volume 30 number 9 september 2012
e d i tor i a l
Visualization of mass cytometry cell
signaling data from 14 cell types
(y axis) identified in primary human
peripheral blood mononuclear cells
treated with 27 kinase inhibitors
(x axis) and 14 stimuli (z axis).
Bodenmiller et al. describe an approach
for multiplexing samples for mass
© 2012 Nature America, Inc. All rights reserved.
cytometry, which facilitates such large-
scale analyses (p 858).
Credit: Erica Savig, Nolan lab.
Gene therapy nears the finish line,
p 807
npg
Foundations get busy, p 818
805 Will the floodgates open for gene therapy?
806 Further confirmation needed
n ews
807 First gene therapy nears landmark European market authorization
808 US funds vaccine centers for biodefense
810 Bayer acquisition spotlights biopesticides
810 Regulatory fog lifts on obesity drugs
812 Amylin’s three-party good-bye
812 Industry cautiously welcomes Supreme Court decision on healthcare overhaul
814 Compulsory license bandwagon gains momentum
814 Biotechs opt for alternative floatation strategy
815 GSK buys partner Human Genome Sciences
815 India’s biosimilar regulations
815 Myriad’s patents redux
816 NIH injects $275 million into undiagnosed diseases and RNA research
816 China’s key R&D programs behind schedule
816 Around the world in a month
817 data page: Drug pipeline: Q212
818 News feature: Patient power
B i oe n trepre n eur
B u i l d i n g a bus i n ess
821 How much risk are you prepared to take?
Mark Van Dyke
op i n i o n a n d comme n t
C O R R E S P O ND E N C E
825 To be or not to be transgenic
826 Broad consent in biobanking
826 An accelerated workflow for untargeted metabolomics using the METLIN database
828 Successful suppression of a field mosquito population by sustained release
of engineered male mosquitoes
Nature Biotechnology (ISSN 1087-0156) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 75 Varick Street,
Fl 9, New York, NY 10013-1917. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 75 Varick Street, Fl 9, New York,
NY 10013-1917. Tel: (212) 726 9335, Fax: (212) 696 9753. Annual subscription rates: USA/Canada: US$250 (personal), US$4,677 (institution), US$5,382
(corporate institution). Canada add 5% GST #104911595RT001; Euro-zone: €202 (personal), €3,713 (institution), €4,634 (corporate institution); Rest of world
(excluding China, Japan, Korea): £130 (personal), £2,400 (institution), £2,990 (corporate institution); Japan: Contact NPG Nature Asia-Pacific, Chiyoda Building,
2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. POSTMASTER: Send address changes to Nature
Biotechnology, Subscriptions Department, 75 Varick Street, 9th Floor, New York, NY 10013-1917. Authorization to photocopy material for internal or personal
use, or internal or personal use of specific clients, is granted by Nature Publishing Group to libraries and others registered with the Copyright Clearance Center
(CCC) Transactional Reporting Service, provided the relevant copyright fee is paid direct to CCC, 222 Rosewood Drive, Danvers, MA 01923, USA. Identification
code for Nature Biotechnology: 1087-0156/04. Back issues: US$45, Canada add 7% for GST. CPC PUB AGREEMENT #40032744. Printed by Publishers
Press, Inc., Lebanon Junction, KY, USA. Copyright © 2012 Nature America, Inc. All rights reserved. Printed in USA.
i
 volume 30 number 9 september 2012

feature
WNT
pate n ts
831 Teva v. AstraZeneca and secret prior art under 102(g)(2)
Sandra Lee & Michael Knierim
834 Recent patent applications in biosensors

Frizzled
N E W S A ND V I E W S
835 Modulating WNT receptor turnover for tissue repair
Therapeutic targeting of adult stem Arie Abo & Hans Clevers
cells, p 835
836 RNA-mediated programmable DNA cleavage
Rodolphe Barrangou
838 Silicon dreams of cells into symbols
Jeremy Gunawardena
841 Research highlights
© 2012 Nature America, Inc. All rights reserved.

computat i o n a l b i o l og y

perspect i ve
842 Navigating cancer network attractors for tumor-specific therapy
Pau Creixell, Erwin M Schoof, Janine T Erler & Rune Linding

research
Understanding teratoma risk from cell
perspect i ve
therapies, p 849
849 Lessons from human teratomas to guide development of safe stem cell therapies
Justine J Cunningham, Thomas M Ulbright, Martin F Pera & Leendert H J Looijenga

ARTICLES
858 Multiplexed mass cytometry profiling of cellular states perturbed by small-
npg

molecule regulators
Bernd Bodenmiller, Eli R Zunder, Rachel Finck, Tiffany J Chen, Erica S Savig,
Robert V Bruggner, Erin F Simonds, Sean C Bendall, Karen Sachs, Peter O Krutzik
& Garry P Nolan
868 Combinatorial discovery of polymers resistant to bacterial attachment
Andrew L Hook, Chien-Yi Chang, Jing Yang, Jeni Luckett, Alan Cockayne,
Steve Atkinson, Ying Mei, Roger Bayston, Derek J Irvine, Robert Langer,
Daniel G Anderson, Paul Williams, Martyn C Davies & Morgan R Alexander

Discovering bacteria-resistant
materials, p 868

nature biotechnology iii


Differentiating stem cells to lung
epithelia, p 876
© 2012 Nature America, Inc. All rights reserved.
Polyethyleneimine as a mucosal
adjuvant, p 883
npg
Dual fertilizer and weed control using
phosphite, p 889
nature biotechnology
volume 30 number 9 september 2012
l etters
876 Directed differentiation of human pluripotent stem cells into mature airway
epithelia expressing functional CFTR protein
Amy P Wong, Christine E Bear, Stephanie Chin, Peter Pasceri, Tadeo O Thompson,
Ling-Jun Huan, Felix Ratjen, James Ellis & Janet Rossant
883 Polyethyleneimine is a potent mucosal adjuvant for viral glycoprotein antigens
Frank Wegmann, Kate H Gartlan, Ali M Harandi, Sarah A Brinckmann,
Margherita Coccia, William R Hillson, Wai Ling Kok, Suzanne Cole, Ling-Pei Ho,
Teresa Lambe, Manoj Puthia, Catharina Svanborg, Erin M Scherer, George Krashias,
Adam Williams, Joseph N Blattman, Philip D Greenberg, Richard A Flavell,
Amin E Moghaddam, Neil C Sheppard & Quentin J Sattentau
889 Engineering phosphorus metabolism in plants to produce a dual fertilization and
weed control system
Damar Lizbeth López-Arredondo & Luis Herrera-Estrella
careers a n d recru i tme n t
894 Theory to practice: real world case-based learning for management degrees
Maria Theodosiou, Jean-Philippe Rennard & Arsia Amir-Aslani
896 people
v
 in this issue

27 kinase inhibitors. The unprecedented scale of the study allowed

Combinatorial discovery of bacteria- the authors to assess the kinase- and cell type–selectivity of inhibi-
resistant materials tors and differential responses of cells collected from eight human
Bacterial infections donors. In principle, the approach could be used to multiplex tens
associated with medical of thousands of samples, facilitating new uses of mass cytometry as a
devices such as catheters, readout for drug screens and genome-wide RNA interference studies.
heart valves and prosthetic [Articles, p. 858] CM
joints pose a substantial
burden of disease, so Engineering fertilization and weed control
preventing such infections
through the use of Phosphorus is an essential and non-
materials that reduce renewable element that is included in
bacterial attachment, and commercial fertilizers to improve crop
© 2012 Nature America, Inc. All rights reserved.

subsequent formation of yields, because most soils lack suffi-

biofilms and growth, is
an attractive option for
intervention. Alexander and colleagues present a high-throughput
microarray approach to screen hundreds of polymeric materials
for those that restrict the adherence and growth of bacteria. They
find a new group of structurally related materials comprising ester
and cyclic hydrocarbon moieties that dramatically reduce the
attachment of the pathogenic species Pseudomonas aeruginosa,
Staphylococcus aureus and Escherichia coli to coated surfaces.
The authors show that the discovered materials, when used to
coat silicone catheters, outperform a commercial silver hydrogel
coating in a mouse infection model, revealing the potential of
these materials for clinical translation. [Articles, p. 868] SJ
Multiplexing mass cytometry
npg
cient phosphate for maximum plant
growth. But as phosphorus is nonre-
newable, new approaches to the provi-
sion of phosphorus for plant growth are
required. Phosphite is a reduced form of
phosphorus that is more soluble than
orthophosphate, but it is not used as
a fertilizer because it inhibits plant growth. López-Arredondo and
Herrera-Estrella report the production of transgenic Arabidopsis
and tobacco that express a bacterial phosphite-oxidoreductase
gene, enabling these plants to use phosphite as the sole source of
phosphorus. The transgenic plants can metabolize phosphite sup-
plied in sterile and nonsterile soil, as well as when it is applied as
a foliar fertilizer. Because they lack the phosphite-oxidoreductase
gene, weeds are inhibited by phosphite, and the authors show that
phosphite could hold promise as a dual fertilization and weed control
system for plants transgenic for this trait. Adoption of this approach

Mass cytometry can has the potential to reduce the amount of nonrenewable phosphorus
currently measure 34 ors
used in commercial farming. [Letters, p. 889] SJ
ibit
parameters on cells. 27
inh

But samples must be

analyzed one at a time,
which hampers large-
scale studies of many
samples across several
conditions. Nolan and
colleagues describe the
use of seven metal ion tags to label cells in each well of a 96-well
plate with a unique combination of tags. This allows the entire plate
to be analyzed in a single instrument run, thereby increasing sample
throughput while leaving 27 parameters to characterize the cells. The
authors harness this multiplexing capability to quantify 14 phosphor-
ylation sites in 1,344 treated samples of primary human peripheral
blood mononuclear cells. The cells are from time-course or dose-
response experiments using different combinations of 12 stimuli and
Written by Kathy Aschheim, Michael Francisco, Susan Jones,
Jason Kreisberg & Craig Mak
Mature lung epithelial cells from stem cells
Until now, cystic fibrosis
research has been ham-
pered by the lack of a
ready supply of lung epi-
thelial cells expressing
the mutated forms of
the cystic fibrosis trans-
membrane conductance
regulator (CFTR) gene that give rise to the disease. Rossant and col-
leagues provide a solution by devising a protocol for differentiating
human pluripotent stem cells to mature CFTR-expressing proximal
airway epithelia and applying it to induced pluripotent stem cells
(iPSCs) derived from cystic fibrosis patients. The protocol retraces
the pathway of in vivo lung development. Pluripotent stem cells are
converted stepwise into definitive endoderm, anterior foregut (by
treatment with fibroblast growth factor (FGF)2 and sonic hedgehog)
and proximal lung progenitor cells (by treatment with FGF7, FGF10,

nature biotechnology volume 30 number 9 SEPTEMBER 2012 vii

 i n t h is iss u e

FGF18 and of bone morphogenetic protein 4). Finally, an air-liquid mixed with viral glycoproteins, PEI
interface culture is used to differentiate the proximal lung progenitor forms large, spherical complexes that
cells into mature epithelial cells expressing the appropriate mark- are taken up by dendritic cells, result-
ers, including CFTR, which is properly localized to the apical plasma ing in the production of high-avidity
membrane and shows cAMP-regulated channel activity. To dem- antibodies and a primarily TH2-
onstrate the utility of their approach in drug screening, the authors biased response in mice. Complexes
generate CFTR-expressing cells from induced pluripotent stem cells of PEI with either a herpes simplex
derived from patients with the most common cystic fibrosis mutation virus type-2 glycoprotein or influ-
(F508del) and show that the phenotypic defect is partly rescued by an enza hemagglutinin administered
experimental cystic fibrosis drug. [Letters, p. 876] KA intranasally are shown to protect
mice against an otherwise lethal viral
challenge. Unlike other experimental
Polyethyleneimine as a mucosal adjuvant mucosal adjuvants, such as cholera
Adjuvants can greatly increase the immunogenicity of vaccines. toxin or α-galactosylceramide, PEI
Although many important pathogens gain entry at mucosal surfaces, triggers the release of double-stranded DNA from dying cells, leading
there are no mucosal adjuvants licensed for use in people. Sattentau and to activation of cytoplasmic DNA sensors and the transcription factor
colleagues now report that polyethyleneimine (PEI), a cationic polymer Irf3, a key regulator of interferon alpha and beta. Future studies will
used for DNA delivery in vitro and in vivo, has the unexpected property determine whether PEI is an effective adjuvant for a wider range of
of functioning as a mucosal adjuvant for viral protein antigens. When antigens in humans. [Letters, p. 883]  JK
© 2012 Nature America, Inc. All rights reserved.

Patent roundup
Next month in
A recent Federal Circuit decision concerning an oft-neglected
prior art provision could have a substantial impact on the
patentability of life science inventions. Lee and Knierim break • Mammalian cells as building blocks in synthetic
down Teva v. AstraZeneca and the future of Section 102(g)(2). biology
[Patents, p. 831] MF • Detecting newly synthesized secreted proteins
• TRICEPS pulls down receptors
Recent patent applications in biosensors.
[New Patents, p. 834] MF • CRISPR processing for predictable gene expression
npg

viii volume 30 number 9 SEPTEMBER 2012 nature biotechnology


Will the floodgates open for gene therapy?
In a matter of days, a momentous event will occur: a gene therapy will, for the first time anywhere in the Western
hemisphere, be available commercially with full marketing approval.
T owards the end of July, the European Medicines Agency’s Committee
for Medicinal Products for Human Use (CHMP) recommended the
approval in the European Union of a treatment called Glybera (alipogene
tiparvovec) from a small Dutch company, uniQure biopharma. Unless mat-
ters get untypically administratively complicated, it should be only a matter
of days before the European Commission endorses this recommendation,
© 2012 Nature America, Inc. All rights reserved.
making the approval official.
Admittedly, the event will probably not compare in terms of impact on
the public’s imagination with the first manned moon landing, the first test
tube baby or even the sequencing of the human genome. But within its own
limited firmament, the first gene therapy to be sold legally in Europe will be
an affair of some significance.
Gene therapy enthusiasts are comparing Glybera’s imminent approval
to the US approval in 1986 of Orthoclone OKT3, the first monoclonal
antibody (mAb) approved for use in humans. Back in those simpler times,
enthusiasts then thought OKT3 would open the floodgates for a plethora
of therapeutic antibodies.
And in a way, it did. But first there was only a trickle. The flood came much
later. In that sense, the comparison between Glybera and OKT3 may well be
appropriate. OKT3 was very much a prototypic antibody therapeutic. It was
the first clinical product of the unpatented Köhler and Milstein technology
that had been described some 11 years earlier. Researchers had developed
mAbs against a vast range of targets and had already shown, preclinically
at least, the exquisite specificity of many of them. Boosters of the technol-
npg
ogy fondly, and correctly, imagined a procession of mAbs for all sorts of
conditions marching upon a market desperate for drugs targeting highly
defined molecular markers. In the end, it did happen…only the timing
was a bit wrong. The problem was, OKT3 sidestepped one fundamental
deficiency of 1986 murine antibody drugs.
OKT3 was approved in an era when clinicians had not defined the limita-
tions that the human-anti-mouse-antibody (HAMA) effect would impose
on the use of mouse-derived antibodies. The basic problem seems obvious
now: in response to mouse antigens, humans mount an immune reaction
that tends, at the least, to severely limit repeated dosing of the same (or
other) mouse antibodies. Not only that, but the immune response itself
can be dangerous.
OKT3 conveniently circumvented this drawback because its target was
CD3 on T cells, the binding of which leads to apoptosis. As a treatment for
the suppression of organ transplant rejection, the murine mAb not only was
used in patients who already were immunosuppressed but also was given as
an acute treatment for the first 7 days after transplantation. Thus, neutral-
izing antibodies were not a problem. HAMA? SchmAMA! OKT3 was OK.
Glybera may be OK, too. It is a gene therapy for lipoprotein lipase (LPL)
deficiency, an ultra-rare inherited disorder that affects no more than two
people per million in the general population. Affected people can’t break
down fat and try to manage their disease by strictly limiting fat in their diet.
nature biotechnology volume 30 number 9 september 2012 805
Editorial
That is difficult, leading to life-threatening pancreatitis in a high propor-
tion of patients. Glybera addresses the underlying LPL deficiency through
multiple intramuscular injections of an adeno-associated virus vector that
delivers functional LPL genes to muscle cells.
Getting Glybera to market has been a struggle. The company that had
started its development—a now-defunct Dutch startup called Amsterdam
Molecular Therapeutics (AMT)—submitted its approval file in December
2009. The gene therapy protocol, which includes repeated administration
of the functional gene and immunosuppression to prevent ‘rejection’ of the
treatment didn’t go down well with regulators.
After nearly two years of head scratching, see-sawing and scientific
bu­reaucracy, the CHMP declared the treatment “non-approvable” in October
2011, causing AMT to halve its workforce to sit out the process. In February,
as AMT was irreversibly bound for liquidation, Dutch investor Forbion
put forward €6 ($7.5) million to found uniQure biopharma as a vehicle to
acquire AMT’s assets, including Glybera.
It subsequently became clear that there was some wiggle room that would
allow Glybera onto the market. The CHMP recommended marketing autho-
rization only under “exceptional circumstances,” under which uniQure will
have to monitor patient outcomes and feed the information straight to the
European Medicines Agency.
Jörn Aldag, the CEO of uniQure (and before that the CEO of AMT),
maintains that the approval of Glybera puts gene therapy “where [mAbs]
were in the late 1990s, when they were just visible, but tiny.” He predicts a
“steep growth curve” (p. 807).
He may well be right. The relatively rich pipeline of gene therapy candi-
dates already in human trials suggests there may be a surge in the number
of gene therapies approved over the next few years.
What is less certain, however, is whether there will be a noticeable
flood of patients treated. Many of the gene therapies in clinical develop-
ment are treatments for very rare, single-gene deficiencies: Leber’s con-
genital amaurosis, severe combined immunodeficiency, Wiskott-Aldrich
syndrome and the like. Among these Mendelian disorders, the hemophilias—
themselves hardly common conditions—represent a pinnacle of frequency.
Even for these kinds of conditions, there may be HAMA moments in the
future—potential immune attenuation of efficacy, safety issues related to
constructs or vectors, not to mention competition from existing enzyme
replacement therapies that have been available for many years. For more
complex illnesses—and gene therapy is being explored for heart disease
and cancer, too—it remains far from clear that the technical challenges
will be quickly overcome.
So Glybera’s approval is unlikely to augur a flood of gene therapy drugs
for blockbuster markets that transform the drug business in the manner
of mAbs. But even if gene therapies turn out to be nichebusters rather
than blockbusters, their successful commercialization is a remarkable
achievement nonetheless.
 editorial
Further confirmation needed
A new mechanism for independently replicating research findings is one of several changes required to improve the
quality of the biomedical literature.
O ver the past year, the reputation of the biomedical literature has
taken a bit of a beating. Controversy has centered around studies by
researchers at Bayer Pharmaceuticals and Amgen, which independently
have shown that a troublingly high number of papers in certain areas
of translational research cannot be replicated. Last month, as a response
to this problem, the Palo Alto, California–based Science Exchange
announced the launch of the Reproducibility Initiative (http://www.
reproducibilityinitiative.org/). This initiative is one way in which online
© 2012 Nature America, Inc. All rights reserved.
platforms can facilitate rapid and independent corroboration of published
results. But major progress in improving the reproducibility of research
will likely require more sweeping changes to the way in which science is
published and validated.
Published research findings are often modified or refuted by subse-
quent evidence. There is nothing unusual about this; it is the way scientific
knowledge progresses. But there is an increasing concern that publication
bias toward positive results, rising competition to rush findings into print,
an overemphasis on publishing conceptual breakthroughs in high-impact
journals and a lack of incentives for academic researchers to retract irre-
producible findings may be increasing the incidence of false claims in the
literature. All of which has implications for translational research.
Misleading papers result in considerable expenditure of time, money
and effort by researchers following false trails. This affects the careers
of postdocs and academics. It affects companies and investors, present-
ing yet another barrier for the translation of academic discoveries into
new medicines by diverting funds away from real advances. And most
troublingly, it can result in patients being exposed to drugs on the basis of
npg
wrong information. In the past year, two studies have brought into sharp
focus just how bad the problem may be.
In September 2011, a team of researchers at Bayer provided a retro-
spective survey of four years of work in oncology, women’s health and
cardiovascular target validation (Nat. Rev. Drug. Discov. 10, 712, 2011).
They asked 23 of their R&D scientists to tally papers they’d acted upon
and whether or not the findings had panned out. The analysis revealed
that only ~20–25% of the relevant published data could be corroborated
internally.
Five months later, a collaboration involving Amgen scientists published
the results of their efforts to replicate findings from recent publications in
the clinical oncology literature (Nature 483, 531–533, 2012). The data were
disturbing. Of 53 papers, only 6 (11%) were reproducible. A particularly
troubling aspect was the disclosure that in return for cooperation, several
of the authors required the company to sign a confidentiality agreement
preventing the identity of their paper from being revealed.
Which brings us to the Reproducibility Initiative.
The effort takes advantage of Science Exchange’s existing network of
>1,000 core facilities and contract research organizations. After authors of
an original publication submit their study design, the initiative matches the
work to qualified facilities (the identity of which is masked from authors),
which then attempt to replicate the studies for a fee. Once the results are
806 volume 30 number 9 September 2012 nature biotechnology
returned, the initiative’s advisory board determines whether the study
has earned a ‘certification of reproducibility’ (authors cannot appeal the
board’s decision).
It is entirely the author’s prerogative as to whether the findings are writ-
ten up; if they are, they can be published in a special section of PLoS
ONE. Nature Publishing Group and Rockefeller University Press have also
agreed to link from the original publication to the PLoS ONE paper, and
Figshare (http://figshare.com/) will host the data from the verification.
Whether authors will elect to publish findings that fail to replicate their
original results is unclear—if they don’t, then the arrangement clearly fails
to address the problem of research reproducibility.
Another big question is, who pays for the work? According to Science
Exchange, initially authors will be paying. But this is less than optimal,
given the competing interest and the scarcity of research funding.
A different scenario would be for funding agencies to bankroll the
effort. Alternatively, tech transfer offices—or some other, richer part of
the author’s institution—could support validation work as a means of
making academic assets more attractive to potential licensing partners.
With the current vogue for ‘capital efficiency’, perhaps venture capitalists
and pharma companies will use the service if it’s cheaper than replicating
work in-house.
So what about the other papers, validation of which won’t find support
from investors, companies, institutions or funders?
Online commenting on papers—available on Nature for the past 2.5
years but yet to be rolled out to other Nature research journals—remains
an easy way for the community to highlight problematic studies. More
could also be done during the peer review process in terms of rejecting
papers in which only ‘representative’ data are reported or proper statistical
analysis is not used (Nat. Neurosci. 14, 1105–1107, 2011).
Journal editors and reviewers can also do more to ensure that the rel-
evant information is captured about an experimental protocol and condi-
tions, instrument settings and parameters. Too often, there is more than
a little secret sauce to the procedures followed in a laboratory. Detailed
description and the publication of full protocols (together with videos
depicting experiments) could help.
Perhaps most importantly, different research communities need to
come together to address particularly troublesome research questions in
their field; in oncology, for example, a clear need exists for guidance on
the preclinical cancer models to use in particular settings and to address
particular questions.
Clearly, more should be done to increase the quality of published work.
The Reproducibility Initiative, and other efforts like it (e.g., http://www.
sciencecheck.org/), will help by validating results and ensuring that sup-
porting data are placed in openly accessible repositories. Greater atten-
tion also needs to be paid during peer review to the completeness of the
experimental protocol disclosed, the supporting data and the robustness of
the authors’ analysis. Most of all, a change in publication culture is needed.
Sometimes replication is as important as discovery.
 news
in this section
Industry cautiously Biotechs opt for GSK buys partner
welcomes Supreme alternative floatation Human Genome
Court decision on strategy p814 Sciences p815
healthcare
overhaul p812

First gene therapy nears landmark European market authorization

Some 22 years after the first US Food and
Drug Administration–approved trial of
a gene therapy, a genetic medicine for the
ultra-rare inherited disorder lipopro-
tein lipase (LPL) deficiency is poised for
final approval in Europe. If the European
Medicines Agency (EMA) gives its formal
authorization at the end of this month,
the LPL deficiency treatment, alipogene
tiparvovec (Glybera), will be the first gene
© 2012 Nature America, Inc. All rights reserved.
and clinically important reduction in the
occurrence of acute pancreatitis, the most
insidious symptom of LPL deficiency. Given
the length of response, Aldag believes there
should be annualized payments in the
same range as enzyme-replacement thera-
pies, which patients receive on a consistent
basis. Currently in Europe, those types of
medicines are priced at €150,000–450,000
($187,000–561,000) per annum. (LPL defi-

therapy to be approved for sale in a Western
market. Although the challenges of admin-
istering and monitoring novel gene thera-
pies will likely limit clinical applications to
indications poorly served by conventional
medicines in the near term, the decision
potentially opens the way for a renaissance
in genetic medicines.
Approval will also be a watershed moment
from a regulatory standpoint, as the EMA’s
assessment of alipogene has followed pro-
tracted deliberations in which the file was
examined and voted on four times, and the
conventions—if not the rules—governing
EMA procedures were thrown out of the
window (Box 1).
“It’s been a long ride, but an important
one, and the outcome is terrific,” says Jörn
npg
ciency cannot be treated with a replacement
therapy because the enzyme’s half-life is too
short.)
Similar pricing would make alipogene the
most expensive approved medicine of all time
for a single treatment, so it bears watching
how this will play out with pricing and reim-
bursement bodies. In addition, there are
onerous post-marketing requirements that
will ensure patients who have received ali-
pogene continue to be followed. uniQure will
Fat particles (pictured) accumulate in the
have to maintain a patient registry and put in blood stream of people with lipoprotein lipase
place a risk management plan under which deficiency. Source: uniQure
no more than two patients per month will be
treated, at a limited number of expert centers.
The company intends to follow-up its cost uniQure dear, but it also, “represents a
success with the EMA by submitting the very substantial investment on the regulatory
alipogene file to the US Food and Drug side,” which has parallels with the amount of

Aldag, CEO of Amsterdam-based uniQure,
the developer of alipogene tiparvovec.
But guiding the gene therapy through
the byzantine and meticulous European
regulatory process has been expensive.
uniQure (formerly Amsterdam Molecular
Therapeutics) invested €50 ($62) million
on the development of alipogene, including
€15 ($19) million spent after the product
was first submitted to EMA in December
2009. Though the extremely low incidence
of LPL deficiency (1–2 afflicted per million
people) would seem to leave uniQure with
little chance of recouping this investment,
the many profitable approved products for
rare diseases have blazed a pricing path ali-
pogene can follow, though it needs tweaking.
Alipogene, a recombinant adeno-associated
viral (AAV) vector expressing the Ser447X
variant of human LPL, is given in a single
sitting by means of several injections into
the leg muscles, resulting in a long-term
Administration (FDA) and Health Canada.
Based on talks with those regulatory agen-
cies, Aldag says he’s “pretty certain they will
accept the same data.”
If alipogene can clear the final hurdle of
formal market authorization this month, it
will be a landmark for others in the gene
therapy field (Table 1), providing valida-
tion that a gene therapy can indeed be com-
mercialized. According to Aldag, the field is
primed for growth. “Gene therapy is at the
position where monoclonal antibodies were
in the late 1990s, when they were just visible,
but tiny. It will now follow a similar, very
steep, growth curve,” he predicts.
Jeff Ostrove, president and CEO of a
US gene therapy specialist, Ceregene, of
San Diego, agrees with this assessment.
“[Alipogene] is a very positive dry run for
the entire field and really in many ways is
the very positive tip of an iceberg.” Ostrove
notes that reaching this point has not only
time and effort the FDA and other agencies
put into establishing the safety of monoclo-
nal antibodies (mAbs) and biopharmaceu-
ticals. “That work by the regulators is very
much paying off in terms of getting approvals
[of mAbs] now,” Ostrove says.
Sander van Deventer, co-founder of
uniQure’s forerunner, Amsterdam Molecular
Therapeutics, and who earlier in his career
was the first to administer an anti-tumor
necrosis factor (TNF)-a mAb, to patients
with Crohn’s disease, is in a unique position
to compare the regulators’ views of mAbs
in the 1990s versus today’s gene therapies.
That pilot antibody, Remicade (infliximab),
secured FDA approval in 1998, but “the
attitude of many people to [mAbs] prior to
1995 was exactly the same as to gene ther-
apy today. Now that’s going to change,” van
Deventer says.
The road to approval for alipogene has not
been easy. After the EMA’s second rejection

nature biotechnology volume 30 number 9 SEPTEMBER 2012 807

 NEWS

in brief Table 1 Selected companies with gene therapies in clinical development

Company Product Indication Clinical status
US funds vaccine centers for uniQure Alipogene (Glybera) Lipoprotein lipase Pending approval
biodefense deficiency
AMT-060 Hemophilia B Phase 1/2
The US Department of Health and Human
Services (HHS) has allocated $400 million Advantagene (Auburndale, ProstAtak (AAV expressing herpes Prostate cancer Phase 3
for three Centers for Innovation in Advanced Massachusetts) simplex virus thymidine kinase Glioma Phase 2
combined with valacyclovir)
Development and Manufacturing for vaccine Pancreatic cancer Phase 1/2
production, but some are concerned that
GlaxoSmithKline (London) 2696273 (GIADAl retroviral vector ADA deficiency in Phase 3
excessive focus on influenza may leave the encoding adenosine deaminase severe combined
nation less prepared against other threats. (ADA) for ex vivo hematopoietic immunodeficiency
A 2010 report prepared by HHS proposed stem cell therapy) (SCID)
establishing such centers to accelerate Vical (San Diego) Allovectin (4,853-bp DNA plasmid Melanoma Phase 3
development of medical countermeasures encoding human MHC class I
against chemical, biological, radiological HLA-B7 and chimpanzee b-2-
and nuclear threats, develop new vaccine microglobulin plus cytofectin
production technologies, and enable rapid DMRIE/DOPE)
manufacture of existing vaccines. On June 19, Applied Genetic AGTC-0106 (AAV vector carrying Alpha-1 antitrypsin Phase 2
the HHS Biomedical Advanced Research and Technologies (Alachua, alpha-1 antitrypsin gene) deficiency
Development Authority announced allocations Florida)
of $163 million for a center in Maryland, Bluebird Bio Lenti-D (lentiviral vector encod- Adrenoleukodystrophy Phase 2
$60 million for a center in North Carolina ing human ATP binding cassette
and $176 million for a center in Texas. Each subfamily D for ex vivo therapy of
center partners with companies—Novartis’s
© 2012 Nature America, Inc. All rights reserved.

hematopoietic stem cells)

vaccine facility in Holly Springs, North Carolina, Lentiglobin (lentiviral vector encod- Beta thalassemia Phase 2
GlaxoSmithKline’s Vaccines of Marietta, ing human beta-globin gene or a
Pennsylvania, and Emergent BioSolutions hybrid A-gamma/beta-globin for
Manufacturing in Baltimore, with researchers ex vivo hematopoietic stem cell
at Texas A&M (College Station), Michigan State gene therapy)
University (Flint), Duke (Durham) and North Celladon (San Diego) Mydicar (AAV1 encoding Cardiomyopathy Phase 2
Carolina State (Raleigh)—to accelerate response sarcoplasmic reticulum ATPase 2a)
to future pandemics. Ceregene Cere-120 (AAV2 encoding Parkinson’s disease Phase 2
Maj. Gen. (ret.) Philip Russell, former director neurturin)
of the Walter Reed Army Institute of Research,
Cere-110 (AAV2 encoding nerve Alzheimer’s Phase 2
an expert consulted by HHS, expressed growth factor) disease
disappointment in the implementation.
Genzyme (Cambridge, AAV-hAADC-2 (AAV2 encoding Parkinson’s disease Phase 2
“They combined influenza preparedness with
Massachusetts) the human aromatic l-amino acid
biodefense against very strong recommendations
decarboxylase gene)
from me and a lot of other people,” he says.
“They’re totally unrelated problems, and Oxford BioMedica ProSavin (equine infectious anemia Parkinson’s disease Phase 2
virus encoding aromatic l-amino (complete)
totally unrelated issues.” When this plan
acid decarboxylase, tyrosine hydrox-
was devised, more funds were available for
ylase and GTP cyclohydrolase I
preventing flu outbreak than for bioterrorism,
making this emphasis an expedient strategy. Shenzhen SiBiono Gendicine (replication-defective Lung cancer Phase 2
GeneTech (Shenzen, China) adenovirus vector encoding human
npg

However, he believes that supporting efforts

P53)
to mass produce and distribute flu vaccine
The US National Institutes of Health Clinicaltrials.gov website lists 2,965 gene therapy trials, of which 1,103 are listed as
will do little to facilitate R&D against novel recruiting. The vast majority of these are being run by academic investigators. AAV, adeno-associated virus; MHC, major histo-
threats. “We envisioned a single organization compatibility complex; DMRIE, 1,2-dimyristyloxypropyl-3-dimethyl-hydroxyethylammonium bromide; DOPE, dioleyl phosphati-
totally dedicated to advanced development and dyl ethanolamine.
manufacturing of biodefense vaccines,” says
Russell. “But it seems like technical capabilities
were secondary to lobbying power and decision of the gene drug in November 2011, uniQure [gene therapy] doses are so tiny, and this means
making.” Michael Eisenstein stopped all investment in the product and you won’t need the same multimillion dollar
abandoned the clinical and manufacturing scale of investment in manufacturing facilities.”
in their words infrastructure. The company is now rebuild-
ing, and Aldag says he hopes the first patient
Alipogene is most obviously a role model
for other gene therapies for monogenic
“It could be the to receive a commercial gene therapy will be inherited disorders, but it also paves the
biggest upside treated in the first half of next year. way for products in broader, multifactorial,
surprise in the history
On a broader level, there remains more noninherited diseases, such as ProSavin,
of the pharmaceutical
and biotech industry.”
spadework to do in terms of agreeing upon developed by Oxford, UK–based Oxford
Following the flop reimbursement, convincing physicians and BioMedica, which delivers genes for the three
of Pfizer/J&J’s others to push through the wider adoption enzymes that are needed to generate dopa-
bapineuzumab’s of gene therapy, admits Ena Prosser, partner mine, the neurotransmitter that is depleted
in August, ISI at Fountain Healthcare Partners of Dublin, a in Parkinson’s disease. “We are basking in
Group analyst Mark
venture capital firm that has invested in gene reflected sunshine because there are now
Schoenebaum ponders
the likely aftermath were Eli Lilly’s Alzheimer’s therapy companies. Even so, she notes one more ticks in the box for gene therapy,” says
anti-b-amyloid antibody, solanezumab, advantage some gene therapies will have—com- Stuart Naylor, the company’s CSO.
approved. (Bloomberg, 8 August 2012) pared with mAbs and in terms of uptake, “the Although alipogene is targeted at an

808 volume 30 number 9 SEPTEMBER 2012 nature biotechnology

 news

Box 1 A regulatory marathon

The alipogene file was submitted in December 2009 and suffered its first rejection
by the EMA’s Committee for Medicinal Products for Human Use (CHMP) on June 23,
2011, after experts on another EMA body, the Committee for Advanced Therapeutics
(CAT) questioned its clinical efficacy, as only 27 patients had been treated, though it
acknowledged that the product’s AAV vector and associated manufacturing protocol met
safety standards. When uniQure requested a reexamination of the file, CAT suggested
its concerns could be addressed by post-marketing surveillance, but the CHMP was not
convinced, and rejected alipogene for a second time in November 2011.
At this point, uniQure considered the product dead, especially given the failures of
other companies, such as Ark Therapeutics and its adenoviral gene therapy sitimagene
ceradenovec (Cerepro) in malignant glioma, to negotiate the EMA approval process. But
in January 2012, the European Commission, with no official public explanation, told
the committee to think again. These deliberations resulted in a 16–15 majority in favor
of approving alipogene when the CHMP voted for the third time in April. Although the
vote was favorable, because 1 of the 32 CHMP members was absent on the day of the
vote and the rule book requires an absolute majority of 17 positives, alipogene again
got a thumbs down.
At this point, the EMA rewrote the rules, and rather than referring this decision to the
European Commission, told both CAT and the CHMP to take another look. Rules were
© 2012 Nature America, Inc. All rights reserved.

further breached because even though the official EMA line is that it won’t look at new
information when reexamining a file, it did weigh further follow-up data. This extended
the patient data considered from 2.4 years to more than 4 years when the file was
scrutinized for the fourth time in July. NM

extremely small group of patients, Naylor uniQure was able to negotiate repeated
believes the level of internal debate about rejections of alipogene by the EMA only
the treatment among the EMA’s various because its investors kept the faith; in truth,
committees and expert groups means the the gene therapy field has never been flush
agency is better suited to provide guidance with cash. That is beginning to change, how-
on the clinical development of other gene ever, as clinical trials deliver compelling
therapies. “This means better reassurance evidence of long-lasting effects, as seen with
that you don’t invest money in the wrong ProSavin, where Parkinson’s disease patients
place,” he says. have seen sustained improvements in symp-
Similarly, Nigel Parker, CEO of gene toms and in some cases were able to reduce
therapy startup, FKD their doses of L-dopa.
Therapies OY of Kuopio, This increased finan-
npg

