogenesis are based, namely that overexpression of only two oncogenes and a telomerase catalytic subunit is suffi-cient to create a malignant human tumor21, simply does not hold up experimentally for a diploid human cell22.

Massive genomic imbalances and methylationThe Human Genome Project used diploid genomes in which the principles of mutation, reversion, suppression and methylation are reasonably understood, and in which the tran-scriptomes, proteomes and higher-order sys-tems are characteristic of particular cell types at equilibrium. In contrast, all the genomes of the thousands of solid tumors that have been examined microscopically or molecularly depart from diploidy23–25, as do multiple myelomas26. Some regions of a cancer genome are differentially amplified, others are deleted and many are rearranged.

Solid tumors consist of a heterogeneous population of aneuploid and/or segmentally aneuploid cells, where spontaneous mitotic nondisjunction at a single cell division can inexorably change the dosages of thousands of genes, microRNAs as well as other noncoding entities and the dosage-mediated interactions of thousands of noncoding single nucleotide polymorphisms, all without any mutational input whatsoever. Equally important, cancer genomes undergo massive changes in hyper- and hypomethylation27 leading to large altera-tions in gene activity. These clinically profound genome-wide methylation changes in genes and regulatory regions can occur completely independently of mutation.

This huge departure from diploidy com-bined with a changing methylome introduces novel properties to cellular networks, such that conventional interpretations of phenotypic change through mutations have little trac-tion. Transcriptomes emanating from variably aneuploid genomes are subjected to pertur-bations that far exceed anything that single gene effects can muster. The descriptions of homo-, hetero- and hemizygosity, dominance and recessiveness, and neomorphic, antimor-phic, hypermorphic and hypomorphic alleles lose their conventional meaning because allelic and methylomic dosages are mere parameters in perturbed networks where the key is net-work flux.

It is in such aneuploid contexts that multi-drug resistance rapidly develops, even when major multidrug resistance genes are experi-mentally deleted from a genome28. As deficien-cies involved in loss of heterozygosity cannot revert, and because back-mutation frequencies are low, conventional mutational interpreta-tions of the rapidity of multidrug resistance and reversion do not hold29.

The real killer—metastasisPrimary tumors are heterogeneous at all levels30,31 and many remain dormant32. The heterogeneity is illustrated by microarray

data showing that the correlation among the expression profiles for three different parts of the same kidney tumor is poor33. The hetero-geneity among tumors from the same indivi-dual is also extensive. In prostate cancer, there are often many distinct foci in the same pros-tate, only one of which may be invasive and have a deleterious effect on the patient34. Thus sequence data derived from a primary tumor are problematic because the important func-tional variation between cells is obliterated. Single-cell data are far more informative.

Analysis of single disseminated tumor cells after curative resection of the primary breast cancer reveals that disseminated cells can exhibit changes completely different from those observed in the primary tumor15. Similarly, when bone marrow micrometastases are com-pared with the primary colorectal tumors from the same patient, cells disseminated to the bone marrow do not always carry the same K-ras mutations as the primary tumor35.

Money well spent?No one doubts that primary tumors accumu-late somatic mutations over time. However, the Achilles’ heel of cancer is not the mutational baggage train of the primary tumor, but the genomic imbalances and methylation changes of the deadly cohort of cells that metastasize in different genetic backgrounds. As a megapro-ject in advancing cancer research and ultimate cures, the human cancer genome project thus is fundamentally flawed.

First, as the mutational spectra of the pri-mary tumor and its metastatic derivatives may only partially overlap, therapeutic strategies developed for specifically targeting mutations in primary tumors are unlikely to eradicate cells that have already left the primary tumor and are evolving along different genomic tra-jectories. The clinically significant entity is not the primary tumor per se but the rare cells within it that give rise to metastases and the particular genetic background within which they occur36–40.

Second, there is growing evidence for the profound clinical effects of genome-wide methylation changes in genes and regulatory regions41. These changes can take place com-pletely independently of oncogenic or tumor suppressor or mismatch repair mutations and would not be detected by a human cancer genome sequencing effort.

Thus, although a mutation-cataloging research megaproject may be a diverting occupation for sequencing centers and gene hunters, leading scientists should think care-fully before they tout its therapeutic promise to patients and politicians. The simple truth is that the money would be much better spent

COMMENTARY©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


if research priorities were reevaluated. A good place to start would be to dismiss the falla-cious notion that single mutations in primary tumors are the optimal starting point for research that would lead to the discovery of new, more effective cancer drugs. The clinical reality is that it is not single genes, but rather the properties of aneuploid-based methylated networks that allow metastatic cancer cells to explore novel niches in different genetic back-grounds and to rapidly become resistant to drug-based therapies.

1. Pollack, A. New York Times, March 28, A1 (2005). 2. Klein, C.A. Adv. Cancer Res. 89, 35–67 (2003).3. Leaf, C. Fortune, 149, 76–97 (2004).4. SEER Program (www.seer.cancer.gov) SEER* Stat

Database; SEER 18 Regs, Nov 2004 Sub (1973-2002 varying), National Cancer Institute, DCCPS, released December 2004.

5. Vogelstein, B. & Kinzler, K.W. Nat. Med. 10, 789–799 (2004).

6. SEER Program (http://www.seer.cancer.gov) SEER* Stat

Database; SEER 18 Regs, Nov 2004 Sub (1973-2002 varying), National Cancer Institute, DCCPS, released December 2004.

7. Sonnenschein, C. & Soto, A.M. Mol. Carcinogen. 29, 205–211 (2000).

8. Harris, H. Nature 427, 201 (2004).9. Marx, J. Science 304, 658–659 (2004).10. Mohsin, S.K. et al. J. Clin Oncol. 23, 2460–2468

(2005).11. Hurwitz, H. N. Engl J. Med. 350, 2335–2342

(2004).12. Ozcelik, C. et al. Proc. Natl. Acad. Sci. USA 99, 8880–

8885 (2002).13. Bendell, J.C. et al. Cancer 97, 2972–2977 (2003).14. Hofmann, W.-K. et al. Lancet 359, 481–486

(2002).15. Schmidt-Kittler, O. et al. Proc. Natl. Acad. Sci. USA

100, 7737–7742 (2003).16. Klein, C.A. Cell Cycle 3, 29–31 (2004).17. Samuels, Y. et al. Science 304, 554 (2004).18. Paez, J.G. et al. Science 304, 1497–1500 (2004).19. Rajagopolan, H. et al. Nature 428, 77–81 (2004).20. Barbier, M. et al. Nature 413, 796 (2001).21. Hahn, W.C. et al. Nature 400, 464–468 (1999).22. Akagi, T. et al. Proc. Natl. Acad. Sci. USA 100,

13567–13572 (2003).23. Pollack, J.R. et al. Proc. Natl. Acad. Sci. USA 99,

12963–12968 (2002).

24. Aneuploidy Conference Abstracts. Cell. Oncol. 26, 171–269 (2004).

25. Duesberg, P. et al. Cell Cycle 3, 823–828 (2004).26. Fonseca, R. Blood 102, 2562–2567 (2003).27. Rush, L.J. Blood 97, 3226–3233 (2001).28. Duesberg, P. et al. Proc. Natl. Acad. Sci. USA 98,

11283–11288 (2001).29. Duesberg, P. et al. Proc. Natl. Acad. Sci. USA 97,

14295–14300 (2000).30. Al-Hajj, M. et al. Proc. Natl. Acad. Sci. USA 100,

3983–3988 (2003).31. Al-Hajj, M. et al. Curr. Opin. Genet. Dev. 14, 43–47

(2004).32. Folkman, J. & Kalluri, R. Nature 427, 787 (2004).33. Vasselli, J.R. et al. Proc. Natl. Acad. Sci. USA 100,

6958–6963 (2003).34. Masters, J.R.W. & Lakhani, S.R. Nature 404, 921

(2000).35. Tortola, S. et al. J. Clin. Oncol. 19, 2837–2843

(2001).36. Fidler, I.J. & Kripke, M.L. Nat. Genet. 34, 23 (2003).37. Hunter, K. et al. Nat. Genet. 34, 23–24 (2003).38. Ramaswamy, S. et al. Nat. Genet. 34, 25 (2003).39. Dick, J.E. Proc. Natl. Acad. Sci. USA 100, 3547–3549

(2003).40. Kondo, T. et al. Proc. Natl. Acad. Sci. USA 101, 781–

786 (2004).41. Egger, G. Nature 429, 457–463 (2004).

COMMENTARY©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Beware the biotech barkerTom Jacobs

Biotech company executives’ public statements can tell you a lot, but even the most experienced investors have trouble separating their spin from fact. This is especially difficult because most biotechs are development-stage com-panies lacking profits, dependent for years on selling hope and dreams. Yet knowing the dif-ference between story and substance is the only way to determine whether you are investing or speculating—gambling, really—and making money and avoiding losses. Management is not necessarily to blame, but investors must recognize the game.

Filling the cookie jarThe game for development-stage biotechs has one invariable rule: like teenagers, they never have enough cash. Once public, companies desperate to fund a decade or more of lead product research and development must raise money through selling more of their shares and/or borrowing money (by selling their debt, often with the provision of converting debt to shares; Nat. Biotechnol. 21, 855, 2003). The greater the demand for shares, the higher the share price, the more money to be made by selling new shares.

But with no profits to form a sound financial footing to a stock’s valuation, there is only hope and dreams—air—to boost the stock price, so the hotter the air, the better. Company execu-tives perform a relentless round of public rela-tions designed to garner attention. A wonderful story can vault a stock price to the heavens, but even if a company’s stock is stuck in one of the frequent cycles where investors desert biotech in droves, the story can still snare will-ing lenders. And so for hopeful biotechs, noth-ing is more important than the story and the storyteller.

Step right up!All development-stage biotech bigwigs have to develop their barker skills, tirelessly and convincingly telling a better story to divert investor attention from other rides and attract them to their own. But you and I must hear their message warily. Company management covets not us but institutional investors, the deep-pocketed buyers that can lay out the mil-lions or even billions to buy enough shares or debt to make a real difference. Compared with them, you and I, so-called retail investors with 10, 100 or 1,000 shares, are mere insects, yet we hear the same pitch. Executive statements may come with legal disclaimers, but not “This is for Wall Street, not John/Jane Q. Investor.” The responsibility for caution is ours.

Barker par excellenceConsider perhaps the greatest biotech barker in biotech history, former Celera Genomics (Rockville, MD, and S. San Francisco, CA, USA; NYSE:CRA) president, J. Craig Venter. Thanks to management at Applera (Norwalk, CT, USA), the parent of both Applied Biosystems (Foster City, CA, USA; NYSE:ABI) and Celera, Venter in 1999 took the helm of Celera and also the tide in the affairs of men—the incredible public attention focused on sequencing the human genome—at its flood.

It was a perfect storm, a once-in-a-millen-nium combination of public attention and investor interest in biotech. Venter was the consummate biotech barker, the rebellious, larger-than-life wild genomist who blew apart the scientific establishment with his shotgun method of gene sequencing. The hungry media went nuts. The government’s program was wrong, slow and perhaps even harmful in its delay! Celera’s celerity would bring us all per-sonal DNA cards, individualized medicine and investor riches!

The stock rose from $7.34 in June 1999 to a high close of $247 on March 6, 2000, about 35 times in less than nine months—an unparal-leled gestation period. Whatever the combina-tion of company spin and media attention, it did

the job for Celera. At the height of the frenzy, Celera sold almost $1 billion in stock to large institutional investors.

When the carnival leaves townToday, the stock is around $10, down 96% from its March 2000 peak, Venter has departed, and Celera’s genomic information business now belongs to sister company Applied Biosystems. Celera has become a drug maker with a pipe-line but no products, along with a 50% share of Celera Diagnostics (Alameda, CA, USA) and its products.

The cash raised at the carnival’s peak pro-vided quite a lifeline. Even with this year’s expected $135 million to $150 million cash burn, Celera’s stash should last another five years. Many retail investors sport more serious burns, because they did not take the profits the carnival brought them.

A rule of thumb for speculationThus, it’s key to know the difference between investing and speculation, between the cold hard nature of company numbers in a quar-terly or annual filing, and a pied piper CEO. We should invest all or most of our money by estimating what a profitable company is worth based on current product sales and profits and estimates of a reasonable future.

We should only speculate, if at all, with small amounts we can afford to lose, buying but sparingly shares of unprofitable compa-nies with promising stories. And when we do speculate, we should be happy, and take any profits flighted to us, with the hot air. I make it a general practice to consider selling half my shares of a speculation if and when the price doubles from my purchase price. Then no matter what happens, I’ve not lost anything, and—to use a gambling phrase—I’m playing with the house’s money.

Let’s resolve to know the difference between investing and speculating and to know that while executives of development-stage biotechs may sing siren songs of biotech love, they aren’t for thee and me.

Tom Jacobs is cofounder of Complete Growth Investor (http://www.completegrowth.com). He welcomes your comments at [email protected]. Tom owns no shares of companies mentioned in this article.

I N V E S T O R ’ S L A B©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Navigating an ethical patchwork—human gene banksKaren J Maschke

Population genetics research collaborations are reaching increasingly across national boundaries to access human tissue repositories. Will discrepancies in national policies on informed consent and IP rights hinder progress?

Several countries are establishing banks of human blood samples, or biobanks, associated with electronic health records in the hope that population genetics can speed the identification of disease susceptibility genes or diagnos-tic biomarkers. Concurrently, biotech companies are independently amassing private collections of DNA and tissues or seeking to collaborate with public population databases and biobanks. The collection and storage of biologi-cal samples raises several ethical and policy issues about access to, and use of, these samples, particularly around issues related to informed consent. However, no binding international regulatory framework addresses these issues. Instead, a patchwork of national laws, regulations and ethics advisory body guidelines govern the collection, storage and research use of biological samples. A review of policies at the major biobanks in North America, Europe and Asia reveals that although wide agreement exists on issues such as informed consent and patient con-fidentiality, there is no consensus on consent procedures for new research on previously stored samples, on informing individuals about (or even providing them access to) the results of research carried out on their samples, or on intellectual property (IP) restrictions on biospecimens and data.

The rise of the gene bankAccording to a 1999 Rand report1, over 300 million samples obtained primarily during

routine clinical and surgical procedures are stored in the United States in a wide variety of public and private institutions. Human biological samples have also been collected in Europe and elsewhere, though little is known about where and how many samples are stored. In Scandinavian countries, for example, some long-standing public healthcare systems have been stockpiling collections of human tissues and blood for decades. However, not all of these samples are readily accessible to research-ers and some may be unusable for purposes of genetics research because they were not stored properly, they were not properly annotated, or consent had not been obtained for their

use in research. As a consequence, a new wave of national and local initia-tives across the globe aims to establish genetics research biobanks (collections of blood or tissue samples) that can be linked to medical, genealogical or lifestyle information about a specific population gathered using a specific consent process.

These biobank initiatives take the form of national efforts to collect health and genetic data on large populations (e.g., the Estonian Genome Project, Quebec CARTaGENE, UK Biobank, Singapore Tissue Network or Biobank Japan; see Table 1), collections of tis-sues from focused population groups created by provincial clinics or private ventures (see Box 1 and Table 2), or public-private partnerships (e.g., the collaboration between the Icelandic parliament (Althingi) and US-owned deCODE Genetics (Reykjavik, Iceland) or the partnership between the Swedish Medical Biobank and UmanGenomics in Umeå, Sweden).

Apart from initiatives aimed at large populations at the national level (see Table 1), a diverse range of bio-

banks are also being created provincially. In the United States, for example Northwestern University (Chicago, IL, USA) created the NUgene Project to collect and store DNA sam-ples and healthcare information from associ-ated hospitals and clinics. In Wisconsin, the Marshfield Clinic has initiated the Marshfield Personalized Medicine Project to collect DNA samples from 40,000 local citizens. Biobanks also exist at hospitals and clinics affiliated with Duke University (Durham, NC), the University of Alabama (Birmingham) and the Mayo Clinic, Rochester, Minnesota.

Other projects are underway to focus on specific diseases. For example, an Alzheimer’s

Karen J. Maschke is the associate for ethics and science policy at the Hastings Center, 21 Malcolm Gordon Road, Garrison, New York 10524-5555, USA. e-mail: [email protected]

And

rew

Bro

okes

/Cor

bis

F E AT U R E©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Genebank has been set up and is jointly spon-sored by the US National Institute on Aging and the Alzheimer’s Association. To facilitate cancer research, The US National Cancer Institute (NCI; Bethesda, MD, USA) is cur-rently setting up a National Biospecimen Network, which aims to centralize provin-cial repositories, such as those at Duke or the Mayo Clinic, into one place2. In Europe, there

are existing projects such as the UK’s National Cancer Tissue Resource.

Added to these initiatives are biobanks set up to study race (e.g., Howard University’s Genomic Research in African Diaspora Biobank), twins (e.g., the European GenomEUtwin project, a population-based project involving Danish, Finnish, Italian, Dutch and Swedish twins) or to provide anthropologic information, such as

the recently announced Genographic Project, which aims to collect DNA samples from over 100,000 people worldwide to trace human migrations throughout the globe.

Informed consentFive countries—Estonia, Iceland, Norway, Sweden and the United Kingdom—have national legislation governing the collection,

Table 1 Population-based biobank projectsProject (location)

Funding Description Consent IP rights URL

CARTaGENE (Quebec, Canada)

Supported by public money from Genome Canada.

Commencement immi-nent. DNA to be extracted and stored from 50,000+ adults in Quebec between 25 and 74 years.

Multi-layered options, samples double-coded.Results not shared with donors.

Researchers or biobank permitted to obtain IP on inventions.

http://www.rmga.qc.ca/en/cartagene.htm

Estonian Genome Project (Tartu, Estonia)

Supported initially by private money from EGeen (Mountain View, CA, USA). Since January 2004, the project has received addi-tional funding from the Estonian government.

Initiated in October 2002. DNA to be extracted from blood samples of 1 million Estonian adults and chil-dren together with health and genealogical data.

Blanket consent. Samples coded. Donor option for results.

Owned by EGP, IP policy unknown (since for profit company participation ter-minated in Dec. 2004).

http://www.genomics.ee/

Latvian Genome Project(Riga, Latvia)

Supported by the Latvian Genome Foundation.

Pilot project initiated in 2002 with 60,000 pilot samples.

Blanket consent. Samples coded.

IP rights owned by Latvian Genome Foundation, which plans to market access to database.

http://bmc.biomed.lu.lv/gene/

Icelandic Biobank (Reykjavík, Iceland)

A public-private col-laboration with deCODE (Reykjavík).

Blood samples to be col-lected from 270,000 Icelandic citizens and linked to Iceland Health Sector Database and genealogical records.

Informed consent, donor informed of objectives. Presumed consent of pre-viously collected samples. Results not shared with donors.

deCODE has a 12-year exclusive license to rights over samples, but does not own them. Icelandic health system takes a share of any profits (capped at $1 million per year).

http://www.decode.com/

UK Biobank (Manchester, UK)

Publicly funded by the Wellcome Trust, the Medical Research Council, the UK Department of Health, and Scottish Executive.

DNA, medical records and lifestyle questionnaires to be acquired from 500,000 UK adult volunteers between 45 and 69 years old to begin in 2006. Subjects to be followed for 30 years.

Blanket consent. Donors not provided with research results.

WellcomeTrust owns samples and database. IP policies not in place.

http://www.ukbiobank.ac.uk/

Medical Biobank (Umeå, Västerbotten, Sweden)

Publicly funded by Swedish National Healthcare System.

Contains over 85,000 DNA samples from individuals of 40, 50 and 60 years of age in Västerbotten county together with medical records.

Informed consent acquired from previous donors for each new project; UmanGenomics has access only to coded sam-ples. Donors not provided with research results.

UmanGenomics has exclusive rights to genetic samples from existing Medical Biobank and exclusive right to commer-cialize information derived from Biobank.

http://www.biobanks.se/medicalbiobank.htm

Singapore Tissue Network (Biopolis, Singapore)

Publicly funded since March 2002 by Singapore Biomedical Research Council, Agency for Science, Technology and Research, Ministry of Health and the Genome Institute of Singapore.

Genomic information from Singapore population groups via a network of collaborating organizations and hospitals.

Informed consent acquired from donors for each new project using previously collected samples.

The Genome Institute of Singapore will avoid any commercialization of the project.

http://www.stn.org.sg/

Biobank Japan(Kanagawa, Japan)

Initiated in 2003 with public funding from the Japanese Ministry of Education, Culture, Sports, Science and Technology.

DNA samples to be acquired from 300,000 Japanese individuals of 20+ years of age suf-fering from 30 common illnesses.

Full informed consent acquired from donors. Donors provided with research results. If comprehensive consent obtained, use of existing samples allowed for new research.

IP policies not yet in place.

http://www.src.riken.go.jp/eng/src/project/person.html

FEATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


storage and use of biological samples (Table 3). Elsewhere, regulations governing research with humans and/or specific guidelines issued by national ethics advisory bodies or regulatory agencies apply (e.g., see Table 4).

Voluntary informed consent from individu-als who participate in research is a core ethi-cal principle of human research ethics. This must be balanced with the need to facilitate research using biological samples without

undue logistical and ethical hurdles (see Box 2). Extending this principle to research with biological samples means giving individuals the opportunity to decide whether they are willing to let researchers collect, store and study their

Following a firestorm of controversy in 1998 when the Icelandic government agreed to provide deCODE Genetics with exclusive rights to the nation’s medical and genealogical records, the biotech sector has been increasingly involved in collaborations with public population databases and biobanks. In Sweden, the Medical Biobank has also granted UmanGenomics exclusive rights to national medical and genetic data and, elsewhere, the Latvian Genome Project (and initially the Estonian Genome Project) was initiated with the aim of providing economic benefit from associated biotech ventures.

Several business models are emerging in the sector. Certain companies are setting up their own tissue repositories and selling this information to interested parties; others are negotiating access to samples and patient information in public population databases and tissue repositories, and using these to identify new targets or biomarkers for internal research and development programs; others are providing expertise and technology to support public institutions or drug companies undertaking clinical work; and yet others are pursuing combinations of the above (see Table 2).

The handling of human samples and medical records by corporations presents some unique ethical challenges. In the United States, although companies provide assurances that they safeguard donor privacy and interests, there is no legal requirement for companies to protect human subjects. And it is not clear what happens to patient medical records and confidential information when a private biobank goes bankrupt. Often guarantees made by a company during sample acquisition are not legally binding on the trustee in a bankruptcy. When DNA Sciences of Fremont, California, went bankrupt in 2003, samples and data from 18,000 donors were part of its assets; in this case, DNA Sciences and its assets were bought by Genaissance Pharmaceuticals (New Haven) for $1.3 million. In 2001, a court in Japan auctioned off a human cell collection that a scientific society had used as collateral on a loan.

Reconciling the commercial imperatives of private companies with the goals of public tissue banks to improve the health of local populations also presents problems. After intense criticism, deCODE Genetics agreed to give the Icelandic health system

a portion of any profits accrued (with a cap of about $1 million a year). IDgene of Jerusalem, a now-defunct Israeli company collecting samples from Ashkenazi Jews, also had a similar arrangement to donate a percentage of future profits to the healthcare of Israeli society.

In the case of the Icelandic project, perhaps the most contentious issue was its reliance on ‘presumed consent,’ whereby the entire population’s health records are automatically included in the database unless a citizen specifically requests otherwise. To date, up to 7,000 individuals have opted out. But deCODE is by no means unique in prompting intense public opposition. In 2000, Australia’s Autugen (now AGT Bioscience, Victoria) made an aborted attempt to form a collaboration with the Tongan Ministry of Health. This project fell through after the Ministry failed to give the company access to Tongan medical records; tribal leaders and opposition groups claimed there was insufficient public consultation and the informed consent procedures failed to take account of Tongan traditions in which the extended family is involved in decision making.

Box 1 Ethics and the private sector

Table 2 Selected biotech companies involved in biobanks and genetics researchCompany Activity

AGT Biosciences (formerly Autogen, Victoria, Australia)

Internal discovery program accesses a unique DNA collection from world-wide populations (>44,000 samples). External research collaborations with pharma.

Ardais (Lexington, MA, USA)

Provides informatics support and advisory services to facilitate biospecimen collection, management and distribution by biobanks and academics and drug companies involved in clinical work.

deCODE genetics (Reykjavík, Iceland)

Contracted by the Icelandic government to put the health records of all 270,000 citizens into a single database (~7,000 citizens have thus far elected to opt out). Using database and Icelandic Biobank, plans to carry out gene association and founder studies for internal drug target discovery and development program and external research collaborations.

EGeen (Tartu, Estonia) Analysis of disease and drug response data using DNA and biomarker profiles from donors in the Estonian Genome Project. Initial focus on hyper-tension.

First Genetic Trust(Chicago, IL, USA)

Provides informatics support and advisory services to biobank developers seeking to protect human subjects and ensure data privacy. Partnered with Howard University to launch the Genomic Research in African Diaspora Biobank.

Genizon Biosciences(Quebec, Canada)

A bank of 50,000 Québecois patients and relatives with 28 different dis-eases. Partnership with Myriad Genetics (Salt Lake City, UT, USA) and Perlegen Sciences (Mountain View, CA, USA).

Genomics Collaborative(Cambridge, MA, USA)

Offers access to a clinically annotated tissue bank of >120,000 donors from around the globe. Provides tissue banking and consulting services to drug industry. Technical/consulting services for the Singapore Tissue Network.

Newfound Genomics (St. John’s, Canada)

Banking samples from individuals in Newfoundland and Labrador provinces to study the causes of diabetes and obesity.

UmanGenomics(Umeå, Sweden)

Exclusive commercial access to Medical Biobank. DNA and plasma samples from more than 70,000 Swedish donors with detailed medical and lifestyle histories.

F EATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


samples. In addition, individual consent may vary depending on what personal confidential information is connected with a sample: sam-ples can be linked to an individual’s medical records, codified so that an individual’s infor-mation can be retrieved only using a specific key, anonymized so that connected personal information cannot be retrieved at a later date or anonymous where no personal information is connected to the sample (see Box 3).

A survey of the main national biobanks reveals consensus across the policies that informed con-sent is required when collecting samples. The Act on Biobanks in Iceland (see Table 3) is, how-ever, unique in that it permits presumed consent (that is, consent is assumed unless otherwise indicated) to govern the storage of samples in a biobank if they were obtained for the purpose of clinical tests or treatment.

Consent options. Although informed con-sent is universally accepted, there is considerable divergence on policies regarding the use of con-sent options. With consent options, individuals have the option to decide whether researchers can recontact them to obtain consent for future studies with their sample and might have the option of opting out of future research. In the United States, the Health Insurance Portability and Accountability Act (HIPPA) Privacy Rule implemented in April 2003 states that patients have to authorize the use of protected health information for each use of their data. When that is impossible, three alternatives are pos-sible: first, an anonymized data set can be con-structed (see Box 3); second, a ‘limited’ data set can be provided with identifiers linked to a data use agreement making the institution liable for violations by the recipient; or third, an Institutional Review Board (IRB) can issue a waiver of authorization indicating that the study poses minimum risk to privacy and the research could not practicably be carried out without access to, and use of, the information.

An approach taken for the collection of biological samples by the Icelandic BioBank involves the use of two consent forms. Individuals who sign consent form ‘a’ autho-rize researchers to use a blood sample for specific research with the understanding that when the research ends, the sample will be destroyed. Consent form ‘b’ authorizes researchers to use a blood sample for specific research and for additional similar research if

approved by the Data Protection Commission and the National Bioethics Committee. By signing consent form ‘b,’ individuals also give consent for DNA to be extracted, coded, stored and used for any research approved by the Data Protection Commission and the National Bioethics Committee3.

The Quebec CARTaGENE project (see Table 1) also uses consent options. Participants in this population-based semi-longitudinal project will give general consent for use of their anonymized samples. An additional consent mechanism will offer choices for opting into three specific research activities.

Unlike the projects in Iceland and Quebec, the Estonian Genome Project and the UK Biobank use a blanket consent approach, that is, consent for unspecified uses of samples. The proposed Ethics Governance Framework (EGF) for the UK Biobank recommends that individ-uals not be given the opportunity to choose which data about themselves will be used or what kind of research can be conducted with their samples. Instead, the EGF recommends that individuals be told they can either opt in or out of the project, with the understanding that if they opt in their samples may be used in the future for unspecified research.

Table 3 Several national laws pertaining to confidentiality of medical data linked with tissue repositories

Country Law/rule (year enacted)

US Health Insurance Portability and Accountability Act(HIPAA; 1996); associated Privacy Rule (2003)

UK Human Tissue Act (2004)

Estonia Human Genes Research Act (2000)

Latvia Human Genome Research Law (2002)

Iceland Act on Biobanks No. 110/2000 (2000); Act on a Health Sector Database No. 139/1998 (1998); Act on the Rights of Patients No. 74/1997 (1997)

Sweden Act on Biobanks (2002)

Norway The Norwegian Act on Biobanks (2003)

Box 2 Balancing individual rights with research progress

From an individual rights perspective, informed consent should be required for research with all types biological samples, whether they are linked with healthcare information or coded, anonymized or anonymous (see Box 3, for explanation). On the other hand, the medical and public health perspectives favor relaxing the informed consent requirement so that important research can go forward11. When potential research harms are minimal or nonexistent, advocates for these perspectives place greater weight on the needs of science than on the right of self-determination, particularly for use of coded, anonymized and anonymous samples. Research harms are said to be minimal or nonexistent when using anonymized and anonymous samples, and when using coded samples if researchers do not have access to the code. Under these circumstances, the need to affirm and protect the right of self-determination is considered to be less compelling than when samples are identifiable.

However, for human dignity and privacy to have meaning, individuals ought to have self-determination over their body and its materials, especially if research carries potential harms. Many commentators assert that research with coded, anonymous and anonymized samples may be harmful to individuals, families and communities. Information derived from population-based studies might be used to discriminate against or stigmatize individuals from these populations. Moreover, because there is cultural variation in the way the body and its materials are viewed, special handling practices may be required when samples are obtained from certain populations.

Even where there is consensus for obtaining informed consent, disagreement exists over whether individuals should have the opportunity to opt in or out of certain types of research and to specify their preference regarding notification of research results. From the individual rights perspective, the more options individuals have the better, whereas a blanket consent approach gives researchers greater latitude in how they will use samples for which specific consent was not obtained, and affirms medical and public health goals of advancing science.

FEATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


In Germany, the National Ethics Council (NEC) recommends that the purpose of the research should determine whether consent options are offered. The NEC contends that the absence of consent options does not vio-late the right of self determination because individuals can decide whether or not they want to participate in genetics research under specified conditions.

Norway also favors a conditional approach to consent options: risk of research harms, sensitivity of the biological material and the vulnerability of the research subject are factors that determine what information is provided to individuals and the specificity of consent.

None of the policies examined appear to explicitly endorse a consent option that gives individuals the opportunity to decide whether their identifiable sample can later be anony-mized for research use.

New uses of stored samples. There is no uniform approach regarding consent for new uses of stored samples. Policies in Canada, Germany, Norway, the Netherlands and the United States permit use of stored samples without consent if the samples are not iden-tifiable. Iceland’s policy confers decision-making authority about the need for consent for new use of stored samples to its National Bioethics Committee. In Estonia and the United Kingdom, consent is not required for new use of stored samples because individuals gave blanket consent for research with their samples at the time of collection.

Under the newly enacted UK Human Tissue Act of 2004, exceptions to the requirement for informed consent include the provision that consent is not required if samples are ano-nymized and if an ethics review committee approves the research.

Minors and incapacitated adults. Policies in Denmark, Estonia, Sweden, the Netherlands and the United Kingdom contain special rules for minors and incapacitated adults. In the Netherlands, consent must be obtained from both parents or legal representatives for chil-

dren less than 12 years old; children between the ages of 12 and 18 must give consent, along with both of their parents or legal representa-tives. Parents in Estonia must give consent for children over the age of seven.

Regulations in the United States governing research with children require parental con-sent and child or adolescent assent when cer-tain conditions are met. However, it is unclear whether IRBs in the United States require researchers to obtain assent from children/adolescents whose samples are collected or whether adolescents who reach the age of legal maturity are contacted to obtain their consent for ongoing use of their samples.

The UK Biobank and the Quebec CARTaGENE projects will collect samples from adults, presumably only from adults with deci-sional capacity. The proposed EGF for the UK Biobanks states the “UK Biobank will have to have clear policies on how to respect partici-pants’ wishes if they become incapacitated or die” but does not say whether samples will be collected from incapacitated adults with sur-rogate permission.

Access to research resultsWhether individuals should have access to information obtained from research with their identifiable samples is still open to debate.

Focus groups and surveys indicate that some people want access to research results and that the opportunity to obtain personal informa-tion is what motivates some of them to con-sent to use of their samples4. However, only the Estonian Genome Project requires researchers to give individuals access to personalized infor-mation if they want it.

The proposed EGF for the UK Biobank says individual research results should not be given to participants because the Biobank is a research project, not a healthcare project. The German NEC mentions the issue of research results in its report “Biobanks for Research,” but does not make a recommendation for or against providing results to individuals.

Consultative approachesAlthough there has been extensive discus-sion about community consultation for genetics research, none of the national poli-cies examined requires this approach when researchers collect samples from socially identifiable groups. The German NEC men-tions the notion of group consent, but notes that problems associated with research on indigenous populations are not present in Germany. Although several of the European countries examined here do not have indig-enous populations and traditionally have had homogenous populations, nearly all of these countries have experienced an influx of immigrants from the Middle East, Africa and Asia. Consequently, researchers and biobank programs may need to address cultural issues surrounding the collection and use of genetic samples from these and other populations.

If community consultation is defined as interaction with the broader public, rather than only with socially identifiable com-munities, then only two large-scale biobank projects, the UK Biobank and CARTaGENE, have incorporated some form of community consultation into the planning and develop-ment of the projects. Various communication

Box 3 Tiered linking of healthcare data to DNA samples

Complicating the consent issue are the various methods that are used for coding and identifying DNA samples in biobanks. Some stored samples either contain personal identifiers or can contain a code that links them to the individual source (identifiable samples). Others have identifiers, but will be anonymized at some later time (anonymized samples). A last category is a sample collected without any link to personal identifiers or for which all identifying information has been destroyed (anonymous samples).

Although anonymous and anonymized samples retain some value for population-based genetics research, it is difficult to pool genetic data for pedigree/family members and look for disease markers. Thus, genetic information linked to individual medical records is of greatest potential for marker discovery and for developing personalized treatment and prevention therapies.

Table 4 International guidelines covering ethics for gene banksOrganization Guideline

World Medical Association (WMA)

Declaration of Helsinki (2000)Declaration of Ethical Considerations regarding Health Databases (2002)

Council of Europe Recommendation on Human Tissue Banks (1994)Convention on Human Rights and Biomedicine (1997)

UNESCO Draft Report on the Collection, Treatment, Storage and Use of Genetic Data (2001)Declaration on the Human Genome and Human Rights (1997)

Human Genome Organization Statement on DNA Sampling: Control and Access (1998)Statement on Benefit Sharing (2000)

FEATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


strategies were used in Iceland and Estonia to inform the public about the projects, though no formal consultative approach was used in establishing the projects’ collection, stor-age and access polices4. In contrast, the UK Biobank and CARTaGENE projects used a “partnership or collaborative approach” in addressing consent and other issues. Both projects used focus groups and surveys to measure public attitudes toward the projects and the ethical issues involved4.

Ownership and IP rightsBiological samples, personal information about sample donors, genealogies and the discover-ies and inventions resulting from research with samples and associated data are likely to have commercial value. Yet, the general principle in international and national law is that there are no property rights on the human body.

For example, in the case of Moore v. Regents of the University of California5, a California court ruled that individuals do not have an ownership interest in their cells after they are removed. More recently, a Florida court sided with the Moore ruling in holding that individuals do not retain an ownership interest in their donated tissue samples6. However, institutions and other entities that store samples and genetic databases have asserted control over samples and data-bases and over the management of IP rights.

The Estonian Genome Project was estab-lished as a public-private partnership between the Estonian Genome Project Foundation (EGPF) and EGeen, a biotech company in the United States. Under the original arrangement, the EGPF would own the samples and genetic data, and EGeen would be the exclusive com-mercial licensee of the database. In December 2004, EGPF and EGeen terminated this agree-ment, which means that EGeen is no longer the primary funder of the project. Media accounts indicate that because of economic realities, EGeen wanted to redirect the project to focus on particular priority diseases rather than follow the original agreement to conduct population-based research. EGPF has not pro-vided information about continued financing for the project or about commercial rights to the data and inventions resulting from research under new financing agreements.

In January 2005, the UK Biobank issued a draft IP and access policy that recommends how IP rights should be managed. IP accru-ing from the creation and development of the biobank is vested in the UK Biobank. When IP arises out of research using the biobank, it is vested in the researcher, his or her institution, or their assignee. In Quebec, the CARTaGENE project does not confer ownership rights to researchers or to the biobank, but the project permits these parties to obtain intellectual

property rights over inventions obtained from genetic information.

The Icelandic Act on Biobanks states that the biobank operator does not own the sam-ples, but has rights over them. Pursuant to the Icelandic Health Sector Database Act (see Table 3), the Icelandic government issued an exclusive license to deCODE Genetics to establish and operate the Icelandic Health Sector Database, which includes the right to sell access to the database7 (deCODE is registered and head-quartered in Reykjavik, Iceland, and is a wholly owned subsidiary of deCODE Genetics, regis-tered in Delaware, USA). Legislation in Iceland does not place special legal restrictions on the licensee’s freedom to negotiate IP rights for itself8. Although deCODE Genetics is conduct-ing research with samples obtained from over 100,000 individuals, it is unclear whether the Health Sector Database is fully operational. In November 2003, the Icelandic Supreme Court ruled that the Health Sector Database Act did not adequately protect personal privacy in accordance with the Iceland Constitution’s pri-vacy guarantee. The Health Sector Database Act permits every citizen’s health data to be entered into the database without informed consent, but gives individuals the opportunity to opt out of the database. In the lawsuit heard by the court, the relative of a deceased individual argued that her right to privacy was violated because information about her could be inferred from data related to hereditary characteristics of her deceased relative, whose health data had been entered into the database (Box 4).

ConclusionsAll of the policies discussed above require some form of ethical oversight for the col-lection, storage and research use of biological samples. Moreover, there is an emerging con-sensus that individuals should be told about policies for sample ownership, managing IP rights and protecting privacy and confidenti-ality of genetic information.

The ethical situation is far less clear for bio-banks containing stored samples that were collected without consent for use in genetics research. There is a notable lack of consen-sus on whether informed consent should be required for research with stored samples that were collected without consent for research or that were collected for research that differs from the proposed study. Even so, research with stored samples is going forward, and new samples are being collected for storage in research biobanks.

It remains to be seen whether variation in consent policies will hinder these activities. Organized opposition to the large-scale pro-jects in Iceland and the United Kingdom and

Box 4 Debate moves into the courts

Although the case law involving research with human biological samples is sparse, some ethical and policy controversies are finding their way into the courts.

Iceland. In November 2003, the Icelandic Supreme Court ruled that the Icelandic Health Sector Database Act (see Table 3) does not adequately protect personal privacy in accordance with the Icelandic Constitution’s requirement that “Everyone shall enjoy the privacy of his or her life, home, and family.” The Supreme Court agreed with the plaintiff’s claim that she has a personal interest in preventing her deceased father’s medical records being entered into the database because it might be possible to infer information about her from his records. The court ruled that under the Act’s presumed consent provision, deceased individuals and their living relatives are at a disadvantage in their ability to opt out of the Health Sector Database, which will contain the medical and genealogical records of all Icelanders and will link this information to genetic information obtained from the tissue samples of the Icelandic population deposited in a biobank.

United States. Two lawsuits seeking $75 million were filed in 2004 against the University of Arizona, its Institutional Review Board, several genetic researchers and other named defendants. Several members of the Havasupai Tribe filed one of the lawsuits, and the tribe itself filed the other one. The plaintiffs allege that blood samples obtained from tribal members for diabetes research were used without their consent for secondary research to identify an association between certain gene variants and schizophrenia and for ancestral migration studies. Additional charges against the defendants include allegations of breach of fiduciary duty, fraud and misrepresentation, infliction of emotional distress, conversion, violation of civil rights and various claims of negligence. The federal lawsuits are pending in the US District Court for Arizona.

FEATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


public opinion surveys and lawsuits in the United States and Iceland reveal public con-cern about consent issues, confidentiality of genetic information and the type of research conducted with their samples.

Some commentators have called for harmo-nizing policies for research with human tissue across national boundaries9. Several transna-tional biobanking projects are attempting to develop a minimal threshold of policy harmo-nization within their respective spheres. These projects include GenomEUtwin; Population Project in Genomics (P3G), an open-access database that will integrate data from the UK Biobank, the Estonian Genome Project, CARTaGENE and GenomeEUtwin; and the International HapMap Consortium, which involves research groups from Canada, China, Japan, Nigeria, the United Kingdom and the

United States. Whether these transnational attempts at harmonization will succeed, or whether an international regulatory framework is possible, remains to be seen.

There is a growing need for more explicit, enforceable and coordinated international pol-icy guidelines. Consent issues remain a potential barrier to harmonization, as do national poli-cies regarding data protection, genetic privacy and IP rights. Moreover, as the failed attempts to develop an international regulatory frame-work regarding human cloning reveals, global policymaking is fraught with deeply entrenched value conflicts and geopolitical realities10.

1. Eiseman, E. & Haga, S.B. Handbook of Human Tissue Sources: A National Resource of Human Tissue Samples (RAND, Santa Monica, CA, 1999).

2. Bouchie, A. Coming soon—a global grid for cancer research. Nat. Biotechnol. 22, 1071–1073 (2004).

3. Árnason, V. Coding and consent: moral challenges of

the database project in Iceland. Bioethics 18, 27–49 (2004).

4. Godard, B. et al. Strategies for consulting with the community: the cases of four large-scale genetic data-bases. Science and Engineering Ethics 10, 457–477 (2004).

5. Moore v. Regents of the University of California, 793 P.2d 479 (Cal. 1990).

6. Greenberg v. Miami Children’s Hosp. Research Inst., Inc., 64 F. Supp. 2d 1064 (S.D.Fla. 2003).

7. Árnason, E. Personal identifiability in the Icelandic Health Sector Database. J. Inform. Law Technol. 2, (2002). http://www2.warwick.ac.uk/fac/soc/law/elj/jilt/2002_2/arnason/

8. Kaye, J. et al. Population genetic databases: a compara-tive analysis of the law in Iceland, Sweden, Estonia, and the UK. Trames 8, 15–13 (2004).

9. Bauer, K. et al. Ethical issues in tissue banking for research: a brief review of existing organizational poli-cies. Theor. Med. Bioethics 25, 113–142 (2004).

10. Maschke, K.J. & Murray, T.H. Ethical issues in tissue banking for research: the prospects and pitfalls of set-ting international standards. Theoretical Medicine and Bioethics 25, 142–155 (2004).

11. Andrews, L. Future Perfect: Confronting Decisions about Genetics (Columbia University Press, NY, 2002).

F EATURE©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


The challenge to patent law of pure chemicalprotein synthesisAaron Xavier Fellmeth

The emergence of pure chemical protein synthesis as a commercially viable method of drug design and production will create serious problems in the patent system.

In the last five years, advances in the chemical synthesis of proteins has reached the point

where it is conceivable that, in the near future, the vanguard of pharmaceutical development will be not recombinant genetics but chem-istry. Because nearly all biological processes are controlled by the molecular recognition of peptides and proteins, the understand-ing and manipulation of these substances is central to neurobiology, enzymology, immunology, pharmacology and molecular biology and biochemistry more generally. Although experiments in chemically synthe-sizing peptides from their constituent amino acids without transcription date back some 100 years, only in the last 15 have chemists advanced from synthesizing amino acids, simple sugars and purines to short peptide chains, such as acyl carrier protein1, and now to medium-sized proteins of 200 amino acids or more, such as erythropoietin2–4. Smaller synthetic peptides have become widely avail-able as important commercial products, from sweet aspartame to clinical hormones such as oxytocin, calcitonin and gonadotropin-releasing hormone (GnRH) super-agonists. Now that it is becoming commercially feasi-ble to synthesize larger proteins chemically, a serious intellectual property challenge looms on the horizon. There is an urgent need to consider the role that chemical synthesis will assume in pharmacology, where recombi-nant genetics is still the standard, and where patents over “isolated and purified” natural peptides and proteins have long been granted with wild abandon.

It is important to realize that chemically syn-thesized proteins need not have precisely the same molecular makeup as their recombinant counterparts. The advantages of chemical syn-thesis of proteins over recombinant produc-tion are several. First and most important in terms of drug design, chemical synthesis offers predictability and customizability not avail-able to geneticists. Through protein research, chemists can rationally design peptide ligands whose biochemical properties can be predicted a priori. Customization now takes chemists in directions so far foreign to geneticists. Chemists can attach unnatural amino acids, pseudopep-tides or polymers to proteins at binding sites to suppress undesired characteristics. For example, the coupling of polyethylene glycol (PEG) polymers has yielded fully bioactive derivatives with reduced immunogenicity5. Polymers may also be added as macromolecu-lar carriers for target-specific drug delivery or sustained release of drugs, with increased solubility and other beneficial pharmacologi-

cal effects6. Polymeric and unnatural amino acid tweaking may change the functionality of the protein, resulting in entirely new charac-teristics or modifications to current bioactive characteristics, such as increased or decreased longevity or reactivity.

Another advantage of chemical synthesis relates to the tricky matter of producing glyco-proteins. Chemical synthesis offers an oppor-tunity for greater homogeneity of glycoforms than is currently possible in recombinant proteins7. Finally, chemical synthesis tends to yield proteins having a high degree of purity. Although automated techniques can typically purify recombinant proteins with great success, there is less risk of DNA impurities or endo-toxins contaminating a batch of chemically synthesized proteins.

Protein patentsDespite the differences between pure chemical synthesis of proteins and recombinant produc-tion, there are and will continue to be cases in

Aaron Xavier Fellmeth is at Arizona State University College of Law, P.O. Box 877906, Tempe, Arizona 85287-7906, USA.e-mail: [email protected]

Cor

bis

PAT E N T S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


which chemically synthesized proteins are iden-tical to recombinant proteins. Unfortunately, the United States Patent and Trademark Office (USPTO), with the sanction of the US Court of Appeals for the Federal Circuit (but, tellingly, not the US Supreme Court), has permitted patents on naturally occurring proteins when “isolated and purified” from their natural solu-tion or matrix. Already, hundreds of naturally occurring proteins have been patented8–10. Such proteins as erythropoetin11, mammary transforming protein12 and human chemo-tactic protein13 have been patented by the first researchers who showed that they puri-fied and recombinantly reproduced the natu-ral substance and found a substantial use for it. Such patents are highly problematic as a matter of both policy and law. They tend to impose tollbooth costs on the development of products and processes relating to biological functions, disorders and diseases involving the protein at issue. This raises the cost of bio-logical and medical research unnecessarily by distributing funds from the developers of new and useful technologies to researchers who have merely identified and purified a protein without inventing anything new, contrary to the requirements of the Patent Act.

The unnecessary costs imposed by patents on naturally occurring proteins raise an espe-cially thorny problem for chemists seeking new ways to synthesize such proteins. A patent on a recombinant protein typically claims the iso-lated and purified protein regardless of how it is produced. In reality, of course, the patentee merely mimicked the natural process of protein production by expressing the DNA artificially. The more difficult task of chemical synthesis produces the same protein in an entirely differ-ent way. As a result, a patent on a recombinant protein may block the commercialization of an identical protein produced by pure chemical synthesis. To the extent that the protein has been substantially modified in the process of chemical synthesis (e.g., by docking polymers to the protein or including unnatural amino acids), it is at least possible that the protein will fall outside the scope of the patent claims. But where chemically synthesized proteins sub-stantially preserve their natural form, a patent battle will inevitably ensue, thereby preempt-ing (or raising the costs of) the development of chemically synthesized drugs.

Substantial transformation of proteinsIn an earlier article, Linda Demaine and I explained the origins of the aberrational pat-ent policy that permits patents on “isolated and purified” naturally occurring biochemicals14. There, we explained at length why such pat-ents are invalid as claiming products of nature.

We also proposed a test for patentability—the “substantial transformation test”—based on whether the patent applicant had transformed the molecule at issue so as to give it a new and different biological function. The purpose of this test was to preserve the traditional prohibition on patenting either products of nature or merely purified, preexisting substances. These doctrines serve several important public policies, including inter alia prohibiting private persons from claim-ing monopolies over products not invented by them, preventing arbitrary upstream monopo-lies that impose gatekeeper costs and inefficien-cies on vast fields of downstream research not invented by the patent owner and preempting costly confusion and litigation over blocking or overlapping patents covering different aspects of preexisting natural molecules.

The chemical synthesis of proteins presents a paradigmatic case of why patents on natu-rally occurring biochemicals were tradition-ally—and should continue to be—disallowed. When a biologist first reproduces a newlydiscovered natural protein through recombinant

expression, he has done several things, none of which necessarily results in an ‘invention.’ By identifying the structure and chemistry of the protein, he merely reveals a fact about nature. This discovery is unpatentable as lacking statu-tory subject matter. By reproducing it recombi-nantly, he may merely use well-known methods for identifying and reproducing it. Assuming the reproduction process is not patentable (because, for example, the process is one commonly used), a purified version of a naturally occurring pro-tein is equally not properly an invention in the sense of the patent law. Besides generally lack-ing the required quality of nonobviousness15, it, too, lacks patentable subject matter. The recombinant protein has the same structure and performs the same biological function as its naturally occurring analog.

Patent avenues to policy outcomesBy granting patents on such proteins, the pat-ent system necessarily preempts or discourages the development of new methods of reproduc-ing these natural proteins. Chemical synthesis

is a prototypical example of such a method. When a chemist reproduces the same protein through pure synthesis, there is again no prod-uct invention (although there may be a process invention), because the product is the same as the purified recombinant protein. If the pat-ent system has allowed the biologist to patent the protein, the chemist not only can obtain no patent on the synthesized version of the same protein, but he cannot commercialize the syn-thesized protein at all without the permission of the biologist. Yet, the process used by the chemist is entirely different from that used by the biologist. The biologist cannot reasonably claim that the chemist has infringed a process patent, because pure chemical synthesis in no way resembles recombinant protein produc-tion. And the protein itself already existed in nature. The ineluctable deduction is that the notion that the chemist producing a pure protein synthetically infringes the biologist’s ‘invention’ (a purified, naturally occurring pro-tein reproduced recombinantly) is a legal and scientific absurdity. Yet, it logically follows from the patent theories now commonly accepted by the USPTO and the Federal Circuit.

If any product is patentable, it is a protein that has been substantially transformed by way of physical alteration so as to perform a new biological function. Biologists and espe-cially proteomic chemists have already made significant advances in understanding struc-tural motifs and post-translational modifi-cation. This knowledge has resulted in, and foreshadows more, great advances in design-ing new methods of altering proteins, result-ing in longer-lasting biochemical effects, fewer unintended side effects, greater bioactivity and other benefits. The patent law should encour-age such inventions, but it should also require sufficient proof that the subject matter claimed in the patent application is indeed substantially different from the natural protein. In this way, the benefits of encouraging invention can be preserved without imposing unnecessary costs on researchers.

Beyond the substantial transformation of proteins and peptides, researchers retain other avenues for patenting proteomic inventions. The most common is the process patent. As a kind of tollbooth for research on any product using the patented process, such patents can be extremely lucrative, and the field remains relatively open for patents on novel, efficient methods of designing and synthesizing pro-teins, both existing and new. Among such pat-ents are Genentech’s patent on a method for purifying proteins through Protein A chroma-tography16 and Stephen Kent’s novel method for preparing modified proteins through liga-tion of segmented peptides17. Because process

By granting patents on [“isolated and purified” naturally occurring] proteins, the patent system necessarily preempts or discourages the development of new methods of reproducing these natural proteins.

PATENTS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


patents do not encompass natural products, the tollbooth effect does not inhibit new develop-ments in the field of proteomics insofar as an incentive remains to invent around the process patent.

There are, in short, many ways to synthesize or produce a protein, but inventing around a protein itself is exceedingly difficult. The physical and chemical properties of proteins are highly determined by their structure, and nearly all natural proteins produced by the body are necessary to it in the form produced. Whatever incentive patents on natural proteins provide for researchers to investigate the struc-ture and function of these proteins (which is a necessary precondition to modern rational drug design in any case) is more than offset by the costs imposed on wide fields of future research involving that protein. Chemists who cannot pay royalties and wish to avoid litiga-tion are limited to one significant avenue of research—the invention and synthesis of unnatural proteins.

ConclusionsImproved methods of pure chemical protein synthesis foreshadow a clash between existing patentees of “isolated and purified” naturally occurring proteins that have been produced recombinantly and researchers chemically synthesizing the same protein through new methods. The US patent system has created the possibility of this conflict through the myopic decision to allow patents to claim purified, naturally occurring biochemicals regardless of how they were produced. Probably it never occurred to the judges who created this excep-tion to the general rule that such products are unpatentable or that proteins could be pro-duced in any manner other than recombinant genetics. The impending litigation should serve as a wake-up call to Congress, the USPTO and the Federal Circuit. The very last science policy a sound patent system should pursue is the dis-couragement of new methods for creating nat-ural biochemicals that are essential to advances in the biomedical sciences.

1. Hancock, W.S. et al. J. Bio. Chem. 247, 6224 (1972).

2. Warren, J.D. et al. J. Am. Chem. Soc. 126, 6576 (2004).

3. Kent, S.B.H. et al. Science 299, 884 (2003).4. Kent, S.B.H. et al. Science 266, 776 (1994).5. Burnham, N.L. Amer. J. Hosp. Pharm. 51, 210

(1994).6. Hudecz, F. in Self-Assembling Peptide Systems in

Biology, Medicine and Engineering (ed. Aggeli, A. et al.) 139–140 (Kluwer Academic Publishers, Dordrecht, Netherlands, 2001).

7. Kochendoerfer. G.G. et al. Science 299, 884 (2003).8. US Patent No. 6,806,065 (Oct. 19, 2004) (claiming

inter alia “an isolated nucleic acid molecule encoding a Rickettsia felis outer membrane protein”).

9. US Patent No. 6,800,473 (Oct. 5, 2004) (claiming inter alia human cathepsin L2 protein and its coding gene).

10. US Patent No. 6,794,500 (Sept. 21, 2004) (claiming RNA-binding protein).

11. Amgen, Inc. v. Chugai Pharmaceutical Co., 927 F.2d 1200 (Fed. Cir.), cert. denied, 502 US 856 (1991).

12. US Patent No. 6,057,434 (May 2, 2000).13. US Patent No. 5,880,263 (Mar. 9, 1999).14. Demaine, L.J. & Fellmeth, A.X. Stanford Law Rev. 55,

303, 357 (2002).15. 35 USC § 103 (2004).16. US Patent No. 6,797,814 (Sept. 28, 2004).17. US Patent No. 6,476,190 (Nov. 5, 2002).

PATENTS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Recent patent applications in microarraysPatent # Subject Assignee Inventor(s) Priority

application date

Publication date

WO 200518796 A reaction-conducting system comprising a porousreaction substrate with its top surface bounded to a rigid support having multiple through-going holes, which form reaction zones; useful, e.g., for polymer and peptidesynthesis reactions and as a microarray support.

PamGene (Hertogenbosch,The Netherlands)

Kievits T, Ruijtenbeek R, van Beuningen MG

8/21/2003 3/3/2005

US 20050048580 An array comprising a substrate having multiple addresses, each address comprising a nucleic acid encoding anaffinity-tagged test amino acid sequence, a translation effector and a binding agent that recognizes theaffinity tag. The binding agent is preferably attached tothe substrate; useful for high-throughput analyses ofprotein interactions.

Harvard College (Cambridge, MA, USA)

LaBaer J, Lau AY 8/3/2004 3/3/2005

US 20050048554 A microarray with a platform comprising a solid substrate and one consecutive hybrid film coating the surface of the solid substrate, where the film comprises alternating polycationic and polyanionic polymer layers; useful for identifying protein-protein interactions, for drug screening, for characterizing antibodies and in enzyme assays.

Zhou J; Zhou X Zhou J, Zhou X 8/18/2004 3/3/2005

US 20050046758 A method for transcribing biomolecular patterns,involving two-dimensionally arranging biomolecules ona board, forming a thin-film layer made of an inorganic substance on the biomolecules, forming a supporting layer on the thin film layer, and peeling the thin film layer and the supporting layer off of the biomolecules together;useful for manufacturing biochips or devices usingquantum dots or photonic crystals.

Omron KK (Tokyo); Aoyama S;Matsushita T; Nisjikawa T;Norioka S; Tsuda Y; Wazawa T

Aoyama S,Matsushita T, Nisjikawa T,Norioka S, Tsuda Y, Wazawa T

7/29/2003 3/3/2005

US 20050048648 A medium for reformulating biological membranesthat enhances assay performance and is useful forfabricating and prolonging the shelf-life of biologicalmembrane arrays.

Fang Y; Ferrie AM Fang Y, Ferrie AM 8/29/2003 3/3/2005

US 20050048531 An array of nucleic acid probes, where each probeconsists essentially of any of 127811 fully definednucleotide sequences; useful for genetic analysis.

Affymetrix(Santa Clara, CA, USA)

Lockhart DJ,Mack DH, Mittman M

9/17/1998 3/3/2005

WO 200516869 New dendrimer compounds useful in biochips fordetecting a target compound, and for methods ofdiagnosis and biochemical analysis.

Pohang Iron and Steel Co.; Pohang University of Science and Technology Foundation(Pohang, Korea)

Choi KY, Choi YS, Hong BJ, Kwon SH, Oh SJ, Park JW,Youn TO

8/19/2003 2/24/2005

US 20050042363 A microarray with a macroporous polymer substrate,manufactured by obtaining a macroporous polymer substrate and coating a surface with the substrate. The substrates have high immobilization capacity for large biomolecules and better accessibility of analytes to the immobilized biomolecules.

Chernov BK;Gemmell MA;Golovaj B; Kukhtin AV; Yershov GM

Chernov BK,Gemmell MA,Golovaj B,Kukhtin AV,Yershov GM

8/18/2003 2/24/2005

WO 200514852 A microarray of immobilized biomolecules comprisinga surface carrying a pattern of separated regions, each containing several spots of biomolecules; for use inanalysis and diagnosis.

SusTech GmbH & Co. (Darmstadt, Germany)

Groll J, Levi S, Moeller M, Rong H

7/18/2003 2/17/2005

KR 2004094982 A method for highly concentrating a target material in a sample using a scanning probe microscope to manufacture a highly integrated nano-bioarray.

Sogang University (Seoul, Korea)

Choi JU, Chun BS, Nam YS, Oh BG

5/6/2003 11/12/2004

Source: Derwent Information, Alexandria, VA. The status of each application is slightly different from country to country. For further details, contact Derwent Information, 1725 Duke Street, Suite 250, Alexandria, VA 22314. Tel: 1 (800) DERWENT (info.derwent.com).

PAT E N T S PATENTS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Therapeutic antibody gene transferWayne A Marasco

AAV vectors containing the combined furin-site and 2A ‘self-cleaving’ peptide facilitate high-level expression of monoclonal antibodies in vivo.

The field of human antibody engineering has seen important recent advances, including de novo methods for isolating high-affinity human antibodies and the creation of transgenic mice expressing human antibodies. Yet serious bot-tlenecks in the development process remain, related primarily to the costs and time involved in producing and manufacturing human mono-clonal antibodies (mAbs) both at the preclinical and clinical scales. In this issue, Fang et al.1 pro-vide convincing evidence that in vivo therapeu-tic antibody gene transfer is indeed possible—at least at the preclinical level, potentially accelerat-ing the translation of therapeutic mAbs from bench to bedside.

There are now around 20 US Food and Drug Administration (FDA)–approved therapeutic mAbs on the market today for the treatment of cancer and of autoimmune, inflammatory and infectious diseases, and many more mAbs are in preclinical and clinical development. It has become increasingly evident over the last several years that human mAb therapies are here to stay. And why shouldn’t they be? They have a long track record of safety in human trials, the FDA is quite familiar with them and specific manufac-turing guidelines are established. In addition, as the humanization2 of immunoglobulins (IgG) increases—for example, by conversion of rodent IgGs to chimeric IgGs to CDR-grafted IgGs to ‘fully human’ IgGs—they show reduced immu-nogenicity and improved therapeutic efficacy.

The trick in mAb production has been to find ways of generating stoichiometric amounts of both the heavy and light chains in the mAb-producing cells, especially because an imbal-

ance in chain production can be toxic to the cells or can result in nonfunctional IgG chains in the supernatant that can complicate the puri-fication process (Fig. 1). Fang et al. hit upon their simple yet elegant technological advance while trying to find a solution to the limited cloning space that exists in recombinant adeno-associated virus (rAAV) vectors, an attractive vector system for achieving long-term gene transfer in vivo3. Many investigators have relied on the use of internal ribosomal entry sites (IRES) to allow cap-independent expres-sion of the second gene of interest in a bicis-

tronic cassette. However, as is well known, this method is somewhat unpredictable, is both cell and gene dependent and can often result in lower expression of the gene encoded by the second cistron4.

In the current study, Fang et al. employed a modification of the 2A self-processing sequence derived from the foot-and-mouth disease virus to express a full-length mAb from a single open reading frame driven by a single promoter. The 2A sequence mediates enzyme-indepen-dent ‘cleavage’ to separate polypeptides during the post-translation process5. The particular

Wayne A. Marasco is in the Department of Cancer Immunology & AIDS, Dana-Farber Cancer Institute, Harvard Medical School, 44 Binney Street, Boston, Massachusetts 02115, USA. e-mail: [email protected]

AAV vector HF2AL

EncapsidatedrAAV

Mouse tumormodel

Tail veininjectionof rAAV

Promotor Ab heavy chain

Start codon

Cotransfection of 293 cells

Vector purification

+ AAV2 rep/AAV8 cap+ ‘Helper’ virus functions Adv E2a, E4, VA RNAs

Stop codon

Ab light chain Poly A2A

Fc

a

b

c

d

Variable

Antigen binding sites

Constant

Lightchain

Heavychain

Furin cleavage site

Figure 1 Pathway to therapeutic antibody gene transfer. (a) The complexity of monoclonal antibody production lies in the fact that immunoglobulins of the gamma family, the most commonly used in the clinical setting, are composed of four chains: two identical heavy and two identical light chains, which heterodimerize and homodimerize in the endoplasmic reticulum to form the mature four-chain molecule of IgG. Each heavy and light chain contains an N-terminal variable region and a C-terminal constant region. (b) The rAAV8-HF2AL mAb expression system can accommodate new variable region genes for different mAbs through PCR cloning with specific PCR primers. (c) Cotransfection of 293 cells with AAV-HF2AL vector and plasmids encoding AAV structural genes and adenovirus helper functions is followed by purification steps to produce rAAV particles. (d) rAAV is injected intravenously into mice bearing subcutaneous human or mouse tumors, and therapeutic levels of mAbs are produced. B

ob C

rimi

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


sequence used leads to cleavage between the last two amino acids at its C terminus. One of the tested configurations (heavy chain–2A sequence–light chain (H2AL)) yielded equiva-lent amounts of both heavy and light chains, with a C-terminal 21-amino-acid ‘tail’ on the heavy chain, as expected, and the secreted anti-bodies were functional.

Extending the work of others who have used furin cleavage sites to process heterologous polypeptides6, the investigators showed that addition of a four-amino-acid furin cleavage site immediately proximal to the 2A sequence (HF2AL) resulted in a second cleavage event, so that the final mature heavy chain contained only two extra amino acids at its C terminus both in vitro and in vivo. Impressively, this configura-tion allowed higher levels of secretion of func-tional antibodies.

These experimental findings alone provide a useful advance in the field of antibody engi-neering. However, Fang et al. went one impor-tant step further by incorporating their single mAb expression cassette into a newly reported rAAV serotype 8 vector. Recently isolated from rhesus macaques, this vector has the capacity to transduce hepatocytes and skeletal cells with very high efficiency when delivered directly by portal vein injection or intravenous infusion7,8. By combining these experimental systems, the investigators achieved remarkably high levels (>1,000 ug/ml) and long-term expression (>140 days) of an anti-VEGFR2 mAb in mice and demonstrated therapeutic efficacy against two tumor cell lines in two mouse tumor models.

The significance of this study for the field of therapeutic mAb development is manifold. First, the rAAV8-HF2AL mAb expression sys-tem should provide investigators with a rapid and relatively straightforward way to evaluate potential therapeutic mAbs with less cost and labor than are required to produce sufficient mAbs in vitro for eventual in vivo testing in animal models of disease, at least at the early stages of the discovery process. This applies most readily to mAbs that have attained lead-candidate status through in vitro biological studies. However, it is easy to imagine that the system could be adapted to accept new variable-region genes from human single-chain antibody (scFv) and Fab libraries that have been selected against a target protein of interest.

Second, this system could be adapted to eval-uate combination mAb therapies—an area of active investigation, particularly for cancer and infectious diseases. In this experimental setting, one could test important and underexplored questions concerning the initial timing and relative dosing of two or more mAbs.

Third, as the authors imply, the system may facilitate the manufacture of therapeutic mAbs

by providing a way of generating cell lines that produce high-titer, stable antibodies.

Before this elegant work can advance the cause of in vivo therapeutic antibody gene transfer, however, several important obstacles must be overcome. First, chronic diseases such as cancer and HIV-1/AIDS, for which mAb therapies hold great promise, are characterized by genetic instability, and it is well established that immune pressure can lead to phenotypically altered tumors or viral escape mutants, respec-tively. Whether high levels of circulating mAbs in vivo would prevent or accelerate this process remains to be determined.

A second concern is the immune response to the vector. Heterologous antisera raised against AAV serotypes 1–6 are not neutralizing against AAV8, and therefore prior immunity in humans to AAV serotypes 1–6 is unlikely to interfere with in vivo gene transfer9. Nevertheless, the high pro-miscuity of AAV8 gene transfer in vivo suggests that the consequences of potentially transducing unintended populations of cells must be com-pletely evaluated from the safety standpoint.

Furthermore, the use of tissue-specific promoters to restrict gene expression should be explored.

Finally, the 23-amino-acid 2A self-processing peptide, although cleaved, is a foreign sequence. Processing and presentation of this sequence by major histocompatibility complexes class I and II could still occur and prove detrimental to the host.

Nevertheless, although many questions remain, this study offers new tools and avenues of investigation that should accelerate the pro-cess of therapeutic mAb discovery and shorten the time needed to reach the clinic.

1. Fang, J. et al. Nat. Biotechnol. 23, 584–590 (2005).

2. Lobo, E.D., Hansen R.J. & Balthasar J.P. J. Pharm. Sci. 93, 2645–2668 (2004).

3. Flotte, T.R. Gene Ther. 11, 805–810 (2004).4. de Felipe, P. Genet. Vaccines & Ther. 2, 13 (2004).5. de Felipe, P. & Ryan, M.D. Traffic 5, 616–626

(2004).6. Gaken, J. et al. Gene Ther. 7, 1979–1985 (2000).7. Gao, G.-P. et al. Proc. Natl. Acad. Sci. USA 99,

11854–11859 (2002).8. Nakai, H. et al. J. Virol. 79, 214–224 (2005).9. Jooss, K. & Chirmule, N. Gene Ther. 10, 955–963

(2003).