Finland, says, “It’s clear uniQure was able to cial interest was exempli-
now that if you have got negotiate repeated fied in July when Bluebird
the right clinical data, rejections of alipogene Bio, of Cambridge,
all the technical barriers Massachusetts, raised
and safety concerns over
by the EMA only $60 million in an oversub-
gene-based medicines because its investors scribed Series D financ-
[have been addressed] kept the faith; in truth, ing.The backers are a roll
and the products are the gene therapy field call of blue chip investors,
approvable.” with five new funds join-
In addition, alipogene has never been flush ing the four existing ven-
is important in high- with cash. ture capitalists, and—in a
lighting the potentially sign that gene therapy is
superior therapeutic power of gene therapy edging further toward becoming a commer-
over existing treatments, Parker says. For cial proposition—Shire, of Dublin, made a
example, FKD’s phase 2-ready bladder can- strategic investment also.
cer treatment Instiladrin delivers the gene With many imponderables remaining—pric-
encoding interferon-a2b, prompting the ing, partnering and the number of patients—it
endogenous generation of the anti-cancer is too soon to say what the financial returns will
drug for seven days after a single treatment. be, but says Aldag, “We can tell people we’re
In comparison, when interferon-a2b is No. 1; we’ve done it. Gene therapy is something
injected directly, it is cleared from bladder the EMA permits us to bring to patients.”
cells within 24 hours. Nuala Moran, London

nature biotechnology volume 30 number 9 SEPTEMBER 2012 809

 NEWS

in brief Regulatory fog lifts on obesity drugs

Bayer acquisition spotlights
biopesticides
The July acquisition of AgraQuest, a 16-year-old
biopesticide company, by Bayer CropSciences
(BCS), provides validation to a sector that has
attracted interest from major agbiotech firms.
The Monheim am Rhein, Germany–based
agriculture giant agreed to acquire AgraQuest
for $425 million plus milestones. This follows
a series of deals between the biotech and large
agriculture firms, such as BASF and Monsanto,
and Bayer’s 2010 acquisition of the Israeli
biopesticide firm AgroGreen. The news is
welcome to the Davis, California, area, where
the company’s R&D site is expected to expand.
The acquisition of AgraQuest gives BCS
access to the company’s ‘green’ product lines,
which will be integrated into its platforms for
crop protection and pest control. AgraQuest
touts its line of fungicides and insecticides as
comparatively immune to resistance problems.
Standard single-chemical agents attack only
© 2012 Nature America, Inc. All rights reserved.
In the span of a summer month, the US Food metic inhibitor of prohibitin-1 receptor, adi-
and Drug Administration (FDA) approved its potide, in phase 1 testing. “When we licensed
first two obesity drugs in 13 years, a period adipotide from MD Anderson a year and a
in which it rejected any weight loss drug put half ago, big pharma didn’t want to talk about
before it and withdrew products from the [partnering] because of regulatory struggles.”
market due to psychiatric and cardiovascular There is a striking difference “between now
safety concerns. The recent approvals have and even six months ago as to how big pharma
rejuvenated a field that had been thin on hope companies are approaching the obesity mar-
and brought pharma and investors back into ket,” he says.
the space (Table 1). The increased attention is welcome, but
That’s not to say safety concerns around for obesity drugs to thrive, they must move
obesity drugs have been put to rest. Indeed beyond the mistakes and modalities of the
the two drugs—Belviq (lorcaserin) from past. Fenfluramine and dexfenfluramine—two
San Diego-based Arena Pharmaceuticals, nonspecific serotonin (5-hydroxytryptamine;
approved June 27, and Qsymia (topira- 5-HT) receptor antagonists—were pulled from
mate/phentermine) from Mountain View, the market in 1997 after being associated with
California–based Vivus and partner Tokyo- valvular heart disease.
based Eisai Pharmaceuticals, approved July Belviq is a derivative of nor-dexfenflura-
17—were delayed by nearly two years, in part mine. Unlike its parent molecule, it selec-

one pathway, whereas microbiological agents
produce a variety of active compounds.
AgraQuest’s flagship product, Serenade,
contains a patented strain of Bacillus subtilis
that produces more than 30 active lipopeptides,
according to company literature. These protect
against a variety of fungi and bacteria, including
fire blight, botrytis, sour rot, rust, sclerotinia,
powdery mildew, bacterial spot and white mold.
Mark Faust, a consultant and principal
of Cincinnati-based Echelon Management
International, sees value in companies that
establish synergies among approaches versus
those trying to create magic bullets. “Magic
bullets don’t work forever. Evolution happens.
Things change. I think these synergistic
approaches tend to last longer and be more
resistant to evolution,” he says. Jim Kling
npg
owing to concerns about cardiovascular side tively activates 5-HT2C receptors rather than
effects. But Thomas Hughes, president and the 5-HT2A and 5-HT2B receptors, which are
CEO of Cambridge, Massachusetts–based thought to be involved in hallucinogenic side
Zafgen, which is developing a drug for obesity, effects and heart valve defects, respectively.
says the “fog has lifted.” Qsymia is an extended release formulation
“There was a period of many years of fail- of two existing drugs, topiramate and phen-
ure in the pharma space, unwillingness in the termine. Both Belviq and Qsymia target the
regulatory space to approve obesity drugs, central nervous system (CNS) and are attached
and there were companies struggling to get to extensive post-marketing cardiovascular
approval. I would meet people who said they studies. Another drug, Contrave (naltrexone/
would never invest in obesity,” Hughes says, bupropion) from La Jolla, California–based
adding that now the investor climate is “abso- Orexigen Therapeutics, also targets the CNS,
lutely different.” and the FDA required a cardiovascular out-
Big pharma began showing interest in the comes trial to consider approvability because
winter, after the FDA Advisory Committee of an increase in blood pressure and pulse rate
voted 20–2 on February 22 in favor of approv- seen in its clinical program. The company
ing Qsymia, says Christopher Anzalone, expects to resubmit its New Drug Application
president and CEO ofObesityPasadena, trends
CA–basedamond
(NDA)U.S.basedadults
on an interim analysis of that

in their words Arrowhead Research, which has BRFSS, 1990, 2000, 2010 second half of 2013.
a peptidomi- safety trial in the
(*BMI ≥30, or about 30lbs. overweight for 5´4˝ person)
“We employ 5,000
1990 2000
people in China
already, a few years
ago there were only a
few hundred. …And
that’s good for Basel.”
Severin Schwan,
Roche’s CEO offers
scant consolation for 2010
the thousand made
redundant at Roche’s
shuttered R&D center in Nutley, New Jersey.
(Sonntags Zeitung, 29 July 2012)
“The point is not for doctors to castigate people,
but to understand how people are responding
to their treatments.” George Savage, co-founder
of Proteus Digital Health of Redwood City,
Calif., describes the potential of his company’s
No data <10% <10–14% 15–19% 20–24% 25–29% ≥30%
recently approved pill/sensor, which signals
a smart phone after ingestion, to impact
compliance.. (International Science Times,
Obesity trends among US adults in 1990, 2000 and 2010. Obese is characterized as a body mass
1 August 2012)
index (BMI) >30. Source: Center for Disease Control, Atlanta

810 volume 30 number 9 SEPTEMBER 2012 nature biotechnology

 news

Table 1 Selected compounds in development for obesity—phase 2 and beyond

Company name
Orexigen Therapeutics
Novo Nordisk (Bagsvaerd, Denmark)
Shionogi (Osaka)
TransTech Pharmaceuticals, High Point, NC HPP404 (small-molecule antagonist)
NT Life Sciences (Kadmon & Nano Terra
joint venture)
7TM Pharma (Horsholm, Denmark)
*Not in active development.
Product (modality) Molecular target Latest stage of development
Contrave naltrexone SR/bupropion SR Mu opioid receptor (OPRM1) (MOR) Registration
(small-molecule antagonist)
Victoza liraglutide (peptide agonist) GLP-1 receptor Phase 3
Velneperit (small-molecule antagonist) Neuropeptide Y receptor Y5 (NPY5R) Phase 2b (in Japan)
Histamine H3 receptor (HRH3) Phase 2
KD026 (small-molecule antagonist)* Microsomal triglyceride transfer protein (MTP) Phase 2
Obineptide (small-molecule Neuropeptide Y receptor Y2 (NPY2R) and Phase 1/2
antagonist)* Neuropeptide Y receptor Y4 (NPY4R)
TM30339 (small-molecule Neuropeptide Y receptor Y4 (NPY4R) Phase 1/2
antagonist)*

To some degree these side effects are to
be expected in drugs targeting the CNS. To
move beyond that “the next step tends to be
taking lower doses of those drugs and com-
bining them together to increase efficacy
© 2012 Nature America, Inc. All rights reserved.
the production of new fatty acid molecules by
the liver and helping convert stored fats into
energy. Combined data from two phase 1 stud-
ies in 28 obese women showed that beloranib
resulted in 4.3 kg weight loss versus 0.6 kg
Tartaglia, who also is a partner at Third
Rock Ventures in Boston expects to see more
investment, including from his firm, in the
obesity space. Beyond Ember, Third Rock
has invested in Zafgen and Boston-based

while decreasing side effects—we saw that in
hypertension and it was a step forward,” says
Frank Greenway, professor and chief of the
outpatient clinic at Pennington Biomedical
Research Center in Baton Rouge, Louisiana.
“That’s kind of where we are now with obesity
drugs. Where we want to get to is drugs that
act on the periphery, where the disease is, like
angiotensin receptor blockers in hyperten-
sion.”
The newer obesity drugs still in devel-
opment plan to do just that. Arrowhead
Research’s peptide adipotide has two func-
tional domains: one targets prohibitin-1
receptor, which is found on the surface of
adipose vascular endothelial cells but not
on the surface of other cells, and the sec-
ond causes apoptosis by disrupting mito-
npg
weight gain for placebo over 25 days.
Zafgen’s Hughes says the weight loss seen
in those trials, as well as preclinical studies
in dogs, was about 1 kg/week and was linear.
The company plans to start a phase 2 trial of
the compound in 100–120 obese patients in
September. Zafgen, however, has the addi-
tional hurdle of being a first-in-class medi-
cine, and it will need to blaze a regulatory
path of its own. “So we need to figure out
how to do that without spending half a bil-
lion dollars and taking seven years,” he notes.
Steven Smith, co-director of transla-
tional research at the Sanford-Burnham
Medical Research Institute’s Diabetes and
Obesity Center in Orlando, Florida, says
MetAP2 “is a target I have a lot of enthusi-
asm around right now,” though he also likes
Rhythm Pharmaceuticals, whose RM-493,
a small-peptide melanocortin 4 receptor
(MC4R) agonist, is in phase 1 testing. The
trick to investing in these companies, he
says, is to “derisk early.” The firm spent about
“one and a half years using seed money to
do studies to make sure there wouldn’t likely
be cardiovascular risks before committing
$34 million in a series A round last December”
to Ember, Tartaglia says.
There are safety risks beyond cardiovascular,
of course. 7TM Pharma of Lungby, Denmark,
is focusing on TM38837, an antagonist of
peripheral cannabinoid receptor type 1 (CB1).
At least four CB1 antagonists—including two
that had completed phase 3 testing and a third
that was in phase 3 testing—were shelved after
the approach was shown to increase risk of

chondrial membranes inside the cells.
The compound thus selectively destroys
the blood supply that supports the growth
of unhealthy white fat. In one preclinical
study, adipotide decreased the body weight
of obese mice by 30% after 28 days of daily
injections, Anzalone says, but there was no
effect on lean animals, “so there is no con-
cern about wasting away if a skinny person
takes the drug.” Data from the phase 1 trial
in 39 obese prostate cancer patients, testing
tumor aggressiveness against weight loss, are
expected in mid-2013.
Zafgen’s small-molecule beloranib also
circumvents the CNS. The compound is an
inhibitor of methionine aminopeptidase 2
(MetAP2). In obese people, the liver normally
packages circulating lipids as fat and trans-
ports that fat to adipose tissue. MetAP2 inhi-
bition directly affects metabolism by reducing
Ember Therapeutics’ irisin variants. Irisin is
a naturally occurring hormone that is pro-
duced in muscle cells during exercise. The
Boston-based company’s founders, including
Bruce Spiegelman at Harvard, Cambridge,
Massachusetts, discovered that irisin also
converts white fat cells into brown fat, which
burns calories into heat. When fed a high-
fat diet, mice with overexpressed irisin had
lower weight gain than mice with normal iri-
sin levels (Nature 481, 463–358, 2012).
Ember’s interim CEO Louis Tartaglia says
the company aims to begin phase 1 trials of
an irisin variant in 2014, and he thinks the
way to battle obesity is through polyphar-
macology. Ember’s drugs work on energy
expenditure, and “a drug working on energy
expenditure like our drug plus an appetite
suppressant [such as Qsymia or Belviq] will
ultimately be the way to treat obesity,” he says.
psychiatric disorders, including suicidality.
Christian Elling, vice president of development
for 7TM, says the psychiatric effects were due
to the compounds’ actions of CB1 in the brain,
whereas the weight loss effects were due to both
CNS and peripheral receptors. TM38837 is
currently in phase 1 trials.
When TM38837 moves into phase 2 will
depend on financing, which is the company’s
biggest challenge for developing the product.
Elling is hopeful that the two recent approv-
als will benefit 7TM. “For our compound, we
have the whole CB1 history to contend with
and we need to convince investors that this
is a good idea,” he comments. The approv-
als “will not necessarily make it easy to get
the financing, but they will help. We are still
far away from achieving standard-of-care in
obesity.”
Aaron Bouchie, Ithaca, New York

nature biotechnology volume 30 number 9 SEPTEMBER 2012 811

 NEWS
in brief
Amylin’s three-party
good-bye
The buyout of Amylin Pharmaceuticals by
Bristol-Myers Squibb (BMS) completed in
August called for BMS to pay $31 per share
and gather up ownership of Amylin for about
$5.3 billion. But the uniquely structured
deal also brought in a third party (London-
based AstraZeneca) and added a $1.7-billion
payout by BMS to cover both Amylin debt and
a contractual obligation to Eli Lilly, putting
the total deal value at $7 billion.
New York–based BMS immediately
packaged Amylin’s products together into
a new collaboration with AstraZeneca, in
which the latter pays the now wholly-owned
subsidiary Amylin $3.4 billion, and any
forthcoming profits and losses are split
equally between AstraZeneca and BMS.
The assets in the collaboration are Amylin’s
© 2012 Nature America, Inc. All rights reserved.
GLP-1 agonists, Byetta (exenatide) and
Bydureon (exenatide extended release);
metreleptin, being reviewed by the FDA for
diabetes and/or hypertriglyceridemia in rare
forms of inherited or acquired lipodystrophy;
and Symlin (pramlintide acetate), approved
for type 1 and 2 diabetes in patients with
inadequate glycemic control already taking
meal-time insulin. The three approved
products combined to sell about $830
million worldwide in 2011.
The side payments to Indianapolis-based
Eli Lilly in the buyout stem from Amylin’s
initial development and commercialization
partnership for Byetta with Eli Lilly, signed
in 2002, with a total potential value of $300
million. But in May 2011, Amylin sued Lilly
for engaging in anticompetitive acts and for
breaching that original 2002 agreement,
as Lilly earlier in the year had formed a
npg
collaboration with Boehringer Ingelheim in
Ingelheim, Germany. The two companies were
to jointly develop and commercialize two
oral diabetes agents, Boehringer Ingelheim’s
linagliptin and BI10773, and Lilly’s two
basal insulin analogs, LY2605541 and
LY2963016, with an option to co-develop
and co-commercialize Lilly’s anti-TGF-b
monoclonal antibody. Lilly planned to use the
same sales force to sell both exenatide and
the direct competitor Boehringer Ingelheim’s
linagliptin, and this is the detail that caused
Amylin to file suit. The issue was resolved
when Lilly and Amylin ended the alliance
and Amylin regained rights to exenatide.
Having full rights made Amylin an attractive
acquisition target, except that the breakup
called for Amylin to, among other things, pay
15% of global sales of exenatide until the
sum reached $1.2 billion, plus interest. Any
potential suitor needed to take on that long-
term financial commitment, and thus, BMS’s
multiparty buyout deal, with money flowing in
three directions. Brady Huggett
812 volume 30 number 9 SEPTEMBER 2012 nature biotechnology
Industry cautiously welcomes Supreme
Court decision on healthcare overhaul
This summer’s ruling by the US Supreme Court the same time, PPACA now at least provides
to uphold President Barack Obama’s healthcare a legal framework for the US Food and Drug
reform law has been welcomed by the drug Administration (FDA) to come up with a path-
industry. Industry executives say The Patient way for biosimilars manufacturers.
Protection and Affordable Care Act (PPACA), How much the drug market will expand in
which requires most Americans to obtain 2014 when PPACA comes into force is unclear.
health insurance and
includes a handful of
provisions that affect
drug development,
isn’t perfect and may
not ultimately ben-
efit biotechs’ bottom
line, but the ruling
at least ends a highly
Associated Press, Getty Images
fractionalized and
protracted battle.
“The biggest enemy
of the economy
and our industry is
uncertainty,” says Ron
Cohen, founder of
Although the general population remains generally split on the healthcare
Acorda Therapeutics reform, most in biotech industry view it as a positive.
in Ardsley, New York.
“It’s good that we have closure.” At the same More than 53 million Americans are unin-
time, companies are scrambling to ensure that sured, and the healthcare reform law aims to
they will be compliant with the law’s provisions cover more than half of them. To that end, the
on reporting physician payments. law expands Medicaid to include people at or
The PPACA legislation was enacted in 2010, below 133% of the federal poverty level. This
but was promptly challenged by 26 states and a bodes well for companies like GlycoMimetics
trade group for small businesses. In a 5–4 deci- in Gaithersburg, Maryland, which is develop-
sion on June 28 this year, the Supreme Court ing a small-molecule drug to treat vasoocclusive
deemed the law constitutional, upholding it crisis of sickle cell disease. Most people with
almost in its entirety. The court based its deci- the disease access treatment through Medicaid
sion on the US Congress’s power to impose and Medicare, says Rachel King, CEO of
taxes. The law requires individuals to obtain GlycoMimetics. An expansion of Medicaid
health insurance by 2014 or pay a penalty, and would be a way to give more people access to
that financial penalty “may reasonably be char- the company’s drug candidate, if it is approved.
acterized as a tax,” Chief Justice John Roberts But although the Medicaid expansion was
wrote in the majority opinion. “Because the initially mandatory for states, the Supreme
Constitution permits such a tax, it is not our Court in its decision deemed the provision
role to forbid it, or to pass upon its wisdom or overly coercive, and effectively made the pro-
fairness,” he wrote. gram voluntary for states. Kings says she is
A few key provisions stand out for innovative concerned that if states opt out, that will limit
biotech companies. Most notably, drug makers’ patients’ access. Indeed, several state gover-
markets should expand as more Americans nors have already said publicly they would
become insured and gain access to medicines. not participate in the expansion, even though
Small companies developing therapeutics the federal government would pay for 100%
also receive grants and tax credits for their of the additional costs through 2020. The US
projects, which may prove crucial for cash- Congressional Budget Office in July estimated
starved startups (Box 1). What’s more, inno- that in 2022, 3 million fewer people would be
vative drugs will receive 12 years of exclusivity insured due to state opt-outs.
against competition from biosimilars. The lan- Also offsetting the benefits of a bigger ros-
guage demarcating extra exclusivity for biolog- ter of insured Americans are the $80 billion
ics should provide innovative companies, and in rebates and fees drug makers must pay on
the investors who place their money in them, their commercialized products to help fund the
with more certainty, say industry executives. At plan. “In that sense, it costs us money. But on
 news

June 2012 October 2012 January 2013 January 2014 January 2015 January 2016

Patient-Centered The Independent New funding provided PPACA in force. US One year into Federal support to
Outcomes Research Payment Advisory to state Medicaid citizens able to buy implementation states ends, which
Institute issues first Board formed aimed at programs that choose coverage directly in biopharma industry may lead certain
grants extending the life of to cover preventive affordable insurance gets first indications of cash-strapped states
Medicare Trust Fund services for patients exchanges offering how much new to opt out of Medicare
begins operation. choice of health plans business healthcare
PhRMA has stated reform will provide
elimination of board as
a top priority

the other hand, bringing additional people into
the system is helpful,” says Cohen at Acorda,
which has two commercialized products. “It
may end up being a wash, and that’s fine.”
Biotechs are continuing to dig through the
more obscure sections of the law and finding
measures that affect their businesses. Sections
3001 and 3008, for example, includes incen-
tives for hospitals to reduce infections. That’s
good for S. San Francisco–based KaloBios
© 2012 Nature America, Inc. All rights reserved.
transthyretin (TTR)-mediated amyloidosis,
hemophilia and beta thalassemia. Maraganore
says that some biotech companies could expect
a market increase of 20% due to the elimination
of the caps, but it will vary depending on the
type of drug and the age of the patient popula-
tion.
The law eliminates other caps as well.
Insurance companies can’t exclude people for
having a preexisting condition, nor can they
sures are meant to help spot improper industry-
doctor relationships, but the paperwork could
prove challenging for small companies with few
resources, Cohen says.
The law’s 2.3% tax on revenue for device
manufacturers is proving difficult for the
smaller companies in that sector, particularly
those that have revenues but aren’t yet profit-
able. Congress included the tax in the law to
help pay for the expansion of health coverage.

Pharmaceuticals, which is developing a thera-
peutic monoclonal antibody to prevent pneu-
monia, an infection that often occurs in people
who are on ventilators in hospitals. “The gov-
ernment is pushing to reduce preventable infec-
tions,” says David Pritchard, CEO of KaloBios.
“And we’ve got a drug that prevents infection.”
Biotech executives working in orphan, or
rare, diseases say the law’s insurance market
reforms are key for them. The law bans insur-
ance companies from putting annual or lifetime
limits on individual policies—a boon to people
with continually high healthcare costs and,
by extension, the companies that make treat-
ments for them. “Patients with rare diseases
often hit these caps early in life,” says John
Maraganore, CEO of Alnylam Pharmaceuticals
in Cambridge, Massachusetts, which is devel-
npg
discontinue coverage after someone receives a
bad diagnosis. And children can now stay on
their parents’ policies until they are 26 years old.
This is a key reform for companies like Sarepta
Therapeutics in Cambridge, Massachusetts,
which is developing a therapy for people with
Duchenne muscular dystrophy, who suffer the
worst effects of the disease in their teens and
early twenties, and rarely live to see age 30.
It is not all good news, however. Likely head-
aches for industry in the law relate to what are
known as the ‘sunshine provisions’. The Act
requires companies to report all payments to
doctors and hospitals that are over $10. “It’s
a huge administrative burden,” says Cohen,
which hired new people as a result. “Every sin-
gle time you buy [a doctor] a lunch you have a
reporting obligation,” and separate reports have
But it is unclear yet whether the presumed mar-
ket expansion will offset that tax, says Samuel
Lynch, founder of BioMimetic Therapeutics in
Franklin, Tennessee, which makes recombinant
human platelet-derived growth factor products
that are regulated as devices. Lynch says the
tax will likely force him to pass along some of
the additional costs to hospitals in the form of
higher prices for his products. Lynch’s overall
impression of the Act: “I’m not a fan.”
In one of the more indirect effects of the law,
biotech companies should expect the momen-
tum to pick up on comparative effectiveness,
the practice of governments and insurance pro-
viders judging drugs based on relative health
or cost benefits. “The flip side of some of these
changes and the removal of the caps is there
will be a more active debate around health

oping treatment for rare diseases, such as to be filed with each state, he says. The mea-
Box 1 Where are they now?
As a measure of Obama’s PPACA, legislators in 2010 allocated $1 billion for grants
and tax credits to small companies developing therapeutic products. But the sum was
spread out across nearly 3,000 companies and 4,600 projects, leaving most with awards
of less than $1 million. For the larger of these awardees, the grant “might have been a
percentage on one month’s burn rate,” says Glen Giovannetti, global life sciences leader
at Ernst & Young.
But for Remedy Pharmaceuticals in New York, the grant money was a godsend. The
$733,437 the company received through the program “moved the whole company
forward,” says co-founder Sven Jacobson. “It came at a time when very little money
was flowing around and people were scared to invest.” After investors partly matched
the award, Jacobson was able to design a small pilot study of his product. According
to Jacobson, the results of this pilot helped him raise another $3.7 million from angel
investors and to plan a larger study.
The Therapeutic Discovery awards were a one-shot deal, but the Biotechnology Industry
Organization is pushing for Congress to extend and expand the program because it turned
out to be so popular. If there is a next time, the organization has proposed that the
government should grant larger awards to fewer companies.
outcomes and comparative effectiveness,” says
Maraganore. Companies developing early-stage
drugs should not only collect clinical data but
also data around the economic benefits of
their candidates, he says. The Act established a
government vehicle for conducting compara-
tive effectiveness research called the Patient-
Centered Outcomes Research Institute, which
issued its first grants in June.
Many provisions in the Act will surely be
tweaked over time (Table 1), and the possibil-
ity that parts of the law could be repealed by
legislators after the November elections still
looms. In the meantime, some small biotechs
will wait to see how the larger companies com-
ply with the law. “There’s an election coming
up and things may change again,” says Lynch at
Biomimetic. “As a small company you want to
wait to the extent that you can to see how these
things are being implemented by larger compa-
nies so that [you] don’t get too far out in front.”
EW
Emily Waltz, Nashville, Tennessee

nature biotechnology volume 30 number 9 SEPTEMBER 2012 813

 NEWS
in brief
Compulsory license
bandwagon gains momentum
India’s decision in March to grant its first-ever
compulsory license, which allows a company
to make and market a drug that the patent
holder has not been able to make sufficiently
affordable and accessible, has drawn cheers
from healthcare activists and opprobrium from
the pharma sector. The licensee, Hyderabad-
based Natco Pharma, will sell its generic version
of Bayer’s liver and kidney cancer drug Nexavar
(sorafenib) for 3% of the patented drug’s price
in return for paying 6% royalty on sales to
Leverkusen-based Bayer. Nata Menabde, India’s
representative to the World Health Organization,
told CNBC-TV18, “India has taken a good
political stand on compulsory licenses and
we respect that move. [A compulsory license
is] an important tool that governments have
in their arsenal and they should be using it
as appropriate, as per national legislation,
and keep public health interests above any
© 2012 Nature America, Inc. All rights reserved.
other interests.” Pharma companies and
Western patent offices such as the US Patent
and Trademark office have registered their
opposition, and Bayer appealed the Indian
government’s decision. Court arguments were
heard in August. However, until the decision is
reversed, the precedent means that other Indian
companies may seek similar arrangements
with the government. Compulsory licenses
have previously been used in Brazil, Thailand
and South Africa, and just last month China
amended its intellectual property laws to allow
for compulsory licenses. Michael Francisco
in their words
“We cannot on the one
npg
hand be marketing
our green image and
on the other hand
growing GM crops.
European consumers
will not be fooled by
this duality.” Gillian
Westbrook, manager
of the Irish Organic
Farmers and Growers
Association, after the Irish Agriculture and
Food Development Authority approved testing
of a blight-resistant transgenic potato. (The
Independent, 2 August 2012)
“I personally believe those [biologics] are the
future of drug development in our industry,
they just need time to continue to germinate.”
Following AstraZeneca’s $15.6 billion
acquisition and doubling of the workforce
at MedImmune, CEO Peter Greenleaf of the
Gaithersburg, Md., biotech emphasizes the
need for a long term perspective, given just one
approved product (a swine flu vaccine) and no
drug programs in late development. (Washington
Post, 5 August 2012)
814 volume 30 number 9 SEPTEMBER 2012 nature biotechnology
Biotechs opt for alternative floatation strategy
The Jumpstart Our Business Startups
(JOBS) Act, passed into law in the US
in April, has revitalized the practice
known as a Form 10 filing. This type
of public floatation is considered
more efficient and less risky than
a traditional initial public offering
(IPO), and though it existed before the
JOBS Act, there has been heightened
interest in the process since the
legislation passed. “The JOBS Act put
into the public consciousness that
it’s important to find new ways to do
IPOs,” says William Hicks, a partner
in the New York office of the law firm
Mintz Levin, and that has led to a
renewed focus on Form 10, adding, “It
could become the new normal for” for
emerging growth companies.
In a Form 10 fundraising, a
company prepares a prospectus
document to go public, but raises
The rising popularity of Form 10 filings is
letting companies consider raising funds and
money through a private placement
going public in a stepped process. before filing the prospectus. Because
the private placement is predicated
on also going public—and therefore
eventually having a publicly traded stock—the process broadens the investor base
for the private placement by allowing the participation of so-called ‘crossover’
investors who would normally invest only in public companies that have greater
liquidity than those in the biotech sector.
A key characteristic of the Form 10 process is that the company decides when
its stock begins trading, as opposed to a traditional IPO, where a company goes
through a filing and disclosure process, gains momentum and then prices its shares
at the end—when perhaps the climate for the deal has changed. “People are using
Form 10 as part of a conscious de-risk strategy,” says Hicks. “If the market’s not
good, with a traditional IPO you can have a horrible deal and not get what you
thought you were going to get. With this process, you raise the money first, then you
go public.” (Though, much like with a shelf offering for public companies, firms
are not required to sell shares on the open
market.)
Form 10 can also be used to create a
Using the Form 10
public shell company into which a biotech process for going
firm can merge—a strategy used successfully public “could become
by Los Angeles-based Cougar Biotechnology the new normal.”
(now a division of Johnson & Johnson) in
2006. This is cleaner than a reverse merger
with an existing company, which may carry the risk of litigation or other liabilities.
The strategy also may also give companies more leverage in acquisition
discussions. “Now, the threshold for doing a traditional IPO is so high,” it
isn’t a viable exit option for investors and potential buyers know it, says Hicks.
Ovascience, a fertility company in Cambridge, Massachusetts; KaloBios
Pharmaceuticals, an antibody developer in San Francisco; and Fresh Medical
Laboratories, a diagnostics company in Salt Lake City, Utah, are among the
companies with recent Form 10 filings. If those and others are successful, Form
10 could provide a consistent, alternative exit path for investors, and help small
biotechs put pressure on buyers in negotiations.
Mark Ratner, Cambridge, Massachusetts
 news

GSK buys partner Human Genome Sciences

London-based GlaxoSmithKline
(GSK) acquired for $3.6 billion
long-time collaborator Human
Genome Sciences (HGS),
the developer of Benlysta
(belimumab), a human
antibody that targets BLyS
(B lymphocyte stimulator)
approved in 2011 as the first
new drug for lupus in almost
50 years (Nat. Biotechnol. 29,
292, 2011). The partnership,
which began in 1993, has
generated two other compounds
now in late-stage trials at
GSK: darapladib, an inhibitor
of lipoprotein-associated
phospholipase A2 discovered by
GSK using HGS technology, for
© 2012 Nature America, Inc. All rights reserved.
in brief
India’s biosimilar regulations
New guidelines for “similar biologics”
launched August 15 will help address local
patient access to expensive drugs while serv-
ing to attract international manufacturers to
India. The Department of Biotechnology and
Central Drugs Standard Control Organisation
developed the regulatory framework with input
from industry and academia. Under the new
guidelines, manufacturers must prove similar-
ity to a reference biologic already approved
in India or licensed and sold for at least four
years in a regulated market. The biosimilar
should be comparable to the innovator drug in
safety, efficacy and quality as demonstrated
by analytical and clinical trials and the pre-
clinical and clinical studies should also be
comparative in nature. Approval “without
The buyout of HGS continues a run on big biotechs
involved clinical trials” is possible if manu-
by pharma over the last several years. Source: Human
facturers prove close similarity to its reference
Genome Sciences
product, and show consistency in production
process. Anurag Rathorem at Indian Institute

treating cardiovascular diseases including atherosclerosis, and albiglutide, a glucagon- of Technology in New Delhi, who contributed
to the guidance says, “We took both the US
like peptide-1 (GLP-1) receptor antagonist for type 2 diabetes, created by HGS using
and European guidelines into consideration
albumin-fusion technology and licensed to GSK in 2004. GSK initially launched a $13 while drafting.” The guidelines have received
per share tender offer for HGS in April 2012. HGS’ management advised shareholders a cautious welcome from industry. “The
to reject that bid, claiming among other things that it did not adequately reflect the requirement for comparative clinical trials
value to HGS of the three products, though it also cited synergies GSK would obtain in will certainly have an impact on the compa-
the buyout and the benefit of HGS’ $2.6 billion in net operating loss carryforwards and ny’s budget allocation,” says Geena Malhotra,
chief of research at Mumbai-based Cipla.
R&D tax credits. HGS opened the door to competing offers from other companies, but
Recent deals between Mumbai-based Emcure
GSK then upped its bid to $14.25 per share, or the final $3.6-billion price. “It was a Pharma and Basel-based Roche and between
foregone conclusion that HGS was going to be acquired by GSK,” given the terms of Merck Serono and Dr. Reddy’s Laboratories in
their partnership, says David Brindley of The European Centre for Accelerating Medical Hyderabad are a testament to India’s growing
Innovations at the University of Oxford. “From an investors’ point of view, I’d go as attractiveness as a biosimilars manufacturing
far as to say it was formulaic.” Had HGS been as valuable to anyone else, “there’s no hub. Killugudi Jayaraman
way that there would not have been a bidding war,” he says. With its global marketing
infrastructure, GSK may be in better position to maximize the value of Benlysta. With net Myriad’s patents redux
sales of $52.3 million in 2011 and $31.2 million in first-quarter 2012, HGS’ launch of The biotech industry breathed a collective
the drug had fallen below analyst estimates. sigh of relief with the news that on August
If not for GSK’s support, Benlysta might never have seen the light of day. The 16 the US Court of Appeals for the Federal
npg

Court in New York upheld the Myriad patents

companies began working together in 1993 when GSK (then SmithKline Beecham)
committed up to a whopping $125 million for rights to HGS’ mRNA-based discovery
process, among the first industry models for a broad-based genomics approach linking
protein discovery with disease. But over the next decade, despite the funding from GSK
and a series of large public stock offerings, HGS’ own product development efforts
foundered. Then, in 2005, GSK became HGS’ partner for Benlysta’s development,
remaining stalwart even after a failed Phase 2 trial a few months later threatened the
compound’s future and initiating the first of two dramatically successful Phase 3 studies
in 2007 (Nat. Biotechnol. 27, 779, 2009). Mark Ratner, Cambridge, Massachusetts
in their words
“I never thought this at the university for developing chimeric T cell
would happen, that cancer therapy. (Bloomberg Businessweek,
the pharma industry 7 August 2012)
would get into ultra- “Out of our research in France, we haven’t really
personalized therapy.” developed a new molecule in 20 years.”
The University of Chris Viehbacher, Sanofi’s CEO, explains why
Pennsylvania’s Carl he is firing workers in Toulouse and Montpellier
June after Novartis much to the chagrin of French politicians,
penned a $20 million labor leaders and researchers. (Bloomberg
deal to fund a center Businessweek, 9 August 2012)
on BRCA1 and BRCA2. The court came to the
same 2–1 decision in 2011, but was asked
to revisit the case by the Supreme Court,
following its ruling on a related case. (In
Mayo Collaborative Sciences v. Prometheus
Labs, the Supreme Court found a certain
diagnostic to represent a law of nature, and
hence not be eligible for patent.) As before,
the appeals court reversed two earlier district
court findings—that Myriad’s DNA test are
products of nature, and that a method for
screening potential cancer therapeutics
employs basic scientific principles. Judge
Kimberly Moore wrote that Congress has
authorized an “expansive scope of patentable
subject matter…and the USPTO [US Patent
and Trademark Office] has allowed patents on
isolated DNA sequences for decades,” saying
it is a matter of policy, not for the courts to
decide. Dan Vorhaus, editor of The Genomics
Law Report, says “Those are the types of extra-
legal policy-oriented questions that are at the
heart of the Myriad litigation and will not be
decided by litigation...[but] by policy makers.”
he says. Laura DeFrancesco

nature biotechnology volume 30 number 9 SEPTEMBER 2012 815

 NEWS

in brief
NIH injects $275 million
into undiagnosed diseases
and RNA research
In July, the US National Institutes of
Health (NIH)’s Common Fund announced
pledges of $145 million for research into
undiagnosed diseases and $130 million to
study extracellular RNA communication.
These areas represent “key roadblocks
in biomedical research and…emerging
scientific opportunities ripe for Common
Fund investment,” according to Elizabeth
Wilder, director of the NIH’s Office of Strategic
Coordination, which manages the program.
The Common Fund, established in 2006, is
a $545-million pot of discretionary money, to
support research that cuts across institutes.
The NIH’s Undiagnosed Diseases Program
(UDP) joined the National Human Genome
Research Institute, the NIH Office of Rare
Diseases Research and the NIH Clinical Center
© 2012 Nature America, Inc. All rights reserved.
gain understanding of the mechanisms underlying
these conditions, and improve their diagnosis and
treatment. An estimated 6% of the US population
suffers from such conditions. The Extracellular
RNA Communication program, meanwhile,
will explore how RNA is released from cells
and packaged for transport, and then interacts
with and influences specific cell types and
functions. This offers a chance to “transform our
understanding of endocrinology and intercellular
communication,” Wilder says, as well as provide
opportunities for diagnostics and therapeutics.
This initiative will launch in fiscal 2013 and run
for five years. Further details on the Undiagnosed
Diseases Program are expected in fiscal year
2014. Malorye Allison
China’s key R&D programs
behind schedule
Ninety three per cent of projects that comprise
yuan ($95 billion) in three stages by 2020 to
attain the world’s leadership in key fields. In
a report presented on June 27 to the National
People’s Congress, China’s legislature, Auditor
General Jiayi Liu described an audit of 8 of the
16 S&T schemes, although he did not reveal
which projects had been audited. According to
Liu’s report, 134 projects failed to meet their
milestones, yet continued to receive funding for
their next stages. Furthermore, 582 of the
2,401 sets of research results randomly
examined by CNAO had been replaced by results
from other projects. For the drug development
scheme, the central government invested nearly
6 billion yuan ($882.5 million) in support of
more than 900 drug development projects
and technology platforms by the end of 2011.
A Tianjin-based bio-entrepreneur, Zailin Yu,
president of SinoBiotech, whose project was
among those audited by CNAO, says “The
reason [that so many projects failed to reach
their milestones] is because the scheme’s goal

together in an effort to serve patients with
mystery diseases that have eluded diagnoses.
Since its inception in 2008, more than 200
cases have been tackled by the UDP. With
the new funding, UDP will go extramural (the
projects up until now have been intramural) to
establish a network of medical research centers
that will leverage advances in genomics to
China’s National Key Science and Technology
(S&T) Schemes had not finished their tasks by
2011, the end of the first stage, according to
the Chinese National Audit Office (CNAO) in
Beijing. Launched in 2009 and funding a range
of activities from innovative drug development
to airplane manufacturing, China’s 16 key
S&T schemes plan to spend some 600 billion
to develop the world’s leading innovative drugs
is unrealistic.” Rigid financial management and
late allotment of funding also contribute to the
failure to finish tasks, he adds. Liu reported that
institutions and researchers involved in the
audit have promised to correct their problems
and improve management of their research
funding. Hepeng Jia

Around the world in a month

KAZAKHSTAN
Scientists at the National
Biotechnology Center in Kazakhstan
develop carrots producing a recombinant
tuberculosis vaccine. Laboratory tests in animals
npg

are said to show eating the carrots improves

tuberculosis immunity.
CHINA
Researchers from the Beijing
University of Agriculture report
breeding two female cloned calves with
adipocyte fatty acid binding protein in their
meat. The binding protein is supposed to
enhance the taste of the meat.