Bringing amyloid into focusTodd E Golde & Brian J Bacskai

Amyloid deposits can be rapidly detected in the brains of living mice using a novel ligand and near-infrared fluorescence imaging.

Until recently using clinical imaging tech-nologies such as positron emission tomogra-phy (PET) and magnetic resonance imaging (MRI), the amyloid plaques that accumulate in the brains of patients with Alzheimer dis-ease have been difficult, if not impossible, to detect in vivo. In this issue Hintersteiner et al.1 describe a different approach to imaging amy-loid. Using a near infrared (NIR) fluorescence probe that crosses the blood-brain barrier and binds amyloid plaques in the brains of mice, the amount of amyloid can be cost-effectively estimated using near infrared fluorescence imaging. Eventually such an approach may be adapted to visualize amyloid in humans.

The ability to image disease processes in living humans is one of the major technologic advances of modern medicine. In the context of disease management, in vivo imaging is one of the many, and often most informative, modalities that can be used to diagnose diseases and evaluate treatment outcomes. Largely because of costs, imaging is less commonly used to predict risk for the development of disease in asymptomatic individuals. Although not always thought of as such, tests that rely on imaging are fundamen-tally biomarker studies. As with any biomarker assay, the utility of such tests depends on the sensitivity and specificity of the biomarker and on the sensitivity and specificity of the test used to measure that biomarker. The former issue is extremely important to recognize. No matter how good the assay, its predictive ability is only as good as the predictive ability of the biomarker being studied.

Deposition of the amyloid β-peptide into a fibrillar β-sheet structure referred to as amyloid is a diagnostic hallmark of the post-mortem Alzheimer-disease brain. In Alzheimer disease, amyloid β deposits as amyloid in senile plaques

Todd E. Golde is in the Department of Neuroscience, Mayo Clinic, Mayo Clinic College of Medicine, 4500 San Pablo Road, Jacksonville, Florida 32224, USA, and Brian J. Bacskai is in the Alzheimer’s disease Research Unit, Mass. General Hospital, 114 16th St. Charlestown, Massachusetts 02129, USA.e-mail: [email protected] and [email protected]

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


sequence used leads to cleavage between the last two amino acids at its C terminus. One of the tested configurations (heavy chain–2A sequence–light chain (H2AL)) yielded equiva-lent amounts of both heavy and light chains, with a C-terminal 21-amino-acid ‘tail’ on the heavy chain, as expected, and the secreted anti-bodies were functional.

Extending the work of others who have used furin cleavage sites to process heterologous polypeptides6, the investigators showed that addition of a four-amino-acid furin cleavage site immediately proximal to the 2A sequence (HF2AL) resulted in a second cleavage event, so that the final mature heavy chain contained only two extra amino acids at its C terminus both in vitro and in vivo. Impressively, this configura-tion allowed higher levels of secretion of func-tional antibodies.

These experimental findings alone provide a useful advance in the field of antibody engi-neering. However, Fang et al. went one impor-tant step further by incorporating their single mAb expression cassette into a newly reported rAAV serotype 8 vector. Recently isolated from rhesus macaques, this vector has the capacity to transduce hepatocytes and skeletal cells with very high efficiency when delivered directly by portal vein injection or intravenous infusion7,8. By combining these experimental systems, the investigators achieved remarkably high levels (>1,000 ug/ml) and long-term expression (>140 days) of an anti-VEGFR2 mAb in mice and demonstrated therapeutic efficacy against two tumor cell lines in two mouse tumor models.

The significance of this study for the field of therapeutic mAb development is manifold. First, the rAAV8-HF2AL mAb expression sys-tem should provide investigators with a rapid and relatively straightforward way to evaluate potential therapeutic mAbs with less cost and labor than are required to produce sufficient mAbs in vitro for eventual in vivo testing in animal models of disease, at least at the early stages of the discovery process. This applies most readily to mAbs that have attained lead-candidate status through in vitro biological studies. However, it is easy to imagine that the system could be adapted to accept new variable-region genes from human single-chain antibody (scFv) and Fab libraries that have been selected against a target protein of interest.

Second, this system could be adapted to eval-uate combination mAb therapies—an area of active investigation, particularly for cancer and infectious diseases. In this experimental setting, one could test important and underexplored questions concerning the initial timing and relative dosing of two or more mAbs.

Third, as the authors imply, the system may facilitate the manufacture of therapeutic mAbs

by providing a way of generating cell lines that produce high-titer, stable antibodies.

Before this elegant work can advance the cause of in vivo therapeutic antibody gene transfer, however, several important obstacles must be overcome. First, chronic diseases such as cancer and HIV-1/AIDS, for which mAb therapies hold great promise, are characterized by genetic instability, and it is well established that immune pressure can lead to phenotypically altered tumors or viral escape mutants, respec-tively. Whether high levels of circulating mAbs in vivo would prevent or accelerate this process remains to be determined.

A second concern is the immune response to the vector. Heterologous antisera raised against AAV serotypes 1–6 are not neutralizing against AAV8, and therefore prior immunity in humans to AAV serotypes 1–6 is unlikely to interfere with in vivo gene transfer9. Nevertheless, the high pro-miscuity of AAV8 gene transfer in vivo suggests that the consequences of potentially transducing unintended populations of cells must be com-pletely evaluated from the safety standpoint.

Furthermore, the use of tissue-specific promoters to restrict gene expression should be explored.

Finally, the 23-amino-acid 2A self-processing peptide, although cleaved, is a foreign sequence. Processing and presentation of this sequence by major histocompatibility complexes class I and II could still occur and prove detrimental to the host.

Nevertheless, although many questions remain, this study offers new tools and avenues of investigation that should accelerate the pro-cess of therapeutic mAb discovery and shorten the time needed to reach the clinic.

1. Fang, J. et al. Nat. Biotechnol. 23, 584–590 (2005).

2. Lobo, E.D., Hansen R.J. & Balthasar J.P. J. Pharm. Sci. 93, 2645–2668 (2004).

3. Flotte, T.R. Gene Ther. 11, 805–810 (2004).4. de Felipe, P. Genet. Vaccines & Ther. 2, 13 (2004).5. de Felipe, P. & Ryan, M.D. Traffic 5, 616–626

(2004).6. Gaken, J. et al. Gene Ther. 7, 1979–1985 (2000).7. Gao, G.-P. et al. Proc. Natl. Acad. Sci. USA 99,

11854–11859 (2002).8. Nakai, H. et al. J. Virol. 79, 214–224 (2005).9. Jooss, K. & Chirmule, N. Gene Ther. 10, 955–963

(2003).

Bringing amyloid into focusTodd E Golde & Brian J Bacskai

Amyloid deposits can be rapidly detected in the brains of living mice using a novel ligand and near-infrared fluorescence imaging.

Until recently using clinical imaging tech-nologies such as positron emission tomogra-phy (PET) and magnetic resonance imaging (MRI), the amyloid plaques that accumulate in the brains of patients with Alzheimer dis-ease have been difficult, if not impossible, to detect in vivo. In this issue Hintersteiner et al.1 describe a different approach to imaging amy-loid. Using a near infrared (NIR) fluorescence probe that crosses the blood-brain barrier and binds amyloid plaques in the brains of mice, the amount of amyloid can be cost-effectively estimated using near infrared fluorescence imaging. Eventually such an approach may be adapted to visualize amyloid in humans.

The ability to image disease processes in living humans is one of the major technologic advances of modern medicine. In the context of disease management, in vivo imaging is one of the many, and often most informative, modalities that can be used to diagnose diseases and evaluate treatment outcomes. Largely because of costs, imaging is less commonly used to predict risk for the development of disease in asymptomatic individuals. Although not always thought of as such, tests that rely on imaging are fundamen-tally biomarker studies. As with any biomarker assay, the utility of such tests depends on the sensitivity and specificity of the biomarker and on the sensitivity and specificity of the test used to measure that biomarker. The former issue is extremely important to recognize. No matter how good the assay, its predictive ability is only as good as the predictive ability of the biomarker being studied.

Deposition of the amyloid β-peptide into a fibrillar β-sheet structure referred to as amyloid is a diagnostic hallmark of the post-mortem Alzheimer-disease brain. In Alzheimer disease, amyloid β deposits as amyloid in senile plaques

Todd E. Golde is in the Department of Neuroscience, Mayo Clinic, Mayo Clinic College of Medicine, 4500 San Pablo Road, Jacksonville, Florida 32224, USA, and Brian J. Bacskai is in the Alzheimer’s disease Research Unit, Mass. General Hospital, 114 16th St. Charlestown, Massachusetts 02129, USA.e-mail: [email protected] and [email protected]

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


tions on the information that this approach can provide. For example, it is unlikely that PET scans will ever be able to detect individual plaques or even to provide precise local-ization of amyloid deposition within structural subregions of the brain.

Other studies suggest that combining MRI with an amy-loid ligand labeled with 1H or 19F may enable visualization of amyloid plaques6–8. The most recent report shows that MRI can detect an 19F-labeled amyloid dye known as FSB and that FSB labeling is supe-rior to other MRI methods for detecting amyloid plaques in living mice9. Although such approaches provide more spatial resolution than PET studies and avoid the use of radioactive isotopes with short half-lives, they are still subject to some of the techni-cal limitations of traditional MRI studies and, at least with current ligands, to suboptimal signal-to-noise levels.

Hintersteiner et al. describe an alternative, all-optical approach to imaging an amyloid ligand in vivo. Using a novel fluores-cent amyloid-binding dye, AOI987 (Fig. 1a), which absorbs and emits in the NIR spectrum, they visualize amyloid load in the brain of living mice using NIR fluorescence imaging (Fig. 1b). Fluorescent imaging in live animals with shorter wavelengths of light is severely limited because of tissue autofluorescence and scattering. However, NIR imaging solves this problem by reducing the background and scat-tering through biological tissue10. Although the spatial resolution in the work of Hintersteiner et al. is limited, the authors do demonstrate that the fluorescence signal intensity increases with increasing plaque load in the mice and that amyloid deposition can be detected in mice as young as 9 months. Moreover images could be acquired in short time periods ranging from 0.5 to 3.0 seconds, allowing measurement at multiple time points after dosing to calculate specific binding.

Despite the lack of spatial resolution, the ability to noninvasively quantify amyloid β deposited as amyloid in living mice using NIR imaging is quite exciting. Unlike PET and MRI imaging technologies, which require significant hardware investments, highly skilled operators,

Amyloid-bindingligands

Intravenousinjection

of AOI987

ImagingNear-infrared

light

CCDcamera

a

b

HO

F

FSB

OH

HO

O

O

HO

AOI987

O N

O

BF4-

N

N

N

O

O

O

+

S

N

11CH3HO

PIB

N

SN

H

Congo red

NH2

H2NH

H

O

O

O

O

O

O

N

NN

S

S

N N

Figure 1 Amyloid ligands and near infrared fluorescence imaging of plaques. (a) Examples of amyloid-binding ligands. (b) As described by Hintersteiner et al., the near-infrared probe AOI987, injected systemically in mice, crosses the blood-brain barrier and specifically labels amyloid plaque. Only a light source, appropriate filters and a sensitive charge-coupled-device camera are required to measure the near-infrared fluorescence emitted by the amyloid-binding probe.

and, more variably, in cerebral vessels. The amy-loid in senile plaques forms a spherical core that ranges from ~2 to ~200 µm, but is typically 20–60 µm in diameter. Accumulation of amyloid β in the brain is also hypothesized to be the cause of Alzheimer disease, although there is a great deal of debate as to whether the toxic form is amyloid β that accumulates as visible amyloid deposits or as smaller soluble extracellular or intracellular aggregates. In any case, there is a great deal of evidence from genetic, biochemi-cal, pathologic and animal modeling studies to support the hypothesis that accumulation of amyloid β in the brain is the initiating event in Alzheimer-disease pathogenesis2. Thus, in the-ory, amyloid β deposition as amyloid appears to be a good diagnostic biomarker for Alzheimer disease and may also be a good predictive bio-marker of the disease.

Given the potential of amyloid as a bio-marker, a number of groups have been developing approaches to visualize plaques. Conceptually, the simplest approach has been to use MRI to try to directly visualize amyloid plaques. Despite recent impressive technologi-cal advances in MRI resolution, this approach remains challenging because of the plaques’ small size. Several recent reports suggest that this may now be feasible, at least in transgenic mouse models and in mouse and human tis-sue slices3,4. However, only large plaques (>50 µM) can currently be visualized, and the scans require hours to complete. Thus, additional advances in MRI technology would be needed to translate this approach to humans or even to make it more widely accessible for preclinical studies of amyloid β deposition in mice.

Compounds such as Congo red (Fig. 1a) and thioflavin S have been used for many years to recognize amyloid deposits in post-mortem tissue. Thus, another approach to visualizing amyloid β deposited in vivo is to modify known amyloid-binding compounds or identify new ones. The most advanced of these ligand-based approaches uses Pittsburgh compound B (PIB) to visualize plaque burden in living humans with PET5. PIB is a thiofla-vin derivative labeled with 11C so that it can be used as a PET probe (Fig. 1a). PIB is the only amyloid probe currently used to image amyloid in Alzheimer disease patients, but because of its only recent approval for humans and somewhat limited availability, the clinical utility of PET imaging with PIB has not been definitively established.

Although PIB has been used to image amyloid plaque in humans in vivo and in mice tissue ex vivo, microPET with PIB has not been able to detect amyloid in the brains of living Alzheimer-disease mouse models. Moreover, the inherent resolution limits of PET impose certain restric-

and, in the case of PET, short-lived radioligands, in vivo NIR imaging as described in this study is quite inexpensive and could be set up in virtu-ally any laboratory.

It should also be possible to improve the spatial resolution. NIR imaging using the more elaborate techniques and equipment of opti-cal tomography (which was not done in this study) has attained a spatial resolution of 1 mm or better, approximately the resolution of PET imaging11. It may also be possible to improve upon the fluorescent signal by improving the amyloid probe. For example, an amyloid dye that fluoresced more intensely, changed spectra, or both, upon binding to amyloid could reduce background and improve contrast.

Currently, the major constraint on NIR fluo-rescence imaging is the paucity of fluorescent probes. Probes described in the literature are ligands tagged with NIR fluorophores. Thus, on a more general level, the finding of Hintersteiner et al. is rather exciting simply because they identify an NIR fluorescent ligand that binds the biomarker of interest. This proof-of-concept study will likely engage others in the search for better NIR imaging agents for amy-loid and other targets.

Bob

Crim

i

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Reverse engineering gene regulatory networksAlexander J Hartemink

An information theoretic algorithm that prunes away potentially indirect interactions allows for improved reconstruction of biological networks.

Alexander J. Hartemink is in the Department of Computer Science, Duke University, Box 90129, Durham, North Carolina 27708-0129, USA.e-mail: [email protected]

NIR imaging has been used for many years to look at functional parameters in the human brain (for example, saturations of hemoglo-bin)12,13. Although it is unlikely that NIR imag-ing will permit imaging of the entire brain, it is possible to image several centimeters below the human skull, potentially enabling, for example, detection of amyloid in the cortex. If amyloid imaging were used to diagnose Alzheimer dis-ease, it is likely that simply detecting plaques in areas of the brain known to be affected by Alzheimer disease would be good enough.

The development of amyloid probes that can be imaged in vivo is almost certain to expedite the preclinical and clinical evaluation of novel Alzheimer-disease therapeutics that target amy-loid β. Such ligands may also be useful in the diagnosis of atypical Alzheimer-disease cases. But current clinical diagnosis of Alzheimer dis-ease is reasonably accurate, so it is unlikely that amyloid imaging will become a routine diagnos-tic modality unless it were relatively quick, safe and inexpensive. With further advances in the technology and ligands, NIR imaging of amy-loid may fulfill these criteria14.

There is evidence to suggest that amy-loid deposition predates the clinical signs of Alzheimer disease by years or even decades; however, the exact temporal relationship between amyloid deposition and cognitive dys-function remains to be established. The utility of

existing amyloid probes for detecting very early stages of amyloid deposition in the brain of humans has not yet been determined, although most believe that significant improvements in sensitivity will be needed. As it is almost certain that Alzheimer disease will be easier to prevent than treat, a refined version of current amyloid imaging methods may ultimately be the diag-nostic tool used to determine both who needs prophylactic treatment and when that treatment should be initiated.

1. Hintersteiner, M. et al. Nat. Biotechnol. 23, 577–583 (2005).

2. Golde, T.E. J. Clin. Invest. 111, 11–18 (2003).3. Jack, C.R. Jr. et al. Magn. Reson. Med. 52, 1263–1271

(2004).4. Lee, S.P., Falangola, M.F., Nixon, R.A., Duff, K. &

Helpern, J.A. Magn. Reson. Med. 52, 538–544 (2004).

5. Klunk, W.E. et al. Ann. Neurol. 55, 306–319 (2004).6. Poduslo, J.F. et al. Biochemistry 43, 6064–6075

(2004).7. Skovronsky, D.M. et al. Proc. Natl. Acad. Sci. USA 97,

7609–7614 (2000).8. Wadghiri, Y.Z. et al. Magn. Reson. Med. 50, 293–302

(2003).9. Higuchi, M. et al. Nat. Neurosci. 8, 527–533 (2005).10. Frangioni, J.V. Curr. Opin. Chem. Biol. 7, 626–634

(2003).11. Graves, E.E., Ripoll, J., Weissleder, R. & Ntziachristos, V.

Med. Phys. 30, 901–911 (2003).12. Strangman, G., Boas, D.A. & Sutton, J.P. Biol. Psychiatry

52, 679–693 (2002).13. Pouratian, N. et al. Magn. Reson. Med. 47, 766–776

(2002).14. Skoch, J., Dunn, A., Hyman, B.T. & Bacskai, B.J.

J. Biomed. Opt. 10, 011007 (2005).

Biological systems are wondrously and notori-ously complex. Over the last fifty years, mole-cular biology has helped to reveal the vast and stunning array of components in biological systems. Now, we face the even more daunt-ing challenge of systems biology: determining how all these puzzle pieces come together to create living systems. A recent paper by Basso et al.1 published in Nature Genetics describes a statistical algorithm for more compactly and

more accurately reverse engineering networks describing pair-wise interactions among genes and thin protein products. The network they recover from gene expression profiles of a variety of human B-cell populations suggests that the B-cell regulatory network has both a scale-free and hierarchical architecture, implying the presence of a few ‘hubs’ that are highly connected and preferentially connected to one another.

Reverse engineering is the process of eluci-dating the structure of a system by reasoning backwards from observations of its behavior. In reverse engineering biological networks, one of the first hurdles to overcome is semantic. The term ‘network’ has come to mean different things throughout biology, and the semantic

overload is magnified when computational and statistical interpretations are added. Even in networks whose nodes are ostensibly the same objects (for examples, genes or their pro-tein products), the network edges can mean vastly different things and should be inter-preted with care. As just one example, edges can either be undirected (without an orienta-tion) to capture relations that are symmetric or directed (with an orientation) to capture relations that are asymmetric.

An undirected edge between two genes may indicate that the genes are coexpressed or coregulated, participate in a common pathway or regulatory ‘module’ or share a common bio-logical function, location or process; or that their protein products coprecipitate, directly bind one another, or assemble into the same complex (a problematic term in its own right). On the other hand, a directed edge between two genes may be used to represent a step in a metabolic pathway, signal transduction cascade, or stage of develop-ment; or it may indicate a causal control or a regulatory relationship.

This semantic caveat is important in trying to understand the myriad methods that have been proposed in the last decade for reverse engi-neering biological networks from system-wide data, especially gene expression data. Within this broader context, the ARACNe algorithm of Basso et al. is most closely related to an earlier method for producing ‘relevance networks’2,3. Both sets of authors use a pair-wise mutual information criterion across gene expression profiles to recover edges that are undirected, but ARACNe improves on this somewhat by using the data processing inequality to prune out interactions suspected to be indirect.

After using synthetic data to assess the accuracy of their ARACNe algorithm, Basso et al. apply it to a rather sizable set of gene expression array data, collected from human B-cell populations with a variety of pheno-types, including both normal and malignantly transformed cells at different stages in the ger-minal center reaction process, from naive cells in the mantle zone to differentiated memory or plasma cells. This results in a network with about 129,000 undirected interactions between pairs of genes. Owing to the obvi-ous complexity of such a network, the authors choose to focus on two simpler aspects: a sta-tistical summary of the (global) connectiv-ity distribution among all the nodes in the network, an approach that is quite in vogue; and a more detailed look at a specific (local) portion of the network centered around the proto-oncogene MYC, chosen both because of its clinical importance and because of the wealth of information available for corrobo-ratory purposes.

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Reverse engineering gene regulatory networksAlexander J Hartemink

An information theoretic algorithm that prunes away potentially indirect interactions allows for improved reconstruction of biological networks.

Alexander J. Hartemink is in the Department of Computer Science, Duke University, Box 90129, Durham, North Carolina 27708-0129, USA.e-mail: [email protected]

NIR imaging has been used for many years to look at functional parameters in the human brain (for example, saturations of hemoglo-bin)12,13. Although it is unlikely that NIR imag-ing will permit imaging of the entire brain, it is possible to image several centimeters below the human skull, potentially enabling, for example, detection of amyloid in the cortex. If amyloid imaging were used to diagnose Alzheimer dis-ease, it is likely that simply detecting plaques in areas of the brain known to be affected by Alzheimer disease would be good enough.

The development of amyloid probes that can be imaged in vivo is almost certain to expedite the preclinical and clinical evaluation of novel Alzheimer-disease therapeutics that target amy-loid β. Such ligands may also be useful in the diagnosis of atypical Alzheimer-disease cases. But current clinical diagnosis of Alzheimer dis-ease is reasonably accurate, so it is unlikely that amyloid imaging will become a routine diagnos-tic modality unless it were relatively quick, safe and inexpensive. With further advances in the technology and ligands, NIR imaging of amy-loid may fulfill these criteria14.

There is evidence to suggest that amy-loid deposition predates the clinical signs of Alzheimer disease by years or even decades; however, the exact temporal relationship between amyloid deposition and cognitive dys-function remains to be established. The utility of

existing amyloid probes for detecting very early stages of amyloid deposition in the brain of humans has not yet been determined, although most believe that significant improvements in sensitivity will be needed. As it is almost certain that Alzheimer disease will be easier to prevent than treat, a refined version of current amyloid imaging methods may ultimately be the diag-nostic tool used to determine both who needs prophylactic treatment and when that treatment should be initiated.

1. Hintersteiner, M. et al. Nat. Biotechnol. 23, 577–583 (2005).

2. Golde, T.E. J. Clin. Invest. 111, 11–18 (2003).3. Jack, C.R. Jr. et al. Magn. Reson. Med. 52, 1263–1271

(2004).4. Lee, S.P., Falangola, M.F., Nixon, R.A., Duff, K. &

Helpern, J.A. Magn. Reson. Med. 52, 538–544 (2004).

5. Klunk, W.E. et al. Ann. Neurol. 55, 306–319 (2004).6. Poduslo, J.F. et al. Biochemistry 43, 6064–6075

(2004).7. Skovronsky, D.M. et al. Proc. Natl. Acad. Sci. USA 97,

7609–7614 (2000).8. Wadghiri, Y.Z. et al. Magn. Reson. Med. 50, 293–302

(2003).9. Higuchi, M. et al. Nat. Neurosci. 8, 527–533 (2005).10. Frangioni, J.V. Curr. Opin. Chem. Biol. 7, 626–634

(2003).11. Graves, E.E., Ripoll, J., Weissleder, R. & Ntziachristos, V.

Med. Phys. 30, 901–911 (2003).12. Strangman, G., Boas, D.A. & Sutton, J.P. Biol. Psychiatry

52, 679–693 (2002).13. Pouratian, N. et al. Magn. Reson. Med. 47, 766–776

(2002).14. Skoch, J., Dunn, A., Hyman, B.T. & Bacskai, B.J.

J. Biomed. Opt. 10, 011007 (2005).

Biological systems are wondrously and notori-ously complex. Over the last fifty years, mole-cular biology has helped to reveal the vast and stunning array of components in biological systems. Now, we face the even more daunt-ing challenge of systems biology: determining how all these puzzle pieces come together to create living systems. A recent paper by Basso et al.1 published in Nature Genetics describes a statistical algorithm for more compactly and

more accurately reverse engineering networks describing pair-wise interactions among genes and thin protein products. The network they recover from gene expression profiles of a variety of human B-cell populations suggests that the B-cell regulatory network has both a scale-free and hierarchical architecture, implying the presence of a few ‘hubs’ that are highly connected and preferentially connected to one another.

Reverse engineering is the process of eluci-dating the structure of a system by reasoning backwards from observations of its behavior. In reverse engineering biological networks, one of the first hurdles to overcome is semantic. The term ‘network’ has come to mean different things throughout biology, and the semantic

overload is magnified when computational and statistical interpretations are added. Even in networks whose nodes are ostensibly the same objects (for examples, genes or their pro-tein products), the network edges can mean vastly different things and should be inter-preted with care. As just one example, edges can either be undirected (without an orienta-tion) to capture relations that are symmetric or directed (with an orientation) to capture relations that are asymmetric.

An undirected edge between two genes may indicate that the genes are coexpressed or coregulated, participate in a common pathway or regulatory ‘module’ or share a common bio-logical function, location or process; or that their protein products coprecipitate, directly bind one another, or assemble into the same complex (a problematic term in its own right). On the other hand, a directed edge between two genes may be used to represent a step in a metabolic pathway, signal transduction cascade, or stage of develop-ment; or it may indicate a causal control or a regulatory relationship.

This semantic caveat is important in trying to understand the myriad methods that have been proposed in the last decade for reverse engi-neering biological networks from system-wide data, especially gene expression data. Within this broader context, the ARACNe algorithm of Basso et al. is most closely related to an earlier method for producing ‘relevance networks’2,3. Both sets of authors use a pair-wise mutual information criterion across gene expression profiles to recover edges that are undirected, but ARACNe improves on this somewhat by using the data processing inequality to prune out interactions suspected to be indirect.

After using synthetic data to assess the accuracy of their ARACNe algorithm, Basso et al. apply it to a rather sizable set of gene expression array data, collected from human B-cell populations with a variety of pheno-types, including both normal and malignantly transformed cells at different stages in the ger-minal center reaction process, from naive cells in the mantle zone to differentiated memory or plasma cells. This results in a network with about 129,000 undirected interactions between pairs of genes. Owing to the obvi-ous complexity of such a network, the authors choose to focus on two simpler aspects: a sta-tistical summary of the (global) connectiv-ity distribution among all the nodes in the network, an approach that is quite in vogue; and a more detailed look at a specific (local) portion of the network centered around the proto-oncogene MYC, chosen both because of its clinical importance and because of the wealth of information available for corrobo-ratory purposes.

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Figure 1 Comparison of the performance of ARACNe and a Bayesian network inference algorithm in reverse engineering a synthetic gene regulatory network. (a) The synthetic gene regulatory network used for assessing the reconstruction accuracy of network inference algorithms7. The network has 8 disconnected nodes serving as a negative control (not shown) and 12 interconnected nodes, including a cyclic loop formed by nodes 0, 2, 3 and 7 (regulatory interactions between two genes (nodes) are shown as arrows (edges); black and green arrows represent up- and downregulation, respectively). Reverse engineering using a Bayesian network inference algorithm to recover a dynamic Bayesian network (DBN) on the full data set results in a reconstructed network with 100% accuracy, as reported earlier7 (blue arrows indicate correct edges with correct orientation; no incorrect edges were recovered). Reverse engineering using ARACNe on the full data set results in a reconstructed network with the same 13 correct edges as the DBN reconstruction, but without orientations; it includes two incorrect edges, between nodes 3 and 4 and nodes 7 and 8 (correct and incorrect edges are represented by blue and red, respectively; the ARACNe network is reproduced from Basso et al.). (b) Performance of ARACNe and a Bayesian network inference algorithm on subsets of the full data set. Sensitivity and precision are plotted as a function of the number of samples used for the analysis. At roughly the same sensitivity, the Bayesian network inference algorithm appears to exhibit better precision over a wide range of sample sizes (the ARACNe plots are reproduced from Basso et al.).

DBN sensitivity

Sen

sitiv

ity a

nd p

reci

sion

(%

)

Sample size400100

80

60

40

20

0

100Syntheticnetwork

a bDynamic Bayesian

networkARACNenetwork

800 1400 2000

ARACNe sensitivity ARACNe precision

DBN precision

0

7

3

2

1

4

5

8 9

6

10

11

0

7

3

2

1

4

5

8 9

6

10

11

0

7

3

2

1

4

5

8 9

6

10

11

Analyzing the connectivity distribution of a network is currently popular for two reasons. First, it is a sensible first step in reasoning about networks so large that they are difficult to under-stand otherwise; for all intents and purposes, the interactions recovered by a tool like ARACNe are impossible to visualize directly in a way that facilitates insight. Second, articles and books4 suggesting that many kinds of networks—bio-logical, social, and engineered—are scale-free have recently been published in a flurry. Indeed, the network recovered by ARACNe from B-cell expression profiles has a connectivity distribu-tion that suggests that it, too, is scale-free. Basso et al. appropriately caution that the reported connectivity distribution is not conclusive because an explainable saturation occurs in the ‘low interaction count’ portion of the curve, resulting in a distribution that is scale-free over only one order of magnitude. Nevertheless, the results are consistent with a hypothesis that this network is scale-free.

As for the subnetwork centered around MYC, it contains 56 genes adjacent to MYC, termed ‘first neighbors,’ along with 2,007 genes adjacent to these first neighbors, termed ‘second neigh-bors.’ Even a comparatively small subnetwork of this size is still a challenge to visualize insight-fully, so the authors assess its quality in two ways. First, they determine whether the genes of the subnetwork are enriched for specific cel-lular process categories in the Gene Ontology database5, which they are. Second—and this is a wonderful strength of the paper—the authors experimentally validate some of the first neigh-bors of MYC.

The list of MYC first neighbors was pruned to exclude those with lowest mutual information scores, those that do not contain MYC binding sites near the transcription initiation site, and those already known to be bound directly by MYC. The remaining 12 genes were tested for direct MYC binding using a standard chromatin immunoprecipitation assay, and 11 predictions were positive. Although the authors’ resultant claim of over 90% specificity for ARACNe is perhaps optimistic as they excluded predic-tions with lowest mutual information scores and, more important, predictions not known to contain a MYC binding site, the results are still extremely encouraging. The success of this kind of experimental validation lends credence both to ARACNe and also to computational approaches more generally.

In closing, two further points should be made. First, this paper provides evidence confirming a simple intuition that many in this field have had, namely that gene expression data need not necessarily be collected from perturbation experiments for reverse engineering to be suc-cessful. Although perturbation experiments are certainly useful for network inference, they are also costly, and in some cases infeasible for either technical or ethical reasons. Basso et al. demonstrate that as long as the available data explore a wide range in the ‘expression space’ of the system, biologically meaningful interactions can be recovered by computational algorithms.

Second, the authors of this paper should be commended for evaluating the performance of ARACNe on synthetic data6, and indeed, it performs nobly. However, they seem to misrep-

resent the performance of Bayesian networks on the same synthetic data. They report that ARACNe offers “substantially higher precision” in comparison with Bayesian networks, whereas we have observed exactly the opposite (Fig. 1). The discrepancy is most likely due to the fact that Basso et al. used a static Bayesian net-work in place of a more appropriate dynamic Bayesian network. This is only a minor quibble because the B-cell expression profiles examined in the remainder of the study are of quite a dif-ferent character from the synthetic expression data in many regards, and it is not clear which method would be best suited to network infer-ence in the B-cell context. Indeed, given the earlier caveat that the networks recovered by these and other methods typically have differ-ent semantics, it is likely that multiple methods will be needed to completely understand the regulation and dysregulation of B-cell differ-entiation, as well as other similar problems in systems biology.

1 Basso, K. et al. Nat. Gen. 37, 382–390 (2005).2. Butte, A.J. & Kohane, I.S. Pac. Symp. Biocomput.

2000, 418–429 (2000).3. Butte, A.J., Tamayo, P., Slonim, D., Golub, T.R. &

Kohane, I.S. Proc. Natl. Acad. Sci. USA 97, 12182–12186 (2000).

4. Barabasi, A.L. Linked: How Everything Is Connected to Everything Else and What It Means (Plume, New York, NY, 2003).

5. The Gene Ontology Consortium. Gene Ontology: tool for the unification of biology. Nat. Gen. 25, 25–29 (2000).

6. Smith, V.A., Jarvis, E.D. & Hartemink, A.J. Bioinformatics 18, S216–S224 (2002).

7. Yu, J., Smith, V.A., Wang, P.P., Hartemink A.J. & Jarvis, E.D. in 3rd International Conference on Systems Biology (Karolinska Institute, Stockholm, Sweden, 2002).

Bob

Crim

i

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Tunable antibodiesLouis M Weiner & Paul Carter

A recent Keystone meeting highlighted the progress in generating effective anti-cancer antibodies by manipulating antibody binding affinity and their ability to support ADCC.

Monoclonal antibodies (mAbs) have become the protein therapeutics of choice for targeting cancer, and increasingly for other indications. Recent setbacks involving the withdrawal of Biogen Idec’s (Cambridge, MA, USA) IgG4 mAb (Tysabri, natalizumab) for use in mul-tiple sclerosis, however, serve to emphasize the incompleteness of our knowledge of the functional consequences of mAb-mediated target binding. In addition, for those mAbs (e.g., IgG1 isotype) that trigger antibody-dependent cellular cytotoxicity (ADCC)—the process by which immune cells (e.g., natural killer (NK) cells, macrophages and neutro-phils) are recruited to kill target cells such as tumors—the biological and therapeutic implications of enhancing cytotoxicity remain unclear. Several of many excellent presenta-tions at a recent Keystone meeting in Santa Fe on antibody-based therapeutics for cancer1 emphasized progress in our ability to modify different domains of mAbs and mAb frag-ments to both influence target affinity and modify ADCC.

Alteration of the mAb antigen-combining site by site-directed mutagenesis2 or random mutagenesis with yeast surface display3,4 is now widely used to increase antibody affinity for target antigens. There is also the possibility that these approaches will be supplemented by other novel approaches. One such approach may be to exploit small-molecule specificity and capacity by modifying the IgG structure to contain defined chemical groups with their own binding and biologic properties. This reprograms the mAb’s exquisite target-ing power while retaining the advantages of IgG with respect to pharmacokinetics, mechanisms of clearance and a larger surface of interaction with the target of the chemical entity than is possessed by the unmodified small molecule (Carlos Barbas III; Scripps Research Institute, La Jolla, CA, USA).

Although the range of protein-engineering approaches available for increasing affinity of mAbs to their targets continues to grow, much work remains in relating affinity to biologi-cal and therapeutic potency. To date, most work has proceeded with the assumption that higher affinity antibodies, by virtue of prolonged tumor retention, will have superior tumor targeting and efficacy properties.

This dogma was challenged many years ago by Weinstein et al.5, who proposed and demonstrated the existence of a binding site barrier that impedes the penetration of anti-bodies into tumor masses because durable, high-affinity interactions between the anti-body and its target block the diffusion of such antibodies throughout the tumor mass. This hypothesis was supported and extended sev-eral years ago by Adams et al.6, who showed that single-chain (sc)Fv molecules target-ing HER2/neu, a member of the epidermal growth factor receptor (EGFR) family, with high affinities exhibited unimproved quanti-tative tumor targeting, less tumor targeting

specificity and diminished penetration into solid tumor masses than did their lower-affinity counterparts. As revealed at the meeting, it is now becoming clear that IgG molecules derived by James Marks (University of California, San Francisco) from the same scFv molecules used in the earlier report6 show similar results with respect to tumor retention and tumor targeting specificity (Gregory Adams; Fox Chase Cancer Center, Philadelphia). This indicates that the impact of binding site affinity on tumor targeting is not related to the antigen-binding format (e.g., scFv versus IgG), an observation with broad potential significance as all antibod-ies approved for use in oncology and most in clinical development are in IgG format.

Similarly, high-affinity variants of an anti-carcinoembryonic antigen (CEA) scFv have been generated by yeast surface display7. As with the HER2/neu targeting system described by Adams et al.6, high-affinity scFvs do not target CEA-expressing tumors any more effi-ciently than their low-affinity counterparts

Louis M. Weiner is at the Fox Chase Cancer Center, 333 Cottman Avenue, Philadelphia, PA 19111, USA, and Paul Carter is at Seattle Genetics Corporation, 21823 30th Drive SE, Bothell, Washington 98021, USA. email: [email protected] and [email protected]

Increase affinity of antigen binding site

Antigenbinding site

Antigenbinding site

Fab

Fc

Variable

Constant• Impede tumor penetration• Indirectly improve ADCC• Affect cellular internalization

Increase affinity of Fc receptor binding• Modify pharmacokinetics (FcRn)• Improve ADCC (FcγRIII)

Lightchain

Heavychain

Figure 1 Antibody structure can be manipulated to fit an intended therapeutic application. When antibodies are used to inhibit signaling by blocking ligand-receptor interactions, desirable properties will include high affinity for the tumor antigen, prolonged blood residence (which can be accomplished by increased binding to FcRn); the capacity to mediate ADCC may or may not be critical. When the intended therapeutic mechanism is ADCC, high affinity, multivalent binding to the tumor antigen, high-affinity binding to activating Fc receptors, diminished binding to inhibitory Fc receptors, and prolonged blood residence are desirable; the capacity to mediate signaling by the target cell may be important as well. When antibodies are used as carrier vehicles for immunoconjugates, a moderate affinity for the tumor target is desirable as this is associated with improved tumor penetration. The choice of target antigen and epitope on that antigen are important determinants of internalization, which is frequently required for optimal therapeutic efficacy of immunoconjugates. In contrast to unconjugated antibodies, a shorter blood residence time is frequently desirable, and can be achieved either through the use of antibody fragments or by engineering IgG antibodies with low affinity binding to FcRn.

Bob

Crim

i

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


(K. Dane Wittrup; Massachusetts Institute for Technology, Cambridge, MA, USA; Kerry Chester and Richard Begent; Royal Free Hospital, London). Wittrup postulates that antibody consumption by tumor cells, in addition to high antigen-binding affinity, can pose a major ‘barrier’ to tumor penetration. Antibody consumption depends upon the rates of internalization and catabolism versus recycling of the antibody/antigen complex to the cell surface. These possibilities are under active investigation by experimentation and mathematical modeling.

These observations have potentially impor-tant implications for the design of therapeutic antibodies. Thus, when rapid and complete tumor penetration is essential (e.g., for immunotoxins), a high-affinity binding interaction may be undesirable. In other situations, however, high-affinity binding to tumor antigens can indeed be useful. Work in one of our laboratories (Louis Weiner; Fox Chase Cancer Center, Philadelphia) has shown that both high-affinity binding and multivalency contribute to the ability of antibodies to mediate ADCC. Bispecific anti-bodies targeting HER2/neu and FcγRIII are more efficient mediators of ADCC against HER2/neu-expressing tumor cell lines when the affinity for HER2/neu is higher and when such binding is multivalent.

Focus is extending from the antigen-bind-ing variable domains of mAb molecules, to their Fc region, which directly participates in activating complement via the classical path-way and in recruiting immune cells in ADCC. For example, CD20 is a tetra-spanning mem-brane protein expressed on malignant B cells (but not on plasma cells) that is the target for Genentech (S. San Francisco, CA, USA)/Biogen Idec’s (San Diego, CA, USA) ritux-imab (Rituxan) for B-cell non-Hodgkin’s lymphoma. Fully human antibodies have been identified that are even more potent than rituximab in complement-mediated cytotoxicity assays (Martin Glennie; Tenovus Research Laboratories, Southampton, UK). Preliminary clinical results in lymphoma patients seem quite promising, with objec-tive clinical responses described by Jan van de Winkel (Genmab, Utrecht, Netherlands).

Manipulations of mAb Fc regions can also be exploited to influence their pharmacoki-netics properties. By engineering Fc regions to alter the binding affinity for the major histo-compatibility complex (MHC) class I-related receptor FcRn, it is now possible to produce significant changes in the blood clearance dynamics of IgG molecules (E. Sally Ward; University of Texas Southwestern, Dallas, TX, USA). This brings closer to reality the

engineering of IgG molecules so that their clearance from the circulation is tuned to match the desired clearance profile for the intended biological function and therapeutic application of the molecule. As infrequent dosing is generally more convenient, it is likely that therapy requiring the sustained presence of circulating antibody (e.g., blocking ligand–receptor interactions) will benefit from the presence of such long-lived antibodies.

Work is also progressing in developing mAbs that more efficiently bind to FcγRIII and thereby mediate ADCC. FcγRIII is expressed by NK cells and other leukocyte populations, and activates target cell lysis or phagocyto-sis. Elegant in silico algorithms are now being applied to introduce up to four amino acid mutations into the Fc domain to selectively tune the affinity for FcγRIII, and other Fcγ receptors. (Bassil Dahiyat; Xencor, Monrovia, CA, USA). The process consists of a combined computational and experimental method using a proprietary technology as a compu-tational screen to search the entire sequence space. By eliminating sequences incompat-ible with the protein fold, Protein Design Automation rapidly reduces the number of sequences to a size amenable to experimental screening, resulting in a library that can be constructed and experimentally screened to select for variants with modified properties such as improved binding to FcγRIII. Anti-tumor antibodies containing such high-affin-ity Fc domains are very efficient mediators of ADCC. An alternative strategy to enhance ADCC by mAb is to engineer production cell lines to tune the Fc glycosylation—increase bisecting N-acetylglucosamine and decrease fucosylation (Pablo Umaña; Glycart, Schlieren-Zürich, Switzerland).

The ultimate relevance of ADCC to thera-peutic benefit, however, remains inferential, although progress is also being made here. For example, individuals with FcγRIII poly-morphisms that improve the capacity to medi-ate in vitro ADCC by host natural killer cells using standard cytotoxicity assays have been observed to exhibit higher objective response rates than otherwise similar patients lacking those polymorphisms (Ronald Levy; Stanford University, CA, USA; refs. 8,9). Ongoing clini-cal trials with antibodies designed to have more efficient interactions with FcγRIII should address this consideration more directly.

In related work, antibody Fc region inter-actions with cellular Fc receptors have been modified to enhance antigen presentation by dendritic cells based upon preferential association of the antibodies with activating or inhibitory Fc receptors (Raphael Clynes; Columbia University, New York); such

presentation can be biased to promote or inhibit the generation of cytotoxic T-cell responses against the targeted antigen; for example, an antibody that preferentially asso-ciates with the activating isoform, FcγRIIA, is more likely to induce TH1-dependent T-cell responses, whereas preferential association with FcγRIIB is more likely to induce TH2 –dependent T-cell responses. Indeed, results from a phase 2 clinical trial in metastatic breast cancer run by one of our groups (Louis Weiner), which tested a bispecific mAb target-ing HER2/neu and FcγRIII (which promote ADCC), indicate that some of the people treated with the bispecific mAb developed host antibodies directed against HER2/neu-express-ing tumor cells as well as CD4+ and CD8+ T-cell immune responses against the HER2/neu extracellular and intracellular domains.

The available evidence thus supports the contention that the induction of ADCC can lead to adaptive immune responses. The capacity to measure antibody and T-cell responses in the peripheral blood of treated patients makes it possible to use these ‘foot-prints’ of immune response to determine the relevance of ADCC to therapeutic benefits of selected mAbs. For example, consider a clinical trial of an unconjugated monoclonal antibody with the capacity to mediate in vitro ADCC, and imagine that treatment with the antibody provides significant clinical benefit to some, but not all treated patients. Several such antibodies (e.g., rituximab, trastuzumab, cetuximab) are routinely used in cancer ther-apy today. It should prove possible to directly test the proposition that ADCC underlies at least some of the observed therapeutic benefit of such antibodies by tallying the induction of anti-tumor immune responses and determin-ing if such responses occur more frequently in patients who clinically benefit from antibody therapy. More importantly, if such relation-ships are indeed demonstrated, it may prove possible to further shape, expand and prolong antibody-induced immune responses to the added clinical benefit of treated patients.

1. Keystone Symposium: Antibody-based Therapeutics for Cancer, Hilton Hotel, Santa Fe, New Mexico, February 17–22, 2005.

2. Schier, R. et al. J. Mol. Biol. 263, 551–567 (1996).

3. Colby, D.W. et al. Methods Enzymol. 388, 348–358 (2004).

4. Weaver-Feldhaus, J.M. FEBS Lett. 564, 24–34 (2004).

5. Weinstein, J.N. et al. Ann. NY Acad. Sci. 507, 199–210 (1987).

6. Adams G.P. et al. Cancer Res. 61, 4750–4755 (2001).

7. Graff, C.P. et al. Protein Eng. Des. Sel. 17, 293–304 (2004).

8. Cartron, G. et al. Blood 99, 754–758 (2002).9. Weng, W.K. & Levy, R. J. Clin. Oncol. 21, 3940–3947

(2001).

NEWS AND V IEWS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


SNPs on beadsEffective genotyping studies will require hundreds of thousands of SNP assays. The latest estimate from the haplotyping project puts the number of genetic variants in humans at 10 million, of which 200,000 to 300,000 might be sufficient to genotype an individual. Although arrays that can perform massive multiplexing of SNPs assays exist, technologies are still needed to simplify the procedure. In their whole genome genotyping assay (WGA), researchers from Illumina, Ambion and Prognosys Bioscience describe a platform that combines several simple technologies to assay an unlimited number of sites at single base resolution. WGA uses whole genome amplification, followed by capture of fragments on high-density bead arrays containing sets of probes for each allele. Primer extension with biotin-labeled nucleotides labels the captured strand, followed by signal amplification with streptavidin-phycoerythrin conjugates. In a pilot study, the authors demonstrate that their approach can genotype over 800 loci with 99% accuracy, which compares favorably with existing techniques. By increasing the number and density of hybridization sites, they estimate that they could assay 500,000 sites in a single experiment. The assay may also be applicable to other kinds of studies, such as expression profiling and loss of het-erogeneity measurements. (Nat. Genet. 37, 549–554, 2005) LD

Signals and noiseTwo papers in Science describe the use of synthetic gene regulatory cas-cades in Escherichia coli for determining what intrinsic and extrinsic

factors contribute to gene expression fluctuations and how these fluc-tuations or ‘noise’ are transmitted through gene cascades within single cells. Rosenfeld et al. investigate the quantitative relationship between transcription factor concentration and the production rate of down-stream targets in a synthetic ‘λ-cascade’ system; Pedraza et al. determine how noise propagates in a gene network incorporating the lac repres-sor (LacI) and the tet repressor (TetR). To allow real-time ‘quantitative’ measurement of gene expression, the two teams tracked fluorescence of E. coli strains engineered with fusions of different fluorescent reporter genes, each linked to a specific protein in one of the pathways. Rosenfeld et al. then examined how factors, such as biochemical parameters, noise and varying cellular states, affect gene regulation; Pedraza et al. identi-fied and measured what determines noise propagation in a single gene, including noise from other cellular factors affecting global gene expres-sion (extrinsic noise), noise from fluctuations due to factors inherent to the expression of an individual gene (intrinsic noise) and noise transmit-ted from upstream genes. Both studies provide important insights for understanding noise in cellular gene expression and its implications for the design of synthetic networks. (Science 307, 1962–1965, 2005; Science 307, 1965–1969, 2005) NC

Smaller carriers deliverSynthetic, positively charged and high-molecular-mass polyeth-ylenimines (PEIs) have been a mainstay for liposomal delivery of drugs and nucleic acids. Thus far, however, these have proven to be much less efficient and organ-specific than viral vectors. Now, Thomas et. al. have significantly increased the delivery efficiency and specificity of PEIs by removing N-acyl groups from commer-cial PEIs. By deacylating a commercial, 25-kDa linear PEI (PEI25), they found the aliphatic polyamine to be 21 times more efficient in vitro. Three other de novo synthesized PEIs devoid of N-acyl groups proved even more efficient. When applied to mice, deacylated PEI25 delivered DNA with a 1,500-fold increase in specificity to lungs and an overall 10,000-fold jump in efficiency compared to the acylated molecule. Finally, to prove the therapeutic potential of deacyl-ated PEIs, the authors applied one of the de novo synthesized PEIs, PEI87, together with a short interfering RNA (siRNA) to combat influenza in mouse and observed a 94% drop of virus titers in lung.(Proc. Natl. Acad. Sci. USA 102 5679–5684, 2005) MZ

mAbs against West Nile virusThe spread of West Nile virus, a zoonotic agent that can cause men-ingitis and encephalitis in the elderly and immunocompromised, is adding urgency to the search for new treatments to supplement human polyclonal antibodies. To develop a monoclonal antibody (mAb) with potent virus neutralizing capacity, Diamond and coworkers raised 46 mAbs by immunizing mice with the West Nile virus envelope (E) protein. Error-prone PCR mutagenesis was then used to rapidly map mutations in domain III (DIII) of E protein displayed on the surface of yeast that affect mAb binding. For in vitro protection assays, only two of the 46 mAbs rapidly neutralized the virus in human adrenal carcinoma cells. The most potent of these mAbs, E16, boosted survival of a West Nile Virus Disease mouse model from 10% to 90%. E16 further mapped to the same neutral-izing epitope as human antibodies derived from people convalescing from West Nile viral infection, suggesting its potential as a human therapeutic. Indeed, chimeric mouse-human antibodies based on E16 increased the survival of wild-type mice from 46% to 67%.(Nat. Med. 11, 522–530, 2005) TM

Cell therapy for AlzheimerA phase I clinical trial in eight individuals with mild Alzheimer disease appears to have slowed the decline of cognitive function. The treatment involved transplantation into the brain of autologous fibroblasts genetically modified to express nerve growth factor (NGF). Although the benefits of NGF therapy have been extensively validated in animal models of cholinergic neuronal degeneration, clinical applications have been hindered by a lack of precise delivery methods, as indiscriminate delivery can cause serious side effects. Tuszynski and colleagues found that injection of NGF-expressing fibroblasts into the cholinergic basal forebrain had no adverse consequences during the monitoring period of 18–24 months. Moreover, the yearly rate of decline of cognitive function over an average of 22 months, as measured by the Mini-Mental Status Examination, decreased to 49% of the pretreatment rate. A second test, the Alzheimer Disease Assessment Scale–Cognitive subcomponent, indicated a modest benefit in the first 12 months after treatment. The authors caution that NGF therapy is not a cure for Alzheimer disease because it targets only one component of the pathological process, the cholinergic neuron. Nevertheless, if the results of this small noncontrolled trial are reproducible, the approach represents an important therapeutic advance.(Nat. Med. 11, 545–549, 2005) KA

Research Highlights written by Kathy Aschheim, Nadia Cervoni,Laura DeFrancesco, Teresa Moogan and Mark Zipkin.

RESEARCH H IGHL IGHTS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Systematic interpretation of genetic interactions using protein networksRyan Kelley1,2 & Trey Ideker1,2

Genetic interaction analysis,in which two mutations have a combined effect not exhibited by either mutation alone, is a powerful and widespread tool for establishing functional linkages between genes. In the yeast Saccharomyces cerevisiae, ongoing screens have generated >4,800 such genetic interaction data. We demonstrate that by combining these data with information on protein-protein, prote in-DNA or metabolic networks, it is possible to uncover physical mechanisms behind many of the observed genetic effects. Using a probabilistic model, we found that 1,922 genetic interactions are significantly associated with either between- or within-pathway explanations encoded in the physical networks, covering ~40% of known genetic interactions. These models predict new functions for 343 proteins and suggest that between-pathway explanations are better than within-pathway explanations at interpreting genetic interactions identified in systematic screens. This study provides a road map for how genetic and physical interactions can be integrated to reveal pathway organization and function.

A major biological challenge is to interpret observed genetic interac-tions in a physical cellular context1–3. There are several major types of genetic interactions: synthetic-lethal interactions, in which muta-tions in two nonessential genes are lethal when combined; suppressor interactions, in which one mutation is lethal but when combined with a second, cell viability is restored; and an array of other effects such as enhancement and epistasis. Genetic interactions have been used extensively to shed light on pathway organization in model organisms1–4. In humans, genetic interactions are critical in link-age analysis of complex diseases5 and in discovery of new pharma-ceuticals6. Although genetic interactions are classically identified by mutant screens7, recent studies have applied systematic ‘reverse’ methods such as synthetic genetic arrays (SGA)8 or synthetic lethal analysis by microarrays (SLAM)9 to catalog ~4,000 synthetic-lethal and synthetic-sick interactions in Saccharomyces cerevisiae.

Because of the high-throughput nature of SGA, discovery of new genetic interactions is largely automated. However, interpreting the

functional significance of each result remains a relatively slow process. The problem is compounded by the large number of genetic interac-tions measured when screening one gene versus all others (~34 on average10) as well as possible false positives if the interactions are not confirmed by tetrad or random spore analysis. Thus, without further methods to aid in characterizing synthetic lethals, large-scale interpretation is a daunting prospect.

A promising solution may be to integrate synthetic lethals with other types of high-throughput interactions. For instance, direct physical interactions among proteins are being mapped by systematic two-hybrid11–15 or immunoprecipitation studies16,17, whereas physi-cal interactions between transcription factors and promoter sites are determined using chromatin-immunoprecipitation in conjunction with DNA microarrays18,19. These interactions comprise a physical network, which correlates with the network of genetic interactions and provides potential clues as to the mechanisms behind particular synthetic-lethal effects. Previous studies have demonstrated this cor-related structure in yeast, by showing that two proteins in the same region of the genetic network are likely to also physically interact8,10, that genes with similar patterns of genetic interactions often occur within the same protein complex10 and that a protein with many interactions in the physical network typically has many interactions in the genetic network also20.

These studies suggest that it may be possible to interpret observed synthetic-lethal relationships explicitly using physical interactions. In this regard, previous authors1,21 have noted that synthetic-lethal interactions are typically associated with one of three types of physical interpretations: between-pathway models, within-pathway models and indirect effects (Box 1).

Here, we demonstrate a computational framework for assem-bling genetic and physical interactions into models corresponding to between- versus within-pathway interpretations. Regions of the physi-cal network that correspond to each type of model are identified using a probabilistic scoring scheme. These models predict new protein func-tions and suggest that genetic interactions are more likely to bridge redundant or complementary processes than to combine additively within the same process.

Construction of genetic and physical networksWe assembled a genetic interaction network from two primary data sources (Fig. 1). The first was generated by SGA, a large-scale screen10 crossing 132 yeast gene deletion strains versus each of the ~4,700 avail-able deletion strains22 and resulting in 2,012 observed synthetic-lethal

1Program in Bioinformatics, 2Department of Bioengineering, University of California, San Diego, 9500 Gilman Dr., San Diego, California 92093-0412, USA. Correspondence should be addressed to T.I. ([email protected]).

Published online 5 May 2005; doi:10.1038/nbt1096

A N A LY S I S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


interactions and 2,113 synthetic-sick interactions. The second data source consisted of an additional 687 synthetic-lethal interactions culled from the literature and catalogued at the Munich Information Center for Protein Sequences (MIPS)23. The combined genetic network syn-thesizing these data consisted of 1,434 proteins (genes) linked by 4,812 synthetic-lethal interactions.

We also assembled a physical network of 5,993 yeast proteins con-nected by physical interactions of three types: 15,429 protein-protein interactions (the two proteins a and b display physical binding); 5,869 protein-DNA regulatory interactions (a binds upstream of the gene encoding b) and 6,306 shared-reaction metabolic relationships (a and b are enzymes that operate on at least one metabolite in common). The protein-protein interactions were downloaded from the DIP data-base24 as of July 2004 and predominantly included data from large-scale experiments13,15–17. The protein-DNA interactions were obtained from a large-scale chromatin-immunoprecipitation study of 106 transcription factors18 (interactions with P = 0.001). Enzymatic reactions linked by common metabolites were obtained from KEGG25, excluding metabolite cofactors such as ATP or H2O (listed in Supplementary Table 1 online). The combined physical network covered 94.4% of all proteins in the genetic network. Both networks are provided at http://www.cellcircuits.org/Kelley2005/ in Cytoscape26 (SIF) format.

Between-pathway interpretations for genetic interactionsPreliminary statistical analyses confirm a limited relationship between genetic and physical interactions (see Supplementary Fig. 1 online and

Tong et al.8,10), but demonstrate a need for structured models to effi-ciently separate signal from noise. Towards this goal, we implemented a probabilistic modeling procedure to capture the between-pathway interpretation of genetic interactions. This procedure involved a search for pairs of physical pathways that were densely connected by genetic interactions, in which a ‘pathway’ was loosely defined as any densely con-nected set of proteins in the physical network (this definition generically covers many network structures, including protein complexes). Pairs of pathways (constituting a single network model; see Fig. 1) were assigned a score proportional to the density of physical interactions falling within each pathway and the density of genetic interactions bridging between pathways (Box 2). This search generated 360 significant models cover-ing 401 pathways and incorporating a total of 1,573 genetic interactions (196 MIPS, 687 SGA synthetic lethal, 690 SGA synthetic sick) and 1,931 physical interactions (1,248 protein binding, 77 regulatory, 606 shared reaction). Significance of these models was assessed by comparison to random genetic and physical networks. Detailed information for all models is provided in Supplementary Tables 2 and 3 online and at http://www.cellcircuits.org/Kelley2005/.

Pooling diverse genetic and physical interaction data sets widens the search but also has the potential to decrease the coverage of network models, because not all data sets may be equally predictive and high-scoring network models are more likely to arise at random in large net-works. To investigate the effect of data pooling, we repeated the search on a smaller network comprising large-scale synthetic-lethal (SGA) and protein-binding (DIP) interactions only. This reduced search identified 20 models containing a total of 137 synthetic-lethal and 120 protein-binding interactions (Fig. 2). In comparison to the complete search, fewer protein-binding and SGA synthetic-lethal interactions were incorporated into models, demonstrating the synergy obtained by data pooling (although models generated by the restricted search performed somewhat better in validation). Supplementary Table 4 online analyzes the impact of removing each physical and genetic data set from the modeling procedure.

Within-pathway interpretationsWe next searched the physical and genetic networks for within-pathway explanations. This procedure assigned a high score to single sets of pro-teins that were densely connected by both physical and genetic interac-tions (see Fig. 1, Box 2 and Supplementary Fig. 2 online). This search yielded 91 significant models. In all, these contained 272 MIPS, 225 SGA synthetic lethal and 169 SGA synthetic-sick interactions associated with 318 protein-binding, 37 regulatory and 36 shared-reaction interactions. Four representative within-pathway models are shown in Figure 3.

Box 1 Interpretations of genetic interactions

Between-pathway interpretations. The genetic interaction bridges genes operating in two pathways with redundant or complementary functions. Deletion of either gene is expected to abrogate the function of one but not both pathways.

Within-pathway interpretations. The genetic interaction occurs between protein subunits within a single pathway. A single gene is dispensable for the function of the overall pathway, but the additive effects of several gene deletions are lethal.

Indirect effects. The synthetic lethal phenotype is not mediated by a localized mechanism in the physical network. Indirect effects can occur because a deletion phenotype represents not just the absence of one particular gene, but also the response of the cell to its absence, involving many diverse pathways21.

Network modelidentification

Within-pathway

Between-pathwaySignificant

models

GeneticPhysical

Validation

TypeSynthetic lethalSynthetic sick Synthetic lethal

SourceSGASGAMIPS

No.2,0122,113 724

Genetic network

Type/directionProtein-protein Protein-DNAReaction-reaction

SourceDIPLee et al.KEGG

No.15,4295,8696,306

Physical network

Number of models 360 between 91 within

Enriched functionsPredicted interactions

Number of interactions 1,922 genetic 2,082 physical

Figure 1 Method overview. A combined physical and genetic network is searched to identify between- or within-pathway models of genetic interactions. The between-pathway model implies two groups of proteins (pathways) with many physical connections within each pathway (solid blue links) and genetic interactions spanning between pathways (dotted red links). The within-pathway model implies many physical and genetic interactions within the same group of proteins. In the search, 360 and 91 network models were identified that correspond to between- or within-pathway searches, respectively.

ANALYS IS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Functional enrichment of modelsAs initial validation of the between- and within-pathway models, we found that both types were significantly enriched for particular func-tional annotations recorded in the Gene Ontology database27. Two-hun-dred and fifty-one out of 401 pathways in between-pathway models were enriched for proteins with a common Molecular Function, Biological Process or Cellular Component annotation using the hypergeometric test (P = 0.05; Bonferroni-corrected for multiple testing)28. Similarly, 52 of the 91 within-pathway models were enriched for Gene Ontology annotations. Moreover, these functional enrichments were higher than expected based on the physical interaction network alone (see Supplementary Table 5 online).

Prediction of new protein functionsHaving established that proteins in many of the between- and within-pathway models were enriched for specific annotations, we used this con-cept to predict new protein functions. Specifically, for physical pathways in which a majority of proteins were already assigned a common significant annotation, we predicted this term for the remaining proteins in the path-way. To eliminate overly general predictions, significance was assigned only to those terms that were enriched at a level of P = 0.05 and were associated with fewer than 100 yeast proteins overall.

For between-pathway models, this approach predicted 745 molecular function, biological process or cellular component annotations among 282 proteins. In comparison, the within-pathway models predicted 285

annotations involving 127 proteins, bringing the total to 973 annotations for 343 proteins accounting for repeated predictions. A list of novel func-tional predictions is provided in Supplementary Table 6 online. Less than a quarter of these predictions were attainable using a similar approach based on the physical network only (Supplementary Table 7 online).

Accuracy of these predictions was estimated using cross validation29. Using a standard five-way procedure, the set of yeast proteins was parti-tioned such that annotations were hidden for one-fifth of the proteins and annotations for the remaining four-fifths of proteins were used to predict the hidden information. Each prediction for a protein in the ‘hidden set’ was scored as a success or failure depending on whether it recovered a hid-den annotation. Using this approach, the success rate was estimated to be 63% for between-pathway models, 69% for within-pathway models.

Prediction of new genetic interactionsFinally, we investigated whether the network models could predict the existence of new genetic interactions (Fig. 4). According to the between-pathway model, proteins in one pathway genetically interact with many of the same partners in a second pathway. This leads to the occur-rence of ‘complete bipartite motifs’ in the genetic interaction network, defined as four-protein subnetworks in which the first two proteins are connected to the second two proteins by all four possible genetic interactions (Fig. 4a; see Milo et al.30 for an introduction to network motifs). When an incomplete motif (IM) is observed, for which only three of the four genetic interactions are present, the motif implies

Box 2 Scoring the models

Scoring within-pathway explanations. The within-pathway model implies dense interactions within a single group of proteins in both the physical and genetic networks. We adopt a previously described log-odds score37 to assess the likelihood that a group of proteins is more densely connected than would be expected at random:

where V is a set of proteins and E a set of interactions among those proteins (genetic or physical). IE(a,b) is an indicator function which equals 1 if and only if the interaction (a,b) occurs in E and otherwise 0. For Modeldense, interactions are expected to occur with high probability (β) for every pair of proteins in V. In this work, β is set to 0.9 (Supplementary Fig. 2 shows how the results depend on choice of β). For Modelrandom, the probability of observing each interaction (ra,b) is determined by estimating the fraction of all networks with identical degree distribution which also contain that interaction. Comparable random networks are generated by ‘crossing’ pairs of edges in a process similar to that described by Milo et al.30 In this randomization, only edges of the same type are allowed to be crossed. In addition, for undirected types, either interacting node is allowed to serve as the ‘source’ in crossing the edges. Such randomization generates a family of random networks which resemble the original network and corrects for the presence of highly connected proteins, which score highly under both models. The interaction density is evaluated independently for the physical and genetic networks,

yielding an overall score for the within-pathway model:S = Swithin(V,Ephysical) + Swithin(V,Egenetic).

Scoring between-pathway explanations. The between-pathway model implies dense genetic interactions connecting two separate, nonoverlapping groups of proteins, where each group is densely connected by physical interactions. The density of physical interactions is scored independently within two sets of proteins V1 and V2 using the above function S. A related log-odds score is used to evaluate the probability that the genetic interactions Egenetic bridging between these sets are denser than random:

The final scoring function for the between-pathway model is then:

Search and Significance. Sets of proteins that are well explained by either the within-pathway or between-pathway models are identified using a greedy network search procedure. The search is as previously described by Sharan et al.37 except that it is seeded from each pair of genetically interacting proteins. Pathways that share more than 50% of genetic interactions with a higher-scoring result are discarded. To determine the significance threshold, identical searches are performed over 100 random trials in which both the genetic and physical networks are randomized as described above. Models that score higher than the maximal-scoring models in 95% of random trials are reported as significant.

Swithin(V,E) = logP (V, E ⎜ Modeldense)

P (V, E ⎜ Modelrandom)

= log

Π I

I

E(a,b) + (1 – ) (1 – IE(a,b))β

(a,b)∈V × V

Πra,b ra,bE (a,b) + (1 – ) (1 – IE(a,b))(a,b)∈V × V

β

Sbetween(V1,V2,Egenetic) = log

Π β IE

(a,b) + (1 – ) (1 – IE (a,b))β(a,b)∈V × V 1 2

genetic genetic

Π IE

(a,b) + (1 – ) (1 – IE (a,b))(a,b)∈V × V 1 2

genetic geneticra,b ra,b

S(V1, V2, Eall) = Swithin(Vi, Ephysical) + Sbetween(V1, V2, Egenetic)Σi =1,2

ANALYS IS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


that the remaining interaction is true. Physical network information is incorporated by requiring that valid incomplete motifs fall within (i.e., are subgraphs of) a between-pathway model.

We applied the technique of five-way cross-validation to estimate the accuracy of genetic interaction prediction versus the minimum number of required incomplete motifs (Fig. 5). In each of five cross-validation trials, approximately one-fifth of the genetic interaction data were withheld, including both positive and negative interactions measured for each genetic ‘bait’ in SGA. These positive and negative interactions were subsequently used to test prediction accuracy. For instance, at a prediction threshold of eight or more incomplete motifs, the between-pathway models predicted 43 new genetic interactions with 87% estimated accuracy (Fig. 5). To assess the contribution of the physical models in the prediction process, we also predicted ‘naive’ genetic interactions by relaxing the requirement that incomplete motifs

fall in a between-pathway model. The estimated accuracy fell to 5% for these naive predictions, evaluated at the same threshold of eight incomplete motifs.

For the within-pathway models, genetic interactions were implied between proteins that had genetic interactions with one or more com-mon neighbors (Fig. 4b). The physical network was incorporated by restricting the proteins and neighbors to fall into a single within-path-way model. The number of common neighbors was used as a measure of confidence in the implied genetic interaction, and cross validation was used to estimate the prediction accuracy as a function of this number. The maximal prediction accuracy was 38%, achieved at a prediction threshold of three or more common neighbors (Supplementary Fig. 3 online). The corresponding success rate for naive predictions, made without constraining the proteins to occur in within-pathway models, was 15%. Thus, both types of models enhance the accuracy of prediction

of genetic interactions, but between-pathway models appear to be better predictors than within-pathway models.

Preponderance of between-pathway interactionsGiven a systematic approach for associating genetic interactions with physical interpre-tations, it is of interest to ask which type of interpretation is most common. Focusing on large-scale SGA measurements, roughly three-and-a-half times as many genetic interactions are associated with between- as opposed to within-pathway models (1,377 versus 394 SGA interactions). These figures can be viewed as an a priori expectation that a newly deter-mined SGA interaction will fall between versus within pathways, suggesting that SGA interac-tions typically span between multiple physical network regions instead of occurring within a single complex or pathway. One reason for the preference towards between-pathway models may be that SGA interactions are mainly tar-geted to nonessential genes (due to their use of complete gene deletions as opposed to, e.g., point mutations made by classical tech-niques).