MALAYSIA
Researchers in Malaysia and Japan describe
a method of making ethanol from the empty
fruit bunches left over from palm oil production. The
empty bunches are an abundant waste product of palm
BRAZIL oil production in Southeast Asia, with 15 million metric
Vienna-based See Algae Technology tons collected annually in Malaysia alone.
plans to construct what it says will be
the world's first seaweed-to-biofuel production
plant in the Brazilian state of Pernambuco. The
$9.8-million facility will be built next year and will
use carbon emissions from ethanol production
to grow the seaweed.

816 volume 30 number 9 SEPTEMBER 2012 nature biotechnology

 data page

Drug pipeline: Q212

Laura DeFrancesco

In the first two quarters of 2012, there have been 12 approvals agent for Alzheimer’s, Amyvid, were of note. Sodium glucose trans-
(NME and biologics) compared with 30 for 2011. The appovals of porter 2 inhibitors continue to impress in the clinic. But trials of
two obesity drugs, Qsymia and Belviq, a plant produced replace- Pfizer’s flagship Alzheimer’s therapy bapineuzumab were finally
ment enzyme for Gaucher’s disease, Elelyso, and a PET imaging suspended.

Notable regulatory approvals (Q2 2012)

Drug/company Indication Approval and drug information
FDA approvals by lead indication area
Amyvid (florbetapir Alzheimer’s 4/6/12, FDA. 18F-labeled small molecule allows detec-
70
F-18)/Eli Lilly imaging disease tion of amyloid beta in patients through positron
emission tomography
SOM230/Novartis Cushing’s 4/25/12, Europe. Cyclohexapeptide somatostatin analog 60
syndrome with unique ability to bind somatostatin receptor subtypes

Number of FDA approvals

1-5 50
Stendra (avanafil)/ Erectile 4/12/12, FDA. Phosphodiesterase 5 inhibitor with faster
Vivus dysfunction onset than marketed drugs 40
Elelyso (taliglucerase- Gaucher’s 5/1/12. Human glucocerebrosidase expressed in carrot
alpha)/Pfizer disease cell line 30
Perjeta (pertuzumab)/ Breast cancer 6/8/12, FDA. Humanized mAb that inhibits HER2
Roche dimerization by binding different epitope from Herceptin 20
Belviq (lorcaserin Obesity 6/27/12, FDA. Small-molecule selective agonist of only
hydrochloride)/Arena 5-hydroxy tryptamine 2C receptor subtype
10
Pharmaceuticals
© 2012 Nature America, Inc. All rights reserved.

Myrbetriq Overactive 6/28/12, FDA. Small-molecule beta 3 adrenergic agonist,

(mirabegron)/ bladder relaxing smooth muscle 0

11
95

96

97

98

99

00

01

02

03

04
05

06

07
08

09
10

*
12
Astellas Pharma

20
19

19

19

19

19

20

20

20

20

20
20

20

20
20

20
20

20
Qsymia (phentermine Obesity 7/12/12, FDA. Amphetamine appetite suppressor and Year
and topiramate)/Vivus anticonvulsant with unknown mechanism in obesity
Source: BioMedTracker, a service of Sagient Research (http://www.biomedtracker.com/). FDA, US
■ Infectious disease (108) ■ Cardiovascular (47) ■ Ophthalmology (20) ■ Obstetrics/gynecology (13)
Food and Drug Administration; EMA, European Medicines Agency.
■ Oncology (83) ■ Endocrine (39) ■ Psychiatry (20) ■ Dermatology (9)
■ Autoimmune/immunology (47) ■ Hematology (21) ■ Gastroenterology (17) ■ Rheumatology (4)
Notable regulatory setbacks (Q2 2012) ■ Respiratory (16) ■ Other (98)
■ Neurology (54) ■ Metabolic (24)
Drug/company Indication Setback summary
Setipiprant/ Allergic rhinitis
4/2/12. Phase 3 study of small-molecule antagonist of
Actelion chemoattractant receptor-homologous molecule expressed Source: FDA; BioMedTracker, a service of Sagient Research (http://www.biomedtracker.com/).
on TH2 cells (CRTH2) failed to show efficacy *2012 partial year from Jan. 1 to Aug. 11. Numbers in parentheses after legends are total
Vyndaqel (tafamidis Familial amyloid 6/18/12. FDA issued a complete response letter requesting approvals since 1995.
meglumine)/Pfizer polyneuropathy more efficacy data, despite EMA approval. Small molecule,
originally developed by FoldRx, stabilizes transthyretin,
inhibiting fibril formation
BMS-986094/ Hepatitis C 8/1/10. Phase 2b trial of nucleoside inhibitor of
Bristol-Myers RNA-dependent RNA polymerase (NS5b), originally
Squibb developed by Inhibitex, halted due to safety issues Notable trial results (Q2 2012)
Bapineuzumab/ Alzheimer's 8/6/12. Phase 3 trial of anti-beta amyloid humanized mAb Drug/company Indication Results
Pfizer and Johnson disease discontinued after failing to meet co-primary endpoints Canagliflozin/Johnson Diabetes Phase 3 results of this sodium glucose transporter 2
& Johnson & Johnson inhibitor improved glycemia in patients with moder-
Talactoferin/ Non-small cell 8/6/12. Phase 3 trial of dendritic cell activator (recombi- ate renal impairment (Am. Diabetes Assoc., Abstract
Agennix lung cancer nant human lactoferrin) suspended when overall survival of 41LB, 2012)
patients did not improve Dapagliflozin/Bristol- Diabetes Phase 3 results of this sodium glucose transporter 2
npg

Cytofab Sepsis 8/8/12. Company halted further development on this sheep Myers Squibb inhibitor demonstrated that 10% of TD2 patients with
(AZD9773)/ polyclonal anti-TNF alpha antibody after phase 2b failed to cardiovascular disease achieved three endpoints (Am.
AstraZeneca show improvement over placebo Diabetes Assoc., Abstract 1114-P, 2012)
Ganitumab (AMG Pancreatic 8/8/12. Company halted phase 3 trial when human mAb Amatuximab/Esai Mesothelioma Phase 2 results of chimeric mAb against mesothelin dem-
479)/Amgen cancer against insulin-like growth factor-1 was unlikely to meet onstrated 39% with partial response, 51% with stable
overall survival endpoint disease with mean progression free survival of 6.1 months
Source: BioMedTracker, a service of Sagient Research (http://www.biomedtracker.com/). FDA, US (ASCO, Abstract 7030, 2012)
Food and Drug Administration; EMA, European Medicines Agency. AMG386/Amgen Renal cell carci- Phase 2 results of fusion protein (peptibody) targeting
noma angiopoietins demonstrated overall response rate of
58-59% (ASCO, Abstract 4606, 2012)
Notable upcoming regulatory decisions (Q4 2012) GLPG0634/Galapagos Rheumatoid Phase 2 results of selective JAK1 inhbitor showed over-
Drug/company Indication Upcoming catalyst arthritis all improvement in ACR20 in 83% of patients receiving
Glybera (alipogene Lipoprotein 9/25/12, EMA. Glybera, a Ser447X variant of human drug versus 33% receiving placebo (EUNETHYDIS Int.
tiparvovec/UniQure lipase deficiencylipoprotein lipase gene delivered by AAV vector Conf., Abstract OP0263, 2012)
Linaclotide/Ironwood Irritable bowl 9/8/12, PDUFA. Linaclotide, a 14- mer peptide Ipragliflozin/Astellas Diabetes Phase 2 trial of this sodium glucose transporter2 inhibi-
syndrome agonist of guanylate cyclase-C that increases fluid tor shows that patients reduce A1C and body weight
secretion into the intestine (Am. Diabetes Assoc., Abstract 1046-P, 2012)
Metreleptin/Amylin Lipodystrophy 10/3/12, PDUFA. Metreleptin is recombinant human Miravirsen/Santaris Hepatitis C Phase 2 trial of oligonucleotide targeting MIR-122
leptin, which affects blood glucose and triglycerides demonstrated prolonged dose dependent reduction in
Jetrea/Thrombogenics Thyroid cancer 10/17/12, PDUFA. Jetrea is recombinant microplas- HCV RNA (Eur. Assoc. Study Liver, Abstract 58, 2012)
min, a truncated stable form of plasmin PF-332991/Pfizer Sarcoma Phase 2 results of cyclin dependent kinase 4/6 inhibi-
Enzalutamide/Astellas Prostate cancer 11/21/12, PDUFA. Enzalutamide, an androgen recep- tor demonstrated 70% progression free survival at
tor antagonist of higher affinity than bicalutamide 12 weeks (ASCO, Abstract 10002, 2012).
Cabozantinib/Exelixis Thyroid cancer 11/29/12, PDUFA. Cabozantinib inhibits hepatocyte Yttrium-90 clivatu- Pancreatic Phase 2 trial of mAb targeting mucin antigen labeled
growth factor receptor MET, RET and vascular endo- zumab tetraxetan/ cancer with yttrium-90 demonstrated disease control in 58%
thelial growth factor receptor 2 Immunomedics of patients in 7.7 months (Soc. Nucl. Med., Abstract
Raxibacumab (ABthrax)/ Anthrax 12/15/12, PDUFA. Abthrax, a human monoclonal anti- 495, 2012)
Human Genome Sciences body that blocks anthrax from binding to cell surface Source: BioMedTracker, a service of Sagient Research (http://www.biomedtracker.com/). FDA, US
BG-12 (dimethyl Multiple sclerosis 12/28/12, PDUFA. BG-12, a second-generation Food and Drug Administration; EMA, European Medicines Agency.
fumarate)/Biogen Idec fumarate that is immunomodulatory
Kynamro (mipomersen Dyslipidemia/ 12/31/12, EU. Mipomersen, 2ʹ-O-(2-methoxy)
sodium)/Sanofi hypercholester- ethyl-modified ribose antisense oligo against
olemia apoB-100
Source: BioMedTracker, a service of Sagient Research (http://www.biomedtracker.com/). FDA, US
Food and Drug Administration; EMA, European Medicines Agency. Laura DeFrancesco is Senior Editor, Nature Biotechnology.

nature biotechnology volume 30 number 9 SEPTEMBER 2012 817

 N E W S fe at u re
Patient power
Patient foundations are not only exploring new funding models but
also catalyzing translational research, with notable successes.
Jim Kling reports.
A few weeks ago, the US Food and Drug
Administration (FDA) approved Kyprolis
(carfilzomib), an irreversible proteasome inhib-
itor developed by Onyx Pharmaceuticals of
S. San Francisco, California. The success of
this drug program owes a great deal to the
Multiple Myeloma Research Foundation
(MMRF) of Norwalk, Connecticut, which
supported carfilzomib from its earliest stages.
It is indicative of the growing credibility of
an increasing number of patient foundations
that are galvanizing translational research,
© 2012 Nature America, Inc. All rights reserved.
particularly in areas poorly supported by the
private sector. Unwilling to wait for academic
groups to move basic research into the clinic,
or for industry to swoop in, these foundations
are bankrolling preclinical development and
beyond. Not only are they compensating for
gaps in existing traditional funding sources,
but they are also exploring new models to
coordinate research efforts, improve clinical
trial recruitment, capture intellectual property
(IP), overhaul biobanking and counter data
hoarding by academics and companies.
Found in translation
Like the MMRF, the Myelin Repair Foundation
(MRF) of Saratoga, California, is taking a more
hands-on approach than a traditional patient-
centered foundation would. The foundation
npg
brought together the leading research groups
in the field to form a network—the Accelerated
Research Collaboration—with a goal of getting
a drug into the clinic by 2014. Just last June, it
announced the initiation of a phase 1 clinical
trial—an autologous stem cell therapy devel-
oped by Robert Miller at Case Western Reserve
University in Cleveland, Ohio—putting them
several years ahead of schedule.
In addition to the $45 million that the MRF
has raised, some 80% of which has been handed
out to the eight academic centers compris-
ing their network, the foundation unveiled in
January its own Translational Medicine Center,
where myelin repair assays can be performed
on promising drug candidates coming from
anywhere, including industry. For example, in
February, the foundation signed a deal with
a company, ENDECE Neural of Mequon,
Wisconsin, to test its compound libraries for the
ability to reverse myelin damage. “They were a
cancer company with no thought of entering the
818 volume 30 number 9 September 2012 nature biotechnology
‘Why can’t we do it? Maybe it’s time we became
more aggressive’,” says Beall.
Breaking with tradition
Traditionally, foundations have focused on
patient education and support, allocating a rela-
tively small portion of their budgets to research.
“We don’t behave quite like the legacy organi-
zations,” says MRF’s Johnson. When legacy
neurological space, but they realized that the foundations handed out research money, they
pathway overlapped with work we were doing. followed the US National Institutes of Health
So they came to us to test compounds in our (NIH) model, accepting applications for basic
assays,” says Scott Johnson, president, CEO and research funds but then backing off from financ-
founder of the MRF. ing further development work. But now, some
The MRF doesn’t stop at R&D. It also ensures foundations see their role as not just shepherd-
that discoveries have ing drug candidates
patent protection to along the tangled path
encourage further to drug development
investment. Johnson but actually changing
says the foundation the way translational
has encountered little research and clini-
resistance from mem- cal development are
ber universities. Most done. “What’s dif-
have limited budgets, ferent about us and
forcing university pat- a few other, newer
ent offices to prioritize organizations is that
their efforts in areas, The new philanthropist. Kathy Giusti, founder of we look at the entire
such as hardware and the MMRF, participates in a plenary session at a continuum all the
Partnering for Cures meeting in November, 2011.
software, that have way from discovery
Source: MMRF.
faster payoffs than biology that might
drug research. The be done in academic
foundation pays for patent submissions and labs, to thinking about how to cross the transla-
splits royalties with the university. “It’s a no- tion gap, to working with pharmaceutical com-
brainer. It’s almost like new money that they panies to move drugs forward,” says Johnson.
wouldn’t have gotten at all,” says Johnson. Formerly a serial entrepreneur and business
The inspiration for these activities derives consultant, Johnson founded the MRF in 2002
in part from the success of the Cystic Fibrosis with a patient’s perspective; he has been living
(CF) Foundation of Bethesda, Maryland, which with MS for over 30 years. When he became
invested about $75 million in Kalydeco (iva- more active in MS research, he was surprised
caftor), the Vertex Pharmaceuticals drug that to find that no one had a research plan. “This
was approved by the FDA in January. In the was one of the most stunning things to me.
1990s, the foundation recognized that indus- Coming from business, you have a business
try wouldn’t get serious about the field without plan. I assumed if you were doing research that
more incentives, so it took several steps: it cre- you had a research plan to attack this disease,”
ated a network of patients at clinical sites in the he says. Under his guidance, the MRF has devel-
Seattle area to assist with the organization of oped its own plan. “We sat down our experts and
clinical trials; and it established a screening proj- defined what is known, and prioritized ques-
ect for oral drug candidates that could open up tions to answer, and that became the basis for
the abnormal ion channels that plague patients the initial experiments that we funded.”
with CF. In addition to the $75 million in sup- In fact, these agencies are being run like for-
port of Kalydeco, the foundation spent an addi- profit agencies, with deals structured more like
tional $70 million in partnership with Vertex those of a business than a charity. As the CF
to develop a second-generation drug. “Vertex Foundations’ Beall describes it: “We don’t just
would tell you that they wouldn’t be in the cystic give them the money: we have milestones and
fibrosis area if it were not for the foundation de- they have incentives to perform. If they don’t
risking their early efforts,” says Robert J. Beall, perform we can walk away, and we have reach-
president and CEO of the CF Foundation. in rights we can use [if we walk away]. We get
“A lot of these [patient foundation] boards royalties back, and use them to fund other ven-
are run by parents. They see that the CF ture philanthropy.”
Foundation [has helped develop] a drug that Another nonprofit, the CHDI Foundation,
changes the course of a disease, and they ask, which works with Huntington’s disease
 news fe at u re

Table 1 How foundations are catalyzing R&D progress

Foundation Initiative Description
MMRF CoMMpass Provides comprehensive tracking of genomic changes in
1,000 patients with multiple myeloma from bone marrow
and peripheral blood collected at baseline, at suspected
complete response, and at relapse or progression
CHDI Foundation PLoS Currents Ensures data from CHDI projects are rapidly lodged in
Huntington’s Disease open repositories under a Creative Commons license.
(http://currents.plos.org/)
Parent Project PLoS Currents Promotes rapid communication of new research results
Muscular Dystrophy Muscular Dystropy and operational analyses derived from the study or man-
agement of all types of muscular dystrophy (http://cur-
rents.plos.org/md/)
The Addi & Cassi Cyclodexin IND Promotes clinical testing of an approved drug re-purposed
Fund (Reno, NV) for a rare condition neglected by commercial sponsors. In
2009, FDA approval given for compassionate-use investi-
gational new drug (IND) request to use hydroxypropyl beta-
cyclodextrin in Niemann-Pick Type C disease
researchers, has also established its own trans- subgroups and then say ‘Hey, we’ve got a clinical
lational unit to perform screening, absorp- trial for you’. There will be hypothesis-driven tri-
tion, distribution, metabolism and excretion als that we can conduct within our clinical net-
(ADME) characterization, pharmacokinetics work,” says Quinn Young.
© 2012 Nature America, Inc. All rights reserved.
clinical progress and to assess therapeutic
efficacy. The Michael J. Fox Foundation is
addressing this with a five-year, $45 mil-
lion longitudinal biomarker study called the
Parkinson’s Progression Markers Initiative
(PPMI). Launched in 2010, the study includes
400 untreated patients with Parkinson’s dis-
ease and 200 controls, each of whom will
undergo a battery of tests and assessments. All
data will be freely available in ‘real time’ on a
portal (http://www.PPMI-info.org); research-
ers need not be funded by the Michael J. Fox
funding to access the samples. As novel bio-
markers are discovered, they may be incorpo-
rated into the study.
The project was modeled in part after the
Alzheimer’s Disease NeuroImaging (ADNI)
study, according to Todd Sherer, CEO of The
Michael J. Fox Foundation. Begun in 2003, with
backing from the NIH’s National Institute on
Aging and the National Institute of Bioimaging

and other assays. The foundation approaches
companies working on mechanisms or targets
that could affect the neurological processes
related to Huntington’s disease. It then pays
for compounds to be tested. “We tell them the
results and they own the IP. It’s hard for people
to say no to that if they’re confident that you’ll
respect confidentiality and IP and that you’ll
do high-quality work,” says Robi Blumenstein,
president at CHDI.
Tackling clinical trials
Clinical trials present many challenges for drug
developers, from recruiting patients to validat-
ing clinical endpoints and finding suitable bio-
markers. Foundations are taking steps to address
these problems.
In June, the MMRF began its most ambi-
npg
Cambridge, Massachusetts–based Millennium
Pharmaceuticals and Onyx Pharmaceuticals
have made a multiyear, multimillion-dollar
commitment, according to Quinn Young, even
though the results will not be patentable. “The
benefit to the companies is that they will have
exclusive access to the data for six months. At
the end of the sixth month, it goes into the pub-
lic domain in a portal that we’re building,” says
Quinn Young. Participating academic centers
also have data access during the sixth month.
In addition, companies will have information
on their own drug, in real time, as well as their
competitors’ drugs. “I think that the data are of
interest on a number of levels, from translational
medicine to clinical development, all the way to
marketing,” Quinn Young opines.
The program is expected to cost more than
and Bioengineering, ANDI was able to engage
the pharmaceutical industry as partners in
the effort to screen patients with Alzheimer’s
disease using various imaging modalities and
to conduct genetic and cognitive assessments.
“They slowly got companies to come on board
and get comfortable with this precompetitive
arrangement. That was one of the things in
our favor—companies had had exposure to the
Alzheimer’s initiative. The other was our posi-
tion as a neutral body. There was a high trust
factor that we had no motive in our coordinat-
ing role,” says Sherer. Companies contributing
to PPMI include multinationals, such as Pfizer
of New York, GlaxoSmithKline of London,
Merck of Whitehouse Station, New Jersey,
Genentech/Roche of Basel, GE Healthcare of
Waukesha, Wisconsin, Covance of Princeton,

tious clinical program, called CoMMpass,
which will track 1,000 patients newly diag-
nosed with multiple myeloma, recording all
clinical data, including biopsies to be per-
formed at every relapse over five years. “We
want to understand the profile of patients that
are truly at high risk [of relapse]. About 20%
of patients who initially respond to therapy
relapse very quickly and have a much poorer
prognosis. We want to understand what that
population looks like,” says Anne Quinn
Young, vice president of strategic alliances at
the Multiple Myeloma Research Foundation.
Researchers can also track responses in those
patients to various treatments.
The program includes 41 academic and com-
munity medical centers, and testing will be per-
formed by the Translational Genomics Research
Institute (Phoenix, Arizona). Participants will
also be screened for various known mutations,
and newly identified mutations will be added as
they are discovered. “We may be able to identify
$40 million, largely because of the costs of
assays and sequencing. “No one else would
have the incentive to do something like this, and
no single center would see enough patients to
get this done. From a funding perspective, it’s
tough to get these types of trials funded by the
government, and a single company wouldn’t be
motivated enough to do it,” says Quinn Young.
The availability of a patient database is accel-
erating the pace of clinical development. Nearly
40 trials have opened to date, and Quinn Young
says they have started 60% faster than other
oncology trials. “That’s based on a single-site
experience at Vanderbilt, but it has involved
dozens of trials. Before we refined the process
there, the rate at which we opened trials was very
similar to [that of others in the field],” Quinn
Young adds.
Other foundations are chipping away at
different problems with clinical trials. One
issue for Parkinson’s disease drug develop-
ment is the dearth of biomarkers to track
New Jersey, and Abbott of Deerfield, Illinois, as
well as biotechs like Biogen-Idec of Cambridge,
Massachusetts, and Avid Radiopharmaceuticals
of Philadelphia.
Along similar lines, CHDI is conducting an
observational study to characterize patients in
anticipation of clinical trials. It will collect clini-
cal and personal information from patients, as
well as polymorphisms and other genetic data.
Twelve thousand subjects are currently enrolled.
The data will be available for future clinical tri-
als. “We’ll have [patients] preregistered and
characterized so that we can enroll quickly for
trials,” says Blumenstein. The foundation has
also taken steps to expedite informed consent.
“It’s less scientific content than organization
content,” he adds.
Working with regulators
Moving beyond clinical trials, foundations are
also engaging with the FDA to work through
regulatory hurdles, such as establishing

nature biotechnology volume 30 number 9 september 2012 819

 N E W S fe at u re
appropriate outcomes for clinical trials. The
Michael J. Fox Foundation, for example, will
begin an effort to validate clinical outcome
measures for Parkinson’s disease cognition,
which in turn could be used for late-stage
clinical trials. Such efforts have the potential
to streamline the regulatory process.
Companies can then use the protocols in their
clinical trials, rather than developing their own.
“When they go to the FDA, they can refer to a
foundation-funded study for why they selected
a particular outcome measure,” says Sherer. The
agency is more likely to accept the outcome
measure at face value if a neutral foundation
developed it, he adds.
Other foundation efforts are broader in scope.
In the field of artificial pancreas systems, the
Juvenile Diabetes Research Foundation (JDRF),
headquartered in New York, recognized that
there were no FDA-established guidelines for
approval of such systems. Last June, the foun-
© 2012 Nature America, Inc. All rights reserved.
dation consulted with companies and academ-
ics and produced a draft guidance document,
which they then presented to the FDA to kick-
start its internal process. The JDRF secured a
commitment from the FDA, which issued its
own draft guidance in December 2011. “We
probably accelerated the process by several
years,” believes Richard Insel, chief scientific
officer at the JDRF.
Foundations are in a unique position to gain a
regulatory agency’s confidence, agrees Timothy
Coetzee, chief research officer at the National
Multiple Sclerosis Society. “We don’t have a
commercial stake… the FDA has engaged with
patient advocate organizations, saying, ‘We want
to partner with you to bring people together’.
They need to hear what the scientific com-
munity is thinking, what the stakeholders are
npg
thinking. Foundations have the ability to touch
all those worlds in a manner that really isn’t
biased,” says Coetzee.
“[Foundations] are really well connected and
in tune with the needs of patients and the needs
of R&D organizations,” says Mark Shearman,
vice president of global research and early devel-
opment at EMD-Serono, a contributor to the
National MS Society’s Fast Forward initiative.
This initiative, started in 2007, provides fund-
ing for early-stage biotechs and academics with
technology ready to license.
Promoting sharing
Fierce competition among academic and indus-
try researchers lead groups to keep a tight hold
on their data. Furthermore, negative studies
often don’t see the light of day. “It’s understand-
able. It’s hard (for an) investigator to go through
820 volume 30 number 9 September 2012 nature biotechnology
the agony of writing up negative results,” says
CHDI’s Blumenstein. Such papers would be
hard to place in most journals, anyway.
Foundations are in a position to change
that. The scientific advisory board of MRF’s
Accelerated Research Collaboration meets with
principal investigators two to three times a year.
“There’s a huge amount of intellectual firepower
to discuss [projects]. The more you open up, the
more benefit it is to [the principal investigator]
as well as the team. And it prevents them from
doing research that’s already been done,” says
Johnson.
Openness hasn’t always been an easy sell.
Johnson says that it took a couple of years to
convince researchers to share data immediately,
“but now some of our PIs have joint lab meet-
ings,” he says.
CHDI has a similar attitude. “We’re in a good
position to see [negative results] because we
fund so much work in Huntington’s disease.
We know if we already funded someone to do
an experiment and it didn’t work. We can put
those investigators in touch with each other,”
says Blumenstein. CHDI encourages patenting
of therapeutic candidates, but animal models,
assays and other tools developed must be made
publicly available to other researchers. “We’ve
tried to institutionalize the concept of sharing,”
says Blumenstein.
The Accelerated Cure Project for multiple
sclerosis (ACPMS; Waltham, Massachusetts),
a foundation started in 2001 by tech entre-
preneur Art Mellor, provides researchers
with research tools and highly characterized
biological samples from about 3,000 subjects
from ten multiple sclerosis clinical centers in
the United States. Upon request, aliquots of the
samples are sent out to researcher; in return, all
data must be submitted to the repository data-
base for general use. Over 60 research studies
have used the resources to date. “We certainly
couldn’t have afforded to fund those studies
directly, but we greatly reduced the cost and
time to do them,” says Robert McBurney, CEO
of ACPMS.
The open-source nature of the project may
have turned some companies off, but there is
evidence that it hasn’t been much of a barrier.
Of 60 studies, about 20 were done by companies,
according to McBurney. “It wasn’t as big a prob-
lem as we might have anticipated.”
Researchers have an exclusivity period to
secure IP before turning over the data. The
open source–style system puts a premium on
data quality, and as a result, research applica-
tions are judged not on the hypothesis but the
quality of the study design. “We put a lot of effort
into making sure that we collect really good data
associated with the samples,” says McBurney. He
envisions the program as a potential model for
other diseases. The approach “doesn’t have to be
limited to MS,” he says.
Brain trusts
Foundations are also becoming a repository
of expertise. Many have made substantial
investments in internal scientific acumen, in
the form of advisory boards and employees.
“Companies come to foundations to get insight
into what models to use, or to get a gut check
on the potential of a therapy. We see hundreds
of applications every year, and we interact with
hundreds of thought leaders external to the
foundation, so we have a very good bird’s-eye
view of what’s going on in Parkinson’s disease
research. We see that as a resource beyond
funding that research foundations are bringing
to the table,” says Sherer.
Many of those partnerships involve bio-
tech companies looking to de-risk early-stage
programs, hoping for funds to fill gaps in an
early-stage development program. But the
companies are often looking for more than
just money, according to Insel. “We have a lot
of insight and knowledge of type 1 diabetes.
We fund over $100 million a year in research,
and about one-third of our funding goes out-
side the United States. We have a pretty big
overview of the type 1 diabetes landscape, as
well as insights into the unmet medical needs
of patients.”
Pharma companies are on board with this.
Through its relationship with the National
Multiple Sclerosis Society’s Fast Forward initia-
tive, EMD-Serono of Rockland, Massachusetts,
has funded a variety of early-stage companies
and academic groups that are working on a
range of biological mechanisms and targets.
“I don’t know if it would be possible to get
this sort of diversity in another setting,” says
Shearman.
The research is also well targeted, accord-
ing to Shearman. “[The foundation] really
has the patient in mind, so there’s no risk that
we’re going to be side-tracked by doing aca-
demic research for research’s sake,” says Steve
Arkinstall, senior vice president and head of
US research and global external innovation at
EMD-Serono.
In fact, a big-picture view is critical to the
success of foundations, Johnson says: “I think
the real secret sauce here is driving compounds
through the entire process, not haphazardly
funding in unrelated areas.”
Jim Kling, Bellingham, Washington
 building a business

How much risk are you prepared to take?

Mark Van Dyke
Any faculty member wishing to commercialize a discovery or technology should contemplate the pros and cons of
pursuing the entrepreneurial route or licensing to an outside entity.

L ike life and business, academic research

is all about risk. Risk is involved when
deciding which students to take into a facul-
ty’s research group, whom to collaborate with,
© 2012 Nature America, Inc. All rights reserved.
If you have a good idea of where your tech-
nology sits relative to the rest of the world and
know you want to move your idea into a for-
pursued if the university principals have a very
good understanding of the value of the tech-
nology, the potential market and a substantial
profit entity, you might find that universities amount of data about the asset(s).

what grants to submit and when and where to
publish. It also comes in play with the direc-
tion of research projects and potential rewards.
There is similar risk involved with taking
one’s research out of the lab and into a com-
mercial entity—a complicated process. When
Winston-Salem, North Carolina–based
KeraNetics, where I am CSO, was formed in
2008, we spent nearly a year meeting and talk-
ing with the university’s Office of Technology
Asset Management, the Legal Office, the
Office of Research, the Dean of the School of
Medicine and the Conflict of Interest (COI)
Management Director. Each group had dif-
ferent goals and concerns, and each indepen-
dently gauged the risk involved in launching
KeraNetics. Here I will use that experience
to provide an idea of how you should evalu-
npg
often default to the standby commercializa-
tion strategy of contacting companies they
know (or perhaps you know) and shopping the
technology as a standard patent license. You
should be aware that this strategy rarely works,
as most companies will view your technology
as risky academic research, unproved by the
standards they use for internal evaluation of
new technologies and for in-licensing deci-
sions (though it helps if you are employed at a
university with a big name in biotech, such as
Harvard University, Massachusetts Institute of
Technology or Stanford University). After all,
for more than a decade many established com-
panies have followed the strategy of acquiring
late-stage technologies that have already been
substantially ‘de-risked’ (usually in another
smaller commercial venture) through the clinic,
The problem for you, of course, is that most
universities are not experienced in getting
research to this stage and typically do not have
the expertise, infrastructure or funding to do
it. Moreover, if capabilities do exist within your
university for limited clinical testing, there
may be only enough capacity to advance one
technology at a time, so you might be forced to
compete for scarce internal resources in an envi-
ronment where commercialization decision-
making experience is typically lacking.
Often the best option is for you to become
the commercialization champion, in which
case starting a company moves to the top of
your list of options. But be warned: launching
a startup will require a high level of commit-
ment, so consider all the personal and profes-
sional cons before making such an important

ate the risk involved in pursuing the different the US Food and Drug Administration (FDA) decision. At the onset, you should ask yourself
paths to commercializing your technology or or even the market. Notwithstanding the uphill what your ultimate goal is: do you want to stay
discovery. challenge, this strategy can only be successfully in academia and have a commercial outlet for

Considering a startup
Technology transfer offices (TTOs) often ask a Box 1 The pointed questions
few basic questions (Box 1) when approached
by faculty with an idea, experimental results or When you approach your technology transfer office with a new technology or asset to
an invention disclosure. If you cannot answer commercialize, the officers will likely have a list of questions to ask you. Be prepared. Ask
these questions, then you are not ready to yourself:
What is the invention?
consider a commercialization plan. But even
What is it used for?
before you go to the TTO, the first questions
Why is it useful?
need to be ‘why do I want to launch a startup?’
What is the current state for development?
and ‘is my technology ready?’.
Are there other applications or uses for the invention?
What competitive technologies exist?
Mark Van Dyke (formerly of Wake Forest
What companies or other users of the invention may be interested in further developing,
University School of Medicine) is currently an
licensing or buying the invention?
Associate Professor at Virginia Polytechnic
What have you published or presented publically that relates to this invention?
Institute and State University in Blacksburg,
What unpublished data do you have?
Virginia, USA, and CSO at KeraNetics LLC,
What research plans do you have for further development, and do you have the funding
Winston Salem, North Carolina, USA.
to pursue these plans?
e-mail: mvandyk5@vt.edu

nature biotechnology volume 30 number 9 september 2012 821

 building a business
your research, or do you want to create a new
career for yourself if the company is successful?
Depending on your answers, your thought
processes, strategies and actions may change.
Above all, be honest with yourself about your
intentions and with those you involve in the
process.
Here is one big con: done correctly, com-
mercializing your research will easily double
your workload. If you are the inventor, you will
need to help develop the business plan, identify
market opportunities, assess competitive tech-
nologies, educate potential investors, evaluate
initial hires, translate many lab processes into
an industrial setting, begin the early scientific
activities of the company and, in some cases,
act as the company’s first CEO or (like me) the
CSO. In fact, creating a startup often comes with
80-hour work weeks, which drag on for three
or more years, all while you are often trying to
teach classes, run a research group, keep your lab
© 2012 Nature America, Inc. All rights reserved.
funded and remain active in the scientific com-
munity. And most of this often happens during
non-regular business hours, and is accompanied
by an atmosphere of constant urgency.
You should also expect that the university will
have concerns about your COI and conflict of
commitment. Many universities have strict pol-
icies that help keep the focus of faculty on more
scholarly activities. Universities tend to adopt
one of two approaches: either avoidance of con-
flict (or even the appearance of conflict) or strict
management of conflict. If you are at an institu-
tion that practices the former and you want to
be actively engaged in entrepreneurial endeav-
ors, you will need to find a new job; conflict of
interest is inevitable. If your institution practices
the latter approach, you can simply be mindful
of the university policies, address stakeholder
npg
concerns, be transparent in business dealings
that involve university interests and protect the
interests of the trainees entrusted to your care.
In the case of my involvement in KeraNetics,
a progressive COI management plan had been
developed that included a person tasked with
oversight who was essentially embedded into
my research program and reported through
the university’s Office of Research. His role was
to attend our group meetings; review our data,
especially anything presented in public forums;
ensure that disclosure of potential COI was
given in public forums; meet with trainees to
review their research progress and to ensure no
barriers to scholarship were present; and inde-
pendently report to the university and attest to
our following of the management plan and rules
of the institution.
Building it
For most academics it is hard to separate passion
for their research from the practical realities
822 volume 30 number 9 september 2012 nature biotechnology
of the marketplace. For example, when start-
ing a company, you must consider the end: is
the goal for you and your investors to exit the
startup by being acquired or through an initial
public offering (IPO)? If so, you must consider
what will motivate a favorable decision among
the merger and acquisition (M&A) team or
potential investors and build the company in
that direction.
It is mostly true that universities are bureau-
cratic entities with many political agendas in
play at any one time. It is a difficult place to
overlay business principles and derive the
degree of predictability that a startup needs
to be successful. The unpredictability of uni-
versity politics, their inexperience in dealing
with startup businesses and lack of urgency in
decision-making all increase the risk around
startups.
I found that when research from my lab was
being considered for commercialization, it was
helpful to connect commercialization criteria
to those things most familiar to typical aca-
demic faculty—the five ‘P’s: people, patents,
papers, preliminary data and ‘phunding’.
People. The human factor is one of the most
critical assets in this process. Although the
value of management expertise cannot be
overstated, technical prowess is equally impor-
tant. This is where full commitment from the
faculty champion and/or inventor—whether
that is you or otherwise—is crucial. However,
your initial hires in all areas are important, and
sometimes a pool of talent is available from the
faculty’s own lab. Many academic labs have a
wealth of smart, eager, talented graduate stu-
dents and postdocs who can rise to the chal-
lenge of a startup venture. Their knowledge
can be extensive because they often have con-
tributed greatly to the development and test-
ing of the technology, and their enthusiasm
for building from the ground up is typically
inexhaustible and infectious. Although they
are not generally CEO or CSO material, expe-
rienced postdocs can be critical early hires to
help transition the technology into the com-
pany and begin building in-house expertise.
However, it is worth noting that good scien-
tists do not always make the best initial hires
for a startup company. Skills at the bench
are not the sole criteria for startup success,
so graduate students and postdocs with a
predisposition for business, especially those
who have taken additional business-related
coursework, or preferably the occasional MD
or PhD-MBA student, have high potential
in a startup. If your university has a business
school, you will often find these types of grad-
uate students either enrolled in dual programs
or taking a few business classes out of pure
interest. If there is an entrepreneurial nature to
your work, it is a good idea to keep an eye out
for these types of students and recruit them
to your lab.
The choice of a CEO is one of the most
important initial decisions a startup com-
pany will make. Many investors will attest to
the value of an experienced CEO when deal-
ing with the uncertainties of a startup. In
basic terms, you need to consider whether a
high-powered technical or scientific expert
is the best choice or someone with the abil-
ity to manage the business and raise money.
Obviously someone with both abilities would
be ideal, but these individuals are rare and not
easy for most startups to find. If the company’s
business plan calls for additional research to
be funded largely through competitive grants
and resources provided through the university,
a scientific CEO may make the most sense.
However, if the startup plans to raise most of
its funding through equity investment, a CEO
with business expertise and connections to
capital is more desirable. At KeraNetics, an ini-
tial emphasis was placed on finding someone
with an ability to raise money but ultimately a
senior scientist was quickly hired who could
guide the company scientifically as well as lead
early grant efforts.
You may be asking yourself, ‘where do I find
such people and how can I, an academic scien-
tist, find an entrepreneurial network of such
individuals?’. A good place to start is at your
own university. Your TTO is the obvious place
as they should be familiar with the regional
business community. If they have extended
their professional network through organi-
zations such as the Association of University
Technology Managers, they may be able to
put you in contact with more experienced
individuals and perhaps even possible CEO
candidates. Other potential startup manage-
ment may be found through your university’s
School of Business. Some of the faculty may
be active or formerly active entrepreneurs, or
will likely have an extensive network of con-
tacts. Schools of Business are also a great place
to get assistance in business plan development
from either faculty or students. During the
early stages of KeraNetics, students from the
Wake Forest University Schools of Business
did several class-related projects to evaluate
the company’s potential products, markets and
business development strategies.
A third option is to speak with your scien-
tific colleagues that have traveled the entre-
preneurial road. Many scientific societies have
groups within them that focus on industrial
science or the commercialization of research.
National meetings are great venues to meet and
speak with members of these groups.

Patents. Patents are the most important pri-
mary asset of the company, and you will need
proper protection of the technology as well as
freedom to operate in an opportune area of
the market. In most instances, the process of
conducting a legal and/or business analysis of
the patent or patents in your research area is an
excellent place to start.
If the claims of the patent(s) behind your
critical technology give exclusive access to
a large unmet medical need and freedom
to operate, then you should feel confident.
Investors will require you to show that the
technology is indeed ‘first in class’, so you will
need to understand the current state of the
art and incorporate competitive technologies
into research plans. KeraNetics had a substan-
tial portfolio of 20 initial patent applications
when it started, and an extensive freedom to
operate analysis was conducted soon after the
company was formed.
© 2012 Nature America, Inc. All rights reserved.
Papers. Peer-reviewed reports, especially
those in high-impact journals, legitimize your
research and give credibility to the technology
behind your startup to investors. They also help
support the patents and are often the basis for
regulatory activities, such as establishing clas-
sification of the technology via its intended
mode of action. Key papers, particularly those
using large-animal models, can serve as safety
and efficacy data for obtaining FDA approval
for human clinical trials. A solid understand-
ing of the basic science related to the technol-
ogy is vital to these and other processes, such
as product and/or process optimization and
developing a pipeline of future products. At
KeraNetics, for example, large-animal data
were in place for three different products
npg
when the company was formed, as well as half
a dozen key papers either published or under
review.
Preliminary data. The more data behind your
technology or asset, the better. It is important
because it helps establish a technology platform
that will be the future of the startup company.
More importantly, preliminary data fuel grant
submissions, which can provide critical, non-
dilutive funding to a startup while the company
is operating under negative cash flow.
When KeraNetics was founded, three years
of unpublished data were available, mostly
in cell culture and rodent models, and this
was used for several early grant applications.
Although slow to start, the company was
awarded more than $8 million in grants in its
first three years.
Phunding. The importance of funding is obvi-
ous, but perhaps for more reasons than some
nature biotechnology volume 30 number 9 september 2012 823
would think. Having a robust, well-funded
academic research program (in addition to
the funding the startup itself will need) within
the university that can support and expand
the capabilities of your startup can enhance
your ability to address the unexpected or take
advantage of new opportunities. This helps
reduce risk. In the US, such funding takes
the form of traditional grants from agencies,
such as the Department of Defense (DoD), the
National Science Foundation (NSF) and the
National Institutes of Health (NIH).
On the company funding side, it will be
key for you to structure the startup so that it
is competitive for nondilutive grant funding.
The role of grant reviewer should already be
familiar to you, so it can be relatively easy to
evaluate a startup’s ability to hit the goals of a
grant. The problem exists in the nature of a
startup. Such early-stage ventures often have
no track record of commercialization, no fed-
eral funding, no assets, no technical employees
of their own, and no research space indepen-
dent of the founder’s academic lab, so what
would make them competitive for funding?
If the research is the only novelty, why would
the research fit better as a company-funded
grant rather than a traditional NIH grant to
the principal investigator?
The answer is that your company must offer
a high potential for commercialization of the
technology. This can happen only with the
infrastructure that comes with a substantive
startup, such as working capital, independent
lab space, equipment and a few employees on
the payroll. Fortunately, many universities now
offer incubator space that can be rented at low
cost and give access to core facilities that would
otherwise be beyond the financial means of a
typical startup. This is also another reason why
some initial capital is essential, not only for
paying the rent but also for hiring employees
and perhaps acquiring a few pieces of essential
research or production equipment.
When most people think of startup grant
funding, they naturally gravitate to the ubiq-
uitous Small Business Innovation Research
(SBIR) and Small Business Technology
Transfer (STTR) programs at many govern-
ment agencies, but there are many other
funding agencies that support effective
industry-academic partnerships. Many DoD
Program Announcements, for example, do
not exclude for-profit entities from eligibility.
To the extent that the startup can demonstrate
an effective, functional partnership, an ability
to meet the needs of the DoD better than an
academic group alone and preferably a track
record of having worked through the issues
associated with such arrangements with the
university, the company will be successful.
building a business
While no funding source should be over-
looked, the startup may find that non-SBIR
or non-STTR mechanisms offer longer-term
grant periods and higher maximum funding
caps.
Beyond grants and the initial funding, it is
best to raise additional equity only when there
is demonstrable progress, such as completing
a pivotal preclinical study. Raising money on
bad news can be extremely detrimental to a
startup, so if funding is to be raised in several
tranches, it is best if it happens after success-
ful binary events. But in general, always raise
more money than you think you will need.
KeraNetics has gone through three funding
rounds and fortunately has always had good
news on which to raise money. A fairly steady
stream of incoming grant funding has helped
in this regard, as well as favorable data from
pivotal studies.
You must have critical mass in all these areas
before launching your startup. But what con-
stitutes critical mass? Is one patent, one paper
or one grant enough? This will differ depend-
ing on the technology. Again, it is best to begin
with the end in mind and look to other suc-
cessful academic startups for guidance. For
example, if the university has a track record
of building successful startups based on single
patents or narrowly focused technologies,
chances are they can help find management
and early investors.
Focus and expecting change
These five Ps are just guidelines—the one truth
in biotech startups is that things will change.
Having a well thought-out, carefully consid-
ered business plan is important for maintain-
ing the focus and direction of the company.
Academics are naturally curious and are often
tempted to follow new discoveries. This can be
a mistake in an academic startup; it is critical
that a company stay on track and be aware of
necessary changes to the business plan or new
opportunities. Again, experienced management
can anticipate the need to adjust the business
plan before change is forced upon the startup.
Funding helps mitigate risk, but the distrac-
tion of raising money, whether from investors
or grants, can often divert the company from
execution of its business plan. The key is to
raise enough money initially, and if additional
funding is required, raise it as infrequently as
possible to avoid dilution or the dreaded ‘down
round’ in which funding is more costly to the
current investors, including you as a founder.
Critical mass in the five Ps will help immensely
when raising initial capital and will make that
money some of the cheapest to obtain.
However, contingencies are difficult to
envision and execute if the technology does
 building a business
not lend itself to change or new opportuni-
ties. Many biotech companies have survived
the long and risky road to regulatory approval
by executing contingencies in their business
plan. These can range from orphan status for
a narrow application of a drug to development
of diagnostic tools based on the company’s
research. Startup companies will often consider
non-biomedical uses for their technology so
that the uncertainty of regulatory approval can
be avoided for these products. These strategies
can range from research-use-only products to
veterinary and even non-medical applications.
Finally, it is always best to maintain a resil-
ient demeanor and instill this in the company’s
employees. The startup game is not for every-
© 2012 Nature America, Inc. All rights reserved.
npg
824 volume 30 number 9 september 2012 nature biotechnology
one, and academic faculty need to carefully
consider their own professional and personal
goals before venturing into this often chaotic
space. Many faculty, especially those with
a history of strong institutional support for
their research, may be uncomfortable with
the notion that the startup company may
only have months’, weeks’ or even days’ worth
of operating capital in the bank. If a startup
seems like an unattractive option at this point,
you may want to consider a more arms-length
relationship with the startup, a simple licens-
ing deal with an existing company or handing
the technology off to someone with a higher
risk tolerance, such as a close colleague or
trusted postdoc. However, a startup company
offers wonderful educational opportunities to
graduate students and postdocs, it can greatly
expand the translational potential of one’s
research, and ultimately it can lead to commer-
cialization of technologies that are of benefit
to patients—a worthwhile goal for the ultimate
dispensation of your life’s work.
COMPETING FINANCIAL INTERESTS
The author declares competing financial interests:
details are available at http://www.nature.com/
doifinder/10.1038/nbt.2330.
For more content on bioentrepreneurism,
visit our Trade Secrets blog.
http://blogs.nature.com/trade_secrets/

To be or not to be transgenic
To the Editor:
It is sad and ironic that even though much
progress has been made in deciphering
the genetic content of food plants and
modifying their genomes for the betterment of
humankind, many of the principles of modern
plant genetics, firmly established decades
ago, are now so easily
forgotten or ignored. Such
is the case with many of the
© 2012 Nature America, Inc. All rights reserved.
alarmist arguments raised
in the News Feature by
Emily Waltz1 in the March
issue entitled “Tiptoeing
around transgenics.” Waltz
focuses on the controversy
surrounding the regulation
of modern (and, in
fact, not-so-modern)
biotechnological techniques,
such as those that alter single
base pairs by replacing one
nucleotide with another (that is, create single-
nucleotide polymorphisms or SNPs). We feel it
is important to stress that such genetic changes
must be viewed in a historical and biological
context to understand why calls for new layers
of regulation over technologies that introduce
npg
SNPs and other changes are unwarranted.
The most relevant counterargument
to the need for regulation is the fact that
mutations normally happen. Mutations
occur spontaneously in nature, and their
rate can be increased by the use of mutagens.
On the whole, mutation is a good thing, for
without mutation, we would still be biofilm
on the bottom of the ocean. Although typical
mutation rates are quite low when calculated
on a gene or base-pair basis, they are high
enough that new mutations are the rule rather
than the exception. For example in Arabidopsis
thaliana, the mutation rate per base pair per
generation is estimated to be 7 per billion base
pairs2. Given that there are 125,000,000 base
pairs in the A. thaliana genome, 1.75 new SNP
mutations are expected per generation per
diploid plant. Although SNPs appear to occur
at about the same rate in all plants, crop plants
have larger genomes, and thus more SNPs. Just
one average hectare of 240,000 soybean plants
nature biotechnology volume 30 number 9 september 2012 825
correspondence
can be expected to contain about 1.8 million
novel SNPs.
Not only are SNPs commonplace but
techniques that create SNPs have a long history
of safe use by breeders. Before the advent of the
techniques described by Waltz, the only tool
available to breeders to alter DNA sequences
was the use of radiation
and chemical mutagens.
The Food and Agricultural
Organization (Rome) and
International Atomic Energy
Agency maintain a database
(http://www-infocris.iaea.
org/MVD/) that currently
lists 2,543 known plant
varieties developed through
mutagenesis, including
many common or widely
grown and consumed crop
plants, of which 14% were
derived with chemical
mutatagens3. Chemical mutagens are still
used to create the same kind of SNPs4 that are
cited as a cause for concern in the Waltz News
Feature. Although the genetic basis and extent
of SNPs for the mutant phenotype are usually
unknown when mutagenesis is employed, the
resulting crops are considered as safe as others
and are not subject to premarket regulatory
review.
Breeders depend on sheer luck to find an
alteration in the gene encoding their trait
of interest when they employ mutagenesis,
and they must accept random alterations
elsewhere in the genome whenever these do
not affect crop growth, performance and yield
to an unacceptable point. Today, by using in
vitro techniques, breeders have the ability to
target the gene of interest, and not introduce
unintended and unwanted mutations
elsewhere in the genome. If anything,
therefore, modern techniques should decrease
concerns for safety, not raise them.
This leads us to the question of whether
SNPs alter protein safety. SNPs accumulate
in plants and animals. One simple means of
quantifying SNP formation is by comparing
SNP differences between pairs of genotypes
or varieties5. For example, 0.05% of bases in
soybean coding regions are SNPs, or one SNP
per 2,000 bp in coding regions; the frequency
in noncoding regions is 0.5%, or 1 per
191 bp6. This is similar to the level of SNPs
in the human genome. Such crops as maize
are much more diverse; in this cereal, SNPs
account for as many as 1.3% of base pairs7.
Tenaillon et al.8 have estimated that any two
alleles of a maize gene for a 300–400-amino-
acid protein would differ by 3.5 amino
acids due to SNP accumulation. Within a
diverse population, there are likely to be
15–20-amino-acid differences between
proteins from two alleles of a single maize
gene. It is therefore not surprising that
attempts at protein engineering have not
converted enzymes into toxins, as toxic
proteins have substantial structural differences
from other proteins and need to perform
specific physiological roles to act as toxins9.
SNPs are thus really minor variations
compared with the larger-scale changes
that have accumulated in crops during
domestication and breeding. A case in point
are the elongated tomatoes on today’s market
(which could fall under the category of
cisgenics, another technology highlighted by
Waltz). However, in the tomato’s case, its DNA
got copied and moved to another location
in the genome through naturally occurring
mechanisms10, most probably after the
tomato’s arrival in Spain11.
The Waltz article also discusses the
new technology developed by Pioneer
HiBred (Des Moines, IA, USA) in which
transgenic plants produce nontransgenic,
male sterile plants that are used in hybrid
production. The argument is made that
although these plants are not transgenic
per se, they should be viewed as such. But if
this same ‘sins of the fathers’ argument were
applied elsewhere in agriculture, humans
should not consume modern-day tomatoes
because their parents contained a toxin.
This nonsensical argument is not applied to
conventional plant varieties and, therefore,
there is no reason why it should be applied to
transgenics.
Despite millennia of plant genetic
modification, thus far we have not found
 correspondence
a single verified report whereby breeding
or radiation and/or chemical mutagenesis
resulted in a toxin, allergen or other hazard
that was not known to exist before. These
facts support the conclusion that DNA
insertions and other types of mutations
do not pose unreasonable risks to the
environment or to human and animal
health, regardless of how they came about.
COMPETING FINANCIAL INTERESTS
The authors declare competing financial interests:
details are available at http://www.nature.com/
doifinder/10.1038/nbt.2347.
Wayne A Parrott1, Joseph M Jez2 &
L Curtis Hannah3
1University of Georgia, Athens, Georgia, USA.
2Washington University, St. Louis, Missouri, USA.
Broad consent in biobanking
© 2012 Nature America, Inc. All rights reserved.
To the Editor:
The Feature in the February issue by Scott
et al.1 on the policy challenges of biobanking
characterizes broad specimen donor
informed consent as “ethically contentious.”
A survey of public attitudes is cited. This
same survey found that a significant
percentage of individuals are prepared “to
consent broadly to future research use and to
forego additional choices as a result”2.
With our perspectives in patient advocacy
or at research centers aimed at bringing new
regenerative therapies to patients, we have
consistently emphasized the value of research
donors’ perspectives. In the context of
protocols for creating immortalized cell lines
npg
for banking and distribution, we have also
witnessed support for broad consent. Indeed,
enthusiasm is even more pronounced among
those touched by disease, and patient donors
actually express concern that study-specific
An accelerated workflow for
untargeted metabolomics using the
METLIN database
To the Editor:
Metabolites, which are typically recognized as
small molecules that are involved in cellular
reactions, provide a functional signature
of phenotype that is complementary to the
upstream biochemical information obtained
826 volume 30 number 9 September 2012 nature biotechnology
3University of Florida, Gainesville, Florida, USA.
e-mail: wparrott@uga.edu
1. Waltz, E. Nat. Biotechnol. 30, 215–217 (2012).
2. Ossowski, S. et al. Science 327, 92–94 (2010).
3. Ahloowalia, B.S., Maluszynski, M. & Nichterlein, K.
Euphytica 135, 187–204 (2004).
4. Balyan, H.S., Sreenivasulu, N., Riera-Lizarazu, O.,
Azhaguvel, R. & Kianian, S.F. in Advances in Agronomy,
Vol. 98 (ed. Sparks, D.L.) 357–414 (2008).
5. Tajima, F. Genetics 105, 437–460 (1983).
6. Huan, N.V., Sugimoto, H. & Harada, K. Breed. Sci. 55,
441–446 (2005).
7. Ross-Ibarra, J., Morrell, P.L. & Gaut, B.S. Proc. Natl.
Acad. Sci. USA 104, 8641–8648 (2007).
8. Tenaillon, M.I. et al. Proc. Natl. Acad. Sci. USA 98,
9161–9166 (2001).
9. Pariza, M.W. & Cook, M. Regul. Toxicol. Pharmacol. 56,
332–342 (2010).
10. Xiao, H., Jiang, N., Schaffner, E., Stockinger, E.J. &
van der Knaap, E. Science 319, 1527–1530 (2008).
11. Rodriguez, G.R. et al. Plant Physiol. 156, 275–285
(2011).
consent can be burdensome and impede
research.
This experience suggests to us that broad
consent is ethically responsible, provided
there is comprehensive oversight and a robust
informed consent process. With the continued
support of donors, we look forward to
applying this model in biobanking efforts.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Chris Hempel1, Geoffrey Lomax2 &
Steve Peckman3
1Addi & Cassi Fund, Reno, Nevada, USA.
2California Institute for Regenerative Medicine,
San Francisco, California, USA. 3Eli and Edythe
Broad Center of Regenerative Medicine, University
of California, Los Angeles, California, USA.
e-mail: glomax@cirm.ca.gov
1. Scott, C.T. et al. Nat. Biotechnol. 30, 141–147
(2012).
2. Simon, C.M. et al. Genet. Med. 13, 821–831 (2011).
from genes, transcripts and proteins. The
high correlation between metabolites and
phenotype has created a surge of interest in
the field that is reflected in the number of
metabolomic publications growing from just
a few articles in 1999 to over 5,000 in 2011.
Although relatively new compared with
its genomic and proteomic predecessors,
research in the field of metabolomics has
already led to the discovery of biomarkers for
disease, fundamental insights into cellular
biochemistry and clues related to disease
pathogenesis1,2.
The success of metabolomics over the
past decade has relied largely on advances in
mass spectrometry instrumentation, which
make it possible to detect thousands of
metabolites simultaneously from a biological
sample. Coupled with developments in
bioinformatic tools such as XCMS Online
(https://xcmsonline.scripps.edu/)3, it
has now become relatively routine to
comprehensively compare the intensities of
thousands of metabolite peaks in one sample
group to those in another in an untargeted
manner. This approach, called untargeted
metabolomics, has the potential to implicate
unexpected pathways with a unique
phenotype or disease process.
Despite the attractiveness of having a
comprehensive and unbiased approach
for profiling metabolites that is analogous
to those used in the other ‘omic’ sciences,
an overwhelming proportion of the
metabolomic community exclusively uses a
targeted platform in which only a specified
list of metabolites is measured. The benefit
of such a targeted platform is speed. Unlike
the untargeted platform, after the targeted
mass spectrometry methods are established,
minimal effort and resources are required
to profile these specific metabolites over a
large number of samples. In contrast, the
major bottleneck of untargeted metabolomics
has been the challenge of determining
the identities of the peaks found to be
dysregulated in the untargeted profiling data.
Traditionally, the untargeted
metabolomic platform involves multiple
steps (Fig. 1). The first step is acquiring
global mass spectrometry data for each of
the samples. Next, these data are analyzed
using bioinformatic software that performs
quantitative analyses to find peaks that
are significantly different between sample
groups. The investigator then typically
searches the mass-to-charge (m/z) ratios of
the peaks of interest manually in metabolite
databases. Searches that return hits within
the mass accuracy of the instrument are
considered to be putative identifications.
To confirm the identifications, tandem
mass spectrometry (MS/MS) data from the
research sample are then compared to the
MS/MS data of a commercial standard. To
obtain the MS/MS data, a targeted MS/MS
analysis is typically performed on one of

Autonomous metabolomic workflow
Step 1
Data acquistion
MS and MS/MS
XCMS Online
Traditional metabolomic workflow
Step 1 Step 2 Step 3
MS Comparative Manual MS
acquisition analysis database
search
Figure 1 Schematic representation of the traditional metabolomic workflow involving six steps
and the new METLIN-based workflow with only two steps. In the two-step autonomous workflow,
mass spectrometry (MS) and MS/MS data are acquired simultaneously during profiling and
searched in the METLIN database for automated identification, thereby reducing the time of the
workflow from days or weeks to minutes or hours.
the research samples for which the peak
was determined to be upregulated. The
© 2012 Nature America, Inc. All rights reserved.
fragmentation pattern of the MS/MS data
is then manually compared with that of the
MS/MS data from a commercial standard
(however, not all commercial standards, for
example, stereoisomers, can be resolved by
MS/MS data alone).
To facilitate identification of metabolites
in the untargeted workflow, we launched a
freely accessible metabolite database called
METLIN4 in 2004 (http://metlin.scripps.edu/)
that incorporates MS/MS data from model
compounds. Recently, other metabolite
databases such as the Human Metabolome
Database (HMDB)5, MassBank6 and
LipidMaps7 have also begun incorporating
MS/MS data for standard compounds. These
repositories allow investigators to compare
MS/MS data from their research samples
npg
to MS/MS data from model compounds
catalogued in the database and thereby
improve the speed, efficiency and cost
effectiveness of untargeted studies.
Over the past 7 years, our objective has
been to generate a sufficiently large MS/MS
data library that can be used in an automated
manner to revise the traditional untargeted
metabolomic workflow (Fig. 1). Since we
originally reported the establishment of
METLIN in 2005, we have increased the
number of MS/MS spectra that are included
in the database by a factor of 150. As of
April 2012, METLIN contains MS/MS data
on >10,000 distinct metabolites at four
different collision energies. These data were
collected using an electrospray ionization/
quadrupole time-of-flight (ESI-QTOF)
mass spectrometer in both the positive and
negative detection modes, representing a
total number of >48,000 high-resolution
spectra. To estimate the current coverage
nature biotechnology volume 30 number 9 september 2012 827
correspondence
database that has advanced functionality
Step 2 to automate metabolite identification and
reduce the labor-intensive bottleneck that has
Quantitative comparative analysis
and metabolite identification traditionally been associated with untargeted
metabolite profiling. Instead of manually
comparing the MS/MS data from research
samples to the MS/MS data of commercial
standards, the new version of METLIN
allows metabolomic investigators to upload
Step 4 Step 5 Step 6
their MS/MS data to the METLIN database
Manual Targeted MS/MS Manual Comparison
aquisition MS/MS spectra
ID so the comparisons can be performed in an
automated way. By using automated MS/
MS data matching, metabolite identities
can be confirmed much more efficiently
and quickly compared with the traditional
untargeted metabolomic workflow. The
quality of the match between the MS/MS data
from the research sample and the MS/MS
data from the METLIN library is measured
of physiologically relevant metabolites in by a newly introduced METLIN scoring
METLIN and the other three largest databases system, which is based on a modified version
available (HMDB, MassBank and LipidMaps), of the established X-Rank scoring system9. To
metabolites were isolated from Escherichia evaluate the correlation of METLIN MS/MS
coli and standard human serum using defined data to MS/MS data acquired using different
protocols8. Samples were analyzed in both instrument platforms, we performed a
the positive and negative modes with an ESI- comparative experiment was performed
QTOF mass spectrometer (Supplementary using 23 metabolite standards. The
Methods). Each peak detected (excluding compounds were measured on five different
isotopes) was searched in each of the four instruments, and the resulting spectra were
databases. Figure 2 shows the number of hits matched against the METLIN database.
for each database and also the subset of the Based on the modified X-Rank scoring
hits for which MS/MS data were available to system, the correct result was returned as the
confirm the metabolite identification. first hit for 90 out of the 101 spectra (89.1%;
In addition to its increased size, here Supplementary Table 1 and Supplementary
we describe a new version of the METLIN Figs 1–27).
Database hits for E. coli and human serum
Total
6,000
METLIN
Total
METLIN Human
5,000
E. coli serum
Hits with MS/MS Hits with MS/MS
Hits without MS/MS Hits without MS/MS
4,000
Number of hits
LipidMaps LipidMaps
3,000
HMDB HMDB
2,000
MassBank
MassBank
1,000
0
Figure 2 Estimate of the physiological relevance of metabolite coverage in metabolomic databases.
Metabolites from human serum and E. coli were isolated and analyzed in both the positive and negative
modes by ESI-QTOF mass spectrometry, and the mass of each metabolite was searched with a tolerance
of 5 parts per million in the METLIN, LipidMaps, HMDB and MassBank databases. LipidMaps contains
data primarily on lipids, which is only a subset of the metabolome, but was included in the comparison
for the sake of completeness. For human serum, 12,170 features were detected and searched, and for
E. coli, 11,641 features were detected and searched. The number of hits on the basis of accurate mass
are shown in light blue and light red for E. coli and human serum, respectively. The subset of those
hits that also contained MS/MS data are shown in dark blue and dark red for E. coli and human serum,
respectively.
 correspondence
Some classes of metabolites produce
characteristic fragments or neutral losses
in their MS/MS spectra that can be used as
signatures for unique chemical functional
groups. For example, the MS/MS spectra
of phosphatidylcholines are characterized
by a fragment at m/z 184.07. For instances
in which the MS/MS data uploaded by a
user do not match any compound in the
database, the new version of the METLIN
database will search the MS/MS data for
characteristic fragments that can be used
for molecular classification. The search can
also be performed manually by accessing
the ‘fragment search’ or ‘neutral loss
search’ options. These tools provide a new
mechanism by which unknown metabolites
can be chemically classified, and they take
advantage of the large amount of MS/MS data
in the library.
To highlight the new database
© 2012 Nature America, Inc. All rights reserved.
functionalities, we performed MS/MS on
select peaks from the metabolite extracts of
E. coli and human serum. These data were
uploaded to the METLIN database, and
fragment matching was performed using
the automated feature described above.
Representative examples of metabolites
identified on the basis of the mass
spectrometry and MS/MS data using this
method are shown in Supplementary Figures
28–32. The compounds identified ranged
from lipids to smaller, polar metabolites.
Additionally, representative examples of
unknown compounds that were classified by
characteristic fragments are also shown.
With the combination of the METLIN
functionalities described here and the
increasing speed of QTOF instrumentation
npg
for performing MS/MS, there is the potential
to reduce the untargeted metabolomic
workflow to just two steps (Fig. 1). Using
high–scan-speed QTOF instruments,
mass spectrometry and MS/MS data can
be acquired simultaneously in a single
run. Quantitative information can then
be extracted from the data using the
bioinformatic software XCMS Online, and
metabolites can be identified simultaneously
by matching the MS/MS data with MS/MS
data in the METLIN database in an
automated fashion, an approach that is self-
directed or autonomous in nature.
With this truncated workflow, the time
needed to perform untargeted profiling and
the subsequent metabolite identification
may be reduced to minutes or hours as
compared to the days or weeks needed with
the traditional workflow. The results shown
here from automated MS/MS matching
highlight the applicability of the method for
828 volume 30 number 9 September 2012 nature biotechnology
performing high-throughput, untargeted
metabolomics using this type of accelerated
workflow. Moreover, we have shown that the
coverage of the METLIN database enables
the characterization and identification
of thousands of naturally occurring
metabolites in biological samples. Thus, the
new METLIN database has the potential
to expedite the workflow for untargeted
metabolomics as more investigators obtain
mass spectrometry instrumentation that
can produce high-quality MS/MS data with
increasing speed and sensitivity.
Note: Supplementary information is available at http://
www.nature.com/doifinder/10.1038/nbt.2348.
ACKNOWLEDGMENTS
This work was supported by the California Institute of
Regenerative Medicine (TR1-01219), the US National
Institutes of Health (R24 EY017540, P30 MH062261,
RC1 HL101034 and P01 DA026146) and the US
National Institutes of Health National Institute on
Aging L30 AG0 038036 (G.J.P.). Financial support
was also received from the US Department of Energy
(grants FG02-07ER64325 and DE-AC0205CH11231).
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Successful suppression of a field
mosquito population by sustained
release of engineered male
mosquitoes
To the Editor:
Our paper published last year described the
results of preliminary release experiments
showing that engineered
sterile male mosquitoes
could mate with females
in a wild population in
the Cayman Islands1. This
trial was supported by
simple simulation models
indicating that sustained
release of sufficient numbers
of such males should
substantially suppress a
target population within
a few weeks or months2–4.
In the following letter, we
describe a field release
experiment testing this proposition.
The sterile insect technique is an
environmentally friendly, species-specific
method of pest control that is used to
Ralf Tautenhahn1,2, Kevin Cho1,2,
Winnie Uritboonthai1,2, Zhengjiang Zhu1,2,
Gary J Patti3–5 & Gary Siuzdak1,2
1Department of Chemistry, Center for
Metabolomics, The Scripps Research Institute,
La Jolla, California, USA. 2Department of
Molecular Biology, The Scripps Research Institute,
La Jolla, California, USA. 3Department of
Chemistry, Washington University School of
Medicine, St. Louis, Missouri, USA. 4Department
of Genetics, Washington University School of
Medicine, St. Louis, Missouri, USA. 5Department
of Medicine, Washington University School of
Medicine, St. Louis, Missouri, USA.
e-mail: gjpattij@wustl.edu or siuzdak@scripps.edu
1. Yanes, O. et al. Nat. Chem. Biol. 6, 411–417 (2010).
2. Patti, G.J. et al. Nat. Chem. Biol. 8, 232-234 (2012).
3. Smith, C.A., Want, E.J., O’Maille, G., Abagyan, R. &
Siuzdak, G. Anal. Chem. 78, 779–787 (2006).
4. Smith, C.A. et al. Ther. Drug Monit. 27, 747–751
(2005).
5. Wishart, D.S. et al. Nucleic Acids Res. 37, D603–D610
(2009).
6. Horai, H. et al. J. Mass Spectrom. 45, 703–714 (2010).
7. Sud, M. et al. Nucleic Acids Res. 35, D527–D532
(2007).
8. Yanes, O., Tautenhahn, R., Patti, G.J. & Siuzdak, G.
Anal. Chem. 83, 2152–2161 (2011).
9. Mylonas, R. et al. Anal. Chem. 81, 7604–7610 (2009).
successfully control several agricultural
pest insects5. Large numbers of sterile
insects are released to mate with their wild
counterparts and thereby
reduce their reproductive
potential. However,
despite its attractive
features, this technique
is not in operational use
against mosquitoes, in
part because of damaging
effects of sterilizing doses
of radiation on the released
mosquitoes6–8. Following a
similar principle, we have
proposed that engineered
males carrying a dominant
lethal transgene could
be released to mate with wild females; the
resulting progeny would die as a result of the
lethal effect of the transgene. We named this
system RIDL (release of insects carrying a
 correspondence

dominant lethal gene)9,10. The Aedes aegypti a

RIDL strain, OX513A, has a single transgenic
sequence encoding a red fluorescent marker
and tetracycline-repressible late-acting
dominant lethality3.
Approximately 3.3 million engineered
OX513A males (male mosquitoes do not
bite) were released in a 23-week period George ●
East
in 2010 in a field site in Grand Cayman, a ● Town
End
British Overseas Territory in the Caribbean. ● Bodden 5 km
The preliminary release experiments Town

showed that OX513A males could mate b

with wild females1 and, together with
simulation models3, indicated a minimum
release rate of 3,150 males per hectare (ha) Area A
per week to induce population collapse in
the absence of immigration. We aimed for Area B
twin targets of the release of >4,000 males
per ha per week and a 10:1 ratio of sterile to
wild males in the field. Area C
Releasing large numbers of engineered
© 2012 Nature America, Inc. All rights reserved.

sterile males into a wild mosquito population 500 m

should have several measurable effects on
the wild population, including, in temporal c Release rate/ha/week

order, an increase of the male-to-female ratio, 16,000

Period 1 Period 2 Period 3
females mating with engineered males and 14,000
12,000 Pupal release
suppression of the target population. The 10,000
8,000 Adult release
mosquito population—OX513A and wild— 6,000
4,000
was monitored using adult traps and ovitraps, 2,000
which mimic natural oviposition sites 0
ay ay n n l l l g g p p ct
(Supplementary Notes). Adult traps were M M Ju Ju Ju Ju Ju Au Au Se Se O
5 19 3 17 31 9
also used to monitor the numbers of OX513A 8 22 14 28 11 25
males in the field, as an input of males should
Figure 1 Field site and mosquito releases in 2010. (a) Map of Grand Cayman showing the locations of
change the sex ratio of the population. Larvae
East End and Bodden Town, which were the treatment and external control sites, respectively, and the
hatched from field-collected eggs from capital, George Town. (b) Aerial photograph of East End showing areas A, B and C. (c) Weekly numbers
ovitraps were screened for fluorescence to of adult males released per hectare from direct adult release (solid bars) and emerging from pupal
determine paternity (fluorescent larvae had deployment (shaded bars). Releases occurred 3 times per week; release numbers shown for each week
an OX513A father, and nonfluorescent larvae are the sum of these three releases. During period 1, all treatment areas received treatment, during
had a wild-type father). These monitoring period 2, areas A and B were treated, and during period 3, only area A was treated.
npg

methods were used to evaluate the field trial

endpoints: sterile-to-wild male ratio ≥10:1,
fluorescence ratio ≥50% and statistically
significant population suppression relative to
control site(s).
Initial releases of OX513A males across
a 55-ha area (areas A, B and C, period 1;
Fig. 1) and then across a reduced 32-ha area
(areas A and B, period 2) were restricted by
production difficulties to an average of 1,400
(95% CI 990–1,800) males per ha per week
and 3,900 (95% CI 2,600–5,300) males per
ha per week, respectively (Fig. 1c). These
release rates also did not achieve the target
OX513A-to-wild male ratios, reaching a 1.9:1
ratio (95% bootstrap CI 1.2:1 to 2.8:1, n =
967) in period 1 and a 4.8:1 OX513A-to-wild
ratio (95% bootstrap CI 2.6:1 to 7.9:1,
n = 1,994) in period 2 (Fig. 2a). We therefore
further reduced the release area to 16 ha
(area A, period 3). In this final period, we
released ~14,000 (95% CI 13,700–14,500)
males per ha per week for the first 3 weeks in areas C (likelihood ratio (LR) P = 0.63)
and later reduced this to ~7,700 (95% CI and B (LR P = 0.13) but were significantly
6,900–8,500) adult males per ha per week, lower (~40%) than those in an external
which was supplemented with ~4,900 (95% untreated control site (LR P = 0.026). After
CI 3,800–6,000) adults eclosing from ~5,600 that time point, the index in area A was
(95% CI 4,500–6,800) pupae deployed in highly significantly lower than those in all
field. This gave a mean OX513A-to-wild other areas (P < 0.0001 for each pairwise
male ratio in the release area of 25.2:1 (95% comparison).
bootstrap CI 17.8:1 to 34.9:1, n = 3,155) Over the last 7 weeks of the release period,
in the first 4 weeks of period 3. We also the mean ovitrap index in the untreated
found an increase in the proportion of the areas was 49% (95% CI 43–55%; Fig. 2c). In
field-collected eggs carrying the fluorescent contrast, the mean ovitrap index in area A
marker, with a peak of 88% (Fig. 2b), was 10% (95% CI 7–14%), which is an 80%
implying that the majority of wild females reduction relative to the untreated areas,
were mating with OX513A males. indicating strong population suppression
We used the ovitrap index—the in the treated area during this period. The
proportion of ovitraps in each area with one degree of suppression that is possible in
or more eggs after 1 week—as our primary such a trial is limited by immigration of wild
measure of population density (Fig. 2c). females from surrounding areas, such as area
Until early in period 3, the weekly ovitrap B, as well as, potentially, from eggs laid at an
indices for area A were very similar to those earlier period.

nature biotechnology volume 30 number 9 september 2012 829

 correspondence

a Period 1 2 3 e Figure 2 Effect of periodic release of OX513A male mosquitoes on a wild

population. Release of engineered males should lead to an increase in the
Estimated OX513A:WT adult male ratio OX513A fluorescence profile
male-to-female ratio of the field population, deposition of hybrid eggs after
100
mating of released males with wild females and, finally, suppression of the
target population if sufficient sterility or mortality is induced. (a) The ratio of
10 OX513A to wild-type (WT) adult males estimated from the sex ratio of adults
caught in BG-Sentinel traps. Treatment periods 1–3 are indicated (see also
1 Fig. 1). Increasing the input rate (males per ha per week) and, later, declining
b Percentage of fluorescent larvae wild population numbers resulted in an increase in this ratio over time.
100
(b) Percentage of larvae recovered from ovitraps in treated areas with the
80
RIDL transgene as detected by fluorescence. This percentage is plotted
60
only into September 2010, as the number of eggs collected after that time
40
became too low to act as a reliable measure because of suppression of the
20
target population. (c) Ovitrap index in East End (areas A–C) and the external
0
c Percentage ovitrap index
untreated control site, Bodden Town. Area C received low-level treatment
100 Area A early in the trial (period 1) for a duration and level that were not considered
Area B effective. Because of its close proximity and ecology to area A, area C
80
Area C provided a largely ‘untreated’ internal control that was highly comparable to
60 Bodden Town
area A. From the beginning of August, the ovitrap index in area A declined
40
relative to all control areas. All populations declined later, which is typical of
20 seasonal variation driven by rainfall. However, even after some rainfall from
0 March 2011, by June, the population in area A was still below that in control
d Weekly rainfall in East End (mm) area C. (d) Weekly rainfall in East End in 2010 (mm). (e) Red fluorescence
200
150
of OX513A larvae. Four larvae are shown, of which the upper two are wild-
type and the lower two are OX513A. The larvae were imaged under normal
© 2012 Nature America, Inc. All rights reserved.