Using physical models, it is possible to characterize approximately 40% of the genetic interactions as occurring between or within pathways. Whether the remain-ing interactions belong to between-pathway models, within-pathway models or are best characterized as ‘indirect’ (Box 1) cannot be reliably determined at this stage. For exam-ple, consider the case of two related pathways, each with only one protein required for path-way function. In this case, only the required proteins would be connected by a (single) genetic interaction across the pathways, making it difficult for the between-pathway model to achieve statistical significance.

Further examination of the between-path-way models reveals that many of the genetically linked pathways have clear interdependent functional relationships. For example, pathway

Glycoprotein metabolism

Amino-terminal blocking

Motor activity

Budding

Regulation

Prefoldin complex

Dynactin

Cell cortex

Retromer complex

DNA catabolism

Chromosome

b

M) Prefoldin complex

N) Dyn

actin

com

plex

T) Kinetochore

Q) Cell cortexV) Cell cortex

Y) Retr

omer

comple

xO) Ubiquitination

U) Retrograde transport

Lge1

Bre1Shared proteins

Genetic (SGA)

Physical bindingGenetic bundle

Interactions

a

Yel043w

Ypt6

Rgp1

Ric1

Vps74

Pep8

Vps17

Vps29Vps35

Vps5

Bni4

Chs3

Skt5

Vrp1

Bbc1 Myo5

Rvs167

Bck1 Slt2

Ctf3

Iml3

Ctf19 Chl4

Yll049wJnm1

Nip100

Pac11 Gim3

Pac10

Gim5Yke2

Gim4

Arp1

Figure 2 Between-pathway explanations for genetic interactions. (a) Several high-scoring models are shown (M,N,T; Q,V; O,U,Y). Blue solid and red dotted links indicate physical and genetic interactions, respectively. (b) Bird’s-eye view of all between-pathway models obtained from a search on a reduced network composed of SGA and DIP interactions. Each node [A]–[Z] represents a physical pathway; groups of genetic interactions between pathways are condensed into a single link called a ‘bundle.’ Node colors indicate significant Gene Ontology annotations. Solid gray lines connect pathways that share one or more proteins; such pathways may represent different components of a larger mechanism.

ANALYS IS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


M contains members of the prefoldin complex, which have synthetic-lethal interactions with members of pathways N and T forming parts of the dynactin complex and kinetochore, respectively (Fig. 2a). The prefoldin complex promotes folding of α- and β-tubulin into func-tional microtubules31. These are important for the function of dynactin, an adaptor complex involved in translocating the spindle and other molecular cargos along microtubules32, as well as the kinetochore, which anchors chromosomes to spindle microtubules during metaphase33. Apparently, deletion of proteins in the prefoldin complex reduces micro-tubule stability, leading to synthetic-lethal interactions with pathways that are directly dependent on microtubule function.

These pathways also predict a new function for the uncharacterized protein Yll049w (pathway N). This protein binds Jnm1, a dynactin protein which is required for spindle partitioning in anaphase32. In addition, it has synthetic-lethal interactions with members of the prefoldin complex in a manner similar to dynactin genes. Together, these relationships suggest that Yll049w is associated with dynactin during spindle partitioning. However, because Jnm1 has 12 physical

interactions overall, and Yll049w has a total of 14 interactions in the genetic network, this prediction would have been difficult to make without an integrated approach.

Pathways O, U and Y provide another example of synergistic path-ways linked by genetic interactions (Fig. 2a). Pathways U and Y mediate retrograde transport of proteins to the Golgi apparatus34,35. Pathway O (Bre1, Lge1) is involved in histone ubiquitination and cell size control, where cell size is influenced by the histone ubiquitination activity by an unknown process36. The abundant genetic interactions between path-ways O and U indicate a possible role for retrograde transport in histone ubiquitination, or reciprocally, for histone ubiquitination in retrograde transport. Moreover, the uncharacterized protein Yel043w is physically

BindingRegulatoryReaction

Physical interactions

SGA (SL)

MIPS

Genetic interactions

SGA (SS)

Cellular morphogenesis

Cla4Swi4

Cdc12

Swi6

Bni1

Swe1

Cdc28Gin4

Ste20

Cdc24

Gic2

Bem1

Rsr1

Cdc42

CK2 complex

Cka1

Ckb2Ckb1

Cka2

Spliceosome

Prp11 Prp21

Mud2

Prp9

Msl5

Nucleic acid and related transport

Gle1Nup100

Nup116

Nup84

Nup42

Nup145

Gle2

Between pathway Within pathway

Observed

PredictedMotif 2

Motif 1

Commonneighbors

b b'

a a'

c c'

d d'

e

f

a b

0

20

40

60

80

100

1 8 15 22 29 36

Number of incomplete motifs

Acc

ura

cy (

%)

Between pathwayNaive

Naive predictions

0

20

40

60

80

100

0 50 100 150 200 250 300Number of incomplete motifs

Acc

ura

cy (

%)

Figure 3 Within-pathway explanations for genetic interactions. A total of 91 pathways were identified, of which four examples are displayed. Color is used to indicate the data set from which each interaction was drawn.

Figure 4 Genetic interaction prediction schemes. Two different schemes are proposed for predicting genetic interactions, depending on the underlying network model. Observed genetic interactions are shown in red, while the corresponding predicted genetic interactions are shown in gray. (a) Under the between-pathway model, two incomplete bipartite motifs are shown which predict a genetic interaction between genes b and b′. (b) Under the within-pathway model, common genetic neighbors are used to predict a genetic interaction between genes d and d′. Note that these diagrams contain additional incomplete motifs which have been omitted for clarity: the motifs in a can be rearranged to predict genetic interactions (a to c′) and (c to a′); the motifs in b can be rearranged to predict (e to f).

Figure 5 Success rate of genetic interaction prediction versus the stringency of prediction. Success rate is measured through cross-validation as (predicted positives)/(predicted positives and negatives). Stringency is defined by the minimum number of incomplete bipartite motifs required for prediction. Blue diamonds mark the success rate for predictions in which incomplete motifs must occur in a between-pathway model. The success rate is dramatically higher than for naive predictions (magenta) which predict interactions in the same manner, but are not constrained by the physical network. Even for much more stringent prediction criteria, the success rate of naive predictions fails to exceed that of the between-pathway predictions (inset).

ANALYS IS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


associated with Bre1 and Lge1 and also has the same pattern of genetic interactions, suggesting that the three proteins may function together.

In summary, we have presented a methodology for integrating large-scale genetic and physical networks to capture the physical context behind observed genetic interactions. Approximately 40% of yeast synthetic-lethal genetic interactions can be incorporated into high-level physical pathway models and are approximately three and a half times as likely to span pairs of pathways than to occur within pathways. Further studies will be needed to address other types of genetic effects to extend this approach from yeast to the growing number of other organisms for which protein networks are now available. As systematic approaches generate ever larger databases of interactions across a variety of species, integrative modeling approaches such as the one proposed here will be indispensable for select-ing and organizing the information into predictive models.

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

ACKNOWLEDGMENTSWe thank Jonathan Wang, Owen Ozier and Gopal Ramachandran for preliminary investigations and Vineet Bafna, Ben Raphael and Vikas Bansal for insightful commentary. Craig Mak, Silpa Suthram and Taylor Sittler provided helpful reviews of the text. Funding was provided by the National Institute of General Medical Sciences (GM070743-01) and the National Science Foundation (NSF 0425926).

COMPETING INTERESTS STATEMENTThe authors declare that they have no competing financial interests.

Published online at http://www.nature.com/naturebiotechnology/

1. Guarente, L. Synthetic enhancement in gene interaction: a genetic tool come of age. Trends Genet. 9, 362–366 (1993).

2. Avery, L. & Wasserman, S. Ordering gene function: the interpretation of epistasis in regulatory hierarchies. Trends Genet. 8, 312–316 (1992).

3. Thomas, J.H. Thinking about genetic redundancy. Trends Genet. 9, 395–399 (1993).

4. Hartman, J.L., Garvik, B. & Hartwell, L. Principles for the buffering of genetic variation. Science 291, 1001–1004 (2001).

5. Sham, P. Shifting paradigms in gene-mapping methodology for complex traits. Pharmacogenomics 2, 195–202 (2001).

6. Dolma, S., Lessnick, S.L., Hahn, W.C. & Stockwell, B.R. Identification of genotype-selective antitumor agents using synthetic lethal chemical screening in engineered human tumor cells. Cancer Cell 3, 285–296 (2003).

7. Forsburg, S.L. The art and design of genetic screens: yeast. Nat. Rev. Genet. 2, 659–668 (2001).

8. Tong, A.H. et al. Systematic genetic analysis with ordered arrays of yeast deletion mutants. Science 294, 2364–2368 (2001).

9. Ooi, S.L., Shoemaker, D.D. & Boeke, J.D. DNA helicase gene interaction network defined using synthetic lethality analyzed by microarray. Nat. Genet. 35, 277–286 (2003).

10. Tong, A.H. et al. Global mapping of the yeast genetic interaction network. Science 303, 808–813 (2004).

11. Li, S. et al. A map of the interactome network of the metazoan C. elegans. Science 303, 540–543 (2004).

12. Rain, J.C. et al. The protein-protein interaction map of Helicobacter pylori. Nature 409, 211–215 (2001).

13. Uetz, P. et al. A comprehensive analysis of protein-protein interactions in Saccharomyces cerevisiae. Nature 403, 623–627 (2000).

14. Giot, L. et al. A protein interaction map of Drosophila melanogaster. Science 302, 1727–1736 (2003).

15. Ito, T. et al. Toward a protein-protein interaction map of the budding yeast: a compre-hensive system to examine two-hybrid interactions in all possible combinations between the yeast proteins. Proc. Natl. Acad. Sci. USA 97, 1143–1147 (2000).

16. Ho, Y. et al. Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry. Nature 415, 180–183 (2002).

17. Gavin, A-C., Bösche, M., Krause, R., Grandi, P. & Marzioch, M. Functional organiza-tion of the yeast proteome by systematic analysis of protein complexes. Nature 415, 141–147 (2002).

18. Lee, T.I. et al. Transcriptional regulatory networks in Saccharomyces cerevisiae. Science 298, 799–804 (2002).

19. Harbison, C.T. et al. Transcriptional regulatory code of a eukaryotic genome. Nature 431, 99–104 (2004).

20. Ozier, O., Amin, N. & Ideker, T. Global architecture of genetic interactions on the protein network. Nat. Biotechnol. 21, 490–491 (2003).

21. Tucker, C.L. & Fields, S. Lethal combinations. Nat. Genet. 35, 204–205 (2003).22. Winzeler, E.A. et al. Functional characterization of the S. cerevisiae genome by gene

deletion and parallel analysis. Science 285, 901–906 (1999).23. Mewes, H.W. et al. MIPS: a database for genomes and protein sequences. Nucleic Acids

Res. 30, 31–34 (2002).24. Xenarios, I. et al. DIP, the Database of Interacting Proteins: a research tool for studying

cellular networks of protein interactions. Nucleic Acids Res. 30, 303–305 (2002).25. Kanehisa, M., Goto, S., Kawashima, S. & Nakaya, A. The KEGG databases at GenomeNet.

Nucleic Acids Res. 30, 42–46 (2002).26. Shannon, P. et al. Cytoscape: a software environment for integrated models of biomo-

lecular interaction networks. Genome Res. 13, 2498–2504 (2003).27. Gene Ontology Consortium. Creating the gene ontology resource: design and implementa-

tion. Genome Res. 11, 1425–1433 (2001).28. Zar, J.H. Biostatistical Analysis, edn. 3 (Prentice Hall, New Jersey, 1996).29. Kendall, S.M., Stuart, A. & Ord, J.K. Kendall’s Advanced Theory of Statistics, edn. 5

(Oxford University Press, NY, 1987).30. Milo, R. et al. Network motifs: simple building blocks of complex networks. Science

298, 824–827 (2002).31. Geissler, S., Siegers, K. & Schiebel, E. A novel protein complex promoting formation of

functional alpha- and gamma-tubulin. EMBO J. 17, 952–966 (1998).32. Kahana, J.A. et al. The yeast dynactin complex is involved in partitioning the mitotic

spindle between mother and daughter cells during anaphase B. Mol. Biol. Cell 9, 1741–1756 (1998).

33. Pidoux, A.L. & Allshire, R.C. Centromeres: getting a grip of chromosomes. Curr. Opin. Cell Biol. 12, 308–319 (2000).

34. Pfeffer, S.R. Membrane transport: retromer to the rescue. Curr. Biol. 11, R109–R111 (2001).

35. Siniossoglou, S., Peak-Chew, S.Y. & Pelham, H.R. Ric1p and Rgp1p form a complex that catalyses nucleotide exchange on Ypt6p. EMBO J. 19, 4885–4894 (2000).

36. Hwang, W.W. et al. A conserved RING finger protein required for histone H2B monou-biquitination and cell size control. Mol. Cell 11, 261–266 (2003).

37. Sharan, R., Ideker, T., Kelley, B.P., Shamir, R. & Karp, R.M. Identification of pro-tein complexes by comparative analysis of yeast and bacterial protein interaction data. Proceedings of the Eighth Annual International Conference on Research in Computational Molecular Biology–RECOMB, 282–289 (ACM Press, New York, 2004)

ANALYS IS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Baculovirus as versatile vectors for protein expression in insect and mammalian cellsThomas A Kost1, J Patrick Condreay1 & Donald L Jarvis2

Today, many thousands of recombinant proteins, ranging from cytosolic enzymes to membrane-bound proteins, have been successfully produced in baculovirus-infected insect cells. Yet, in addition to its value in producing recombinant proteins in insect cells and larvae, this viral vector system continues to evolve in new and unexpected ways. This is exemplified by the development of engineered insect cell lines to mimic mammalian cell glycosylation of expressed proteins, baculovirus display strategies and the application of the virus as a mammalian-cell gene delivery vector. Novel vector design and cell engineering approaches will serve to further enhance the value of baculovirus technology.

Over the past 20 years the baculovirus–insect cell expression system has become one of the most widely used systems for routine production of recombinant proteins1–6. A number of technological improvements have eliminated the original tedious procedures required to identify and isolate recombinant viruses, increasing the popularity of the system. These include development of a wide variety of transfer vectors, simpli-fied recombinant virus isolation and quantification methods, advances in cell culture technology and the commercial availability of reagents. These enhancements have resulted in a virus-based expression system that is safe, easy to use and readily amenable to scale-up.

In addition, biotechnology now uses baculoviruses in applications beyond the production of proteins in insect cells and larvae. These include the development of strategies for displaying foreign peptides and proteins on virus particles and the insertion of mammalian cell–active expression cassettes in baculoviruses to express genes efficiently into many different mammalian cell types. Baculoviruses engineered to display foreign peptides and proteins on the viral surface have proven particularly useful as immunogens and both surface display and capsid fusions may provide further opportunities for enhancing and targeting baculovirus-mediated transduction of mammalian cells.

Here, we review recent advances in baculovirus–insect cell protein production, baculovirus display and the development and application of baculoviruses as mammalian-cell gene-delivery vectors (Fig. 1).

Isolation and quantification of recombinant baculovirusesRecombinant baculovirus expression vectors were initially isolated using a highly inefficient homologous recombination process. Insect cells cotransfected with baculovirus and transfer plasmid DNA pro-duced a mixture of parental and recombinant viruses, with a recombi-nation frequency of only about 0.1%. Progeny were usually resolved by

plaque assay and recombinant clones identified microscopically by their distinctive occlusion-negative plaque phenotypes. This was tedious as recombinant plaques, surrounded by a sea of occlusion-positive paren-tal virus plaques, were difficult to identify. A huge improvement came with the development of baculovirus DNA that could be linearized at a unique Bsu36I site in the polyhedrin locus7. When used together with a transfer plasmid to cotransfect insect cells, the linearized viral DNA gave rise to recombinants at a higher frequency, typically around 25%. Later, baculovirus DNAs were engineered to have multiple Bsu36I sites, one within an essential viral gene8. Bsu36I digestion created a large deletion that functionally inactivated the essential gene, thus precluding replication of parental virus and increasing the frequency of recombi-nant virus production to over 90%. This approach was commercialized and the use of predigested viral DNAs became status quo for recombi-nant baculovirus production. Still, baculovirus plaque assays remained an essential part of the technology, as recombinant baculoviruses were most frequently cloned using this approach.

Efforts to eliminate the requirement for a plaque assay in virus isola-tion led to development of an in vivo bacterial transposition method, first described in 1993 and later commercialized as the Bac-to-Bac sys-tem9. This method involves site-specific transposition of a foreign gene from a donor plasmid to a cloned baculoviral DNA, or ‘bacmid’ such that the foreign gene is controlled by the polyhedrin promoter. Since Escherichia coli clones containing recombinant bacmid DNA acquire an antibiotic resistance marker and lose a lacZ marker, they can be easily selected and identified. One simply isolates viral DNA from positive bacterial clones and uses this bacmid DNA to transfect insect cells and produce recombinant virus. Theoretically this method does not require a plaque assay to resolve parental from recombinant virus progeny; however, the virus stock may be, nevertheless, polyclonal. A recent improvement in the frequency of recombinant viruses produced using Tn7-mediated transposition has been described, which may allow for the efficient generation of baculoviral libraries10,11. It is important to note that recent reports have shown that BAC vector sequences can be spontaneously excised from bacmid-derived vectors upon passage in insect cells12,13. In our experience, this has not posed a major problem

1Gene Expression Protein Biochemistry, GlaxoSmithKline R&D, 5 Moore Drive, Research Triangle Park, North Carolina 27709, USA. 2Department of Molecular Biology, University of Wyoming, Laramie, Wyoming 82071, USA. Correspondence should be addressed to T.A.K. ([email protected]).


R E V I E W©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


for routine protein production using bacmid-derived viruses, but could hinder large-scale production systems.

Recently two additional approaches have been described for gen-erating recombinant baculoviruses. An in vitro transposition system, adapted from Bac-to-Bac, commercially known as BaculoDirect, uses a purified transposase to move a foreign gene from a donor plasmid to a viral DNA acceptor, placing the foreign gene under the control of the polyhedrin promoter. The parental viral DNA has a herpes virus thymidine kinase gene, which serves as a negative selection marker that is eliminated upon transposition. Thus, insect cells are transfected with the mixture of parental and recombinant viral DNA created in the test tube and parental viruses are eliminated by gancyclovir treatment. As with Bac-to-Bac, plaque isolation is not required, but the virus stock may be polyclonal and an experienced virologist would probably under-take plaque purification. An alternative method uses a baculovirus with a lethal mutation in orf1629, which encodes an essential gene product required for virus replication. Viral DNA is maintained as a bacmid in E. coli. In this system cotransfection of insect cells with a transfer vec-tor containing the gene of interest and the engineered viral DNA yields 100% recombinant viruses14.

Studies have also been undertaken to develop rapid and facile methods for virus quantification. These efforts have been driven by the use of automated platforms for baculovirus production and expression15–17 and the need to rapidly titrate baculovirus stocks destined for the production of multi-subunit protein complexes and mammalian cell transduction studies. A commercially available immunocytochemical assay using a monoclonal antibody against the baculovirus gp64 protein provides virus titers within 48 h18. Two additional antibody-based assays have been developed that provide rapid virus titrations19,20. Quantification of baculovirus particles by

flow cytometry21 and real-time PCR22 has been described. Several assays that use reporter proteins such as green fluorescent protein (GFP) or β-galactosidase have also been described; however, an inherent drawback of these methods is the need for expression of an additional protein.

Protein glycosylation in the baculovirus–insect cell systemIt is often stated that the baculovirus–insect cell system has eukaryotic protein processing capabilities. It is generally true that insect cells can fold, modify, traffic and assemble newly synthesized polypeptides to pro-duce highly authentic, soluble end products1,2,23. However, it is equally true that insect protein processing pathways are not necessarily equivalent to those of higher eukaryotes. One of the best examples of a similar, but distinct processing pathway is the protein N-glycosylation pathway.

Early research on dipteran insect cells, which has been reviewed in detail elsewhere24,25, established a model of the insect proteinN-glycosylation pathway which, with a few caveats, is still valid today. These studies indicated that insect cells could assemble N-glycans, transfer them to nascent polypeptides and trim the N-glycan precur-sors to produce high mannose or paucimannose end products (Fig. 2). However, the cells failed to elongate the trimmed N-glycans to pro-duce complex products containing terminal galactose and/or sialic acid residues. As recombinant glycoproteins began to be produced, it was recognized that the lepidopteran insect cell lines used as hosts for baculovirus expression vectors followed this general paradigm. In addition, enzyme assays showed that these cell lines had little or none of the galactosyltransferase and sialyltransferase activities involved in N-glycan elongation (Fig. 2). Moreover, it was found that many of these cell lines have an unusual processing activity that converts an inter-mediate common to both the insect and mammalian pathways to the insect-specific paucimannose end product26 (Fig. 2).

Today, it is generally recognized that most baculovirus-expressed recombinant glycoproteins will acquire authentic N-glycans only at sites occupied by high mannose structures on the native mammalian products. In contrast, they are most likely to acquire paucimannose N-glycans at sites occupied by complex, terminally galactosylated and/or sialylated N-glycans on the native product. This latter fact is a clear limi-tation of the baculovirus–insect cell expression system because N-gly-cans, and particularly terminal sialic acids, contribute to glycoprotein functions in many different ways. For some clinical applications, such as in vivo administration of a therapeutic recombinant glycoprotein, the absence of terminal sialic acids would be unacceptable.

Recent trends in the development of the baculovirus–insect cell system include extensive efforts to address this problem, the details of which have been reviewed elsewhere27–31. However, an overview of selected devel-opments will be of interest to investigators using the baculovirus–insect cell system for recombinant glycoprotein production. An early step was the development of expression plasmids and methods for transforming lepidopteran insect cell lines containing stably integrated, constitutively expressed foreign genes32,33. These studies set the stage for the creation of transgenic lepidopteran insect cell lines containing mammalian genes encoding N-glycan processing activities that were absent in the paren-tal cell lines. The first transgenic insect cell line of this type was pro-duced by transformation with a bovine β1,4-galactosyltransferase gene. Baculovirus infection of this cell line, but not the parental Sf 9 cell line, led to the production of a foreign protein with terminally β-galactosylatedN-glycans34. Subsequently, a transgenic Sf 9 line encoding both bovine β1,4-galactosyltransferase and rat α2,6-sialyltransferase was isolated that supported the production of terminally α2,6-sialylated N-glycans35. This was a surprising result because the donor substrate required by the rat sialyltransferase, CMP-sialic acid, is not found at detectable levels in Sf 9

Baculovirus

Insect larvae

Insect cells

Mammalian cells

Baculodisplay

Figure 1 Versatility of baculovirus expression vectors. Baculovirus vectors can be used for a variety of applications. These include producing proteins in insect larvae, insect cells and mammalian cells. The insect and mammalian cells in the photomigrographs were treated with baculoviruses expressing GFP. Viruses can also be produced that display peptides or proteins on the surface of viral particles. The red circles on the schematic virus particle represent displayed gp64 fusion proteins.

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


cells36,37. A possible explanation was obtained with the finding that Sf9 cells have a sialic acid scavenging pathway that can support de novo glycoprotein sialylation in transgenic cells expressing the sialyltransferase38. However, detailed structural analyses revealed that only the lower (α1,3) branch of the N-glycans pro-duced by these transgenic insect cells had been elongated. Because monoantennary N-glycans are rarely found on mammalian glycoproteins, another transgenic insect cell line was designed to express five mammalian glycosyltransferases, including one that initiates elongation of the upper (α1,6) N-glycan branch39. These cells, designated SfSWT-1 cells produced biantennary, terminally monosialylated N-glycans. Finally, because a sialic acid scavenging pathway might be an inefficient way to produce CMP-sialic acid for de novo glycoprotein sialylation, a transgenic line designated SfSWT-3 was produced. These cells encode the same five glycosyltransferases of SfSWT-1, plus two murine enzymes that convert the sialic acid precursor, N-acetylman-nosamine, to CMP-sialic acid40. When cultured in the presence of N-acetylmannosamine, this new cell line produced high levels of intra-cellular CMP-sialic acid and a recombinantN-glycoprotein with a highly homogeneous, biantennary, monosialylated side chain.

The transgenic insect cell line approach has begun to address the inability of the baculovirus–insect cell system to produce authentic recombinant N-glycoproteins; however, the humanization of protein processing pathways in insect cells is a work in progress, with many additional developments needed and yet to come.

Enhancing protein expression in insect cellsIn many instances sufficient quantities of functional protein for experimental needs can be readily obtained from baculovirus-infected insect cells. However, this is not always true and for numerous rea-sons increased yields of functional protein are often desirable. Various approaches to increasing production of properly processed proteins were covered in an earlier review on this topic41. A number of studies have documented enhanced protein production following cotrans-fection with baculoviruses expressing chaperone proteins, which are known to aid in the folding and modification of newly synthesized proteins. The expression of correctly assembled Shaker potassium chan-nels in Sf9 cells was enhanced by coexpression of the calcium-binding, lectin-chaperone calnexin together with substitution of the polyhedrin promoter with the weaker basic protein promoter to drive expression of the ion channel42. Coexpression of calreticulin promoted the pro-duction of properly folded human lipoprotein lipase43 and HLA-DR4 tetramers44. Another approach has been to coexpress the chaperone Hsp70 and its cofactors Hsdj and Hsp40. Such coinfections have resulted in increased yields of soluble Epstein-Barr virus replication protein, BZLF145 and functionally active tumor suppression protein LKB146. These studies demonstrate the potential value of coexpressing chaperones to enhance functional protein production.

Significant increases in expression levels have also been reported by the addition of various DNA elements to the virus. The addition of baculovi-rus homology region 1 (hr1)47 and hr3 (ref. 48) sequence regions to the virus genome resulted in increased luciferase production. Incorporation

of a 21-base-pair (bp) element derived from a 5′ untranslated leader sequence of a lobster tropomyosin cDNA that contains the Kozak sequence and A-rich sequence found in the polyhedrin leader sequence into a recombinant virus enhanced the expression of tropomyosin and luciferase 20- and sevenfold, respectively49. As with the addition of hr elements, the effect of this 21-bp element will require further evaluation with additional proteins to determine the general applicability of this approach for enhancing protein yield.

Baculovirus infection of insect cells results in microscopically observ-able cell lysis within 3–5 d after infection. Cell disruption may lead to increased proteolytic activity and other environmental factors that can result in degradation of recombinant protein. In an attempt to over-come this difficulty, a baculovirus with reduced capability for initiating cell lysis was isolated by random mutagenesis and the application of a novel fluorescence resonance energy transfer (FRET)-based assay for selecting the desired mutant50. At 5 d after infection the mutant virus showed only 7% lysis of infected Sf21 cells, whereas the parent virus showed 60% lysis. Using this virus the authors demonstrated that a higher level of compactly folded, engineered luciferase protein could be produced with less degradation as compared to the parental virus. Another approach to reducing protein degradation is the develop-ment of a chitinase and v-cathepsin negative bacmid. Generation of a recombinant virus designed to express the cattle parasite Theileria parva sporozoite surface protein p67 with this bacmid protected the secreted recombinant protein from degradation51.

The baculovirus–insect cell system has been used successfully for the expression of thousands of diverse types of proteins. It has proven particularly valuable for the expression of G protein–coupled receptors (GPCRs)52,53 and coexpression with G proteins has proved valuable for studying receptor–G protein interactions54. The system has also proven very useful for expression of cytochrome p450 enzymes55–57. Irrespective of the protein being produced, a major advantage of the baculovirus–insect cell expression system is the ease of scale-up from the laboratory to a large-scale production system58.

Insect MammalianN yteca- lglu inimasoc d esa N yteca- lglu ynimasoc l

II esarefsnart

N yteca- lg masotcala ni ly tr na refs esalaiS yt sesarefsnar

ysotcalaG l sesarefsnartlaiS yt sesarefsnar

Asn

Asn

nsA

Asn

Asn

GlcNAcMan

GalGalNAc

Fucose

Sialic acid

High mannose

Paucimannose

Sialylated complex

Key to symbols

Figure 2 Overview of processing pathways and major N-glycans produced by insect and mammalian cell systems. The processing pathways in both systems yield a common intermediate. The major insect-cell end product (paucimannose) is produced by further trimming of this intermediate (left-hand branch), whereas the major mammalian-cell end products (including sialylated complex) are produced by elongation of this intermediate (right-hand branch).

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Production of virus-like particles (VLPs) and protein complexesThe baculovirus–insect cell system has been used extensively for the production of VLPs to study viral assembly processes in the absence of infectious virus, produce antigens for immunization and proteins for diagnostic assays and for gene transfer59–62. This approach is particu-larly valuable in those cases where cell culture–based viral replication systems are not available, such as human papilloma virus (HPV) and hepatitis C virus (HCV). The baculovirus–insect cell system allows one to deliver individual viral structural proteins via coinfection with mul-tiple baculoviruses, each expressing a single protein, or via a single virus designed to express multiple proteins63. By varying the multiplicity of infection and by using various promoters, one can attempt to control the amount of expressed proteins to optimize VLP production.

A striking example of the application of this technology is the develop-ment of HPV VLP–based vaccines64,65. VLPs composed of the HPV types 16/18 L1 structural proteins produced in baculovirus-infected insect cells have been shown in clinical trials to be efficacious in preventing cervical infections with HPV-16 and HPV-18, together with the associated cyto-logical abnormalities and lesions66. HCV VLPs have also been successfully produced using the baculovirus–insect cell system67–71. VLPs containing the core, E1 and E2 proteins of HCV resemble putative HCV virions and have been shown to effectively induce HCV-specific humoral and cel-lular immune responses in baboons71. Recently the assembly of human severe acute respiratory syndrome (SARS) coronavirus–like particles in baculovirus-infected insect cells expressing the S, E and M structural pro-teins has been described72,73. Budded VLPs could only be detected in the culture medium when the genes encoding the three proteins were carried by a single recombinant baculovirus73. These results provide impetus for further studies into the assembly and development of candidate VLP-based vaccines against this important disease.

Insect cells infected with recombinant baculoviruses have also been used to produce infectious adeno-associated virus (AAV) type 2 vec-tors74. Insect cells were coinfected with three recombinant baculoviruses, one expressing the AAV replication proteins, a second expressing the AAV structural proteins and a third expressing GFP under the control of a cytomegalovirus (CMV) promoter bounded by the AAV inverted terminal repeat sequences. The yield of functional genome-containing AAV particles per Sf9 cell produced in this system approached 5 × 104 demonstrating the system can produce large quantities of AAV vectors.

Baculovirus displayA variety of strategies have been developed for displaying heterolo-gous peptides or proteins on the surface of baculovirus particles by fusing the peptide or protein to the baculovirus surface glycoprotein, gp6475,76. In most instances the vector is designed so that baculovirus particles contain both wild-type gp64 and gp64 molecules containing the heterologous protein sequence. Baculoviruses displaying gp64-fusion proteins have proven to be very effective immunogens. Since this approach was first used to raise monoclonal antibodies against the nuclear receptors LXRβ and FXR77, it has been used successfully to elicit antibody responses to a variety of displayed proteins. These include human peroxisome proliferator-activated receptor78, Plasmodium ber-ghei circumsporozoite protein79, hemagglutinin protein of Rinderpest virus80, Theila parva sporozoite surface antigen p6781 and foot-and-mouth disease virus proteins82. Baculovirus display strategies have also been used for modification of the viral surface to influence baculovi-rus-mediated transduction of mammalian cells. These studies will be discussed further below. In addition to gp64 fusions, GFP was recently fused to the baculovirus vp39 capsid protein83. Capsid modifications may allow novel approaches for enhancing baculovirus mediated gene delivery into mammalian cells.

It has also been shown that membrane proteins produced in infected insect cells can be incorporated into baculovirus particles in a functional form. This was first observed for the β-2-adrenergic receptor, which was recovered in a functional form complexed with heterotrimeric G pro-teins84 and more recently for the human leukotriene B4 receptor85. This approach has been used successfully to produce a functional γ-secretase complex on the surface of baculovirus particles86. Coinfection of Sf 9 cells with viruses expressing the four putative γ-secretase components resulted in the production of virus particles with γ-secretase activity that was concentrated ~2.5-fold higher in the budded virus particles as compared to Sf 9 cell membranes. These studies show that baculovirus particles can provide a unique scaffold for the assembly and enrichment of functional membrane bound protein complexes.

Recombinant protein production in insect larvaeThe use of baculovirus-infected insect cell larvae as hosts for protein production was first described for the production of α-interferon in 1985 (ref. 87). Since that time larvae have been used successfully to produce a variety of recombinant proteins88–92. This approach has been more widely adopted in Asian countries, including China, Japan and India, where silkworms are abundantly available and more laboratories have experience in growing and maintaining larvae. In most studies the expression vectors were based on Bombyx mori nucleopolyhedrosis virus (BmNPV), which infects the silkworm Bombyx mori. However, there is at least some industrial interest in the larvae of the cabbage looper moth, Trichoplusia ni, as host for recombinant protein produc-tion by Autographa californica nuclear polyhedrosis virus (AcMNPV)-based recombinant baculovirus vectors in the United States, as well. Protein expression levels in baculovirus-infected larvae can be very high, reducing costs for large-scale production. Nevertheless, due to a general unfamiliarity with larval systems and ready access to cell culture facilities, this approach has not gained widespread popularity in most molecular biology laboratories in North America and Europe.

Baculovirus-mediated gene delivery in mammalian cellsThe successful use of recombinant baculoviruses to direct gene expres-sion in mammalian cells was first reported ten years ago93,94. Since we reviewed this subject95 there has been a remarkable increase in pub-lished reports of the use of this system. A number of publications have focused on improvements and demonstration of new cell types suscep-tible to baculovirus transduction, but most have described applications of this technology in areas such as genomics, pharmaceutical screening assays and in vivo applications such as gene therapy. In this section, we discuss advances that have been reported in the past few years.

Host cells and transduction parameters. The use of recombinant baculoviruses containing mammalian cell–active expression cas-settes, commonly referred to as BacMam viruses, for gene delivery to mammalian cells was first demonstrated in cells of liver origin93,94. Subsequently, a number of labs reported gene delivery to a broad range of nonhepatic cell lines and primary cells96–98. Our 2002 review95 contains a table of reported susceptible cells; however, there have been many recent additions. Primary rat chondrocytes are efficiently transduced by baculovirus99 and the transduced cells retain their differentiated state. Mouse primary kidney cells can express genes delivered by baculovirus for up to 20 d100. Hepatic stellate cells from rat and human are transduced at greater than 90% efficiency when the cells are activated by culturing on plastic surfaces, although transduc-tion of fresh cultures is quite low (<20%)101. Human osteosarcoma cell lines have been shown to be extremely good hosts for BacMam-mediated gene delivery102–104 surpassing even hepatoma cell lines in

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


the observed level of gene expression. The efficient transduction of human mesenchymal stem cells has recently been demonstrated105. Though not mammalian, a number of fish cell lines also can be trans-duced at a low level that is improved by the addition of butyrate106. Furthermore, virus can be injected into the embryos of zebrafish to direct gene expression in vivo107.

One major advantage of the BacMam system is that gene delivery can be accomplished in many different cell types by simply adding a viral inoculum (Fig. 3). However, certain cell types that are not effi-ciently transduced can no doubt benefit by optimization of transduc-tion conditions. Several parameters have been explored to improve transduction of HeLa cells108. Dilution of the virus inoculum with phosphate buffered saline rather than mammalian cell growth media as well as prolonged incubation of cells with the virus at low tempera-ture (25 °C rather than 37 °C) led to increases in gene expression. The addition of virus to HeLa cells in mid-log phase, followed by further additions of virus after several days allowed recombinant gene expres-sion to be prolonged for over 10 d109. The addition of butyrate, a non-specific inhibitor of histone deacetylase, has been shown to increase BacMam-mediated gene expression in a number of cell types98,110. Butyrate addition and incubation of cells at lower temperatures (34 °C rather than 37 °C) were used to increase protein yields in transduced Chinese hamster ovary (CHO) cells grown in suspension culture110. These investigators employed a modified CHO cell line that expresses the adenovirus E1A gene to further activate the CMV immediate early promoter and found that yields of micrograms of protein per ml of culture could be achieved. Enhancement of expression in trans-fected cells was observed using plasmids containing either the CMV or Drosophila melanogaster hsp70 gene promoter together with the baculoviral hr1 sequence111. A modest two- to threefold improvement in expression was observed in baculovirus transduction of mamma-lian cells when the hr1 sequence was added to the hsp70 promoter. A comparison of BacMam gene delivery efficiency using four different promoters and four cell lines, in the presence and absence of the histone deacetylase inhibitor trichostatin A, demonstrated the cell line–dependent nature of the transduction process112.

Baculovirus entry. The entry mechanism and fate of baculovirus in mammalian cells are not well understood95, although some insights have been gained in recent years. When rat hepatocytes were cultured on collagen in medium containing DMSO to produce tight islands of cells, only cells on the periphery were transduced with baculovirus113. EGTA treatment disrupts the tight junctions, causes internalization of tight junction proteins, and allows the virus to transduce internal cells in these islands114. These findings imply that baculoviruses access the hepatocytes through the basolateral surface, which has implications for in vivo gene delivery to polarized epithelia. The fate of the virus after cell entry has been probed with a novel capsid protein-GFP fusion virus83. In cells that are transduced efficiently, the labeled viral nucleocapsid can be visualized in the nucleus by 4 h after transduction. However, in cells that are not transduced, the capsids do not travel to the nucleus suggesting that in certain cell types a defect in nuclear transport may block efficient baculovirus-mediated gene delivery.

The ability of the baculovirus envelope protein gp64 to accept fusions of other proteins and display them on the virion surface, as discussed above, has been exploited in attempts to target BacMam transduction of particular cells76. In general these attempts have led to increased binding of virus to cells, but no increased gene expression as compared to unmodified virions. However, a virus containing a fusion of avidin to gp64 has been shown to provide a higher transduction frequency in some cells115. A combination approach to improved transduction is

suggested by these viral protein fusion technologies76. Fusions to the viral surface protein can be used to target virus to certain cell types, whereas fusion of nuclear targeting sequences to the capsid protein might facilitate nuclear delivery of the viral DNA.

In vivo gene delivery. BacMam viruses would appear to be ideal vec-tors for in vivo applications because they cannot replicate in mam-malian cells, but can efficiently deliver genes into many cells types. Unfortunately, theory runs headlong into the reality that baculovi-ruses are rapidly inactivated by human serum complement116. A num-ber of methods have been demonstrated to overcome this limitation and allow systemic delivery of virus in vivo117. After it was observed that a soluble form of complement receptor type 1 (sCR1) would protect baculovirus from complement inactivation in vitro118, sCR1 and virus were injected into the portal vein of mice, resulting in trans-gene expression in the livers of treated animals119. In these studies significant toxicity associated with viral delivery was attributed to the possible induction of inflammatory cytokines produced in response to the virus. It has been suggested that baculovirus particles pseudo-typed with the vesicular stomatitis virus (VSV)-G protein are more resistant to complement than unmodified virus120. Viruses containing VSV-G were, in fact, more resistant to inactivation by human, rabbit, guinea pig, hamster and mouse serum, but not rat121. This modified virus could be used for in vivo gene transfer to cells in the cerebrum and testis, though no comparisons were made with the unmodified virus to assess improvements in efficiency. Another method used to accomplish in vivo gene delivery by baculovirus is to evade the com-plement system by careful choice of the route of administration117. Stereotaxic injection of virus into the striatum and vitreous body of rats resulted in transgene expression in neuronal cells distant from the injection site demonstrating that the virus can be transported through the axons of neurons122. A comparison of transgene delivery

SO 2-U KHB

S 2-soa392 KEH

Figure 3 Photomicrograph of mammalian cells transduced with a BacMam virus expressing GFP. Virus as described in Condreay et al.98. The cells were transduced with 100 plaque-forming units of virus per cell and photographed 24 hrs after virus addition. U-2 OS, human osteosarcoma, BHK, baby hamster kidney, HEK 293, human embryonic kidney andSaos-2, human osteosarcoma cells. Transduction frequency of these cell types is routinely greater than 90% as measured by the number of fluorescent green cells.

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


by baculovirus and adenovirus after injection into rat brain revealed that BacMam vectors targeted the choroid plexus, while adenovirus delivered genes to glial cells123.

Baculoviruses as immunizing reagentsThe use of baculoviruses to express antigens under the control of mam-malian cell–active promoters and elicit immune responses in vivo was first demonstrated by Aoki and colleagues124. Intramuscular inocula-tion with virus that expressed the pseudorabies gB protein elicited an anti-gB antibody response in mice. Similarly, antigen-specific immune responses were elicited with viruses expressing the HCV E2 protein and carcinoembryonic antigen as immunogens125. A VSV-G pseudo-typed virus gave a similar response with tenfold less virus. A baculo-virus that expresses the influenza hemagglutinin protein elicited an immune response when delivered by intramuscular or interperitoneal injection126. When either the hemagglutinin virus or wild-type virus were delivered intranasally, immunity was induced in the animals that protected them from a lethal challenge of influenza virus. This phenomenon was reported previously127. Treatment with baculovirus preparations could induce interferon production in cultured cells thus protecting the cells from infection by VSV. Similarly, mice injected with baculovirus became protected from a lethal challenge with encepha-lomyocarditis virus.

Gene function studies. The broad range of host cells susceptible to BacMam transduction, including primary cell cultures95, make the BacMam approach an excellent choice for in vitro studies of gene function. Baculovirus gene delivery has been used to deliver herpes virus genes to cells to elucidate their function128,129. Gene products can be studied in isolation from the parental herpes virus to determine their contribution to processes that happen during a normal infec-tion, such as the mechanism behind the herpes virus US3-mediated block of apoptosis130. Baculoviruses delivering truncated forms131 or mutant forms132 of glycoprotein D have been used to complement glycoprotein D–defective herpes viruses and dissect domains of the

glycoprotein D protein that are involved in different events in infec-tion. Delivery of the CMV IE2 gene with a recombinant baculovirus facilitated studies on the role of this gene product in regulating viral gene expression133 and cellular gene expression that might be involved in viral pathogenicity134.

The in vitro study of hepatitis B virus (HBV) and HCV virus is hin-dered by the lack of cell culture systems to propagate these viruses. For HBV this problem was overcome by using a baculovirus to deliver the HBV genome to hepatoma cells and launch a productive HBV infection135. This approach has been used to investigate the action of anti-viral compounds on HBV infection136, probe the characteristics of drug-resistant mutants of HBV137 and characterize the reinitiation of viral infection after anti-viral treatment138. Analogous to HBV deliv-ery, baculoviruses have been used to deliver a cDNA copy of the HCV genome to cells139. Although HCV RNA replication or virions were not detected, the HCV polyprotein was produced and properly processed. McCormick and colleagues have extended this observation and pro-duced baculoviruses with inducible expression of HCV genomes to study viral replication140,141.

The discovery that BacMam viruses could be used to deliver genes to specific regions of the developing zebrafish has made it possible to elucidate gene function by observing the effects of misexpressing pro-teins in the whole animal107,142. When virus is injected into early-stage embryos, transgene expression is disseminated, but when it is injected at later stages, expression is limited to the injection site107.

RNA interference has become a widely used tool to study the effects of reducing the expression of targeted gene products. A prerequisite for successful RNA interference is efficient delivery of the interfering RNA. Recently a baculovirus engineered to express short hairpin RNA from the U6 promoter has been used to reduce expression of lamin A/C RNA and protein in human cell lines (Saos2, HepG2 and Huh7) and primary hepatic stellate cells143. These results indicate that baculoviruses provide another useful delivery approach for delivering interfering RNAs.

Cell-based assays. The development of cell-based assays for high-throughput screening of chemical libraries has traditionally involved using stable cell lines producing the target protein(s). In some instances developing and culturing these cell lines is problematic due to the del-eterious effects of the target gene product(s). There are a number of recent reports illustrating how BacMam-mediated transient expression can be used to overcome this difficulty for cell-based assay development in automated facilities. Assays to screen for modulators of protein activ-ity have been described for ion channels144, nuclear receptors104,145 and GPCRs146,147. We have previously mentioned viral replication assays suitable for this purpose136. Jenkinson et al.148. have described a novel assay using BacMam-delivered human immunodeficiency virus gene products to mimic the receptor-mediated membrane fusion event required for viral infection that is amenable to a high-throughput screening format. Another interesting application involves using fixed monolayers of BacMam-GCPR–transduced cells to screen for antibod-ies destined for immunohistochemistry studies149. The versatility of the BacMam system for pharmaceutical screening assays is illustrated in Figure 4. Large libraries of these viral reagents can be easily generated making gene delivery to the proper cell type for assay a simple liquid delivery step that can be performed on automated platforms.

Conclusions and future developmentsThe baculovirus-insect cell expression system has proven to be an extremely valuable tool for recombinant protein production. Ongoing improvements in vector design and simplification of recombinant virus isolation techniques, combined with the relative ease of small

Recombinantbaculovirus DNA

Sf9 insect cells

Automated assay platform

Target protein virus stocks

GPCRs ICs NRs Transduced cells

Figure 4 Production of BacMam viruses expressing G protein–coupled receptor (GPCR), ion channel (IC) and nuclear receptor (NR) target proteins and transduction of mammalian cells in microtiter plate-based high-throughput assay formats. The system allows for a high level of flexibilityin terms of target proteins, cell types and assay formats.

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


and large scale culture of insect cells, have resulted in widespread use of this system. Many laboratories are beginning to automate the pro-duction of large numbers of viruses and protein production schemes using advanced cloning methods, robotic liquid handling and protein purification instruments. Genetic transformation has been used to create transgenic insect cells with humanized protein glycosylation pathways, which can be used as improved hosts for baculovirus expression vectors, enabling the production of more authentic mam-malian N-glycoproteins. In the future, we can expect to see analogous transgenic insect cell lines engineered in other ways to improve their protein processing capabilities and enhance their ability to produce properly folded and modified recombinant proteins. In addition, the application of the baculovirus–insect cell system for the production of VLPs and functional multi-subunit complexes will continue to provide reagents that are difficult to produce in any other way.

Perhaps the most unexpected development discussed in this review is the increasing use of recombinant baculoviruses as gene delivery vectors for mammalian cells. We believe this application will ultimately become as commonplace as the use of recombinant baculoviruses for recombinant protein production. This relatively new gene delivery approach offers a number of advantages including: the inability of the virus to replicate in mammalian cells, the absence or virtual absence of cytotoxicity, technical simplicity and a superior biosafety profile150 as compared to mammalian cell–derived viral vectors. The character-ization of this gene delivery application is in its infancy and it will be especially important to gain additional understanding of the mecha-nisms involved in viral transduction. An increased knowledge of the baculovirus-cell interactions that result in efficient gene transfer will direct future efforts to enhance vector design and/or engineer cells to increase transduction efficiency. Finally, the application of baculodis-play technologies may also prove useful to improve and extend the variety of host cells that can be efficiently transduced.

ACKNOWLEDGMENTST.A.K. and J.P.C. thank John Gray, Mike Romanos and John Reardon at GlaxoSmithKline for their continued support and encouragement of the development and application of baculovirus technology. D.L.J. thanks the National Institutes of Health (GM49734) and the National Science Foundation (BES9814157 and BES9818001) for generously supporting baculovirus–insect cell work in his laboratory. The authors appreciate the assistance of James Frye in preparing the figures.

COMPETING INTERESTS STATEMENTThe authors declare competing financial interests (see the Nature Biotechnology website for details).


1. Luckow, V.L. & Summers, M.D. Trends in the development of baculovirus expression vectors. Bio/Technology 6, 47–55 (1988).

2. O’Reilly, D.R., Miller, L.K. & Luckow, V.A. Baculovirus Expression Vectors: A Laboratory Manual (W.H. Freeman, New York, 1992).

3. Miller, L.K. (ed.). The Baculoviruses (Plenum Press, New York, 1997).4. Possee, R.D. Baculoviruses as expression vectors. Curr. Opin. Biotechnol. 8, 569–

572 (1997).5. Kost, T.A. & Condreay, J.P. Recombinant baculoviruses as expression vectors for

insect and mammalian cells. Curr. Opin. Biotechnol. 10, 428–433 (1999).6. Freisen, D.P. & Miller, L.K. Insect viruses. in Fields’ Virology, edn. 4, (eds. Knipe,

D.M. et al.) 1871–1940 (Lippincott Williams and Wilkins, Philadelphia, 2001).7. Kitts, P.A., Ayres, M.D. & Possee, R.D. Linearization of baculovirus DNA enhances

the recovery of recombinant virus expression vectors. Nucleic Acids Res. 18, 5667–5672 (1990).

8. Kitts, P.A. & Possee, R.D. A method for producing recombinant baculovirus expres-sion vectors at high frequency. Biotechniques 14, 810–817 (1993).

9. Luckow, V.A., Lee, S.C., Barry, G.F. & Olins, P.O. Efficient generation of infectious recombinant baculoviruses by site-specific transposon-mediated insertion of for-eign genes into a baculovirus genome propagated in Escherichia coli. J. Virol. 67, 4566–4579 (1993).

10. Airenne, K.J., Peltomaa, E., Hytönen, V.P., Laitinen, O.H. & Ylä-Herttuala, S. Improved generation of recombinant baculovirus genomes in Escherichia coli. Nucleic Acids Res. 31, e101 (2003).

11. Laitinen et al. A multipurpose vector system for the screening of libraries in bacteria, insect and mammalian cells and expression in vivo. Nucleic Acids Res. 33, e42 (2005).

12. Pijlman, G.P., van den Born, E., Martens, D.E. & Vlak, J.M. Autographa californica baculoviruses with large genomic deletions are rapidly generated in infected insect cells. Virology 283, 132–138 (2001).

13. Pijlman, G.P., van Schijndel, J.E. & Vlak, J.M. Spontaneous excision of BAC vector sequences from bacmid-derived baculovirus expression vectors upon passage in insect cells. J. Gen. Virol. 84, 2669–2678 (2003).

14. Zhao, Y., Chapman, D.A. & Jones, I.M. Improving baculovirus recombination. Nucleic Acids Res. 31, e6 (2003).

15. Chambers, S.P., Austen, D.A., Fulghum, J.R. & Kim, W.M. High-throughput screen-ing for soluble recombinant expressed kinases in Escherichia coli and insect cells. Protein Expr. Purif. 36, 40–47 (2004).

16. Page, R. et al. Scalable high-throughput micro-expression device for recombinant proteins. Biotechniques 37, 364–368 (2004).

17. Bahia, D., Cheung, R., Buchs, M., Geisse, S. & Hunt, I. Optimization of insect cell growth in deep-well blocks: development of a high-throughput insect cell expression screen. Protein Expr. Purif. 39, 61–70 (2005).

18. Kitts, P.A. & Green, G. An immunological assay for determination of baculovirus titers in 48 hours. Anal. Biochem. 268, 173–178 (1999).

19. Kwon, M.S., Dojima, T., Toriyama, M. & Park, E.Y. Development of an antibody-based assay for determination of baculovirus titers in 10 hours. Biotechnol. Prog. 18, 647–651 (2002).

20. Mulvania, T., Hayes, B. & Hedin, D. A flow cytometric assay for rapid, accurate determination of baculovirus titers. BioProcessing J. 3, 47–53 (2004).

21. Shen, C.F., Meghrous, J. & Kamen, A. Quantitation of baculovirus particles by flow cytometry. J. Virol. Methods 105, 321–330 (2002).

22. Lo, H.R. & Chao, Y.C. Rapid titer determination of baculovirus by quantitative real-time polymerase chain reaction. Biotechnol. Prog. 20, 354–360 (2004).

23. Jarvis, D.L. Baculovirus expression vectors. in The Baculoviruses. (ed. Miller, L.K.) 389–431 (Plenum Press, New York, 1997).

24. Marz, L., Altmann, F., Staudacher, E. & Kubelka, V. Protein glycosylation in insects. in Glycoproteins, vol. 29a. (eds. Montreuil, J., Vliegenthart, J.F.G. & Schachter, H.) 543–563 (Elsevier, Amsterdam, 1995).

25. Marchal, I., Jarvis, D.L., Cacan, R. & Verbert, A. Glycoproteins from insect cells: sialylated or not? Biol. Chem. 382, 151–159 (2001).

26. Altmann, F., Schwihla, H., Staudacher, E., Glossl, J. & Marz, L. Insect cells contain an unusual, membrane-bound β-N-acetylglucosaminidase probably involved in the processing of protein N-glycans. J. Biol. Chem. 270, 17344–17349 (1995).

27. Jarvis, D.L., Kawar, Z.S. & Hollister, J.R. Engineering N-glycosylation pathways in the baculovirus-insect cell system. Curr. Opin. Biotechnol. 9, 528–533 (1998).

28. Jarvis, D.L. Humanizing recombinant glycoprotein production in the baculovirus-insect cell expression system. Virology 310, 1–7 (2003).

29. Tomiya, N., Betenbaugh, M.J. & Lee, Y.C. Humanization of lepidopteran insect-cell-produced glycoproteins. Acc. Chem. Res. 36, 613–620 (2003).

30. Tomiya, N., Narang, S., Lee, Y.C. & Betenbaugh, M.J. Comparing N-glycan process-ing in mammalian cell lines to native and engineered lepidopteran insect cell lines. Glycoconj. J. 21, 343–360 (2004).

31. Betenbaugh, M.J., Tomiya, N., Narang, S., Hsu, J.T. & Lee, Y.C. Biosynthesis of human-type N-glycans in heterologous systems. Curr. Opin. Struct. Biol. 14, 601–606 (2004).

32. Jarvis, D.L., Fleming, J.A., Kovacs, G.R., Summers, M.D. & Guarino, L.A. Use of early baculovirus promoters for continuous expression and efficient processing of foreign gene products in stably transformed lepidopteran cells. Bio/Technology 8, 950–955 (1990).

33. Jarvis, D.L., Weinkauf, C. & Guarino, L.A. Immediate early baculovirus vectors for foreign gene expression in transformed or infected insect cells. Protein Expr. Purif. 8, 191–203 (1996).

34. Hollister, J.R., Shaper, J.H. & Jarvis, D.L. Stable expression of mammalian beta 1,4-galactosyltransferase extends the N-glycosylation pathway in insect cells. Glycobiology 8, 473–480 (1998).

35. Hollister, J. & Jarvis, D.L. Engineering lepidopteran insect cells for sialoglycoprotein production by genetic transformation with mammalian β1,4-galactosyltransferase and a2,6-sialyltransferase genes. Glycobiology 11, 1–9 (2001).

36. Hooker, A.D. et al. Constraints on the transport and glycosylation of recombinant IFN-gamma in chinese hamster ovary and insect cells. Biotech. Bioeng. 63, 559–572 (1999).

37. Tomiya, N., Ailor, E., Lawrence, S.M., Betenbaugh, M.J. & Lee, Y.C. Determination of nucleotides and sugar nucleotides involved in protein glycosylation by high-per-formance anion-exchange chromatography: sugar nucleotide contents in cultured insect cells and mammalian cells. Anal. Biochem. 293, 129–137 (2001).

38. Hollister, J., Conradt, H.O. & Jarvis, D.L. Evidence for a sialic acid salvaging pathway in lepidopteran insect cell lines. Glycobiology 13, 487–495 (2003).

39. Hollister, J., Grabenhorst, E., Nimtz, M., Conradt, H.O. & Jarvis, D.L. Engineering the protein N-glycosylation pathway in insect cells for production of biantennary, complex N-glycans. Biochemistry 41, 15093–15104 (2002).

40. Aumiller, J.J., Hollister, J.R. & Jarvis, D.L. A transgenic lepidopteran insect cell line engineered to produce CMP-sialic acid and sialoglycoproteins. Glycobiology 13, 497–507 (2003).

41. Ailor, E. & Betenbaugh, M.J. Modifying secretion and post-translational processing in insect cells. Curr. Opin. Biotechnol. 10, 142–145 (1999).

42. Higgins, M.K., Demir, M. & Tate, C.G. Calnexin co-expression and the use of weaker promoters increase expression of correctly assembled shaker potassium channel in

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


insect cells. Biochim. Biophys. Acta 1610, 124–132 (2003).43. Zhang, L., Wu, G., Tate, C.G., Lookene, A. & Olivecrona, G. Calreticulin promotes

folding/dimerization of human lipoprotein lipase expressed in insect cells. (Sf21). J. Biol. Chem. 278, 29344–29351 (2003).

44. Fourneau, J.M., Cohen, H. & van Endert, P.M. A chaperone assisted high yield sys-tem for the production of HLA-DR4 tetramers in insect cells. J. Immunol. Methods 285, 253–264 (2004).

45. Yokoyama, N. et al. Co-expression of human chaperones Hsp70 and Hsdj or Hsp40 cofactor increases solubility of overexpressed target proteins in insect cells. Biochim. Biophys. Acta 1493, 119–124 (2000).

46. Martinez-Torrecuadrada, J.L. et al. An efficient expression system for the production of functionally active human LKB1. J. Biotechnol. 115, 23–34 (2005).

47. Venkaiah, B., Viswanathan, P., Habib, S. & Hasnain, S.E. An additional copy of the homologous region (hr1) sequence in the Autographa californica multinucleocapsid polyhedrosis virus genome promotes hyperexpression of foreign genes. Biochemistry 43, 8143–8151 (2004).

48. Chen, Y. et al. A constitutive super-enhancer: homologous region 3 of bombyx mori nucleopolyhedrosis virus. Biochem. Biophys. Res. Commun. 318, 1039–1044 (2004).

49. Sano, K., Maeda, K., Oki, M., & Maeda, Y. Enhancement of protein expression in insect cells by a lobster tropomysin cDNA leader sequence. FEBS Lett. 532, 143–146 (2002).

50. Ho, Y., Lo, H.R., Lee, T.C., Wu, C.P.Y. & Chao, Y.C. Enhancement of correct protein folding in vivo by a non-lytic baculovirus. Biochem. J. 382, 695–702 (2004).

51. Kaba, S.A., Salcedo, A.M., Wafula, P.O., Vlak, J.M. & van Oers, M.M. Development of a chitinase and v–cathepsin negative bacmid for improved integrity of secreted recombinant proteins. J. Virol. Methods 122, 113–118 (2004).

52. Sarramegna, V., Talmont, P., Demange, P. & Milon, A. Heterologous expression of G-protein-coupled receptors: comparison of expression systems from the standpoint of large-scale production and purification. Cell. Mol. Life Sci. 60, 1529–1546 (2003).

53. Massotte, D. Protein-coupled receptor overexpression with the baculovirus-insect cell expression system: a tool for structural and functional studies. Biochim. Biophys. Acta 1610, 77–89 (2003).

54. Gazi, L., Nickolls, S.A. & Strange, P.G. Functional coupling of the human dopamine D2 receptor with Gαi1, Gαi2, Gαi3 and Gα0 G proteins: evidence for agonist regulation of G protein selectivity. Br. J. Pharmacol. 138, 775–786 (2003).

55. Hood, S.R., Shah, G. & Jones, P. Expression of cytochrome P450 in a baculovirus system. Methods Mol. Biol. 107, 203–218 (1998).

56. Rushmore, T.H., Reider, P.J., Slaughter, D., Assang, C. & Shou, M. Bioreactor sys-tems in drug metabolism: synthesis of cytochrome P450-generated metabolites. Metab. Eng. 2, 115–125 (2000).

57. Sanchez, R.I. et al. Cytochrome p450 3a4 is the major enzyme involved in the metabolism of the substance P receptor antagonist aprepitant. Drug Metab. Dispos. 32, 1287–1292 (2004).

58. Ikonomou, L., Schneider, Y.J. & Agathos, S.N. Insect cell culture for industrial pro-duction of recombinant proteins. Appl. Microbiol. Biotechnol. 62, 1–20 (2003).

59. Maranga, L., Cruz, P.E., Aunins, J.A. & Carrondo, M.J.T. Production of core and virus-like-particles with baculovirus infected insect cells. Adv. Biochem. Eng. Biotechnol. 74, 183–206 (2002).

60. Noad, R. & Roy, P. Virus-like particles as immunogens. Trends Microbiol. 11, 438–444 (2003).

61. Garcea, R.L. & Gissman, L. Virus-like particles as vaccines and vessels for the delivery of small molecules. Curr. Opin. Biotechnol. 15, 513–517 (2004).

62. Petry, H., Goldmann, C., Ast, O. & Luke, W. The use of virus-like particles for gene transfer. Curr. Opin. Mol. Ther. 5, 524–528 (2003).

63. Berger, I., Fitzgerald, D.J. & Richmond, T.J. Baculovirus expression system for heterologous multiprotein complexes. Nat. Biotechnol. 22, 1583–1587 (2004).

64. Fausch, S.C., DaSilva, D.M., Eiben, G.L., LePoole, I.C. & Kast, W.M. HPV protein/peptide vaccines: from animal models to clinical trials. Front. Biosci. 8, s81–s91 (2003).

65. Stanley, M.A. Progress in prophylactic and therapeutic vaccines for human papilloma infection. Expert Rev. Vaccines 2, 381–389 (2003).

66. Harper, D.M. et al. Efficacy of a bivalent L1 virus-like particle vaccine in prevention of infection with human papilloma types 16 and 18 in young women: a randomised controlled trial. Lancet 364, 1757–1765 (2004).

67. Baumert, T.F., Ito, S., Wong, D.T. & Liang, T.J. Hepatitis C virus structural proteins assemble into virus-like particles in insect cells. J. Virol. 72, 3827–3836 (1998).

68. Xiang, J. et al. Recombinant hepatitis C virus-like particles expressed by baculovirus: utility in cell-binding and antibody detection assay. J. Med. Virol. 68, 537–543 (2002).

69. Choi, S.H., Kim, S.Y., Park, K.J., Kim, Y.J. & Hwang, S.B. Hepatitis C virus core protein is efficiently released into the culture medium in insect cells. J. Biochem. Mol. Biol. 37, 735–740 (2004).

70. Girard, C. et al. Effect of the 5′ non-translated region on self assembly of hepatitis C virus genotype 1a structural proteins produced in insect cells. J. Gen. Virol. 85, 3659–3670 (2004).

71. Jeong, S.H. et al. Immunization with hepatitis C virus-like particles induces humoral and cellular immune responses in nonhuman primates. J. Virol. 78, 6995–7003 (2004).

72. Ho, Y., Lin, P.H., Liu, C.Y.Y., Lee, S.P. & Chao, Y.C. Assembly of human severe acute respiratory syndrome coronavirus-like particles. Biochem. Biophys. Res. Commun. 318, 833–838 (2004).

73. Mortola, E. & Roy, P. Efficient assembly and release of SARS coronavirus-like particles

by a heterologous expression system. FEBS Lett. 576, 174–178 (2004).74. Urabe, M., Ding, C. & Kotin, R.M. Insect cells as a factory to produce adeno-associ-

ated virus type 2 vectors. Hum. Gene Ther. 13, 1935–1943 (2002).75. Boublik, Y., Di Bonito, P. & Jones, I.M. Eukaryotic virus display: engineering the

major surface glycoprotein of the Autographa californica nuclear polyhedrosis virus (AcNPV) for the presentation of foreign proteins on the virus surface. Bio/Technology 13, 1079–1084 (1995).

76. Oker-Blom, C., Airenne, K.J. & Grabherr, R. Baculovirus display strategies: emerging tools for eukaryotic libraries and gene delivery. Brief. Funct. Genomic. Proteomic. 2, 244–253 (2003).

77. Lindley, K. et al. Production of monoclonal antibodies using recombinant baculovirus displaying gp64-fusion proteins. J. Immunol. Methods 234, 123–135 (2000).

78. Tanaka, T. et al. The generation of monoclonal antibodies against human peroxi-some proliferator-activated receptors (PPARs). J. Atheroscler. Thromb. 9, 233–242 (2002).

79. Yoshida, S. et al. Baculovirus virions displaying Plasmodium berghei circumsprozoite protein protect mice against malaria sporozoite infection. Virology 316, 161–170 (2003).

80. Rahman, M.M., Shaila, M.S. & Gopinathan, K.P. Baculovirus display of fusion pro-tein of Pestes des petits ruminants virus and hemagglutination protein of rinderpest virus and immunogenicity of the displayed proteins in mouse model. Virology 317, 36–49 (2003).

81. Kaba, S.A., Nene, V., Musoke, A.J., Vlak, J.M. & van Oers, M.M. Fusion to green fluorescent protein improves expression of Theileria parva sporozoite surface antigen p67 in insect cells. Parasitology 125, 497–505 (2002).

82. Tami, C. et al. Immunological properties of FMDV-gp64 fusion proteins expressed on SF9 cell and baculovirus surfaces. Vaccine 23, 840–845 (2004).

83. Kukkonen, S.P. et al. Baculovirus capsid display: a novel tool for transduction imag-ing. Mol. Ther. 8, 853–862 (2003).

84. Loisel, T.P. et al. Recovery of homogeneous and functional beta 2-adrenergic recep-tors from extracellular baculovirus particles. Nat. Biotechnol. 15, 1300–1304 (1997).

85. Masuda, K. et al. A combinatorial G protein-coupled receptor reconstitution system on budded baculovirus. Evidence for Galpha and Galphao coupling to a human leukotriene B4 receptor. J. Biol. Chem. 278, 24552–24562 (2003).

86. Hayashi, I. et al. Selective reconstitution and the recovery of functional γ-secre-tase complex on budded baculovirus particles. J. Biol. Chem. 279, 38040–38046 (2004).

87. Maeda, S. et al. Production of human alpha-interferon in silkworm using a baculo-virus vector. Nature 315, 592–594 (1985).

88. Wu, D. et al. Expression of biologically active recombinant equine interferon-γ by two different expression systems using insect cells and silkworm larvae. Cytokine 20, 63–69 (2002).

89. Sehgal, D., Malik, P.S. & Jameel, S. Purification and diagnostic utility of a recombi-nant hepatitis E virus capsid protein expressed in insect larvae. Protein Expr. Purif. 27, 27–34 (2003).

90. Muneta, Y. et al. Expression and one-step purification of bovine interleukin-21 (IL-21) in silkworms using a hybrid baculovirus expression system. Biotechnol. Lett. 26, 1453–1458 (2004).

91. Nagaya, H. et al. Establishment of a large-scale purification procedure for purifica-tion of recombinant bovine interferon-tau produced by a silkworm-baculovirus gene expression system. J. Vet. Med. Sci. 66, 1395–1401 (2004).

92. Motohashi, T. Shimojima, T., Fukagawa, T., Maenaka, K. & Park, E.Y. Efficient large-scale production of larvae and pupae of silkworm by Bombyx mori nuclear polyhedrosis virus bacmid system. Biochem. Biophys. Res. Commun. 326, 564–569 (2005).

93. Hofmann, C. et al. Efficient gene transfer into human hepatocytes by baculovirus vectors. Proc. Natl. Acad. Sci. USA 92, 10099–10103 (1995).

94. Boyce, F.M. & Bucher, N.L. Baculovirus-mediated gene transfer into mammalian cells. Proc. Natl. Acad. Sci. USA 93, 2348–2352 (1996).