100
50 illumination (left) and epifluorescence (right). Red fluorescence of the
0
OX513A larvae is caused by expression of DsRed2 (ref. 3).
Ap
r ay n
Ju
l g p ct r
ov Dec Jan Feb Ma Ap
r ay un
M Ju
1 Au Se O N 1 M 1J
1 1 1 1 1 1 1 1 1 1 1 1

The time delay inherent in sterile-male– ACKNOWLEDGMENTS 1Mosquito Research and Control Unit, Grand

release methods (the population reduction We thank L. New and J. Renmant for assistance in egg
production, Z. Ebanks and E. Ebanks for technical
occurs in the progeny of released insects),
help in Grand Cayman, T. Matthews and E. Moxon
combined with female monogamy, means for administrative support and all staff and students
that we were releasing a sufficient number
at the Mosquito Research Control Unit and Oxitec
of males of sufficient quality to achieve for their help and support during this study. We
thank G. Labbé and P. Gray for comments on the
suppression at least 4–6 weeks before the
manuscript and the Lands and Survey Department
point at which population suppression was of the Cayman Islands Government for permission
detected (5 August 2010; Fig. 2c). This to use imagery and data. A.F.H. thanks Adapco,
indicates that suppression was achieved by Bayer and Central Life Sciences for supporting her
releasing approximately 3,500 males per ha PhD studentship. I.B. and A.C. are PhD students
supported by Biotechnology and Biological Sciences
per week, which gave a male-to-female ratio
Research Council Collaborative Awards in Science
in adult traps of 3:1, corresponding to an
npg
Cayman, Cayman Islands. 2Vector Group,
Liverpool School of Tropical Medicine, Liverpool,
UK. 3Oxitec Limited, Oxford, UK. 4Department
of Zoology, University of Oxford, Oxford, UK.
5Medical Entomology Unit, Infectious Diseases
Research Centre, Institute for Medical Research,
Kuala Lumpur, Malaysia. 6Medical Research
Council Centre for Outbreak Analysis and
Modelling, Department of Infectious Disease
Epidemiology, Faculty of Medicine, Imperial
College London, St Mary’s Campus, London, UK.
e-mail: luke.alphey@oxitec.com

and Engineering. C.A.D. acknowledges the UK
overflooding ratio of ~5:1. Medical Research Council for Centre funding
This trial was conducted in an area support.
with no conventional control targeting
A. aegypti and a relatively high initial COMPETING FINANCIAL INTERESTS
The authors declare competing financial interests:
population density; larviciding and removal details are available at http://www.nature.com/
of breeding sites within an integrated doifinder/10.1038/nbt.2350.
program would greatly reduce the number
of sterile mosquitoes required. The positive
Angela F Harris1,2, Andrew R McKemey3,
outcome and successful demonstration of Derric Nimmo3, Zoe Curtis3, Isaac Black3,4,
population suppression is encouraging for Siân A Morgan3, Marco Neira Oviedo3,
genetic control strategies in general and, Renaud Lacroix3, Neil Naish3, Neil I Morrison3,
in particular, validates the potential of Amandine Collado3,4, Jessica Stevenson3,
OX513A RIDL mosquitoes for population Sarah Scaife3, Tarig Dafa’alla3,
suppression. Guoliang Fu3,4, Caroline Phillips3,
Andrea Miles3, Norzahira Raduan3,5,
Note: Supplementary information is available at http:// Nick Kelly1, Camilla Beech3, Christl A Donnelly6,
www.nature.com/doifinder/10.1038/nbt.2350. William D Petrie1 & Luke Alphey3,4
1. Harris, A.F. et al. Nat. Biotechnol. 29, 1034–1037
(2011).
2. Atkinson, M.P. et al. Proc. Natl. Acad. Sci. USA 104,
9540–9545 (2007).
3. Phuc, H.K. et al. BMC Biol. 5, 11 (2007).
4. Alphey, N., Alphey, L. & Bonsall, M.B. PLoS ONE 6,
e25384 (2011).
5. Dyck, V.A., Hendrichs, J. & Robinson, A.S. Sterile
Insect Technique: Principles and Practice in Area-Wide
Integrated Pest Management (Springer Netherlands,
2005).
6. Alphey, L. et al. Vector Borne Zoonotic Dis. 20,
295–311 (2010).
7. Helinski, M., Parker, A. & Knols, B.G.J. Malar. J. 5, 41
(2006).
8. Benedict, M.Q. & Robinson, A.S. Trends Parasitol. 19,
349–355 (2003).
9. Alphey, L., Nimmo, D., O’Connell, S. & Alphey, N. in
Transgenesis and the Management of Vector-Borne
Disease Vol. 627 (ed. Aksoy, S.) 93–103 (Landes
Bioscience, 2008).
10. Thomas, D.D., Donnelly, C.A., Wood, R.J. & Alphey, L.S.
Science 287, 2474–2476 (2000).

830 volume 30 number 9 September 2012 nature biotechnology


Teva v. AstraZeneca and secret prior art under
102(g)(2)
Sandra Lee & Michael Knierim
A recent Federal Circuit decision and the reasoning behind it could have a significant impact on the patentability of
other life science inventions, even after changes in the patent law.
O n December 1, 2011, the United States
© 2012 Nature America, Inc. All rights reserved.
Court of Appeals for the Federal Circuit
issued its opinion in Teva v. AstraZeneca1 con-
cerning an oft-neglected prior art provision—
35 USC § 102(g)(2). On its face, the opinion
reiterates well-established principles relating to
102(g)(2) prior art, which is often referred to
as ‘secret’ prior art because it lacks a publicity
requirement. Thus, something not yet publicly
disclosed as required by Section 102(a) and
(b) can be prior art under Section 102(g)(2).
However, lurking within the language of the
opinion’s text, there are two easily overlooked,
but nevertheless striking, facets to this case: the
court’s broad characterization of the invention,
and the fact that AstraZeneca actually invali-
dated Teva’s claim without meeting each and
every limitation.
The decision came just months after passage
npg
of the America Invents Act (AIA), the long-
awaited patent reform law which will effectively
remove Section 102(g) from Title 35 of the US
Code on March 16, 2013 (ref. 2). Although
some might be tempted to dismiss this opinion
as obviated by the new provisions of the AIA, the
decision thus will have relevance at least until
the current statute is replaced. Moreover, cer-
tain continuing applications claiming priority
before the effective date will also be governed
by the existing language of Section 102 (ref. 3).
With the long pendency times currently expe-
rienced at the US Patent and Trademark Office,
and the years before pending applications will
be involved in litigation, the existing language
of Section 102 will be relevant for many years to
come, both in prosecution and litigation.
Sandra Lee is a partner and Michael Knierim
is an associate at Baker Botts, New York, New
York, USA.
e-mail: sandra.lee@bakerbotts.com
nature biotechnology volume 30 number 9 SEPTEMBER 2012 831
Background
Section 102(g) was originally intended to codify
the law of interferences4. Nevertheless, courts
have long interpreted it as capable of being used
as a prior art provision, not limited to interfer-
ences5. In 1999, Congress amended Section
102(g), separating it into two parts to clarify its
dual role as both an interference provision and a
prior art provision6. Subsection 102(g)(2), being
the prior art provision, presently reads, “A per-
son shall be entitled to a patent unless…before
such person’s invention thereof, the invention
was made in this country by another inventor
who had not abandoned, suppressed, or con-
cealed it. In determining priority of invention
under this subsection, there shall be considered
not only the respective dates of conception and
reduction to practice of the invention, but also
the reasonable diligence of one who was first to
conceive and last to reduce to practice, from a
time prior to conception by the other.”
Section 102(g)(2) has two main requirements.
First, the inventive activity must occur in the
United States. Second, the invention must be
continuously used and “not abandoned, sup-
pressed, or concealed….” Thus, the Section
102(g)(2) analysis is different from that of the
other provisions of Section 102. Instead of first
establishing that a reference is prior art, and then
determining if the reference contains each and
every limitation of the claim, Section 102(g)(2)
requires a determination that an inventor has
made “the invention.”
Teva v. AstraZeneca
Prior art under Section 102(g)(2) often arises
when a party has reduced to practice a prod-
uct before the effective filing date of a patent.
It should be noted that although this gives
Section 102(g)(2) the flavor of prior use,
102(g)(2) is not a prior use defense. Indeed,
p at e n t s
this was the case in Teva v. AstraZeneca. The
accused product was Crestor (rosuvastatin),
a well-known stabilized statin formulation
for the treatment of high cholesterol. Statins
are widely known to treat high cholesterol,
but are inherently unstable. To be medically
viable, statins must be manufactured in sta-
bilized formulations.
AstraZeneca first produced the accused
product in mid-1999. The product contained
the following ingredients: (i) a statin; (ii) cro-
spovidone, which is an amido group containing
polymeric compound; and (iii) tribasic calcium
phosphate, which has a stabilizing effect on
statin formulations. At the time of the first pro-
duction, AstraZeneca understood that tribasic
calcium phosphate was a stabilizer. However,
AstraZeneca failed to appreciate that crospo-
vidone had a stabilizing effect, but instead was
adding it as a disintegrant.
The claim at issue was directed to a stabilized
statin formulation, where the stabilizing com-
pound is either an amido-group or amino-group
containing polymeric (AGCP) compound or
combination thereof. Teva filed a provisional
US patent application on April 10, 2000. It fol-
lowed up a year later with a utility application,
which issued in May of 2003. A reissue patent
followed in 2007. Teva filed suit in 2008. The
claim at issue requires (i) a statin; (ii) at least one
AGCP compound; and (iii) no other stabilizing
compound.
AstraZeneca filed a motion for summary
judgment of invalidity under 102(g)(2). For
purposes of the motion, AstraZeneca stipulated
infringement. The motion was successful, and
Teva appealed to the Federal Circuit. On appeal,
Teva raised four arguments. The first and central
argument was that Section 102(g)(2) required
AstraZeneca to show an appreciation at the time
of invention that crospovidone had a stabilizing
 pat e n t s
effect. In support of this argument, Teva relied
on case law holding that accidental discovery
does not give rise to prior inventive activity for
purposes of Section 102(g)(2) (ref. 7).
Teva’s second argument was that the US
District Court for the Eastern District of
Pennsylvania had implicitly construed the
claim in an overbroad fashion to include a
product containing a plurality of stabilizing
compounds. This argument seems counter-
intuitive, particularly in light of the fact that
neither party disputes that Crestor contains
tribasic calcium phosphate, a stabilizer that is
not an AGCP compound. Closely paralleling
its second argument, Teva’s third argument was
that the district court had improperly applied
inherency precedent from the 102(b) context
without requiring AstraZeneca to prove appre-
ciation. And finally, Teva’s fourth argument was
that, assuming AstraZeneca was in fact first to
invent, AstraZeneca suppressed or concealed
© 2012 Nature America, Inc. All rights reserved.
its invention by failing to disclose that crospo-
vidone stabilized the formulation.
The court dealt with Teva’s arguments in turn.
The court held that to establish prior invention,
the party asserting it must prove that it appreci-
ated what it had made. The prior inventor does
not need to know everything about how or why
the invention worked, or conceive of its inven-
tion in the same words as the patentee would
later use to claim it1. Thus, the court held that
AstraZeneca needed only to appreciate that its
invention was stable and to identify the compo-
nents of the invention.
To arrive at this position, the court cited
Dow Chemical Co. v. Astro-Valcour, Inc.8,
Mycogen Plant Sciences v. Monsanto Co.9 and
Invitrogen Corp. v. Clontech Labs10. In Dow
and Mycogen, the court found prior inventive
npg
activity sufficient for Section 102(g)(2) even
though the prior inventor did not appreciate
certain aspects of the invention. In Dow, the
prior inventor did not appreciate the patent-
ability of his invention, but recognized what
it was and that it was beneficial. In Mycogen,
the prior inventor had also appreciated what
he had made, but described his invention as a
process of modifying the frequency of nucleo-
tides instead of modifying the frequency of
codons. In Invitrogen, which Teva had relied
heavily upon, the court held that prior inven-
tive activity consisting of randomly altering a
panel of mutant genes falls within a category
of accidental duplication cases insufficient for
prior inventive activity under 102(g)(2).
Notwithstanding the differences in out-
come, the court held that “Dow, Mycogen Plant
Sciences, and Invitrogen are consistent applica-
tions of the same rule”1. The court noted that
the distinguishing characteristic to establish
prior invention is that an inventor needs to
832 volume 30 number 9 SEPTEMBER 2012 nature biotechnology
prove appreciation of what it had made, not
why or how it worked. Additionally, the inven-
tor need not understand the invention in the
same terminology a plaintiff subsequently
uses to claim it. Applying this distinction to
Teva’s remaining arguments, the court held
that Teva’s second argument regarding broad
claim construction is tantamount to argu-
ing that AstraZeneca needed to understand
its invention in the same terms used in the
asserted claims. As to Teva’s third argument,
the court noted that this case does not raise
issues of inherency. Rather, AstraZeneca’s con-
cession of infringement established that the
accused product is an embodiment within the
scope of the claims. Finally, the court rejected
Teva’s fourth argument, which presupposed
that AstraZeneca needed to appreciate the sta-
bilizing effect of crospovidone, as that issue had
already been decided to the contrary.
Implications
The court’s holding and reasoning appears to
be in line with precedent, as it is not a new idea
that Section 102(g) requires appreciation of
what was made, but not why it works. In 1972,
Section 102(g)(2) will be
applicable to many cases for
years to come. However, the
Federal Circuit opinion in Teva v.
AstraZeneca provides an early
reference point from which to
compare the changes brought
about by the America Invents Act.
the Court of Customs and Patent Appeals, the
Federal Circuit’s predecessor court, albeit in the
interference context, held that “[a]n inventor
need not understand precisely why his inven-
tion works in order to achieve an actual reduc-
tion to practice”11.
Claim 1 has three main limitations. The
claim requires a statin, a stabilizing effective
amount of at least one AGCP compound,
and “wherein said stabilized pharmaceuti-
cal composition does not contain a stabiliz-
ing effective amount of another stabilizer or
a combination or other stabilizers.”12 Crestor
contains a statin, and thus the first limitation
is met. It also contains crospovidone, which
is an AGCP compound and has a stabilizing
effect. Thus, the second limitation is met.
Finally, it contains tribasic calcium phos-
phate, which is a stabilizing compound, but
is not an AGCP compound. Thus, Crestor
does not meet each and every limitation of
the claim, as the claim includes the negative
limitation. Yet, the court still held that pro-
duction of Crestor by AstraZeneca in 1999
was sufficient inventive activity to invalidate
Teva’s claim.
The language of the opinion provides two
plausible justifications for this outcome. On
one hand, the court stated that “because of
AstraZeneca’s limited concession of infringe-
ment, there is no question that the amount of
crospovidone AstraZeneca’s drug contained
falls within the scope of the asserted claims
as defined by the limitation ‘stabilizing effec-
tive amount’”1. This raises the consideration
that but for AstraZeneca’s concession, the
court would not be bound to accept that
Crestor is an embodiment within the scope
of the claims, and thus perhaps not Section
102(g)(2) prior art.
On the other hand, the court characterized
“the invention” as a broad concept sufficient
to invalidate a narrower claim. That is, the
court considered the invention as a formula-
tion including an AGCP stabilizer, and Teva
simply expressed the invention in a differ-
ent way by claiming an additional negative
limitation. This justification is buttressed by
the court’s extensive citation to the proposi-
tion that a prior inventor need not describe
his invention in the same language as subse-
quently claimed.
The America Invents Act
As noted previously, Section 102(g)(2) will be
applicable to many cases for years to come.
However, the Federal Circuit opinion in Teva v.
AstraZeneca provides an early reference point
from which to compare the changes brought
about by the AIA.
For claims filed after March 16, 2013, the
new Section 102 will apply. There is, of course,
an exception for certain continuing applica-
tions claiming priority before the effective
date of the AIA. Notwithstanding the excep-
tion, the new Section 102 lacks a direct analog
to old 102(g). However, the AIA does provide
a limited substitute to Section 102(g)(2). The
prior user rights defense, formerly limited to
business methods, has been expanded and
reinvigorated with the passage of the AIA. The
“prior commercial use defense,” as it is now
referred to, expands the defense to all catego-
ries of patentable subject matter. Although the
prior commercial use defense in some respects
is similar to Section 102(g)(2), there are some
notable differences.
A major difference is that the prior com-
mercial use defense is not a prior art pro-
vision—successfully asserting it will not
invalidate a patent13. However, although this
is a personal defense, the sale or disposition

of a useful end result by a person entitled to
assert the defense shall exhaust the patent
owner’s rights with respect to the purchaser.
To raise the prior commercial use defense, a
defendant must commercially use the subject
matter that would otherwise infringe a patent
in the United States at least one year before the
earlier of the effective filing date or Section
102(b) public disclosure date of the patent
being asserted. Finally, if the defense is raised
without a reasonable basis, the case will be
made exceptional for the purpose of awarding
attorneys fees14.
Conclusions
Although the central holding that Section
102(g)(2) requires an appreciation of what
was made, but not why or how it works,
the Federal Circuit has taken this line of
precedent one step further. In conjunction
© 2012 Nature America, Inc. All rights reserved.
npg
nature biotechnology volume 30 number 9 SEPTEMBER 2012 833
with precedent that holds, for purposes of
Section 102(g)(2), that a prior inventor need
not describe his prior invention in the same
words as subsequently claimed, the court has
invalidated a patent with Section 102(g)(2)
secret prior art that failed to meet each and
every limitation of the claim. In light of the
plausible justifications for its outcome, this
case’s impact on other provisions of Section
102 remains uncertain, but could prove to
be significant.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
1. Teva v. AstraZeneca. No. 2011–1091 (Fed. Cir., Dec.
1, 2011).
2. H.R. 1249, 112th Cong. § 3 (2011).
3. Chisum, D.S. Priority among competing patent appli-
cants under the American Invents Act, § II.A (2011).
<http://papers.ssrn.com/sol3/papers.cfm?abstract_
id=1969592>
pat e n t s
4. Rich, G.S. Speech to the New York Patent Law
Association (Nov. 6, 1951).
5. See Thomson, S.A. v. Quixote Corp., 166 F.3d 1172,
n. 3 (Fed. Cir. 1999).
6. S. 1948 § 4806 (1999).
7. See Invitrogen Corp. v. Clontech Labs., Inc., 429 F.3d
1052 (Fed. Cir. 2005).
8. Dow Chemical Co. v. Astro-Valcour, Inc., 267 F.3d
1334 (Fed. Cir. 2001).
9. Mycogen Plant Sciences v. Monsanto Co., 243 F.3d
1316 (Fed. Cir. 2001).
10. Invitrogen Corp. v. Clontech Labs., 429 F.3d 1052
(Fed. Cir. 2005).
11. Parker v. Frilette, 462 F.2d 544, 547 (CCPA 1972).
12. US Patent No. RE39,502, Claim 1 (emphasis added)
13. H.R. 1249 at § 5 (amending 35 USC § 273(g) to recite
“[a] patent shall not be deemed to be invalid under
section 102 or 103 solely because a defense is raised
or established under this section.”).
14. H.R. 1249 at § 5 (amending 35 USC § 273(f) to recite
“[i]f the defense under this section is pleaded by a
person who is found to infringe the patent and who
subsequently fails to demonstrate a reasonable basis
for asserting the defense, the court shall find the case
exceptional for the purpose of awarding attorney fees
under section 285.”).
 pat e n t s

Recent patent applications in biosensors

Priority
Patent number Description Assignee Inventor application date Publication date
WO 2012098758 A detection apparatus for a biosensor with National Institute of Akiyama S, 1/20/2011 7/26/2012
an optical prism incident surface and a Advanced Industrial Fujimaki M,
detection plate adhesion surface that are set Science and Technology Nagata K
at a specific angle; useful in, e.g., the medical (Tokyo)
field.
MX 2010012079 A regenerable biosensor for determining the Autonomous University Salas BV, 1/13/2011 7/19/2012
enzyme activity of protease. The biosensor is of Baja California Stoytcheva MS,
provided with a resonator that is provided with (Mexicali, Mexico) Zlatev RK
a quartz microbalance unit, where the quartz
microbalance unit is provided with a coercive
power of 10,000–40,000 units, ensuring a
simple and efficient biosensor.
WO 2012096162 A sensor chip for measuring properties of Panasonic Hashimotodani K, 1/13/2011 7/19/2012
substances, e.g. cells, comprising a diaphragm (Osaka, Japan) Nakano Y,
provided with two different surfaces and Nakatani M, Oka T,
through-hole(s) formed between surfaces, Ushio H, Yamada Y,
where a portion of the surfaces and through- Yamamoto T
hole is covered with a silicon layer.
US 20120178178 A biosensor cartridge, e.g., mass-based sensor, Samsung Electronics Choi YS, Do Jae P, 1/6/2011 7/12/2012
optical sensor, electrical sensor, quartz crystal (Suwon, S. Korea) Han KY, Kim SK,
microbalance, cantilever sensor and surface Lee HJ, Lee JH,
acoustic wave sensor, for automating biosensor Lee JN, Lee SS,
© 2012 Nature America, Inc. All rights reserved.

processes to detect biological material from a Lee YH

biological sample, e.g., blood.
US 20120168306 A biosensor system for performing affinity- University of Texas Hassibi A, Jang B, 11/17/2008 7/5/2012
based detection to detect an analyte in a System (Austin, TX, Manickam A
biological sample. Uses include but are not USA)
limited to toxin, hormone, DNA strand, protein
and bacteria.
JP 2012122744 A method for manufacturing a biosensor Murata Manufacturing Tanigawa M 12/6/2010 6/28/2012
involving adhering an electrode layer and a Co. (Kyoto, Japan)
cover layer on the surface of a spacer layer by
hot melt adhesive.
KR 2012057808 A biosensor useful for detecting gonyautoxin Korea Ocean R&D Jae HL, Lee TK, 11/29/2010 6/7/2012
(GTX) from phytoplankton, comprising anti- Institute (Ansan, Man C
GTX antiserum separated from a combination S. Korea)
of protein and GTX from immunized mouse.
KR 2012057596 A sensibility diagnostic system comprising Yonsei University Hyo IJ, Jung HL 12/30/2011 6/5/2012
a sensibility diagnostic chip equipped with Industrial-Academic
a biosensor, where the sensibility diagnostic Co-operative Foundation
unit converts emotions into an emotional (Seoul)
quotient based on the received signals from
the subject. The system is useful for diagnosis
npg

of sensibility, e.g., stress.

KR 2012055009 A biosensor useful for multiplexed diagnostics, Electronics and Bong PH, Moon YJ, 11/22/2010 5/31/2012
comprising a plane light source, specific Telecommunications Park JW
antibody, detection area, color charge-coupled Research Institute
device camera and long-pass filter. (Daejon, S. Korea)
CN 102477416 A high-salt lipase useful in the printing, Zhang B Fang J, Liu G, Su D, 9/5/2011 5/30/2012
petrochemical and environmental industries, Zhang B
single-cell protein production, cosmetic
production, biological material, biosensor
and biomedical fields. The high-salt lipase is
cost effective and has strong selectivity and
specificity with improved enzyme activity,
stability and practicability.
Source: Thomson Scientific Search Service. The status of each application is slightly different from country to country. For further details, contact Thomson Scientific, 1800 Diagonal
Road, Suite 250, Alexandria, Virginia 22314, USA. Tel: 1 (800) 337-9368 (http://www.thomson.com/scientific).

834 volume 30 number 9 September 2012 nature biotechnology

 news and views

Modulating WNT receptor turnover for tissue repair

Arie Abo & Hans Clevers
New mechanistic insight into the regulation of WNT signaling supports efforts to target this pathway in adult stem
cells for regenerative medicine.

Experimental cell therapies often involve cell a b

transplantation, but for some conditions it may
be advantageous to follow a simpler strategy of
Enterocytes
enhancing tissue repair by coaxing endo­genous RSpo
RSpo
stem cells to proliferate and differentiate. This WNT
© 2012 Nature America, Inc. All rights reserved.

approach has shown promise in animal models

Internalization Ubiquitnation
of disease, where potentiation of the Wnt
pathway in stem cells by the RSpondin (RSpo)
family of secreted proteins stimulates tissue TA
compartment LGR5 ZNRF3 LRP5/6 Frizzled
regeneration and healing. In a recent study in
Nature, Hao et al.1 reveal important new infor-
mation about the mechanism by which RSpo Membrane clearance Degradation Activation
BMI1+
proteins boost WNT signaling. They show that +4 position
activation of the WNT receptor is inhibited by Paneth cell

its interaction with ZNRF3, a transmembrane LGR5+

crypt stem cell
E3 ubiquitin ligase. RSpo proteins effectively
sequester ZNRF3 by inducing dimerization of
ZNRF3 with the LGR4 receptor, thereby activat- Figure 1 Modulating LGR5+ crypt stem cells with RSpo proteins for regeneration of the gastrointestinal
ing WNT signaling (Fig. 1). Although Hao et epithelium. (a) The architecture of the crypt. LGR5+ cells (yellow) reside between the Paneth cells
(green) at the bottom of the crypt, and BMI1+ cells (red) reside at position +4. LGR5+ cells are WNT
al.1 worked with HEK293 cells rather than with
dependent and can give rise to all cell lineages of the gastrointestinal epithelium. (b) Molecular
stem cells, expression of LGR receptors in adults mechanism by which RSpo proteins activate LGR5+ stem cells by activating WNT. In the absence of
appears to be restricted to stem cells. Thus, their RSpo proteins, ZNRF3 ubiquitinates the WNT receptor Frizzled, targeting it for degradation. In the
study has broad implications for regenerative
npg

medicine as it identifies a new RSpo-protein tar-
get that may be expressed in adult stem cells and
further validates the importance of WNT com-
plex turnover for therapeutic intervention.
The Wnt signaling pathway regulates the
expression of genes that are critical for embryonic
development and tissue homeostasis. However,
the possibility of exploiting this pathway for
regenerative medicine has been hampered by
difficulties in producing recombinant WNTs in
suf­ficient quantities for systemic administration.
In one recent study using local administration,
Arie Abo is at the California Institute for
Regenerative Medicine, San Francisco,
California, USA, and Hans Clevers is at the
Hubrecht Institute, Developmental Biology
and Stem Cell Research, Royal Netherlands
Academy of Arts and Sciences Utrecht, 3584CT
Utrecht, The Netherlands.
e-mail: aabo@cirm.ca.gov or
h.clevers@hubrecht.eu
presence of RSpo proteins, ZNRF3 dimerizes with LGR5 receptor, leading to removal of ZNRF3 from
the cell surface, accumulation of WNT receptor complex on the membrane and enhancement of WNT
signaling. TA, transit-amplifying cells.
Wnt3a delivered in liposomes was shown to
stimulate skeletal mesenchymal stem cells and
promote bone repair2. In another strategy, acti-
vation of Wnt signaling with antibodies against
the Wnt pathway molecule Dkk1 was used to
enhance bone formation in a mouse model of
rheumatoid arthritis3. RSpo proteins were the
first Wnt signaling modulators to be synthesized
in relatively large amounts and administered
systemically to animals (A.A. and colleagues)4–6.
This work demonstrated the regenerative poten-
tial of RSpo proteins in models of rheumatoid
arthritis, oral mucositis, and experimental colitis
and showed that tissue regeneration is a direct
result of potentiating the Wnt pathway in stem
cells, which triggers expansion and regeneration
of the epithelial layer and bone.
The missing molecular link between RSpo
proteins and stem cells emerged with our
recent studies implicating the stem cell markers
Lgr4 and Lgr5 in RSpo protein–mediated Wnt
signaling7. Lgr4 and Lgr5 are homologous
orphan G-protein-coupled receptors that have
been identified as specific markers of adult
stem cells. In the gastrointestinal tract, they
are expressed only on stem cells, which reside
among the Paneth cells at the bottom of the
crypt (H.C. and colleagues)8 (Fig. 1). Crypt
stem cells play a critical role in homeostasis and
regeneration of the gastrointestinal epithelial
lining. Moreover, single crypt stem cells isolated
from the mouse intestine can reconstitute the
crypt-villus structure ex vivo when appropriate
growth factors are provided7. If RSpo proteins
are added to cultured mouse Lgr5+ cells, the
cells form large crypt-villus organoids7, simi-
lar to intestinal crypt and villi structures, that
contain fully polarized enterocytes with mature

nature biotechnology volume 30 number 9 september 2012 835

 news and v iews
brush borders, Goblet cells and enteroendocrine
cells. All RSpo family members (RSpo1–4) bind
with high affinity to LGR4 and LGR5 receptors
and thereby potentiate WNT signaling (H.C.
and colleagues)9,10. However, the mechanisms
underlying this activity of RSpo proteins have
not been well understood.
The WNT receptor complex is a heterodimer
of the LRP6 and Frizzled receptors. RSpo pro-
teins were previously shown to activate WNT
signaling by blocking LRP6 internalization
through a mechanism that involves inhibition of
the interaction of DKK1 and Kremen (A.A. and
colleagues)11. Dkk1 inhibits Wnt signaling by
binding to and inducing dimerization of Lrp6
with the transmembrane proteins Kremen1 and
Kremen2, followed by rapid endocytosis of Lrp6
from the plasma membrane12. The new work by
Hao et al.1 reveals a second, similar mechanism
for regulating WNT receptor complex turnover,
which operates not through DKK1/Kremen but
© 2012 Nature America, Inc. All rights reserved.
through the ZNRF3 axis (Fig. 1).
The authors began by carrying out a genomic
analysis to discover genes that are upregulated
in primary colorectal tumors exhibiting hyper-
activity of the WNT pathway, which identified
the candidate genes ZNRF3 and RNF43. These
genes were found to be functional homologs
of each other that negatively regulate WNT
signaling. Knocking down ZNRF3 in HEK293
cells led to higher cell-surface levels of the
LRP6 and Frizzled receptors, whereas over­
expressing ZNRF3 promoted ubiquitinylation
of the Frizzled receptor and its subsequent
degradation. Experiments with RSpo proteins
showed that RSpo 2, 3 and 4 increased cell-
surface levels of the LRP6 and Frizzled
receptors and that RSpo1 decreased ubiquit-
inylation of the Frizzled4 receptor.
npg
Hao et al.1 also investigated the roles of LGR4
and LGR5. RSpo protein–mediated turnover
of Frizzled receptors required LGR receptors.
Moreover, ubiquitinylation and cell-surface
expression of Frizzled receptors 4, 5 and 8 were
abolished when LGR4 receptor was knocked
down with short interfering RNA or when the
cells were treated with RSpo proteins. In addi-
tional experiments to unravel the links between
LGR receptors, RSpo proteins and ZNRF3, RSpo
proteins were found to mediate interaction of
LGR4 and ZNFR3 and to induce membrane
clearance of ZNRF3.  Furthermore, dimeriza-
tion of LGR4 and ZNRF3 either artificially or
by RSpo proteins led to endocytosis of ZNRF3
and accumulation of WNT receptor complex
on the cell surface.
Enhancing WNT signaling to promote stem
cell proliferation and differentiation is emerg-
ing as a promising strategy to promote regen-
eration of a variety of tissues. The results of
Hao et al.1 further validate this approach and
836 volume 30 number 9 september 2012 nature biotechnology
identify ZNRF3 as a potential target for thera-
peutic intervention.  Although the authors did
not study stem cells, they demonstrated the role
of ZNRF3 in regulating the WNT pathway not
simply in cultured cells but also by knockdown
experiments in zebrafish and mice, providing
important in vivo validation.
Whichever WNT-signaling targets and drugs
ultimately prove most effective, enhancing tis-
sue repair by activating endogenous stem cells
is likely to be beneficial for a diverse group of
disorders. In the gastrointestinal tract, several
diseases are associated with injuries of the epi-
thelial lining, including inflammatory bowel dis-
ease (comprising Crohn’s disease and ulcerative
colitis), which involves destruction of mucosal
integrity and chronic exposure to gut microbial
flora.  Current therapies for inflammatory bowel
disease all work through an anti-inflammatory
mechanism using agents such as nonsteroidal
anti-inflammatory drugs or biologics that target
the pro-inflammatory cytokine tumor necrosis
factor alpha. However, additional therapies are
clearly needed to treat primary and secondary,
nonresponsive patients. Although the role of
stem cells in gastrointestinal disease has not yet
been elucidated, targeting epithelial stem cells to
accelerate mucosal repair should be beneficial
in some cases. So far, the function of Lgr5+ stem
cells has been studied mostly in stomach, small
intestine, colon and hair follicle, but this marker
has also been detected in mammary gland,
brain and, after tissue damage, in pancreas and
liver.  The roles of these cells in tissue regenera-
tion is the subject of ongoing research (H.C.).
In developing strategies for translating WNT
signaling biology to regenerative medicine, we
must not overlook the potential risks associated
with overactivation of this pathway. WNT
RNA-mediated programmable DNA
cleavage
Rodolphe Barrangou
Genome engineering has a new tool—endonucleases involved in bacterial
adaptive immunity that can be reprogrammed with customizable small,
noncoding RNAs.
Genome engineering methods rely on site-
specific endonucleases that trigger sequence
modification by DNA-repair systems at the
Rodolphe Barrangou is at DuPont Nutrition
and Health, Madison, Wisconsin, USA.
e-mail: rodolphe.barrangou@dupont.com
activation has been reported in numerous can-
cers, including breast and colon cancer. One
of the first mutations in colon cancer is in the
tumor suppressor adenopolyposis coli gene, a
mutation that leads to constitutive activation
of the WNT pathway, and tumorigenesis may
be accelerated by chronic inflammation in the
colon.  Therefore, in inflammatory bowel dis-
ease patients, treatment with agents such as
RSpo proteins may reduce the risk of tumor­
igenesis by repairing damaged mucosa and
thereby  minimizing exposure to inflamma-
tory reactive oxidants5. More generally, WNT
signaling modulators such as RSpo proteins
may be safer than approaches that use WNT
ligands because they potentiate the regenera-
tion of endogenous damaged tissue without
violating the order of events essential for tis-
sue restoration. Thus, WNT modulators could
be used to potentiate WNT-dependent repair
in specific tissues rather than throughout the
body, avoiding potential side effects caused by
growth factors such as WNT ligands.
COMPETING financial INTERESTS
The authors declare no competing financial interests.
1. Hao, H.X. et al. Nature 485, 195–200 (2012).
2. Minear, S. et al. Sci. Transl. Med. 2, 29–30 (2010).
3. Diarra, D. et al. Nat. Med. 13, 156–163 (2007).
4. Zhao, J. et al. Gastroenterology 132, 1331–1343
(2007).
5. Zhao, J. et al. Proc. Natl. Acad. Sci. USA 106,
2331–2336 (2009).
6. Krönke, G. et al. Arthritis Rheum. 62, 2303–2312
(2010).
7. Sato, T. et al. Nature 459, 262–265 (2009).
8. Barker, N. et al. Nature 449, 1003–1007 (2007).
9. Carmon, K.S. et al. Proc. Natl. Acad. Sci. USA 108,
11452–11457 (2011).
10. de Lau, W. et al. Nature 476, 293–297 (2011).
11. Binnerts, M.E. et al. Proc. Natl. Acad. Sci. USA 104,
14700–14705 (2007).
12. Mao, B. et al. Nature 417, 664–667 (2002).
cleavage site. Enzymes that are currently
widely used for genome engineering include
zinc finger nucleases (ZFNs) and transcription
activator-like effector nucleases (TALENs),
both of which can be designed to target spe-
cific genome sequences. These proteins have
to be engineered for each application to
 news and v iews

a Cas9 genes (cas) constitute the CRISPR-Cas system.