95. Kost, T.A. & Condreay, J.P. Recombinant baculoviruses as mammalian cell gene-delivery vectors. Trends Biotechnol. 20, 173–180 (2002).

96. Shoji, I. et al. Efficient gene transfer into various mammalian cells, including non-hepatic cells, by baculovirus vectors. J. Gen. Virol. 78, 2657–2664 (1997).

97. Yap, C.C. et al. A hybrid baculovirus-T7 RNA polymerase system for recovery of an infectious virus from cDNA. Virology 231, 192–200 (1997).

98. Condreay, J.P., Witherspoon, S.M., Clay, W.C. & Kost, T.A. Transient and stable gene expression in mammalian cells transduced with a recombinant baculovirus vector. Proc. Natl. Acad. Sci. USA 96, 127–132 (1999).

99. Ho, Y.C., Chen, H.C., Wang, K.C. & Hu, Y.C. Highly efficient baculovirus-mediated gene transfer into rat chondrocytes. Biotechnol. Bioeng. 88, 643–651 (2004).

100. Liang, C.Y., Wang, H.Z., Li, T.X., Hu, Z.H. & Chen, X.W. High efficiency gene transfer into mammalian kidney cells using baculovirus vectors. Arch. Virol. 149, 51–60 (2004).

101. Gao, R. et al. High efficiency gene transfer into cultured primary rat and human hepatic stellate cells using baculovirus vectors. Liver 22, 15–22 (2002).

102. Song, S.U. & Boyce, F.M. Combination treatment for osteosarcoma with baculoviral vector mediated gene therapy (p53) and chemotherapy (adriamycin). Exp. Mol. Med. 33, 46–53 (2001).

103. Song, S.U. et al. Effective transduction of osteogenic sarcoma cells by a baculovirus vector. J. Gen. Virol. 84, 697–703 (2003).

104. Clay, W.C. et al. Recombinant baculoviruses used to study estrogen receptor function in human osteosarcoma cells. Assay Drug Dev. Technol. 1, 801–810 (2003).

105. Ho, Y.C., Chung, Y.C., Hwang, S.M., Wang, K.C. & Hu, Y.C. Transgene expression

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


and differentiation of baculovirus-transduced human mesenchymal stem cells.J. Gene Med.; published online 11 February 2005 (doi: 10.1002/jgm.729).

106. Leisy, D.J., Lewis, T.D., Leong, J.A. & Rohrmann, G.F. Transduction of cultured fish cells with recombinant baculoviruses. J. Gen. Virol. 84, 1173–1178 (2003).

107. Wagle, M. & Jesuthasan, S. Baculovirus-mediated gene expression in zebrafish. Mar. Biotechnol. (NY) 5, 58–63 (2003).

108. Hsu, C.S., Ho, Y.C., Wang, K.C. & Hu, Y.C. Investigation of optimal transduction conditions for baculovirus-mediated gene delivery into mammalian cells. Biotechnol. Bioeng. 88, 42–51 (2004).

109. Hu, Y.C., Tsai, C.T., Chang, Y.J. & Huang, J.H. Enhancement and prolongation of baculovirus-mediated expression in mammalian cells: focuses on strategic infection and feeding. Biotechnol. Prog. 19, 373–379 (2003).

110. Ramos, L. et al. Rapid expression of recombinant proteins in modified CHO cells using the baculovirus system. Cytotechnology 38, 37–41 (2002).

111. Viswanathan, P. et al. The homologous region sequence (hr1) of Autographa cali-fornica multinucleocapsid polyhedrosis virus can enhance transcription from non-baculoviral promoters in mammalian cells. J. Biol. Chem. 278, 52564–52571 (2003).

112. Spenger, A., Ernst, W., Condreay, J.P., Kost, T.A. & Grabherr, R. Influence of pro-moter choice and trichostatin A treatment on expression of baculovirus delivered genes in mammalian cells. Protein Expr. Purif. 38, 17–23 (2004).

113. Bilello, J.P., Delaney, W.E., Boyce, F.M. & Isom, H.C. Transient disruption of inter-cellular junctions enables baculovirus entry into nondividing hepatocytes. J. Virol. 75, 9857–9871 (2001).

114. Bilello, J.P., Cable, E.E., Myers, R.L. & Isom, H.C. Role of paracellular junction complexes in baculovirus-mediated gene transfer to nondividing rat hepatocytes. Gene Ther. 10, 733–749 (2003).

115. Räty, J.K. et al. Enhanced gene delivery by avidin-displaying baculovirus. Mol. Ther. 9, 282–291 (2004).

116. Hofmann, C. & Strauss, M. Baculovirus-mediated gene transfer in the presence of human serum or blood facilitated by inhibition of the complement system. Gene Ther. 5, 531–536 (1998).

117. Huser, A. & Hofmann, C. Baculovirus vectors: novel mammalian cell gene-delivery vehicles and their applications. Am. J. Pharmacogenomics 3, 53–63 (2003).

118. Hofmann, C., Huser, A., Lehnert, W. & Strauss, M. Protection of baculovirus-vectors against complement-mediated inactivation by recombinant soluble complement receptor type 1. Biol. Chem. 380, 393–395 (1999).

119. Hoare, J., Waddington, S., Thomas, H.C., Coutelle, C. & McGarvey, M.J. Complement inhibition rescued mice allowing observation of transgene expression following intra-portal delivery of baculovirus in mice. J. Gene Med. 7, 325–333 (2005).

120. Barsoum, J., Brown, R., McKee, M. & Boyce, F.M. Efficient transduction of mam-malian cells by a recombinant baculovirus having the vesicular stomatitis virus G glycoprotein. Hum. Gene Ther. 8, 2011–2018 (1997).

121. Tani, H. et al. In vitro and in vivo gene delivery by recombinant baculoviruses.J. Virol. 77, 9799–9808 (2003).

122. Li, Y., Wang, X., Guo, H. & Wang, S. Axonal transport of recombinant baculovirus vectors. Mol. Ther. 10, 1121–1129 (2004).

123. Lehtolainen, P., Tyynela, K., Kannasto, J., Airenne, K.J. & Ylä-Herttuala, S. Baculoviruses exhibit restricted cell type specificity in rat brain: a comparison of baculovirus- and adenovirus-mediated intracerebral gene transfer in vivo. Gene Ther. 9, 1693–1699 (2002).

124. Aoki, H. et al. Induction of antibodies in mice by a recombinant baculovirus express-ing pseudorabies virus glycoprotein B in mammalian cells. Vet. Microbiol. 68, 197–207 (1999).

125. Facciabene, A., Aurisicchio, L. & La Monica, N. Baculovirus vectors elicit antigen-specific immune responses in mice. J. Virol. 78, 8663–8672 (2004).

126. Abe, T. et al. Baculovirus induces an innate immune response and confers protec-tion from lethal influenza virus infection in mice. J. Immunol. 171, 1133–1139 (2003).

127. Gronowski, A.M., Hilbert, D.M., Sheehan, K.C., Garotta, G. & Schreiber, R.D. Baculovirus stimulates antiviral effects in mammalian cells. J. Virol. 73, 9944–9951 (1999).

128. Ye, G.J., Vaughan, K.T., Vallee, R.B. & Roizman, B. The herpes simplex virus 1 U(L)34 protein interacts with a cytoplasmic dynein intermediate chain and targets nuclear membrane. J. Virol. 74, 1355–1363 (2000).

129. Zhou, G., Galvan, V., Campadelli-Fiume, G. & Roizman, B. Glycoprotein D or J delivered in trans blocks apoptosis in SK-N-SH cells induced by a herpes simplex virus 1 mutant lacking intact genes expressing both glycoproteins. J. Virol. 74, 11782–11791 (2000).

130. Benetti, L. & Roizman, B. Herpes simplex virus protein kinase US3 activates and functionally overlaps protein kinase A to block apoptosis. Proc. Natl. Acad. Sci. USA 101, 9411–9416 (2004).

131. Zhou, G. & Roizman, B. Truncated forms of glycoprotein D of herpes simplex virus 1 capable of blocking apoptosis and of low-efficiency entry into cells form a heterodi-mer dependent on the presence of a cysteine located in the shared transmembrane domains. J. Virol. 76, 11469–11475 (2002).

132. Zhou, G., Avitabile, E., Campadelli-Fiume, G. & Roizman, B. The domains of glyco-protein D required to block apoptosis induced by herpes simplex virus 1 are largely distinct from those involved in cell-cell fusion and binding to nectin1. J. Virol. 77, 3759–3767 (2003).

133. Kronschnabl, M., Marschall, M. & Stamminger, T. Efficient and tightly regulated expression systems for the human cytomegalovirus major transactivator protein IE2p86 in permissive cells. Virus Res. 83, 89–102 (2002).

134. Kronschnabl, M. & Stamminger, T. Synergistic induction of intercellular adhesion molecule-1 by the human cytomegalovirus transactivators IE2p86 and pp71 is mediated via an Sp1-binding site. J. Gen. Virol. 84, 61–73 (2003).

135. Delaney, W.E. & Isom, H.C. Hepatitis B virus replication in human HepG2 cells mediated by hepatitis B virus recombinant baculovirus. Hepatology 28, 1134–1146 (1998).

136. Delaney, W.E., Miller, T.G. & Isom, H.C. Use of the hepatitis B virus recombinant baculovirus-HepG2 system to study the effects of (–)-beta-2’,3′-dideoxy-3′-thia-cytidine on replication of hepatitis B virus and accumulation of covalently closed circular DNA. Antimicrob. Agents Chemother. 43, 2017–2026 (1999).

137. Gaillard, R.K. et al. Kinetic analysis of wild-type and YMDD mutant hepatitis B virus polymerases and effects of deoxyribonucleotide concentrations on polymerase activity. Antimicrob. Agents Chemother. 46, 1005–1013 (2002).

138. Abdelhamed, A.M., Kelley, C.M., Miller, T.G., Furman, P.A. & Isom, H.C. Rebound of hepatitis B virus replication in HepG2 cells after cessation of antiviral treatment. J. Virol. 76, 8148–8160 (2002).

139. Fipaldini, C., Bellei, B. & La Monica, N. Expression of hepatitis C virus cDNA in human hepatoma cell line mediated by a hybrid baculovirus-HCV vector. Virology 255, 302–311 (1999).

140. McCormick, C.J., Rowlands, D.J. & Harris, M. Efficient delivery and regulable expres-sion of hepatitis C virus full-length and minigenome constructs in hepatocyte-derived cell lines using baculovirus vectors. J. Gen. Virol. 83, 383–394 (2002).

141. McCormick, C.J., Challinor, L., Macdonald, A., Rowlands, D.J. & Harris, M. Introduction of replication-competent hepatitis C virus transcripts using a tetracy-cline-regulable baculovirus delivery system. J. Gen. Virol. 85, 429–439 (2004).

142. Wagle, M. et al. EphrinB2a in the zebrafish retinotectal system. J. Neurobiol. 59, 57–65 (2004).

143. Nicholson, L.J., Philippe, M., Paine, A.J., Mann, D.A. & Dolphin, C.T. RNA interfer-ence mediated in human primary cells via recombinant baculoviral vector. Mol. Ther. 11, 638–644 (2005).

144. Pfohl, J.L. et al. Titration of KATP channel expression in mammalian cells utilizing recombinant baculovirus transduction. Receptors Channels 8, 99–111 (2002).

145. Boudjelal, M., et al. The application of bacmam technology in nuclear receptor drug discovery. Biotechnol. Ann. Rev. 11, in press, (2005).

146. Ames, R. et al. BacMam recombinant baculoviruses in G protein-coupled receptor drug discovery. Receptors Channels 10, 99–107 (2004).

147. Ames, R. et al. Heterologous expression of G protein-coupled receptors in U-2 OS osteosarcoma cells. Receptors Channels 10, 117–124 (2004).

148. Jenkinson, S. et al. Development of a novel high-throughput surrogate assay to mea-sure HIV envelope/CCR5/CD4-mediated viral/cell fusion using bacmam baculovirus technology. J. Biomol. Screen. 8, 463–470 (2003).

149. Su, J.L. et al. A cell-based time-resolved fluorescence assay for selection of anti-body reagents for G protein-coupled receptor immunohistochemistry. J. Immunol. Methods 291, 123–135 (2004).

150. Kost, T.A., Condreay, J.P. & Mickelson, C.A. Biosafety and viral gene transfer vectors. in Biological Safety, Principles and Practices, edn. 3, (eds. Fleming, D.O. & Hunt, D.L.) 579–597 (ASM Press, Washington, DC, 2000).

REV IEW©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

In vivo detection of amyloid-b deposits by near-infraredimaging using an oxazine-derivative probeMartin Hintersteiner1, Albert Enz2, Peter Frey2, Anne-Lise Jaton2, Willy Kinzy1, Rainer Kneuer1,Ulf Neumann2, Markus Rudin1, Matthias Staufenbiel2, Markus Stoeckli1, Karl-Heinz Wiederhold2 &Hans-Ulrich Gremlich1

As Alzheimer’s disease pathogenesis is associated with the formation of insoluble aggregates of amyloid b-peptide, approaches

allowing the direct, noninvasive visualization of plaque growth in vivo would be beneficial for biomedical research. Here we

describe the synthesis and characterization of the near-infrared fluorescence oxazine dye AOI987, which readily penetrates the

intact blood-brain barrier and binds to amyloid plaques. Using near-infrared fluorescence imaging, we demonstrated specific

interaction of AOI987 with amyloid plaques in APP23 transgenic mice in vivo, as confirmed by postmortem analysis of brain

slices. Quantitative analysis revealed increasing fluorescence signal intensity with increasing plaque load of the animals, and

significant binding of AOI987 was observed for APP23 transgenic mice aged 9 months and older. Thus, AOI987 is an attractive

probe to noninvasively monitor disease progression in animal models of Alzheimer disease and to evaluate effects of potential

Alzheimer disease drugs on the plaque load.

The clinical characteristics of Alzheimer disease are dementia, cogni-tive impairment and memory loss. On a histological level, theaccumulation and deposition of amyloid-b peptides Ab1-40 andAb1-42 into amyloid plaques is considered an important hallmarkin Alzheimer disease pathogenesis1–3. These Ab peptides result fromcleavage of amyloid precursor protein (APP) by proteases called b-and g-secretases. Another pathomorphological hallmark of Alzheimerdisease are neurofibrillary tangles composed of the hyperphosphory-lated, microtubule-associated protein tau (for reviews, see refs. 2,3).Today, clinical diagnosis of Alzheimer disease is based on mental andcognitive examinations of patients4, and definitive confirmation of thedisease is obtained only postmortem by histopathological examinationof brain tissue for amyloid plaques and neurofibrillary tangles. There-fore, approaches allowing the direct visualization of the growth of theamyloid plaque load (and neurofibrillary tangles) in vivo would bebeneficial for the assessment of the disease status.

Imaging of amyloid plaques using structural imaging techniques,such as magnetic resonance imaging (MRI), faces high demandson spatial resolution required for the identification of individualplaques (diameter r50 mm). Although feasible in fixed humanbrain specimens5 and in vivo in transgenic mouse models of Alzheimerdisease6, these approaches are not feasible in the clinic. In addition,the sensitivity of MRI is low due to a small specific contrastbetween plaques and surrounding tissue. Imaging of amyloid plaqueswith high sensitivity, that is, high contrast between amyloid plaque–associated signals and the background, critically depends on thedevelopment of contrast agents that specifically enhance amyloidplaque signals. Such agents would be valuable for early diagnosis

and monitoring of disease progression, as well as for the evaluationof therapeutic interventions. A number of promising candidatecompounds have been described7,8, including plaque-specificligands for positron emission tomography (PET)9–16, for MRI17–19

and for optical imaging (multiphoton microscopy)20–23. Multiphotonmicroscopy is invasive, probes only small fields-of-view, and, hence,is of limited use for in vivo applications. MRI-based approaches havelow sensitivity and low blood-brain barrier penetration of the bulkyMRI ligands. PET is limited, particularly for animal research, by theshort half-life of positron-emitting nuclei and the narrow availabilityof the technology. Nevertheless, with regard to clinical applicationPET currently seems to be the most promising approach24,25. Analternative imaging technology exploits the low absorption coefficientsof light in the near-infrared region of the electromagnetic spectrum.For wavelengths l between 650 and 900 nm, light attenuation istypically one order of magnitude per 10 mm of tissue26. With theavailability of dyes that absorb and fluoresce in this spectral domain,near-infrared fluorescence (NIRF) imaging has become a powerfuland inexpensive tool for noninvasive imaging of target-specific inter-actions in animals27,28.

We have synthesized and characterized the oxazine derivativeAOI987 as an NIRF ligand with the properties required to targetamyloid plaques in vivo. Administration of the dye to APP23 trans-genic mice overexpressing human APP, which develop cerebral amy-loid plaque deposits starting at an age of 6 months29, revealed specificbinding to amyloid plaques: the fluorescence intensity was signifi-cantly higher in APP23 mice as compared with age-matched litter-mates. Moreover, quantitative image analysis demonstrated that the

Published online 17 April 2005; doi:10.1038/nbt1085

1Discovery Technologies, Novartis Institutes for Biomedical Research, CH-4002 Basel, Switzerland. 2Nervous System Department, Novartis Institutes for BiomedicalResearch, CH-4002 Basel, Switzerland. Correspondence and requests for material should be addressed to H.U.G. ([email protected]).

NATURE BIOTECHNOLOGY VOLUME 23 NUMBER 5 MAY 2005 57 7

A RT I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

integrated fluorescence signal reflects the amyloid plaque load ofAPP23 mice.

RESULTS

Synthesis and fluorescence properties of oxazine dyes

NIRF agents for in vivo imaging of amyloid plaques should becharacterized by low molecular weight (o500 Da) and high lipo-philicity (logP 1-4) to be able to cross the blood-brain barrier insufficient amounts for detection. They should exhibit high affinityfor and specific binding to amyloid plaques without being rapidly

degraded. Moreover, fluorescence emissionmaxima should lie in the range 650 o l o900 nm. We hypothesized that ligands basedon a core structure of bridged oxazine dyes30

should fulfill these requirements. Furtherimprovements of solubility and a favorablebathochromic shift of the emission maximumare achieved by incorporation of a hetero-atom in position 1 of the terminal ring inoxazine dyes (data not shown).

We prepared a series of four novel oxazinedyes by a two-step procedure (Fig. 1), withazo or nitroso derivatives serving as keyintermediates. The nitroso derivative ofbuilding block 1a was found to be not acces-sible via nitrosation of 1a, whereas nitrosationof 1b–d proceeded smoothly to give 2b–d.However, reaction of building block 1a withp-nitrobenzenediazonium; tetrafluoroborateyielded azo derivative 2a, an appropriateintermediate for the preparation of hetero-

cyclic oxazine dye 3a. All dyes within this series showed promisingNIRF properties (Fig. 2a). The fluorescence emission maximumslightly increased within the series from 670 nm for AOI987 (dye3a, X ¼ O) to 695 nm for ASG236 (dye 3b, X ¼ S), whereas thequantum yield in mouse serum decreased from 41% to 13%.Three factors made AOI987 (dye 3a) the most interesting dye inthis series: one single absorption and emission maximum at 650and 670 nm, respectively, a moderate absorption coefficient of64,570 M�1cm�1 and most importantly, a very high quantum yieldof 41% (in mouse serum).

X

O OH

–

–

–

–

–

–

OH

OH

OH

OHOH

CH3

CH3

CH3

CH3

CH3

CH3

CH2CH3

C6H4N2O2

CH2CH3

1a

1b

1c

1d

2a

2b

2c2d

O

O OO

HO NN

ON N+

OO

O

S

C

C

CC

S

R1 R2 R3 X

N

H+

MeOHN+

N

N+O

O

orNaNO2

R1

R2

1a–d

X

N

N

R1

R3

R2

2a–d

1a

O

HO N1a

O

HO N1a

O

HO N

1a

BF4–

NaBF4

HCI, EtOH

NaBF4

HCI, EtOH3a

AOI-987

ON

ON N+

BF4–

3c

ASG-237

S ON

ON N+

BF4–

3b

ASG-236

ON

ON N+

BF4–

3d

AMQ-987

Figure 1 General synthesis scheme. A series of novel oxazine dyes 3a–d has been prepared in a

two-step procedure comprising the synthesis of azo intermediate 2a or nitroso derivatives 2b–d and

condensation thereof with 4-methyl-3,4-dihydro-2H-benzo[1,4]oxazin-6-ol (1a).

Solvent λmax(nm) λf (nm) φ (%) c (M/I)ε (I/Mcm)

3a 644 61

61

41

28

28

13

670 1.526 × 10–6

1.531 × 10–6

6.706 × 10–6

5.828 × 10–6

1.344 × 10–6

1.706 × 10–6

670

695

695

677

677

64,570

61,930

13,920

12,880

67,430

56,700

650

659

665

649

658

MeOH

MeOH

MeOH

3a Serum

Serum

Serum

3b

3b

3c

3c

35

30

25

20

15

10

5

0

–5

∆F

∆Fm

ax −

∆F

680

400

350

300

250

200

150

100

460 480 500 520 540 560

50

0700 720 740

Emission wavelength (nm)

876543210

0 1 2 3 4 5 6 7AOI987 (µM)

Kd = 0.22 ± 0.13 µM

Thioflavin (3 µM)

+ 0.125 µM AOI987AOI987AOI987AOI987

+ 0.375 µM+ 1.120 µM+ 3.250 µM

+ aggregated Aß

Flu

ores

cenc

e in

tens

ity

Wavelength (nm)

a

d e

b c

Figure 2 Fluorescence characterization and in vitro binding studies. (a) Fluorescence properties of dyes 3a–c in methanol and mouse serum.

(b) Colocalization of plaques, fluorescently labeled with AOI987 with the optical image of the same region in a cryostat brain section. The results

demonstrate that AOI987 labels Ab plaques with high sensitivity and clarity. Scale bars, 1 mm (large panel) and 100 mm (lower panels). (c) In vitro staining

of amyloid plaques labeled with AOI987 (left; dye 3a), ASG236 (middle; dye 3b) and ASG237 (right; dye 3c) at two different concentrations (0.001%

upper row and 0.0001% lower row). Scale bar, 100 mm. (d) Difference fluorescence emission spectra of AOI987 (dye 3a) in buffer and bound to aggregated

Ab. Excitation wavelength was 650 nm. Fluorescence intensity data at 705 nm were used for calculation of binding constants (inset). (e) Thioflavin T

displacement from the thioflavin-Ab-complex (open circles) by increasing amounts of AOI987 (dye 3a). Free thioflavin shows no fluorescence (filled squares).

5 78 VOLUME 23 NUMBER 5 MAY 2005 NATURE BIOTECHNOLOGY

A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

In vitro binding properties to amyloid plaques

Binding of various oxazine dyes (compounds 3a–d) to amyloidplaques was determined by incubating brain slices of APP23 micefor 5 min at 20 1C with aqueous solutions of the compound atdifferent concentrations (10�2% to 10�4% (wt/vol)). The tissuesections were analyzed using brightfield and fluorescence microscopy(Fig. 2b). The fluorescent spots were found to colocalize with Abdeposits stained with silver methenamine, demonstrating the highlyspecific staining of amyloid plaques by the oxazine dyes. Although allnew dyes (3a–d) displayed binding to the plaques with high specificity,as reflected by a high contrast-to-background fluorescence, AOI987(dye 3a) yielded superior sensitivity with excellent contrast-to-noise

ratios down to a concentration of 0.0001% (wt/vol) (Fig. 2c). Bindingof AOI987 (dye 3a) to aggregated Ab peptide Ab1-40 in solution wasinvestigated by differential fluorescence spectroscopy and thioflavin Tdisplacement (Fig. 2d,e). The fluorescence spectrum of AOI987showed slightly reduced fluorescence intensity in the range 680 and720 nm upon Ab peptide binding. Difference spectra were recorded atdifferent AOI987 concentrations, and the magnitude of the fluores-cence shift was found to be dependent on the concentration ofaggregated Ab peptide. Saturation was observed at high concentra-tions of AOI987, and the Kd value for AOI987 was estimated to be0.2 mM. Aggregated Ab peptide was incubated with thioflavin T tostudy related thioflavin T displacement. The measured fluorescenceintensity with a maximum between 460 and 560 nm (excitation445 nm) reflected thioflavin bound to aggregated Ab peptide. Addi-tion of AOI987 caused a concentration-dependent reduction of thefluorescence intensity, indicating displacement of thioflavin fromaggregated Ab peptide. Measurement of displacement curves allowedthe calculation of the apparent binding constant for AOI987 toaggregated Ab peptide. As the excitation and detection wavelengthsof AOI987 (650 nm and 680–720 nm, respectively) are well separatedfrom those of thioflavin T, quenching effects can be excluded as beingthe source of the thioflavin T fluorescence intensity decrease. Theexperiments were carried out at thioflavin concentrations between0.37 mM–3 mM. The apparent binding constants for AOI987 werefound to be in the range of 0.1 mM–0.2 mM for all thioflavinconcentrations and did not increase with higher thioflavin concentra-tions. This indicates that the displacement of thioflavin from aggre-gated Ab occurs not via a simple competitive mechanism because inthat case, an increase of the apparent binding constant for AOI987with higher thioflavin concentrations would be expected.

1,000

100

100.0 0.5 1.0

Time (h)1.5 2.0

Brain

Plasma

LOQ brain

AO

I987

con

cent

ratio

n(p

mol

/ml o

r g

± s

.d.)

Figure 3 Concentration of AOI987 (dye 3a) in brain and plasma of

anesthetized wild-type mice (C57BL/6Jico) at various time points

(5–120 min) after a single intravenous administration of 3 mg/kg. Values

represent pmol/ml plasma or pmol/g brain 7 s.d.; n ¼ 4. The limit of

quantification (LOQ) was 3 pmol/ml plasma and 15 pmol/g brain.

4095.

2597.

1100.

4095.

2597.

1100.

4095.

2597.

1100.

4095.

2597.

1100.

4095.

2647.

1200.

4095.

2647.

1200.

4095.

2597.

1100.

4095.

2597.

1100.

4095.

2322.

550.0

4095.

2322.

550.0

4095.

2322.

550.0

4095.

2322.

550.0

A0I987_A_190603_A3_30 SL 1/1 A0I987_A_190603_A3_60 SL 1/1 A0I987_A_190603_A3_120 SL 1/1 A0I987_A_190603_A3_240 SL 1/1

A0I987_C_190603_A4_30 SL 1/1 A0I987_C_190603_A4_60 SL 1/1 A0I987_C_190603_A4_120 SL 1/1 A0I987_C_190603_A4_240 SL 1/1

A0I987_A_190603_A5_30 SL 1/1 A0I987_A_190603_A5_60 SL 1/1 A0I987_A_190603_A5_120 SL 1/1 A0I987_A_190603_A5_240 SL 1/1

30 min 60 min

Transgenic

Wild type120 min 240 min

a b

Figure 4 In vivo imaging of amyloid-b deposits. (a) Representative images of female 17-month-old APP23 transgenic (top row) and wild-type (middle row)

mice, injected i.v. with 0.1 mg/kg AOI987 (dye 3a). The images were recorded 30, 60, 120 and 240 min after the injection of the fluorescent dye. In the

bottom row, corresponding images of a female 17-month-old transgenic APP23 mouse treated with 0.9% saline only are shown. Scale bar, 1 cm; color

scale bars in arbitrary units. (b) NIRF microscopy of air-dried cryotome sections (20 mm thickness) of 16-month-old female mice that have been dosed with

0.1 mg/kg AOI987, i.v. The brains were excised and fixed 4 h after dye administration; left, APP23 transgenic; right: wild-type mouse. Scale bar, 100 mm.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Blood-brain barrier penetration

Blood-brain barrier penetration of AOI987 (dye 3a) was assessed usinganesthetized wild-type mice (C57BL/6Jico). The animals were killed atvarious time points (n ¼ 4 each, 5 min to 2 h) after intravenous dyeadministration. AOI987 levels were determined in plasma and braintissue extracts using liquid chromatography/mass spectrometry meth-ods (Fig. 3). Maximal plasma concentration of AOI987 was achievedwithin 15 min after administration. The dye was rapidly eliminatedfrom circulation, with residual plasma levels being less than 10% ofthe initial concentration 2 h after administration. The eliminationof AOI987 from the brain compartment was slightly slower (Fig. 3).The brain concentration of AOI987 exceeded that of the correspond-ing plasma level at all time points measured. These results indicate arapid and significant penetration of AOI987 through the intact blood-brain barrier.

In vivo NIRF imaging

Female 17-month-old APP23 transgenic mice (n ¼ 3 to 5) and age-matched wild-type littermates (n ¼ 2 to 4) have been used to assessthe potential of the oxazine dyes 3a–d for specific amyloid plaqueimaging in vivo using NIRF imaging. Various amounts of the dyes(0.1, 1 and 3 mg/kg; vehicle, 0.9% saline) have been injected i.v. intothe tail vein. Immediately after dosing, an intense fluorescence signalcould be detected in the brain. The disappearance of the fluorescencesignal was substantially slower in transgenic APP23 as compared withwild-type animals. This is illustrated in Figure 4a by images recorded30, 60, 120 and 240 min after i.v. injection with the previouslyestablished optimal dose of 0.1 mg/kg AOI987 (dye 3a). The temporal

behavior of the fluorescence signals was observed to be equal in6-month-old APP23 transgenic and wild-type mice: histologicalanalysis revealed a minimal amyloid plaque load in these younganimals. It was only at the age of 9 months that differences betweenAPP23 transgenic and age-matched wild-type littermates becameobvious. APP23 mice treated with the vehicle 0.9% saline showedsolely background fluorescence at all time points (Fig. 4a).

In vivo results were confirmed by ex vivo measurements on brainsprepared 4 h after dye administration, immediately after acquiring thefinal in vivo NIRF image. Fluorescence microscopy demonstratedselective in vivo staining of amyloid plaques with AOI987 in APP23mice, whereas the brain sections of wild-type animals showed nofluorescence signal (Fig. 4b).

To exclude the possibility that the different temporal behavior offluorescence signals might be due to altered dye metabolism/elimina-tion in APP23 mice as compared with nontransgenic littermatecontrols, we compared the time course of fluorescence signals origi-nating from brain and thorax of 16-month-old female transgenicand wild-type mice. After i.v. administration of AOI987 (dye 3a; at0.1 mg/kg i.v.), identical rates of signal disappearance were observedfor the thoracic region of APP23 and control mice as well as for thebrain of wild-type mice. However, brain signals of APP23 animalsdecreased at a significantly slower rate (Fig. 5), indicative of specificdye retention in cerebral tissue of transgenic mice.

Striking differences were observed when comparing the dyes 3a–dunder in vivo conditions. Brain-specific signal enhancement wasobserved for compounds 3b and 3c, although the intensity differencesbetween APP23 and wild-type mice were less pronounced as com-pared with AOI987 (dye 3a): the initial fluorescence signal wassignificantly weaker and decreased faster than that of AOI987. Nobrain-associated signal was observed after administration of dye 3d,whereas for peripheral tissues fluorescence intensities were comparableto those of the other dyes, indicating that compound 3d did not passthe blood-brain barrier. Among the oxazine compounds tested,AOI987 (dye 3a) showed the greatest potential for imaging of amyloidplaques with in vivo NIRF imaging.

Quantification of in vivo NIRF imaging

Semi-quantitative information was derived from NIRF images bynormalizing the fluorescence intensity to that of the first measurementpoint recorded 30 min after dye administration (Iref, Fig. 6a).Statistically significant differences between transgenic and wild-typemice were obtained at the 120-min as well as at the 240-min point

120

100

80

60

40

20

00 30 60 90 120

Time (min)

Rel

ativ

e flu

ores

cenc

e si

gnal

(%

)

150 180 210 240 270

Figure 5 Time course of fluorescence signals originating from brain and

thorax of 16-month-old female APP23 transgenic and wild-type mice;

red, transgenic brain; blue, wild-type brain; magenta, transgenic thorax;

green, wild-type thorax.

120 0.5

0.4

0.3

0.2

0.1

0

70

60

50

40

30

20

100

80

60

40

20

00 30 60 90 120 150 180 210 240 270 0 30 60 90 120 150 180 210 240 270 8 10 12 14 16 18 20 22R

elat

ive

fluor

esce

nce

sign

al (

%)

Time (min) Time (min) Age (months)

Spe

cific

bin

ding

Sig

nal i

nten

sity

(a.

u.)

a b c

Figure 6 Semi-quantitative in vivo imaging of amyloid-b deposits in living mice. (a) Relative signal activities of 10-month-old APP23 transgenic (upper)and wild-type (lower) mice as a function of time after i.v. injection of 0.1 mg/kg AOI987 (dye 3a). Statistically significant differences between transgenic

and wild-type mice were obtained at the 120- as well as at the 240-min point (P ¼ o0.001, one-way ANOVA). (b) Specific binding of AOI987, that is,

fluorescence signal (transgenic mice) minus fluorescence signal (nontransgenic control mice) divided by fluorescence signal (transgenic mice), as a function

of time after i.v. injection of 0.1 mg/kg AOI987 to 16-month-old APP23 mice. (c) AOI987 fluorescence signal intensity as a function of the age of female

APP23 transgenic mice. Each data point represents the mean fluorescence intensity of a set of ten female APP23 mice of different ages. The corresponding

values for age-matched wild-type animals were subtracted to correct for nonspecific binding. Two independent measurement series have been collated into

the figure illustrating the reproducibility of the measurements. Values are given as mean 7 s.e.m.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

(P ¼ o0.001, one-way ANOVA). The relative intensity parameter(Irel

fl ðtÞ ¼ IflðtÞ=Iref ) displayed remarkable reproducibility, with coeffi-cients of variation of 7% (three measurements within 11 d with four14-month-old female APP23 transgenic mice, measurements at240 min after i.v. administration of 0.1 mg/kg AOI987).

The comparison of the fluorescence intensities observed in APP23and wild-type mice provided a measure of the specific binding ofAOI987 (dye 3a) to amyloid plaques, as shown for 16-month-oldtransgenic APP23 mice that are injected i.v. with 0.1 mg/kg AOI987(Fig. 6b). In the results shown here, we define specific binding as thefluorescence signal of transgenic mice minus the fluorescence signal ofnontransgenic mice, divided by the fluorescence signal of transgenicmice, thereby accounting for the nonspecifically bound dye fraction.

The sensitivity of the in vivo assessment of the amyloid plaque loadusing AOI987 (dye 3a) as a plaque-specific dye (dose 0.1mg/kg, i.v.)was evaluated in a serial study involving female APP23 mice ofdifferent ages (Fig. 6c). The fluorescence intensity increased withage as expected based on the well-documented increase in the amyloidplaque load with age in this animal model of Alzheimer disease29

(Fig. 6c). Two parameters were used to assess the plaque load: therelative signal intensities Irel

fl ð4hÞ measured 4 h after dosing and thearea-under-the curve or rate-of-the-clearance curves. For both para-meters, the corresponding values for age-matched wild-type animalswere subtracted to correct for nonspecific binding. A sigmoidalrelationship between the NIRF measures and the age of the animals,that is, the plaque load, was found (Fig. 6c).

DISCUSSION

The oxazine dye AOI987 (dye 3a) fulfills all the requirements forselective in vivo staining of Ab plaques. The compound readilypenetrates the blood-brain barrier, and the estimated binding constantto aggregated amyloid-b peptide is 0.2 mM. Moreover, the fluorescenceproperties are favorable, with absorption and emission maxima at 650and 670 nm respectively and a high quantum yield of 41%, aprerequisite for high sensitivity. The mode of target interaction iscurrently unknown; intercalation of the planar oxazine dye with b-sheet structure observed in amyloid plaques seems a plausible model.

To evaluate the feasibility of specific plaque labeling in vivo, wecarried out experiments with APP23 transgenic mice. This animalmodel reproduces aspects of Alzheimer disease such as the amyloid-bpeptide deposition in amyloid plaques. These deposits start to appearat the age of 6 months predominantly in the neocortex and hippo-campus. Their number and size increases with age; at 24 months theyoccupy substantial regions of the cortex, hippocampus and thalamus.The amyloid plaques are associated with gliosis and dystrophicneuritis. Cognitive performance of APP23 mice declines with age,suggesting a link to the pathomorphological and pathophysiologicalchanges described29.

Experiments using NIRF imaging demonstrated that amyloidplaques could be observed in living APP23 mice with intact craniumand blood-brain barrier after i.v. administration of AOI987 at dosesranging from 0.1 to 3 mg/kg. Good discrimination between specificand nonspecific dye accumulation was observed 4 h after dosing.Specific labeling of amyloid plaques could be confirmed unambigu-ously by ex vivo analysis of thin sections of the frozen APP23 brains,which were removed immediately after NIRF imaging. The colocaliza-tion of the AOI987 fluorescence with the silver-methenamine stainingas well as the lack of staining of other brain structures demonstrated aspecific interaction of AOI987 with the amyloid plaques.

Biological tissue is a highly scattering medium, the diffuse nature oflight propagation in tissue preventing the accurate determination of

dye distribution and concentration from a single NIRF measurement.Spatial resolution of optical imaging decreases with increasing depthof the fluorescent light source; although microscopic resolution can beachieved when probing superficial structures (or tissue slices), resolu-tion of individual plaques when imaging through the intact craniumand scalp is not feasible. Hence, quantification of the amyloid plaqueload by NIRF imaging is achieved by spatial integration of fluores-cence intensities, analogous to the quantification of receptor densitiesin classic receptor neuroimaging studies.

By normalizing the measured fluorescence reflectance signal inten-sities to the first imaging time point at 30 min after i.v. administrationof the dye, we have derived a reliable intensity parameter Irel

fl ðtÞ,thereby eliminating unavoidable differences among individual ani-mals. Fluorescence intensity measures for APP23 mice were correctedfor the nonspecific dye distribution by subtracting the respectiveintensity values for age-matched wild-type animals. Significant specificbinding was observed for APP23 mice 9 months and older. Thefluorescence intensity and, hence, the extent of specific dye accumula-tion increased with the age of the animals as expected.

The noninvasiveness of plaque-specific NIRF imaging allowsrepeated measurements over the life span of a mouse, providing anideal tool for monitoring progression of amyloid plaque formation inanimal models of Alzheimer disease such as APP23 transgenic mice.The approach is also valuable for evaluation of potential Alzheimerdisease drugs in vivo: therapy response can be compared in the sameanimal to a reference state before treatment. Such paired designsshould increase the statistical significance of the studies.

The described reflectance fluorescence imaging technique (RFI) isinherently qualitative, despite the semi-quantitative analysis described.RFI signals are heavily surface weighted and therefore insensitive toalterations of dye concentrations in deeper layers. Although themethod allowed the sensitive assessment of the amyloid plaque loadwith increasing age of APP23 mice, it remains to be shown whetherthe sensitivity is sufficient to detect minor changes in the amyloidplaque load as induced by a potential Alzheimer disease drug. Never-theless, RFI has been used to semi-quantitatively demonstrate drugeffects in superficial structures, that is, the inhibition of proteaseactivity in subcutaneously implanted tumors31 and in animal modelsof rheumatoid arthritis32. Improved quantitative data should beobtainable using optical tomography techniques26,33. Preliminaryexperiments with AOI987 in APP23 mice using optical tomographicmethods confirmed results obtained with RFI.

Compared with optical imaging methods (RFI or optical tomo-graphy), small animal PET provides clearly superior quantification,high sensitivity and full three-dimensional information. A numberof potential amyloid plaque-sensitive PET ligands have been devel-oped, with some of them in clinical development13–16,25. Yet, theobvious methodological advances of small animal PET are counter-balanced by the cost effectiveness of NIRF imaging34. Another strikingadvantage of optical imaging is the experimental simplicity due to theuse of stable dyes. This should allow high-throughput screening ofpotential drug candidates. Therefore, for animal studies, opticaltechniques, and in particular optical tomography, which provideboth high sensitivity and improved quantification, offer significantadvantages when compared with the demands on infrastructurerequired for PET studies.

In summary, we have described a series of oxazine dyes thatreadily penetrate the intact blood-brain barrier and show preferentialbinding to amyloid plaques. In particular, the dye AOI987 (dye 3a)has optical properties, such as maximum absorption/emission inthe near infrared domain and high quantum yield, that are favorable


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

for in vivo imaging. Using NIRF imaging, we demonstrated specificinteraction of AOI987 with amyloid plaques in APP23 transgenicmice in vivo, as confirmed by postmortem analysis of brainslices. Semi-quantitative analysis revealed increasing fluorescencesignal intensity with increasing plaque load of the animals.Thus, the plaque-specific oxazine NIRF dye AOI987 is an attractiveprobe to noninvasively monitor disease progression in animalmodels of Alzheimer disease and to evaluate the effectiveness ofpotential drugs.

METHODSChemical synthesis. Reagents used for chemical syntheses were purchased

from Sigma Chemical and were used without further purification. Fluorescence

properties of dyes (3a–d) were determined at Solvias AG.

4-methyl-7-(4-nitro-phenylazo)-3,4-dihydro-2H-benzo[1,4]oxazin-6-ol (2a):

4-nitro-benzenediazonium tetrafluoroborate (574 mg, 2.42 mmol) was dis-

solved in 10% H2SO4 (400 ml) and added to a solution of 4-methyl-3,4-

dihydro-2H-benzo[1,4]oxazine-6-ol35 (1a) (400 mg, 2.42 mmol) in methanol

(2 ml). The reaction mixture was stirred for 30 min at 20 1C, neutralized with

25% aqueous ammonia and the red precipitate was spun down. The crude azo

intermediate was purified by recrystallization in n-butanol to give the title

compound as a red powder (720 mg, 95%).

AOI987 3a: Azo intermediate 2a (560 mg, 1.78 mmol) and 4-methyl-3,4-

dihydro-2H-benzo[1,4]oxazin-6-ol (1a) (326 mg, 1.96 mmol) were dissolved

in a mixture of ethanol/water (10:1, 10 ml). After the addition of 32% HCl

(700 ml) the reaction mixture was stirred at 70 1C under reflux for 1 h and

subsequently the solution was concentrated under reduced pressure. The

residue was dissolved in water and treated with a saturated solution of sodium

tetrafluoroborate. The precipitate was spun down and purified by column

chromatography (SiO2, dichloromethane/methanol ¼ 10:2) to give 480 mg

(1.17 mmol, 65%), blue crystals.

Rf ¼ 0.65 [CH2Cl2/methanol (10:2)], m.p.: 238 7 2 1C.

UV/VIS: lmax ¼ 644 nm (methanol).1H-NMR (500 MHz): d ¼ 3.37 (s, 6 H), 3.80 (t, 4 H, 3J ¼ 4.8 Hz), 4.37

(t, 4 H, 3J ¼ 4.8 Hz,), 7.03 (s, 2 H), 7.21 (s, 2 H).13C-NMR (125 MHz): d ¼ 48.66, 63.30, 95.10, 112.8, 134.5, 145.2,

145.7, 146.8.

MS (70 eV, EI): m/z (%) ¼ 324.2 (100) [M]+, 310 (3).

HRMS (C18H18N3O3, M+): theor.: 324.1348

found: 324.1349

ASG236 3b, ASG237 3c, AMQ987 3d: The same procedure described for the

preparation of AOI987 (3a) (Fig. 1) was carried out to prepare oxazine dyes

3b–d using building blocks 1b–d.

UV/VIS and fluorescence analysis. UV/VIS and fluorescence spectra were

recorded and analyzed. For UV/VIS spectra, a Lambda 19 spectrometer from

Perkin Elmer equipped with cells of 1.0-cm path length was used. The scan rate

was 120 nm/min. The fluorescence data were obtained on a Spex Fluorolog

(I.S.A.) spectrometer equipped with a cooled R928 detector (Slit 1 mm).

Quantum yields were determined by using Cresylviolett (exc., 595 nm; yield,

0.54) as standard.

Animal preparation. Female APP23 transgenic mice (3 r n r 5)29 and age-

matched wild-type littermates (2 r n r 4) aged 6, 9, 10, 12, 14, 16, 17 and

21 months were used at around 25-g body weight. For NIRF experiments, the

animals were anesthetized using 1.5% isoflurane (Abbott) in nitrous oxide/

oxygen, 2:1, administered with a face mask. The duration of the anesthesia was

3 min. The dyes were administered intravenously (i.v.) into the tail vein at

various doses (vehicle, 0.9% saline; injection volume, 10 ml/kg). NIRF images

were recorded 30, 60, 120 and 240 min after dye administration. To avoid

problems caused by light scattering due to the fur, we shaved animals for the

NIRF experiments.

To determine the optimal dose, various doses of AOI987 (dye 3a) were

administered i.v. (3, 1, 0.3, 0.1 and 0.01 mg/kg) to female APP23 transgenic and

nontransgenic mice. At 4 h after probe application, the difference of the

normalized signal intensities Irel between transgenic and nontransgenic mice

was highest for a dosage of 0.1 mg/kg. The dye was completely washed out 24 h

after i.v. application. We found AOI987 in peripheral organs, yet observed no

acute toxicity at any of the doses used.

NIRF imaging. In vivo NIRF imaging was carried out using a ‘bonSAI’

fluorescence reflectance small animal imager prototype (Siemens AG, Medical

Solutions). For fluorescence excitation, three laser diodes at 660 nm with a total

power of 10 mW/cm2 have been used yielding a uniform illumination of the

whole animal. The fluorescent light emitted from the sample (mouse) was

detected by a charge-coupled device (CCD) camera (Hamamatsu ORCA)

equipped with a focusing lens system (macro lens 60 mm, 1:2.8, Nikon).

The image matrix comprised 532 � 256 pixels. A bandpass filter was used for

the selection of the detection wavelength (700 nm). Data collection, for

example, integration times, ranged from 0.5 to 3.0 s depending on the

fluorescence intensity. The experiment was controlled by a PC using the

Siemens SYNGO software. NIRF images have been evaluated quantitatively

using region-of-interest analysis tools provided by BioMap 3.05, an image

analysis software tool (http://www.maldi-msi.org) developed in-house. Auto-

fluorescence measured immediately before dye administration turned out to be

negligible. All experiments have been carried out in adherence to the Swiss laws

of animal protection.

Quantitative image analysis. Direct determination of dye concentrations from

reflectance NIRF images is not feasible. We therefore used signal intensities

relative to the initial measurement point at 30 min, that is, Irel(t) ¼ I(t)/

I(30 min). Signal intensities displayed an exponential decay within error limits:

exponential regression yielded the initial signal intensity value for t ¼ 0, which

was the reference for the subsequent analyses. To estimate specific dye binding

in a specific organ, relative intensity measurements in APP23 mice were

corrected for nonspecific dye accumulation: a specific binding parameter at

time t was calculated according to Bsp(t)¼{Irel(t; tg) – Irel (t; wt)}/Irel (t; tg)

where the tg and wt represent transgenic and wild-type animals. Bsp values

increased as a function of time and reached a plateau between 150 and 240 min

after dye administration. Based on the time courses, quantification was typically

carried out at 240 min after dosing.

Ex vivo analysis. Mice were killed by decapitation immediately after NIRF

imaging; the brain was removed and immediately frozen on dry ice. Brain

sections were cut in a cryotome (20 mm thick, thaw mounted and air dried) and

analyzed using the same fluorescence microscopy setup as described below for

in vitro staining analyses.

In vitro staining analysis. Cryostat brain sections (20 mm thick, fixed in

formaldehyde and air dried after washing in PBS and ethanol) were incubated

at 20 1C for 5 min with a 50% ethanol solution containing different concen-

trations of contrast agent (0.1% to 0.0001%). Stained sections were then

washed in PBS solution and investigated both before and after washing with

50% ethanol using fluorescence microscopy with a Cy5 filter set mounted on a

Olympus BX51 and illumination with a halogen lamp.

Blood-brain barrier penetration. AOI987 was administered intravenously to

male C57BL/6Jico mice (23–25 g) and groups of n ¼ 4 were killed by

decapitation after 5, 15, 30, 60 and 120 min. Trunk blood was collected in

EDTA-containing Eppendorf tubes, centrifuged and, after removal, the plasma

was frozen and stored at �70 1C. The brain was removed and immediately

frozen on dry ice and stored at �70 1C until analysis.

AOI987, spiked with internal standard (NVP-AAG561), was extracted from

the plasma and brain using dichloromethane and, after drying under nitrogen,

was redissolved in acetonitrile and separated, isocratically, by high-performance

liquid chromatography on a C18 reversed-phase column. Quantitative analysis

was performed by selected ion recording (SIR) over the respective protonated

molecular ions [M+H+] of NVP-AOI987-DA-1, (m/z 324.4 7 0.5) and NVP-

AAG561 (m/z 411.3 7 0.5). The peaks obtained were automatically integrated

using the software facilities of the analytical system MassLab (Finnigan). The

peak area was chosen as the chromatographic signal for quantification.

In vitro fluorescence binding assay. AOI987 was added in several aliquots to a

prepared solution of fibrillary Ab1-40 in buffer or to buffer alone36,37.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Fluorescence emission spectra (690–750 nm, excitation at 650 nm) were

recorded using a LS50B luminescence photometer (Perkin Elmer) in 25 mM

phosphate buffer, pH 7.4 containing 120 mM NaCl. Difference spectra between

Ab1-40 solution and buffer were calculated for each AOI987 concentration.

AOI987 concentration was plotted versus fluorescence difference at 705 nm and

the maximum signal was calculated by extrapolating the data to infinite

AOI987 concentration. The binding constant of AOI to fibrillar Ab1-40 was

calculated from a plot of AOI987 concentration versus Fmax-F.

Thioflavin T displacement experiments were done using the same buffer and

aggregated Ab preparation. After recording spectra for thioflavin alone and in

the presence of aggregated Ab, AOI987 was added in several aliquots, to a final

concentration of 3.25 mM. Fluorescence intensity values at 490 nm were plotted

versus concentration of AOI987, and binding constants were calculated using

the Origin software package (OriginLab).

ACKNOWLEDGMENTSThe authors would like to thank Alexandra Suter for her excellenttechnical assistance.


Received 29 October 2004; accepted 18 February 2005


1. Meyer-Luehmann, M. et al. Extracellular amyloid formation and associated pathology inneuronal grafts. Nat. Neurosci. 6, 1–8 (2003).

2. Nussbaum, R.L. & Ellis, C.E. Alzheimer’s disease and Parkinson’s disease. N. Engl.J. Med. 348, 1356–1364 (2003).

3. Hardy, J. & Selkoe, D.J. The amyloid hypothesis of Alzheimer’s disease: progress andproblems on the road to therapeutics. Science 297, 353–356 (2002).

4. Zamrini, E., De Santi, S. & Tolar, M. Imaging is superior to cognitive testing for earlydiagnosis of Alzheimer’s disease. Neurobiol. Aging 25, 685–691 (2004).

5. Benveniste, H., Einstein, G., Kim, K.R., Hulette, C. & Johnson, G.A. Detection ofneuritic plaques in Alzheimer’s disease by magnetic resonance microscopy. Proc. Natl.Acad. Sci. USA 96, 14079–14084 (1999).

6. Zhang, J. et al. Detection of amyloid plaques in mouse models of Alzheimer’s disease bymagnetic resonance imaging. Magn. Reson. Med. 51, 452–457 (2004).

7. Mathis, C.A., Wang, Y. & Klunk, W.E. Imaging b-amyloid plaques and neurofibrillarytangles in the aging human brain. Curr. Pharm. Des. 10, 1469–1492 (2004).

8. Bacskai, B.J., Klunk, W.E., Mathis, C.A. & Hyman, B.T. Imaging amyloid-b depositsin vivo. J. Cereb. Blood Flow Metab. 22, 1035–1041 (2002).

9. Ono, M. et al. 11C-labeled stilbene derivatives as Ab-aggregate-specific PET imagingagents for Alzheimer’s disease. Nucl. Med. Biol. 30, 565–571 (2003).

10. Suemoto, T. et al. In vivo labeling of amyloid with BF-108. Neurosci. Res. 48, 65–74(2004).

11. Okamura, N. et al. Styrylbenzoxazole derivatives for in vivo imaging of amyloid plaquesin the brain. J. Neurosci. 24, 2535–2541 (2004).

12. Klunk, W.E. et al. Uncharged thioflavin-T derivatives bind to amyloid-beta protein withhigh affinity and readily enter the brain. Life Sci. 69, 1471–1484 (2001).

13. Mathis, C.A. et al. Synthesis and evaluation of 11C-labeled 6-substituted 2-arylben-zothiazoles as amyloid imaging agents. J. Med. Chem. 46, 2740–2754 (2003).

14. Mathis, C.A. et al. A lipophilic thioflavin-T derivative for positron emission tomography(PET) imaging of amyloid in brain. Bioorg. Med. Chem. Lett. 12, 295–298 (2002).

15. Agdeppa, E.D. et al. Binding characteristics of radiofluorinated 6-dialkylamino-2-naphthylethylidene derivatives as positron emission tomography imaging probes forb-amyloid plaques in Alzheimer’s disease. J. Neurosci. 21, RC189 1–5 (2001).

16. Agdeppa, E.D. et al. 2-Dialkylamino-6-acylmalononitrile substituted naphthalenes(DDNP analogs): novel diagnostic and therapeutic tools in Alzheimer’s disease.Mol. Imaging Biol. 5, 404–417 (2003).

17. Wadghiri, Y.Z. et al. Detection of Alzheimer’s amyloid in transgenic mice usingmagnetic resonance microimaging. Magn. Reson. Med. 50, 293–302 (2003).

18. Poduslo, J.F. et al. Design and chemical synthesis of a magnetic resonance contrastagent with enhanced in vitro binding, high blood-brain barrier permeability, and in vivotargeting to Alzheimer’s disease amyloid plaques. Biochemistry 43, 6064–6075(2004).

19. Poduslo, J.F. et al. Molecular targeting of Alzheimer’s amyloid plaques for contrast-enhanced magnetic resonance imaging. Neurobiol. Dis. 11, 315–329 (2002).

20. Bacskai, B.J. et al. Imaging of amyloid-b deposits in brains of living mice permits directobservation of clearance of plaques with immunotherapy. Nat. Med. 7, 369–372(2001).

21. Klunk, W. et al. Imaging Ab plaques in living transgenic mice with multiphotonmicroscopy and methoxy-X04, a systemically administered Congo red derivative.J. Neuropathol. Exp. Neurol. 61, 797–805 (2002).

22. Bacskai, B.J. et al. Four-dimensional multiphoton imaging of brain entry, amyloidbinding, and clearance of an amyloid-b ligand in transgenic mice. Proc. Natl. Acad.Sci. USA 100, 12462–12467 (2003).

23. Styren, S.D., Hamilton, R.L., Styren, G.C. & Klunk, W.E. X-34, a fluorescent derivativeof Congo red: a novel histochemical stain for Alzheimer’s disease pathology.J. Histochem. Cytochem. 48, 1223–1232 (2000).

24. Shoghi-Jadid, K. et al. Localization of neurofibrillary tangles and beta-amyloid plaquesin the brains of living patients with Alzheimer disease. Am. J. Geriatr. Psychiatry 10,24–35 (2002).

25. Klunk, W.E. et al. Imaging brain amyloid in Alzheimer’s disease with Pittsburghcompound-B. Ann. Neurol. 55, 306–319 (2004).

26. Weissleder, R. & Ntziachristos, V. Shedding light onto live molecular targets. Nat. Med.9, 123–128 (2003).

27. Ntziachristos, V., Bremer, C. & Weissleder, R. Fluorescence imaging with near-infraredlight: new technological advances that enable in vivo molecular imaging. Eur. Radiol.13, 195–208 (2003).

28. Funovics, M., Weissleder, R. & Tung, C.-H. Protease sensors for bioimaging. Anal.Bioanal. Chem. 377, 956–963 (2003).

29. Sturchler-Pierrat, C. et al. Two amyloid precursor protein transgenic mouse models withAlzheimer disease-like pathology. Proc. Natl. Acad. Sci. USA 94, 13287–13292(1997).

30. Neuweiler, H. et al. Detection of individual p53-autoantibodies by using quenchedpeptide-based molecular probes. Angew. Chem. Int. Edn. Engl. 41, 4769–4773(2002).

31. Bremer, C., Tung, C.H. & Weissleder, R. In vivo molecular target assessment of matrixmetalloproteinase inhibition. Nat. Med. 7, 743–748 (2001).

32. Wunder, A. et al. In vivo imaging of protease activity in arthritis: a novel approach formonitoring treatment response. Arthritis Rheum. 50, 2459–2465 (2004).

33. Graves, E.E. et al. A submillimeter resolution fluorescence molecular imaging systemfor small animal imaging. Med. Phys. 30, 901–911 (2003).

34. Rudin, M. & Weissleder, R. Molecular imaging in drug discovery and development.Nat. Rev. Drug Discov. 2, 123–131 (2003).

35. Daves, G. & McMurtrey, K.D. Koenig’s adducts of N-alkyl(aryl)-aminoethanols andquinones. 3,4-Dihydro-4-alkyl(aryl)-8a-hydroxy-2H–1,4-benzoxazine-6(8aH)-ones.J. Org. Chem. 35, 4252–4253 (1970).

36. Klunk, W.E. & Pettegrew, J.W. Development of small molecule probes for thebeta-amyloid protein of Alzheimer’s disease. Neurobiol. Aging 15, 691–698 (1994).

37. Levine, H. III Thioflavin T interaction with synthetic Alzheimer’s disease b-amyloidpeptides: detection of amyloid aggregation in solution. Protein Sci. 2, 404–410(1993).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Stable antibody expression at therapeutic levelsusing the 2A peptide

Jianmin Fang, Jing-Jing Qian, Saili Yi, Thomas C Harding, Guang Huan Tu, Melinda VanRoey & Karin Jooss

Therapeutic monoclonal antibodies (mAbs) are currently being developed for the treatment of cancer and other diseases. Despite

clinical success, widespread application of mAb therapies may be limited by manufacturing capabilities. In this paper, we

describe a mAb delivery system that allows continuous production of a full-length antibody at high-concentrations in vivo after

gene transfer. The mAb is expressed from a single open reading frame by linking the heavy and light chains with a 2A self-

processing peptide derived from the foot-and-mouth disease virus. Using this expression system, we generated a recombinant

adeno-associated virus vector encoding the VEGFR2-neutralizing mAb DC101 (rAAV8-DC101). A single dose of rAAV8-DC101

resulted in long-term expression of >1,000 lg/ml of DC101 in mice, demonstrating significant anti-tumor efficacy. This report

describes the first feasible gene therapy approach for stable delivery of mAbs at therapeutic levels, which may serve as an

attractive alternative to direct injection of mAbs.

mAbs have become important therapeutic agents for the treatment ofcancer, inflammation and infectious disease. With technical advance-ments in antibody engineering such as human antibody phage dis-play1, mice transgenic for human IgG2 and antibody humanizationtechniques1, highly specific human monoclonal antibodies can bereadily generated for various disease targets.

Chronic diseases such as cancer require long-term therapies, andmAbs are often infused into people with cancer, frequently at highdoses over a long period of time, which can induce adverse effects3. Analternative approach to long-term delivery of therapeutic antibodies isto express the antibodies in vivo after gene transfer. For mostantibodies, however, therapeutic serum concentrations range fromone to several hundred mg/ml, levels that have been impossible toachieve using gene transfer. Unfortunately, the recombinant adeno-associated virus (rAAV) vector, which is an attractive vector system forachieving long-term gene transfer in vivo4, cannot accommodateconventional antibody expression cassettes that drive the mAb heavyand light chains from two individual promoters, because the vectorcannot package more than B5 kb efficiently.

A potentially advantageous approach for in vivo delivery of anti-bodies is to express mAb heavy and light chains in a bicistronic vectorthat uses a single promoter. The conventional method for bicistronicexpression cassettes, however, uses internal ribosomal entry sites(IRES) that leads to substantially lower expression of the secondgene than the catabolite activator protein (CAP)-dependent firstgene5. In this study, we describe an antibody expression systemthat uses the foot-and-mouth-disease virus (FMDV)-derived 2Aself-processing sequence to express full-length antibodies from a singleopen reading frame (ORF). 2A sequences are oligopeptides located

between the P1 and P2 proteins in some members of the picornavirusfamily and can undergo self-cleavage to generate the mature viralproteins P1 and P2. Among various 2A or 2A-like sequences, FMDV2A is particularly short (minimum of 13 amino acids) and is able to‘cleave’ at its own C terminus between the last two amino acidsthrough an enzyme-independent but undefined mechanism, probablyby ribosomal skip, during protein translation6–11. Using a FMDV 2Asequence adjacent to a furin cleavage site to link the antibodyheavy and light chain sequences, we were able to engineer a mAbexpression cassette that, in the context of AAV-mediated gene transfer,results in high levels of full-length, functional monoclonal antibodiesin vitro and in vivo. Sustained mAb serum levels of 41,000 mg/ml wereachieved in mice with a single administration of an rAAV8 vectorexpressing DC101, an anti-angiogenic mAb targeting vascularendothelial cell growth factor receptor-2 (VEGFR2 or Flk-1)12.The rAAV8 mediated gene transfer of DC101 resulted in signifi-cant (P o 0.001) anti-tumor efficacy in two tumor models, demon-strating the generation of functional antibodies in vivo using thisexpression system.

RESULTS

2A-mediated mAb expression from a single ORF

DC101, a rat anti-mouse VEGFR2 (Flk-1) IgG1 mAb, which has beenwell characterized for its anti-angiogenic effects in mouse tumormodels12, was chosen as a model antibody to evaluate expression offull-length antibodies from a single ORF using the FMDV 2A self-processing peptide. An expression cassette termed H2AL, in which theheavy and light chain sequences of the DC101 mAb were linkedtogether by the FMDV 2A self-cleavage sequence, was generated and


Department of Preclinical Oncology and Immunology, Cell Genesys, Inc., 500 Forbes Blvd., S. San Francisco, California 94080, USA. Correspondence should beaddressed to J.F. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

cloned into an expression plasmid driven by the CAG promoter(Fig. 1). The CAG promoter is comprised of the cytomegalovirus(CMV) immediate early enhancer region, chicken b-actin promoter/splice donor and rabbit b-globin enhancer13 and is constitutivelyactive. A control plasmid, which linked the DC101 heavy and lightchains by an IRES derived from encephalomyocarditis virus (EMCV)(Fig. 1b, HIRESL), was also generated. Both plasmids were transientlytransfected into human embryonic kidney (HEK) 293 cells, and theantibody concentrations in the supernatants were determined byenzyme-linked immunosorbent assay (ELISA) after 48 h. The H2ALconstruct resulted in about 1.6 mg/ml of DC101, whereas the HIRESLconstruct resulted in 0.1 mg/ml of the mAbs (Fig. 2a). Thus, the 2Asequence efficiently facilitates antibody heavy and light chain expres-sion from a single ORF.

To further characterize the DC101 heavy and light chains expressedfrom the 2A-containing construct, the proteins in the supernatants oftransiently transfected HEK 293 cells were separated on SDS-PAGEgels under both reducing and nonreducing conditions and subjectedto western blot analysis using a polyclonal goat anti-rat IgG antibodythat recognizes antibody heavy and light chains. Under reducingconditions, two protein bands at molecular weights of B55 and25 kDa were detected, corresponding to the IgG heavy and lightchain proteins of DC101, respectively (Fig. 2b). Bands of similarsize were detected in samples from the parental DC101 hybridoma

cells. The heavy chain protein from the H2AL plasmid migratesslightly more slowly than the native DC101 antibody heavychain (Fig. 2b) due to 23–amino acid residues that are derived fromthe 2A sequence and remain after cleavage. Under nonreducingconditions, a single band of approximately 160 kDa was detected inthe cell culture supernatant of H2AL-transfected HEK 293 cells(Fig. 2c), which is the expected size of a dimerized full-lengthantibody containing two heavy and two light chains. The molecularweight of the mAb expressed from the H2AL plasmid was slightlyhigher than that of the native antibody, because of the additionalamino acid residues at the C terminus of the heavy chain. Noadditional protein bands, which are expected when the ratio betweenheavy and light chains is imbalanced, were detected under nonredu-cing conditions (Fig. 2c), suggesting that the antibodies in the super-natant of H2AL transfected cells are properly dimerized and neitherheavy nor light chains are in excess.

Biological activity of mAbs expressed from 2A plasmids

The biological activity of the antibodies expressed from the2A-containing plasmid was evaluated using a binding assay thatmeasures mAb-binding activity to immobilized mVEGFR2 protein(see Methods). DC101 antibodies expressed from the H2AL constructin HEK 293 cells were able to recognize the mouse VEGFR2 withsimilar binding activity as the parental antibodies (Fig. 3a) and werecapable of blocking the VEGF and mVEGFR2 interaction in a dose-dependent manner in a VEGF-mVEGFR2 (ligand-receptor) bindingassay with the same potency as the native antibodies (Fig. 3b). Thus,antibodies expressed from the 2A-containing construct retain fullbiological activity.

Removal of 2A-derived amino acid residues by furin cleavage

Since the 2A self-processing cleavage occurs between the last twoamino acids at the C terminus of the 2A peptide, the first protein inthe cassette (that is, the heavy chain in the H2AL construct) has 23additional amino acid residues at its C terminus (Fig. 1a). Toeliminate possible adverse effects caused by the remaining 2A residues,a DC101 expression cassette was engineered that included a furincleavage site sequence (RAKR), located between the 2A sequence andthe mAb heavy chain (HF2AL) (Fig. 1b, lower construct).

mAb was expressed from the HF2AL construct in HEK 293cells after transient plasmid transfection (Fig. 2a). The heavychain proteins in the supernatants were separated by SDS-PAGEgel and analyzed by western blot using a goat anti-rat IgGantibody. The antibody heavy chains expressed from the HF2ALconstruct have a similar molecular weight as the native antibodyheavy chains, suggesting successful cleavage at the furin cleavagesite (Fig. 2b). Furthermore, the antibodies expressed from theHF2AL construct appeared as a single band at a molecular weightcorresponding to the antibody dimer (Fig. 2c) and demonstrated fullbiological activity in antibody binding (Fig. 3a) and neutralizationassays (Fig. 3b).

To confirm successful removal of the residual 2A amino acids, wegenerated a construct that contained six histidine residues (his-tag) atthe C terminus of the antibody light chain in the HF2AL cassette. Thehis-tagged antibody was expressed in vivo by hydrodynamic injectionof the HF2AL expression plasmid into mice, and mAb was purifiedfrom mouse serum using a nickel column purification system. Thepurified antibody appears as two protein bands in a reducing SDS-PAGE gel at B52 and 25 kDa (data not shown), corresponding to theantibody heavy and light chains, respectively. The heavy chain bandwas excised from the gel, digested with trypsin and analyzed by mass

Stop codon2AFurin

cleavagesite

Start codon

Promoter

Promoter

Promoter

HF2AL Ab heavy chain

Ab heavy chain

Ab heavy chain

Ab light chain

Ab light chain

Ab light chain

Poly A

Poly A

Poly A

Stop codon

Stop codonStart codon

Stop codonStart codon Startcodon

HIRESL IRES

2A

H2AL

2A residues

2A residue (1 aa)2A residues

Heavy and lightchains after signal-peptide removalduring secretion

Protein translationand self-processing

Cleavage site

2A

mRNA

Signal peptide Heavy chain Signal peptide

Light chaina

b

Figure 1 Full-length mAb expression cassette using the FMDV 2A sequence.