RuvC
CRISPR-Cas provides RNA-mediated adap-
tive immunity against viruses and plasmids in
bacteria and archaea2. Immunity is acquired
Target DNA PAM through the integration of invasive genetic ele-
ment sequences as ‘spacers’ into the CRISPR
crRNA locus of the host. Arrays of CRISPR spacers are
HNH transcribed and processed into small interfer-
tracrRNA
ing CRISPR RNAs (crRNAs) that guide Cas
proteins to mediate cleavage of homologous
Cas9 i dsDNA sequences2–4.
b RuvC Among the three types of CRISPR-Cas
RuvC+
HNH– systems that have been identified based on
cas content and sequences5, type I- and type
Target DNA PAM ii III-mediated immunity rely on a large and
multiprotein Cas complex, whereas type II
RNA chimera systems solely (and conveniently) rely on a
RuvC–
HNH+ single Cas protein, Cas9. Although the target
HNH
sequence is defined by the CRISPR spacer,
iii
RuvC+ a strict requirement in many CRISPR-Cas
HNH+ systems is the presence of a proto-spacer–
© 2012 Nature America, Inc. All rights reserved.

c associated motif (PAM) sequence, typically

2–4 nucleotides (nt), located in the immedi-
FokI ate vicinity of the spacer target. Most cur-
rent applications of CRISPR-Cas systems
Target DNA focus on increasing the breadth of phage
resistance in industrial bacteria, leveraging
FokI CRISPR spacer hypervariability for genotyp-
TAL-effector
DNA binding ing and epidemiological surveys and exploit-
domain ing iterative spacer additions for analysis of
host-virus dynamics6. The ability of func-
tional CRISPR-Cas systems to be trans-
ferred across even distant bacterial genera
d and species had previously set the stage for
FokI
ZFPs bind DNA
heterologous applications7.
Jinek et al.1 show that the Cas9-crRNA
Target DNA ribonucleoprotein complex mediates specific
FokI
DNA cleavage for CRISPR-encoded immunity
in bacteria. A trans-activating CRISPR RNA
npg

Figure 1 crRNA-guided DNA cleavage by Cas9. (a) The Cas9 protein (light blue) combines with (crRNA), termed tracrRNA, pairs with another
crRNA (red) and tracrRNA (orange) to form a ribonucleoprotein interfering complex. The crRNA distinct crRNA that guides Cas9 to homo­l­
sequence guides the interference complex to a complementary sequence in the target DNA (dark ogous target sequences (Fig. 1a). In another
blue). Once an R-loop is formed (see open structure), the HNH and RuvC Cas9 domains nick the
series of experiments, the authors demon-
complementary and noncomplementary strands, respectively, 3 nt away from the PAM sequence.
(b) Reprogramming of DNA cleavage by an RNA chimera molecule that includes partial tracrRNA linked strate that the endonuclease HNH and RuvC
to a target-specific customized crRNA (purple). Single-stranded DNA nicking can be achieved by mutation (a Holliday junction resolvase) domains of
of either the HNH motif (i) or the RuvC motif (ii), whereas double-stranded cleavage can be achieved Cas9 nick the complementary and noncomple-
using the two wild-type domains (iii). (c) TALEN DNA cleavage. TAL effector DNA binding domains on the mentary DNA strands, respectively, generating
left and on the right recognize the left and right target sequence, respectively, and trigger dsDNA cleavage a dsDNA blunt cleavage. The ability to nick or
through the FokI nuclease domain. (d) ZFN DNA cleavage. Zinc finger proteins (ZFPs) recognize target
cleave DNA at defined and discrete locations in
DNA sequences and trigger dsDNA cleavage through the FokI nuclease domain.
a DNA sequence opens new avenues for edit-
ing and assembling DNA sequences in vitro.
reprogram the cleavage site to specifically tar- synthetic RNA-mediated reprogramming of In an elegant proof-of-concept experiment, the
get the DNA sequence of interest. In a recent cleavage specificity. Several immediate applica- authors demonstrate that Cas9 DNA cleavage
paper in Science, Jinek et al.1 find that two RNA tions of this approach can be envisioned, nota- specificity (minimally defined by 13-nt perfect
molecules direct DNA cleavage in a ribonucleo- bly customized single-strand DNA nicking and base-pairing between crRNA and target DNA)
protein complex that forms part of the adaptive double-stranded (ds)DNA cleavage for genome can be reprogrammed to target a green fluores-
immune system of bacteria, and that a single engineering and editing. cent protein gene using a chimeric RNA fusion
RNA chimera can be engineered to repro- The molecular machinery in the Jinek et al.1 of the crRNA 3′ end with the tracrRNA 5′ end
gram sequence specificity. This synthetic tour study is part of the clustered regularly inter- (Fig. 1b). Arguably, this combines the flex-
de force introduces a new endonuclease family spaced short palindromic repeats (CRISPR), ibility of RNA interference with the power of
to the genome engineering arsenal that enables which, together with CRISPR-associated DNA restriction enzymes to generate single- or

nature biotechnology volume 30 number 9 september 2012 837

 news and v iews
double-stranded DNA breaks in any sequence
for genome targeting or editing.
Several studies have shown that ZFNs
and TALENs (Fig. 1c,d) can be engineered
to induce mutations at specific locations in
genome sequences8,9 in plants (Zea mays),
bacteria (Escherichia coli), model animals
(Drosophila melanogaster), human embryonic
stem cell lines and induced pluripotent stem
cell lines. However, these proteins have to be
redesigned for every different DNA sequence
targeted. The 3-nt recognition motif of ZFNs
can be promiscuous and generate off-target
cleavage, and target frequency has been esti-
mated at ~1 in every 500 bp8. The ability to
modulate TALEN specificity by modification
of the repeat variable di-residues within their
amino acid sequence has been an advantage
of TALENS compared with ZFNs9. However,
the nonspecific nuclease domain of FokI in
TALENs can generate nonspecific and non­
© 2012 Nature America, Inc. All rights reserved.
targeted cleavage. Also, the sole product is
strictly dsDNA cleavage, and target frequencies
have been estimated at ~1 in every 35 bp8.
Chimeric RNA-Cas9 systems could provide
several advantages over ZFNs and TALENs,
including sequence-recognition specificity (by
customized crRNA sequences), PAM dinucleo­
tides that can be frequent (GG in the case of
Jinek et al.1), the convenience of redesign-
ing nucleotides in nucleic acid sequences as
opposed to modifying amino acids in protein
sequences, and the ability to nick either or both
DNA strands. The latter provides a molecular
basis for precise genome editing by a program-
mable DNA nicking enzyme10. Because Cas9
has two distinct domains (RuvC and HNH),
which each nick a specific DNA strand, wild-
type Cas9 and functional domain mutants
npg
can generate either dsDNA breaks or single-
stranded DNA nicks, as desired (Fig. 1b).
Single-stranded breaks are repaired by error-
free homologous recombination at the nicking
site, producing a precise mutation, as opposed
to error-prone, nonhomologous, end-joining
repair for dsDNA cleavage, which frequently
creates insertions and deletions after ZFN or
TALEN cleavage.
Molecular analysis of Cas9 cleavage has
shown that both linearized and super-
coiled DNA can be cleaved, suggesting that
multiple distinct chimeric RNAs can be con-
currently or sequentially used to process DNA
at multiple target sites using a single enzyme,
opening avenues for genome stacking and
shuffling. Moreover, specific cleavage sites can
be engineered to generate DNA molecules that
have custom extremities for iterative genome
build-up. Accordingly, crRNA-Cas–directed
nicking and cleavage set the stage for more
precise DNA surgery and genome editing.
838 volume 30 number 9 september 2012 nature biotechnology
Although immediate applications of this
new tool include customized DNA nicking
and/or cleavage in bacteria, there are intriguing
possibilities for genome editing and genome
engineering of eukaryotes. This will require
testing whether crRNA-Cas systems can effi-
ciently cleave chromatin DNA in vivo and be
readily transferred into organisms of inter-
est, notably yeast and fungi, but also plants,
for crop and agricultural applications, and
human cells, for medical purposes. Only
the future will tell whether this program-
mable molecular scalpel can outcompete
ZFN and TALEN DNA scissors for precise
genomic surgery.
Silicon dreams of cells into
symbols
Jeremy Gunawardena
Diverse mathematical models combine to create a comprehensive whole-cell
computational model of a human pathogen.
Writing in Cell, a team from Stanford
University and the J. Craig Venter Institute
has reported a whole-cell simulation of
Mycoplasma genitalium, a urogenital parasite
adored by synthetic biologists for its reduced
genome, a mere snip at 525 genes. This auda-
cious accomplishment by Karr et al.1 of what
has been called “a grand challenge of the 21st
century”2 goes beyond previous attempts3 in its
comprehensiveness and especially in capturing
the elusive and subtle chicken-and-egg recur-
sion through which a cell creates itself.
Models of living processes have a long
history in science and fiction. From Jacques de
Vaucason’s automata to The Terminator, our
creations are rooted in a yearning to control the
uncontrollable. In the afterglow of cybernetics,
Arthur Guyton built a model of the systemic
blood circulation4 that still inspires awe, if
not actual use. Physiologists and bioengineers
have long been assembling a Virtual Heart5
and have even been aspiring to a Virtual Rat
(http://www.systemscenters.org/centers/
virtual-physiological-rat/). Industrial appli-
cations include the computational platform
PhysioLab, developed by the biotech company
Jeremy Gunawardena is in the Department
of Systems Biology, Harvard Medical School,
Boston, Massachusetts, USA.
e-mail: jeremy@hms.harvard.edu
COMPETING FINANCIAL INTERESTS
The author declares competing financial interests:
details are available in the online version of the paper.
1. Jinek, M. et al. Science advance online publication,
doi:10.1126/science.1225829 (28 June 2012).
2. Wiedenheft, B., Sternberg, S.H. & Doudna, J.A. Nature
482, 331–338 (2012).
3. Deltcheva, E. et al. Nature 471, 602–607 (2011).
4. Garneau, J.E. et al. Nature 468, 67–71 (2010).
5. Makarova, K.S. et al. Nat. Rev. Microbiol. 9, 467–477
(2011).
6. Andersson, A.F. & Banfield, J.F. Science 320,
1047–1050 (2008).
7. Sapranauskas, R. et al. Nucleic Acids Res. 39,
9275–9282 (2011).
8. Cermak, T. et al. Nucleic Acids Res. 39, 7879–7889
(2011).
9. Miller, J.C. et al. Nat. Biotechnol. 29, 143–148 (2011).
10. Kim, E. et al. Genome Res. 22, 1327–1333 (2012).
Entelos, which models dynamic interactions
among organ systems in diseases like diabetes6.
The origins of the M. genitalium simulation
lie in the constraint-based models of whole-cell
metabolism developed by Bernhard Palsson7.
Such a constraint-based model implements one
of the 28 processes into which Karr et al.1 have
partitioned the cell’s operations (Fig. 1). These
include processes that track exchanges with
the extracellular medium and all the metabolic
fluxes, the state of the supercoiled chromosome,
transcription of all active genes, processing of all
mRNAs, translation of all proteins, formation of
all macromolecular complexes including RNA
polymerases and ribosomes, and progress of
cytokinesis and FtsZ polymerization. It would
be a nightmare to tie all of these processes
together without a beautiful idea: between
timesteps of one second, the processes operate
independently of each other. What links them
are 16 variables that record the overall cellular
state (e.g., counts of all mRNAs, protein mono-
mers, protein complexes). The variables are read
by the processes at the start of each timestep
and written back at the end, in a loop that
repeats until the cell divides. This takes slightly
longer than M. genitalium itself does, 10 h
on a 128-node cluster that runs the publicly
available, object-oriented Matlab code.
The one-second time-scale separation
enables two vital features. First, it allows pain-
less implementation of the subtle recursion
 news and v iews

16 Cell variables Mathematical models 28 Cell processes

Translation

RNA DNA
Chromosome Replication
Condensation
Probabilistic binding
Supercoiling
Transcript Replication initiation
RNA Transcriptional regulation
Markov model /
Transcription
Polypeptide probabilistic binding
Segregation
Protein monomer Terminal organelle
Boolean
Protein activation
Protein

Complex
Host interaction
RNA pol Aminoacylation
Start Ribosome Mass action
Complexation
Protein folding Cell Yes Finish
FtsZ ring Ribosome assembly division?
Protein modification
Metabolite

Metabolic reaction Protein translocation

Protein processing I and II
Metabolite RNA modification
No
Stochastic Poisson
RNA decay
process
Geometry Protein decay
DNA damage
Host DNA repair
Other

RNA processing
Mass
© 2012 Nature America, Inc. All rights reserved.

Stimulus ODE FtsZ polymerization

Constraint-based models Metabolism

Time
Geometric analysis Cytokinesis

Repeat

Figure 1 Whole-cell simulation of M. genitalium. The functionality of the cell was divided into 28 processes (right column). Each process was independently
modeled using the most appropriate mathematical representation (middle column). The background color of each process matches the background color of
the mathematical representation used to model it. The simulation is broken down into one-second timesteps. At the start of each timestep, the processes read
the values of the 16 cell variables (left column), carry out their computations independently of each other and then update the cell variables at the end of the
timestep. This procedure integrates the different processes, which influence each other only through the cell variables, and is carried out repeatedly until the
simulation terminates at cell division, as measured by the progress of FtsZ polymerization. RNA pol, RNA polymerase; ODE, ordinary differential equation.

mentioned above, by which proteins make
the ingredients out of which they themselves
are made. Robert Rosen struggled with the
profound challenges of such “metabolic
closure”8,9, and successful implementation
npg
the complexity, all that can be deduced from a
model is implicit in its assumptions. Implicit
does not mean obvious: the assumptions
underlying the whole numbers, 1, 2, 3,…,
have kept mathematicians on their toes since
Because models are only as good as their
assumptions, they can fall short of expectations.
Constraint-based models are sometimes touted
as requiring no assumptions beyond reaction
stoichiometries. Such models do well at pre-

of it distinguishes this model from any other.
Second, it allows each of the 28 processes to
be independently implemented by the most
appropriate method (e.g., constraint-based
models, stochastic processes, discrete models,
computer code, ordinary differential equa-
tions) (Fig. 1). The right horse can be chosen
for the right course. The 1,900 resulting param-
eters were estimated using experimental data
garnered from a variety of organisms, from
Mycoplasma pneumoniae to Escherichia coli.
Karr et al.1 had to wade through over 900 pub-
lications in the process. The model is neither
simple nor elegant but it is effective. Knockout
of individual genes in silico identifies essential
genes with 80% accuracy compared to data
from M. genitalium.
Although such accuracy is encouraging,
what we always demand of a model is emergent
novelty. What we often forget is that a model
is not a description of reality; it is a descrip-
tion of our assumptions about reality10. Despite
Diophantus’s Arithmetica. The assumptions
of the M. genitalium model are not so pro-
found. However, they have not been brought
together in this way before. The novelty arises
from looking inwards, from learning whether
we really understand what we think we know
about how metabolism, transcription, transla-
tion and replication recursively create a new
cell. Indeed, the model yields a very interesting
prediction of this kind. Simulations show that
replication initiation, and replication itself, vary
substantially in duration between stochastically
different cells, whereas cell cycle durations are
tightly clustered around nine hours. According
to the model, cells that take longer to initiate
replication accumulate a greater pool of the
dNTPs needed to make DNA so they replicate
faster, making up for lost time. Metabolism
coordinates the cell cycle, independently of
genetic regulation. It is a beguiling prediction,
but whether artifact or real biology remains to
be tested.
dicting exchange fluxes with the environment
but, as for the internal fluxes, they may not get
even the directions correct. If thermodynam-
ics is not put in, it does not magically emerge.
Similarly, the M. genitalium model cannot pro-
vide insights into lysine acetylation11 because
that is not part of its program. Models cannot
supplant the kind of experiments undertaken
by Luis Serrano and colleagues, whose compre-
hensive unraveling of the metabolome12, tran-
scriptome13 and proteome14 of M. pneumoniae
revealed how noncoding RNAs and multifunc-
tional enzymes can compensate for a reduced
genome. The expectation that, with enough
details, a model will miraculously spring to
life and make such experiments unnecessary
is the stuff of fiction.
This is not to say that models cannot tell us
about things we do not yet know exist. Physicists
have spent the gross domestic product of a
small country confirming the physical exis-
tence of something conjured up conceptually

nature biotechnology volume 30 number 9 september 2012 839

 news and v iews
in Peter Higgs’ 1964 theoretical paper. Much
the same conjuring has happened in biology.
Michaelis and Menten conjured up enzyme-
substrate complexes more than 30 years before
Britton Chance showed they existed, using a
simple model now familiar to all biochemists10.
Mendel conjured up the discrete particles later
called genes using high school–level algebra.
These kinds of models trade detail for abstrac-
tion; they focus on a particular question to the
exclusion of all else. They require more inspira-
tion, less perspiration.
Karr et al.1 already have their sights set on
E. coli, a more experimentally tractable organ-
ism of interest to a broad range of biologists.
With 4,288 genes, a division time of 30 min
and far more copious data, computational
power may become the limiting resource.
One wonders whether specialized hardware,
like that which David Shaw and colleagues
have exploited so spectacularly in molecu-
© 2012 Nature America, Inc. All rights reserved.
lar dynamics15, will eventually be required,
npg
840 volume 30 number 9 september 2012 nature biotechnology
if whole-cell simulation is to aspire to
mammalian complexity.
Having pulled off a tour-de-force of com-
putational biology, it is a shame that Karr
et al.1 succumb to genocentrism in their title.
The model no more “predicts phenotype from
geno­type” than DNA makes RNA makes pro-
tein. DNA does not make anything. For that you
need a cell. How strange it is that cell theory,
the first true theory in biology, predating both
Darwin’s evolution and Mendel’s genetics, is so
readily ignored in our current fixation with the
genome, only one of the cell’s many components.
The irony is that the paper itself is profoundly
antigenocentric; metabolic closure could hardly
be otherwise8. Moreover, the real paper—the
supplementary information, all 120 pages
of it—is modern cell bio­logy compiled into
instructions, symbols and formulas. The authors
have published a new kind of molecular and cell
biology textbook—one that is executable. Our
students will want to download the app.
COMPETING FINANCIAL INTERESTS
The author declares no competing financial interests.
1. Karr, J.R. et al. Cell 150, 389–401 (2012).
2. Tomita, M. Trends Biotechnol. 19, 205–210
(2001).
3. Ishii, N., Robert, M., Nakayama, Y., Kanai, A. &
Tomita, M. J. Biotechnol. 113, 281–294 (2004).
4. Guyton, A.C., Coleman, T.G. & Granger, H.J. Annu. Rev.
Physiol. 34, 13–44 (1972).
5. Noble, D. Physiology (Bethesda) 19, 191–197 (2004).
6. Shoda, L. et al. Clin. Exp. Immunol. 161, 250–267
(2010).
7. Edwards, J.S., Covert, M. & Palsson, B.O. Environ.
Microbiol. 4, 133–140 (2002).
8. Rosen, R. Life Itself: a Comprehensive Enquiry into
the Nature, Origin and Fabrication of Life. Complexity
in Ecological Systems (Columbia University Press,
2005).
9. Letelier, J.-C. & Soto-Andrade, J. J. Theor. Biol. 238,
949–961 (2006).
10. Gunawardena, J. Mol. Biol. Cell 23, 517–519
(2012).
11. van Noort, V. et al. Mol. Syst. Biol. 8, 571 (2012).
12. Yus, E. et al. Science 326, 1263–1268 (2009).
13. Güell, M. et al. Science 326, 1268–1271 (2009).
14. Kühner, S. et al. Science 326, 1235–1240 (2009).
15. Dror, R.O., Dirks, R.M., Grossman, J.P., Xu, H. &
Shaw, D.E. Annu. Rev. Biophys. 41, 429–452 (2012).
 perspective
Navigating cancer network attractors for tumor-
specific therapy
Pau Creixell1, Erwin M Schoof1, Janine T Erler2 & Rune Linding1
Cells employ highly dynamic signaling networks to drive
biological decision processes. Perturbations to these signaling
networks may attract cells to new malignant signaling and
phenotypic states, termed cancer network attractors, that
result in cancer development. As different cancer cells reach
© 2012 Nature America, Inc. All rights reserved.
these malignant states by accumulating different molecular
alterations, uncovering these mechanisms represents a grand
challenge in cancer biology. Addressing this challenge will
require new systems-based strategies that capture the intrinsic
properties of cancer signaling networks and provide deeper
understanding of the processes by which genetic lesions
perturb these networks and lead to disease phenotypes.
Network biology will help circumvent fundamental obstacles
in cancer treatment, such as drug resistance and metastasis,
empowering personalized and tumor-specific cancer therapies.
Cells are constantly computing decisions based on the integration of
different cues that reach them at various times. In contrast to single-
cell organisms, in multicellular organisms, cellular decisions should,
ultimately, benefit the organism as a whole, even if that implies that an
individual cell will have to decide to commit suicide. In line with this
unique feature, signaling networks have evolved during multicellular
evolution to allow cells to integrate cues and make decisions that ensure
npg
cooperative behavior between them. By hijacking these mechanisms,
cancer cells escape cooperative rules and transition from a game gov-
erned by Nash equilibria1,2 between all cells into a new scenario where
cancer cells decide their behavior purely based on their own benefit,
or as phrased by Hanahan and Weinberg3, “become masters of their
own destinies.” Given the central role played by signaling networks in
the integration of cues to compute any cellular responses, we argue that
cancer is not simply a disease with a genetic basis, but is one ultimately
driven by perturbations at the signaling network level, and that both the
‘cue-signal-response’ rules of cellular decision-making and the switch
in strategy from cooperative to selfish are major, hitherto understudied,
hallmarks of cancer3,4.
In this article, we dissect the strategies cancer cells use to become
‘selfish’ and drive disease. We first review how genetic lesions can lead to
altered protein function, which can result in changes to the structure and
1Cellular Signal Integration Group (C-SIG), Center for Biological Sequence
Analysis (CBS), Department of Systems Biology, Technical University of Denmark
(DTU), Lyngby, Denmark. 2Biotech Research & Innovation Centre (BRIC),
University of Copenhagen, Copenhagen, Denmark. Correspondence should be
addressed to J.T.E. (janine.erler@bric.ku.dk) or R.L. (linding@cbs.dtu.dk).
Published online 10 September 2012; doi:10.1038/nbt.2345
842 volume 30 number 9 SEPTEMBER 2012 nature biotechnology
dynamics of signaling networks and ultimately cellular phenotype. Next,
we describe five general properties of cancer signaling networks (Fig. 1)
and define five challenges in cancer network biology and propose
strategies to overcome them (Fig. 2). By meeting these challenges,
network biology may fundamentally advance not only basic biology
but also patient treatment. Finally, we describe how a combination
of relatively new technologies could become a potent cocktail for the
discovery of network drugs, and we discuss the practical implementation
of personalized and tumor-specific cancer therapy.
From genomic lesions to functional network perturbations
Tumor cells often harbor hundreds to thousands of genetic lesions. But
based on the observation that some of these genetic lesions are repeat-
edly observed in several cancers (e.g., BRAF V600E, present in >50%
of all malignant melanomas5), it has been hypothesized that only a few
genetic lesions are causally implicated in cancer development (‘drivers’),
whereas the majority have no functional consequences (‘passengers’)6.
Although this classification has had some use in identifying
mutations that are highly prevalent, it is now apparent that a tumor
is not, under any circumstances, a static and uniform population of
malignant cells. Rather, it is a dynamic ensemble of subpopulations
with different abnormalities undergoing molecular evolution7–9.
Two fundamental principles of cancer signaling networks can explain
why a binary driver/passenger classification may be too simplistic to
accommodate the complex dynamic nature of tumors. First, different
tumors can develop similar phenotypes by acquiring mutations in
different proteins10, in what we term analogous mutations (Fig. 1a).
Second, it has been shown that two different mutations not capable
of causally driving cancer by themselves are able to do so when they
appear in combination within the same cells or even within two
neighboring cells11, in what could be described as two passengers
becoming drivers or, as we refer to them, synthetic oncogenes (Fig. 1b).
Thus, patient-to-patient heterogeneity can be driven by the presence
of different mutations in the same or in different proteins that lead to a
similar signaling state and phenotypic outcome.
Altogether, the intrinsic heterogeneity of tumors makes it a pressing
challenge for cancer network biologists to develop tools to identify
the extent to which combinations of cancer mutations affect protein
function and cellular and phenotypic states (Fig. 2a,b). Even though
several such tools have been developed (reviewed in ref. 12), existing
methods are mainly based on protein structure and/or sequence
conservation. This is at odds with recent findings that show that cancer
mutations tend not to cluster on the most conserved protein regions.
In kinases, for example, mutations typically hit the kinase activation
segment, a functional, yet largely nonconserved protein region13.
 perspecti v e

An insightful example of how to explore

b this sequence-function relationship in protein
domains was carried out by researchers in
the Ranganathan and Yaffe laboratories who,
using methods from statistical mechanics,
a Mutation in A =
healthy cell
Mutation in B =
healthy cell
generated synthetic WW domains de novo
that maintained fold and function17,18.
Synthetic
oncogenes Further supporting a complex sequence-
function relationship, additional studies from
Different mutations but the Ranganathan laboratory demonstrated
same phenotype
Mutation in A + Mutation in B = that, in addition to protein architecture
cancer cell c described as combinations of modules such
as globular domains and linear motifs19–21,
e protein domains themselves often have well-
Response A

Properties of cancer Activation of Activation of defined sectors formed by sparse networks of

signaling networks kinase 1 leads kinase 1 leads
Cue A to one outcome to another outcome residues often linking spatially distant regions
(anti-apoptotic) (pro-apoptotic)
that contribute cooperatively but unequally
to its function22,23. Although some targeted
Cue A studies analyzing several cancer mutations in
Baseline network a single kinase have been conducted24, similar
ate
B A d Network state A Network state B
approaches to those used for WW domains
ate
© 2012 Nature America, Inc. All rights reserved.

ll f ll f
Ce Ce should be pursued to generate high-throughput
experimental studies of cancer mutations in the
Dynamic context of signaling networks. These would
networks
help gain a better understanding of which
amino acid residues can be changed freely
without affecting the protein and network
function and, most importantly, which cannot.
Timepoint 1 Timepoint 2

From network perturbations to cellular

Figure 1 Properties of cancer signaling networks. (a) Analogous mutations. Two different tumors phenotypes
may achieve the same signaling and phenotypic outcome with two different mutations (b) Synthetic The characterization of cellular signaling pro-
oncogenes. Mutations that are not oncogenic on their own can cooperate when appearing together cesses has largely focused on identifying the
to drive tumor formation11; by analogy to synthetic lethality, we call the genes harboring cooperative function of individual genes and proteins. A
mutations, synthetic oncogenes. (c) Multivariate nature of signaling networks. The response of a cell to notable exception is a landmark study25 on the
a specific cue depends on, and can only be predicted by taking into account, the state of the cellular
context dependence of the Jun-activated kinase
signaling networks25. This dependency, known as the multivariate nature of signaling networks, is
often neglected when classifying mutations and genes as oncogenes or tumor suppressors and cancer
(JNK) in apoptosis. Before this work, para-
drivers or passengers. (d) Dynamic networks. Although signaling networks are often represented as doxical results suggested that JNK had a pro-
apoptotic function26, an anti-apoptotic func-
npg

static, it is clear that they are highly dynamic entities. Given that the role of signaling networks in
computing cellular responses is highly dependent on it, and that cancer mutations will perturb it, this
dynamic nature is a critical property of cancer signaling networks. (e) Signaling network landscapes.
The different states that a signaling network occupies can be represented as a landscape (with stable
steady states or attractors represented as valleys and unstable steady states represented as hills), where
the cell constantly gets pushed by signaling cues31,32,39,40. These states drive cellular and disease
phenotypes and represent network drug targets.
Because cancer cells would obtain the greatest fitness advantage
from mutations that target the most-functional residues, we reason
that a better understanding of the functionality of protein residues
would allow more accurate predictions of the consequences of cancer
mutations. Functional residues have been defined as those residues
required for a protein to perform its molecular function(s), in the
sense that they cannot be freely changed without directly affecting
the role(s) of the protein14. Here we extend this definition to include
a more fine-grained and precise definition of protein function as
an ensemble of protein features that together describe the different
functional capabilities of proteins (e.g., ATP binding, substrate
specificity, protein activation or phospho-tyrosine binding). This new
definition would not only adapt well to current studies of sequence-
function associations15,16, but also lead to a better description of the
effects of a mutation affecting such residues (Fig. 2a,b).
tion27 or even a lack of involvement in apopto-
sis28. The systematic approach undertaken by
Janes et al.25 revealed that the phosphorylation
status of JNK (and thus its catalytic activity)
was not sufficient to determine apoptotic com-
mitment; instead, activation of JNK could lead
to both apoptosis and proliferation depending
on the cellular signaling network state at the time of activation. Thus, this
work demonstrated that a protein’s cellular role is not a static property
but rather can only be defined dynamically—that is, its role depends
on the context of the network it is operating within. Similar context
dependencies have been confirmed for other kinases, such as Erk and
MK2. Because of this, which is referred to as the multivariate property
of signaling networks (Fig. 1c), we suggest that it is essential to study
cellular context at the systems level.
Although these multivariate molecular networks seem to have evolved
a complex structure that makes them robust against deletion of a few
proteins29, they are highly dynamic. Thus, a more accurate description
of signaling networks should take into account the fact that a single static
network does not exist unchanged over time. Instead, a cell contains
a dynamic ensemble of networks whose different permutations are
manifested in the cell depending on the different cues the cell is presented

nature biotechnology volume 30 number 9 SEPTEMBER 2012 843

 perspecti v e

with (Fig. 1d). This dynamic nature of signaling networks could, at least
in part, explain why all mutant proteins do not seem to be expressed at a
given point in time30, if a substantial part of the proteome is so dynamic
that it is expressed only when the cell senses a specific cue.
Moreover, according to a general principle of complex systems
introduced in the 1980s31,32, dynamic cellular networks can only exist
in a finite number of states, owing to the constraints that interactions
between nodes impose on one another. These network states can be
represented as landscapes, where most-probable and least-probable
states are represented as valleys and mountains, respectively (Fig. 1e).
Cells are continuously exploring this landscape
and are pushed from one state to another by
different environmental or intracellular cues.
Implications for cancer research
The multivariate nature of signaling networks has
profound implications for cancer research. Just as a Disease genome
We postulate that network-attacking mutations affect the cell not by
perturbing how the signaling landscape is projected to the phenotypic
dimension, but by changing the ensemble of dynamic networks that
can be manifested in a cell and, in consequence, the number and
stability of steady states in the signaling landscape, thus creating new
attractor states that only cancer cells can occupy, also known as cancer
network attractors (Fig. 3). This has additional implications for other
mechanisms, such as oncogene and non-oncogene addiction41, where
cancer cells would be trapped in cancer attractor states and could
escape from them by reverting the genomic aberration that initially
b
A
Feature 1 P P
B C
DP P P
E

Feature 2 Disease networks

it is inaccurate to assign a static function (e.g.,
apoptotic or anti-apoptotic) to a single protein,
it is clear that static interpretations of mutations, Feature 3
that is, driver or passenger mutations, are also
© 2012 Nature America, Inc. All rights reserved.

misleading. For example, given that the pheno- Protein sequence Disease proteome
typic role of JNK strongly depends on network c
state, it is clear that a mutation in JNK (and thus e
probably any other mutation) should not be

Response B
Challenges in cancer
statically labeled as a driver or passenger or as network biology Cue A
an oncogene or tumor suppressor, as such clas-
sifications are context dependent (e.g., disease
or cell-type specific). Several examples, such Drug A Drug B Cue A
as Myc33 or WT1 (ref. 34) gene products that
act as both tumor suppressors and oncogenes,
support this idea. These results underscore the
Rewiring dTumor subpopulations Ce
ll fat
e
B

Ce
ll fat
eA

Immune system
importance of assessing mutations based on their
effects on signaling networks and of developing
novel classification methods to do so. Along these
lines, MAP2K4 (one of the protein kinases that
can phosphorylate and activate JNK) has been
shown to be recurrently lost or mutated in sev- Surrounding Matrix
tissue
eral cancers35–38. These represent prime examples
npg

of mutations that may display ambivalent pheno-

typic impact similar to JNK.
Motivated by the example of MAP2K4 and
many other mutated kinases38, we maintain
that mutations capable of affecting signaling
networks—which we call network-attacking
mutations (Fig. 2c)—are more likely to affect
phenotype than other mutations. Thus, we discuss
a general strategy in which mutations in individual
cancers are assessed based on, first, the likelihood
they will affect protein function, and second,
the cellular role of the signaling network that
they are operating within (Fig. 3). Our strategy
extends the concepts introduced by Waddington
and elaborated by Kauffman and Huang et
al.31,32,39,40, where cancer mutations are turned
into perturbations capable of reshaping these
landscapes. We represent the cellular response
or phenotype as another dimension where each
network state (every point in the landscape) is
constantly projected to and translated into a
cellular decision or phenotypic outcome.
Figure 2 Challenges in cancer network biology. (a) Functional consequences of cancer mutations.
Using an ensemble of protein-function features (e.g., ATP binding, substrate specificity, activation
of the protein kinase or phospho-tyrosine binding), which together represent a comprehensive
description of a protein’s molecular functions, will enable more accurate and predictive evaluation
of cancer mutations. (b) Modeling of disease networks. Although experimental and computational
tools for modeling molecular networks exist, creating more comprehensive, sensitive and
accurate new tools especially designed to model disease-associated networks still represents a
big challenge in network biology. (c) Network-attacking mutations and cancer network attractors.
Network-attacking mutations are mutations that lead to a new cellular phenotype by perturbing
signaling networks either at the network structure or the network dynamics level. Network-
attacking mutations transform signaling networks, generating new possible network states by
changing the number and/or stability of steady states in the signaling landscape 31,32,39,40. These
acquired signaling capabilities lead to alterations in the cell’s normal ‘cue-signal-output’ flow
and thereby drive disease phenotypes (see Fig. 3 for further details). (d) Tumor subpopulations
and micro-environment. The field is only beginning to comprehend the complex interactions that
exist between different co-evolving tumor cell subpopulations and between those cells and the
tumor microenvironment, both of which strongly influence tumor progression. (e) Network-aware
and temporal drugs. As predicted by R.L. and Pawson66 several years ago, new pharmaceutical
strategies that target networks instead of single proteins are becoming available 47,48. We predict
this trend will not only continue, but also include recent advances that highlight the possibility to
‘cure’ networks using time- and order-dependent therapies68. In coming years, the discovery of
resistant, metastatic, tissue or cell-specific networks could lead to an even greater advance in the
field of network medicine (Fig. 5).

844 volume 30 number 9 SEPTEMBER 2012 nature biotechnology

 perspecti v e

Wild type Network-attacking mutation Network-attacking mutation

caused the perturbed landscape. Given affecting network dynamics affecting network structure
the high degree of determinism that exists
No mutation Mutation A Mutation B
between signaling networks, landscapes and
phenotypes, we argue that network-attacking Genome
mutations are at the heart of all new decision-
making capabilities acquired by cancer cells. Kinase A A A
Consequently, in our view, the study of No
P
cue Substrates B C B C B C
both network-attacking mutations and new Signaling
attractor states acquired by cancer cells, that network
A A A
is, cancer network attractors, deserves the Cue P P
B C B C BP C
highest priority from the field. Such studies
should be performed through systematic and
Cue
quantitative sampling of cell dynamics at Signaling Cue
multiple levels (e.g., genomic or epigenetic, landscape Cue Cancer
Cancer attractor
proteomic and phenotypic), followed by attractor

B
A

A
nonlinear interpolation and integrative Phenotypic

te

te

te

te

te

te
fa

fa

fa

fa

fa

fa
ll

ll

ll

ll

ll

ll
outcome

Ce

Ce

Ce

Ce

Ce

Ce
computational modeling (Fig. 4).
The first network-attacking cancer mutation,
Figure 3 Network-attacking cancer mutations. Proteins are the key elements of signaling networks as a
described more than 15 years ago42, was a point result of their ability to integrate external cues and direct the information flow toward a specific cellular
mutation in the kinase domain of RET (M918T), outcome (e.g., epidermal growth factor (EGF) leading to proliferation or tumor necrosis factor alpha
which leads to a switch in peptide specificity. In (TNF-a) leading to apoptosis). Network-attacking mutations affect the ‘cue-signal-output’ cellular
© 2012 Nature America, Inc. All rights reserved.

line with their importance, network-attacking information flow by affecting either the dynamics (middle), for example, by keeping proteins
mutations have attracted more attention in constitutively active, or the structure (right), by affecting protein specificity, of the signaling networks.
recent years43–48. Moreover, information has Signaling networks can be represented as a landscape with the most likely network states represented
as valleys (stable steady states or attractors) and the least likely network states as mountains (unstable
been accumulating steadily about how specific- steady states). Network-attacking mutations dysregulate signaling networks by perturbing the number
ity in signaling networks and modular protein and/or stability of steady states in the landscape, effectively creating new cancer-specific attractors that
domains emerges49–51, leading to the defini- only cancer cells will be able to reach.
tion of determinants of specificity in protein
domains52,53. These determinants, sometimes referred to as specificity- Despite the fact that the number of known cancer network-attacking
determining residues, are residues that can lead to substrate specificity mutations is still relatively low, recent findings suggest that in-frame
changes after mutation. Notably, direct mutagenesis of these determinants mutations are enriched on interaction interfaces57, which implies
of specificity has been used to rewire the entire histidine kinase signal- they are also likely to affect determinants of specificity. Moreover,
ing system in bacteria in a predictive manner54. Recent follow-up work many fusion proteins have been discovered that likely directly rewire
indicates that mutations in determinants of specificity prevent cross-talk or create new network states58. Given the rate at which cancer muta-
and allow protein family expansions55, in a process similar to the one tions are being reported and the development of new computational
powered by negative selection over Src homology 3 (SH3) protein domains methods for systematically identifying these mutations (Fig. 2b),
that show similar specificity56. We propose that similar studies in human we predict a steep increase in the number of network-attacking muta-
signaling networks, coupled with mapping of cancer mutations on these tions that will be uncovered in the coming years.
determinants of specificity, would shed new light on whether signaling
npg

rewiring is a general principle of oncogenesis and tumor progression, Personalized cancer network biology
knowledge of which would in turn be critical as molecular therapies tar- Led by recent advances in sequencing technologies, the amount of data
get proteins and their networks and not genes. on cancer genome mutations is growing exponentially59. Current efforts

Sampling strategy - Interpolation -

Figure 4 Traditional versus network biology Comprehensive Linear
approaches. In more traditional biological
approaches, where only one or a few genes
or proteins are sampled across a limited
[Stimuli y]

[Stimuli y]

set of conditions, there has been limited Traditional biology

approach
success in deriving predictive models across Initial data set
conditions or cell types that would require
High
comprehensive sampling. In contrast, apoptosis [Stimuli x] [Stimuli x]
network biology relies on systematic
[Stimuli y]

sampling across combinations of states

that result in increased performance of a
network model. Unlike classic approaches, Sampling strategy - Interpolation -
in which the system is stimulated with Low Model-driven Nonlinear
single specific cellular cues (e.g., growth [Stimuli x] apoptosis
factor), in the network biology approach, the
Network
[Stimuli y]

[Stimuli y]

multivariate nature of signaling networks and

Network biology model
the nonlinear relationship between signaling approach
input and output can be successfully
elucidated by interrogating the system with
[Stimuli x] [Stimuli x]
multiple orthogonal cues.

nature biotechnology volume 30 number 9 SEPTEMBER 2012 845

 perspecti v e

from the Cancer Genome Atlas and Cancer

A
Genome Project, now under the umbrella of the P
P
International Cancer Genome Consortium60, Single-cell B C
will facilitate the annotation and collection of DNA sequencing
DP P E
P
cancer genome data. We foresee similar waves
Interpretation
of technological progress and the generation
Best combination
of new consortiums in the cancer proteomics CyToF/mass spectrometry
fields in the near future. The establishment
of the Clinical Proteomic Tumor Analysis
Consortium (http://proteomics.cancer.gov/pro- Cell culture
grams/cptacnetwork), and the implementation
Network drugs Resistance
of new approaches61 and labeling techniques62 follow-up
Early diagnostic Phenotyping
optimized for patient samples are encouraging
advances in this direction.
These advances, however, will need to
be coordinated with new algorithmic and
experimental high-throughput methods (e.g., Xenograft model
high-content screening) capable of interpreting
this flood of information because the functional Figure 5 Personalized cancer network biology. The goal of personalized cancer network biology is to
interpretation of the data is currently the main be able to treat each tumor with the best combination of drugs tailored to that tumor. Ideally, early
bottleneck in the field of personalized cancer diagnosis should be followed by the development of tumor-specific cell lines and xenograft models,
© 2012 Nature America, Inc. All rights reserved.

cancer genome sequencing, and proteomic and phenotypic analysis. Combinations of network drugs
network biology. Computational integration should then be tried in the tumor-specific cell line and xenograft model and eventually transferred
of large quantitative data sets is also becoming back to the patient. Continuing to treat the tumor-specific cell culture with the same network drug
increasingly important, and thus there is a combination as is used in the patient may be useful for understanding potential resistance and/or
growing requirement for supercomputing metastasis.
infrastructure with large algorithmic dynamic
range (e.g., next-generation large shared memory systems). Benchmarking high-throughput profiling of phenotypic cell states in the tumor and
and validation of systematic workflows and algorithms is already receiving design of patient-specific combinations of network drugs with resistance
increasing attention through initiatives, such as the DREAM challenge63 follow-up (Fig. 5). Relatively new techniques, such as single-cell and high-
and IMPROVER64. depth sequencing70,71, imaging72 and cytometry time-of-flight73, could
Two emerging areas in network biology that are likely to contribute to prove especially valuable for monitoring the number, properties and
the future of cancer research are the study of cell-cell interactions (Fig. 2d) behavior of different tumor subclones (Fig. 2d). Ideally, network drugs,
and drugs specifically designed to interfere with diseased network such as the aforementioned order- and time-dependent combination68,
dynamics (that is, network drugs; Fig. 2e). should then be chosen based on the interpretation of sequencing as well
R.L. and collaborators65 studied cell-cell interactions by isotopically as the proteomic and phenotypic analysis of tumor cells and tested on
labeling two distinct subpopulations of cells, one expressing ephrin-B1+ the tumor-specific cell lines and xenograft model. The best-performing
and the other Eph-B2+, and carrying out a comprehensive phospho- combination should ultimately be transferred back to the patient (Fig. 5).
proteomic analysis. This strategy facilitated the first measurements This whole process should take the shortest time possible to avoid
of phosphorylation events during the interaction of two cell the evolution of the tumor in the patient and the consequent loss of
npg

subpopulations. The proliferative behavior of cancer cells is still poorly relationship between the primary tumor and the cell line. Tumor-
understood in part because it is difficult to experimentally study the specific cell lines would be kept and treated with the same drugs used
transmission of proliferative factors from one cell to its neighbors3. in the patient to monitor tumor evolution and treat for resistance
Therefore, we argue that a similar isotopic labeling strategy could and/or metastasis as soon as there is enough evidence of it (Fig. 5).
be used to investigate the cooperation between cells with different Ideally, every patient and paired xenograft or cell line should have a
oncogenic lesions that together (that is, synthetic oncogenes; Figs. 1b complete electronic record showing the treatment history to facilitate
and 2d) lead to tumor formation11. retrospective and cross-disease studies74,75.
Combination drugs that interfere with disease networks (so-called
network medicine66) have been shown to lead to a better response than Conclusions
single-hit therapies by causing secondary perturbations to signaling Although we have highlighted some of the challenges that still exist in
networks47,48,67. Recent work by the Yaffe laboratory represents a clear cancer network biology, substantial progress is also being made. For
leap forward within the field of network medicine68,69. Following network example, the usage of patient-derived tumor tissue in animal xenograft
modeling, Yaffe and colleagues68 managed to decode the signaling models to test the response to particular drugs aimed at developing
network dynamics that drive resistance to DNA-damaging chemotherapy. new personalized cancer therapy is rapidly becoming an established
This information was used to sensitize otherwise resistant triple-negative technology76. Surgical orthotopic implantation to transplant tumors
breast cancer cells to conventional DNA-damaging chemotherapy by taken directly from the patient to the corresponding organ of immu-
administering doxorubicin (Adriamycin, Doxil) and erlotinib (Tarceva) nodeficient mice77 is currently one of the most promising methods to
in an order- and time-dependent fashion. This could be considered the enable drug screening in patients. In addition, new clinical trials, such
first example of temporal network drugs (Figs. 2e and 5). as the MD Anderson T9 project78, are under way in which patients are
We predict that personalized or even tumor-specific cancer therapy will given therapy that targets tumor-specific aberrations. Nevertheless,
become a reality in the foreseeable future, starting from early diagnosis of the implementation of the strategy depicted in Figure 5 would benefit
the disease, followed by next-generation sequencing, proteomic analysis, from further developments in technology, funding and legislation. For

846 volume 30 number 9 SEPTEMBER 2012 nature biotechnology


example, generating models for cancer research that represent human
patient diversity79 and mimicking the complexity of tumor microenvi-
ronments (J.T.E. and collaborators)80 remain extraordinary challenges
(Fig. 2), and further research efforts and investments are required.
As cancer biology becomes a ‘big data’ science, similar to physics,
we expect to see more systematic, data-driven research efforts that
will uncover and confront many of the tumor complexities that have
remained elusive so far.
Despite recent predictions of >13 million cancer deaths in 2030
(ref. 81), as discussed in this Perspective, we foresee that within this
timeframe tumor-specific medicine will become a reality, thanks to
a new generation of cancer network biologists who will hopefully
overcome these challenges, positively contributing to the battle
against this devastating disease and the significant reduction of patient
suffering.
ACKNOWLEDGMENTS
We apologize to our colleagues whose work could not be cited due to space
limitations. We thank all members of the C-SIG (DTU), the ErlerLab (BRIC),
M. Yaffe (MIT) and N. Brunner (KU) for critical input on this manuscript. R.L. is
a Lundbeck Foundation Fellow and is supported by a Sapere Aude Starting Grant
from The Danish Council for Independent Research and a Career Development
© 2012 Nature America, Inc. All rights reserved.
Award from Human Frontier Science Program. J.T.E. is supported by a Hallas
Møller Stipend from the Novo Nordisk Foundation. Visit http://www.networkbio.
org/, http://www.lindinglab.org/ and http://www.erlerlab.org/ for more information
on cancer-related network biology.
AUTHOR CONTRIBUTIONS
All correspondence should be addressed to both J.T.E. and R.L.
Competing FINANCIAL interests
The authors declare no competing financial interests.
Published online at http://www.nature.com/doifinder/10.1038/nbt.2345.
Reprints and permissions information is available online at http://www.nature.
com/reprints/index.html.
1. Nash, J.F. Equilibrium points in N-person games. Proc. Natl. Acad. Sci. USA 36,
48–49 (1950).
2. Nash, J.F. Non-cooperative games. Ann. Math. 54, 286–295 (1951).
3. Hanahan, D. & Weinberg, R.A. Hallmarks of cancer: the next generation. Cell 144,
646–674 (2011).
4. Hanahan, D. & Weinberg, R.A. The hallmarks of cancer. Cell 100, 57–70 (2000).
5. Davies, H. et al. Mutations of the BRAF gene in human cancer. Nature 417, 949–954
(2002).
6. Stratton, M.R., Campbell, P.J. & Futreal, P.A. The cancer genome. Nature 458, 719–
npg
724 (2009).
7. Ding, L. et al. Clonal evolution in relapsed acute myeloid leukaemia revealed by whole-
genome sequencing. Nature 481, 506–510 (2012).
8. Gerlinger, M. et al. Intratumor heterogeneity and branched evolution revealed by
multiregion sequencing. N. Engl. J. Med. 366, 883–892 (2012).
9. Hou, Y. et al. Single-cell exome sequencing and monoclonal evolution of a JAK2-
negative myeloproliferative neoplasm. Cell 148, 873–885 (2012).
10. Cancer Genome Atlas Research Network. Integrated genomic analyses of ovarian
carcinoma. Nature 474, 609–615 (2011).
11. Wu, M., Pastor-Pareja, J.C. & Xu, T. Interaction between RasV12 and scribbled clones
induces tumour growth and invasion. Nature 463, 545–548 (2010).
12. Ng, P.C. & Henikoff, S. Predicting the effects of amino acid substitutions on protein
function. Annu. Rev. Genomics Hum. Genet. 7, 61–80 (2006).
13. Dixit, A. et al. Sequence and structure signatures of cancer mutation hotspots in protein
kinases. PLoS ONE 4, e7485 (2009).
14. Pazos, F. & Bang, J.-W. Computational prediction of functionally important regions in
proteins. Curr. Bioinform. 1, 15–23 (2006).
15. Fowler, D.M. et al. High-resolution mapping of protein sequence-function relationships.
Nat. Methods 7, 741–746 (2010).
16. Jensen, L.J. et al. Ab initio prediction of human orphan protein function from post-
translational modifications and localization features. J. Mol. Biol. 319, 1257–1265
(2002).
17. Socolich, M. et al. Evolutionary information for specifying a protein fold. Nature 437,
512–518 (2005).
18. Russ, W., Lowery, D., Mishra, P., Yaffe, M. & Ranganathan, R. Natural-like function in
artificial WW domains. Nature 437, 579–583 (2005).
19. Puntervoll, P. et al. ELM server: a new resource for investigating short functional sites
in modular eukaryotic proteins. Nucleic Acids Res. 31, 3625–3630 (2003).
20. Lim, W.A. & Pawson, T. Phosphotyrosine signaling: evolving a new cellular
communication system. Cell 142, 661–667 (2010).
nature biotechnology volume 30 number 9 SEPTEMBER 2012 847
perspecti v e
21. Seet, B.T., Dikic, I., Zhou, M.M. & Pawson, T. Reading protein modifications with
interaction domains. Nat. Rev. Mol. Cell Biol. 7, 473–483 (2006).
22. Halabi, N., Rivoire, O., Leibler, S. & Ranganathan, R. Protein sectors: Evolutionary
units of three-dimensional structure. Cell 138, 774–786 (2009).
23. Reynolds, K.A., McLaughlin, R. & Ranganathan, R. Hot spots for allosteric regulation
on protein surfaces. Cell 147, 1564–1575 (2011).
24. Wan, P.T. et al. Mechanism of activation of the RAF-ERK signaling pathway by
oncogenic mutations of B-RAF. Cell 116, 855–867 (2004).
25. Janes, K.A. et al. A systems model of signaling identifies a molecular basis set for
cytokine-induced apoptosis. Science 310, 1646–1653 (2005).
26. Lei, K. & Davis, R.J. JNK phosphorylation of Bim-related members of the Bcl2 family
induces Bax-dependent apoptosis. Proc. Natl. Acad. Sci. USA 100, 2432–2437 (2003).
27. Lamb, J.A. et al. JunD mediates survival signaling by the JNK signal transduction
pathway. Mol. Cell 11, 1479–1489 (2003).
28. Abreu-Martin, M.T. et al. Fas activates the JNK pathway in human colonic epithelial
cells: lack of a direct role in apoptosis. Am. J. Physiol. 276, G599 (1999).
29. Jeong, H., Mason, S.P., Barabasi, A.L. & Oltvai, Z.N. Lethality and centrality in protein
networks. Nature 411, 41–42 (2001).
30. Shah, S.P. et al. The clonal and mutational evolution spectrum of primary triple-
negative breast cancers. Nature advance online publication, doi:10.1038/nature10933
(4 April 2012).
31. Kauffman, S. & Levin, S. Towards a general theory of adaptive walks on rugged
landscapes. J. Theor. Biol. 128, 11–45 (1987).
32. Kauffman, S.A. & Weinberger, E.D. The NK model of rugged fitness landscapes and its
application to maturation of the immune response. J. Theor. Biol. 141, 211–245 (1989).
33. Uribesalgo, I., Benitah, S.A. & Di Croce, L. From oncogene to tumor suppressor: The
dual role of Myc in leukemia. Cell Cycle 11, 1757–1764 (2012).
34. Yang, L., Han, Y., Saurez Saiz, F. & Minden, M.D. A tumor suppressor and oncogene:
the WT1 story. Leukemia 21, 868–876 (2007).
35. Ellis, M.J. et al. Whole-genome analysis informs breast cancer response to aromatase
inhibition. Nature 486, 353–360 (2012).
36. Curtis, C. et al. The genomic and transcriptomic architecture of 2000 breast tumours
reveals novel subgroups. Nature 486, 346–352 (2012).
37. Kan, Z. et al. Diverse somatic mutation patterns and pathway alterations in human
cancers. Nature 466, 869–873 (2010).
38. Greenman, C. et al. Pattern of somatic mutation in human cancer genomes. Nature
446, 153–158 (2007).
39. Waddington, C.H. The Strategy of the Genes: a Discussion of Some Aspects of
Theoretical Biology (Allen & Unwin, 1957).
40. Huang, S. & Ingber, D.E. Shape-dependent control of cell growth, differentiation, and
apoptosis: switching between attractors in cell regulatory networks. Exp. Cell Res.
261, 91–103 (2000).
41. Luo, J., Solimini, N.L. & Elledge, S.J. Principles of cancer therapy: oncogene and
non-oncogene addiction. Cell 136, 823–837 (2009).
42. Songyang, Z. et al. Catalytic specificity of protein-tyrosine kinases is critical for
selective signaling. Nature 373, 536–539 (1995).
43. Zhong, Q. et al. Edgetic perturbation models of human inherited disorders. Mol. Syst.
Biol. 5, 321 (2009).
44. Dreze, M. et al. ‘Edgetic’ perturbation of a C. elegans BCL2 ortholog. Nat. Methods
6, 843–849 (2009).
45. Pe’er, D. & Hacohen, N. Principles and strategies for developing network models in
cancer. Cell 144, 864–873 (2011).
46. Vidal, M., Cusick, M.E. & Barabási, A.-L.L. Interactome networks and human disease.
Cell 144, 986–998 (2011).
47. Schoeberl, B. et al. Therapeutically targeting ErbB3: a key node in ligand-induced
activation of the ErbB receptor-PI3K axis. Sci. Signal. 2, ra31 (2009).
48. Huang, P.H. et al. Quantitative analysis of EGFRvIII cellular signaling networks reveals
a combinatorial therapeutic strategy for glioblastoma. Proc. Natl. Acad. Sci. USA 104,
12867–12872 (2007).
49. Miller, M.L.L. et al. Linear motif atlas for phosphorylation-dependent signaling. Sci.
Signal. 1, ra2+ (2008).
50. Linding, R. et al. Systematic discovery of in vivo phosphorylation networks. Cell 129,
1415–1426 (2007).
51. Mok, J. et al. Deciphering protein kinase specificity through large-scale analysis of yeast
phosphorylation site motifs. Sci. Signal. 3, ra12 (2010).
52. Brinkworth, R.I., Breinl, R.A. & Kobe, B. Structural basis and prediction of substrate
specificity in protein serine/threonine kinases. Proc. Natl. Acad. Sci. USA 100, 74–79
(2003).
53. Turk, B.E. Understanding and exploiting substrate recognition by protein kinases. Curr.
Opin. Chem. Biol. 12, 4–10 (2008).
54. Skerker, J.M. et al. Rewiring the specificity of two-component signal transduction
systems. Cell 133, 1043–1054 (2008).
55. Capra, E.J., Perchuk, B.S., Skerker, J.M. & Laub, M.T. Adaptive mutations that prevent
crosstalk enable the expansion of paralogous signaling protein families. Cell 150,
222–232 (2012).
56. Zarrinpar, A., Park, S.H. & Lim, W.A. Optimization of specificity in a cellular protein
interaction network by negative selection. Nature 426, 676–680 (2003).
57. Wang, X. et al. Three-dimensional reconstruction of protein networks provides insight
into human genetic disease. Nat. Biotechnol. 30, 159–164 (2012).
58. Brehme, M. et al. Charting the molecular network of the drug target Bcr-Abl. Proc. Natl.
Acad. Sci. USA 106, 7414–7419 (2009).
59. Wong, K.M.M., Hudson, T.J. & McPherson, J.D. Unraveling the genetics of cancer: genome
sequencing and beyond. Annu. Rev. Genomics Hum. Genet. 12, 407–430 (2011).
 perspecti v e
60. Ledford, H. Big science: the cancer genome challenge. Nature 464, 972–974
(2010).
61. Bensimon, A., Heck, A.J.R. & Aebersold, R. Mass spectrometry-based proteomics and
network biology. Annu. Rev. Biochem. 81, 379–405 (2012).
62. Geiger, T., Cox, J., Ostasiewicz, P., Wisniewski, J.R. & Mann, M. Super-SILAC mix for
quantitative proteomics of human tumor tissue. Nat. Methods 7, 383–385 (2010).
63. Prill, R.J. et al. Towards a rigorous assessment of systems biology models: the DREAM3
challenges. PLoS ONE 5, e9202 (2010).
64. Meyer, P. et al. Verification of systems biology research in the age of collaborative
competition. Nat. Biotechnol. 29, 811–815 (2011).
65. Jørgensen, C. et al. Cell-specific information processing in segregating populations of
Eph receptor ephrin-expressing cells. Science 326, 1502–1509 (2009).
66. Pawson, T. & Linding, R. Network medicine. FEBS Lett. 582, 1266–1270 (2008).
67. Chandarlapaty, S. et al. AKT inhibition relieves feedback suppression of receptor tyrosine
kinase expression and activity. Cancer Cell 19, 58–71 (2011).
68. Lee, M.J. et al. Sequential application of anticancer drugs enhances cell death by rewiring
apoptotic signaling networks. Cell 149, 780–794 (2012).
69. Erler, J.T. & Linding, R. Network medicine strikes a blow against breast cancer. Cell 149,
731–733 (2012).
70. Navin, N. et al. Tumor evolution inferred by single-cell sequencing. Nature 472, 90–94
(2011).
71. Nik-Zainal, S. et al. The life history of 21 breast cancers. Cell 149, 994–1007 (2012).
© 2012 Nature America, Inc. All rights reserved.
npg
848 volume 30 number 9 SEPTEMBER 2012 nature biotechnology
72. Pedersen, M.W. et al. Sym004: a novel synergistic anti-epidermal growth factor receptor
antibody mixture with superior anticancer efficacy. Cancer Res. 70, 588–597 (2010).
73. Bendall, S.C. & Nolan, G.P. From single cells to deep phenotypes in cancer. Nat.
Biotechnol. 30, 639–647 (2012).
74. Roque, F.S. et al. Using electronic patient records to discover disease correlations
and stratify patient cohorts. PLOS Comput. Biol. 7, e1002141 (2011).
75. Jensen, P.B., Jensen, L.J. & Brunak, S. Mining electronic health records: towards
better research applications and clinical care. Nat. Rev. Genet. 13, 395–405
(2012).
76. Blumenthal, R.D. & Goldenberg, D.M. Methods and goals for the use of in vitro and
in vivo chemosensitivity testing. Mol. Biotechnol. 35, 185–197 (2007).
77. Hoffman, R.M. Orthotopic mouse models expressing fluorescent proteins for cancer
drug discovery. Expert Opin. Drug Discov. 5, 851–866 (2010).
78. Gonzalez-Angulo, A.M., Hennessy, B.T. & Mills, G.B. Future of personalized medicine
in oncology: a systems biology approach. J. Clin. Oncol. 28, 2777–2783 (2010).
79. Hunter, K.W. Mouse models of cancer: does the strain matter? Nat. Rev. Cancer 12,
144–149 (2012).
80. Cox, T.R. & Erler, J.T. Remodeling and homeostasis of the extracellular matrix:
implications for fibrotic diseases and cancer. Dis. Model. Mech. 4, 165–178
(2011).
81. WHO. World health organization fact sheet 297 (2012). http://www.who.int/
mediacentre/factsheets/fs297/en/

Lessons from human teratomas to guide development
of safe stem cell therapies
Justine J Cunningham1, Thomas M Ulbright2, Martin F Pera3 & Leendert H J Looijenga4
The potential for the formation of teratomas or other neoplasms
is a major safety roadblock to clinical application of pluripotent
stem cell therapies. Preclinical assessment of the risk of tumor
formation in this context poses considerable scientific and
regulatory challenges, especially because animal xenograft
models may not properly reflect the long-term tumorigenic
© 2012 Nature America, Inc. All rights reserved.
potential of human cells. A better understanding of the biology of
spontaneously occurring teratomas and related tumors in humans
can help to guide efforts to assess and minimize the potential
hazards of embryonic stem cell or induced pluripotent stem cell
therapeutics. Here we review the features of teratomas derived
experimentally from human pluripotent stem cells and argue
that they most closely resemble spontaneous benign teratomas
that occur early in both mouse and human life. The natural
history and pathology of these spontaneously occurring teratomas
provide important clues for preclinical safety assessment and
patient monitoring in trials of stem cell therapies.
The risk that therapeutics derived from human embryonic stem
(hES) cells or induced pluripotent stem (hiPS) cells will give rise
to neoplasms, particularly teratomas (Box 1) and other embryonal
tumors, is a central safety hurdle to the application of these tech-
nologies in regenerative medicine. Eliminating this risk requires
npg
extinguishing the intrinsic features of hES and hiPS cells—
pluripotency and indefinite self-renewal—before transplantation so
the cells undergo differentiation in a uniform, predictable manner.
With current awareness that many cancers may originate from a
stem cell component, there is concern that hES cell–derived trans-
plants might be contaminated with residual undifferentiated (that is,
pluripotent) cells capable of forming benign neoplasms or even of
giving rise to aggressive invasive or ‘malignant’ teratomas.
Preclinical safety assessment of the potential for tumor forma-
tion in stem cell therapy poses challenges for several reasons. First,
animal xenograft models may not accurately predict the fate of grafted
cells in humans. Second, for some applications it may be necessary to
repeatedly treat patients with doses containing billions of cells, and in
1Department of Drug Safety Evaluation, Allergan Inc., Irvine, California, USA.
2Department of Pathology & Laboratory Medicine, Indiana University School of
Medicine and Indiana University Health Partners, Indianapolis, Indiana, USA.
3The University of Melbourne, Walter and Eliza Hall Institute of Medical Research,
Florey Neurosciences Institutes, Melbourne, Victoria, Australia. 4Department
of Pathology, Erasmus MC-University Medical Centre Rotterdam (Daniel den
Hoed Cancer Centre), Josephine Nefkens Institute, Rotterdam, The Netherlands.
Correspondence should be addressed to J.J.C. (cunningham_justine@allergan.com).
Published online 10 September 2012; doi:10.1038/nbt.2329
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 849
perspective
practical terms it may be difficult to guarantee the complete absence of
undifferentiated cells in production at this scale. Third, although undif-
ferentiated pluripotent cells carry the greatest concern, it is uncertain
what risks are posed by other (progenitor) populations. Fourth, unlike
drugs, which have half-lives on the order of hours, cellular therapeutics
may be designed to persist for a large fraction of the patient’s life span,
and current technology for monitoring the long-term fate of grafted
cells in animals or humans has serious limitations.
Lessons learned from the biology of spontaneous human teratomas
offer a unique opportunity to gain insight into potential complications
of hES cell therapies and to answer key questions for the stem cell
field. In this Perspective, we survey the features of teratomas formed
by pluripotent stem cell grafts in animals and compare these experi-
mental tumors to the various classes of teratomas seen clinically. We
highlight those features of spontaneously occurring teratomas that
can help guide efforts to identify, understand and mitigate risks of
pluripotent stem cell therapies.
Study of tumors from human pluripotent stem cells
The assessment of the tumorigenic potential of human pluripotent
stem cells in animal models is carried out for two different objectives.
The most common goal is to assess the developmental potential of
a particular cell line, specifically, to determine whether the isolate is
pluripotent. Indeed the formation of teratomas in animals has become
the gold standard for hES cell pluripotency. However, this relates only
to the cells’ capacity to generate somatic lineages. A smaller number
of studies have examined teratoma formation from the standpoint of
preclinical safety assessment1,2.
The frequency of teratoma formation by human pluripotent stem
cell grafts can be influenced by a number of factors, including the site
of injection3–6, the cell inoculum6,7, the addition of excipients such as
Matrigel5,6 and inoculation of other cell populations into the graft8.
However, few of these variables have been convincingly shown to
influence the histological makeup of teratomas, with the interesting
exception of co-inoculation of human fetal cells (see below). Studies of
human cancer stem cell transplants in mice have demonstrated clearly
that the degree to which the host animal is immunocompromised
can profoundly influence the relationship between the size of the cell
inoculum and the probability of tumor formation9,10. Although the
influence of host immune phenotype on tumor formation has not
been systematically investigated for human pluripotent stem cells, it
is likely important for these cells as well.
In general, when ~1 × 106 undifferentiated diploid human pluripotent
stem cells are inoculated into immunocompromised mice, tumors arise
in a high proportion of recipients within 4 to 6 weeks3–6. The histogenesis
 Perspective

Box 1  Glossary of terms

Carcinoma. A malignant tumor derived from epithelial tissues such as the skin or epithelial components of other organs, such as the gut
and lungs.
Choriocarcinoma. A tumor comprising malignant trophoblastic tissues; may be seen in aberrant gestations (e.g., after hydatidiform mole)
or as a form of type II GCT.
Dysgerminoma. A primitive malignant tumor resembling primordial germ cells that occurs in the ovary; seminoma is the counterpart in
the testis and germinoma in extragonadal sites.
Dysplasia. Literally, bad or abnormal development. In pathology, a term most often used to describe early cellular abnormalities (nuclear
enlargement, hyperchromasia, cellular disorganization) that may progress to a fully malignant, invasive growth. Usually describes a
premalignant cellular condition.
Embryonal carcinoma. A tumor resembling the undifferentiated, highly proliferative, primitive epithelial tissues of the embryo with very
limited differentiation toward either embryonic or extra-embryonic lineages. The cells are characterized by expression of embryonic genes
OCT3/4, NANOG and SOX2.
Extragonadal. Any region of the body that is outside of the gonads, for example, head, chest, spine.
Germinoma. See Dysgerminoma.
Neoplasia. Literally, new development or growth. In pathology, a term used to describe a clonal proliferation of cells that most often
results in the formation of a mass (tumor). Resulting neoplasms may be either benign (nonspreading) or malignant (capable of metasta-
sizing to other sites).
© 2012 Nature America, Inc. All rights reserved.

Nonseminoma. A class of (testicular) GCTs that includes all types other than seminoma, that is, embryonal carcinoma, teratoma,
choriocarcinoma and yolk sac tumor (type II GCT). The distinction between seminoma and nonseminoma reflects clinical differences,
including age of onset, survival and response to treatment. The majority of nonseminomas are malignant.
-oma. Greek, group or mass. In medicine, refers to a neoplasm or tumor.
Sacrococcygeal. The ′tail end′ of the spine that includes the bones of the posterior pelvis, the sacrum and coccyx.
Seminoma. See Dysgerminoma.
Terato-. Greek, wonder, expression of divine power.
Teratos- or Teras-. Greek, deformity or monster.
Teratoma. A monstrous mass/tumor consisting of tissues from all three embryonic germ layers (ectoderm, mesoderm and endoderm), but
pathologists also recognize ‘bidermal’ and ‘monodermal’ teratomas. Simply, they are lesions foreign to their anatomical sites, such as
growth of hair and sebaceous glands within the lungs. These tumors are named for appearance rather than behavior. Most are benign,
whereas those that are malignant have defined histological characteristics.
Teratocarcinoma. An old term for a tumor with elements of teratoma and EC-like cells, the latter representing a malignant, pluripotential
cell population. More accurately, this class of tumor is termed a nonseminoma, which is by definition malignant.
Yolk sac tumor (endodermal sinus tumor). A neoplasm of germ or stem cell origin that resembles yolk sac elements of either extra-
embryonic or embryonic (gut epithelium) type.
npg

of these tumors is poorly understood, but there is evidence that the cells
recapitulate post-implantation embryonic development at an early stage
after engraftment11. The neoplasms that ultimately develop typically con-
tain a range of somatic tissues displaying varying degrees of maturation.
Some neoplasms show substantial organotypic differentiation, with vari-
ous cell types organized into structures resembling fetal or adult tissues,
but often they contain more primitive structures representing early stages
of development, such as neural rosettes (Fig. 1). Commonly there are
abundant cystic lesions of uncertain tissue origin. Although spontane-
ously arising neoplasms that consist largely of teratoma may also contain
elements representative of fetal membranes, such as yolk sac or placenta,
only one study has described these tissues in teratomas derived from the
implantation of diploid hiPS cells into animals12.
Importantly, a very common class of spontaneous tumors in
adults that may contain teratoma, type II testicular germ cell tumors
(GCTs)13 (Table 1), frequently contains undifferentiated malignant
cells called embryonal carcinoma (EC) cells. EC cells are the stem
cells of these tumors and are characterized by expression of pluri­
potency genes such as OCT3/4 (also known as POU5F1) and SOX2.
They have not been described in xenografts of diploid hES cells but
have been found in hiPS cell grafts 12. This is in contrast to tumors
formed by mouse ES14 and iPS cells15, which contain undifferenti-
ated tumor stem cells capable of metastasis.
One report examined the effect of transplanting human fetal
tissue in conjunction with hES cells8. Strikingly, the tumors from
these co-transplanted cells consisted not of teratomas but of primitive
a b c
Figure 1  Teratoma from hES cells showing varying degrees of maturation.
This teratoma from an hES cell transplant in immunocompromised mice
shows a range of somatic tissues displaying varying degrees of maturation.
(a) Primitive neuroectodermal tissue consisting of neural tube–like
structures (arrow). (b) Classic mesodermal tissues can be observed
in many hES cell–derived teratomas, including cartilage (arrow).
(c) Endodermal tissues resembling mucosal structures of the gut
surrounded by smooth muscle (arrow). H&E-stained paraffin sections of
human teratoma from the collection of M.F.P.