(a) Schematic illustration of the biosynthesis of mAb heavy and light chains

by the 2A, peptide-containing expression cassette. (b) Antibody expression

cassettes in which mAb heavy and light chain sequences are linked by the

2A sequence (H2AL), an IRES sequence (HIRESL) or a combination of furin

cleavage site and 2A sequence (HF2AL).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

spectrometry. We detected no peaks that represent peptide frag-ments derived from the 2A sequence (APVK, QTLNFDLLK andLAGDVESNPG) in the mass spectrum, demonstrating that the 2Aresidues had been removed from the antibody heavy chain of theHF2AL construct. Moreover, the mass spectrum analysis revealed a1,039.53-Da fragment corresponding to the C-terminal fragmentsequence, SLSHSPGKRA, that contains the native C-terminalsequence of the antibody heavy chain plus two additional aminoacids (arginine and alanine) derived from the furin cleavage site. Theactual amino acid sequence of the 1039.53 Da fragment was furtherconfirmed by the post-source decay (PSD) analysis. In summary, thesedata demonstrate that the addition of a furin cleavage site to the 2Aself-processing cleavage site results in the removal of the 23–aminoacids that remain after 2A cleavage.

2A-mediated, high-level expression of mAbs in vivo

Using the 2A sequence, the expression cassette for DC101 could be fitinto a rAAV vector. We generated rAAV vectors pseudotyped with thecapsid proteins from AAV serotype-8 (ref. 14) that express the DC101antibody from an H2AL or HF2AL cassette driven by the CAGpromoter, termed rAAV8-DC101(H2AL) and rAAV8-DC101(HF2AL).Mice were injected through the hepatic portal vein with one of three

dose levels of rAAV8-DC101(H2AL) vector (1 � 1011, 2 � 1011 or4 � 1011 vector genomes (vg)/mouse), and DC101 serum levelswere evaluated over time. We detected, 28 d after administrationof the rAAV8-DC101(H2AL) vector, peak serum levels of 3,286and 1,877 mg/ml DC101 in mice that had received 4 � 1011 or2 � 1011 vg/animal, respectively (Fig. 4a). Antibody serum levelsdeclined somewhat thereafter but remained at about 600 mg/ml inthe animals treated with the highest dose of vector throughout the4-month study (Fig. 4a). Interestingly, two- to tenfold higher DC101serum levels were achieved when the antibody was expressed fromthe HF2AL expression cassette (Fig. 4b). Peak expression levelsin mice receiving 4 � 1011 and 2 � 1011 vg/mouse of therAAV-DC101(HF2AL) vector were 48,000 mg/ml and remainedabove 1,000 mg/ml up to 4 months after rAAV vector administration.Furthermore, DC101 antibody expressed from either cassette exhibitedfull biological activity in the antibody binding (Fig. 3a) and neutra-lization assays (Fig. 3b).

We monitored serum alanine aminotransferase (ALT) and aspartateaminotransferase (AST) levels in these animals to evaluate liverfunction after gene transfer and were not able to detect elevatedALT or AST serum levels in all vector-treated mice throughout theentire experiment (data not shown).

Figure 2 In vitro expression of DC101 mAb using

the 2A self-processing sequence–containing

expression plasmids. HEK 293 cells were

transiently transfected with DC101-expressing

plasmids and supernatants were harvested

for protein analyses. (a) ELISA analysis of

supernatants 48 h after transfection of HEK 293

cells with the H2AL, the HIRESL or the HF2ALplasmids (mean 7 s.d.). (b) Western blot

analysis of DC101-containing supernatants under

reducing conditions. Proteins in the supernatants

of the parental hybridoma cells (hybridoma),

the HEK 293 cells transfected with the H2AL

plasmid (H2AL) or the HF2AL plasmid (HF2AL),

or untransfected HEK 293 cells (mock) were

separated by SDS-PAGE under reducing conditions and probed with a goat anti-rat IgG (H+L) polyclonal antibody. (c) Western blot analysis of supernatants

from the parental DC101 hybridoma cells (hybridoma), from HEK 293 cells transfected with the H2AL plasmid (H2AL) or HF2AL plasmid (HF2AL), or from

untransfected HEK 293 cells (mock). Proteins in supernatants were separated in SDS-PAGE under nonreducing conditions and probed with a goat anti-rat

IgG (H+L) polyclonal antibody.

HF2ALHIRESLH2AL0

0.5

1

1.5

2

2.5

3

DC

101

mA

b (µ

g/m

l)

14

17

17

28

28

38

38

49

49

62

62

98

98

188

188

MW(kDa)

MW(kDa)M

ock

Hyb

ridom

a

H2A

L

HF

2AL

Moc

k

Hyb

ridom

a

H2A

L

HF

2AL

a b c

010301003001,0003,000

Antibody (ng/ml)

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

OD

405

OD

405

Control

HF2AL in vivo

HF2AL in vitro

H2AL in vitro

Hybridoma

H2AL in vivoControl

HF2AL in vivo

HF2AL in vitro

H2AL in vitro

Hybridoma

H2AL in vivo

031.2562.51252505001000

Antibody (ng/ml)

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6a b

Figure 3 Biological activity of DC101 mAbs generated from 2A-containing expression cassettes. (a) Antibody binding activity to immobilized Flk-1 protein.

DC101 antibody concentration was determined by ELISA. Binding of the mAb to Flk-1 was detected at an absorbance of 405nm after incubation with an

anti-rat IgG-HRP antibody, followed by the addition of the HRP substrate (mean 7 s.d.). (b) Neutralizing properties of DC101 evaluated in a VEGF-Flk-1

binding assay. VEGF-Flk-1 binding was detected at an absorbance of 405 nm with an anti-Flk-1 antibody conjugated to HRP.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Anti-tumor efficacy of 2A-mediated mAbs in vivo

After demonstrating high-level antibody expression in mice afterrAAV8-mediated gene transfer, we investigated whether DC101generated from the rAAV8 vector results in anti-tumor efficacyin vivo as demonstrated for recombinant DC101 antibodies12. Givenits higher levels of serum DC101 expression in vivo, only the furin-containing rAAV8-DC101(HF2AL) vector was tested.

rAAV8-DC101(HF2AL) or rAAV8-null control vector (2 � 1011 vg/mouse) was injected into nude mice through the tail vein, and DC101antibody serum levels were evaluated by ELISA. At day 24 after vectoradministration (Fig. 5), the mice were injected subcutaneously witheither 1 � 105 cells/mouse of murine B16F10 melanoma cells or 5 �106 cells/mouse of human U87 malignant glioblastoma cells. rAAV8-DC101 injection resulted in milligram levels of mAb in the serum ofmice (Fig. 5a,d). Significant anti-tumor activity was observed in bothmodels in mice treated with the rAAV8-DC101(HF2AL) vector(Fig. 5b,e; P o 0.05), which resulted in a significantly prolongedsurvival time (Fig. 5c,f; P o 0.001). In the B16F10 model, mediansurvival time (MST) increased from 30 d in the control group to41.5 d in the treated group. In the U87 MG model, DC101 antibodygene transfer resulted in tumor dormancy in 6 out of 11 mice formore than 3 months and three mice tumors that had reached avolume of 400–700 mm3 regressed completely (data not shown). Insummary, a single administration of a rAAV8 vector expressing DC101mAb results in stable and high mAb serum levels that are able tocontrol tumor burden.

DISCUSSION

This study demonstrates that a 2A self-processing peptide derivedfrom FMDV facilitates efficient and apparent equimolar expression offull-length antibody heavy and light chains in vitro and in vivo from asingle ORF and that the antibody chains self-assemble to form afunctional antibody. To our knowledge, no one else has shown thatfull-length mAb gene transfer can provide potent anti-tumor activityin vivo, which may be useful as an alternative therapy to directinjection of monoclonal antibodies.

Given the potential clinical benefits of antibody gene therapy, greateffort has been devoted to the expression of full-length antibodiesin vivo after gene transfer15. To achieve therapeutic effects, however,most antibodies require high and sustained serum levels, typicallyranging from one to several hundred mg/ml16–18. Such high mAbserum levels can be achieved only by repeated administration ofhigh doses of recombinant protein, typically ranging from 5 mg/kgto 20 mg/kg of body weight. The high mAb serum concentrationsrequired for clinical efficacy and the typically low expression levelsafter gene transfer have made the development of antibody genetransfer technologies challenging.

Electroporation of mAb plasmids in muscle achieved mAb serumlevels of 1.5 mg/ml in mice19, whereas implantation of ex vivo transducedcells with retroviral vectors, such as myoblasts20 and fibroblasts21,resulted in mAb serum levels of 1–3 mg/ml. rAAV vectors encodingthe heavy and light chains of a human anti-HIV mAb driven by aminimal CMV or EF1 alpha promoter yielded antibody serum levels of

140120100806040200Days

140120100806040200Days

1

10

100

1,000

10,000

100,000

DC

101

(µg/

ml)

1

10

100

1,000

10,000D

C10

1 m

Ab

(µg/

ml)

1 × 1011

2 × 1011

4 × 1011

1 × 1011

2 × 1011

4 × 1011

a b

10090807060504030

40

Days

Days

00

20

60

80

100

120

Per

cent

sur

viva

l

40

0

20

60

80

100

120

Per

cent

sur

viva

l

120100806040200

rAAV8 null

rAAV8 DC101

rAAV8 nullrAAV8 DC101

rAAV8 nullrAAV8 DC101

rAAV8 null

rAAV8 DC101

***

***

U87 MG

U87 MG

U87 MG

B16F10B16F10

B16F10

201713100

Days

3935322825211814117

Days

1,400

1,200

1,000

800

600

400

200

0

Tum

or v

olum

e (m

m3 )

Tum

or v

olum

e (m

m3 )

1,250

1,000

750

500

250

0

7550250

Days

50403020100

Days

1

10

100

1,000

10,000

1

10

100

1,000

10,000

DC

101

(µg/

ml)

DC

101

(µg/

ml)

2 × 1011

2 × 1011

a b c

d e f

*

*

Figure 5 Anti-tumor efficacy after rAAV8-mediated gene transfer of DC101. (a–f) rAAV8-CAG-HF2AL or rAAV8-null control vectors (2 � 1011 vg/mouse) were

administered through the tail vein into NCr nude mice on day 0 and serum DC101 concentrations were determined by rat IgG1 ELISA (mean 7 s.e.m.)

(a,d). On day 24, B16F10 (1 � 105 cells/mouse) (a–c) or U87 (5 � 106 cells/mouse) (d–f) tumor cells were injected subcutaneously into these mice and

tumor volume (mean 7 s.e.m.) (b,e) and survival (c,f) were evaluated. For tumor growth (b,e) and survival (c,f), the day of tumor challenge was regarded

as day 0. *, P o 0.05; ***, P o 0.001.

Figure 4 Expression of DC101 mAb in vivo by

rAAV8 vector-mediated gene transfer. (a,b) rAAV8

CAG H2AL (a) or rAAV8 CAG HF2AL (b) vectors

were administered to NCr nude mice at three

doses (1 � 1011, 2 � 1011 or 4 � 1011 vg/

mouse) through the portal vein. Mice were bled

weekly and DC101 serum levels determined by

ELISA (mean 7 s.e.m.).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

4–5 mg/ml22. The mAb serum levels described in these studies are belowthe therapeutic levels. So far, only high doses of recombinant adenoviralvectors have occasionally achieved mAb serum levels of B200 mg/ml23.However, mAb expression by adenoviral vectors is transient23, due tothe instrinsic immunogenicity of the vector backbone24.

rAAV is a preferred vector system when long-term gene expressionis desired. rAAV vectors can stably transduce host cells and are capableof expressing therapeutic proteins at constant levels in vivo followinga single vector administration. rAAV vectors have been shown toefficiently transduce quiescent cells of the muscle, liver, brain andeye24–27 and are currently under clinical evaluation28–31. One of theproblems of using rAAV vectors as mAb expression vectors, however,is their limited packaging capacity, which makes it difficult to expressmAb heavy and light chains from individual promoters in one vector.We used the short FMDV 2A sequence to mediate antibody heavy andlight chain expression from a single ORF, which enabled us toconstruct a single rAAV vector for the production of a full-lengthmAb. In combining the 2A technology with the recently identifiedAAV8 serotype that efficiently transduces the liver, mAb serum levelsof 41,000 mg/ml were achieved in mice after a single administrationof the rAAV vector. Antibody expression remained at high levels forover 4 months. In summary, this expression system presents a feasiblegene therapy approach for long-term delivery of antibodies at highlevels in vivo. Furthermore, the high concentration of mAb achievedin vivo after rAAV8-mediated gene transfer will make vector manu-facturing possible for human use.

In our study, we were able to show that the 23 additional aminoacids derived from the 2A sequence at the C terminus of the heavychain can be efficiently removed by adding a furin cleavage site next tothe 2A sequence. This modification results in a significant increase ofmAb serum levels and generates an antibody that more closely re-sembles the native protein, thereby eliminating possible adverse effects.

There are currently contrasting results regarding whether theleading signal peptide is necessary for the second protein to enterinto the endoplasmic reticulum (ER) for protein secretion. In yeast,the lack of the signal peptide at the N terminus of the second proteinresulted in cytosolic localization of the protein10. In contrast, inmammalian cells, the second protein can enter the ER lumen withouta signal sequence11. It is currently not clear, how the signal sequence ofthe second protein affects overall antibody expression. Nevertheless,the mAb light chain in our H2AL and HF2AL constructs contains thenative signal peptide at its N terminus. Since the signal peptide iscleaved during protein secretion, inclusion of the leading sequenceremoves the amino acid proline from the N terminus of the antibodylight chain, that is derived from the 2A sequence. Therefore, thesecreted antibody light chain expressed from these cassettes is expectedto retain its native sequence.

In addition to providing an option for an alternative long-term anti-body therapy, we believe that this technology may be of great value forthe generation of mAb producer cell lines. Current technologies usedfor the generation of stable mAb producer cell lines are labor intensiveand time consuming. We are currently evaluating this technology forthe generation of stable mAb producer cell lines and have preliminarydata demonstrating that the 2A technology combined with viral vectorsenables the rapid identification of high mAb producer cell clones.

Furthermore, the 2A-furin/rAAV8 technology described within thispaper may be a useful tool to validate mAb targets in vivo for drugdevelopment, study protein function in vivo by blocking their biolo-gical activity or carry out cell-depletion research to study the functionof a particular cell type in an in vivo system. The 2A-furin/rAAV8technology for mAb gene expression may also have a great potential

for delivery of neutralizing mAbs to induce passive immunity in someinfectious diseases.

In summary, the 2A-furin technology may have broad applicationas an in vivo mAb therapy for the treatment of cancer or other chronicdiseases, and as a research tool for studies in which high and sustainedmAb serum levels are required or for the rapid generation of stablemAb producer cell lines.

METHODSPlasmid construction. Total RNA from DC101 hybridoma cells was purified

using RNeasy kit (Qiagen). First stream cDNA was synthesized from total RNA

using specific primers to heavy or light chain constant region sequences.

Variable regions of the antibody heavy and light chains, including their signal

peptide sequences, were amplified with the rapid amplification of cDNA ends

cloning kit (BD Biosciences Clontech). The VH and VL were cloned into the

pCR 2.1 plasmid using the TA cloning kit (Invitrogen) and sequenced. The

consensus VH and VL sequences were determined based on sequence data from

the clones derived from multiple independent PCRs. Constant regions of both

heavy and light chains were also cloned from cDNA. Variable and constant

regions of heavy and light chains were joined together by PCR reaction to

generate full-length heavy and light chains.

To generate the constructs containing the 2A self-processing sequence, the

cDNA oligo for a 24–amino acid FMDV 2A peptide was synthesized (Bio-

source) based on the sequence APVKQTLNFDLLKLAGDVESNPGP32. The

cDNA encoding antibody heavy chain, 2A and light chain was assembled by

PCR and was cloned into a plasmid downstream of a CAG promoter. The final

plasmid, pH2AL, contains a single ORF consisting of a full-length heavy chain,

the 2A sequence, and full-length light chain (Fig. 1). Both heavy and light

chains include their native signal peptide sequences at their N termini. The

plasmid also includes a bovine growth hormone poly A sequence at the 3¢ end.

For the construct that also contains a furin cleavage site, the sequence that

encodes the furin cleavage site, RAKR, was inserted by PCR between the

antibody heavy chain and the 2A sequence. The cDNA that encodes an antibody

heavy chain, a furin cleavage site (RAKR), the 2A sequence and an antibody light

chain was cloned into the plasmid downstream of the CAG promoter (pHF2AL).

To express full-length antibody heavy and light chains with IRES sequences,

the DC101 heavy chain, an IRES sequence derived from the EMCV33 and the

DC101 light chain were inserted into the plasmid downstream of the CAG

promoter. Both heavy and light chain cDNAs end with a stop codon (Fig. 1).

rAAV vector preparation. Subconfluent HEK 293 cells were cotransfected

using the calcium phosphate method with the pAAV-CAG-DC101 vector

plasmid in combination with the AAV8 serotype helper plasmid p5e18-VD2/8

(ref. 14) and pXX-6 (ref. 34). Forty-eight hours after transfection, cells were

harvested using PBS/EDTA (10 mM) and lysed by three freeze/thaw cycles in

cell lysis buffer (150 mM NaCl, 50 mM HEPES, pH 7.6). Lysates were treated

with 250 U/ml benzonase for 15 min at 37 1C and cellular debris was removed

by centrifugation. The cleared cell lysate was fractionated by ammonium sulfate

precipitation and the rAAV virions were isolated on two sequential CsCl

gradients. The gradient fractions containing rAAV were dialyzed against sterile

PBS containing CaCl2 and MgCl2, and stored at –80 1C. Viral titers were

determined by dot-blot analysis. Briefly, rAAV preparations were treated with

DNaseI followed by proteinase K in the presence of 0.5% SDS and 10 mM

EDTA to liberate the rAAV genomes, followed by phenol chloroform extraction

and ethanol precipitation. Viral DNA was denatured in alkali and applied to a

nylon membrane. Dilutions of the corresponding vector plasmid were used as

standards to determine the rAAV virion copy number. A radioactive probe

specific for the rAAV transgene was hybridized to DNA on the filter and the

filter was exposed to film followed by quantification of radioactivity by a

b-counter (Perkin Elmer). Biological activities of the purified rAAV were

determined by DC101 antibody expression in HEK 293 or HuH7 cells following

rAAV transduction in vitro.

Cell culture and transfection. HEK 293, B16F10, U87MG and DC101

hybridoma cell lines were obtained from ATCC. HuH 7 cells were from Jing-

Hsiung Ou, University of Southern California. HEK 293 cells were cultured in

Iscove’s Modified Dulbecco’s Medium (Invitrogen), supplemented with 3 mM


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

L-glutamine and 10% fetal bovine serum. HuH 7, B16F10, U87 and DC101

hybridoma cells were cultured in DMEM medium (Invitrogen), supplemented

with 3 mM L-glutamine and 10% fetal bovine serum. To produce DC101 mAb

from the hybridoma, exhausted cell culture supernatants were harvested. For

DC101 expression in vitro, plasmid DNA was purified using a plasmid DNA

mega purification kit (Qiagen) and cells were transfected in 6-well tissue

culture plates or 10-cm dishes with FuGene 6 transfection reagent (Roche).

Twenty-four hours after transfection, the cell culture medium was removed and

the cells were fed fresh medium with or without 10% FBS. The supernatants

were collected after 48 or 72 h.

ELISA and western blots. The DC101 antibody concentrations in mouse

serum or cell culture supernatants were determined using a commercial ELISA

assay kit for rat IgG1 (Bethyl Lab). For protein analysis in polyacrylamide gels,

protein samples were separated in precast Tris-glycine gels (Invitrogen) under

reducing or nonreducing conditions. For western blot analysis, proteins in

polyacrylamide gels were transferred to nitrocellulose membranes. The mem-

branes were blocked with 5% nonfat dry milk and incubated with a goat anti-

rat IgG antibody (Calbiochem) conjugated with horseradish peroxidase (HRP).

Protein bands were visualized by exposure on X-ray films (Kodak) after the

membranes were treated with enhanced chemiluminescence solution (Pierce).

Antibody binding and blocking assay. DC101 mAb antibodies were expressed

in the supernatants of hybridoma cells (hybridoma), from HEK 293 cells

transfected with the H2AL (H2AL in vitro) or the HF2AL plasmids (HF2AL

in vitro), or in the sera of mice injected with AAV8 H2AL (H2AL in vivo) or

HF2AL (HF2AL in vivo) vectors. To evaluate binding activity of the DC101

antibody to mVEGFR2 (Flk-1), 96-well ELISA plates were coated with 200 ng/

ml of recombinant Flk-1-Fc protein (R&D Systems). The plates were blocked

with 5% nonfat dry milk and incubated with various concentrations of DC101

antibody. The plates were incubated with goat anti-rat IgG antibody conjugated

to HRP and staining revealed by peroxidase substrate. Medium from naive

HEK 293 cells served as a negative control (control). The plates were read in a

microplate reader at absorbance of 405 nm.

To evaluate the neutralizing effect of DC101 mAb on VEGF-Flk-1 binding,

96-well plates were coated with 500 ng/ml of recombinant human VEGF165

(R&D Systems). Recombinant Flk-1-Fc protein was preincubated with

various concentrations of DC101 mAb expressed from hybridoma cells

(Hybridoma), HEK 293 cells transfected with the H2AL (H2AL in vitro) or

the HF2AL (HF2AL in vitro) plasmids, or sera of mice injected with AAV8

H2AL (H2AL in vivo) or AAV8 HF2AL (HF2AL in vivo) vectors. After blocking

with 5% nonfat dry milk, 50 ng/ml of recombinant Flk-1-Fc (R&D Systems),

which had been preincubated with various concentrations of DC101 antibody,

was added to each well and incubated at 37 1C for 1 h. The plates were washed

with Tris-buffered saline, incubated with biotin-conjugated goat anti-Flk-1

antibody (R&D Systems), washed again and the staining revealed with strepta-

vidin-HRP (DB Pharmingen) and peroxidase substrate. Medium from naive

HEK 293 cells was used as a negative control (control). VEGF-Flk-1 binding

was detected at an absorbance of 405 nm with an anti-Flk-1 antibody

conjugated to HRP.

His-tagged antibody expression, purification and mass spectrum analysis. A

hydrodynamic gene transfer method35 was used to express his-tagged DC101

antibody from plasmid in mice. Briefly, a pAAV-CAG-DC101 antibody HF2AL

construct with 6� histidine residues (his-tag) at the C terminus of the light

chain was constructed. Plasmid DNA (50 mg in 2 ml of PBS) was rapidly

injected into a NCr nu/nu mouse via the tail vein. His-tagged DC101 antibody

in mouse serum was purified using a nickel column (Qiagen). Purified proteins

were separated in a SDS-PAGE gel under reducing conditions and the antibody

heavy chain band was isolated, trypsin digested and analyzed in a mass

spectrometer. To determine the amino acid sequence, we isolated the peptide

fragment peaks from the mass spectrum and analyzed them by post source

decay (PSD) analysis.

Antibody expression in vivo by rAAV vector–mediated gene transfer. Female

NCr nu/nu mice (6–8 weeks old) were obtained from Taconic. All mice were

housed under specific-pathogen-free conditions and treated according to the

Institute for Laboratory Animal Research Guide for the Care and Use of

Laboratory Animals. rAAV vector at 1 � 1011, 2 � 1011 and 4 � 1011 vg/

mouse was injected into mice (n ¼ 5 in each group) via a surgically implanted

portal vein catheter. Mice were bled by alternate retro-orbital puncture at each

scheduled time point for up to 6 months for analysis of DC101 expression.

Blood samples may be collected from the orbital sinus of anesthetized mice at

scheduled intervals.

Mouse tumor models. Female NCr nu/nu mice (n ¼ 10–12 in each group)

were injected with rAAV8-CAG-HF2AL vector through intravenous adminis-

tration via tail veins at a dose of 2 � 1011 vg/mouse in 200 ml of PBS. Mice in

the control group were injected with the same dose of rAAV8-null vector. To

monitor serum DC101 levels, mice were bled weekly by alternate retro-orbital

puncture. At day 24 after rAAV administration, B16F10 melanoma (1 � 105

cells/mouse in 200 ml PBS) or human U87 MG glioma (5 � 106 cells/mouse in

200 ml PBS/Matrigel at 1:1 ratio) cells were implanted subcutaneously into the

flanks of mice. Tumor volume was measured twice a week with a caliper and

calculated by the formula of [(width � length � height)/2]. For survival

studies, end points were based on the pre-established criteria that include

tumor volume, body weight loss, degree of tumor necrosis and the general

health of animals.

ACKNOWLEDGMENTSThe authors would like to thank Mingxia Shi, Sandra Sanchez, Lei Xu, GailColbern and the animal service group of Cell Genesys for technical assistance, JohnLeszyk at the University of Massachusetts Medical School for carrying out massspectrometry analysis and Peter Working for critical reading of the manuscript.

COMPETING INTERESTS STATEMENTThe authors declare competing financial interests (see the Nature Biotechnologywebsite for details).

Received 20 January; accepted 10 March 2005


1. Hudson, P.J. & Souriau, C. Engineered antibodies. Nat. Med. 9, 129–134 (2003).2. Green, L.L. Antibody engineering via genetic engineering of the mouse: XenoMouse

strains are a vehicle for the facile generation of therapeutic human monoclonalantibodies. J. Immunol. Methods 231, 11–23 (1999).

3. Maloney, D.G. et al. IDEC-C2B8 (Rituximab) anti-CD20 monoclonal antibody therapy inpatients with relapsed low-grade non-Hodgkin’s lymphoma. Blood 90, 2188–2195(1997).

4. Grimm, D. & Kay, M.A. From virus evolution to vector revolution: use of naturallyoccurring serotypes of adeno-associated virus (AAV) as novel vectors for human genetherapy. Curr. Gene Ther. 3, 281–304 (2003).

5. Mizuguchi, H., Xu, Z., Ishii-Watabe, A., Uchida, E. & Hayakawa, T. IRES-dependentsecond gene expression is significantly lower than cap-dependent first gene expressionin a bicistronic vector. Mol. Ther. 1, 376–382 (2000).

6. Ryan, M.D. & Drew, J. Foot-and-mouth disease virus 2A oligopeptide mediated cleavageof an artificial polyprotein. EMBO J. 13, 928–933 (1994).

7. Donnelly, M.L., Gani, D., Flint, M., Monaghan, S. & Ryan, M.D. The cleavage activitiesof aphthovirus and cardiovirus 2A proteins. J. Gen. Virol. 78, 13–21 (1997).

8. Donnelly, M.L. et al. Analysis of the aphthovirus 2A/2B polyprotein ‘cleavage’ mechan-ism indicates not a proteolytic reaction, but a novel translational effect: a putativeribosomal ‘skip’. J. Gen. Virol. 82, 1013–1025 (2001).

9. Szymczak, A.L. et al. Correction of multi-gene deficiency in vivo using a single ‘self-cleaving’ 2A peptide-based retroviral vector. Nat. Biotechnol. 22, 589–594 (2004).

10. de Felipe, P., Hughes, L.E., Ryan, M.D. & Brown, J.D. Co-translational, intraribosomalcleavage of polypeptides by the foot-and-mouth disease virus 2A peptide. J. Biol.Chem. 278, 11441–11448 (2003).

11. de Felipe, P. & Ryan, M.D. Targeting of proteins derived from self-processing poly-proteins containing multiple signal sequences. Traffic 5, 616–626 (2004).

12. Prewett, M. et al. Antivascular endothelial growth factor receptor (fetal liver kinase 1)monoclonal antibody inhibits tumor angiogenesis and growth of several mouse andhuman tumors. Cancer Res. 59, 5209–5218 (1999).

13. Niwa, H., Yamamura, K. & Miyazaki, J. Efficient selection for high-expression trans-fectants with a novel eukaryotic vector. Gene 108, 193–199 (1991).

14. Gao, G.P. et al. Novel adeno-associated viruses from rhesus monkeys as vectors forhuman gene therapy. Proc. Natl. Acad. Sci. USA 99, 11854–11859 (2002).

15. Bakker, J.M., Bleeker, W.K. & Parren, W.H.I. Therapeutic antibody gene transfer: anactive approach to passive immunity. Mol. Ther. 10, 411–416 (2004).

16. Davis, T.A. et al. Rituximab anti-CD20 monoclonal antibody therapy in non-Hodgkin’slymphoma: safety and efficacy of re-treatment. J. Clin. Oncol. 18, 3135–3143 (2000).

17. Lin, Y.S. et al. Preclinical pharmacokinetics, interspecies scaling, and tissue distribu-tion of a humanized monoclonal antibody against vascular endothelial growth factor.J. Pharmacol. Exp. Ther. 288, 371–378 (1999).

18. Armbruster, C. et al. A phase I trial with two human monoclonal antibodies (hMAb 2F5,2G12) against HIV-1. AIDS 16, 227–233 (2002).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

19. Perez, N. et al. Regulatable systemic production of monoclonal antibodies by in vivomuscle electroporation. Genet. Vaccines Ther. 2, 2 (2004).

20. Noel, D. et al. In vitro and in vivo secretion of cloned antibodies by genetically modifiedmyogenic cells. Hum. Gene Ther. 8, 1219–1229 (1997).

21. Noel, D., Pelegrin, M., Brockly, F., Lund, A.H. & Piechaczyk, M. Sustained systemicdelivery of monoclonal antibodies by genetically modified skin fibroblasts. J. Invest.Dermatol. 115, 740–745 (2000).

22. Lewis, A.D., Chen, R., Montefiori, D.C., Johnson, P.R. & Clark, K.R. Generation ofneutralizing activity against human immunodeficiency virus type 1 in serum by anti-body gene transfer. J. Virol. 76, 8769–8775 (2002).

23. Noel, D. et al. High in vivo production of a model monoclonal antibody on adenoviralgene transfer. Hum. Gene Ther. 13, 1483–1493 (2002).

24. Jooss, K. & Chirmule, N. Immunity to adenovirus and adeno-associated viral vectors:implications for gene therapy. Gene Ther. 10, 955–963 (2003).

25. Monahan, P.E., Jooss, K. & Sands, M.S. Safety of adeno-associated virus gene therapyvectors: a current evaluation. Expert Opin. Drug Saf. 1, 79–91 (2002).

26. Lu, Y. Recombinant adeno-associated virus as delivery vector for gene therapy–a review.Stem. Cells. Dev. 13, 133–145 (2004).

27. Flotte, T.R. et al. Phase I trial of intramuscular injection of a recombinant adeno-associated virus alpha 1-antitrypsin (rAAV2-CB-hAAT) gene vector to AAT-deficientadults. Hum. Gene Ther. 15, 93–128 (2004).

28. Moss, R.B. et al. Repeated adeno-associated virus serotype 2 aerosol-mediated cysticfibrosis transmembrane regulator gene transfer to the lungs of patients with cysticfibrosis: a multicenter, double-blind, placebo-controlled trial. Chest 125, 509–521(2004).

29. Manno, C.S. et al. AAV-mediated factor IX gene transfer to skeletal muscle in patientswith severe hemophilia B. Blood 101, 2963–2972 (2003).

30. Janson, C. et al. Clinical protocol. Gene therapy of Canavan disease: AAV-2 vector forneurosurgical delivery of aspartoacylase gene (ASPA) to the human brain. Hum. GeneTher. 13, 1391–1412 (2002).

31. Luo, J. et al. J. Subthalamic GAD gene therapy in a Parkinson’s disease rat model.Science 298, 425–429 (2002).

32. Ryan, M.D., King, A.M. & Thomas, G.P. Cleavage of foot-and-mouth disease viruspolyprotein is mediated by residues located within a 19 amino acid sequence. J. Gen.Virol. 72, 2727–2732 (1991).

33. Jang, S.K., Pestova, T.V., Hellen, C.U., Witherell, G.W. & Wimmer, E. Cap-independenttranslation of picornavirus RNAs: structure and function of the internal ribosomal entrysite. Enzyme 44, 292–309 (1990).

34. Xiao, X., Li, J. & Samulski, R.J. Production of high-titer recombinant adeno-associatedvirus vectors in the absence of helper adenovirus. J. Virol. 72, 2224–2232 (1998).

35. Liu, F., Song, Y. & Liu, D. Hydrodynamics-based transfection in animals by systemicadministration of plasmid DNA. Gene Ther. 6, 1258–1266 (1999).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Gene knockdown by large circular antisense forhigh-throughput functional genomicsYun-Han Lee1,4,6, Ik-Jae Moon1,6, Bin Hur1,6, Jeong-Hoh Park1, Kil-Hwan Han1, Seok-Yong Uhm1,Yong-Joo Kim1, Koo-Jeong Kang2, Jong-Wook Park3, Young-Bae Seu4, Young-Ho Kim4 & Jong-Gu Park1,5

Single-stranded genomic DNA of recombinant M13 phages was tested as an antisense molecule and examined for its

usefulness in high-throughput functional genomics. cDNA fragments of various genes (TNF-a, c-myc, c-myb, cdk2 and cdk4)

were independently cloned into phagemid vectors. Using the life cycle of M13 bacteriophages, large circular (LC)-molecules,

antisense to their respective genes, were prepared from the culture supernatant of bacterial transformants. LC-antisense

molecules exhibited enhanced stability, target specificity and no need for target-site searches. High-throughput functional

genomics was then attempted with an LC-antisense library, which was generated by using a phagemid vector that incorporated

a unidirectional subtracted cDNA library derived from liver cancer tissue. We identified 56 genes involved in the growth of

these cells. These results indicate that an antisense sequence as a part of single-stranded LC-genomic DNA of recombinant

M13 phages exhibits effective antisense activity, and may have potential for high-throughput functional genomics.

Gene expression can be specifically reduced or ablated in cells after theuptake of antisense molecules complementary to a specific mRNAsequence. Antisense inhibition of gene expression is believed to beachieved through RNaseH activity after the formation of an antisenseDNA-mRNA duplex or through steric hindrance of movement/binding of the ribosomal complex1. Gene silencing by antisensetreatment has been considered ideal for functional analysis of genesand further for drug target discovery2. Intense efforts have been madeto develop antisense anticancer agents that eliminate aberrant expres-sion of genes involved in tumor initiation and progression3–8. Theefficacy of antisense oligonucleotides (AS-oligos) has been validated inanimal models9–14.

We have previously described a series of distinct antisense molecules,with closed structures lacking exonuclease active sites, resulting inmuch enhanced stability in biologic fluids15,16. These results promptedus to investigate the potential of the single-stranded circular genome ofM13 bacteriophages (phages) as antisense molecules. A recombinantM13 phagemid vector was engineered to produce a single-strandedcircular genome containing an antisense sequence, which was thentested for enhanced stability and specific antisense activity.

Various methods have been devised to study gene expression17–23,however, the information generated has been limited to differential orsequential expression profiles of genes in different tissues or cells.Rapid accumulation of genomic sequence information and expressionprofiling has created a bottleneck in subsequent definitive genefunctionalization and/or target validation. Most definitive functiona-lization of genes has been performed with various conventional

gain-of-function or loss-of-function studies. Loss-of-function studieshave been done either with gene knockdown using conventionalantisense24,25 or its related technologies26–28, or with gene knockoutusing homologous recombination29,30. These approaches are limited inthat they must be done individually. Construction of an extensiveantisense library may provide an answer to this information bottleneckfor massive gene functionalization. AS-oligo libraries have been par-tially established and used to obtain functional data of a large numberof genes. Constructing such a library, however, can be costly and timeconsuming because a target site search must be done31–33. Thus,another approach is needed to facilitate construction of an antisenselibrary. LC-antisense constructs may provide a salient advantage inlibrary construction because they do not require target site searches.

In the present study, we tested the efficacy of single-strandedcircular genomic DNA of M13 phagemids as antisense moleculeswith regard to enhanced stability and target-specific reduction of geneexpression. An LC-antisense library was constructed by using cDNAprepared from hepatoblastoma tissue by subtractive hybridization,which was then used to screen genes involved in the growth of a livercancer cell line using a high-throughput approach.

RESULTS

Construction and purification of LC-antisense molecule

Covalently closed circular antisense molecules are unusually stable andeffective in reducing target gene expression, suggesting the potential ofthe single-stranded circular genome of bacteriophage M13 as anantisense molecule. The F1 replication origin of the M13 phagemid


1WelGENE Inc., 71B 4L, Development Sector 2-3, Sungseo Industrial Park, Dalseogu, Daegu, 704-230, South Korea. 2Department of General Surgery, Dongsan MedicalCenter, Keimyung University, 3Department of Immunology, Keimyung University School of Medicine, 4Department of Microbiology, College of Natural Sciences, KyungpookNational University, 1370 Sangyeokdong, Bookgu, Daegu, 702-701, South Korea. 5Department of Medical Genetic Engineering, Keimyung University School of Medicine,Dongsan Medical Center, 194 Dongsandong, Joonggu, Daegu, 700-712, South Korea. 6These authors contributed equally to this work. Correspondence should beaddressed to J.-G.P. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

was used to generate a single-stranded circular phage genome harbor-ing either the antisense or sense sequence for a target gene. Rat TNF-acDNA was cloned into a pBS KS (�) vector to produce the antisensesequence as a part of the phage genome (see Supplementary Fig. 1online). The phage genomic LC-antisense molecule, designated asTNFa-LCAS in this study, was isolated from the culture supernatantof bacterial cells that were transformed with recombinant phagemidharboring rat TNF-a cDNA34. Large-scale purification of the LC-antisense molecule was done by gel filtration column chromatography.The antisense sequence in the single-stranded phage genomic DNA wasconfirmed by DNA sequencing using the T3 primer (data not shown).LC-antisense molecules to the c-myc, c-myb, cdk2 and cdk4 genes werealso constructed by the same approach and designated as c-myc-LCAS,c-myb-LCAS, cdk2-LCAS and cdk4-LCAS, respectively. Similarly, bothLCSE (single-stranded phage genome containing the sense sequence ofeach target gene) and LCSS (single-stranded phage genome devoid ofan insert sequence) were also prepared as control molecules.

LC-antisense molecules were expected to be resistant to exonu-cleases because of their circular structure15,16. When single-stranded(ss) TNFa-LCAS was incubated with either XhoI or exonuclease III,the antisense molecules were found to be largely intact after 3 h(Fig. 1a). In contrast, when XhoI was added to the double-stranded(ds) recombinant M13 phagemid DNA harboring the TNF-a cDNA,the dsDNA was restriction digested, generating two linear bands of 3.4and 0.5 kb on an agarose gel. Furthermore, the dsDNA was digested tocompletion by the combination of XhoI and exonuclease III, leavingno detectable DNA band. The fact that TNFa-LCAS is ssDNA wasreconfirmed by the efficient digestion of the circular molecules with S1nuclease, which specifically cuts ssDNA regardless of sequence com-position. When TNFa-LCAS was combined with cationic lipids, alarge fraction of the antisense molecules remained intact after anextended period of incubation in fetal bovine serum (FBS), even after24 h incubation in 30% FBS (Fig. 1b).

Effective inhibition of target gene expression by LC-antisense

Encouraged by the enhanced tolerance of LC-antisense molecules tonucleases, we tested TNFa-LCAS for antisense activity. TNFa-LCAS(1.4 nM) was complexed with cationic lipids and added to the ratmonocytic WRT7/P2 cell line in which TNF-a expression was inducedby lipopolysaccharide (LPS) treatment. When treated with TNFa-LCAS, the cells were shown to have a substantially reduced levelof TNF-a mRNA (Fig. 1c). In contrast, cells treated with either

TNFa-LCSE (the sense strand of TNF-a DNA) or LCSS (single-stranded vector genomic DNA) did not show much reduction ofTNF-a mRNA. The RT-PCR band of TNF-a was confirmed bySouthern hybridization with a probe that bound to the internal regionof the amplified DNA fragments (Fig. 1c). To confirm that thetreatment of LC-antisense leads to the eventual blockade of proteinsynthesis from target mRNA, we transfected WRT7/P2 cells withTNFa-LCAS and measured the level of TNF-a protein secreted fromthe transfectants. Commensurate with the reduction of TNF-a mRNAlevel, the level of TNF-a in the cell culture supernatant was also reducedby more than 90% after treatment with TNFa-LCAS (Fig. 1d). Incontrast, none of the two control molecules, TNFa-LCSE and LCSS,significantly reduced the level of TNF-a protein in WRT7/P2 transfec-tants. After observing the effective antisense activity of TNFa-LCAS, weperformed experiments to determine if LC-antisense molecules toother genes, such as c-myc and c-myb, would also block expression oftheir respective target genes. When 1.12 nM of c-myc-LCAS was addedto K562 cells, c-myc mRNA was reduced by about 70% compared tothat obtained after c-myc-LCSE transfection (Fig. 1e). Similarly, treat-ment of 1.12 nM of c-myb-LCAS to K562 cells reduced c-myb mRNAlevel by about 80% (Fig. 1f). The treatment of c-myc-LCAS did notaffect the expression of the c-myb gene and vice versa (Fig. 1e,f). Theseresults show that LC-antisense can efficiently reduce gene expression insmaller amounts than most conventional antisense molecules.

Target specificity and antisense activity of LC-antisense

If LC-antisense is to be effective, it must be target specific, especially inregard to its large length. Thus, sequence specificity of LC-antisensemolecules was examined by an RNase protection assay (RPA) aftertreatment of HeLa cells with cdk2-LCAS. Whereas cdk2-LCAS reducedcdk2 expression in the cells at time points of 24 and 48 h, the antisensedid not substantially affect expression of other genes, cdk1, p16, L32and GAPDH (Fig. 2a). CDK2 levels in HeLa cells transfected withcdk2-LCAS were also examined by western blotting analysis. Whereascdk2-LCAS at a concentration of 0.8 or 1.6 nM reduced the intracel-lular level of CDK2 by more than 80%, an equal amount of cdk2-LCSE

c-myc

c-myb

c-myb-LCSEc-myb-LCAS

β-actin

LipidsLipids (nM)

c-myb

c-myc

β-actin

(nM) 1.121.12 0.560.56 1.121.12 0.560.56

c-myc-LCSEc-myc-LCAS

LCSSrTNFα-LCAS rTNFα-LCSE

*

* *

0

0.05

0.1

0.15

0.2

0.25

Abs

orba

nce

(450

nm

)

rTNF-α

β-actin

rTNF-α

TN

Fα-

LCS

E

TN

Fα-

LCA

S

0.6

2.0

4.3

(kb)

LCS

S

Lipi

ds

Siz

e m

arke

r

Sha

m c

ontr

ol

48 h24 h16 h6 h4 h3 h2 h1 h

30% FBS

0.6

2.04.3

(kb)

Xh

ol/E

xoIII

S1

nucl

ease

Xh

ol

Con

trol

Siz

e m

arke

r

ds ss ds ss ds ss ds ss

a b

c d

e f

Figure 1 Stability and antisense activity of LC-antisense. (a) Characterization

of TNFa-LCAS molecules. Either double-stranded (ds) recombinant

TNFa-phagemid or single-stranded (ss) TNFa-LCAS was incubated with the

restriction endonuclease XhoI, S1 nuclease or XhoI/exonuclease III and run

on a 1% agarose gel with sham-treated controls. (b) Stability test of LC-

antisense molecules. TNFa-LCAS plus cationic lipid complexes were treated

with 30% FBS for different periods of time as indicated and run on a 1%

agarose gel with a sham-treated control. (c) Antisense activity of TNFa-LCAS

on TNF-a mRNA levels in WRT7/P2 cells. RT-PCR analysis was carried out

with two sets of primers, either TNF-a primers or b-actin primers. Southern

blotting, shown in the bottom panel, was carried out to detect TNF-aexpression. (d) Reduced expression of TNF-a by TNFa-LCAS treatment.

ELISA of TNF-a in medium: WRT7/P2 cells transfected with TNFa-LCAS,TNFa-LCSE or LCSS. Each bar value represents the mean 7 s.d. of

triplicate experiments. Statistical significance was calculated with student’s

t-test (analysis of variance, * P o 0.05). (e,f) Indicated amounts of

LC-antisense molecules to c-myc (e) and c-myb (f) were transfected into

K562 cells. Amplified PCR fragments of each target gene were run on a

1% agarose gel and visualized with ethidium bromide staining.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

had no substantial effect (see Supplementary Fig. 2 online). Interest-ingly, the level of cdk4 mRNA was also reduced by the treatment ofcdk2-LCAS. The result may be explained by the coordinate regulationbetween the two G1 phase-specific cell cycle regulators in a sequentialfashion35,36. However, it was necessary to rule out the off-target effectof cdk2-LCAS, therefore, we monitored the changes in cdk2, cdk4 andcdk6 gene expression after either cdk2- or cdk4-LCAS treatment atearlier time points using real-time quantitative RT-PCR. Cdk4 sharesconserved regions in some of its sequence to those of cdk2 and cdk6,the other two cell cycle regulators. Nucleotides 749–848 of the cdk4cDNA (GenBank accession number NM_000075) exhibits 75%sequence similarity to a region of the cdk2 gene (NM_001798) and74% to a region of cdk6 gene (NM_001259), respectively. There is,however, no sequence similarity between cdk2 and cdk6 genes. A time-course experiment was initially carried out by transfection of 2 � 104

HeLa cells by various amounts of cdk2- or cdk4-LCAS to monitor theoptimal concentration of the antisense and time points for effectiveknockdown of their respective gene expression. In addition, the activityof cdk2-LCAS was compared to that obtained with cdk2 siRNA. When

0.8, 1.6 and 2.4 nM of either cdk2- or cdk4-LCAS was added to HeLacells, target mRNA was reduced by about 40–75%, in a dose-dependentmanner at 6, 12 and 24 h (see Supplementary Fig. 3a,b). In contrast,cells that were treated with either LCSE or LCSS did not showsubstantial reduction of either cdk2 or cdk4 gene expression. Theinfluence on the expression of noncognate cdk genes was thenexamined after cdk2- and cdk4-LCAS treatment, respectively. When1.6 nM of cdk2-LCAS was added to HeLa cells, the level of cdk4 mRNAwas not changed substantially at 6 h and 12 h, but was decreased byabout 48% at 24 h (Fig. 2b). Similarly, 50 nM cdk2 siRNA reducedtarget cdk2 RNA by about 48% compared to the sham treatment, butcdk4 RNA was not reduced at earlier time points. However, after the24 h incubation with cdk2 siRNA, cdk4 mRNA was reduced byapproximately 47% (see Supplementary Fig. 3c online).

Further, when 1.6 nM of cdk4-LCAS was added to HeLa cells, theexpression of cdk2 and cdk6 mRNA was not affected at 6 h and 12 h,but was decreased by approximately 64% and 47% at 24h, respectively(Fig. 2c). The lack of off-target expression interference among thefunctionally associated genes, cdk2, cdk4 and cdk6, at earlier time

Figure 2 Target specificity and antisense activity

of LC-antisense. Detection of gene expression

was done after the transfection of LC-antisense

into HeLa. (a) RPA assay for expression of various

genes in HeLa cells. Total RNA was extracted at

24 h or 48 h after treatment with cdk2-LCAS or

at 48 h after sham treatment or LCSS treatment

in 6-well plates. After hybridization of theextracted RNA with biotin-labeled probes or yeast

tRNA (negative control), the samples were run

together with unhybridized probes as markers

on a denaturing polyacrylamide gel. Reference

mRNAs: L32 and GAPDH. Irregular splotches

are shown on L32 and GAPDH bands of the two

right-hand lanes. PISSLRE, a human CDC-2

related protein kinase, (b,c) Real-time RT-PCR

analysis of cdk gene expression in HeLa cells

treated with cdk2-LCAS (b) or cdk4-LCAS (c).

Total RNA was extracted at 6 h, 12 h and 24 h

after treatment with 1.6 nM of LC-antisense or at

24 h after sham, lipids alone, LCSE and LCSS treatments in 24-well plates. In all real-time RT-PCR experiments, expression is calculated relative to b-actin

and is normalized to sham treatment. Each bar value represents the mean 7 s.d. of triplicate experiments. (d,e) Effect of LC-antisense on proliferation of

human cancer cell lines was measured by MTT assays after transfection of LC-antisense molecules. (d) Effects of two types of c-myb-LCAS on K562 cell

proliferation. C-myb-LCAS1 and c-myb-LCAS2 contain 0.5 kb or 1.5 kb of the c-myb cDNA sequence, respectively. (e) Cdk4-LCAS on MCF-7 cell

proliferation. AS, cdk4-LCAS; SE, cdk4-LCSE. Each bar value represents the mean 7 s.d. of triplicate experiments.

0.56 (nM)0.56LCSSSEASAS SE

0.28

0

0.1

0.2

0.3

Abs

orba

nce

(570

nm

)

LCSSLCSEAS2AS1c-myb-LCAS

(0.56 nM)

0.1

0.5

0.9

Abs

orba

nce

(570

nm

)

24 h12 h6 hCdk4-LCAS (1.6 nM)

Cdk2-LCAS (1.6 nM)

Cdk

2-LC

AS

(48

h)

LCS

S (

48 h

)

Sha

m (

48 h

)

Yeas

t tR

NA

Pro

bes

Cdk

2-LC

AS

(24

h)

Cdk4Cdk6Cdk2

Cdk4Cdk2

LCSE (2

4 h)

LCSS (2

4 h)

Lipids

(24

h)

Sham

(24

h)

24 h12 h6 h

LCSE (2

4 h)

LCSS (2

4 h)

Lipids

(24

h)

Sham

(24

h)

0

20

40

60

80

100

120

Nor

mal

ized

exp

ress

ion

0

20

40

60

80

100

120

Nor

mal

ized

exp

ress

ion

GAPDH

L32

p16

PISSLRE

p21

p27Cdk4Cdk3

Cdk2Cdk1

a b d

c e

55

Con

trol

DN

ALi

pids

Sha

m c

ontr

ol5045403530252015105WGSL0

20

40

60

80

100

120

Per

cent

of a

bsor

banc

e (5

70 n

m) HepG2 cell growth

(ix)(viii)(vii)

Treated withLC-antisensemolecules

(iv) (v) (vi)

Controltreatments

(i) Sham control (ii) Lipids (iii) Control DNA

a b

Figure 3 High-throughput functional analysis to identify genes involved in the growth of liver cancer cells. (a) Growth inhibition of HepG2 cells after

transfection with LC-antisense library was examined by light microscopy 4 d after transfection (200� magnification). (i)–(iii), control treatments as indicated;

(iv)–(ix), HepG2 cells treated with different LC-antisense molecules. A representative example of the data acquired from treatments with 6 out of 1,200

kinds of LC-antisense is shown. (b) LC-antisense species of 56 random genes were transfected to a HepG2 cell line in a macroarray configuration. The

transfectants were examined for growth inhibition by MTT assays in triplicate. Cells that were sham-treated, treated with lipids alone and treated with control

DNA plus lipid complexes were assayed simultaneously. Each bar value represents the mean 7 s.d. of triplicate experiments.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

points and between the c-myb and c-myc genes by respective LC-antisense molecules demonstrate the target specificity.

We next tested LC-antisense molecules for growth inhibition ofcancer cell lines by targeting c-myb and cdk4. To accomplish this, twodifferent LC-antisense molecules to c-myb at 0.56 nM (c-myb-LCAS1containing 0.5 kb c-myb antisense and c-myb-LCAS2 containing 1.5 kbc-myb antisense) were added to K562. When the cancer cell linestransfected with the antisense molecules were examined for growthwith the MTT assay, both c-myb-LCAS1 and c-myb-LCAS2 were ableto inhibit cancer cell growth by more than 60% (Fig. 2d). In contrast,c-myb-LCSE and LCSS did not substantially affect K562 cell growth.Similarly, when cdk4-LCAS was added to MCF-7 cells, the cell growthof antisense transfectants was inhibited by more than 70% at antisenseconcentrations of either 0.28 or 0.56 nM. In contrast, cdk4-LCSEshowed only marginal inhibition of cell growth, less than 9% and 15%inhibition at the same concentrations (Fig. 2e). The changes of CDK4levels in MCF-7 cells treated with various amounts of cdk4-LCAS werealso examined with western blot analysis. Whereas cdk4-LCAS at aconcentration of 0.8 or 1.6 nM reduced the CDK4 level by morethan 60%, an equal amount of cdk4-LCSE had no substantial effect(see Supplementary Fig. 4 online). These results demonstrate thatLC-antisense molecules may provide target specificity and effectiveantisense activity.

En masse identification of liver cancer-related genes

The fact that a phagemid vector can be easily used to constructa cDNA library, prompted us to investigate the feasibility of

high-throughput functional genomics using LC-antisense technology.Using an LC-antisense library can be appealing because target sitesearches are not required. Thus, we constructed an LC-antisenselibrary to identify genes that are functionally involved in the growthof liver cancer cells (see Supplementary Fig. 5 online). To improve ourchances of finding genes of interest, we prepared mRNA from bothhepatoblastoma and noncancerous adjacent liver tissues, which wasdifferentially amplified for liver cancer–specific mRNA by a suppres-sion subtractive hybridization method37. Differentially amplifiedcDNAs were unidirectionally cloned into a phagemid vector, andthe cDNA library constructed was transformed into Escherichia colicompetent cells. From the cDNA library of 9,600 transformants, 1,200clones with cDNA inserts of more than 500 base pairs were selected bya simplified method for plasmid isolation38. LC-antisense moleculeswere then purified from the culture supernatant of bacterial compe-tent cells superinfected with helper bacteriophages. The random geneLC-antisense library of 1,200 member species was arrayed for trans-fection in 13 96-well plates that had been seeded with HepG2 cells forfunctional analysis. Each LC-antisense molecule (0.1 mg) was com-plexed with cationic lipids at a ratio of 1:3 (wt/wt) and transfected into7 � 103 HepG2 cells in each well of 96-well plates. Cells were inspectedfor morphological changes with light microscopy (Fig. 3a) andmeasured quantitatively for growth inhibition with an MTT assay4 d after transfection. Of the 1,200 antisense species selected by insertsizes, 153 (B13%) were found to be inhibitory to cancer cell growthin varying degrees. In contrast, cells treated with single-strandedcontrol DNA (devoid of antisense insert sequences) exhibited a mild

Table 1 List of genes involved in liver cancer cell growth

Gene description and putative functional category No. of clones Accession no.

Protein synthesis

Homo sapiens ribosomal protein S25 (RPS25) WGSL5 BC004986

Siboglinum ekmani 18S ribosomal RNA, partial sequence WGSL16 AF315062

H. sapiens ribosomal protein S8 (RPS8) WGSL19 NM_001012

H. sapiens ribosomal protein, large P1 WGSL23 NM_001003

H. sapiens ribosomal protein S24 (RPS24) WGSL33 NM_033022

H. sapiens ribosomal protein S17 (RPS17) WGSL38 M13932

H. sapiens clone IMAGE:3543815 WGSL39 BC020169

H. sapiens ribosomal protein L27 (RPL27) WGSL45 NM_000988

H. sapiens ribosomal protein S5 (RPS5) WGSL46 BC018151

H. sapiens ribosomal protein L35 (RPL35) WGSL55 NM_007209

Translation factors

H. sapiens eukaryotic translation initiation factor 3, subunit 6 interacting protein (EIF3S6IP) WGSL3 NM_016091

H. sapiens eukaryotic translation initiation factor 4A, isoform 2 (EIF4A2) WGSL28 NM_001967

H. sapiens eukaryotic translation elongation factor 1 gamma WGSL41 BC028179

Structural proteins and their regulators

H. sapiens tissue inhibitor of metalloproteinase 1 (TIMP1) WGSL6 XM_033878

H. sapiens clone MGC:5318 IMAGE:2900273 WGSL14 BC006781

H. sapiens beta-2-microglobulin (B2M) WGSL32 NM_004048

H. sapiens syntaxin 7 (STX7) WGSL53 XM_004526

Metabolism

H. sapiens glutamate dehydrogenase 1 (GLUD1) WGSL18 NM_005271

Human liver glutamate dehydrogenase WGSL25 J03248

H. sapiens similar to serine (or cysteine) proteinase inhibitor WGSL34 BC011991

Human mRNA for glutamate dehydrogenase WGSL42 X07769

Other

H. sapiens alpha-fetoprotein (AFP) WGSL8 BC027881

H. sapiens ferritin, light polypeptide (FTL) WGSL9 NM_000146

H. sapiens cutaneous T-cell lymphoma-associated tumor antigen se20-4 (SE20-4) WGSL11 BC024270

H. sapiens apolipoprotein A-II WGSL12 BC005282

H. sapiens heat shock 70 kDa protein 8 (HSPA8) WGSL13 NM_006597

Table 1 continued on following page


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

level of growth inhibition that could also be seen in cells treated witha dsDNA-lipid complex. To eliminate redundancy, we comparedthe sequences of cDNA clones complementary to the 153 growth-inhibiting LC-antisense molecules and searched the GenBank databasefor matching sequences. For instance, LCAS 3, 5, 8, 9, 10 and 23contained the antisense sequences reversely complementary to thecDNA sequences of WGSL3 (nucleotides 1367–1885 of GenBankaccession number NM_016091), WGSL5 (nt 21–489 of BC004986),WGSL8 (nt 1448–2017 of BC027881), WGSL9 (nt 421–863 ofNM_000146), WGSL10 (nt 2119–2446 of NM_024894) andWGSL23 (nt 321–512 of NM_001003) genes, respectively. Therewere 56 unique sequences out of 153 cDNAs, and these were sortedand designated as clones WGSL 1–56. Putative functional categoriza-tion of each gene was then performed by motif-based searches on thebasis of the revealed sequence information (Table 1). The LC-anti-sense molecules derived from the 56 clones were then designated asLCAS 1–56. Functional categorization indicated that 18 out of the 56genes encode proteins of undefined functions. The remaining 38 geneshave previously defined functions. The growth inhibiting activities ofthe 56 LC-antisense molecules were further confirmed by repetitiveMTT assays (Fig. 3b). These 56 genes appear to have functions directlyor indirectly related to cell growth of hepatoblastomas.

Effects of LC-antisense on cell cycle progression and apoptosis

The 56 LC-antisense species inhibitory to cancer cell growth werestudied further to reconfirm their roles and to understand theunderlying molecular mechanisms of their inhibitory effects. We

used flow cytometry analysis to detect changes in cell cycle patterns inHepG2 cells that were treated with the LC-antisense species for48 h. When compared to control treatments, using lipids alone orLCSS plus lipids, 53 (B95%) out of the 56 LC-antisense moleculesexhibited an increased percentage of cells with sub-G0-G1 DNAcontent (Fig. 4a). We then determined whether cell death causedby antisense treatment reflected the induction of apoptosis. HepG2cells treated with LC-antisense molecules were subjected to a DNAfragmentation assay. Twenty seven (LCAS 1, 3, 5, 8, 11, 12, 14, 15, 16,20, 21, 22, 24, 29, 31, 33, 34, 35, 40, 41, 42, 43, 47, 48, 49, 51, and 53)out of the 53 LC-antisense species were found to cause characteristicDNA ladder formation 48 h after the transfection, indicating apop-totic progression caused by the antisense molecules (Fig. 4b).These results suggest that the LC-antisense library system is aneffective means for en masse identification of genes involved in cancercell growth.

Functional validation of the identified genes using other antisense

A large number of genes were rapidly identified to be involved in livercancer cell growth with the LC-antisense library. Functional valida-tions of the genes identified from the liver cancer cells were furthercarried out by using other antisense technologies including siRNA andPS end-capped AS-oligos. We chose an LC-antisense of clone WGSL11 (accession number BC024270; gene description, H. sapiens cuta-neous T-cell lymphoma-associated tumor antigen), as an example, outof the seven LC-antisense molecules that showed differential inhibitionof HepG2 cell growth (data not shown).

Table 1 Continued

Gene description and putative functional category No. of clones Accession no.

H. sapiens haplotype M*2 mitochondrion WGSL21 AF382013

H. sapiens haptoglobin (HP) WGSL26 NM_005143

Human cytochrome P450IIE1 (ethanol-inducible) gene WGSL30 J02843

H. sapiens cytochrome b5 outer mitochondrial membrane precursor WGSL31 BC014431

Human DNA sequence from clone RP4-792D7 on chromosome 1q42.2-43. Contains the 5¢ end of the TARBP1

gene for TAR (HIV) RNA-binding protein 1

WGSL35 AL136124

H. sapiens interferon, gamma-inducible protein 30 (IFI30) WGSL36 XM_038146

H. sapiens fibrinogen, gamma polypeptide (FGG), transcript variant gamma-A WGSL40 NM_000509

H. sapiens hypothetical protein My014 (MY014) WGSL44 NM_030918

Human liver fatty acid binding protein (FABP) WGSL49 M10050

H. sapiens clone MGC:12445 IMAGE:3935036 WGSL50 BC005348

Human gene for heterogeneous nuclear ribonucleoprotein (hnRNP) core protein A1 WGSL51 X12671

H. sapiens FK506 binding protein 3 (25kD) (FKBP3) WGSL52 NM_002013

Undefined functions

Human chromosome 14 DNA sequence BAC R-123M6 of library RPCI-11 from chromosome 14 WGSL1 AL117190

Human DNA sequence from clone CTA-175E3 on chromosome 22q12.1 WGSL2 Z95113

Human DNA sequence from clone RP11-38P6 on chromosome 9 WGSL4 AL354874

H. sapiens BAC clone RP11-620E11 from chromosome 4 WGSL7 AC079926

H. sapiens hypothetical protein FLJ14075 (FLJ14075) WGSL10 NM_024894

H. sapiens PRO2675 mRNA WGSL15 AF119890

H. sapiens mRNA; cDNA DKFZp762B195 WGSL17 AL359585

H. sapiens clone RP11-56O18 from chromosome 2 WGSL20 AC019159

H. sapiens cDNA FLJ35730 fis, highly similar to alpha-1-antichymotrypsin precursor WGSL22 AK093049

H. sapiens BAC clone RP11-449G13 from chromosome 16 WGSL24 AC020716

H. sapiens chromosome 5 clone RP11-412P18 WGSL27 AC091952

H. sapiens clone IMAGE:3923943 WGSL29 BC024924

H. sapiens chromosome 4 clone B366O24 map 4q25 WGSL37 AC004067

H. sapiens BAC clone RP11-360H4 from chromosome 2 WGSL43 AC019086

Human DNA sequence from clone RP11-334A14 on chromosome 1 WGSL47 AL445183

H. sapiens genomic MHC class III complement gene cluster (MCGC@) on chromosome 6 WGSL48 NG_000013

H. sapiens BAC clone CTD-2324K8 from 7p14-p13 WGSL54 AC011230

H. sapiens genomic DNA, chromosome 11q clone:RP11-680L20 WGSL56 AP001102


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

We studied the effect of two siRNAs, 11-1siRNA and 11-2siRNA,to WGSL 11, on target mRNA level. 11-1siRNA, was found tobe effective in target RNA reduction, and was compared with LCAS11 for target RNA reduction and cell proliferation blockade. Quanti-tative downregulation of the target gene expression by LCAS-11and 11-1siRNA was done with real-time RT-PCR to detect targetmRNA levels in HepG2 cells. 24-h treatment with LC-antisense(1.6 nM) and siRNA (50 nM) resulted in about a 70% and 60%reduction of target mRNA, respectively, when compared to thatobtained with sham treatment (Fig. 5a). HepG2 cell growth wasinhibited by about 70% and 67% by 0.8 nM LCAS-11 and 25 nM11-1siRNA, respectively (Fig. 5b).

Next, we set out to reconfirm the validity of the functional data byusing other types of antisense molecules. To find an effective target sitefor antisense inhibition, a series of PS end-capped AS-oligos derivedfrom the clone WGSL11 were designed and evaluated for their abilityto inhibit target gene expression and HepG2 cell growth. Of fiveantisense sequences tested, the most active one, PS 11-1, was selected.When treated with PS 11-1, target mRNA levels were much reducedand cell growth was inhibited by about 70% at a concentration of 1.1mM (Fig. 5c,d). In contrast, cells treated with either mismatch or sense

control end-capped AS-oligos did not show much reduction of targetmRNA and resulted in only marginal cell growth inhibition. The resultwas reconfirmed by Southern hybridization of the RT-PCR band oftarget mRNA (Fig. 5c). These results validate the utility of LC-antisenselibrary in screening genes of interest in a high-throughput mode.

MCisplat

in

LCAS 2

9

LCAS 2

1

LCAS 1

6

LCAS 1

5

LCAS 1

4

LCAS 1

1

Contro

l DNA +

Li.

Sham

cont

rol

M

LCAS 31

22.1%

1,0008006004002000

040

8012

016

020

0

040

8012

016

020

0

040

8012

016

020

0C

ount

s

040

8012

016

020

0

Cou

nts

040

8012

016

020

0

040

8012

016

020

00

4080

120

160

200

040

8012

016

020

0

040

8012

016

020

0

Cou

nts

FL2-A1,0008006004002000

FL2-A

1,0008006004002000

FL2-A

1,00080060040020001,00080060040020001,0008006004002000

1,00080060040020001,00080060040020001,0008006004002000

LCAS 29

47.2%

LCAS 16

43.9%

23.8%

LCAS 11

62.3%

LCAS 14

56.9%

LCAS 15

11.0%

Control DNA + Li.

6.1%

Lipids

3.5%

Sham

030121.001 030121.003 030121.004

021122.021021115.036021115.033

021122.022 021115.037 021122.029

a

b

Lipids11-S11-M311-M211-M111-511-411-311-211-1

100

80

60

40

20

0

HepG2 cell growth

WGSL 11

β-actin

WGSL 11

Lipids

11-S

11-M

3

11-M

2

11-M

1

11-5

11-4

11-3

11-2

11-1

LCAS 11

Sham

11-1

siRNA

Lipids

Sham

NC siRNA

LCSS

LCAS 11

11-1

siRNA

Lipids

Sham

NC siRNA

LCSS

LCAS 11

1.2

1

0.8

0.6

0.4

0.2

0Nor

mal

ized

exp

ress

ion WGSL11 expression

1.2

1

0.8

0.6

0.4

0.2

0

HepG2 cell growth

Per

cent

of a

bsor

banc

e(5

70 n

m)

Per

cent

of a

bsor

banc

e(5

70 n

m)

a

c

d

b Figure 5 Validation of WGSL 11 as a potential target for the inhibition of

cancer cell growth. LCAS 11, siRNA and a series of PS end-capped AS-

oligos directed against WGSL 11 were designed and transfected into HepG2

cells. (a) HepG2 cells were treated with 1.6 nM LCAS 11 or 50 nM 11-

1siRNA or controls in a 24-well plate. Total RNA was isolated and subjected

to real-time RT-PCR 24 h after the antisense treatments. Values representthe average of three independent experiments. Open bars, treatment by

LCAS 11 or controls; solid bars, treatment by 11-1siRNA or controls.

(b) HepG2 cells were treated with 0.8 nM of LCAS 11, 25 nM of 11-

1siRNA or controls in a 96-well plate. The transfectants were examined for

growth inhibition using MTT assays 72 h after the antisense treatments.

Cells treated with sham, lipids alone and control molecules complexed with

lipids were also assayed for comparisons. Open bars, treatment by LCAS 11

or controls; solid bars, treatment by 11-1siRNA or controls. (c) HepG2 cells

were transfected with 0.8 nM LCAS 11 and 1.1 mM the PS end-capped

AS-oligos, complexed with Lipofectin in a 48-well plate. Total RNA was

subjected to RT-PCR 48 h after the AS-oligo treatments. DNA bands in b

were then transferred onto a nylon membrane and subjected to Southern

hybridization. (d) HepG2 cells were transfected with each PS end-capped

AS-oligos (0.16 mM) complexed with Lipofectin at a ratio of 1:2.5 (wt/wt) in

a 96-well plate. MTT assays were carried out to determine the inhibition of

cell growth 72 h after AS-oligos treatment. Each bar value represents the

mean 7 s.d. of triplicate experiments. Sham, sham treated; 11-1–5, treated

with PS end-capped AS-oligos of five different sequences; 11-M1–M3,treated with mismatch control oligos of three different sequences to

PS 11-1; and 11-S, treated with a sense control oligo of PS 11-1.

Figure 4 Effects of LC-antisense on cell cycle progression and apoptotic

induction. A representative example of the data acquired from treatments

with six kinds of LC-antisense is shown. (a) Cell cycle analysis after

transfection of cells with LC-antisense molecules. HepG2 cells and controls

were treated with the LC-antisense molecules, LCAS 11, LCAS 14, LCAS

15, LCAS 16, LCAS 29 and LCAS 31: sham treatment, lipids alone and

control DNA + Li. (lipid) complexes. Cells were harvested at 48 h after

transfection. Functional analysis was performed on an equal number of cells(104 events) by flow cytometry after staining of DNA with propidium iodide.

(b) Induction of apoptotic DNA ladder formation by LC-antisense molecules.

The LC-antisense molecules, LCAS 11, LCAS 14, LCAS 15, LCAS 16, LCAS

21 and LCAS 29, were treated to HepG2 cells along with controls: sham

treatment, control DNA + Li. (lipid) complexes and cisplatin (positive

control). Genomic DNA was extracted 48 h after transfection and run on a

1.6% agarose gel. M, 100 bp DNA ladder size marker.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

DISCUSSION

LC antisense molecules were generated as single-stranded genomicDNA of recombinant bacteriophages, and was tested for stability,antisense activity and, further, for its usefulness in high-throughputfunctional genomics. When LC-antisense molecules to TNF-a mRNAwere used, the antisense molecules were found to be stable in thepresence of nucleases and effective in reduction of target mRNA,which was reflected in the antisense activity that required less than1/10 the amount of most other types of AS-oligos. LC-antisense toTNF-a was also found to substantially reduce production of rat TNF-a in cells, confirming commensurate antisense activity at the proteinlevel. Further, the broad utility of LC-antisense was confirmed withantisense-mediated expression blockade of several other genes (c-myc,c-myb, cdk2 and cdk4) of biological significance.

The enhanced antisense activity of LC-antisense molecules may beexplained in two ways. One reason may be that the long antisensesequence (1,000 bases on average) in the molecules allows theformation of a more stable duplex between the antisense sequenceand the complementary sequence of target mRNA. The lengthy duplexmay serve as a substrate for RNaseH activity for an extended period oftime. Another reason is that mRNA tends to form extensive secondaryand tertiary structures among its own sequences or interact with RNA-binding proteins in the cell cytoplasm, which can make some targetsequences inaccessible. It is more likely, with its long length, thatcertain regions within LC-antisense molecules have a higher chance ofbinding to complementary sequences in target mRNA.

As the antisense sequences are much longer in LC-antisensemolecules, target-specificity of LC-antisense is of critical concern. Toprove sequence specificity of LC-antisense in a rigorous manner, wecarried out both multi-probe RPA and real time RT-PCR. The specificantisense activity was shown by the lack of off-target effects betweenLC-antisense molecules to c-myb and c-myc and was reconfirmed byLC-antisense molecules and siRNA targeting the same gene, cdk2.Because cdk4 expression was downregulated only by cdk4 LCAS butnot by cdk2 LCAS, and vice versa, also indicates sequence specificitybecause the two genes have a region of localized homology. Thedelayed downregulation by functionally associated genes may beexplained by the tightly coordinated regulation between cell cycleregulatory proteins. Perturbed expression of a growth regulatory genehas been reported to alter expression of other genes involved in theG1/S transition phase of cell cycle progression36. Primary reduction ofcdk2 expression may have subsequently lowered the activity of CDK4(ref. 35). Even with its long length, LC-antisense provides sequencespecificity comparable or better to existing antisense technologies. Aswith other antisense technologies including siRNA, there may still besome off-target effects when a large amount of LC-antisense moleculesis used.

Knockdown conferred by antisense provides much faster means forgene functionalization than do the conventional knockout methods.To take advantage of antisense technologies, antisense libraries havebeen constructed using AS-oligos and used for selection of drugtargets39,40. A vector system for expression of ssDNA in mammaliancells was also reported41. More recently, an approach using a genome-wide synthetic siRNA or siRNA expression library was developed forunveiling gene functions42. Although the technology of siRNA hasbeen reported to be effective, the efficiency in target reduction andspecificity appears to be comparable to those of LC-antisense. It shouldbe noted that both AS-oligos and siRNA, unlike LC-antisense, requiretarget site searches that are time consuming and often inconclusive.LC-antisense, as with other antisense, appears to bring about some-what varying degrees of target reduction even with its lower variability

and better efficiency. Thus, antisense activities obtained from distinctLC-antisense molecules need to be analyzed with some prudence.

By using the random gene LC-antisense library, we identified 56genes functionally involved in liver cancer cell growth. Motif-basedsearches suggested that these include genes with novel functions andgenes with defined functions, some of which were, as expected,involved in critical cellular metabolisms in DNA replication, transcrip-tion and translation. Yet, some others contribute to cancer cell growthin addition to their functions that appear irrelevant to cancer cellgrowth. Several genes that we found to be involved in liver cancer cellgrowth in the present study were shown to be overexpressed inhepatocellular carcinoma43,44. In fact, ribosomal proteins P1, S17,L35, fibrinogen gamma polypeptide and elongation factor-1 gamma(WGSL 23, 38, 55, 40 and 41 respectively) were overexpressed in livercancer and involved in protein synthesis. These results suggest thatthese genes, although essential in their housekeeping roles, may havedifferential expression levels in liver cancer tissues and support cellgrowth in liver cancer. Similar findings of a large number of house-keeping genes in the identification of growth-related genes were alsoreported in other functional genomics using expressed antisense inCandida albicans45. If overexpressed above a normal level, these genesmay play an important role in cancer or pathogenic cell growth.Recently, for example, a-fetoprotein (WGSL 8) was reported tostimulate expression of some oncogenes (c-fos, c-jun and N-ras) inliver cancer cells46. It would then be worth targeting these genes tocurb cancer cell growth. Further studies are clearly warranted toinvestigate biochemical processes of protein products of the genes.

An LC-antisense library can be constructed with the unidirectionalcloning of cDNA fragments of known sequences (unigenes) intophagemid vectors. Contrary to the random gene antisense library,each antisense species in the unigene antisense library has a uniquesequence. An advantage of the unigene antisense library is that a largepanel of human genes can be individually targeted without redun-dancy, and genes of constant transcription levels with post-translational modifications may be screened.

The LC-antisense library system may provide a faster, more costeffective and analytically accurate tool for the study of functionalgenomics. It would be interesting to see if the LC-antisense moleculescan also be used in animals as this would potentially provide a usefulapproach for in vivo functional genomics.

METHODSConstruction of recombinant phagemids. Various recombinant phagemids

were constructed according to standard cloning procedure47. WRT7/P2 cells

(1 � 105) were seeded in each well of a 48-well plate. Rat TNF-a expression was

induced in the cells by the treatment of LPS (Sigma-Aldrich) at 30 mg/ml for

4–24 h. Cells were harvested at desired time points to examine the level of

mRNA. The LPS incubation time that induced the highest expression level

of TNF-a was chosen for further experiments. The RT-PCR fragment (708 bp)

of TNF-a that comprises the entire coding sequence was amplified with a pair

of PCR primers (5¢-GATCGTCGACGATGAGCACAGAAAGCATGATCC-3¢and 5¢-GATCGAATTCGTCACAGAGCAATGACTCCAAAG-3¢) and sequence

verified. To construct TNFa-LCAS, the rat TNF-a cDNA fragment was cloned

into the multiple cloning site of the pBluescript (pBS) KS(�) vector (Strata-

gene) using SalI and EcoRI restriction sites in the same direction as the lacZ

gene (see Supplementary Fig. 1 online). Control sense molecules were

constructed similarly. Likewise, cDNA fragments of the c-myc, c-myb, cdk2

and cdk4 genes were amplified with a pair of PCR primers (see Supplementary

Table 1 online) and cloned into the EcoRV site of pBS-KS (+) or (�) vector.

The recombinant phagemids were transformed into Epicurian Coli XL-1 Blue

competent cells (Stratagene) by the calcium-chloride method. Cloning direc-

tion of amplified cDNA fragments were confirmed with both restriction

digestion and DNA sequencing.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Production and purification of either LC-antisense or control molecules. LC-

antisense or control molecules to target genes were produced by overnight

culture of transformed bacterial cells that had previously been infected with

helper bacteriophages, and purified by gel filtration column chromatography.

These methods are described in detail in Supplementary Methods online.

Structural analysis and stability test of LC-antisense molecules. For structur-

al analysis, 1 mg of TNFa-LCAS was treated with XhoI (10 U/mg DNA),

exonuclease III (160 U/mg DNA), or S1 nuclease (10 U/mg DNA) at 37 1C for

3 h, and subjected to phenol extraction, ethanol precipitation and gel electro-

phoresis on a 1% agarose gel. For stability test, 1 mg of the antisense molecules

was tested alone or after complex formation with lipids at a ratio of 1:3 (wt/wt)

of DNA/lipids. We added 30% FBS that was not heat inactivated to the

antisense-lipids complex and incubated it at 37 1C for varying periods of time

for up to 48 h. After incubation with FBS and exonucleases, LC-antisense was

extracted with phenol, precipitated with ethanol and run on a 1% agarose gel.

Transfection of LC-antisense and siRNA. Transfection of LC-antisense or

siRNA was carried out to study the activity of the LC-antisense molecules. For

LC-antisense transfection, cells were seeded on a 6-well (for RPA assay), 24-well

(for real-time PCR) or 48-well plate (for RT-PCR) in an appropriate volume of

culture medium. Cationic lipids, Lipofectamine, Lipofectamine 2000 or Lipo-

fectamine plus reagents (Invitrogen) were mixed with the purified molecules in

various ratios (wt/wt) for transfection into target cells. These lipid-DNA

complexes were mixed with Opti-MEM (Invitrogen) and added to cells

according to the manufacturer’s protocol. After 6 h transfection at 37 1C, the

cells were added with fresh medium and incubated further for up to 48 h at

37 1C before assays. Expression of the Rat TNF-a gene was induced with LPS

treatment (30 mg/ml) to WRTP7/P2 cell transfectants. To compare the effects of

the LC-antisense molecules, identical quantities of lipids alone and control

DNA plus lipid complexes were also added to the same number of cells in a

different well plate and assayed simultaneously. For siRNA transfection, cells

were seeded on a 24-well (for real-time PCR), or 96-well plate (for MTT

reduction assay) in an appropriate volume of culture medium. Cdk2siRNA

(sense sequence, 5¢-GGUACCGAGCUCC UGAAAUCTT-3¢; antisense, 5¢-GAU

UUCAGGAGCUCGGUACCTT-3¢), 11-1siRNA (sense sequence, 5¢-GCAGG

CACUGGAGGAUAUUCTT-3¢; antisense, 5¢-GAAUAUCCUCCAGUGCCUG

CTT-3¢), and 11-2siRNA (sense sequence, 5¢-GAAGCAAGAAAUGAAGAAAC

TT-3¢; antisense, 5¢-GUUUCUUCAUUUCUUGCUUCTT-3¢) duplexes were

synthesized (Bioneer) and transfected into HeLa or HepG2 cells using siPORT

Lipid (Ambion) as recommended by the manufacturer. To compare the effects

of the siRNA molecules, identical quantities of lipids alone and negative control

no. 1 siRNA (Ambion) plus lipid complexes were also added to the same

number of cells in a different well plate and assayed simultaneously.

Detection of target gene transcript. After the transfection of LC-antisense or

siRNA, the change of target gene expression in mRNA level was detected with

RT-PCR, RPA, and real-time quantitative RT-PCR methods. RNA preparation

was carried out with Tri reagent (Molecular Research Center) according to the

protocol recommended by the manufacturer. Purified RNA was subjected to

RT-PCR in a 50-ml reaction volume by using the Access RT-PCR kit (Promega)

and a thermal cycler (MJ Research) as recommended by the manufacturer. A

pair of primers was used to amplify TNF-a, c-myc, c-myb and WGSL 11 genes

(see Supplementary Table 2 online). PCR product was confirmed on a 1%

agarose gel, and quantitative analysis of the amplified DNA was performed with

AlphaImager 1220, a gel documentation apparatus (Alpha Innotech).

To investigate the effect of LC-antisense molecule on the steady-state level of

cdk2 mRNA, RPA was carried out according to the instruction of the

RiboQuant Multi-Probe RPA System (BD Pharmingen). Total RNA was

obtained from the transfectants at 24 h and 48 h after antisense or control

treatment. The antisense RNA probes were synthesized from an hCC-1

template set (BD Pharmingen) in the presence of biotin-16-UTP (Roche).

The biotin-labeled probes were hybridized in excess to 20 mg total RNA in

solution. Unprotected probes and RNA were digested by RNases. The RNA/

probe hybrids were run on a denaturing polyacrylamide gel and then

transferred onto a nylon membrane by a semi-dry blotting unit (Fisher

Scientific). Immobilized hybrids were cross-linked to the membrane by

exposing to UV light. The membrane was incubated with a streptavidin-

horseradish peroxidase reagent before exposure to an X-ray film.

Target gene expression was also measured by real-time quantitative RT-PCR.

Total RNA (1 mg) was reverse transcribed by using random primers supplied in

the Reverse Transcription System (Promega). To quantify gene expression,

cDNA of cdk2, cdk4, cdk6 and WGSL 11 genes were amplified by using

respective pair of primers (see Supplementary Table 3 online), the DyNAmo

HS SYBR Green qPCR Kit (MJ Research), and the DNA Engine Opticon 2

System (MJ Research) according to the manufacturer’s instruction. To normal-

ize the amount of total RNA present in each reaction, b-actin gene was

amplified simultaneously. Triplicate assays were done with RNA samples

isolated from at least two independent experiments.

Detection of polypeptides with ELISA or western blotting. Quantification of

each target protein after TNFa-LCAS, cdk2-LCAS, and cdk4-LCAS treatment

was performed with the enzyme-linked immunosorbent assay (ELISA) or

western blotting analysis (see Supplementary Methods online).

MTT assay to determine inhibition of cell growth. Both LC-antisense

molecules (c-myb and cdk4 LCAS) and siRNA to WGSL11 were studied for

their growth inhibitory effects using the MTT assay. The MTT reagent (3-[4,5-

dimethylthiazol-2-yl]-2,5-diphenyl-tetrazolium bromide) (Sigma-Aldrich)

was diluted with PBS to a concentration of 5 mg/ml, and 100 mg of the diluent

was added to each well containing 100 ml culture medium. Cells were main-

tained in a CO2 incubator at 37 1C for 4 h and treated with an equal amount of

isopropanol (containing 0.1N HCl) at 25 1C for 1 h. The cells were then mea-

sured for absorbance at 570 nm with an ELISA reader, SpectraMAX 190

(Molecular Devices).

Construction of a unidirectional subtracted liver cDNA library. To clone

differentially expressed genes in hepatoblastomas, a cDNA library was con-

structed by using a subtractive hybridization procedure37. The construction of

the library is described in detail in Supplementary Methods online.

Preparation of a liver cancer LC-antisense library. Bacterial competent cells

containing recombinant pBS SK(�) phagemids were plated on Luria-Bertani

agar plates containing 50 mg/ml of ampicillin and 50 mg/ml of tetracycline and

incubated at 37 1C for 16 h. Isolated colonies were seeded in a well of 96-deep-

well plates containing 1.4 ml of 2�YT liquid medium (tryptone 16 g, yeast

extract 10 g, NaCl 10 g per 1,000 ml) added with 50 mg/ml ampicillin. Cells

were cultured for 7 h at 37 1C with vigorous shaking. To produce LC-antisense

molecules from each phagemid, 20 ml of the bacterial culture was transferred to

each well prefilled with 1.4 ml of fresh 2�YT liquid medium containing 9 ml of

helper bacteriophages, M13K07 (New England Biolabs). After 1 h incubation,

4.2 ml of 70 mg/ml kanamycin was added and cultured at 37 1C for 12 h. The

superinfection was carried out in triplicate for each clone to maximize the yield

of antisense molecules in a single purification step. Single-stranded LC-

antisense molecules were purified from the culture supernatant of bacterial

cells using QIAprep 96 M13 Kits and QIAVAC vacuum manifolds (Qiagen)

according to manufacturer’s instructions. To test both quantity and purity, we

ran purified LC-antisense molecules on a 1% agarose gel along with control LC-

molecules derived from pBS SK(�) phagemid without a cDNA insert.

Transfection of an LC-antisense library into a liver cancer cell line. To

identify genes involved in the growth of liver cancer cells, Lipofectamine 2000

was mixed with antisense molecules of the liver cancer–specific LC-antisense

library for transfection into HepG2. The cells (7 � 103) were washed twice with

Opti-MEM, seeded in each well of 96-well plates in 100 ml of Opti-MEM

supplemented with 10% FBS and incubated for 12–18 h at 37 1C in a 5% CO2

incubator. LC-antisense molecules (0.1 mg) were complexed with 0.3 mg of the

cationic lipids, and the antisense molecule plus lipid complexes were added to

the cultured cells. The cultures were exchanged with fresh medium 24 h after

transfection and incubated for 4 d further. To compare the effects of the LC-

antisense molecules on cell proliferation, we also added equal quantities of

lipids alone and control DNA plus lipid complexes to the same number of cells

in a different 96-well plate and assayed them simultaneously. Control DNA was

single-stranded phage genomic DNA lacking a cDNA insert. After the transfec-

tion, microscopic observation or MTT reduction assay was performed to study


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

the effect of antisense molecules on proliferation of cancer cells as described

above. The percentage of growth inhibition of cells in each well treated with

antisense plus lipids complex was calculated by comparing the optical density

with those of sham treatments, using the following formula: 1 � (absorbance of

an experimental well/absorbance of a sham control well) � 100.

Gene identification and sequence motif search. To identify genes comple-

mentary to LC-antisense molecules that inhibited proliferation of liver cancer

cells, we sequenced recombinant phagemids obtained by alkaline lysis from the

5¢ upstream of the (+) strand of cDNA inserts using the T3 primer. Sequences

of cDNA inserts were compared with those of the GenBank database. Poly-

peptides deduced from cDNA sequences were then searched for amino acid

motifs using the ProfileScan Server (http://hits.isb-sib.ch/cgi-bin/PFSCAN).

Treatment of PS end-capped AS-oligos. To reconfirm the functional role of

WGSL11 gene in the cell proliferation of liver cancer, we designed a series of PS

end-capped AS-oligos (see Supplementary Table 4 online) and transfected

HepG2 cells. The procedures are described in detail in Supplementary

Methods online.


ACKNOWLEDGMENTSThis study was supported by generous grants of the CDRC of Korean Science &Engineering Foundation (research grant no. R01-2000-00138, R13-2002-028-01004-0), South Korea, and WelGENE Inc., a biotechnology company foundedby Jong-Gu Park.


Received 15 December 2004; accepted 14 March 2005


1. Dolnick, B.J. Antisense agents in cancer research and therapeutics. Cancer Invest. 9,185–194 (1991).

2. Bennett, C.F. & Cowsert, L.M. Antisense oligonucleotides as a tool for gene functiona-lization and target validation. Biochim. Biophys. Acta 1489, 19–30 (1999).

3. Thompson, C.B., Challoner, P.B., Neiman, P.E. & Groudine, M. Levels ofc-myc oncogene mRNA are invariant throughout the cell cycle. Nature 314, 363–366(1985).

4. Melani, C., Rivoltini, L., Parmiani, G., Calabretta, B. & Colombo, M.P. Inhibition ofproliferation by c-myb antisense oligodeoxynucleotides in colon adenocarcinoma celllines that express c-myb. Cancer Res. 51, 2897–2901 (1991).

5. Anfossi, G., Gewirtz, A.M. & Calabretta, B. An oligomer complementary to c-myb-encoded mRNA inhibits proliferation of human myeloid leukemia cell lines. Proc. Natl.Acad. Sci. USA 86, 3379–3383 (1989).

6. Kamano, H. et al. Effects of the antisense v-myb expression on K562 human leukemiacell proliferation and differentiation. Leuk. Res. 14, 831–839 (1990).

7. Melotti, P. & Calabretta, B. Induction of hematopoietic commitment and erythromyeloiddifferentiation in embryonal stem cells constitutively expressing c-myb. Blood 87,2221–2234 (1996).

8. Ferrari, S. et al. Differential effects of c-myb and c-fos antisense oligodeoxynucleotideson granulocytic differentiation of human myeloid leukemia HL60 cells. Cell GrowthDiffer. 1, 543–548 (1990).

9. Thaler, D.S., Liu, S. & Tombline, G. Extending the chemistry that supports geneticinformation transfer in vivo: phosphorothioate DNA, phosphorothioate RNA, 2¢-O-methyl RNA, and methylphosphonate DNA. Proc. Natl. Acad. Sci. USA 93, 1352–1356 (1996).

10. Graham, M.J. et al. In vivo distribution and metabolism of a phosphorothioateoligonucleotide within rat liver after intravenous administration. J. Pharmacol. Exp.Ther. 286, 447–458 (1998).

11. Zhang, H. et al. Reduction of liver Fas expression by an antisense oligonucleotideprotects mice from fulminant hepatitis. Nat. Biotechnol. 18, 862–867 (2000).

12. Roush, W. Antisense aims for a renaissance. Science 276, 1192–1193 (1997).13. Tomita, N. et al. Transient decrease in high blood pressure by in vivo transfer of

antisense oligodeoxynucleotides against rat angiotensinogen. Hypertension 26, 131–136 (1995).

14. Yacyshyn, B.R. et al. A placebo-controlled trial of ICAM-1 antisense oligonucleotide inthe treatment of Crohn’s disease. Gastroenterology 114, 1133–1142 (1998).

15. Moon, I.J. et al. Target site search and effective inhibition of leukaemic cell growth by acovalently multiple antisense oligonucleotide to c-myb. Biochem. J. 346, 295–303(2000).

16. Moon, I.J. et al. Potent growth inhibition of leukemic cells by novel ribbon-typeantisense oligonucleotides to c-myb. J. Biol. Chem. 275, 4647–4653 (2000).

17. Hieter, P. & Boguski, M. Functional genomics: it’s all how you read It. Science 278,601–602 (1997).

18. Lander, E.S. Array of hope. Nat. Genet. 21, 3–4 (1999).19. Pandey, A. & Mann, M. Proteomics to study genes and genomes. Nature 405, 837–

846 (2000).20. Quackenbush, J. et al. The TIGR Gene Indices: analysis of gene transcript sequences in

highly sampled eukaryotic species. Nucleic Acids Res. 29, 159–164 (2001).21. Velculescu, V.E., Zhang, L., Vogelstein, B. & Kinzler, K.W. Serial analysis of gene

expression. Science 270, 484–487 (1995).22. Oien, K.A., Vass, J.K., Downie, I., Fullarton, G. & Keith, W.N. Profiling, comparison and

validation of gene expression in gastric carcinoma and normal stomach. Oncogene 22,4287–4300 (2003).

23. Prashar, Y. & Weissman, S.M. READS: a method for display of 3¢-end fragmentsof restriction enzyme-digested cDNAs for analysis of differential gene expression.Methods Enzymol. 303, 258–272 (1999).

24. Gottschalk, A.R. et al. p27Kip1 is required for PTEN-induced G1 growth arrest. CancerRes. 61, 2105–2111 (2001).

25. Zeng, J.Z., Wang, H.Y., Chen, Z.J., Ullrich, A. & Wu, M.C. Molecular cloning andcharacterization of a novel gene which is highly expressed in hepatocellular carcinoma.Oncogene 21, 4932–4943 (2002).

26. Ahn, J.D. et al. Inhibitory effects of novel AP-1 decoy oligodeoxynucleotides on vascularsmooth muscle cell proliferation in vitro and neointimal formation in vivo. Circ. Res.90, 1325–1332 (2002).

27. Suyama, E., Kawasaki, H., Nakajima, M. & Taira, K. Identification of genes involved incell invasion by using a library of randomized hybrid ribozymes. Proc. Natl. Acad. Sci.USA 100, 5616–5621 (2003).

28. Caplen, N.J. RNAi as a gene therapy approach. Expert Opin. Biol. Ther. 3, 575–586(2003).

29. Umans, L. et al. Targeted inactivation of the mouse alpha 2-macroglobulin gene.J. Biol. Chem. 270, 19778–19785 (1995).

30. Zimmermann, K. et al. Targeted disruption of the GAS41 gene encoding a putativetranscription factor indicates that GAS41 is essential for cell viability. J. Biol. Chem.277, 18626–18631 (2002).

31. Matsuda, M. et al. Abrogation of the Fc gamma receptor IIA-mediated phagocytic signalby stem-loop Syk antisense oligonucleotides. Mol. Biol. Cell 7, 1095–1106 (1996).

32. Vickers, T.A., Wyatt, J.R. & Freier, S.M. Effects of RNA secondary structure on cellularantisense activity. Nucleic Acids Res. 28, 1340–1347 (2000).

33. Smith, L., Andersen, K.B., Hovgaard, L. & Jaroszewski, J.W. Rational selection ofantisense oligonucleotide sequences. Eur. J. Pharm. Sci. 11, 191–198 (2000).

34. Jupin, I. & Gronenborn, B. Abundant, easy and reproducible production of single-stranded DNA from phagemids using helper phage-infected competent cells. NucleicAcids Res. 23, 535–536 (1995).

35. Mulligan, J.M., Campiani, G., Ramunno, A., Nacci, V. & Zisterer, D.M. Inhibition of G1cyclin-dependent kinase activity during growth arrest of human astrocytoma cells bythe pyrrolo-1,5-benzoxazepine, PBOX-21. Biochim. Biophys. Acta 1639, 43–52(2003).

36. Yu, B., Lane, M.E., Pestell, R.G., Albanese, C. & Wadler, S. Down regulation of cyclinD1 alters cdk4- and cdk2-specific phosphorylation of retinoblastoma protein. Mol. CellBiol. Res. Commun. 3, 352–359 (2000).

37. Diatchenko, L. et al. Suppression subtractive hybridization: a method for generatingdifferentially regulated or tissue-specific cDNA probes and libraries. Proc. Natl. Acad.Sci. USA 93, 6025–6030 (1996).

38. Beuken, E., Vink, C. & Bruggeman, C.A. One-step procedure for screening recombinantplasmids by size. Biotechniques 24, 748–750 (1998).

39. Yonekura, H. et al. Antisense display, a method for functional gene screening:evaluation in a cell-free system and isolation of angiogenesis-related genes. NucleicAcids Res. 27, 2591–2600 (1999).

40. Marcusson, E.G., Vincent, T.M., Hari, K.L., Chiang, M.Y. & Dean, N.M. Case study: useof a library of antisense inhibitors for gene functionalization and drug target validation.Drug Discov. Today: Targets 3, 117–122 (2004).

41. Chen, Y., Ji, Y.-J. & Conrad, C. Expression of ssDNA in mammalian cells. Biotechniques34, 167–171 (2003).

42. Miyagishi, M. & Taira, K. Strategies for generation of a siRNA expression librarydirected against the human genome. Oligonucleotides 13, 325–333 (2003).

43. Shuda, M. et al. Enhanced expression of translation factor mRNAs in hepatocellularcarcinoma. Anticancer Res. 20, 2489–2494 (2000).

44. Fan, B.L. et al. Cloning and identification of fibrinogen gamma polypeptide (FGG) genedifferentially expressed in human hepatocellular carcinoma. Ai Zheng 23, 249–253(2004).

45. De Backer, M.D. et al. An antisense-based functional genomics approach for identifica-tion of genes critical for growth of Candida albicans. Nat. Biotechnol. 19, 235–241(2001).

46. Li, M.S., Li, P.F., Chen, Q., Du, G.G. & Li, G. Alpha-fetoprotein stimulated theexpression of some oncogenes in human hepatocellular carcinoma Bel 7402 cells.World J. Gastroenterol. 10, 819–824 (2004).

47. Sambrook, J. & Russell, D.W. Plasmids and their usefulness in molecular cloning. inMolecular cloning: A laboratory manual 3rd edition (Cold Spring Harbor LaboratoryPress, Cold Spring Harbor, New York, 2001).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Migration and differentiation of neural precursorsderived from human embryonic stem cells in therat brainViviane Tabar1,2, Georgia Panagiotakos1,2, Edward D Greenberg1,2, Bill K Chan1,2, Michel Sadelain3,Philip H Gutin2 & Lorenz Studer1,2

Human embryonic stem (hES) cells provide a potentially

unlimited cell source for regenerative medicine. Recently,

differentiation strategies were developed to direct hES cells

towards neural fates in vitro. However, the interaction

of hES cell progeny with the adult brain environment

remains unexplored. Here we report that hES cell–derived

neural precursors differentiate into neurons, astrocytes

and oligodendrocytes in the normal and lesioned brain

of young adult rats and migrate extensively along white

matter tracts. The differentiation and migration behavior of

hES cell progeny was region specific. The hES cell–derived

neural precursors integrated into the endogenous precursor

pool in the subventricular zone, a site of persistent

neurogenesis. Like adult neural stem cells, hES cell–derived

precursors traveled along the rostral migratory stream to the

olfactory bulb, where they contributed to neurogenesis. We

found no evidence of cell fusion, suggesting that hES cell

progeny are capable of responding appropriately to host cues

in the subventricular zone.

Directing neural differentiation of hES cells in vitro has been ofparticular interest in view of the success of mouse ES cells inpreclinical models of disease1–5. However, the capacity of hES cell–derived precursors to integrate into the adult brain and respond toenvironmental cues in this typically nonpermissive milieu remainsunknown, whereas the behavior of human fetal neural stem cellsin vivo is well described6–9.

We derived stable EGFP-expressing hES cells from lines H1(WA01)10 and HES3 (ES03)11 after transduction with the CPGphosphoglycerate kinase–enhanced green fluorescent protein self-inactivating (mPGK-EGFP SIN)-lentiviral vector. The CPG lentiviralvector expressing enhanced green fluorescent protein (EGFP) undercontrol of the PGK promoter was derived from a multiply attenuatedHIV vector system12 and included a U3 deletion and introduction of acPPT element. Vectors were produced by triple transfection of humanembryonic kidney 293 cells followed by ultracentrifugation and titra-tion as previously described13. Undifferentiated hES cells growing on

mouse embryonic fibroblasts (MEF) were exposed to the virus at a titerof 0.5 � 108 transforming units/ml for 8 h followed by a 16-h recoveryperiod for 3 consecutive days. EGFP was detected by native fluores-cence at day 3 after transduction. Single EGFP-expressing hES cellcolonies were transferred and replated repeatedly until uniform EGFPexpression was observed among all cells within colonies. No loss inEGFP expression was observed during propagation or differentiationfor up to 15 months after transduction (Fig. 1a–c). The absenceof single cell derivation or selection suggests a polyclonal origin ofthe EGFP+ cell population. The phenotypic characteristics anddifferentiation profile of wild-type and EGFP-transduced hES cellswere indistinguishable.

Neural differentiation was induced under serum-free conditions bycoculture on a stromal cell line (MS5)3,14. At 1 month neuralprecursors were isolated, replated feeder-free and maintained for 4weeks in N2/basic fibroblast growth factor/epidermal growth factor(Fig. 1d). At this stage the majority of cells (490%) were immuno-reactive for neural precursor markers (Nestin, Musashi, A2B5) andnegative for undifferentiated ES cell markers such as Oct-4, SSEA-3and SSEA-4. In addition to immature neural precursors (Fig. 1e–f),8% of the cells expressed markers of immature neurons (Fig. 1g) and2% were immunoreactive for glial fibrillary acidic protein (GFAP)suggesting astrocytic identity.

The in vivo behavior of hES cell–derived (WA01-GFP) neuralprecursors was first monitored in the normal brain of 3-month-oldrats. We transplanted 105 EGFP+ neural precursors or equivalent deadcell controls into the striatum (n ¼ 8). The animals received three dailyinjections of (+)-5-bromo-2¢-deoxyuridine (BrdU) (300 mg/kg/dayfor 3 consecutive days) immediately before being killed at 11 weeksafter grafting. To assess whether modulation of host-derived cues isrequired for cell migration, differentiation and integration, we createdlysolecithin lesions in the cingulum of a second group of animals(n ¼ 8). These lesions result in a focal demyelination that elicits aconsistent repair response leading to complete spontaneous remyelina-tion in 6–8 weeks15. Five days after the lesions were made, the animalsreceived the same transplants as the unlesioned group, includingdead cell controls, and were subjected to an identical BrdU regimen.


1Developmental Biology, 2Neurosurgery and 3Medicine, Memorial Sloan Kettering Cancer Center, 1275 York Ave, New York, New York 10021, USA. Correspondenceshould be addressed to V.T. ([email protected]).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

EGFP+ human nuclear antigen (hNA)+ human cells were detectedin all animals grafted with hES cell–derived neural precursors, whereasall dead cell control grafts were devoid of any EGFP+ or hNA+

labeling. Colabeling of human cells with hNA and EGFP was con-firmed by double immunohistochemistry (Supplementary Fig. 1online). Grafts consisted of a core and large numbers of migratingcells (Fig. 2). Most cells in the core were neuronal precursors (TuJ1+,58.4% 7 4.7) and postmitotic neurons (NeuN+, 28.3% 7 2.3,Fig. 2a). The volumes of the graft core ranged from 0.31 to 2.09mm3 (Cavalieri estimator). Stereological counts (fractionator16,17)yielded an average of 325,410 human cells within the graft core(range 95,664–703,296). Migrating human cells followed white mattertracts and were distributed both ipsi- and contralateral to the injectionsite within corpus callosum, fornix, fimbria, striatum and cortex.Stereological counts of the total number of migrating cells amountedto an average of 100,172 cells per animal (range 57,375–182,250).Within the striatum, hES cell–derived cells were located largely(79.6%) in DARPP32-/MBP+ zones. Most human cells in the corpuscallosum expressed nestin (92.6% 7 1.4, Fig. 2b). These migratingcells exhibited a fivefold increase in BrdU uptake compared with cellswithin the graft core (13.4% 7 1.1 versus 2.6 7 0.5%, P o 0.001).Tunel labeling demonstrated a very low level of cell death 11 weeksafter transplantation (o0.1% of all human cells per brain). GFAP+

human cells were rare and mostly visualized in the periphery of thecore graft. Grafted cells were also colabeled for oligodendroglialmarkers such as NG2 (21.7% 7 1.4), cyclic nucleotide phosphodies-terase (CNP), myelin/oligodendrocyte–specific protein (MOSP), O1,O4 and myelin binding protein (o5%) (Supplementary Figs. 2 and 3online). All animals with lysolecithin lesions (n ¼ 8) exhibited fullrepair by the time they were killed regardless of the graft status (dead

versus alive). This was assessed by myelin histochemical stains (Luxolblue) and immunohistochemistry for MBP, MAG and NG2 (data notshown). Although human cells participated in the lesion repair, thelesion did not result in increased recruitment of hES cell–derivedprecursors into the area of demyelination or any detectable change inthe overall differentiation profile of the grafted cells. Panels of humancells labeled for various oligodendrocyte markers in confocal andepifluorescence microscopy are provided as Supplementary Figures 2and 3 online.

b

dc

GFP/NeuN

Dlx2/hNA Dcx/hNA Calretinin/hNA

Nestin/hNA MBP/GFP

fhg

a

e

b c d

f g h

DAPI Oct-3/4

Oct4+, Pax6−,Sox1−, Nestin−

Oct4−, Pax6+,Sox1+, Nestin+,

NCAM+

Oct4, −Sox1−, A2B5+,Nestin+, Musashi+

(Pax6+/−, Tuj1+/−, GFAP+/−)

(+/− Noggin) FGF2 + EGF

KSR medium N2 mediumPolyornithine/lamininMS5 stroma

Neural induction

DAPI/Nestin/GFP DAPI/A2B5/GFP DAPI/Tuj1/Pax6

Neural proliferation

eGFP

hES cells onstromal cells

Day: 0 28 35 42 49 56

P0 P1 P2 P3 P4

Rosetteneural

precursors

Rosette-freeneural

precursors

a

d

e f g

b c

Figure 2 Transplantation of hES-derived neural precursors into the

striatum of the young adult rodent brain. (a) Camera lucida drawing of

a representative coronal section through the brain of a lesioned animal

illustrating the distribution of hNA+ cells 11 weeks post grafting: (b) EGFP+/

NeuN+ neurons in the graft core (green outline in a), (c) Migrating hNA+/

Nestin+ cells in the contralateral corpus callosum, (d) z-stack confocal

image of a EGFP+/MBP+ cell located in the remyelinated cingulum after

lysolecithin lesion. (e) Diagram of a sagittal brain section illustrating thelocation of hNA+/Dlx2+ cells (f) and hNA+/GFAP+ (f inset) in the SVZ.

(g) hNA+/doublecortin+ neuronal precursors in the RMS. (h) z-stack confocal

image of hNA+/Calretinin+ neuron in the olfactory bulb. DAPI nuclear

counterstains are in blue. Scale bars, 10 mm.

Figure 1 In vitro derivation of EGFP+ neural precursors from human ES

cells. (a–c) Stable EGFP expressing hES cells (WA01) 14 months after

transduction with the CPG PGK-EGFP SIN lentivirus. Over 95% of the

Oct4+ undifferentiated hES cells (b) expressed EGFP (c). (d) Schematic

representation of the neural differentiation protocol. HES cells are plated

on MS5 stromal cells from days 0–28, neural rosettes are mechanically

isolated and replated feeder-free on precoated dishes. Neural precursor cells

derived from neural rosettes are propagated in the presence of FGF2 andEGF as attached monolayer cultures and passaged weekly. Cells were

used at 4 weeks of precursor cell proliferation. (e,f) Typical markers for

undifferentiated hES cells, neural rosettes and neural precursor stage are

listed in red, absent markers are in blue. Characteristic markers expressed

at the neural precursor stage were nestin (e) and A2B5 (f). (g) Pax 6

expression was observed in the majority of neural rosettes cells and greatly

reduced at the neural precursor stage. Up to 8% Tuj1+ immature neurons (g)

were observed among the neural precursor cells. EGFP and Pax6 are

depicted in green, DAPI nuclear stain in blue and all other markers in red.

Scale bar in (c): 25 mm and applies to Fig. a–c; scale bar in f, 10 mm, and

applies to e also; scale bar in g, 25 mm.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Whereas the in vitro differentiation of hES cell–derived precursorsinto mature oligodendrocytes has proven challenging18, our datasuggest that hES cell–derived precursors are capable of differentiationinto myelin-expressing oligodendrocytes in the young adult brainwithout the need for demyelination-induced environmental cues.However, the lysolecithin model is suboptimal for evaluating thefunctional capacity of hES cell–derived myelin-producing cells. Futurestudies may include transplantation of hES cell progeny into modelssuch as the shiverer mouse or the myelin-deficient rat, whichconstitute a more significant challenge for remyelination.

None of the grafted animal brains harbored a teratoma (absence ofalpha-fetoprotein, cytokeratin, myosin, SSEA-3 and SSEA-4) or othertypes of tumor as assessed by histological analysis on hematoxylin &eosin (H&E) sections. EGFP expression was maintained 11 weeks aftergrafting (Fig. 1a and Supplementary Figs. 1–3), indicating thatlentiviral-mediated gene transfer might be an efficient way to achievelong-term expression of therapeutic transgenes in hES cell–derivedneural progeny in vivo.

The adult rodent brain harbors sites of persistent neurogenesisincluding the subventricular zone (SVZ) and the dentate gyrus.

Interestingly, human cells were detected in the ipsi- and contralateralSVZ, where they expressed markers typical of ‘transit amplifying’ typeC cells (nestin+, BrdU+, dlx-2+, Fig. 2d), type A migrating neuronalprecursors (doublecortin+) and GFAP+ type B astrocytes, the putativestem cells in the adult brain19 (Fig. 2d inset). Whereas human cellswere detected in the fimbria (the white matter output of thehippocampus), none were found within the dentate gyrus, anothersite with a highly regulated neurogenic niche20. Human cells expres-sing doublecortin (Dcx) were found within the rostral migratorystream (RMS; Fig. 2e) and in the olfactory granular and peri-glomerular layers, where they gave rise to differentiated interneuronsincluding calretinin+ olfactory neurons (Fig. 2f). The region-specificimmunohistochemical profile of the human cells is identical to thatexhibited by endogenous SVZ stem cell progeny. All colocalization wasconfirmed by serial 3D-reconstruction of 0.5- to 0.8-mm confocalsections (Supplementary Figs. 1,2 and 4). No human cells were foundin other regions of the olfactory or orbital lobe, suggesting that theyreached the olfactory bulb selectively by the RMS route rather than byrandom migration. Also noted was an absence of neuronal differentia-tion outside the core graft and the RMS/olfactory bulb area.

hNA/BrdU/DAPI hNA/BrdU/DAPI

hNCAM/DAPI

hNCAM/DAPI hNA/BrdU/GFAP hNA/Dcx/BrdU

hNA/BrdU/DAPI 800

20

10

Hum

an B

rdU

+ c

ells

(avg

/sec

tion)

0

20

10

Hum

an B

rdU

+ c

ells

(avg

/sec

tion)

0

20

10

Hum

an B

rdU

+ c

ells

(avg

/sec

tion)

0

600

400

200

0Prox.RMS

DistalRMS

OB Prox.RMS

DistalRMS

OB

Prox.RMS

DistalRMS

OB

Prox.RMS

DistalRMS

OB

Prox.RMS

DistalRMS

OB

Prox.RMS

DistalRMS

OB

Tota

l Brd

U+

cel

ls(a

vg/s

ectio

n)

800

600

400

200

0

Tota

l Brd

U+

cel

ls(a

vg/s

ectio

n)

800

600

400

200

0

Tota

l Brd

U+

cel

ls(a

vg/s

ectio

n)Day 1

Day 4

Day 28

a

d

e f g

b c h i

j k

l m

Figure 3 Kinetic analysis of human cell migration from the SVZ to the olfactory bulb post transplantation into the SVZ. (a–c) Examples of BrdU+/hNA+

cells (arrows) in the SVZ (a), RMS (b) and olfactory bulb (c). Inset in (c) shows a higher power magnification of the BrdU+ endogenous and human cellsdepicted in the marked area. HNA+/BrdU+ cells are indicated by arrows, olfactory glomeruli by arrowheads. (a–c) Images were taken 4 d after BrdU labeling,

corresponding to 18 d after transplantation; hNA (green), BrdU (red). (d) Multi-image reconstruction of human NCAM+ neural precursors traveling along

the RMS towards the olfactory bulb at day 18 after transplantation. (e) At 42 d after transplantation human cells persisted in the SVZ and were aligned in

human NCAM+ chains. (f–g) Triple immunohistochemistry at the same timepoint revealed BrdU+ human GFAP cells adjacent to the SVZ (f, BrdU in red, hNA

in green, GFAP in blue); and doublecortin+ cells in the RMS (g, BrdU in blue, hNA in green, doublecortin in red). (h–m) Kinetics of endogenous and hES-

derived precursor cell migration to the olfactory bulb: The number of total BrdU+ cells and the number of human BrdU+/hNA+ cells are shown at 1 d (h,i),

4 d (j,k), and 28 d (l,m) after BrdU exposure. Cells were quantified as average number of cells per uniform randomly selected section within a defined area

of the proximal RMS (immediately adjacent to the anterior SVZ), the distal RMS (immediately proximal to the olfactory bulb) and the olfactory bulb (see also

Fig. 4j). BrdU+ human cells within the graft core at the injection site are not included in this analysis, since their peri-SVZ location is in the immediate

vicinity of the injection site. Data are presented as mean 7 s.e.m. Scale bars (a–c) 25 mm except for inset (10 mm); (d) 100 mm; (e–f) 25 mm; (g) 10 mm.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

The capacity of hES cell–derived neural precursors to reach the SVZand to participate in olfactory bulb neurogenesis was further analyzedusing kinetic studies of cell proliferation and migration along theRMS. Unlike the first experiment, in which hES cell–derived neuralprecursors were injected into the striatum, the second experimentinvolved injection into the rostral SVZ. We selected this site to increasethe number of cells migrating along the RMS for time-based quanti-tative analyses. Normal young adult rats (n ¼ 8) were transplanted inthe rostral SVZ with 0.5 � 105 hES cell–derived (WA01; non-GFPexpressing) neural precursors21. Two weeks after grafting, animalsreceived BrdU injections (three doses of 100 mg/kg/8 h) over a24-h period. Animals were killed 1 d, 4 d and 28 d after BrdUadministration. Histological analysis confirmed the progressivemigration of BrdU+ human precursors from the SVZ to the RMSand olfactory bulb in parallel with endogenous precursor cellmigration (Fig. 3a–c).

At day 1, the distribution of all BrdU+ human cells was restricted tothe graft core at the injection site. Because of the proximity of theinjection site to the SVZ, it was difficult to distinguish actualmigration from coincidental localization to the SVZ. Therefore ouranalysis did not include any BrdU-labeled cells in the SVZ but onlythose that migrated to the RMS or olfactory bulb. Endogenous BrdU+

cells were present in the proximal RMS at day 1 and to a lesser extentin the distal RMS and olfactory bulb. This difference betweenendogenous and hES cell–derived cells may be explained by a slowermigration rate of the human cells compared with rat cells or by thepresence of a mitotic population of endogenous rat cells in the RMS.At day 4, the majority of BrdU-labeled human cells were detected inthe RMS, concomitant with an overall increase of endogenous BrdU+

cells observed in that region. Low power images revealed largenumbers of human cells traveling in the RMS, as identified byimmunohistochemistry for human-specific, neural cell–adhesionmolecule (NCAM; Fig. 3d). By week 4, the overall BrdU label was

significantly diminished and mostly confined to the olfactory bulb,consistent with previous work demonstrating a 50% decrease in thenumber of thymidine-labeled cells between days 15 and 45 after theirbirth in the SVZ22. Similarly, the number of human cells that reachand persist in the olfactory bulb by day 28 is low. However, themajority of these human cells in the olfactory bulb were BrdU-labeled(88% on day 28 versus 49% on day 4), demonstrating precursororigin. The progressive increase in the proportion of BrdU+ humancells reaching the olfactory bulb is highly suggestive of a process oftargeted migration from the SVZ that complies with the kinetics of theSVZ population (Fig. 3h–m). The increasing percentage of BrdU+

human cells in the olfactory bulb may be attributed to the slowmigration of the labeled human cells and a high degree of BrdUlabeling achieved by the regimen used. Representative phenotypic fatesof human BrdU+ cells were confirmed by triple immunohistochem-istry (Fig. 3f,g). At the end of week 4, human cells were detected in theSVZ lining the lateral wall of the posterior aspect of the lateralventricle (Fig. 3e). These cells were discontinuous with the graftcore, which was seen at the injection site more rostrally and may besuggestive of human cell contribution to the SVZ similar to the cellsobserved in the SVZ in the first experiment, 11 weeks after intrastriataltransplantation (Fig. 2d).

To achieve labeling rates sufficient for detection of BrdU+ humancells in the olfactory bulb, our BrdU scheme (three discrete doses over24 h) was aimed at labeling rapidly dividing transit amplifying cells23

rather than marking the slow cycling stem cells. hES cell–derived cellsin the SVZ were largely negative for BrdU at day 28 after labeling, aswere the majority of the host SVZ cells. Occasionally, a few BrdU+

human cells were detected in small clusters of human cells lining theSVZ (data not shown). The majority of endogenous and humanBrdU+ cells were located in the distal RMS and olfactory bulb 4 weeksafter labeling. Representative camera lucida drawings of the RMSand olfactory bulb at each time point examined demonstrate the

Prox.RMS

Prox.RMS

DistalRMS OB

DistalRMS OB

Day

1D

ay 4

Day

28

ParasagittalDay 28

Human BrdU + cell

Human BrdU − cell

Endogenous BrdU + cell

Prox.RMS

DistalRMS OB

Injectionsite

Ventricle

a b c ja′ b′ c′

d′ e′ f ′

g ′ h′ i′

d e f

g h i

Figure 4 Camera lucida drawings of representative coronal sections at the level of proximal rostral migratory stream (RMS), distal RMS and olfactory bulb

on days 1, 4 and 28 after BrdU injection. (a–i) Insets from are shown at higher magnification in (a¢–i¢). Blue circles represent endogenous BrdU+ cells,

green triangles represent BrdU- human cells and red squares BrdU+ human cells. These markers indicate the location of individual cells analyzed at an

initial magnification of 400� (one marker per cell). All sections were stained concomitantly for BrdU and human nuclear antigen. Brown outlines in the

olfactory bulb sections represent the olfactory glomeruli. (j) Represents a camera lucida drawing of a parasagittal section taken 28 d after BrdU injection.

Representation in Figures a and a¢ do not include the large number of human BrdU labeled cells that are found in the graft core and immediate vicinity of

the injection site. Note the near exclusive distribution of human cells to the immediate vicinity of the graft, the RMS and the olfactory bulb. Black lines

through the section indicate the approximate level of the coronal sections seen in this figure. Arrows indicate the approximate cell injection site and the

location of the lateral ventricle.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

spatial distribution of human cells, BrdU+ cells and BrdU+ humancells (Fig. 4).

Unlike the striatal grafts, human cells injected into the rostral SVZexhibited a more limited migration pattern and almost exclusivelytraveled along the RMS towards the olfactory bulb, whereas themajority remained within the graft core (Fig. 4j). Although markerexpression and the kinetics of human cell migration to the olfactorybulb presented here are compatible with endogenous SVZ precursorbehavior, future long-term BrdU labeling studies will be required tofully demonstrate the stem cell nature of the human cells integrated inthe SVZ. Whereas mouse and human ES cells have demonstrated apotential for multilineage and neuron subtype specification in vitro,the in vivo neuronal differentiation potential of hES cell–derivedneural precursors in the current study was restricted and similar tothat exhibited by primary neural progenitors. However, the ability ofhES cell–derived neural precursors to appropriately respond to envir-onmental cues in vivo suggests that environmental factors will becrucial in determining the utility of hES cells in regenerative therapies.

Recent studies have demonstrated that stem cell plasticity can bemimicked by a fusion of grafted and host cells24–26. We used twoindependent assays to test for cell fusion: DNA in situ hybridizationwith a rat-specific X chromosome probe combined with immunohis-tochemistry for hNA (Supplementary Fig. 5a online) and genomicin situ hybridization (GISH) of biotin-labeled rat and digoxigenin-labeled human total DNA (Supplementary Fig. 5b). Neither methodshowed evidence of nuclear fusion. Histochemical analysis (cresylviolet and H&E; ten slides/animal) and double-label immunohisto-chemistry (45,000 human cells examined) for hNA, DAPI andvarious cytoplasmic markers (nestin, beta-tubulin, S100-beta) didnot detect any multinuclear human cells.

The capacity of hES cell progeny to migrate extensively in the youngadult brain and differentiate into neurons, astrocytes and oligoden-drocytes underlines the potential utility of hES cells for cell therapies.That the cells integrate into the SVZ progenitor pool in young adultanimals and adopt an adult neural stem cell–like fate demonstratesappropriate interactions with the stem cell niche in the SVZ. Thesefindings also emphasize the potent role of environmental context indetermining the migration behavior and cell fate of hES cells in vivo.

METHODSCell culture and viral transduction. Undifferentiated hES cells (lines H1

(WA01, XY, passages (P) 40–65), and HES327 (ES03, XX, P 50–65) were

cultured on mitotically inactivated MEF (Specialty Media) and maintained

under growth conditions and passaging techniques described previously28. The

CPG PGK-EGFP SIN lentiviral vector was prepared as previously described13.

Native EGFP fluorescence was detected 3 d after transduction. Green colonies

were manually removed and replated repeatedly until uniform EGFP expression

was confirmed. Neural induction was obtained by growing EGFP cells on MS5

cells in knock-out serum replacement (KSR) medium as described previously14.

After 16 d, cells were switched from KSR medium to N2 medium. At 1 month,

neural precursors were mechanically isolated, replated feeder-free onto poly-

ornithine/laminin-coated culture dishes (50–100 � 103 cells/cm2) and main-

tained for 4 weeks in N2 supplemented with FGF-2 and EGF (20 ng/ml each,

R&D). Cells were passaged weekly after exposure to Ca2/Mg2-free HBSS for 1 h

at 25 1C, spun at 200g for 5 min and replated at 50–100 � 103 cells/cm2.

Slightly enhanced neural induction was observed in the presence of Noggin 500

ng/ml Noggin Fc Chimera (R&D), applied from days 0–28.

Animal surgery. All animal experiments were done in accordance with

protocols approved by our Institutional Animal Care and Use Committee

(IACUC) and following National Institutes of Health (NIH) guidelines for

animal welfare. Young adult Sprague Dawley female rats (86- to 92-d old at

time of grafting) were acquired from Taconic and used throughout the study.

Stereotactic implantation of graft cells was performed under full anesthesia

using a mixture of ketamine (Ketaset, Fort Dodge Animal Health) and xylazine

(AnaSed, Lloyd Laboratories). For the lysolecithin lesion, we injected 2 ml of 2%

lysolecithin in PBS (L-a-lyso-lecithin, Calbiochem) into the right cingulum (AP,

�0.3; ML, �2.0; V, �2.6 mm; and TB, �3.0). For the striatal implants,

the animals received a unilateral 1 ml (100,000 cells in sterile HBSS) injection at

the following coordinates: striatum: anteroposterior, �0.3; mediolateral,

�2.0; ventral, 2.8; tooth bar, �3.0. For the BrdU time-based analysis, the

50,000 cells were implanted in the SVZ: AP, +1.6; ML, �1.5; DV, �4.2; TB, �2.3.

All coordinates relative to bregma and ventral coordinates relative to cortex.

Immunosuppression. All rats received cyclosporine (Neoral 100 mg/ml;

Novartis) at 20 mg/kg/day intraperitoneally (i.p.). This regimen was initiated

2 d before grafting and maintained until the day they were killed.

BrdU administration. Two regimens of BrdU (97%; Aldrich) were used in the

different experiments. The first regimen was given to the animals with the

striatal implants. It consisted of 300 mg/kg/day given i.p. for the 3 consecutive

days preceding death (11 weeks after grafting). The second regimen consisted of

three doses of BrdU at 100 mg/kg administered every 8 h for a total of 3 doses

over 24 h. This regimen was initiated 2 weeks after transplantation in the SVZ.

Two animals were selected randomly and killed at the following time points

after BrdU administration: 1 d, 4 d and 4 weeks. BrdU was dissolved in sterile

normal saline and .007 M NaOH.

Tissue processing. Rats were deeply anesthetized with a 25-mg intraperitoneal

injection of pentobarbital solution (Nembutal Sodium Solution, Abbott

Laboratories). They were then transcardially perfused with 0.1% heparinized

normal saline at 4 1C (Sigma) followed by 4% paraformaldehyde (PFA) in PBS

also at 4 1C (pH 7.4). The brains were carefully extracted, post-fixed overnight

in 4% PFA at 4 1C, and subsequently transferred to 30% sucrose at 4 1C until

embedding. Optimal cutting temperature compound (O.C.T. Compound,

Tissue-Tek) was used to embed the brains and sections were cut on a freezing

cryostat and stored at �80 1C.

Immunohistochemistry. Sections were washed briefly with PBS 0.1% BSA

(Sigma). For fluorescence double immunohistochemistry, sections were first

blocked with 10% normal goat serum (Gibco) in PBS and 0.3% Triton X-100

(with the exception of surface antigens, where Triton X-100 was omitted).

Some antibodies required a pretreatment step as follows: 30 min in 2N HCl at

25 1C for BrdU; 3 min in 100% acetone at �20 1C for human nuclear antigen.

Primary antibodies were incubated overnight at 4 1C. Appropriate secondary

antibodies and fluorochromes (AMCA and Alexa conjugates (Molecular

Probes) or Cy-conjugates (Jackson Immunoresearch Labs)) were applied for

1 h at 25 1C followed by PBS washes, DAPI (Molecular Probes) counterstain

and mounted in glycerol. Triple labeling was carried out in the following

sequence: pretreatment for BrdU followed by a sodium borate wash, postfixa-

tion in 2% PFA then acetone at 20 1C for 3 min and incubation with all three

primary antibodies combined overnight. Secondary antibodies were used as

above. The primary antibodies included: rat anti-BrdU (1:40, Abcam, Cam-

bridge, UK); Calretinin (1:2000, Swant); Nestin (gift from R.D. McKay,

1:1,000); Dlx2 (gift from S. Anderson and J. Rubenstein, 1:100); TuJ1

(Covance/BabCo, monoclonal 1:200, polyclonal 1:1,000); human NCAM

(Eric-1, 1:100, Santa Cruz Biotechnology); EGFP (1:500, Molecular Probes);

A2B5 (1:50, Roche Diagnostics); Oct-3/4 (1:100, Santa Cruz Biotechnology);

Pax-6 (1:50, Covance/BabCo); DARPP-32 (1:200); SSEA-3 (MC631, 1:50,

DSHB); SSEA-4 (MC 813-70, 1:75, DSHB); Pancytokeratin (1:300, Sigma);

a-fetoprotein (1:300, Sigma); myosin (1:400, Sigma); CNP (Sternberger, 1:100);

and human nuclear antigen (1:50), GFAP (1:1,000), rat MBP (1:200), MAG

(1:200), NG2 (1:100), Neu-N (1:50), Musashi (1:100), DCX (1:3,000), O1

(1:50), O4 (1:50), MOSP (MAB 328, 1:1,000), Galc (1:50), all from Chemicon.

Confocal sections were imaged on a Leica TCS SO2 AOBS set-up and

reconstructed using a Leica Confocal Software Lite package.

Tunel. Sections were blocked with 3% H2O2 in methanol for 15 min at 25 1C,

rinsed and permeabilized in 0.1% Triton-X in 0.1% sodium citrate for 2 min on

ice. They were then incubated in the Tunel mixture prepared according to the

manufacturer’s instructions (in-situ cell death detection kit POD, Roche


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Diagnostics). In some cases, the FITC label was converted to DAB chromogen

using the manufacturer’s POD converter kit.

Quantification. Graft volumes were estimated using the Cavalieri estimator

probe (Stereo Investigator version 6, Microbrightfield). Total cell number was

assessed separately in the graft core and in the migrating population. Systematic

random sampling was applied to the regions of interest (graft core and areas of

distribution of human cells) as defined on serial sections. The stereological

software was used to design and implement the fractionator probes at a

coefficient of error (Gundersen) of r0.05. For the cell counts within the

RMS and olfactory bulb (Fig. 3), cells were quantified as average number of

cells per uniform randomly selected section within the regions of interest. The

latter were defined as follows: proximal RMS (immediately adjacent to the

anterior SVZ), distal RMS (immediately proximal to the olfactory bulb) and

olfactory bulb (within the olfactory bulb). Data are presented as mean 7 s.e.m.

In situ hybridization (GISH). Total genomic human and rat DNA was

fragmented using nick translation and labeled with digoxigenin and biotin,

respectively. Absence of cross-reactivity was confirmed by hybridization with

normal human and rat cells. Brain sections were pretreated with pepsin (0.05%

in 0.1 M HCl) for 5 min at 37 1C, washed in PBS and dehydrated in graded

alcohol solutions. They were then postfixed in 4% PFA and washed. DNA

probes (180 ng DNA per slide in 20 ml) were diluted in hybridization buffer (2�SSC, 50% formamide, 01% SDS, 1� Denhardt’s, 40 mM sodium phosphate,

pH.7) and applied to the slides. The slides were sealed in rubber cement,

denatured at 80 1C for 7 min and hybridized overnight at 37 1C. The sections

were washed in 2� SSC, permeabilized in 0.1% Tween 4� SSC for 10 min at

37 1C and incubated in mouse anti-digoxigenin (1/500, Chemicon) for 1 h at

25 1C. Appropriate secondary antibodies conjugated to Alexa fluorochromes

(Molecular Probes) were applied for 1 h at 25 1C followed by DAPI counterstain.

For the combined immunofluorescence for hNA and rat-CMS X probe

hybridization, slides were postfixed in 4% PFA for 15 min at 25 1C, pretreated

in cold acetone for 3 min at �20 1C. Incubation with human nuclear antibody

(1/50) was carried out at 4 1C overnight. This was followed by incubation in an

Alexa 555-conjugated secondary antibody and PBS rinses. The sections then

underwent pre-treatment with pepsin as described above, followed by dehy-

dration in serial graded alcohols. In situ hybridization for the rat X chromo-

some was carried out using an FITC-labeled rat X Chromosome probe

(Cambio and ID Labs) and according to manufacturer’s instructions. The

slides were mounted in glycerol.


ACKNOWLEDGMENTSWe thank R. McKay for the nestin antibody, S. Anderson and J. Rubenstein forthe Dlx2 antibody and M. Leversha for the DNA probes and assistance withGISH. Supported by the National Institute of Neurological Disorders andStroke, NIH, R21NS046045, the Michael W. McCarthy Foundation, the M.J. FoxFoundation, and the Kinetics Foundation. M.S. is supported by NIH grantsHL57612 and CA08748.


Received 12 November 2004; accepted 8 March 2005


1. Kim, J.H. et al. Dopamine neurons derived from embryonic stem cells function in ananimal model of Parkinson’s disease. Nature 418, 50–56 (2002).

2. Bjorklund, L.M. et al. Embryonic stem cells develop into functional dopaminergic neu-rons after transplantation in a Parkinson rat model. Proc. Natl. Acad. Sci. USA 99,2344–2349 (2002).

3. Barberi, T. et al Neural subtype specification of fertilization and nuclear transferembryonic stem cells and application in parkinsonian mice. Nat. Biotechnol. 21,1200–1207 (2003).

4. Brustle, O. et al. Embryonic stem cell-derived glial precursors: a source of myelinatingtransplants. Science 285, 754–756 (1999).

5. McDonald, J.W. et al. Transplanted embryonic stem cells survive, differentiate andpromote recovery in injured rat spinal cord. Nat. Med. 5, 1410–1412 (1999).

6. Englund, U., Fricker-Gates, R.A., Lundberg, C., Bjorklund, A. & Wictorin, Y. Transplan-tation of human neural progenitor cells into the neonatal rat brain: extensive migrationand differentiation with long-distance axonal projections. Exp. Neurol. 173, 1–21(2002).

7. Englund, U., Bjorklund, A. & Wictorin, K. Migration patterns and phenotypic differ-entiation of long-term expanded human neural progenitor cells after transplantationinto the adult rat brain. Dev. Brain Res. 134, 123–141 (2002).

8. Caldwell, M.A. et al. Growth factors regulate the survival and fate of cells derived fromhuman neurospheres. Nat. Biotechnol. 19, 475–479 (2001).

9. Svendsen, C.N. et al. Long-term survival of human central nervous system progenitor cellstransplanted into a rat model of Parkinson’s disease. Exp. Neurol. 148, 135–146 (1997).

10. Thomson, J.A. et al. Embryonic stem cell lines derived from human blastocysts.Science 282, 1145–1147 (1998).

11. Reubinoff, B.E. et al. Neural progenitors from human embryonic stem cells. Nat.Biotechnol. 19, 1134–1140 (2001).

12. Zufferey, R., Nagy, D., Mandel, R.J., Naldini, L. & Trono, D. Multiply attenuatedlentiviral vector achieves efficient gene delivery in vivo. Nat. Biotechnol. 15, 871–875(1997).

13. May, C. et al. Therapeutic haemoglobin synthesis in beta-thalassaemic mice expressinglentivirus-encoded human beta-globin. Nature 406, 82–86 (2000).

14. Perrier, A.L. et al. Derivation of midbrain dopamine neurons from human embryonicstem cells. Proc. Natl. Acad. Sci. USA 101, 12543–12548 (2004).

15. Gensert, J.M. & Goldman, J.E. Endogenous progenitors remyelinate demyelinatedaxons in the adult CNS. Neuron 19, 197–203 (1997).

16. West, M.J. Design-based stereological methods for counting neurons. Prog. Brain Res.135, 43–51 (2002).

17. West, M.J. Design based stereological methods for estimating the total number ofobjects in histological material. Folia Morphol. (Warsz. ) 60, 11–19 (2001).

18. Studer, L. Stem cells with brainpower. Nat. Biotechnol. 19, 1117–1118 (2001).19. Doetsch, F., Caille, I., Lim, D.A., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Subven-

tricular zone astrocytes are neural stem cells in the adult mammalian brain. Cell 97,703–716 (1999).

20. Monje, M.L., Toda, H. & Palmer, T.D. Inflammatory blockade restores adult hippocam-pal neurogenesis. Science 302, 1760–1765 (2003).

21. Suhonen, J.O., Peterson, D.A., Ray, J. & Gage, F.H. Differentiation of adult hippo-campus-derived progenitors into olfactory neurons in vivo. Nature 383, 624–627(1996).

22. Petreanu, L. & Alvarez-Buylla, A. Maturation and death of adult-born olfactory bulbgranule neurons: role of olfaction. J. Neurosci. 22, 6106–6113 (2002).

23. Doetsch, F., GarciaVerdugo, J.M. & AlvarezBuylla, A. Cellular composition and three-dimensional organization of the subventricular germinal zone in the adult mammalianbrain. J. Neurosci. 17, 5046–5061 (1997).

24. Medvinsky, A. & Smith, A. Stem cells: fusion brings down barriers. Nature 422, 823–825 (2003).

25. Alvarez-Dolado, M. et al. Fusion of bone-marrow-derived cells with Purkinje neurons,cardiomyocytes and hepatocytes. Nature 425, 968–973 (2003).

26. Weimann, J.M., Johansson, C.B., Trejo, A. & Blau, H.M. Stable reprogrammedheterokaryons form spontaneously in Purkinje neurons after bone marrow transplant.Nat. Cell Biol. 5, 959–966 (2003).

27. Reubinoff, B.E., Pera, M.F., Fong, C.Y., Trounson, A. & Bongso, A. Embryonic stem celllines from human blastocysts: somatic differentiation in vitro. Nat. Biotechnol. 18,399–404 (2000).

28. Zhang, S.C., Wernig, M., Duncan, I.D., Brustle, O. & Thomson, J.A. In vitro differentia-tion of transplantable neural precursors from human embryonic stem cells. Nat.Biotechnol. 19, 1129–1133 (2001).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Transient inhibition of BMP signaling by Noggin inducescardiomyocyte differentiation of mouse embryonicstem cellsShinsuke Yuasa1,2, Yuji Itabashi1, Uichi Koshimizu4, Tomofumi Tanaka4, Keijiro Sugimura4,Masayoshi Kinoshita1, Fumiyuki Hattori2,4, Shin-ichi Fukami3, Takuya Shimazaki3, Hideyuki Okano3,5,Satoshi Ogawa1 & Keiichi Fukuda2

Embryonic stem (ES) cells are a promising source of

cardiomyocytes, but clinical application of ES cells has

been hindered by the lack of reliable selective differentiation

methods. Differentiation into any lineage is partly dependent

on the regulatory mechanisms of normal early development.

Although several signals, including bone morphogenetic

protein (BMP)1,2, Wnt3 and FGF4, are involved in heart

development, scarce evidence is available about the exact

signals that mediate cardiomyocyte differentiation. While

investigating the involvement of BMP signaling in early heart

formation in the mouse, we found that the BMP antagonist

Noggin is transiently but strongly expressed in the heart-

forming region during gastrulation and acts at the level of

induction of mesendoderm to establish conditions conducive

to cardiogenesis. We applied this finding to develop an

effective protocol for obtaining cardiomyocytes from mouse

ES cells by inhibition of BMP signaling.

BMP signaling is crucial in mesodermal induction and cardiacformation1,2. However, simple stimulation with BMP2/BMP4 didnot augment or suppress cardiomyocyte induction from ES cells(data not shown). In the vertebrate nervous system, Noggin andother BMP inhibitors (chordin and follistatin) are involved in neuraldifferentiation in a context-dependent fashion5,6. We hypothesizedthat BMP antagonists may also be involved in cardiomyocyte induc-tion. Here, we performed whole-mount in situ hybridization forvarious BMP antagonists on mouse embryos at different gastrulationstages. The BMP antagonist Noggin was transiently but stronglyexpressed in the heart-forming area (Fig. 1a,b). It was clearlyexpressed at the cardiac crescent at mouse embryo day E7.5 and thelate crescent stage at E8.0, but was barely detectable in the linear hearttube after E8.5. In contrast, the expression of Noggin at the notochordcontinued after E8.5, as reported previously7,8. Sectioning of whole-mount samples from E7.5 and E8.0 showed expression of Noggin in

both the endodermal and mesodermal layers and made clear thatNoggin was derived from the primary heart field (Fig. 1c,d). Thismarked difference in the time course of Noggin expression betweenthe heart-forming region and notochord suggested that transientexpression of Noggin functions in cardiomyocyte differentiation.

We stimulated mouse ES cells in suspension cultures with Noggin invarious ways (Fig. 2a,b). We administered Noggin before or afterembryoid body formation to mimic the transient and strong expres-sion of Noggin at the early gastrulation stage. Discontinuation ofleukemia inhibitory factor (LIF) and addition of Noggin before orafter embryoid body formation did not increase the incidence offormation of spontaneously beating embryoid bodies (Fig. 2b, rows2,3). Interestingly, addition of Noggin on day 0 and discontinuation ofLIF on day 3 slightly but substantially increased the beating embryoidbody incidence (Fig. 2b, row 4), suggesting that the optimal timing forNoggin might be both before and after embryoid body formation.Next, we added Noggin at either �3, 0, +1, +2 or +3 d (Fig. 2b, rows5–9), and LIF before embryoid body formation. Although Noggin atday 0 (Fig. 2b, row 6) slightly increased the beating embryoid bodyincidence, this incidence gradually decreased at the later time points.Based on these results, we administered Noggin at day –3 and day 0from embryoid body formation. This led to a marked increase inbeating embryoid body incidence to 95.3% at 10 d (Fig. 2b, rows10–16), and continued growth of embryoid bodies to day 14. Theseresults suggest that the cardiomyocyte inductive activity of Noggin wasrestricted to the period from 3 d before one day after embryoid bodyformation and that the ES cells must initially be undifferentiated.

This protocol was effective in two independent ES cell lines, EB3and R1, and the optimal concentration of Noggin was 150 ng/ml(Fig. 2c and Supplementary Fig. 1 online). To demonstrate that thiseffect was specific to inhibition of the BMP pathway, we administeredvarious concentrations of BMP2 at day 0 (Fig. 2d). Even low dosesof BMP2 strongly inhibited Noggin-dependent cardiomyocyte induc-tion. To confirm that the inhibition of BMP signaling in the early


1Division of Cardiology, Department of Medicine, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku, Tokyo 160-8582, Japan. 2Department ofRegenerative Medicine and Advanced Cardiac Therapeutics, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku, Tokyo 160-8582, Japan. 3Departmentof Physiology, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku, Tokyo 160-8582, Japan. 4Daiichi Suntory Biomedical Research Co. Ltd., 1-1-1Wakayamadai, Shimamoto-cho, Mishima-gun, Osaka 618–8513, Japan. 5Core Research for Evolutional Science and Technology (CREST), Japan Science and TechnologyAgency (JST), Kawaguchi, Saitama, 332-0012, Japan. Correspondence should be addressed to K.F. ([email protected]).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

phase of differentiation could accelerate cardiomyocyte induction,soluble BMP receptor-1A (BMPR-1A) or another BMP antagonist,chordin, was also administered, and cardiomyocyte induction wasobserved. Both interventions augmented the incidence of beating inindividual embryoid bodies (Fig. 2e). In contrast, administration ofvarious growth factors, including insulin-like growth factor-1 (IGF-1),fibroblast growth factor (FGF2) and BMP2, using the same protocoldid not boost cardiomyocyte induction (Fig. 2f). These resultssuggest that inhibition of BMP signaling in the undifferentiated orimmediate early phase of ES cell differentiation is crucial for cardio-myocyte differentiation.

Next, we examined which step of cardiomyocyte developmentNoggin acted upon. Noggin-treated ES cells expressed markedlyhigher levels of brachyury T than untreated cells, and then showedstrong induction of cardiomyocyte marker gene expression (Nkx2.5and Tbx5). Despite this increase in brachyury-T expression, theexpression of other early mesodermal markers transiently increasedbut then subsequently decreased (Fig. 2g). We also performed wholemount in situ hybridization of the embryoid bodies, and quantifiedthese mesodermal marker-positive cells (Fig. 2h,i). Taken together,these data suggest that Noggin acts principally between the undiffer-entiated and brachyury-T-positive states. Brachyury T is a marker ofmesendodermal progenitors that can differentiate into mesoderm orendoderm depending on culture conditions9. In our experiments,Noggin increased both the proportion of the cells expressing brachy-ury T by 1.8-fold and the level of brachyury-T mRNA percell by sixfold (Fig. 2j). This suggests that an increase in mRNAper cell is essential for cardiomyocyte induction from undifferen-tiated ES cells, and that there may be subpopulations within thebrachyury-T-positive cells that can be distinguished by their levels ofexpression. The increase in cells expressing high levels of brachyury Tthat formed mesendoderm resulted in the large increase in Nkx2.5-positive cells.

To quantify the incidence of cardiomyocyte induction with Noggintreatment, we immunostained for cardiac-specific proteins and

observed the results by confocal laser microscopy. Most cells in theNoggin-treated embryoid bodies stained positive for myosin heavychain (MHC), myosin light chain (MLC), atrial natriuretic peptide(ANP), cardiac troponin I and sarcomeric actinin (Fig. 3a–c). Incontrast, the cardiomyocyte content was markedly lower in the controlor in embryoid bodies treated with other Noggin protocols. Theoptimal Noggin protocol led to synchronous beating of the entireembryoid body (see Supplementary Video online). The isolated cellsexpressed many cardiac markers and had a typical cardiac myocytemorphology. At day 10, the embryoid bodies were attached to thegelatin-coated dishes and stained with anti-MHC antibodies. Therewas an B100-fold increase in the number of cardiomyocytes com-pared with control.

The Noggin protocol efficiently induced expression of cardiactranscription factors, including Nkx2.5, GATA4, TEF1, Tbx5 andMEF2C, whereas expression of the stem cell marker Oct3/4 rapidlydecreased (Fig. 3d). Cardiac-specific proteins were also stronglyinduced, including ANP, brain natriuretic peptide, MLC-2v, MLC-2a, a-MHC, b-MHC and a-cardiac actin. Western blot analysisrevealed that the Noggin-treated embryoid bodies expressed GATA4,troponin I, MLC and ANP at levels that were 10- to 450-fold higherthan those seen with the other protocol (Fig. 3e,f). To investigatewhether this inductive phenomenon was cell autonomous or non-autonomous, we treated ES cells stably transfected with the geneencoding green fluorescent protein (GFP) with Noggin, and thencombined them with untreated GFP– ES cells just before embryoidbody formation. The majority of GFP+, Noggin-treated ES cellsdifferentiated into cardiomyocytes in the embryoid bodies, whereasvery few GFP– untreated ES cells became cardiomyocytes (Fig. 3g,h).These findings suggest that Noggin-mediated induction of cardio-myocyte formation is a cell-autonomous phenomenon.

A number of growth factors and chemical compounds inducecardiomyocyte differentiation of mouse ES cells, including reactiveoxygen species (2.7-fold increase in beating embryoid body inci-dence)10, TGFb plus BMP2 (threefold increase)11, targeting ofRBP-Jk (downstream of notch signaling)-gene (20-fold increase)12,ascorbic acid (fivefold increase)13, as well as IGF-1, FGF, oxytocin,erythropoietin, retinoic acid and dimethyl sulfoxide14–16. To ourknowledge, however, no previous protocol for increasing cardiomyo-cyte differentiation is as efficient as our present protocol (approxi-mately 100-fold increase in the number of cardiomyocytes comparedwith control). The efficiency of our protocol may reflect the fact that itmakes use of endogenous factors and is modeled on in vivo cardio-myoctye induction.

Accumulating evidence implicates BMP signaling as a potent heart-inductive signal. Administration of BMP-2/BMP-4 to explant culturesfrom chicken embryo induces full cardiac differentiation in stages 5–7anterior medial mesoderm, a tissue that is normally not cardio-genic17,18. In contrast, before stage 3 or during early stages ofgastrulation, both BMP2 and BMP4 inhibit cardiomyogenesis19.

NC

Noggin

Noggin Noggin Nkx2.5Nkx2.5

NogginE7.5

E7.5 E7.5

Noggin Nkx2.5E8.0

E7.5

E8.0

E8.25

E8.5

E9.0

Nkx2.5 Nkx2.5

iiiiii

ivvvi

CC CC

CC CCNC

i i

ii ii

iii iii

iv iv

v v

vi vi

a c

b d

Figure 1 Transient expression of noggin at the heart forming area. (a) Whole-

mount in situ hybridization of noggin and Nkx2.5 was performed at mouse

embryo stages E7.5, E8.0, E8.25, E8.5 and E9.0. Note that noggin was

strongly expressed at the cardiac crescent (E7.5) and late crescent stage

(E8.0), but was undetected after E8.5. In contrast, Nkx2.5 was expressed

thereafter. Arrows indicate the heart. (b) The schema of noggin and Nkx2.5

expression at E7.5. CC: cardiac crescent, NC: notochord, LHT: linear heart

tube. (c) Section of samples at E7.5 and E8.0 with the whole mount in situhybridization. i–vi represented the site of the section as shown in a. (d) The

schema of noggin and Nkx2.5 expression at E7.5.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Although BMPs are expressed in lateral plate mesoderm including theanterior lateral plate20, stimulation of ES cells by BMP2 or BMP4 doesnot augment cardiomyocyte differentiation. Together, these findingssuggest that BMPs play multiple roles in mesodermal induction andspecific organ differentiation and that their temporal and spatialexpression is critical in cardiomyocyte induction19.

In the vertebrate nervous system, the local action of Noggin andother BMP inhibitors on BMP signaling is very important in neuralinduction, in patterning during embryonic development and in adultneurogenesis21. In Xenopus laevis gastrula-stage embryos, Noggin andother BMP inhibitors are secreted by the Spemann organizer andinduce neural tissue from dorsal ectoderm7,22,23 by inhibiting ecto-dermal BMPs24. In the developing neural tubes, BMP has been shownto specify the dorsal fates of neural progenitor cells25. BMP inhibitors

are also expressed in ventral somites or in the notochord, suggestingthat some are involved in a counter gradient of BMP activity along thedorso-ventral axis.

Based on the analogy to the central nervous system, we suspectedthat the context-dependent differential action of BMPs in cardiomyo-cyte induction might be explained by local action of Noggin and otherBMP inhibitors. We found that Noggin is transiently but stronglyexpressed at the anterolateral plate in mouse embryos at E7.0–E8.0and is critical in cardiomyocyte induction. The restricted andhighly effective window of Noggin’s inductive action for cardio-myocyte differentiation from ES cells exactly matched the normaldevelopmental conditions in the heart-forming area in E7.0-E8.0embryos. From the present results, we propose that BMP signalingis essential for at least two steps in the cardiomyocyte induction

Days Days

NogginControl

Noggin

Contro

l

Noggin

Contro

l

(%)100

80

60

40

20

0

100

80

60

40

20

0

(%)

Brachyury T

Brachyury T (day 3)(fold)

Cel

l cou

nt

RN

A/C

ell

Nkx 2.5 Tbx 5

3 5 7 3 5 7

8

6

4

2

0

Day 3 Day 5 Day 7 Day 3 Day 5 Day 7NogginControl

Brachyury T

Nkx 2.5

Tbx 5

h i j

Embryonic stem cells (+LIF)

Embryonic stem cells (+LIF +Noggin)

(+Noggin –LIF)

Embryoid body formation in suspension

on gelatin-coated dish

on gelatin-coated dish

Day –3

Day 0

123456789

10111213141516

Day –

3 LI

F

Day –

3 Nog

gin

Day 0

Nog

gin

Day 1

Nog

gin

Day 2

Nog

gin

Day 3

Nog

gin0 20 40 60 80 100

% beating EB incidence

NSNS

NS

NSNS

NS

NSNSNS

NSNS

NS

*

**

*

a b c d e

0

1,50

050

015

0 50 15 5 1.5

0.5 0

20

40

60

80

100In

cide

nce

(%)

NSNSNS

**

*

*

**

*

Noggin (ng/ml)

EB3

Inci

denc

e (%

)

100

80

60

40

20

0

NS

*

**

******

500 50 5 0.

50.

050.

005

Contro

l

BMP2 (ng/ml)

Days Days Days Days Days

Brachyury T Nkx 2.5 Tbx 5 Flk 1 GATA 1(fold) (fold) (fold) (fold) (fold)

800

600

400

200

00 1 2 3 4 5 7 0 1 2 3 4 5 7 0 1 2 3 4 5 7 0 1 2 3 4 5 7 0 1 2 3 4 5 7

100

60

20

0

1,200

1,000

800

400

0

200

100

0

3,000

2,000

1,000

0

Inci

denc

e (%

)

100

80

60

40

20

0

*

**

**** **

******

Days

7 10 13

BMPR1A

Chordin

Noggin

Control

Inci

denc

e (%

)

100

80

60

40

20

0

**

NS NS NS NS NS NS

IGF1

1 nm

IGF1

10 n

m

FGF2 1

nm

FGF2 10

nm

BMP2

1 nm

BMP2

10 n

m

Noggin

Contro

l

f g

Figure 2 Protocol and efficiency of the cardiomyocyte induction from ES cells using noggin, chordin and soluble BMP receptor-1A. (a) Representative

schema of the protocol of cardiomyocyte induction from ES cells. (b) The efficiency of various protocols for noggin (150 ng/ml) exposure were compared.

(c) Dose-efficiency relationship of noggin administration was demonstrated using two different ES cell lines, EB3 and R1 (Supplementary Fig. 1 online).Both lines showed the same dose-efficiency relationship. (d) Administration of low dose of BMP2 abolished the effect of noggin for cardiomyocyte induction,

indicating the BMP2 concentration was critical in this phenomenon. (e) Effect of other BMP antagonists, chordin and soluble BMPR-1A (BMP neutralizing

receptor), on cardiomyocyte induction. Both chordin and BMPR-1A were administered using the same protocol as noggin (150 ng/ml). Both of these BMP

antagonists induced cardiomyocyte induction from ES cells at the same level as noggin, indicating that transient relief from the intrinsic BMP signal is

critical for cardiomyocyte induction. (f) Other factors including IGF-1, FGF-2 and BMP2 did not affect cardiomyocyte induction with this protocol.

(g) Quantitative RT-PCR of early mesodermal markers and cardiac transcription factors. Black column, noggin-treated ES cells; white column, nontreated

ES cells. Each column was normalized by GAPDH. (h) Section of embryoid bodies at 3, 5 and 7 d after embryoid body formation with the whole mount

in situ hybridization. (i) To identify individual cells in embryoid body sections, nuclei were stained with propidium iodide. Cells positive for brachyury T,

Nkx2.5 and Tbx5 were counted. Cell numbers are presented as a percentage of the total. (j) The increased number of brachyury-T-positive cells and the

increased levels of brachyury-T mRNA at day 3 after embryoid body formation are shown. *P o 0.05, **P o 0.01 versus control; NS, not significant.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

process: mesodermal induction26 and cardiomyocyte differentiation1,2.However, between these steps, a transient block of intrinsic BMPsignaling may be the most important step for determining cardiomyo-genic differentiation.

METHODSWhole-mount in situ hybridization. Pregnant ICR wild-type mice were

purchased from Japan CLEA. All experiments were approved by the Keio

University Ethics Committee for Animal Experiments. Mice from embryonic

day (E) 7.5, 8.0, 8.25, 8.5 and 9.0 were removed, and whole-mount in situ

hybridization was performed using digoxigenin-labeled RNA probes as

described27. The full-length cDNAs for mouse Noggin and nkx2.5 (accession

number NM_008711 and NM_008700, respectively) were obtained by RT-PCR

and subcloned into pBluescript plasmid. The cDNAs for mouse Tbx5 and

brachyury T were kindly provided by H. Yamagishi and H. Bernhard,

respectively. The probes were transcribed with T3 or T7 RNA polymerase.

Cell culture. Mouse embryonic fibroblast-free ES cells were used. Undiffer-

entiated ES cells (EB328, R129) were maintained on gelatin-coated dishes in

GMEM supplemented with 10% FBS (Equitechbio), 2 mM L-glutamine,

0.1 mM nonessential amino acids, 1 mM sodium pyruvate, 0.1 mM

2-mercaptoethanol and 2,000 U/ml murine LIF (Chemicon International).

EB3 cells (a kind gift from H. Niwa, Riken, Japan), which carry the blasticidin

S-resistant selection marker gene driven by the Oct3/4 promoter (active in the

undifferentiated status) were maintained in medium containing 20 mg/ml

blasticidin S to eliminate differentiated cells. EB3 is a subline derived from

E14tg2a ES cells30, and was generated by targeted integration of the Oct3/4-

IRES-BSD-pA vector28 into the Oct3/4 allele.

Differentiation of ES cells. ES cells were cultured on gelatin-coated dishes in

a-MEM supplemented with 10% FBS (Equitechbio), 2 mM L-glutamine, 0.1

mM nonessential amino acids, 1 mM sodium pyruvate, 0.1 mM 2-mercap-

toethanol, 2,000 U/ml LIF and 0.15 mg/ml Noggin (Noggin-Fc, R&D) for 3 d.

Then, the cells were trypsinized, and cultured to form spheroids (embryoid

bodies) from a single cell using a three-dimensional culture system in the same

medium as described above minus the LIF on uncoated Petri dishes to induce

embryoid bodies. FGF2, IGF-1, BMP2, chordin and BMP receptor-1A/Fc

(BMPR-1A) were purchased from R&D.

Histological and immunohistochemical analysis. Embryoid bodies

(12–14 d) were fixed in 4% paraformaldehyde for 45 min and embedded

using Tissue-Tek OCT (Sakura Finetek). In some experiments, the isolated

cells were plated on gelatin-coated glass coverslips at low density and fixed in

4% paraformaldehyde for 5 min. The samples were exposed to primary

antibodies including anti-MHC (MF20), anti-troponin I (C-19, Santa Cruz

Myosin

Noggin10 day

Control10 day

Troponin I Actinin MLC ANP

Myosin

15 d

ay10

day

5 da

y

5 da

y

0 da

y

0 da

y

ANP

BNPMLC2vMLC2aα MHC

α Ca actinGAPDH

β MHC

Nkx2.5

Oct 3/4

GATA4

TEF1

Tbx5

MEF2C

GAPDH

Control

Noggin treatment

No treatment

mix7 days

EB formation Immunostaining

Noggin

15 d

ay10

day

15 d

ay10

day

5 da

y

5 da

y

0 da

y

0 da

y

15 d

ay10

day

Troponin I Actinin MLC ANP3,000

Cardiomyocytenumber

2,000

1,000

0

30 600400200

0

2010

0

80

100Actinin (%)

GFP(−) GFP(+)

80

60

40

20

0

1612

840

604020

0

Contro

l

GATA4

MLC ANP

Troponin 1

Noggin

Contro

l

Noggin

Contro

l

NogginControl Noggin

GFP(+) ES cell

GFP(−) ES cell

Actinin GFP TOTO3 MERGE

****

** **

**

a

b

d

g h

e f

c

Figure 3 Expression of stem cell marker,

cardiac transcription factors and cardiac

specific proteins in noggin-treated ES cells.

(a) Immunostaining for anti-MHC, anti-troponin I,

anti-ANP, anti-actinin and anti-MLC are shown.

Most of the cells in the whole embryoid bodies

were stained with cardiomyocyte-specific

antibodies. (b) Isolated cells were stained withthe same antibodies. Red represents nuclear

staining with PI. In the last immunofluorescent

photograph, ANP, MHC and nucleolus are

stained with rhodamine, FITC and DAPI,

respectively. (c) Embryoid bodies were attached

to the gelatin-coated tissue culture plate, and

stained with anti-MHC and examined for the

number of cardiomyocytes. The number of

cardiomyocytes with noggin-treated cells

was 100-fold more than the control cells.

**P o 0.01 versus control. (d) RT-PCR of

Oct3/4, and cardiac transcription factors

including Nkx2.5, GATA4, TEF1, Tbx5 and

MEF2C is shown. Noggin treatment facilitated

the extinction of Oct3/4 and accelerated the

time and degree of cardiac transcription factor

expression. (e) RT-PCR of cardiac-specific

proteins. Noggin treatment augmented their

expression. (f) Western blot analysis of cardiac-specific proteins. **P o 0.01 versus control.

(g) Cell autonomy of the noggin-treated ES cells.

GFP+ ES cells were treated with noggin, mixed

with untreated GFP– ES cells and embryoid

body formation was performed. GFP+ cells

expressed actinin, whereas GFP– cells did not.

(h) Quantitative analysis of g. ANP, atrial

natriuretic peptide; MLC, myosin light chain;

BMP, bone morphogenetic protein; MHC, myosin

heavy chain; GFP, green fluorescent protein; BNP

(brain natriuretic peptide).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Biotechnology; 1:500), anti-actinin (EA-53, Sigma; 1:800), anti-ANP

(CHEMICON; 1:100), and anti-MLC (P-18, Santa Cruz; 1:500). Bound

antibodies were visualized using a secondary antibody conjugated with

Alexa488. Nuclei were stained with 4¢,6-diamidine-2-phenylidole dihydrochlo-

ride (DAPI; Sigma Aldrich) or propidium iodide (PI, Sigma), TOTO3

(Molecular Probes). The percentage of MHC–expressing cells was quantified

using the day-12 embryoid bodies.

RT-PCR and real-time quantitative PCR. Total RNA was extracted using Trizol

reagent (GIBCO) and RT-PCR was performed as described previously28. At

least five replicates were done for each time point. The PCR primers are listed

in the Supplementary Table 1 online. Before quantitative analysis, the linear

range of the PCR cycles was measured for each gene, and the appropriate

number of PCR cycles was determined. GAPDH was used as an internal

control. For quantitative analysis of brachyury T, Nkx2.5, Tbx5, Flk1 and

GATA1 expression, cDNA was used as template in a TaqMan real-time PCR

assay using the ABI Prism 7700 sequence detection system (Applied Biosys-

tems) according to the manufacturer’s instructions. All samples were run in

triplicate. Data were normalized to GAPDH. The primers and TaqMan probe

for brachyury T, Nkx2.5, Tbx5, Flk1 and GATA1 were Mm00436877_m1,

Mm00657783_m1, Mm00803521_m1, Mm00440099_m1, and

Mm00484678_m1 (Applied Biosystems), respectively.

Western blotting. Embryoid bodies were lysed in a buffer containing 20 mmol/l

Tris-HCl (pH 7.4), 100 mmol/l NaCl, 5 mmol/l EDTA, 1.0% Triton X-100, 10%

glycerol, 0.1% SDS, 1.0% deoxycholic acid, 50 mmol/l NaF, 10 mmol/l

Na3P2O7, 1 mmol/l Na3VO4, 1 mmol/l phenylmethylsulfonyl fluoride, 10 mg/

ml aprotinin, and 10 mg/ml leupeptin. Proteins were separated on 5% to 10%

SDS-PAGE. Western blot analysis was performed as described previously29.

Rabbit polyclonal antibodies against GATA4 (Santa Cruz Biotechnology),

troponin I, MLC and ANP were used as primary antibodies, and peroxidase-

conjugated goat anti-rabbit IgG was used as a secondary antibody. Signals were

visualized with an ECL kit (Amersham).

Statistical analysis. The data were processed using StatView J-4.5 software.

Values are reported as means 7 s.d. Comparisons among values for all groups

were performed by one-way ANOVA. The Scheffe’s F test was used to

determine the level of significance. The probability level accepted for signifi-

cance was P o 0.05.


ACKNOWLEDGMENTSThis work was (partially) supported by a grant-in-aid from the 21st centuryCenter of Excellence Program of the Ministry of Education, Culture, Sports,Science and Technology, Japan to Keio University. We are grateful to H. Niwafor kindly providing ES cell line EB3 and T. Yoshizaki and Y. Okada for theirthoughtful advice and discussion.


Received 21 September 2004; accepted 30 March 2005


1. Winnier, G., Blessing, M., Labosky, P.A. & Hogan, B.L. Bone morphogenetic protein-4 isrequired for mesoderm formation and patterning in the mouse. Genes Dev. 9, 2105–2116 (1995).

2. Zhang, H. & Bradley, A. Mice deficient for BMP2 are nonviable and have defectsin amnion/chorion and cardiac development. Development 122, 2977–2986(1996).

3. Marvin, M.J., Di Rocco, G., Gardiner, A., Bush, S.M. & Lassar, A.B. Inhibition of Wntactivity induces heart formation from posterior mesoderm. Genes Dev. 15, 316–327(2001).

4. Mima, T., Ueno, H., Fischman, D.A., Williams, L.T. & Mikawa, T. Fibroblast growthfactor receptor is required for in vivo cardiac myocyte proliferation at early embryonicstages of heart development. Proc. Natl. Acad. Sci. USA 92, 467–471 (1995).

5. Sasai, Y., Lu, B., Steinbeisser, H. & De Robertis, E.M. Regulation of neural induction bythe Chd and Bmp-4 antagonistic patterning signals in Xenopus. Nature 376, 333–336(1995).

6. Lim, D.A. et al. A. Noggin antagonizes BMP signaling to create a niche for adultneurogenesis. Neuron 28, 713–726 (2000).

7. Smith, W.C. & Harland, R.M. Expression cloning of noggin, a new dorsalizing factorlocalized to the Spemann organizer in Xenopus embryos. Cell 70, 829–840 (1992).

8. McMahon, J.A. et al. Noggin-mediated antagonism of BMP signaling is required forgrowth and patterning of the neural tube and somite. Genes Dev. 12, 1438–1452(1998).

9. Kubo, A. et al. Development of definitive endoderm from embryonic stem cells inculture. Development 131, 1651–1662 (2004).

10. Sauer, H., Rahimi, G., Hescheler, J. & Wartenberg, M. Role of reactive oxygen speciesand phosphatidylinositol 3-kinase in cardiomyocyte differentiation of embryonic stemcells. FEBS Lett. 476, 218–223 (2000).

11. Behfar, A. et al. Stem cell differentiation requires a paracrine pathway in the heart.FASEB J. 16, 1558–1566 (2002).

12. Schroeder, T. et al. Recombination signal sequence-binding protein Jkappa altersmesodermal cell fate decisions by suppressing cardiomyogenesis. Proc. Natl. Acad.Sci. USA 100, 4018–4023 (2003).

13. Takahashi, T. et al. Ascorbic acid enhances differentiation of embryonic stem cells intocardiac myocytes. Circulation 107, 1912–1916 (2003).

14. Boheler, K.R. et al. Differentiation of pluripotent embryonic stem cells into cardiomyo-cytes. Circ. Res. 91, 189–201 (2002).

15. Heng, B.C., Haider, H.K., Sim, E.K., Cao, T. & Ng, S.C. Strategies for directing thedifferentiation of stem cells into the cardiomyogenic lineage in vitro. Cardiovasc. Res.62, 34–42 (2004).

16. Sachinidis, A. et al. Cardiac specific differentiation of mouse embryonic stem cells.Cardiovasc. Res. 58, 278–291 (2003).

17. Schultheiss, T.M., Burch, J.B. & Lassar, A.B. A role for bone morphogenetic proteins inthe induction of cardiac myogenesis. Genes Dev. 11, 451–462 (1997).

18. Andree, B., Duprez, D., Vorbusch, B., Arnold, H.H. & Brand, T. BMP-2 induces ectopicexpression of cardiac lineage markers and interferes with somite formation in chickenembryos. Mech. Dev. 70, 119–131 (1998).

19. Ladd, A.N., Yatskievych, T.A. & Antin, P.B. Regulation of avian cardiac myogenesis byactivin/TGFbeta and bone morphogenetic proteins. Dev. Biol. 204, 407–419 (1998).

20. Lyons, K.M., Hogan, B.L. & Robertson, E.J. Colocalization of BMP 7 and BMP 2 RNAssuggests that these factors cooperatively mediate tissue interactions during murinedevelopment. Mech. Dev. 50, 71–83 (1995).

21. Lim, D.A. et al. Noggin antagonizes BMP signaling to create a niche for adultneurogenesis. Neuron 28, 713–726 (2000).

22. Smith, W.C., Knecht, A.K., Wu, M. & Harland, R.M. Secreted noggin protein mimicsthe Spemann organizer in dorsalizing Xenopus mesoderm. Nature 361, 547–549(1993).

23. Lamb, T.M. et al. Neural induction by the secreted polypeptide noggin. Science 262,713–718 (1993).

24. Zimmerman, L.B., De Jesus-Escobar, J.M. & Harland, R.M. The Spemann organizersignal noggin binds and inactivates bone morphogenetic protein 4. Cell 86, 599–606(1996).

25. Liem, K.F. Jr., Jessell, T.M. & Briscoe, J. Regulation of the neural patterning activity ofsonic hedgehog by secreted BMP inhibitors expressed by notochord and somites.Development 127, 4855–4866 (2000).

26. Winnier, G., Blessing, M., Labosky, P.A. & Hogan, B.L. Bone morphogenetic protein-4 isrequired for mesoderm formation and patterning in the mouse. Genes Dev. 9, 2105–2116 (1995).

27. Sasaki, H. & Hogan, B.L. Differential expression of multiple fork head related genesduring gastrulation and axial pattern formation in the mouse embryo. Development118, 47–59 (1993).

28. Niwa, H., Miyazaki, J. & Smith, A.G. Quantitative expression of Oct-3/4 definesdifferentiation, dedifferentiation or self-renewal of ES cells. Nat. Genet. 24, 372–376 (2000).

29. Nagy, A., Rossant, J., Nagy, R., Abramow-Newerly, W. & Roder, J.C. Derivation ofcompletely cell culture-derived mice from early-passage embryonic stem cells. Proc.Natl. Acad. Sci. USA 90, 8424–8428 (1993).

30. Hooper, M., Hardy, K., Handyside, A., Hunter, S. & Monk, M. HPRT-deficient (Lesch-Nyhan) mouse embryos derived from germline colonization by cultured cells. Nature326, 292–295 (1987).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Construction of lycopene-overproducing E. coli strainsby combining systematic and combinatorial geneknockout targetsHal Alper1, Kohei Miyaoku1,2 & Gregory Stephanopoulos1

Identification of genes that affect the product accumulation

phenotype of recombinant strains is an important problem in

industrial strain construction and a central tenet of metabolic

engineering. We have used systematic (model-based) and

combinatorial (transposon-based) methods to identify gene

knockout targets that increase lycopene biosynthesis in strains

of Escherichia coli. We show that these two search strategies

yield two distinct gene sets, which affect product synthesis

either through an increase in precursor availability or

through (largely unknown) kinetic or regulatory mechanisms,

respectively. Exhaustive exploration of all possible

combinations of the above gene sets yielded a unique set of

64 knockout strains spanning the metabolic landscape of

systematic and combinatorial gene knockout targets. This

included a global maximum strain exhibiting an 8.5-fold

product increase over recombinant K12 wild type and a

twofold increase over the engineered parental strain. These

results were further validated in controlled culture conditions.

Optimization of metabolic phenotype often requires the simultaneousrerouting of metabolic intermediates and rewiring of regulatorynetworks. In prior work, this optimization has been accomplishedby the modification of genes with well-defined structural or regulatoryroles in the context of the particular metabolic pathway beingconsidered1–3. Distant genes affecting a metabolic phenotype eitherthrough redistribution of metabolite precursors or indirect kinetic andglobal regulatory effects have been particularly challenging to identify.Models are relatively ineffective in the search for such genes becauseof their inability to capture the genes’ complex, nonlinear kinetic andregulatory interactions. In general, methods for identifying genetictargets are not as powerful as the molecular biological tools that areeffectively used to modify such targets. These issues become moreinvolved when one considers the possibility of multiple gene modula-tions4. In general, the complex nature of the metabolic landscaperaises significant challenges in the development of an optimal searchstrategy because varying genetic backgrounds and culturing condi-tions have a profound impact on the type of gene targets identified byvarious strategies.

Recently, we reported on a method for the rational design of strainsthat identifies single and multiple gene knockout targets based on aglobal stoichiometric analysis. The method was applied successfully toincrease lycopene production in recombinant strains of Escherichiacoli5. Lycopene production was investigated in the context of thenonmevalonate6 pathway in which cells are recombinant, expressingthe crtEBI operon to encode for the polymerization into the 40-carbonmolecule product. The pre-engineered strain used for the studycontained chromosomal overexpressions of dxs, idi and ispFD5

(Fig. 1a). There has been a significant effort to specifically engineerthe isoprenoid pathway and downstream genes7–13; however, in theprevious study5 and this current one, we investigate genome-widegene knockout targets. A total of seven single and multiple stoichio-metric gene deletions, (DgdhA, DaceE, DytjC (gpmB), DfdhF, DgdhADaceE, DgdhA DytjC, DgdhA DaceE DfdhF), were predicted andexperimentally validated to increase lycopene production throughincreasing the supply of precursors and cofactors that are importantin the lycopene pathway5. These seven mutations along with theparental strain comprise the set of eight systematically designedgenotypes. The left panel of Figure 1b depicts the methodology foridentifying these systematic gene knockout targets.

Lycopene production in these systematically identified knockoutstrains was still below the stoichiometric maximum, presumablylimited by unknown kinetic or regulatory factors that are unaccountedfor in stoichiometric models. To identify additional knockout targetsthat affect the lycopene phenotype via regulatory, kinetic or otherunknown mechanisms, we undertook a global transposon librarysearch in the background of the pre-engineered parental strain.Screening this transposon library on glucose plates identified threegene targets that correlated with lycopene overproduction. Uponsequencing, these combinatorial targets were identified as rssB (alsoknown as hnr), yjfP and yjiD. In the case of yjiD, the transposon wasfound to be inserted between the identified promoter region and thegene for yjiD and will henceforth be referred to as DpyjiD. The rightpanel of Figure 1b shows the identity and annotated function of theseselected gene targets along a representative location of the transposoninsertion event. We note that none of the previously identified singlestoichiometric genes surfaced in the combinatorial transposon search


1Department of Chemical Engineering, Massachusetts Institute of Technology, Room 56-469, Cambridge, Massachusetts 02139, USA. 2On leave from MitsubishiChemical Corporation. Correspondence should be addressed to G.S. ([email protected]).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

because of the relatively high threshold of the lycopene accumulationlevel imposed in the selection of candidate strains. Using these threeidentified targets, it is possible to create a total of seven genecombinations of single, double and triple combinatorial target muta-tions (DrssB, DyjfP, DPyjiD, DrssB DyjfP, DrssB DPyjiD, DyjfP DPyjiDand DrssB DyjfP DPyjiD). These seven combinations along with theparental strain constitute the combinatorial strain set comprising atotal of eight strains.

The previous results point to two distinct sets of stoichiometric andcombinatorial gene targets. It is not clear how these targets interactwhen combined. To answer this question, we conducted an exhaustivestudy of the 64 strains comprising all combinations of the eightstoichiometric and eight combinatorial genotypes. These target geneswere modified in the background of the pre-engineered recombinantE. coli strain. The resulting production profiles over the course ofa 48-hour shake-flask fermentation process provided the infor-mation needed for the complete mapping of the lycopene metaboliclandscape (Fig. 2).

Several interesting observations arise from the topology of thismetabolic landscape. First, two global maxima exist, each with

production levels around 11,000 p.p.m. (mg/g dry cell weight). Thefirst strain contains the DgdhA DaceE DfdhF genotype, which is apurely stoichiometrically designed strain. The other maximum isDgdhA DaceE DPyjiD, which is created through the combination ofstoichiometric and combinatorial targets. Second, several local max-imum points are present with production levels ranging from 8,400 to9,400 p.p.m., each formed from the combination of systematic andcombinatorial targets. Third, the left quadrant of the graph indicatesthat the combination or stacking of more than one combinatorialknockout target greatly reduces lycopene levels to below 2,000 p.p.m.,and as low as only 500 p.p.m. for some constructs, which is below theproduction level of a recombinant wild-type E. coli K12 strain. Finally,visual inspection of this landscape suggests a highly nonlinear functionwith many local optima.

Clustering methods have been routinely applied to the analysis ofmicroarray (and other) data to determine sets of genes that exhibitsimilar expression profiles14. Likewise, the technique of hierarchicalclustering may be applied to the metabolic landscape of Figure 2 tocluster gene knockout constructs exhibiting similar production pro-files over the four time points. Presumably, strains clustering most

Pyruvate

Glyceraldehyde3-phosphate

DXP

DMP P

IPP P

dxsidi

isp

genes

Chromosome-based genes Plasmid-based genes

Lycopene

crtEBI

crtE

crtI

crtB

CmR

pAC-LYC

Parental strainoverexpressions

Product = f(stoichiometry, kinetics, regulation)

S • v = b

XX

XXXX

XX

XXXX

Formate dehydrogenase HfdhF

Phosphoglucomutase IIytjC(gpmB)

Pyruvate dehydrogenaseaceE

Glutamate dehydrogenase gdhA

FunctionGene

gdhA

gpm

a/gp

mB

aceE fd

hF talB

Gene targets identified through stoichiometric modeling:

Member, two-component systems, σS degradation

130-aa hypothetical protein

249-aa hypothetical protein

Gene targets identified through combinatorial methods:

rssB hnr)

fP

D

σ70 Promoter

rssB

yjfP

yjiD

XX

XX

XX

XX

XXXX

XX

Transposonlibrary

IS10R IS10R

XX

XX

XX

XX

XXXX

Selection or

XX

KanR

AmpRtnp lacIq

Gene knockout target identification

Systematic Combinatorial

screening

Transposon site

Transposon site

Transposon site

Wild type

∆gdha ∆aceE

∆ gdha

Knockoutbackground

a

b

Figure 1 Systematic and combinatorial gene

knockout target identification. (a) Lycopene

synthesis begins with the condensation of the

key glycolytic intermediates, glyceraldehyde

3-phosphate and pyruvate and continues in a

nearly linear pathway. In the engineered strain

used in this study, the idi, ispFD, and dxs genes

are overexpressed by chromosomal promoter

replacement. To produce lycopene, a cluster

of genes, crtEBI are expressed on a plasmid.

(b) Systematic targets (illustrated on the left)

were identified through the use of global,stoichiometric modeling (as more comprehensive

models are unavailable) to identify gene

knockouts which were predicted in silico to

increase lycopene by increasing either cofactor

or precursor supply5. Combinatorial targets

(illustrated on the right) were identified through

the use of transposon mutagenesis. The gene

rssB is a response regulator responsible for

recruiting the proteolysis of the stationary

phase sigma factor, sS (encoded by rpoS)21,22,

which has previously been implicated in the

overproduction of carotenoids23. The gene

yjfP is a 249–amino acid protein which is

currently not annotated, but has been putatively

categorized as either a nonpeptidase homolog24

or as a putative hydrolase (1st module)25.

Finally, yjiD is a 130–amino acid protein with an

unknown function25. For only the yjiD mutants,

the transposon site was only found between thepromoter region and the gene. These targets were

combined to create the unique set of 64 mutant

strains used in this study.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

closely accumulate product by following similar modes of action inthe mechanism of lycopene production. Upon clustering the entire setof 64 strains, two distinct organizations emerge for the two sets ofgene targets previously identified.

Clustering lycopene profiles (across the four time points) for theeight stoichiometric knockout strains revealed a fairly close, stackeddendrogram (see abscissa of Fig. 3a). When these strains are plottedagainst the lycopene accumulation level, they reveal an expanding

concentric bubble-plot suggesting an additive effect of accumulatinggene deletions. This is in concert with the presumed mode of action inthese strains, namely the increasing availability of precursors andcofactors that are needed for lycopene biosynthesis.

In contrast to Figure 3a, all combinatorial targets, as exemplified byrssB, force a split-tree shape in the dendrogram when performed in thebackground of each of the seven stoichiometric targets (Fig. 3b).Different time courses in lycopene accumulation suggest differentmodes of action for the effect of the combinatorial genes on thisphenotype. Specifically, whereas each construct formed from thedeletion of a single combinatorial target gene tends to exhibit similarbehavior (increased production), the deletion of combinations of thesegenes yields phenotypes that are neither linear nor synergistic. In fact,double and triple knockout constructs arising from these combinatorialtargets exhibit vastly different production profiles from the individualtargets (Fig. 2). This nonlinearity suggests that the combinatorialtargets are disrupting regulatory processes that are relatively incompa-tible, and in certain cases deleterious, when combined.

Biological differences are observed when combinatorial genesare deleted together with stoichiometric ones. Strains in clusterY (Fig. 3b) all exhibit an extended lag phase, which extends to16–18 h before reaching a typical cell density OD 3.5–4.0. In contrast,strains in cluster Z do not posses such a lag phase and exhibit a steadyincrease of lycopene production with time. The average, scaled

0

2,000

4,000

6,000

8,000

10,000

12,000

Paren

talac

eE

gdhA

-ace

EytjCgd

hAfdhF

gdhA

-ytjC

gdhA

-ace

E-fdhF

NonerssByjfPyjiD

yjfP-yjiDrssB-yjiDrssB-yjfP

rssB-yjfP-yjiD

Maximum lycopene production

0 p.p.m.2,000 p.p.m.4,000 p.p.m.6,000 p.p.m.

8,000 p.p.m.10,000 p.p.m.12,000 p.p.m.

Systematic knockouts Combinatorial k

nockouts

Lyco

pene

(p.

p.m

.)

5,000

6,000

7,000

8,000

9,000

10,000

11,000

12,000

5,000

6,000

7,000

8,000

9,000

10,000

11,000

12,000

Systematic targets

Par

enta

l

∆ace

E

∆gdh

A∆a

ceE

∆ytjC

∆gdh

A∆y

tjC

∆gdh

A∆a

ceE

∆fdh

F

∆gdh

A

∆fdh

F

Max

. lyc

open

e (p

.p.m

.)

Max

. lyc

open

e (p

.p.m

.)

Cluster X

Added rssb knockout

Par

enta

l

∆ace

E

∆gdh

A∆a

ceE

∆ytjC

∆gdh

A

∆fdh

F

∆gdh

A∆y

tjC

∆gdh

A∆a

ceE

∆fdh

F

Cluster Y Cluster Z Time (h)0 10 20 30 40 50

Frac

tion

of m

axim

um y

ield

0.0

0.2

0.4

0.6

0.8

1.0

Average profile of the eight strains in cluster X

Average profile of the three strains in cluster Y

Average profile of the five strains in cluster Z

Bars represent the variability exhibited between the different strains in each cluster

a b c

Figure 3 Clustering analysis depicting the interaction of systematic and combinatorial targets. Lycopene production profiles across the 48-h shake-flask

fermentation are clustered, resulting in the dendrograms illustrated. (a) The purely systematic strains have a stacked dendrogram which is visually illustratedwith a concentric bubble plot. Strains that are more tightly clustered have similar modes of action, thus all systematic strains seem to be additive in nature.

This is further evidenced by the close clustering of DfdhF and the parental strain, as the fdhF single knockout was determined from the stoichiometric

analysis to bring about no enhancement of lycopene production. (b) Conversely, the addition of any combinatorial genotype, rssB in this case, decouples the

systematic design and causes a disjoint pattern in the dendrogram and bubble plot. This has the implication that local, metabolic gene targets are more

accessible through a sequential search than global, regulatory targets, which require a simultaneous search that is sensitive to the genetic background of the

strain. c compares the average, relative production profiles for the three clusters shown in a and b. The biological differences in the production profiles for

each of these clusters are evident.

Figure 2 Visualization of the metabolic landscape. Eight systematically

derived knockout genotypes were combined with eight systematically

derived genotypes to create a unique collection of 64 strains. The maximum

lycopene production (in p.p.m.) during the course of a 48-h shake-flask

fermentation is plotted. Among the interesting features of this landscape is

the presence of two global maxima (at around 11,000 p.p.m.) and several

local maxima. Furthermore, certain combinations of combinatorial targets in

most systematically derived genetic backgrounds resulted in a substantialdecrease in lycopene production.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

production profiles for the purely systematic cluster and the twoclusters forced by an rssB deletion are compared in Figure 3c. It isnoted that this branched pattern is exhibited by all strains constructedfrom the deletion of any combinatorial gene in the background of thestoichiometric targets, with different production profiles characteriz-ing each of the clusters.

Drawing from this analysis, it appears that stacking (that is,deleting) combinatorial target genes upon stoichiometric ones leadsto a decoupling of the stoichiometric logic. This decoupling is evidentin analyzing the impact of the deletion of rssB or any other combi-natorial gene, on the shape of the dendrogram obtained fromhierarchical clustering of the lycopene accumulation profiles for theeight stoichiometric strains; it is also quantified by covariance analysis(Supplementary Fig. 1 and Supplementary Discussion).

The exhaustive exploration of the combinations of stoichiometricand combinatorial targets allowed the identification of several inter-esting strains on the basis of their performance in batch shake-flaskcultivations. To better assess the production capacity of these knock-out strains, fed-batch cultivations were carried out in shake-flasks andcontrolled bioreactors with staged glucose feed (Fig. 4). Several strainswere thus evaluated. Optimized shake-flask fermentations highlightthe capability of the global maximum strains to produce upwards of18,000 p.p.m. in 24–40 h (Fig. 4). These global maximum strains werealso grown in 500-ml bioreactors with a similar glucose feeding profileand pH control and showed enhanced lycopene production producingupwards of 23,000 p.p.m. in only 60 h (data not shown). Furtherimprovements are possible through iterative bioreactor optimization.The good correspondence between fermentor and shaker-flasks resultssuggests that the performance of the strains selected by the describedmethod is transferable to larger systems.

Identification of multiple gene targets affecting a particular pheno-type is an open problem. Among the complications are strong non-linear effects, lack of accurate models capable of capturing geneticinteractions and ineffective search strategies. To address these issues inthe context of lycopene production, we undertook an exhaustiveexperimental search to investigate combinations of rationally selectedgenes with those identified through combinatorial methods. A numberof promising strains were obtained, some of which were capable ofproducing upwards of 18,000 p.p.m. (or 18 mg/g dry cell weight) oflycopene in defined glucose medium using simple fed-batch conditions.This value represents a nearly fourfold increase over the parental strainwhen cultured in simple cultivation, a twofold increase over the pre-engineered parental strain in similar conditions and an 8.5-fold increaseover recombinant wild-type K12 E. coli under similar conditions.

The metabolic landscape defined through this unique set of 64knockout strains allows for several observations of importance tometabolic engineering. First, rationally selected stoichiometric geneknockout targets have the potential of generating serious contendersin the quest for maximally producing strains. We note that one ofthe two maximum overproducing strains resulted from the knockoutof three stoichiometric genes (gdhA, aceE, fdhF). Additionally, theknockout of specific combinatorial genes yielded substantiallyenhanced phenotypes in the background of particular stoichiometricknockout genes. Second, whereas combinatorial gene targets holdgreater potential than stoichiometric ones as single knockout mutants,multiple knockouts of the combinatorial gene set led to a distinctdeterioration of the lycopene phenotype. Yet, it proved invaluable inthe creation of some important strains in the landscape. Third, thepresence of many local maxima complicates the nature of the land-scape and raises questions about general sequential search strategies.Previously, sequential search strategies were found to be quite effective

when applied to the space of stoichiometric genes5, which is due totheir overall additive effect on phenotype. Figure 2 suggests that thisresult does not hold when combinatorial genes are also included in thesearch space, necessitating exhaustive combinatorial searches of thetype undertaken in this study. Although identification of optimal genetargets will continue to be a demanding undertaking, searches for genetargets will be significantly aided by advanced models of cell functionaccounting for kinetic and regulatory mechanisms.

It should be noted that the search of this study was limited to theeffect of gene knockout only. Gene knockdown or overexpression addsan extra layer of complexity in the metabolic engineering of over-producing strains and could provide further drastic improvements ofproduct overproduction phenotypes.

This study underscores some important issues optimizing pheno-type. First, high-throughput screening methods combined withdetailed cellular models will aid in efficient strain optimization.Second, combinatorial targets influencing global cellular functionshould be invoked at later stages in the strain improvement processto avoid selecting those with limited utility or incompatible modes ofaction. Finally, metabolic genes seem to have a linear impact in theoverall cellular phenotype whereas the effect of regulatory targets isdefinitely nonlinear and more complex. This work serves as a casestudy aiming to understand the complex interaction of the genotype-phenotype space in the context of product overproduction phenotype.The lessons gained from the exhaustive exploration of systematic andcombinatorial gene knockout sets can help shape future strainimprovement programs as they are tested in diverse systems fordivergent products.

METHODSStrains and media. E. coli K12 PT5-dxs, PT5-idi, PT5-ispFD, provided by

DuPont, was used as the lycopene expression strain when harboring the pAC-

LYC plasmid containing the crtEBI operon15. Overexpression of dxs, idi, and

Lyco

pene

(p.

p.m

.)

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

15 h24 h40 h47 h

K12,Wild-type

Isoprenoidpathway

modification

Globaloptimumstrains

Globalminimum

strain

Time (h)0 10 20 30 40

Glu

cose

(g/

l)

0

1

2

3

4

5Glucose feeding profile

Local maximum

strains

Figure 4 Behavior of selected strains in optimized culturing conditions.

Selected strains from the metabolic landscape were cultured in fed-batch

shake-flasks with increased M9 salts and a staged glucose feed as

represented in the glucose feeding profile. Strains presented from left to

right are K12 (recombinant wild-type); engineered parental strain with dxs,

idi, and ispFD overexpressions; DgdhA DaceE DfdhF; DgdhA DaceE DPyjiD;

DgdhA DaceE DfdhF DrssB DyjfP; DgdhA DaceE DfdhF DyjfP; DgdhA DaceE

DrssB DyjfP DPyjiD. The two global maxima were capable of producing

upwards of 18,000 p.p.m. in 24 to 40 h. Strain behavior was transferable

to these optimized conditions. These results highlight that strains isolatedon solid-media plates retained the lycopene overproduction phenotype in the

course of the scale-up process.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

ispFD was chromosomally incorporated without an antibiotic marker through

promoter delivery. Strains were grown at 37 1C with 225 r.p.m. orbital shaking

in M9-minimal medium16 containing 5 g/l D-glucose and 68 mg/ml chloram-

phenicol. All simple cultures were 50 ml, grown in a 250-ml flask with an 1%

(vol/vol) inoculation from an overnight 5-ml culture and assayed at 15, 24, 39

and 48 h. Optimized shaker-flasks were 50-ml cultures grown in 250-ml flasks

with a 1% (vol/vol) inoculation from an overnight 5-ml culture with glucose

feeds of 5 g/l at 0 and 15 h and 3 g/l at 24 h. The medium for these experiments

was M9-minimal medium16 with double concentrations of all salts except

CaCl2 and MgSO4. All experiments were performed in biological knockout

replicates to validate data and calculate statistical parameters. Glucose mon-

itoring was conducted periodically using an r-Biopharm kit to verify complete

usage of glucose. Cell density was monitored spectrophotometrically at 600 nm.

All PCR products were purchased from Invitrogen and used Taq polymerase.

M9 Minimal salts were purchased from US Biological and all remaining

chemicals were from Sigma-Aldrich.

Transposon library screening and sequencing. Transposon libraries were

generated using the pJA1 vector17. Cells were transformed with between 800

and 1,600 ng of the plasmid, then diluted and plated on M9-glucose-agar plates

(containing 1 mM isopropyl-b-D-thiogalactoside) with a target density of 200

colonies per 150 � 15 mm Petri dishes. Plates were incubated at 37 1C for 36 h,

then allowed to sit at 22 1C. Cells identified as exhibiting increased lycopene

content (more red) were isolated and cultured throughout the culturing

process. The identity of promising targets were sequenced using an altered

version of Thermal Asymmetric Interlaced PCR (TAIL-PCR)18. For the TAIL1

reaction, 1.5 ml of genomic DNA isolated using a DNA purification kit

(Promega) was used as the initial template. The TAIL3 reaction was increased

to 30 cycles. Kanamycin-specific primers: TAIL1, 5¢-TATCAGGACATAG

CGTTGGCTACCCG-3¢; TAIL2, 5¢-CGGCGAATGGGCTGACCGCT-3¢; TAIL3,

5¢-TCGTGCTTTACGGTATCGCCGCTC-3¢. The degenerate primer AD1 was

used as described in the reference. The product of the TAIL3 reaction was

purified by a PCR cleanup kit (Qiagen) after gel visualization. This product was

sequenced using the primer TAIL-seq, 5¢-CATCGCCTTCTATCGCCTTCTT-3¢.Gene target identity was determined through BLAST nucleotide sequence

comparison. Strains identified through transposon mutagenesis were subse-

quently constructed by using PCR product recombination and tested for

maintenance of the lycopene overproduction phenotype.

Knockout construction and verification. Gene deletions were conducted using

PCR product recombination19 using the pKD46 plasmid expressing the lambda

red recombination system and pKD13 as the template for PCR (see Supple-

mentary Table 1 online for primer designs). Gene knockouts were verified

through colony PCR. Phage transduction was used for creating multiple gene

knockout strains. P1vir phage transduction was used to transfer knockout

mutants between strains20. PCR primers used for knockout and verification

may be found in Supplementary Table 1 online.

Lycopene assay. Intracellular lycopene content was extracted from 1 ml of

bacterial culture at the point of total glucose exhaustion. The cell pellet was

washed, and then extracted in 1 ml of acetone at 55 1C for 15 min with

intermittent vortexing. The lycopene content in the supernatant was quantified

through absorbance at 475 nm12 and concentrations were calculated through a

standard curve. The entire extraction process was performed in reduced light

conditions to prevent photobleaching and degradation. Cell mass was calcu-

lated by correlating dry cell with OD600 for use in p.p.m. (mg lycopene/g dry

cell weight) calculations.

Hierarchical clustering routines. A complete linkage hierarchical clustering of

the lycopene time profiles for the entire 8 � 8 strain matrix (containing values

of the maximum lycopene production) using the Euclidean distance as the

similarity metric was performed using Cluster Version 3.0. Dendrograms were

visualized using Java TreeView Version 1.0.8.


ACKNOWLEDGMENTSWe acknowledge financial support of this work by the DuPont-MIT Alliance.In particular, we would like to thank Wonchul Suh for providing the parentalE. coli strain. We also thank Joel Moxley for providing thoughtful suggestionsand Veronica Godoy for providing the initial phage stock.


Received 10 February; accepted 2 March 2005


1. Stephanopoulos, G., Aristidou, A. & Nielsen, J. Metabolic Engineering: Principles andMethodologies (Academic Press, San Diego, CA, 1998).

2. Ostergaard, S., Olsson, L., Johnston, M. & Nielsen, J. Increasing galactose consumptionby Saccharomyces cerevisiae through metabolic engineering of the GAL gene regulatorynetwork. Nat. Biotechnol. 18, 1283–1286 (2000).

3. Stafford, D.E. et al. Optimizing bioconversion pathways through systems analysis andmetabolic engineering. Proc. Natl. Acad. Sci. USA 99, 1801–1806 (2002).

4. Koffas, M.A., Jung, G.Y. & Stephanopoulos, G. Engineering metabolism and productformation in Corynebacterium glutamicum by coordinated gene overexpression. Metab.Eng. 5, 32–41 (2003).

5. Alper, H., Jin, Y.-S., Moxley, J. & Stephanopoulos, G. Identifying gene targets for themetabolic engineering of lycopene biosynthesis in Escherichia coli. Metab. Eng. (in thepress) doi:10.1016/j.ymben.2004.12.003 (2005).

6. Adam, P. et al. Biosynthesis of terpenes: studies on 1-hydroxy-2-methyl-2-(E)-butenyl4-diphosphate reductase. Proc. Natl. Acad. Sci. USA 99, 12108–12113 (2002).

7. Matthews, P.D. & Wurtzel, E.T. Metabolic engineering of carotenoid accumulation inEscherichia coli by modulation of the isoprenoid precursor pool with expression ofdeoxyxylulose phosphate synthase. Appl. Microbiol. Biotechnol. 53, 396–400 (2000).

8. Misawa, N. & Shimada, H. Metabolic engineering for the production of carotenoids innon-carotenogenic bacteria and yeasts. J. Biotechnol. 59, 169–181 (1997).

9. Farmer, W.R. & Liao, J.C. Improving lycopene production in Escherichia coli byengineering metabolic control. Nat. Biotechnol. 18, 533–537 (2000).

10. Smolke, C.D., Martin, V.J.J. & Keasling, J.D. Controlling the metabolic flux through thecarotenoid pathway using directed mRNA processing and stabilization. Metab. Eng. 3,313–321 (2001).

11. Lee, P.C. & Schmidt-Dannert, C. Metabolic engineering towards biotechnologicalproduction of carotenoids in microorganisms. Appl. Microbiol. Biotechnol. 60, 1–11(2002).

12. Kim, S.-W. & Keasling, J.D. Metabolic engineering of the nonmevalonate isopentenyldiphosphate synthesis pathway in Escherichia coli enhances lycopene production.Biotechnol. Bioeng. 72, 408–415 (2001).

13. Jones, K.L., Kim, S.-W. & Keasling, J.D. Low-copy plasmids can perform as well as orbetter than high-copy plasmids for metabolic engineering of bacteria. Metab. Eng. 2,328–338 (2000).

14. Eisen, M., Spellman, P., Brown, P. & Botstein, D. Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. USA 95, 14863–14868 (1998).

15. Cunningham, F.X., Jr., Sun, Z., Chamovitz, D., Hirschberg, J. & Gantt, E. Molecularstructure and enzymatic function of lycopene cyclase from the cyanobacteriumSynechococcus sp strain PCC7942. Plant Cell 6, 1107–1121 (1994).

16. Maniatis, T., Fritsch, E.F. & Sambrook, J. Molecular Cloning: A Laboratory Manual (ColdSpring Harbor Laboratory Press, Cold Spring Harbor, NY, 1982).

17. Badarinarayana, V. et al. Selection analyses of insertional mutants using subgenic-resolution arrays. Nat. Biotechnol. 19, 1060–1065 (2001).

18. Liu, Y.-G. & Whittier, R.F. Thermal asymmetric interlaced pcr: automatable amplifica-tion and sequencing of insert end fragments from pi and yac clones for chromosomewalking. Genomics 25, 674–681 (1995).

19. Datsenko, K.A. & Wanner, B.L. One-step inactivation of chromosomal genes inEscherichia coli K-12 using PCR products. Proc. Natl. Acad. Sci. USA 97, 6640–6645 (2000).

20. Miller, J.H. A Short Course in Bacterial Genetics (Cold Springs Harbor Laboratory Press,Cold Springs Harbor, NY, 1992).

21. Muffler, A., Fischer, D., Altuvia, S., Storz, G. & Hengge-Aronis, R. The responseregulator RssB controls stability of the sigma(S) subunit of RNA polymerase inEscherichia coli. EMBO J. 15, 1333–1339 (1996).

22. Sandmann, G., Woods, W. & Tuveson, R.W. Identification of carotenoids in Erwiniaherbicola and in a transformed Escherichia coli strain. FEMS Microbiol. Lett. 59,77–82 (1990).

23. Becker-Hapak, M., Troxtel, E., Hoerter, J. & Eisenstark, A. RpoS dependent over-expression of carotenoids from Erwinia herbicola in OXYR-deficient Escherichia coli.Biochem. Biophys. Res. Commun. 239, 305–309 (1997).

24. Rawlings, N., Tolle, D. & Barrett, A. MEROPS: the peptidase database. Nucleic AcidsRes. 32, D160–D164 (2004).

25. Serres, M.H. et al. A functional update of the Escherichia coli K-12 genome.Genome Biol. 2, published online 20 August 2001 (doi:10.1186/gb-2001-2-9-research0035).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Quantitative mouse brain proteomics usingculture-derived isotope tags as internal standardsYasushi Ishihama1,2, Toshitaka Sato1,2, Tsuyoshi Tabata1, Norimasa Miyamoto1, Koji Sagane1, Takeshi Nagasu1

& Yoshiya Oda1

An important challenge for proteomics is to be able to

compare absolute protein levels across biological samples1,2.

Here we introduce an approach based on the use of culture-

derived isotope tags (CDITs) for quantitative tissue proteome

analysis. We cultured Neuro2A cells in a stable isotope-

enriched medium and mixed them with mouse brain samples

to serve as internal standards. Using CDITs, we identified and

quantified a total of 1,000 proteins, 97–98% of which were

expressed in both mouse whole brain and Neuro2A cells.

CDITs also allow comprehensive and absolute protein

quantification. Synthetic unlabeled peptides were used to

quantify the corresponding proteins labeled with stable

isotopes in Neuro2A cells, and the results were used to obtain

the absolute amounts of 103 proteins in mouse whole brain.

The expression levels correlated well with those in Neuro2A

cells. Thus, the use of CDITs allows both relative and absolute

quantitative proteome studies.

Relative expression levels of cellular proteins under different condi-tions, as well as protein identities, are routinely determined by meansof mass spectrometry (MS) in conjunction with stable isotopiclabeling of either proteins or proteolytic peptides2. MS-based quanti-fication is performed by labeling one sample with a light isotope and

designating it as the reference, and subsequently labeling all othersamples with a heavy isotope. Quantitative proteomic strategies haveproven particularly advantageous for the discrimination of targetproteins from contaminants copurified nonspecifically3–5. In vivolabeling strategies for MS-based proteomics entail growing cells in amedium in which an essential nutrient is labeled with a stable isotope.Quantitative data are obtained after analysis of amounts of digestedpeptide6 or proteins7. The advantages of this approach are simplifiedsample manipulation and comprehensive labeling. Recently, an in vivolabeling method for mammals by long-term administration of a dietenriched in a stable isotope was reported8. Although this is an

Ratio 1 = green/blue Ratio 2 = red/blue

Tissue 1/tissue 2 = green/red = ratio 1/ratio 2

Tissue 1 Tissue 2Isotope-labeled cells(CDIT cells)

m/zm/z m/zm/z

Combine

Extract/separate

Digest proteins

Analyze by MSCompare

m/zm/z

Combine

Extract/separate

Digest proteins

Analyze by MS

Compare

From tissue 1 From tissue 2

LL

H(H)

L

L

H

(H)

a

b

Figure 1 Strategy of quantitative mouse brain proteomics using CDITs.

(a) Quantitative tissue proteome analysis using stable isotope-labeled

cultured cells as global internal standards. Tissue samples 1 and 2 are mixed

with cultured cells early in the process to obviate the variations during

sample preparation. After protein extraction and separation, digested proteins

are analyzed by mass spectrometry to identify and quantify proteins. The ratio

between the two isotopic distributions (one from a tissue sample and one

from cultured cells labeled with isotopes) can then be determined from the

mass spectra. Changes of protein level in two tissue samples are estimated

by calculating the ratio of the two ratios, ratio 1/ratio 2, a procedure which

cancels out the internal standards (cultured cells). (b) Method for semi-

quantifying a protein found in a tissue sample, but not in the cells cultured

with stable isotopes. The ratio of a target peptide, which does not have a

corresponding labeled peak in cultured cells, is obtained by using the peak

ratio against an isotope-labeled, cultured-cell-derived peptide of different

sequence, but with the closest (ideally the same) retention time in LC/MS.


1Laboratory of Seeds Finding Technology, Eisai Co., Ltd., 5-1-3 Tokodai, Tsukuba, Ibaraki 300-2635, Japan. 2These authors contributed equally to this work.Correspondence should be addressed to Y.O. ([email protected]).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

interesting approach, it takes a long time (44 d) to obtain the labeledrat, and some tissues, such as brain, are not completely labeled withstable isotopes.

We have developed an alternative quantitative approach for study-ing the proteome of mammalian tissues based on the application ofthe stable isotope labeling by amino acids in cell culture (SILAC)9–11

approach to the generation of an internal standard. We applied CDITsto quantify the mouse brain proteome by using mouse neuro-blastoma Neuro2A cells cultured in 13C-labeled leucine-rich mediumas an internal standard (Fig. 1). To validate our methodology, we usedan affinity matrix-immobilized E7070-like compound, which isknown to enrich cytosolic malate dehydrogenase (cMDH)3, to deter-mine levels of cMDH. We diluted 0.1 ml of the soluble fraction of amixture of wild-type brain with an appropriate amount of labeledNeuro2A cells in 0.9 ml of PBS, and the solution was loaded onto anE7070-immobilized column. The column was washed with 1 Msodium chloride in PBS, and cMDH was eluted with 10 mMNADH in PBS. The eluted fraction was concentrated by ultrafiltration,and then cMDH was separated on SDS-PAGE. The cMDH band wasexcised and in-gel-digested with trypsin. We also prepared ADAM22(ref. 12) knockout mouse brain samples mixed with Neuro2A cells,and purified cMDH from them. We repeated the affinity purificationprocedures five times (total of ten samples, five wild-type and fiveADAM22 knockout samples). Tryptic peaks of cMDH were measuredwith matrix-assisted laser desorption ionization (MALDI)-MS, andpeaks due to leucine-containing peptides from brain cMDH werecompared with those from Neuro2A cells. Although affinity purifica-tion steps are usually variable, precision was extremely good (thecoefficient of variation was 4.11%; see Supplementary Table 1online). Next, the amount of brain tissue was changed to confirmthe linearity of the method. Peak intensity ratios of cMDH were foundto be linearly correlated to relative amount (R ¼ 0.994) in the rangefrom 1:20 to 5:1 (brain cMDH versus Neuro2A cMDH, see Supple-mentary Table 2 online). To confirm the suitability of Neuro2A cellsas internal standard cells for the mouse brain proteome, we identifiedmany proteins extracted from Neuro2A cells and mouse whole brain.A total of 957 different proteins identified by leucine-containingpeptides were expressed in mouse whole brain. Among them, 14proteins in the brain were not found in Neuro2A cells. There was 98%or more overlap among the 1,000 proteins in mouse whole brain andNeuro2A cells (Fig. 2a).

For a larger scale validation experiment, 44 different proteins inwild-type mouse brains were carefully quantified by manually mea-suring peak ratios with the corresponding proteins in Neuro2A cells.The observed quantitative value was very close to the expected value(see Fig. 2b and part 1 of Supplementary Table 3 online). Since theCDIT approach allows the calculation of a ratio of ratios to obtainrelative quantitative values, variation in the assessment of MS spectrais minimized by cancellation of systematic errors of internal standardintensities13. Since the dynamic range of MS detector is generally quitenarrow, two or three different amounts of Neuro2A should be addedinto brain samples to increase the number of quantified proteins (seepart 2 of Supplementary Table 3 online). When the extracts fromNeuro2A were equivalently added to brain extracts, two-thirds ofidentified proteins in wild-type mouse whole brains had an appro-priate expression level (0.1 o peak ratio o 10) in Neuro2A cells.

Although in vivo labeling has advantages, chemical tagging strate-gies like isotope-coded affinity tags (ICATs)14 allow for quantitativetissue proteomics. Therefore we compared the performance of clea-vable ICATs with the CDIT approach by using the same amount ofmouse brain extracts. The major innovation of the ICAT approach wasto purify cysteine-containing peptides by using an affinity tag forreducing the complexity of a peptide mixture by about a factor of 10.But our results indicate that the CDIT approach allows the efficientidentification and quantification of more proteins (602 proteins, theaverage ratio was 0.99 7 0.41) than the ICAT method (339 proteins,the average ratio was 1.26 7 0.33) (Fig. 2c and Supplementary Table 4online), suggesting the applicability of the method in the context ofreal world complex samples. Considering the inefficiency of ICAT,there might be two reasons for the difference. First, time-consumingand variable steps are required to attach chemical tags and removeexcess reagents, which can lead to sample losses2,15; second, thepresence of chemical tags like ICAT makes tandem MS (MS/MS)spectra complicated and thus peak identification becomes difficult3,16.

Although an isotope-labeled peptide with the same sequence as thetarget was found for almost every sequence in this analysis, it is likelythat some proteins found in brain, especially less abundant proteins,are not expressed in Neuro2A cells. In this situation, the peak ratiobased on the corresponding isotope-labeled peak from Neuro2A cellscannot be calculated, so we selected an isotope-labeled peptide with adifferent sequence from Neuro2A cells as the internal standard forcalculation of the peak ratio (Fig. 1b). We calculated peak ratios for

939 414

Brain: 953 Neuro2A: 943 0

1

2

3

4

5

6

0 2 4 6

Mixing ratio (Neuro2A/brain)

Nor

mal

ized

MS

sign

al r

atio

226 113376

CDIT: 602 ICAT: 339

a b c

Figure 2 Validation data of CDIT strategy. (a) Number of identified and quantified proteins from wild-type whole mouse brain and Neuro2A cells. Aftercombining brain extracts with Neuro2A cell lysates, proteins were digested with trypsin, and then peptides were analyzed by LC/MS. Identification was

performed based on leucine-containing peptides, because leucine residues were labeled with stable isotopes and the peak ratios of those peptides were

measured for quantification. When there were no corresponding peaks from labeled Neuro2A cells, we judged that those proteins were expressed in brains,

but not in Neuro2A cells. (b) The measured average ratio of 44 different proteins in unlabeled mouse brain to those in labeled Neuro2A cells versus the

expected ratio showing the linearity and precision of the method. Neuro2A cell lysates were combined with three different amounts of whole brain extracts

in the ratios were 1:5, 1:1 and 5:1. The detailed data are in part 2 of Supplementary Table 3 online. (c) Summary of the number of proteins identified

and quantified by the CDIT method and cleavable ICAT method. Proteins were extracted from wild-type whole mouse brains. Starting protein amounts were

300 mg for CDIT, 300 mg for light ICAT labeling and 300 mg for heavy ICAT labeling. For the CDIT approach, the same amount of total proteins extracted

from labeled Neuro2A cells were added as internal standards. The detailed data are in Supplementary Table 4 online.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

155 different-sequence pairs and same-sequence pairs; the differencebetween them was 1.29 7 19.3% (see Supplementary Table 5-1online). This result showed that quantitative precision became worsewhen different-sequence peptides were selected as internal standardsinstead of same-sequence, isotope-labeled peptides. Then, the targetpeptide in a brain was normalized to the average ratio calculated byfour different sequences selected from Neuro2A at a similar elutiontime (see part 2 of Supplementary Table 5 online). The difference wasonly 1.02 7 28.9% (n ¼ 35); therefore normalization for ‘semi-quantification’ was possible by using different-sequence peptides.Indeed, in other areas such as pharmacokinetic studies there aremany examples of the successful use of internal standards with adifferent structure from that of the target molecules17–21.

We next explored whether the CDIT method is an efficient tool forexamining protein expression levels in animal tissues. The systematictreatment of animals with kainate induces generalized tonic-clonicseizures, which are due to necrosis and apoptosis of brain cells22,23.Isolated hippocampus treated with kainate was compared with non-treated counterparts. After adding Neuro2A cells, the hippocampuswas fractionated into five parts. In total, 598 proteins were identifiedand quantified (see Supplementary Table 6 online). As in the case ofwhole brain, 497% of the proteins found in mouse hippocampuswere expressed in Neuro2A cells. Although the mechanisms under-lying kainate neurotoxicity are still not well understood, the expres-sion levels of 21 different proteins were changed more than twofold(see Supplementary Table 7 online). But these relatively abundantproteins and metabolic enzymes could well be linked indirectlythrough general toxicity with kainate treatment.

Absolute concentrations of proteins in samples are also importantin addition to the relative concentrations between two samples.Conventionally, antibodies, enzymatic assays and staining dyes havebeen used to measure absolute protein amounts. Another techniquefor absolute quantification is to use MS after spiking known amountsof isotopically labeled analytes, so-called isotope dilution24 orAQUA25. One of the proteolytic peptides of a particular target proteinis synthesized using isotope-labeled reagents and then the absolute

amount is measured. However, this method is difficult to apply toquantitative analysis of large numbers of protein, because peptidesynthesis usually requires a tenfold excess of reagents (expensiveisotopically labeled reagents in this case), and also because the scaleof conventional peptide synthesis is microgram to milligram, whereasMS requires only femtogram to picogram amounts of peptides. WithCDITs, absolute amounts of target proteins in brain can be calculated,if absolute quantification of target proteins in Neuro2A cells isconducted in advance (Fig. 3a). Since all the proteins in the Neuro2Acells are already labeled with stable isotopes, conventional unlabeledsynthetic peptides can be used to identify the absolute amounts oftarget proteins. It has been found that the AQUA approach does notwork for in-gel-digested proteins because of low recovery in thedigestion or extraction step26. In our strategy, quantified syntheticunlabeled peptides and the labeled cultured cells are used in the firststep, in which cells are lysed and proteins are extracted by ultrasonica-tion without any purification step. Then, tryptic digestion is done inthe solution to maximize recovery. Because the contents of trypticpeptides from the same protein should be equivalent, the content ofthe tryptic peptide, which is quantified by matching to the unlabeledpeptide, is the same as that of other tryptic peptides from the protein.Therefore, we used all labeled peptides from the cultured cells forquantification of the unlabeled tissue sample spiked with isotope-labeled cultured cells at the second step (Fig. 3a). In addition, we canuse purification steps in addition to the in-gel digestion withoutaffecting the quantification because the internal standards are notpeptides, but proteins, unlike in the AQUA method. This approachwas applied to mouse whole brain in conjunction with CDIT-labeledcells. We successfully quantified 103 proteins in mouse brain, and theirexpression levels correlated well with those of Neuro2A cells (Fig. 3band Supplementary Table 8 online). The absolute amounts ofproteins in a particular organelle are not calculated from the amountsof proteins in the whole lysate of the cultured cells, but are obtainedby measuring the expressed amounts of proteins in the target organellein advance. Because the cellular fractionation is not highly reprodu-cible, the resultant amounts of proteins in the sample organelle are

11 10 100 1,000 10,000

1,000

10

Absolute quantification oftarget proteins in CDIT cells

CDIT cells(labeled)

Labeled protein

Labeled peptides

Peptidesynthesizer

Unlabeled peptide(known amount)

Trypsin

m/z

m/z

TissueCDIT cells(labeled & quantified)

Any peptides can be used for absolute quantification

Extract/separate/digest

Target protein

Protein amounts in Neuro2A (pmol/mg)

Pro

tein

am

ou

nts

inm

ou

se b

rain

(p

mo

l/mg

)

100

10,000a b

Figure 3 Total scheme for absolute quantification using CDITs. (a) Absolute quantification using amplified isotope double dilution. An unlabeled synthetic

peptide is used as an internal standard for a target protein expressed in CDIT cells. (b) Expression of 103 proteins in mouse cultured cell (Neuro2A) and

brain (wild type). The plot is based on data listed in Supplementary Table 8 online.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

influenced by the variations of two experiments, that is, fractionationfor the cultured cells and for the samples spiked with culturedcells. Nevertheless, this approach is very attractive compared withconventional isotope dilution methods as a comprehensive approachdue to its much lower cost. Moreover, because the amounts of targetproteins in the cultured cells are calculated, the absolute amountsof corresponding proteins in the samples can be compared at theprotein level, instead of at the peptide level, which should givebetter reliability.

In conclusion, the CDIT-based method is a simple, convenient andcost-effective approach for relative and absolute quantification of atissue proteome. The American Type Culture Collection (http://www.atcc.org/) provides more than 1,000 mammalian cell lines bytissue source, and some of them, which can grow in stable isotope richmedium, are probably useful as internal standards for their tissues oforigin, and proteins that do not have a ‘shared peptide’ can be semi-quantified by using a different-sequence peptide from labeled celllines. Finally, it is easy to obtain large numbers of cells (41010 cells)from animal tissues, but there is a limit to the scale of cell culture ingeneral laboratories. Although the large-scale preparation of CDIT-labeled cells could be a hurdle, many laboratories carrying outproteomic studies routinely handle 107–109 cells, which is the scalerequired for this method. The CDIT approach thus represents a robustalternative for tissue proteomics with the potential to provide relativeand absolute quantitative data on proteins.

METHODSMaterials and reagents. Neuro2A cells were grown to a density of 7 � 107 cells/

15-cm diameter dish in RPMI-1640 medium (Sigma) deficient in L-leucine,

and U-13C �6 labeled L-leucine (Cambridge Isotope Laboratories) was added

to the culture. TPCK-treated, sequencing-grade modified trypsin was obtained

from Promega. 5-cyclohexyl-1-pentyl-beta-D-maltoside (CYMAL-5) was

obtained from Anatrace. Negative gel stain MS kit was obtained from Wako.

Gels with a thickness of 1.0 mm (Tris-HCl, 5–20% acrylamide gradient gel)

were obtained from DRC. All other reagents were of analytical grade.

Mouse brain sample preparation. All mice were treated ethically according to

the rules of the Eisai Co., Ltd. Animal Use and Care Committee. Seventeen

male C57BL/6 mice were used at 8–16 weeks old. Kainate in saline buffer was

administered intraperitoneally (i.p.) at a dose of 30 mg/kg. The same volume of

the solution in PBS was administered i.p. to control mice. These mice were

killed 2–3 h after the injection by rapid decapitation. The brain was removed,

the hippocampus was isolated on ice, and the protein amounts were measured

by means of micro bicinchoninic acid (BCA) assay (Pierce). Each experimental

hippocampus was combined with 4 � 109 labeled Neuro2A cells and the

mixture was suspended in 0.32 M sucrose solution containing 1 mM sodium

hydrogen carbonate and protease inhibitor cocktail (Roche Diagnostics). The

suspension was homogenized in a Teflon Potter-type homogenizer and cen-

trifuged at 710g for 10 min to remove the nuclear fraction. The supernatant was

centrifuged at 13,800g for 10 min to separate the soluble fraction and insoluble

materials. The pellets were resuspended in 0.32 M sucrose solution, and the

suspension was layered on 1.2 M sucrose solution then centrifuged at 82,500g

for 2 h to separate cytosol, trafficking and secretion-related organelles, and

mitochondrial fractions. For whole brain analysis, frozen mouse brain with

added Neuro2A cells was homogenized with a protease inhibitor cocktail,

sonicated and then centrifuged at 100,000g for 1 h at 4 1C. The supernatant and

the pellet were collected as the soluble fraction and membrane-nuclear fraction,

respectively. Proteins from the pellet fraction were extracted with 8 M urea

containing 1% CHAPS and 1 M sodium chloride. Protein solutions were

concentrated with an Ultrafree-MC centrifugal filter (10,000 Da nominal

molecular weight limit (NMWL)) (Millipore) to 100 ml. After SDS-PAGE,

each lane was cut into twelve equal pieces, and in-gel digestion was carried

out27. For cleavable ICAT analysis, ICAT kits were used according to the

protocol recommended by Applied Biosystems.

Mass spectrometric analysis. The dried samples were desalted with C18 Stage-

tips28 and then redissolved in 20 ml of acetonitrile/water/TFA, 5:95:0.1, for

liquid chromatography (LC)/MS analysis using a ‘stone-arch’ column29. The

eluent was directed to an ESI ion trap mass spectrometer (ThermoFinnigan

Model LCQ) with a lab-made nano-spray ion source at a flow rate of 1 ml/min

after flow splitting, or to an ESI QqTOF mass spectrometer (Applied Biosys-

tems Model QSTAR pulsar i) with a lab-made nano-spray ion source at a flow

rate of 200 nl/min after flow splitting. A linear gradient of B from 5–30% was

run, using mobile phase A of 0.5% acetic acid and mobile phase B of 0.5%

acetic acid/acetonitrile, 20:80.

Data processing. MS/MS data were analyzed by MASCOT (Matrix Sciences)

and Sonar MS/MS (ProteoMetrics). After protein identification by MS, leucine-

containing peptides were extracted from the search results. Their m/z and scan

number information were used to extract mass chromatograms and in-house

software determined the peak areas of peptides and manual confirmation was

done to correct peak areas. Each peak was quantified relative to its correspond-

ing isotope-labeled peak from Neuro2A cells, which were used as comprehen-

sive internal standards to normalize the variations of sample preparation and

analysis. Finally the amount of each peak was compared in different tissue

samples relative to Neuro2A cells.

Procedures for absolute quantification. Proteins from Neuro2A cells labeled

with 13C6-Leu were dissolved in Tris buffer (pH 9) and 8 M urea, then reduced,

alkylated and digested with Lys-C (Wako), followed by dilution with 50 mM

ammonium bicarbonate buffer (pH 9.0) and digestion with trypsin. Candidates

for peptide synthesis containing at least one leucine and one tyrosine, but not

methionine or cysteine, were selected considering the sequences of tryptic

peptides from proteins expressed in Neuro2A cells. One hundred and twenty-

two peptides were synthesized using a Shimadzu PSSM-8 with F-moc

chemistry and were purified by preparative high-performance liquid chroma-

tography (HPLC). Amino acid analysis, peptide mass measurement and HPLC-

UV were carried out for purity and structural confirmation. Known amounts of

these peptides were spiked into the peptide mixtures from Neuro2A cells and

LC/MS analyses were carried out to obtain the ratio of labeled peptides to the

unlabeled peptides. The spiked amounts were adjusted to obtain a ratio in the

range of 0.1–10. The absolute amounts of mouse brain proteins were calculated

from the peak ratio between Neuro2A labeled peptides and unlabeled brain

peptides in the mass spectra.


ACKNOWLEDGMENTSThis work was supported by funds from New Energy and Industrial TechnologyDevelopment Organization, Japan (NEDO).




1. Aebersold, R. & Mann, M. Mass spectrometry-based proteomics. Nature 422, 198–207(2003).

2. Sechi, S. & Oda, Y. Quantitative proteomics using mass spectrometry. Curr. Opin.Chem. Biol. 7, 70–77 (2003).

3. Oda, Y. et al. Quantitative chemical proteomics for identifying candidate drug targets.Anal. Chem. 75, 2159–2165 (2003).

4. Ranish, J.A. et al. The study of macromolecular complexes by quantitative proteomics.Nat. Genet. 33, 349–355 (2003).

5. Blagoev, B. et al. A proteomics strategy to elucidate functional protein-protein inter-actions applied to EGF signaling. Nat. Biotechnol. 21, 315–318 (2003).

6. Oda, Y., Huang, K., Cross, F.R., Cowburn, D. & Chait, B.T. Accurate quantitation ofprotein expression and site-specific phosphorylation. Proc. Natl. Acad. Sci. USA 96,6591–6596 (1999).

7. Pasa-Tolic, L. et al. High-thoughput proteome-wide precision measurements of proteinexpression using mass spectrometry. J. Am. Chem. Soc. 121, 7949–7950 (1999).

8. Wu, C.C., MacCoss, M.J., Howell, K.E., Matthews, D.E. & Yates, J.R. 3rd. Metaboliclabeling of mammalian organisms with stable isotopes for quantitative proteomicanalysis. Anal. Chem. 76, 4951–4959 (2004).

9. Jiang, H. & English, A.M. Quantitative analysis of the yeast proteome by incorporationof isotopically labeled leucine. J. Proteome Res. 1, 345–350 (2002).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

10. Ong, S.E. et al. Stable isotope labeling by amino acids in cell culture, SILAC,as a simple and accurate approach to expression proteomics. Mol. Cell. Proteomics 1,376–386 (2002).

11. Zhu, H., Pan, S., Gu, S., Bradbury, E.M. & Chen, X. Amino acid residue specific stableisotope labeling for quantitative proteomics. Rapid Mass Commun. Mass Spectrom.16, 2115–2123 (2002).

12. Sagane, K., Yamazaki, K., Mizui, Y. & Tanaka, I. Cloning and chromosomal mapping ofmouse ADAM11, ADAM22 and ADAM23. Gene 236, 79–86 (1999).

13. MacCoss, M.J., Wu, C.C., Liu, H., Sadygov, R. & Yates, J.R. 3rd. A correlation algorithmfor the automated quantitative analysis of shotgun proteomics data. Anal. Chem. 75,6912–6921 (2003).

14. Gygi, S.P. et al. Quantitative analysis of complex protein mixtures using isotope-codedaffinity tags. Nat. Biotechnol. 17, 994–999 (1999).

15. Sakai, J., Kojima, S., Yanagi, K. & Kanaoka, M. (18)O-labeling quantitative proteomicsusing an ion trap mass spectrometer. Proteomics 5, 16–23 (2005).

16. Parker, K.C. et al. Depth of proteome issues: a yeast isotope-coded affinity tag reagentstudy. Mol. Cell. Proteomics 3, 625–659 (2004).

17. Mano, N., Oda, Y., Yamada, K., Asakawa, N. & Katayama, K. Simultaneous quantitativedetermination method for sphingolipid metabolites by liquid chromatography/ionsprayionization tandem mass spectrometry. Anal. Biochem. 244, 291–300 (1997).

18. Lensmeyer, G.L. & Poquette, M.A. Therapeutic monitoring of tacrolimus concentrationsin blood: semi-automated extraction and liquid chromatography-electrospray ionizationmass spectrometry. Ther. Drug. Monit. 23, 239–249 (2001).

19. Gunawan, S., Griswold, M.P. & Kahn, D.G. Liquid chromatographic-tandem massspectrometric determination of amprenavir (agenerase) in serum/plasma of humanimmunodeficiency virus type-1 infected patients receiving combination antiretroviraltherapy. J. Chromatogr. A. 914, 1–4 (2001).

20. Cass, R.T., Villa, J.S., Karr, D.E. & Schmidt, D.E. Jr. Rapid bioanalysis of vancomycinin serum and urine by high-performance liquid chromatography tandem massspectrometry using on-line sample extraction and parallel analytical columns. RapidCommun. Mass Spectrom. 15, 406–412 (2001).

21. Wilkinson, A.P., Wahala, K. & Williamson, G. Identification and quantification ofpolyphenol phytoestrogens in foods and human biological fluids. J. Chromatogr. B777, 93–109 (2002).

22. Collingridge, G.L. & Isaac, J.T. Functional roles of protein interactions with AMPA andkainate receptors. Neurosci. Res. 47, 3–15 (2003).

23. Lerma, J. Roles and rules of kainate receptors in synaptic transmission. Nat. Rev.Neurosci. 4, 481–495 (2003).

24. Barr, J.R. et al. Isotope dilution—mass spectrometric quantification of specificproteins: model application with apolipoprotein A-I. Clin. Chem. 42, 1676–1682(1996).

25. Gerber, S.A., Rush, J., Stemman, O., Kirschner, M.W. & Gygi, S.P. Absolute quantifica-tion of proteins and phosphoproteins from cell lysates by tandem MS. Proc. Natl. Acad.Sci. USA 100, 6940–6945 (2003).

26. Havlis, J. & Shevchenko, A. Absolute quantification of proteins in solutions and inpolyacrylamide gels by mass spectrometry. Anal. Chem. 76, 3029–3036 (2004).

27. Katayama, H. et al. Efficient in-gel digestion procedure using 5-cyclohexyl-1-pentyl-beta-D-maltoside as an additive for gel-based membrane proteomics. Rapid Commun.Mass Spectrom. 18, 2388–2394 (2004).

28. Rappsilber, J., Ishihama, Y. & Mann, M. Stop and go extraction tips for matrix-assistedlaser desorption/ionization, nanoelectrospray, and LC/MS sample pretreatment inproteomics. Anal. Chem. 75, 663–670 (2003).

29. Ishihama, Y., Rappsilber, J., Andersen, J.S. & Mann, M. Microcolumns with self-assembled particle frits for proteomics. J. Chromatogr. A 979, 233–239 (2002).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Enzyme family–specific and activity-based screeningof chemical libraries using enzyme microarraysDaniel P Funeriu1, Jorg Eppinger2, Lucile Denizot1, Masato Miyake1 & Jun Miyake1

The potential of protein microarrays1 in high-throughput

screening (HTS) still remains largely unfulfilled, essentially

because of the difficulty of extracting meaningful, quantitative

data from such experiments2,3. In the particular case of

enzyme microarrays3, low-molecular-weight fluorescent affinity

labels4–10 (FALs) can function as ideally suited activity probes

of the microarrayed enzymes. FALs form covalent bonds with

enzymes in an activity-dependent manner and therefore can

be used to characterize enzyme activity at each enzyme’s

address, as predetermined by the microarraying process11.

Relying on this principle3, we introduce herein thematic

enzyme microarrays (TEMA). In a kinetic setup we used

TEMAs to determine the full set of kinetic constants and

the reaction mechanism between the microarrayed enzymes

(the theme of the microarray) and a family-wide FAL. Based

on this kinetic understanding, in an HTS setup we established

the practical and theoretical methodology for quantitative,

multiplexed determination of the inhibition profile of

compounds from a chemical library against each microarrayed

enzyme. Finally, in a validation setup, Kiapp values and

inhibitor profiles were confirmed and refined.

Protein microarray technologies typically focus on information con-tent. The higher the multiplexing power, the more relevant andvaluable the information extracted but also the more challenging thetask, since each protein’s behavior on the surface is difficult to probe,control and predict, which results in a need for multiple rounds ofvalidation. Designed for use in the HTS activity-based discovery andselectivity profiling of enzyme inhibitors in one-inhibitor-versus-n-enzymes format, TEMAs (Fig. 1) provide an enzyme microarrayplatform in which the multiplex power is purposely reduced tothe components of a single enzyme family. Indeed, due to theenzymes’ different environmental requirements for activity, microar-rays embedding large numbers of different and diverse enzymesmay well be neither relevant nor realistic for highly parallel activity-based investigations.

We set out to fully validate and demonstrate the capabilities ofTEMA technology on the cathepsin cysteine-protease family. Thedifferential involvement in disease (such as osteoporosis, arthritis,tumor invasiveness and parasital infections) of members of the

cathepsin family12–14 continues to generate important efforts for theidentification of strong, specific inhibitors15–17. Therefore, the avail-ability of a cathepsin microarray-based system for inhibitor discoveryand profiling is of direct practical interest. Moreover, existing FAL-based studies8 of this family provide a suitable starting point fortechnology validation. Several characteristics of the cathepsin family(such as fragility and pH instability) are also important for technologyvalidation, allowing the challenge of TEMAs in an application-relevant, nontrivial context. Cathepsins C, H, L, S, K and B (fromthree different sources) were microarrayed in duplicate within 48identical subarrays of a hydrogel-aldehyde functionalized glass slide,according to the pattern described in Figure 2. The prepared micro-arrays were used in the three experimental setups below, together withan epoxide containing FAL8 (Fig. 2d; Supplementary Methods onlinefor synthesis).

In the first experimental setup, the kinetic setup, after blocking andbuffer preincubation of each subarray, FAL solutions at four differentconcentrations were added concomitantly to four subarrays, at distincttime intervals (Fig. 1a), and the entire microarray was vigorouslywashed 30 s after FAL was last dispensed, such as to prevent anyfurther reaction of noncovalently bound FAL. The fluorescence of eachaddress was measured (Fig. 2a,b) and the data for each enzymeanalyzed according to a previously described algorithm3. The progresscurves obtained (Fig. 2e for cathepsin L, Supplementary Figs. 1–8online for all enzymes) were fitted to a series of possible theoreticalmodels. The best fit revealed that for all enzymes, the reaction betweenthe microarrayed enzymes and the FAL is a combination of aMichaelis-Menten–derived mechanism and a nonenzymatic back-ground reaction3,18 (Fig. 2c). The dependence of initial velocitiesvini on the concentration of the FAL, (Fig. 2f) in all cases adheres towhat is expected for an enzyme-catalyzed reaction. The characteristickinetic constants obtained are reported in Supplementary Table 1online (see Supplementary Methods online for calculation proce-dures). The extracted KM

app values allow the cathepsins to be sorted bytheir efficiency in converting the FAL in the following order (KM

app

(mM)): H (16), K (15), B (6.5), S (5.4), L (5.3), C (1.2). To test themethod’s robustness, the experiment was repeated three times with thesame batch of enzymes at more than 1-week intervals. The derivedkinetic constants of the individual experiments were found to differ byless than the experimental uncertainty.


1Research Institute for Cell Engineering, National Institute of Advanced Industrial Science and Technology, 3-11-46 Nakouji, Amagasaki, Hyogo, 661-0974, Japan.2ForschungsDozentur Molekulare Katalyse, Lehrstuhl fur Anorganische Chemie, Technische Universitat Munchen, Lichtenbergstr. 4, 85748 Garching, Germany.Correspondence should be addressed to D.P.F. ([email protected]) or J.E. ([email protected]).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Several conclusions can be drawn from the kinetic analysis of theprogress curves. Most importantly, they confirm the enzymatic natureof the reaction responsible for the fluorescent labeling of all themicroarrayed proteases by the FAL. The contribution of the none-nzymatic background reaction to the overall signal is below 10% forup to 80% conversion of the enzymes. An exception to this trend isrepresented by cathepsin C (Supplementary Fig. 4). For cathepsin Sand recombinant cathepsin B, the relative contribution of the back-ground reaction increases if the microarray is dried for prolonged

times after blocking. Because the background reaction presumablyresults from the attack of an SH group of a deactivated fold of theenzyme on the epoxide of the FAL, we infer that on the microarray’ssurface there is a loss of enzymatic activity attributed to foldingcollapse upon drying. Correspondingly, only in these two cases is theconcentration of active enzyme, c(E), reduced by 60–70% and theconversion of the remaining active enzyme, as expressed by the pseudosecond-order constant k2nd, slowed by 40–50%. These observationsunderline two important aspects. First, whereas the simple surface

No inhibition Unspecificinhibition inhibition

SpecificKinetic constants

Kinetic setup HTS setup Validation setup

Inhibitor titration

Data analysisData analysisData analysis

Washing and scanning Washing and scanning Washing and scanning

FAL additionFAL additionFAL addition

1. Blocking

1. Blocking

2. Inhibitor preincubation

2. Preincubation

Subarray

1. Blocking2. Preincubation with decreasing

inhibitor concentrations

a b c

Figure 1 Schematic explanation of TEMA technology. Different members of an enzyme family are microarrayed within identical subarrays on a functionalized

glass slide. (a) In the kinetic setup, after blocking and preincubation with reaction buffer, four different concentrations of FAL (one row of subarrays for each

concentration) are reacted for 12 different reaction times (one column of subarrays for each reaction time) at one concentration and one time per subarray.

Analysis of the data from this setup confirms the enzymatic nature of the reaction between the microarray enzymes and the FAL and the kinetic constants that

characterize the activity of each enzyme are extracted. This kinetic characterization is used in defining a set of experimental conditions under which one can

run a meaningful HTS experiment. (b) In the HTS setup, each of the subarrays is preincubated with a potential inhibitor (colored) or a blank (light blue) and

subsequently treated with FAL at a concentration and for a reaction time defined by the kinetic constants derived in a. Analysis of the data provides each

inhibitor’s activity profile against all enzymes of the subarray, importantly, under exactly the same experimental conditions, reducing reproducibility problems

associated with HTS. (c) In the validation setup, the subarrays within one row of subarrays are preincubated with different concentrations of an inhibitor

(including a blank) and subsequently treated with FAL in a manner similar to b. Analysis of the data provides an inhibitor titration curve, from which refined

Kiapp values can be calculated.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

chemistry that we use reliably maintains the activity of the micro-arrayed enzymes under optimized conditions, further advances insurface chemistry19 will increase the TEMAs’ robustness upon storage.Second, they reveal that mere fluorescence observation at an enzyme’saddress is not necessarily related to enzymatic ‘activity’, making onetime point measurements unsuitable for such claims.

The second experimental setup, the HTS setup (Fig. 1b) isestablished with the goal of quantifying the inhibition profile of acompound against the entire panel of microarrayed enzymes from onetime point measurement. In general, Ki

app values are most straight-forwardly extracted from initial velocities (vini) determined afterequilibration of the enzyme/inhibitor system. However, because ofthe impossibility of directly measuring vini, the intrinsically differentreactivity of the microarrayed enzymes towards the FAL and thenecessity of taking into consideration the background reaction, multi-plexing the direct determination of Ki

app as recently described3,20 canresult in significantly distorted inhibition profiles. If this differentkinetic-related behavior of the enzymes is ignored, the error of theKi

app derived from initial velocity ratios was calculated to possiblyamount to up to 800% for the case under study. Therefore, to reliablyquantify the inhibition profiles for the one-inhibitor/several-enzymescase from one-time-point experiments, we measure the exact con-centration of the reaction product Pi (instead of vini) at a given time;we then apply the above complete kinetic description of the reactionbetween the microarrayed enzymes and the FAL by including thebinding equilibrium for a reversible inhibitor. The kinetic data

obtained above allow estimation of whether,and under which conditions, the chosen FALis appropriate for the multiplexed determina-tion of inhibitor activity21 according to thismethodology: indeed, if two enzymes react atvery different velocities, the fast-reactingenzyme would impose a short reaction timefor Pi measurements (before saturation),which in turn may not be enough for theslow-reacting enzyme to reach a sufficientsignal/noise ratio. The FAL used in thisstudy, at a final concentration of 0.4 mMand reaction time of 120 s satisfies theseconditions. This method is suitable to gen-erate inhibitor profiles independent of themechanism of inhibition.

Microarrays were prepared similar to thoseused for kinetic studies. Eight compoundsknown to be cysteine-protease inhibitors (seeSupplementary Methods online for inhibitordescription) were randomly distributedamong 186 compounds with unknown inhi-bitor activity. Seven of the known inhibitorswere screened twice on different microarrays.After preincubating each subarray of theTEMA with one of the library members, theFAL was reacted for 2 min and the microarraywas washed and scanned. The chosen inhi-bitor concentrations (3 mM), in combinationwith average experimental errors of 10%,allowed a range between 80 nM and 2 mMin which Ki

app can be quantified (Fig. 3).From such experiments it is intrinsically pos-sible to derive only one of the two kineticconstants kon and koff that characterize the

inhibition process. Therefore kon was assumed to have the calculatedvalue of the corresponding constant for the FAL-binding process (k+1),and koff was obtained from the inhibition studies. The systematic erroron the calculated Ki

app values obtained under this assumption and forour experimental conditions is at most 25% if kon is within two ordersof magnitude of k+1 (Supplementary Methods). All the knowninhibitors were correctly found in this experimental setup (Fig. 3).Comparison with literature values16,17,22–24 (Supplementary Table 2online) reveals that known inhibition constants and profiles are wellreflected in this assay. They are reproducible, as demonstrated by thedata obtained on the seven inhibitors that were screened twice. Somevery weak inhibition by so far unknown compounds could be observed(Ki

app 4 800 nM). No false positives occurred for strong inhibitors.In the third experimental setup, the validation setup, Ki

app valuesare refined and inhibitor profiles confirmed (Fig. 1c). After thedemonstration that Ki

app can be obtained in a HTS format usingthe accurately established conditions above, we set to demonstrate thatthis method is also suited for the refinement of the calculatedinhibition constants for several enzymes in parallel. This setup isparticularly useful and necessary for compounds whose range of Ki

app

values for the different enzymes is too broad to be measured from thedata acquired from a single concentration experiment, as discussed inthe HTS setup. In this setup, each subarray is preincubated withdifferent inhibitor concentrations, typically ranging from 120 mM to0.11 nM, followed by addition of the FAL at a final concentration of0.42 mM for 2 min (Fig. 4). For high inhibitor concentrations, the

0.5

1.5

4.5

13.5

c(FAL) = 13.5 µMc(FAL) = 4.5 µMc(FAL) = 1.5 µMc(FAL) = 0.5 µM

22

B, hum

B, bov

B, rec

C, bov

H, bov

L, hum

S, rec

K, rec

0.5 3 4 5 7 10 13 171 20

FAL

E + S ES k+1; k–1

ES P kcat

Et + S P kside

c(FAL) (µM)

0

5

10

15

20

25

0 2 4 6 8 10 12 140 42 6 8 10

1.4

1.2

1.0

0.8

0.6

0.4

0.2

0

Time (min)

c (P

rodu

ct)

(nM

)

Vin

i / c

(E)

/ (10

–3 s

–1)

Time (min)

c(F

AL)

(µM

)

a

c d

b

e f

t

Figure 2 Results of the kinetic setup experiments. (a) Fluorescent scanner image of a microarray

processed as described in the kinetic setup. (b) Typical image of a subarray, and the identity of the

microarrayed cathepsins (each enzyme in duplicate). (c) Michaelis-Menten–derived, best-fit reaction

model obtained for all the microarrayed enzymes. For numerical values of k+1, k�1, kcat and kside see

Supplementary Table 1 online. (d) Structure of the FAL used in this study. (e) Time progress curves

derived from the microarray in a for cathepsin L. (f) Plot of initial velocities nini versus [FAL] for all the

microarrayed enzymes.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Kiapp can be determined as described recently3,20. Importantly, for

inhibitor concentrations close to or below the total enzyme concen-tration, one has to take into account the fact that the ‘effective’concentration of inhibitor to which a given enzyme is exposed issignificantly depleted by other enzymes’ binding of this inhibitor. Thisaffects the determined inhibition constants in the interesting case ofstrong inhibitors (when nearly all inhibitor is enzyme bound) if c(I)o 5*c(Et) (with Et the total concentration of all active enzymes). Tocorrect for this effect, we set up an algorithm (SupplementaryMethods) that alternately calculates the Ki

app values for each enzymeas described for the HTS setup and then uses those values todetermine the concentration of free and bound inhibitor. Thoseconcentrations are used to recalculate the Ki

app for each enzyme;this iterative process convergences to accurate Ki

app values. Neglectingthis effect can shift the determined Ki

app values dramatically; in thecase of leupeptin for example, the determined constants are more than

one order of magnitude too low if the effect of parallel competition isneglected. The inhibition constants obtained in the microarray-basedvalidation experiment (Supplementary Table 3 online) very closelycorrespond with the literature values of solution-state studies. Thisindirectly confirms that the studied, active enzymes are immobilizedin a native-like state, with little interference of the solid support ontheir activity.

To acquire information about the method’s robustness under abroad range of conditions, we evaluated the microarrayed enzyme’sbehavior towards environmental factors, in particular surfactantconcentration and pH (Supplementary Fig. 9 online). Thesubarrays were preincubated at different SDS concentrations ordifferent pH and subsequently reacted with FAL solutions for2 min. Cathepsin S and H were found to rapidly loose activity withincreasing SDS concentration, whereas cathepsin K’s activity reaches asharp maximum around the critical micelle concentration of SDS.This effect is similar for cathepsin B, C and L yet much lesspronounced. The fact that environmental conditions can significantlyenhance the solid-support-bound active fold of the enzyme indicatessome degree of conformational dynamism for the covalently boundenzyme. The cathepsins’ higher activity at low pH was confirmed;cathepsin L was particularly sensitive to the pH increase, loosing all itsactivity at pH 4 6.5. In contrast, cathepsin S and C maintainedsubstantial activity at slightly basic pH, in agreement with otherstudies25. Because of higher reactivity of the epoxide group at lowpH, the contribution of the background reaction to the total signalincreased at low pH. Therefore, the fluorescent signal at lower pH washigher than what would be expected from the classic bell-shapedactivity versus pH curves obtained by alternative methods25.

In conclusion, we have shown that TEMA technology can be usedto generate an inhibitor’s profile against a family of enzymes in oneexperiment. The method is suited to HTS experiments where itprovides not only binary yes/no answers, but gives access to at leastsemi-quantitative Ki

app values. By adapting the experimental setup, theexact quantification of on-chip inhibition constants is possible.Miniaturization, multiplexing power, straightforward automatization,throughput, low sample consumption, complemented by high robust-ness demonstrate the advantages of TEMAs over classical, well-plateassays. The quantitative, multiplexed data that it provides, as well asthe inhibition mode3 are crucial for distinguishing between the morepromising, enzyme-specific inhibitors and molecular motives thattarget the studied enzymes indiscriminatingly. This is of particularimportance since it provides an integrated system for the earlyidentification of cross-reactive inhibitors, saving considerable effortsfor their further evaluation. Whereas in this study we use a relativelysmall library of molecules for technology validation, we estimate thatunder appropriate automation on the order of 105 compounds can beprofiled daily against an entire family of enzyme targets. Althoughrequiring purified enzymes, this technology adds high-throughputcapabilities to existing FAL-based gel profiling methods8,21. In prin-ciple, TEMA is general for families of enzymes that form a covalentintermediate with their substrate. However, successful extension ofTEMAs towards a comprehensive coverage of the targetable enzymefamilies depends on two essential factors: firstly, as our kinetic studyhas clearly shown, a family-wide FAL with finely tuned properties andlow background reactivity must be employed. Although significantrecent advances in FAL-related research5–10 few kinetic data relative tothe described FALs are available21. Secondly, although the sensitivity ofthe method is such that only a tiny amount of properly folded enzymemust be present on the surface (0.05 fmoles/spot), the method willsubstantially benefit from advances in the active field of surface

80

200

400

800

2,000

No inhibitor CNI E-64

SLKHCBhum Bbov Brec

E-64

Cat I

PD

CNI

CAL VI

100

ALLN

Leu

CBZL

180CBZCO

a

b

Figure 3 Results of the HTS setup experiment. (a) Enzyme inhibitor

profiles resulting from on-chip parallel screening of an inhibitor library

(one compound/subarray) against eight cysteine proteases of the cathepsin

family. The setting of the method (inhibitor concentration used) allows a

quantification of the inhibition constant Kiapp in the range between 80 and

2,000 nM (shades of blue, see scale bar on the right). Higher values are

shown as white, lower values as orange bars. Roughly, for the microarrays

presented herein, an inhibitor concentration c(I), allows Kiapp values between

c(I) and c(I)/500 to be estimated. Known cysteine protease inhibitors in the

library are identified at the left of the diagram. All known inhibitors were

found in this experimental setup. Their determined profiles match well

the available literature data. (b) The resulting fluorescence intensities

after preincubation with a selective (CNI: NC-D-Pro-Leu-(OBn)) and an

unselective (E-64) inhibitor compared to the noninhibited reactivity (FAL-

incubation time: 2 min). To test data reproducibility and to ensure the

absence of systematic errors, in this setup the positions of cathepsin B

human and cathepsin B bovine were reversed as compared to Figure 2b.


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

chemistry19. Moreover, in the case of enzymes that do not formcovalently-bound intermediates with their substrates, we are currentlyinvestigating the use of alternatives to FALs, such as tagged photo-reactive, mechanism-based inhibitors26–28. However, as our kineticstudies have shown, a detailed analysis and understanding of the on-chip processes is needed before robust use of TEMAs is possible.

METHODSGeneral. The compounds used for HTS screening were purchased from

Nanosyn Inc. Eight known cathepsin inhibitors and a parent compound of

one of the inhibitors, known for its inability to inhibit cathepsins, were

randomly distributed among the screened compounds. These eight compounds

were either commercially available (see below) or synthesized according to

literature procedures.

Microarray preparation. The enzymes were microarrayed using a 200 mm pin

and a commercial microarrayer (LabNext) on a hydrogel-aldehyde (NoAb

Diagnostics) functionalized glass slides (Matsunami Glass) comprised of 48

subarrays (12 rows and 4 columns) at about 1 mg/ml in the respective buffer at

73% humidity and incubated for about 1 h in a wet atmosphere (humidity

from 60–80%), at 25 1C. The subarrays are spatially separated by a hydrophobic

coating, which allows them to be treated both individually and collectively. The

surface was blocked with 2% BSA (wt/vol) in 50 mM TRIS, pH 5.5 (TRISA) for

10 min. Then the microarray was dip-washed with TRISA and dried by

centrifugation at 15g for 2 min. Other surfaces, such as aldehyde, hydrogel-

NHS, NHS, hydrogel-BSA-NHS provided significantly inferior results, resulting

either from poorer immobilization or enzyme denaturation onto the surface.

Microarray processing. Kinetic setup. After the microarray preparation

described above, we preincubated each sub-array with 8 ml reaction buffer

(TRISA containing 5 mM CaCl2, 5 mM MgCl2, 2 mM DTT, 2% DMSO

(vol/vol)) for 30 min. Preliminary experiments have established that the

enzymes’ activity is largely independent of the preincubation time in the range

of 5 to 90 min. After the preincubation, 4 ml of a solution of FAL in reaction

buffer was concomitantly added at the times described into the text, using a

multichannel pipette, at four different stock concentrations (13.5 mM, 4.5 mM,

1.5 mM, 0.5 mM) to a column of four subarrays.

Final concentrations of FAL: 4.5, 1.5, 0.5, 0.166 mM.

Because of significant adsorption of the FAL to the

plastic-ware used (pipette tips, well-plates, centri-

fuge tubes) all the plasticware was preconditioned

with a 20 mM solution of FAL. The FAL concentra-

tion was estimated by UV measurements of the

used solutions both before and after the experi-

ment. At the end of the kinetic experiment the

microarray is first flow washed with a 80 1C

solution of SDS buffer (35 mM SDS, 10 g/l glycine)

at pH 5.5 (wash buffer), then submersed in a 80 1C

solution of wash buffer and vigorously shaken for

3 min, sonicated for 2 min, abundantly washed

with ddH2O and dried by centrifugation at 15g for

2 min. Repeated washings under these conditions

resulted in no significant differences in fluores-

cence. The obtained microarray was scanned using

an ArrayWorks microarray scanner (Applied Pre-

cision) and the fluorescence quantified using

Imagene software (Biodiscovery).

HTS setup. After the microarray preparat-

ion described above, each subarray was preincu-

bated for 90 min with 8 ml reaction buffer

containing the appropriate concentration of inhi-

bitor. In the experiments described herein

the inhibitor concentration was chosen to be

3 mM. After the preincubation, we added 4 ml of

a solution of FAL in reaction buffer at 1.25 mM

using a multichannel pipetting device. Final con-

centrations were 2 mM for c(I) and 0.41 mM for

c(FAL). The microarray was washed 2 min after FAL addition, as described

in the kinetic setup.

Validation setup. Each subarray was preincubated with 8 ml reaction buffer

containing decreasing inhibitor concentrations (120 mM, 30 mM, 7.5 mM,

1.87 mM, 0.47 mM, 117 nM, 30.3 nM, 7.32 nM, 0 nM, 1.83 nM, 0.45 nM,

0.11 nM) for 90 min. Then 4 ml of a solution of FAL in reaction buffer at

1.25 mM was concomitantly added to the subarrays using a multichannel

pipette. The microarray was washed 2 min after FAL addition, as described in

the kinetic setup. In the check of the technology’s robustness, each subarray was

preincubated for 90 min with 8 ml reaction buffer containing decreasing

concentrations of SDS and glycine (Supplementary Fig. 9) or 8 ml reaction

buffer at different pH (from 3.5 to 9 by 0.5 pH units increment). Then 4 ml of a

solution of FAL in reaction buffer at appropriate pH at 1.25 mM is concomi-

tantly added to the subarrays using a multi-channel pipette. 2 min after FAL

addition, the microarray was washed as described in the kinetic setup. Further

details of the experimental setup and data analysis procedure are provided in

the Supplementary Methods online.


ACKNOWLEDGMENTSWe thank Robert Menard for providing recombinant cathepsin B andrecombinant cathepsin L. This work was supported by the National Institute ofAdvanced Industrial Science and Technology of Japan and by the Stifterverbandfur die Deutsche Wissenschaft (Projekt-Nr. 11047: ForschungsDozenturMolekulare Katalyse).

COMPETING INTERESTS STATEMENTThe authors declare competing financial interests (see the Nature Biotechnologywebsite for details).



1. Kambhampati, D. (ed.). Protein microarray technology (Wiley-VCH, Heidelberg, 2004).2. Zhu, H. et al. Global analysis of protein activities using proteome chips. Science 293,

2101–2105 (2001).

10610510410310210110010–1

c(CNI) (nM)10610510410310210110010–1

c(E-64) (nM)

1.2

1.0

0.8

0.6

0.4

0.2

0

Pi/P

max

1.2

1.0

0.8

0.6

0.4

0.2

0

Pi/P

max

Cath BCath CCath HCath KCath LCath S

Cath BCath CCath HCath KCath LCath S

–1.6–0.60.4Inh(–)1.42.43.44.45.46.4

log4(c(Inh) (nM))

7.48.4

CNI

E-64a

b

Figure 4 Results of the validation setup experiment for two inhibitors (CNI, reversible and E-64,

irreversible). (a) Image of the relevant subarrays for E-64 and CNI. To avoid possible systematic errors,

we placed the 0 inhibitor concentration (blank) subarrays fourth from right. Enzyme identity is the

same as in Figure 3. (b) Evolution of the ratio between the initial product formation (Pi/Pmax) as a

function of inhibitor concentration, derived for the two microarray experiments presented in a. It can

be estimated that for the microarrays described herein, a minimum inhibitor concentration of about0.1 nM (1fmole inhibitor) is necessary to observe an effect for Ki

app values in the nM range. For

simplification, only one curve for cathepsin B is shown (source: human).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

3. Eppinger, J., Funeriu, D.P., Miyake, M., Denizot, L. & Miyake, J. Enzyme microarrays:on-chip determination of inhibition constants based on affinity-label detection ofenzymatic activity. Angew. Chem. Int. Ed. 43, 3806–3810 (2004).

4. Chen, G.Y., Uttamchandani, M., Zhu, Q., Wang, G. & Yao, S.Q. Developing a strategyfor activity-based detection of enzymes in a protein microarray. ChemBioChem 4,336–339 (2003).

5. Liu, Y., Patricelli, M.P. & Cravatt, B.F. Activity-based protein profiling: the serinehydrolases. Proc. Natl. Acad. Sci. USA 96, 14694–14699 (1999).

6. Campbell, D.A. & Szardenings, A.K. Functional profiling of the proteome with affinitylabels. Curr. Opin. Chem. Biol. 7, 296–303 (2003).

7. Speers, A.E. & Cravatt, B.F. Profiling enzyme activities in vivo using click chemistrymethods. Chem. Biol. 5, 535–546 (2004).

8. Greenbaum, D.C. et al. Small molecule affinity fingerprinting. A tool for enzyme familysubclassification, target identification, and inhibitor design. Chem. Biol. 9, 1085–1094 (2002).

9. Jessani, N. & Cravatt, B.F. The development and application of methods for activity-based protein profiling. Curr. Opin. Chem. Biol. 8, 54–59 (2004).

10. Goulet, B. et al. A cathepsin L isoform that is devoid of a signal peptide localizes to thenucleus in S phase and processes the CDP/Cux transcription factor. Mol. Cell 14, 207–219 (2004).

11. Gosalia, D.N. & Diamond, S.L. Printing chemical libraries on microarrays for fluidphase nanoliter reactions. Proc. Natl. Acad. Sci. USA 100, 8721–8726 (2003).

12. Berdowska, I. Cysteine proteases as disease markers. Clin. Chim. Acta 342, 41–69(2004).

13. Greenbaum, D.C. et al. A role for the protease falcipain 1 in host cell invasion by thehuman malaria parasite. Science 298, 2002–2006 (2002).

14. Lecaille, F., Kaleta, J. & Bromme, D. Human and parasitic papain-like cysteineproteases: their role in physiology and pathology and recent developments in inhibitordesign. Chem. Rev. 102, 4459–4488 (2002).

15. Kang, K. & Kim, W. Recent developments of cathepsin inhibitors and their selectivity.Exp. Opin. Therap. Pat. 12, 419–432 (2002).

16. Powers, J.C., Asgian, J.L., Ekici, O.D. & James, K.E. Irreversible inhibitors of serine,cysteine, and threonine proteases. Chem. Rev. 102, 4639–4750 (2002).

17. Otto, H.-H. & Schirmeister, T. Cysteine proteases and their inhibitors. Chem. Rev. 97,133–171 (1997).

18. Kuzmic, P. Program DYNAFIT for the analysis of enzyme kinetic data: application toHIV proteinase. Anal. Biochem. 237, 260–273 (1996).

19. Carrillo, A., Gujraty, K.V. & Kane, R.S. Surfaces and substrates. in MicroarrayTechnology and Its Applications (eds. Mueller, U.R. & Nicolau, D.V.) 45–61 (SpringerGmbH, Berlin, 2005).

20. Kuzmic, P. et al. High-throughput screening of enzyme inhibitors: automatic determi-nation of tight-binding inhibition constants. Anal. Biochem. 281, 62–67 (2000).

21. Leung, D., Hardouin, C., Boger, D.L. & Cravatt, B.F. Discovering potent and selectivereversible inhibitors of enzymes in complex proteomes. Nat. Biotechnol. 21, 687–691(2003).

22. Sasaki, T. et al. Inhibitory effect of di- and tripeptidyl aldehydes on calpains andcathepsins. J. Enzyme Inhib. 3, 195–201 (1990).

23. Yamashita, D.S. et al. Structure and design of potent and selective cathepsinK inhibitors. J. Am. Chem. Soc. 119, 11351–11352 (1997).

24. Rydzewski, R.M. et al. Peptidic 1-cyanopyrrolidines: synthesis and SAR of a series ofpotent, selective cathepsin inhibitors. Bioorg. Med. Chem. 10, 3277–3284 (2002).

25. Kirschke, H., Barrett, A.J. & Rawlings, N.D. Proteinases 1: lysosomal cysteineproteases. Protein Profile 2, 1581–1643 (1995).

26. Gilbert, B.A. & Rando, R.R. Modular design of biotinylated photoaffinity probes:synthesis and utilization of a biotinylated pepstatin photoprobe. J. Am. Chem. Soc.117, 8061–8066 (1995).

27. Hagenstein, M.C. et al. Affinity-based tagging of protein families with reversibleinhibitors: a concept for functional proteomics. Angew. Chem. Int. Ed. 42, 5635–5638 (2003).

28. Saghatelian, A., Jessani, N., Joseph, A., Humphrey, M. & Cravatt, B.F. Activity-basedprobes for the proteomic profiling of metalloproteases. Proc. Natl. Acad. Sci. USA 101,10000–10005 (2004).


L E T T E R S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Antibodies

Mouse anti-human antibodies

Allophycocyanin-, fluorescein- and phy-coerythrin-conjugated mouse anti-human CCL3/MIP-1α antibodies are available from R&D Systems. CCL3, a member of the CC or β-chemokine subfamily, was origi-nally purified from the conditioned media of LPS-stimulated macrophages. It acts as a chemoattractant to a variety of cell types including monocytes, T cells, B cells and eosinophils.http://www.RandDsystems.com/

Rabbit monoclonal antibodiesSix new rabbit monoclonal antibodies have been added to Vector Laboratories’ line of primary antibodies for immunohistochemical stain-ing: CD3, COX-2, Cyclin D1, Ki67, Estrogen Receptor and Progesterone Receptor. These antibodies provide excellent results on formalin-fixed, paraffin embedded tissue sections using standard immunohistochemistry methods. They can be applied to sections just like mouse monoclonals, and then detected using anti-rab-bit IgG secondary detection reagents.http://www.vectorlabs.com/

Antibodies for cell-signaling research

Epitomics’ rabbit monoclonal antibod-ies offer more diverse epitope recognition, improved response to less immunogenic antigens, and greatly improved response to rodent proteins. All Epitomics RabMAbs offer high affinity and are extensively characterized and tested for use in WB, IHC, ICC, IF, IP and flow cytometry. Antibodies are available for key proteins involved in various cell-signaling pathways including apoptosis, cell cycle con-trol, cytokine signaling and many phospho-specific proteins (Above: Phospho-EGFR (pY1173) RabMAb).http://www.epitomics.com/

Gene cloning

Human ORF collection

Open Biosystems offers a collection of clones containing full open reading frames (ORFs) for over 8,000 human genes, created by the Center for Cancer Systems Biology of the Dana-Farber Institute. Derived from fully sequenced Mammalian Gene Collection full-length cDNAs, they are cloned into recombi-national entry vectors. ORF clones save time by allowing users to skip PCR, cloning into an expression vector, and verifying the ends of the ORF DNA sequence. The GatewayT entry vectors ensure easy transfer into pro-

karyotic, mammalian, viral or insect expres-sion systems.http://www.openbiosystems.com/

Broad specificity antibodiesThree antibodies from Upstate are validated to recognize over 80 serine/threonine and tyrosine kinases: 4G10 antibodies detect twice as many phospho-proteins as PY-20 and PT-66 using the most frequently refer-enced phospho-tyrosine antibody available; MPM2 antibodies are able to recognize more than 40 mitotic proteins; and highly specific phospho-MBP monoclonal antibodies and conjugates are available as detection reagents for numerous kinase assays.http://www.upstate.com/

Comprehensive human genes

GenScript’s GenPool ORF collection includes every single human gene with known sequence. All the ORFs are cloned into the pDream 2.1 vector, which allows expression in bacteria, Sf9 cells and mammalian cells. Each gene has a FLAG tag to facilitate purifi-cation and detection, which is also removable if needed. pDream2.1 vector is also compat-ible with Gateway and LIC systems.http://www.genscript.com/

NEW ON THE MARKET©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


US biotechnology companies and foreign nationals: the changing dynamics of access to H-1B visasA Stephen Dahms & Stephen C Trow

Recent changes in the H-1B visa program have left biotech employers shorthanded and confused.

Regional biotechnology industry clusters cite access to a skilled labor pool as one of

the top two or three most significant hurdles to commercialization. In the United States, a shortfall in the labor pool has required access to foreign workers in a variety of areas and expertise. Starting with efforts in California, the industry was surveyed in 1998–2001, which led to findings that, depending upon the region, between 6% and 10% of the US biotechnology workforce had H-1B visas, with an estimate of 18,000 in the biotech industry nationally, and projected needs of 25% annual increases in some of the clusters. Relatedly, the US Department of Commerce reported in 2003 that the biotech industry workforce grew annually by 12% over the period 1997–2003.

Biotechnology industry surveys of H-1B visa usage1 also found that 80% of biotech-nology H-1Bs were from US universities, 75% were graduate-degreed (40% PhD, 35% MS, 20% BS and 5% MD), 85% eventually acquired permanent residency in the US and companies were spending an average of $10,200 on each H-1B worker for processing fees and legal expenses through to a green card, leading to the conclusion that US com-panies had spent in excess of $150 million over the previous five years to acquire and keep their H-1B workers. These and other survey data also clearly showed that the H-1B worker skill sets sought by biotech companies

identically match the most pressing employ-ment needs of the industry overall and that their compensation was equal to, or in most cases, higher than US nationals.

Recent historyAccess to H-1B visas is limited by a ‘cap’ or annual limit on the number of new foreign workers who can be granted H-1B visa sta-tus. Increasing demand for H-1B visa status first exceeded the cap in September 1997, and then caused major disruptions in May 1998 when processing of petitions for new H-1B workers was suspended for the remain-ing five months of fiscal year (FY) 1998. In response to pressure exerted by the infor-mation technology (IT) industry and the Biotechnology Industry Organization (BIO; Washington, DC, USA), the H-1B cap was raised from 65,000 to 115,000 in FY 1999 and then again to 195,000 in FY 2001, with a provision that the cap would revert to 65,000 on October 1, 2004.

Starting in mid-2001, discussions began within BIO about legislative strategies to raise the cap from 65,000 back to a level that would assure access to these talented foreign nation-als. This initiative was derailed by the events of September 11, after which any discussion of increasing the entry of foreign workers was counter to public and congressional opinion.

Although there has not been a biotechnology industry H-1B needs survey since early 2001, it is thought that demands are increasing in the range of 3,500 to 5,000 per year. Fortunately, the law raising the cap for FY 2001 also exempted from the cap all H-1B workers employed at an institution of higher education or a related or affiliated nonprofit entity, or at a nonprofit research organization or a governmental research organization, thus removing approxi-mately 10,000 workers from the H-1B cap.

The current situationOn October 1, 2004, when FY 2005 began and the H-1B cap reverted to 65,000, the US Citizenship and Immigration Services (USCIS) announced that it had received enough H-1B petitions during the preceding six months to use up the entire supply for FY 2005. The USCIS indicated that it would continue to pro-cess petitions that it received before October 1, but it would not accept any new H-1B petitions that are subject to the cap for FY 2005. It also announced that it would start to accept peti-tions for FY 2006 on April 1, 2005, six months before the start of FY 2006. Many employers expected that the USCIS would receive enough petitions to exhaust the FY 2006 supply well before October 1, 2005, and made plans to file their petitions in April 2005.

In December 2004, the FY 2005 Omnibus Appropriations Act exempted from the H-1B cap up to 20,000 foreign nationals per year who have earned a master’s degree or higher from a US university. This change was prompted by renewed pressure from the IT, semiconductor and engineering industrial sectors, especially the National Association of Manufacturers (which BIO has partnered with on the H-1B issue since the mid-1990s).

The new exemption is vitally important for foreign students who graduated from US uni-versities during 2004 and are now working in F-1 or J-1 visa status with practical training authorization that expires before October 1, 2005. The exemption was scheduled to take effect on March 8, 2005, and it was expected that the 20,000 new slots would be taken quickly, so employers made plans to file peti-tions for qualifying workers on March 8. There were no changes in the other exemptions from the H-1B cap, so employers continued with their plans to file petitions for nonexempt workers on April 1.

A. Stephen Dahms is at the California State University System Biotechnology Program, San Diego State University, San Diego, California, USA and Stephen C. Trow is at Trow & Rahal, PC, an immigration law firm in Washington, DC, USA.e-mail: [email protected] or [email protected]

CAREERS AND RECRUITMENT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


However, on March 4, 2005, the USCIS announced that it was not ready to accept petitions for these 20,000 new H-1B slots as scheduled on March 8, and then it announced to great surprise that these 20,000 slots would not be limited to foreign nationals with a master’s/PhD degree from a US university. The USCIS did not publish a rationale for this dramatic change, but it appears to be based on a determination that at least 20,000 peti-tions for master’s/PhD graduates of US uni-versities had already been approved during FY 2005, making the 20,000 new slots avail-able to any qualified applicants. In addition, in late March the USCIS announced that it had mistakenly approved at least 10,000 more H-1B petitions during FY 2005 than authorized by the 65,000 cap. This series of announcements left employers wondering whether the 10,000 excess approvals would be deducted from the 20,000 new slots, leav-ing only 10,000 new slots to be given to any and all qualified applicants.

As of April 20 the USCIS had still not announced the filing date and procedure for the 20,000 new H-1B slots that were expected to be available on March 8, nor had it indi-cated whether the 10,000 excess approvals would be deducted from the 20,000 new slots. The prospects for recent master’s/PhD gradu-ates of US universities to avoid gaps in their work authorization before October 1, 2005, are looking much worse, whereas graduates of foreign universities and bachelor’s degree graduates of US universities can now hope for a windfall H-1B approval before October 1. Regardless of how these issues are resolved,

the delay and uncertainty have made life more difficult for employers and workers seeking H-1B status.

Fee hikes and usageEmployers struggling with the limited supply of H-1B visas and confusion over eligibil-ity for the 20,000 new slots are also facing sharply higher filing fees for H-1B petitions. The FY 2005 Omnibus Appropriations Act reinstated a ‘training’ fee that had lapsed in October 2003 and increased that fee from $1,000 to $1,500 for most H-1B peti-tions. The training fee is reduced to $750 for employers that have no more than 25 full-time equivalent employees, including employees of affiliates and subsidiaries. This fee will fund job training and scholarships for US workers, and government processing of H-1B cases.

The Appropriations Act also imposed an additional $500 ‘fraud prevention’ fee for each petition seeking an initial grant of H-1B status or authorization to change employers in H-1B status. This will provide additional funding for visa fraud prevention programs at USCIS and other government agencies. These new fees are separate from the mandatory $185 base fee for an H-1B visa petition, and the optional $1,000 fee for premium processing (faster service) from the USCIS. The total filing fee for a large employer seeking premium processing of an H-1B visa petition is now $3,185.

Over $500 million of H-1B training fees have been collected and routed by the US Department of Labor since 1999 to fund new training programs to help reduce dependency

upon foreign nationals, but unfortunately none of these funds has been directed at the graduate-degree education that the biotech-nology industry needs. Biotech employers are paying H-1B training fees to create programs to relieve their dependency on foreign work-ers, yet most H-1B workers in the biotech and high-tech industries are coming from US educational institutions. The problem arises on the supply side of the labor equation, not the demand side. Currently, over 25% of all PhDs in the US are foreign nationals, and over 50% of all graduate students are foreign students. Funding graduate-level training programs will not reduce the demand for H-1B workers if US students decline to enroll in those programs.

ConclusionsClearly, the H-1B visa program provides a temporary solution to shortages in the national and domestic biotech labor pools—shortages that mirror the inadequate produc-tion of appropriately trained US nationals by US institutions of higher learning. The reality is that universities have inadequate resources for expanding their training pipelines, espe-cially in specialized areas that follow the basic-research phase of company product development. Efforts should be directed toward influencing greater congressional and federal agency attention to these impor-tant topics, especially an increase in the H-1B cap and effective use of the very sizable H-1B fee–derived training funds.

1. Sevier, E.D. & Dahms, A.S. Nat. Biotechnol. 20, 955–956 (2002).

CAREERS AND RECRU ITMENT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy


Stressgen Biotechnologies (Victoria, BC, Canada) has appointed Gregory M. McKee as the company’s president and CEO, succeeding Dan Korpolinski who is leaving to pursue other interests. Mr. McKee, who has served as Stressgen’s chief financial officer and vice president of corporate development for the past two years, will focus on the commercialization of HspE7, the company’s lead product candidate targeting a broad spectrum of human papillomavirus (HPV)-related diseases. Mr. McKee previously served as senior director, corporate

development for Valentis, as well as director of Genzyme’s operations in Asia.

Michael J. Astrue has been appointed to the board of directors of ArQule (Woburn, MA, USA). Mr. Astrue is currently president and CEO of Transkaryotic Therapies, and previ-ously served as vice president, secretary and general counsel for Biogen as well as chairman of the Massachusetts Biotechnology Council.

Benitec (Mountain View, CA, USA) has named Michael Catelani chief financial officer. Mr. Catelani had served as a consultant to the company since January 2005. Mr.

Catelani previously held senior financial management positions including vice presi-dent and CFO at Axon Instruments.

David Chiswell has been named nonexecu-tive chairman of DanioLabs (Cambridge, UK), replacing Roger Brimblecombe who has retired from the DanioLabs board. Dr. Chiswell was a founder and CEO of Cambridge Antibody Technologies. In addition to this new chairmanship, he also serves as chairman of Arrow Therapeutics, Sosei Co. and the UK BioIndustry Association, and as a nonexecu-tive director of Arakis Ltd.

Innovive Pharmaceuticals (New York) has announced the appointment of Adam R. Craig as vice president and chief medical officer. He joins the company from ArQule where he served as medical director and vice president, clinical development. Dr. Craig was also senior direc-tor, clinical development for Ilex Oncology and medical advisor, oncology for Antisoma.

Illumina (San Diego, CA, USA) has appointed Scott D. Kahn as vice president and chief information officer, a new position. Dr. Kahn joins the company from Accelrys, where he served as chief scientific officer. Illumina also named Paul Grint to its board of directors. Dr. Grint is currently senior vice president and chief medical officer of Zephyr Sciences, and previously served as vice president and head of clinical R&D for Pfizer in La Jolla, California.

Bryan Koontz has been promoted to senior vice president and general manager of discov-ery informatics at Tripos (St. Louis, MO, USA). He succeeds Trevor Heritage, who is leaving to pursue other interests. The cofounder, and formerly CEO, of Optive Research, Mr. Koontz joined Tripos as vice president of marketing and corporate development when Tripos acquired Optive in January 2005.

John T. Henderson has been elected chair-man of the board of Myriad Genetics (Salt Lake City, UT, USA). He succeeds Dale Stringfellow who recently died of complica-tions associated with pancreatic cancer. Dr. Henderson, a board member since March 2004, was previously with Pfizer for over 25 years, most recently as a vice president in the pharmaceuticals group. In addition, Myriad’s board of directors has selected Linda S. Wilson to fill the vacancy on the audit com-mittee left by Dr. Stringfellow’s passing. Dr. Wilson has been a director since 1999. She was the seventh president of Radcliffe College, and served as vice president for research at the University of Michigan, and in similar roles at the University of Illinois and Washington University.

Topigen Pharma-ceuticals (Montreal, Quebec, Canada) has announced the appointment of G. John Mohr as chief business officer. Mr. Mohr was most recently a corporate

officer and vice president of business develop-ment and licensing at AtheroGenics, and pre-viously president, US operations at Fournier Pharma.

Protagen (Dortmund, Germany) has appointed Stefan Müllner as chief scientific officer, succeeding Helmut E. Meyer. Dr. Müllner previously served in positions in R&D management at Henkel and Hoechst. Dr. Meyer will join Protagen’s scientific advi-sory board and the board of directors, replac-ing Achim Riemann, former CEO of Arthur D. Little Germany.

Tercica (S. San Francisco, CA, USA) has appointed Chris E. Rivera to the newly created position of senior vice president, commer-cial operations. He joins the company from Corixa, where he was vice president of sales. Mr. Rivera also served as senior vice president of Genzyme Therapeutics.

August M. Watanabe has been named to the board of directors of Ambrx (San Diego, CA, USA). Dr. Watanabe was formerly a senior executive at Eli Lilly, where he occupied the positions of executive vice president of sci-ence and technology and president of Lilly Research Laboratories, and was a member of the board of directors.

Keith Yamamoto has been appointed as chair-person to the scientific advisory board of Sirna Therapeutics (Boulder, CO, USA). Dr. Yamamoto is currently executive vice dean at the University of California, San Francisco School of Medicine. He has been a member of the UCSF faculty for more than 25 years, serving as director of the biochemistry and molecular biology graduate program from 1988 to 2001 and as chairperson of the depart-ment of cellular and molecular pharmacology from 1994 to 2003.

P E O P L E©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

ebio

tech

nolo

gy

Date post:	08-Dec-2016
Category:	Documents
Upload:	dangnhan
View:	264 times
Download:	7 times

NPG Biotechnology Volume 23 Issue 5 May

Documents