850 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 perspective

Table 1  The five types of human GCTs including teratomas

Type I II III IV V
Appearance

Phenotype Teratoma/yolk-sac tumor Seminoma/nonseminoma/ Spermatocytic seminoma Dermoid cyst Gestational trophoblastic
dysgerminoma/germinoma neoplasia (hydatidiform
mole & choriocarcinoma)
Elements of teratoma Yes Yes No Yes No
Incidence 0.12/100,000 6–11/100,000 0.1/100,000 3/100,000 7.5/100,000
Gender bias Females Males Males (exclusively) Females Females (exclusively)
Usual age of onset Neonates and children ~12–35 years >50 years Children/adults Fertile women
(adolescents to adults)
Cell of origin Stem cell/PGC/gonocyte PGC/gonocyte Spermatogonia/spermatocyte Primary/secondary oocyte Empty ovum fertilized
by sperm
Imprinting status Biparental/partially erased Erased Male Maternal Paternal
Genotype Diploid/aneuploid Tri-tetraploid/aneuploid Diploid/aneuploid Diploid/tri-tetraploid Diploid
Columns I, II and IV represent those GCTs that may contain elements of teratoma. Images are hematoxylin and eosin (H&E)-stained sections of human GCTs from the collection of
T.M.U. Table adapted from Oosterhuis and Looijenga13. Incidence, gender bias and usual age of onset were derived from the International Agency for Cancer Research82.
© 2012 Nature America, Inc. All rights reserved.

undifferentiated cells with a diploid karyotype that lacked markers of
pluripotent stem cells. Although the precise nature of these tumors
and their biological significance remain unknown, this finding sug-
gests that the microenvironment of the graft can profoundly influence
the histology of the tumor. This is also demonstrated in type II GCTs,
the histology of which is influenced by the anatomical localization
of the gonad, whereas mainly the seminoma phenotype occurs in
nonscrotal testis (as is the case for the ovarian counterpart).
During propagation in vitro, human pluripotent stem cells can
acquire characteristic genetic abnormalities, such as overrepresenta-
tion of chromosome 12 or 17 (refs. 16,17). Many of these abnormali-
ties are also found in some classes of spontaneously arising type II
GCTs that may contain a component of teratoma. Several reports have
shown that cell lines bearing these characteristic genetic abnormali-
ties form tumors with fewer differentiated elements, and at a lower
cell inoculum, compared with diploid cells18–20. There are also some
indications that tumors from genetically abnormal hES cell cultures
npg
however, whether such lesions should be regarded as neoplasms dis-
tinct from teratomas or, rather, as ‘monodermal’ teratomas, that is,
lineage-restricted proliferations that are otherwise similar to classic
teratomas (see below). Importantly, any form of malignant or benign
neoplasm that arises from a graft presents a safety hazard. Unwanted
ectopic graft tissue can impinge on vital structures directly or indi-
rectly (secrete neurotransmitters, hormones or induce inflammation,
fibrosis or malignant transformation). However, patients with benign
spontaneous teratomas, even large ones, are generally cured by surgi-
cal resection alone25,26.
The risk of tumor formation associated with any pluripotent stem
cell–derived therapy requires the development of strategies to assure
the absence of residual stem cells. The challenge occurs in trying to
guarantee that a product is completely free of undifferentiated cells
or other partially differentiated cells that could generate a tumor.
Below we discuss features of spontaneous teratomas in humans that
may illuminate efforts to estimate the risks of pluripotent stem cell–

can contain elements of EC21,22, although this is as yet an infrequent
observation.
The primary risk of pluripotent stem cell therapy is often consid-
ered to be teratoma formation, but there is also the potential that
other types of embryonal neoplasms corresponding to specific tis-
sue lineages might arise after engraftment. Hypothetically, these
could develop from progenitor populations that possess varying
degrees of replicative capacity. Some examples include the forma-
tion of cystic lesions in rats inoculated with oligodendrocyte precur-
sors23 and the formation of primitive neural tissue after engraftment
of hES cell–derived dopaminergic neurons24. It remains debatable,
derived therapies.
Spontaneous teratomas in humans
Most spontaneous teratomas belong to a subgroup of a broader and
complex class of neoplasms referred to as GCTs. Recently, a new clas-
sification of GCTs was proposed based on genetics and site of origin
as well as morphology13,27. Five types were identified (Table 1), three
of which— types I, II and IV—may contain elements of teratoma.
The predominantly gonadal distribution and origin of types II and
IV teratomas indicate that they likely arise out of multiple defective
processes during germ cell development. In contrast, the origin of

Table 2  Similarities between Type I teratomas and hES cell–derived teratomas

Type Phenotype Cytological features Cell of origin Imprinting status Genotype Common aneuploidies
Type I teratoma Mature or immature ter- Immature organoid PGC or pluripotent Biparental or partially Diploid Loss of (parts) of 1p, 4, 6q;
atoma (unless complicated differentiation with stem cell erased gain of (parts) of 1q, 12p13,
by yolk sac tumor) or without yolk 20q, 22 (only observed in yolk
sac elements sac tumor)
hES cell teratoma Mature or immature Immature organoid Pluripotent cell from Biparental Diploid 3, 5, 8, 12, 13, 17, 18, 20,
teratoma (with or without differentiation with or ICM of blastocyst (monoallelic) 21, X
cystic elements) without cysts
ICM, inner cell mass; PGC, primordial germ cell.

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 851

 Perspective

Figure 2  Genomic imprinting and developmental origin of GCTs. Germ cell tumors (I–IV)
Schematic of normal (black) and aberrant (red) germ cell development. Normal development
Embryogenesis starts with fertilization and formation of a population
GI Fertilization GI
of pluripotent stem cells (epiblast cells), which give rise to the various
differentiation lineages (somatic and extra-embryonic) in vivo and to ES +
cells in culture. All pluripotent stem cells have a bi-parental pattern of
genomic imprinting (GI) (yellow), resulting from a pure male (blue) and Proliferation

Not present

Not present
female (purple) mature germ cell. GI is an epigenetic mechanism that ES
differentially marks the maternal and paternal alleles of a specific set of Differentiation
~100 genes, such that only one parental allele of an imprinted gene is ES
expressed. Imprinting marks inherited from either parent must be erased TE/YST
Specification
TE/YST
and reset during development of the germ line, an event that occurs once (I) Germ cell Compartment
(I)
the primordial germ cells have migrated to the gonads. The different types PGC
Migration
of GCTs originate from different stages of stem or germ cell development.
Erasement
Type I are teratomas (TE) and yolk sac tumors (YST) found in neonates Gonocyte
SE/NS DG/NS
and infants that originate from a stem or primordial germ cell with a bi- Sex determination
(II) (II)
parental or partially erased pattern of GI. Type II are the seminomas (SE)/ (Pre-)
dysgerminomas (DG) and nonseminomas (NS) diagnosed in adolescents Spermatogonium Oogonium
and young adults that originate from an erased germ cell. Type III are the (A,B)
spermatocytic seminomas (SS), predominantly in elderly males. Type IV Prim. oocyte
are dermoid cysts (DC). Type V are hydatidiform moles (MH). The timing of
birth, start of meiosis and puberty are indicated.
SS Prim. spermatocyte DC
(III) (IV)
© 2012 Nature America, Inc. All rights reserved.

Sec. spermatocyte Sec. oocyte

type I tumors is less clear; they may arise either from pluripotent cells

Germ cell

Germ cell
Spermatid Ovum

Soma

Soma
of the peri-implantation embryo that do not undergo normal pattern-
Testis Ovary
ing or from an early germ cell precursor. Thus, the term ‘teratoma’
designates a generic class of neoplasms with similar histopathological
appearance but potentially diverse origins. MH
(V)

Teratomas formed by pluripotent stem cells resemble type I GCTs
Histologic and genomic analysis suggests that teratomas in the
type I GCT category, but not type II or IV, are the most plausible clini-
cal equivalent of an hES cell–derived teratoma. Human pluripotent
stem cells are phenotypically similar to the human EC cells found in
type II GCTs, and some previous reports have suggested that experi-
mental teratomas derived from karyotypically abnormal hES cells
mimic type II GCTs16. Certainly, some features of type II GCTs are
similar to teratomas derived from abnormal hES cells, including an
immature tissue phenotype, appearance of carcinoma-like or EC cells
and aneuploidies (gains or losses in chromosomes X, 12 and 17)16.
But teratomas derived from diploid hES or hiPS cells do not resemble
npg
Sperm Mature oocyte
Birth Start meiosis Start puberty
tetraploid (Table 1). The exception are type IV teratomas, which
frequently show loss of heterozygosity for polymorphic markers,
supporting their derivation from a diploid, post-meiotic I germ
cell, with any given chromosome being either of mostly maternal or
paternal origin32.
Similarities in differentiation patterns. Differentiation within
type I and hES cell teratomas results in the development of mature,

type II GCTs. In fact, type II GCTs originate from a malignant
­intratubular germ cell28 with an aneuploid DNA content29. In contrast,
type I mature teratomas generally have no malignant potential.
Type IV teratomas should not be equated with pluripotent stem
cell teratomas either, despite their content of mature, organoid
tissue types. They are almost exclusively found in the ovaries and arise
from an oocyte that has undergone parthenogenesis. These lesions
are usually cystic and form skin-like structures (dermoid cysts or
mature cystic teratomas).
Type I GCTs are divided into two classes: teratomas, which are dip-
loid, and yolk sac tumors, which are typically aneuploid. They can
arise in a wide variety of body regions, with the anatomical site playing
a part in their pathogenesis. Sacrococcygeal teratomas are the most
common extra-gonadal type I GCT30 and are the likely parallel of hES
cell–derived teratomas. That these tumors might originate from an
early germ cell is supported by experiments in mice31. Evidence sup-
porting the close relationship between type I GCTs and teratomas from
pluripotent stem cells is discussed below and summarized in Table 2.
Karyotypic similarities. Both type I and hES cell teratomas contain
diploid, karyotypically normal cells. By contrast, most teratomas
within the other types of GCTs are typically aneuploid, triploid or
organoid cellular structures. Type I teratomas characteristically
­consist of mature tissues having organ-like relationships to each other,
including, for example, smooth muscle encircling, in an ­intestinal-
like fashion, enteric-type epithelium resting on a ­supporting ­lamina
propria, or pancreatic tissue having normal ductal and islet cell
components admixed with acinar tissue. Remarkably, teratomas
derived from karyo­typically normal hES cells show an organoid
pattern of differentiation that extends to the development of tissue
niches within the tumor. This suggests that developing structures
within hES cell teratomas retain their normal interactive signals to
co-instruct the specification of surrounding regions33,34, mimicking
embryonic development. We (L.H.J.L) and others have seen OCT3/4+
cells in type I teratomas, but they are negative for CD30 (refs. 35–37),
indicating that they do not contain EC cells.
Shared cell of origin? Type I teratomas and hES cell–derived terato-
mas may have a similar cell of origin—a pluripotent stem cell from
the inner cell mass or an early germ cell. hES cells are generated by
culturing the inner cell mass. However, Zwaka and Thomson specu-
lated that during ES cell establishment, the pluripotent self-renewing
population arises from an obligate early germ cell intermediate
that differentiates from the inner cell mass38. Recent evidence from

852 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 perspective

mouse studies supports this hypothesis39,40. Table 3  Histologic features associated with malignancy in human teratomas
Likewise, the sacrococcygeal teratoma is Malignant features Description Tissue types of concern Examples
hypothesized to originate from a stem cell Primitive Cellular components with Primitive neuroectodermal
from the inner cell mass or a remnant of the ­embryonic-type densely packed chromatin rosettes or tubules; primitive
epiblast (a derivative of the inner cell mass), elements showing both apoptosis and nephrogenic tissues
mitotic activity
or from a very early primordial germ cell.
These stem cells are thought to have escaped
normal regulatory control and become mis-
placed during embryogenesis41,42. It has been
suggested that sacrococcygeal teratomas Extra-embryonic Spider web-like network of Yolk sac elements
express pluripotent markers without indica- tissues cells with vacuolated cyto-
plasm and abundant basement
tions of malignant transformation36, but this membrane deposits; may also
has to be confirmed in independent studies. have tubular, cystic, papillary
Further support for a very early plur­i­ and solid patterns of cells
potent or germ cell origin of sacrococcygeal
teratomas comes from studies of genomic
imprinting (Fig. 2). Type I teratomas show a Vascular space Endothelial-lined spaces Intravascular thrombi
invasion containing clumps of
pattern of imprinting that is either somatic or
neoplastic cells
partially erased, and therefore consistent with
an origin in the peri-implantation epiblast or
early primordial germ cell13,27,43. By contrast,
© 2012 Nature America, Inc. All rights reserved.

type II teratomas have a completely erased

genomic imprint, consistent with a gonocyte
origin, whereas type IV teratomas show com- Phenotype of Any of a variety of neoplasms Established pathological
plete maternal imprinting, consistent with a malignancy as seen at numerous hallmarks of malignancy
13,27,43 body sites
post-meiotic germ cell origin . These
data suggest that type I tumors originate
from pluripotent stem cells or early primor-
dial germ cells. In hES and hiPS cell cultures,
imprinting is variable, with partial erasure of
imprinting at some loci in some cell lines44. Images are H&E-stained sections of human teratomas from the collection of T.M.U.
Although this variability is often attributed
to growth under artificial conditions in vitro,
it could also be consistent with a developmental status between that genetic, epigenetic and histological features of these usually benign
of an epiblast cell and one that has embarked on the pathway to germ tumors that are strongly predictive of progression to malignancy.
cell formation and erasure of imprinting. The appearance of such features in hES or hiPS cell–derived terato-
mas would therefore be suggestive of the presence of transformed
Similarities in dysregulation. Finally, cells giving rise to both type I cells in the cultures.
teratomas and hES cell–derived teratomas might share similar chro-
npg

mosomal aberrations and cellular hallmarks when malignant progres- Histological hallmarks of malignancy in teratomas. Malignant
sion is observed16,45–48. Morphologically, these aberrations manifest changes in human teratomas in general correlate with the pres-
as distinct features in sacrococcygeal teratomas, including the pres- ence of several key histological hallmarks (Table 3 and Fig. 3). The
ence of pockets of OCT3/4+ cells, yolk-sac elements and, in some grading of teratomas was originally established for ovarian cases
instances, neuroectoderm. Not surprisingly, disabling p53 improves and ranges from grade 0 (absence of embryonic-type elements) to
hiPS cell generation but also leads to enhanced tumor formation in grade 3 (substantial amounts of immature elements)50. Malignant
chimeric mice49. These topics are discussed further below. behavior is almost exclusively confined to those tumors that are
grade 2 or 3 (‘high’ grade)51 and is based on the selection of the
Lessons from teratomas for stem cell translation slide that contains the greatest amount of immature neuroectoderm.
The demonstrated potential of cultured human pluripotent stem Similar criteria have also been applied to sacrococcygeal teratomas
cells to accumulate genetic lesions characteristic of malignant (type I) and found to correlate with clinical behavior52 but are not
transformation has been a major concern for the field. Advances in useful for type II teratomas. Karyotypic abnormalities (gain of auto-
genomics technology have enabled genome-wide sequencing of hES somal and X chromosomes) in immature ovarian teratomas have
and hiPS cells, and, as a consequence, in addition to gross karyotypic correlated with large volumes of embryonic-type neuroectoderm,
changes, there are increasing reports of small amplifications, dele- with normal chromosomal findings in tumors lacking or having
tions and point mutations occurring during prolonged periods of small amounts of such tissue53,54. In addition, in sacrococcygeal
culture16,17. The emerging challenge for safety assessment lies in the teratomas, high-grade immaturity correlates with the development
interpretation of the significance of these changes and their potential of a yolk sac tumor component, which is considered to be the ‘true’
effects on cell behavior. Biological assays that enable the detection indicator of malignant behavior in this disease 45,52. Karyotypic
of a malignant phenotype in pluripotent stem cells are therefore studies of sacrococcygeal teratomas lacking a yolk sac tumor com-
of enormous value. Here, reference to spontaneously occurring ponent have yielded normal results, with chromosomal abnormali-
teratomas can inform our understanding because there are certain ties correlating only with the evolution of yolk sac tumor 46,54–58.

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 853

 Perspective

Figure 3  Histologic features characteristic of malignant progression

in human teratomas. (a) This benign teratoma in a child shows an a b
organoid replication of pancreatic tissue, including large ducts, lobules
of acinar tissue and occasional islets. Scale bar, 500 µm. (b) Primitive
neuroectodermal tissue (arrow, upper left field) consisting of neural-
type tubules lined by stratified columnar cells is interspersed with more
mature tissues, including a cellular fibrous stroma and cystic gland
Scale bar, 200 µm. (c) Overgrowth of primitive neuroectodermal tissue
in this testicular teratoma from a young man results in a primitive
neuroectodermal tumor (PNET). Scale bar, 200 µm. (d) Endodermal yolk-
sac elements appear as loosely arranged cords of cells (arrow, upper field)
and represent the classic hallmark of a malignant change in an otherwise
pure teratoma. Scale bar, 100 µm. H&E-stained paraffin sections of
human teratoma from the collection of T.M.U.

Importantly, the presence of primitive embryonic tissue components,

in the absence of the other criteria, classifies a lesion as an imma- c d
ture teratoma and potentially malignant.
Primitive neuroectodermal elements have been observed in
grafts of pluripotent stem cells (Fig. 1), as have yolk sac elements12,
consistent with their similarity to the type I GCT. It is not clear
whether yolk sac elements are rare or whether they have been under-
© 2012 Nature America, Inc. All rights reserved.

reported because they were unrecognized. Thus, accurate assessment

of the malignant potential of teratomatous lesions requires a trained
pathologist who has extensive experience in the spectrum of human
germ cell neoplasia.

Genetic and epigenetic changes predictive of malignancy. Previous

studies have noted that frequently observed karyotypic abnormalities
in pluripotent stem cells, in particular overrepresentation of chromo-
somes 12, 17 and X16, or functional inactivation of the TP53 pathway,
are characteristic of malignant type II GCTs13,59. Malignancy in spon-
taneous type I human teratomas is most often correlated with yolk sac
elements that contain similar and other aneuploidies (Table 2)46,55–58.
The remainder of the type I teratomas contain diploid cells with no
current genetic data to explain pathogenesis. Several studies have
demonstrated altered growth and differentiation properties of aneu-
ploid hES cells60,61. Only one study has shown that genetic aberrations
in hES cell lines alter the behavior of cells derived from them. In this
report, differentiated astrocytes derived from an aneuploid hES cell
npg
Preclinical and clinical assessment of safety
The preclinical safety assessment of cellular therapeutics currently
relies on in vitro and in vivo tests. The key genetic and histological
features that characterize malignant progression of type I teratomas
in humans are discussed above and should be considered in safety
assessment when examining cell lines or teratoma xenografts derived
from them. Currently, the main objectives of in vitro safety testing are
to assess the presence of stem cells, the purity of the differentiated

line had properties analogous to astrocytoma and glioblastoma62. In
principle, any genetic alterations could confer a metastatic potential
and should be used as an exclusion criterion.
Could epigenetic change account for the appearance of a benign
type I teratoma? There may be subtle changes in the epigenome inde-
pendent of DNA promoter methylation, possibly relating to polycomb
genes, that silence tumor suppressor gene activity in cells within ter-
atomas63. In pluripotent stem cells, polycomb genes act as repressors
of transcription factors that drive differentiation64. Compared with
normal hES cells, EC cells show two specific repressive polycomb
marks in their chromatin: dimethylated H3K9 and trimethylated
H3K9 (ref. 65). These epigenetic marks may function to inhibit dif-
ferentiation and are associated with hypermethylation of DNA and
inactivation of tumor suppressors in adult malignancies. Although
these observations were limited to one EC cell line derived from a
type II malignant GCT, similar epigenetic changes may be predictive
of malignant behavior in hES or hiPS cells. It is reported that EC cells
in type II GCTs have an intermediate DNA methylation pattern 66,67
related to activation of the de novo methyltransferase DNMT3L68.
No data are available so far regarding the type I GCT. In this context,
analysis of additional markers, including SSEA-5 in hES and hiPS
cells69, is of interest.
population and the presence of cells with genetic alterations related
to malignancy. Highly sensitive approaches to achieving the first two
of these aims are discussed below. High-throughput assays to monitor
genetic or epigenetic alterations across panels of oncogenes or tumor
suppressors, or losses or gains of genetic material, are available now
but are limited in their ability to pick up changes in minority cell
populations. There is a need to develop more- sensitive in vitro assays
of tumorigenic potential. These might include biological assays of
cell behavior, such as changes in self renewal, growth or cell survival,
or tests to measure expression of protein markers characteristic of
transformed cells.
In vivo assays are aimed at assessment of tumor formation in immuno­
compromised animals. These tests are limited by two factors. First, as
in assays of cancer stem cells70, insufficient sensitivity may underes-
timate tumorigenic potential (owing to residual innate or acquired
immunity in immunocompromised animals or possible requirements
for species-specific microenvironmental factors to support tumor
growth). Second, the relatively short life span of rodent test species
precludes long-term monitoring of grafts that may persist in patients
for decades. It is difficult to see how these limitations can be overcome
in routine practice, so the development of surrogate markers for cell
transformation is a priority.

854 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 perspective

Table 4  Methods for assaying residual pluripotent stem cells in clinical product or patient monitoring
Molecular Potential stage of application
Method of detection compartment Sensitivity Limitations Preclinical CMC Clinical
Methods with accepted clinical utility
ASO-qRT-PCR Tumor DNA 0.001%, that is, 1 CTC in Requires large number of samples for repeated Yes Yes Yes
100,000 normal83 testing to assure statistical certainty
Flow cytometry Tumor cell 0.01%, that is, 1 CTC in Requires four- to six-color flow, necessitating Yes Yes Yes
10,000 normal84 multiple cell surface markers
ELISAs Tumor protein Ultrasensitive assays detect Requires unique protein expression & correlation of Yes Yes Yes
in sub-pg/ml range85 protein signal with cell number
MRI Tumor cell Masses >0.3 cm Unknown effect of imaging labels on stem cell Yes No Yes
phenotype or genotoxic potential
FDG-PET Tumor size Masses >1 cm (ref. 86) Poor spatial resolution No No Yes

Methods with emerging evidence

qRT-PCR Tumor miRNA Limit of detection down to Requires identification of miRNAs with known Yes Yes Yes
10 copies of miRNA87 association with pluripotent cells
Immuno-PCR (TPA) Tumor protein Limit of detection in Requires unique protein expression & correlation of Yes Yes Yes
femtogram range88 protein signal with cell number
Fluorescent nanocrystals Tumor miRNA Limit   of detection in Requires identification of miRNAs with known Yes Yes Yes
& cation exchange ­femtomole range89 association with pluripotent cells
Nanoparticle surface Tumor miRNA Limit of detection in Requires identification of miRNAs with known Yes Yes Yes
© 2012 Nature America, Inc. All rights reserved.

plasmon resonance attomole range90 association with pluripotent cells

Bioluminescence (BLI) Tumor cell Limit of detection to be Requires demonstration that vectors used to label Yes No No
determined91,92 cells have no effect on cell product profile
CMC, product chemistry, manufacturing and controls; CTC, circulating tumor cell; ELISA, enzyme-linked immunosorbent assay; miRNA, microRNA; MRI, magnetic resonance
imaging; PET, positron emission tomography; TPA, TaqMan protein assay.

Thresholds in safety assessment. A key hurdle is determining the
threshold level of residual stem cells that poses a risk for teratoma
development. Because there is so much variation among protocols
for teratoma formation in the stem cell literature, estimates of the
threshold for teratoma formation range anywhere from 2 to >10,000
cells6,7,71. Results from xenograft data, even in immunocompromised
or humanized animals, would probably overestimate the true value.
The natural history of spontaneous type I teratomas suggests that this
threshold is more likely to be within the low end of the range, maybe
even <10 cells. As noted above, type I teratomas are thought to arise
either from epiblast cells that have failed to undergo differentiation
or from early primordial germ cells. Either of these cell populations
npg
­ roliferating cell populations such as hES or hiPS cells72, and whose
p
persistence in vivo may be limited. Serum biomarkers could provide
a more sensitive and simpler method for monitoring the safety and
predicting the risk of teratoma formation in preclinical studies and
in patients receiving hES cell–derived therapies, as they do in the
diagnosis and monitoring of cancer.
Certain genes that are expressed mainly during development and
not in adult tissues are often reexpressed by transformed cancer cells.
Although no definitive teratoma biomarker currently exists, proteins
such as alpha-fetoprotein, beta human chorionic gonadotropin and
placental alkaline phosphatase are clinically approved markers for
some histologically defined forms of GCT. The sensitivity and selectiv-

is quite small, and founder cells of the tumors likely represent only a
small subpopulation within them, as type I tumors are observed in
patients who had normal development of the soma and germ line. If
the founder population of type I GCTs is within the range of 10–100
cells, then detection of contamination in stem cell therapeutics must
be accurate in the range of 1 in 107 to 1 in 108 cells for applications
requiring an inoculum of 1 × 109 cells. Highly sensitive technologies
will be required to achieve such levels of detection (Table 4). Because
a spontaneous tumor is similar to an autograft, these considerations
would apply most closely to hiPS cell therapy. In hES cell therapy, the
use of allografts combined with immunosuppression further compli-
cates the assessment of threshold values.
Methods to evaluate and monitor pluripotent cell–derived trans-
plants. Despite the marked progress in imaging technologies, today’s
science falls short of readily visualizing a single or small cluster of
neoplastic cells in patients, a level of sensitivity that may be neces-
sary for monitoring stem cell transplants. Positron emission tomog-
raphy (PET) typically resolves masses ≥1 cm, a size that could contain
millions of cells. Tracking cells with magnetic resonance imaging
(MRI) requires labeling them with contrast agents whose physico-
chemical properties have an as yet unknown genotoxic potential for
ity of some of these marker assays are impressive; measurement of cir-
culating beta human chorionic gonadotropin in pregnancy can detect
the presence of 102 cells producing the hormone. Alpha-fetoprotein,
in particular, has become a useful diagnostic for the development and
recurrence of yolk sac elements in patients with type I teratomas52,73.
A plethora of secreted and cell-surface molecules74–76, transcription
factors77–79 and micro RNAs80,81 that are characteristic of pluripotent
stem cells or their early differentiated progeny have now been identi-
fied, and it is easy to envision that sensitive multiplex assays for such
markers could be used in preclinical or phase 1 safety assessment.
Such assays could be used to detect the presence of pluripotent stem
cells or lineage-committed progenitors with proliferative capacity and
neoplastic potential (Table 4). In addition to conventional immuno­
assays for serum proteins, new and highly sensitive assay technologies
that can exploit these opportunities are emerging. Furthermore, with
the advent of ultrasensitive methods for monitoring preneoplastic
mutations from circulating DNA in plasma samples, it would be
feasible to detect metastatic spread of a transplant.
CONCLUSIONS
Assessment of the risk of tumor formation by pluripotent stem cell–
based therapies is an essential part of safety evaluation. Unfortunately,

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 855

 Perspective
at present the basic scientific knowledge underlying efforts to develop
a sound regulatory framework in this field is rudimentary. Studies of
the origin, pathology, genetics and clinical management of terato-
mas in humans provide rich opportunities for reverse translational
research to drive future laboratory studies. Although parallels exist
between hES cells, hiPS cells and the EC cells of malignant human
teratomas that commonly occur in adolescents or young adults, the
tumors that arise from cultured pluripotent stem cells have more in
common with benign (type I) GCTs of early childhood. The natural
history of these spontaneously occurring tumors provides guidelines
for preclinical safety assessment of hES or hiPS cell lines and for the
development of measures to monitor stem cell grafts in preclinical
studies and in phase 1 trials.
Acknowledgments
J.J.C. would like to thank T. Dervieux for critical review and feedback on the
manuscript. L.H.J.L. would like to thank J. Wolter Oosterhuis for the wonderful
discussions of the last decades, significantly contributing to the work performed.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Published online at http://www.nature.com/doifinder/10.1038/nbt.2329.
© 2012 Nature America, Inc. All rights reserved.
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
1. Laflamme, M.A. et al. Cardiomyocytes derived from human embryonic stem cells in
pro-survival factors enhance function of infarcted rat hearts. Nat. Biotechnol. 25,
1015–1024 (2007).
2. Lu, B. et al. Long-term safety and function of RPE from human embryonic stem cells
in preclinical models of macular degeneration. Stem Cells 27, 2126–2135 (2009).
3. Przyborski, S.A. Differentiation of human embryonic stem cells after transplantation
in immune-deficient mice. Stem Cells 23, 1242–1250 (2005).
4. Cooke, M.J., Stojkovic, M. & Przyborski, S.A. Growth of teratomas derived from
human pluripotent stem cells is influenced by the graft site. Stem Cells Dev. 15,
254–259 (2006).
5. Prokhorova, T.A. et al. Teratoma formation by human embryonic stem cells is site-
dependent and enhanced by the presence of Matrigel. Stem Cells Dev. 18, 47–54
(2009).
6. Hentze, H. et al. Teratoma formation by human embryonic stem cells: evaluation
of essential parameters for future safety studies. Stem Cell Res. (Amst.) 2, 198–210
(2009).
7. Lee, A.S. et al. Effects of cell number on teratoma formation by human embryonic
stem cells. Cell Cycle 8, 2608–2612 (2009).
8. Shih, C.C., Forman, S.J., Chu, P. & Slovak, M. Human embryonic stem cells are
prone to generate primitive, undifferentiated tumors in engrafted human fetal tissues
npg
in severe combined immunodeficient mice. Stem Cells Dev. 16, 893–902 (2007).
9. Kelly, P.N., Dakic, A., Adams, J.M., Nutt, S.L. & Strasser, A. Tumor growth need
not be driven by rare cancer stem cells. Science 317, 337 (2007).
10. Quintana, E. et al. Efficient tumour formation by single human melanoma cells.
Nature 456, 593–598 (2008).
11. Gertow, K. et al. Early events in xenograft development from the human embryonic
stem cell line HS181–resemblance with an initial multiple epiblast formation.
PLoS ONE 6, e27741 (2011).
12. Griscelli, F. et al. Malignant germ cell-like tumors, expressing Ki-1 antigen (CD30),
are revealed during in vivo differentiation of partially reprogrammed human-induced
pluripotent stem cells. Am. J. Pathol. 180, 2084–2096 (2012).
13. Oosterhuis, J.W. & Looijenga, L.H. Testicular germ-cell tumours in a broader
perspective. Nat. Rev. Cancer 5, 210–222 (2005).
14. Rossant, J. & Papaioannou, V.E. The relationship between embryonic, embryonal
carcinoma and embryo-derived stem cells. Cell Differ. 15, 155–161 (1984).
15. Fu, W. et al. Residual undifferentiated cells during differentiation of induced
pluripotent stem cells in vitro and in vivo. Stem Cells Dev. 21, 521–529
(2011).
16. Baker, D.E. et al. Adaptation to culture of human embryonic stem cells and
oncogenesis in vivo. Nat. Biotechnol. 25, 207–215 (2007).
17. Harrison, N.J., Baker, D. & Andrews, P.W. Culture adaptation of embryonic stem
cells echoes germ cell malignancy. Int. J. Androl. 30, 275–281 (2007).
18. Fazeli, A. et al. Altered patterns of differentiation in karyotypically abnormal human
embryonic stem cells. Int. J. Dev. Biol. 55, 175–180 (2011).
19. Tavakoli, T. et al. Self-renewal and differentiation capabilities are variable between
human embryonic stem cell lines I3, I6 and BG01V. BMC Cell Biol. 10, 44
(2009).
20. Moon, S.H. et al. Effect of chromosome instability on the maintenance and
differentiation of human embryonic stem cells in vitro and in vivo. Stem Cell Res. 6,
50–59 (2011).
856 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology
21. Adewumi, O. et al. Characterization of human embryonic stem cell lines by the
International Stem Cell Initiative. Nat. Biotechnol. 25, 803–816 (2007).
22. Andrews, P.W. et al. Embryonic stem (ES) cells and embryonal carcinoma (EC)
cells: opposite sides of the same coin. Biochem. Soc. Trans. 33, 1526–1530
(2005).
23. Akst, J. Cysts stall stem cell trial. Scientist (Aug). 27 (2009).
24. Roy, N.S. et al. Functional engraftment of human ES cell-derived dopaminergic
neurons enriched by coculture with telomerase-immortalized midbrain astrocytes.
Nat. Med. 12, 1259–1268 (2006).
25. De Backer, A., Madern, G.C., van de Ven, C.P., Tibboel, D. & Hazebroek, F.W.
Strategy for management of newborns with cervical teratoma. J. Perinat. Med. 32,
500–508 (2004).
26. De Backer, A. et al. Study of the factors associated with recurrence in children
with sacrococcygeal teratoma. J. Pediatr. Surg. 41, 173–181 (2006).
27. Oosterhuis, J.W., Looijenga, H.J., van Echten, J. & de Jong, B. Chromosomal
constitution and developmental potential of human germ cell tumors and teratomas.
Cancer Genet. Cytogenet. 95, 96–102 (1997).
28. Simmonds, P.D. et al. Primary pure teratoma of the testis. J. Urol. 155, 939–942
(1996).
29. de Graaff, W.E. et al. Ploidy of testicular carcinoma in situ. Lab. Invest. 66,
166–168 (1992).
30. Gobel, U. et al. Germ-cell tumors in childhood and adolescence. GPOH MAKEI and
the MAHO study groups. Ann. Oncol. 11, 263–271 (2000).
31. Runyan, C., Gu, Y., Shoemaker, A., Looijenga, L. & Wylie, C. The distribution and
behavior of extragonadal primordial germ cells in Bax mutant mice suggest a novel
origin for sacrococcygeal germ cell tumors. Int. J. Dev. Biol. 52, 333–344
(2008).
32. Vortmeyer, A.O. et al. Microdissection-based analysis of mature ovarian teratoma.
Am. J. Pathol. 154, 987–991 (1999).
33. Gertow, K. et al. Organized development from human embryonic stem cells after
injection into immunodeficient mice. Stem Cells Dev. 13, 421–435 (2004).
34. Blum, B. & Benvenisty, N. Clonal analysis of human embryonic stem cell
differentiation into teratomas. Stem Cells 25, 1924–1930 (2007).
35. Busch, C., Oppitz, M., Wehrmann, M., Schweizer, P. & Drews, U. Immunohistochemical
localization of Nanog and Oct4 in stem cell compartments of human sacrococcygeal
teratomas. Histopathology 52, 717–730 (2008).
36. Busch, C. et al. Isolation of three stem cell lines from human sacrococcygeal
teratomas. J. Pathol. 217, 589–596 (2009).
37. Rijlaarsdam, M.A. et al. Specific detection of OCT3/4 isoform A/B/B1 expression
in solid (germ cell) tumours and cell lines: confirmation of OCT3/4 specificity for
germ cell tumours. Br. J. Cancer 105, 854–863 (2011).
38. Zwaka, T.P. & Thomson, J.A. A germ cell origin of embryonic stem cells? Development
132, 227–233 (2005).
39. West, J.A., Park, I.H., Daley, G.Q. & Geijsen, N. In vitro generation of germ cells
from murine embryonic stem cells. Nat. Protoc. 1, 2026–2036 (2006).
40. Chu, L.F., Surani, M.A., Jaenisch, R. & Zwaka, T.P. Blimp1 expression predicts
embryonic stem cell development in vitro. Curr. Biol. 21, 1759–1765
(2011).
41. O’Rahilly, R. & Muller, F. Neurulation in the normal human embryo. Ciba Found.
Symp. 181, 70–82 (1994).
42. Sano, K. Pathogenesis of intracranial germ cell tumors reconsidered. J. Neurosurg.
90, 258–264 (1999).
43. Oosterhuis, J.W., Stoop, H., Honecker, F. & Looijenga, L.H. Why human extragonadal
germ cell tumours occur in the midline of the body: old concepts, new perspectives.
Int. J. Androl. 30, 256–264 (2007).
44. Blum, B. & Benvenisty, N. The tumorigenicity of human embryonic stem cells.
Adv. Cancer Res. 100, 133–158 (2008).
45. Heifetz, S.A. et al. Immature teratomas in children: pathologic considerations: a
report from the combined Pediatric Oncology Group/Children’s Cancer Group.
Am. J. Surg. Pathol. 22, 1115–1124 (1998).
46. Veltman, I.M., Schepens, M.T., Looijenga, L.H., Strong, L.C. & van Kessel, A.G.
Germ cell tumours in neonates and infants: a distinct subgroup? APMIS 111,
152–160 (2003).
47. Spits, C. et al. Recurrent chromosomal abnormalities in human embryonic stem
cells. Nat. Biotechnol. 26, 1361–1363 (2008).
48. Golas, M.M., Gunawan, B., Raab, B.W., Fuzesi, L. & Lange, B. Malignant
transformation of an untreated congenital sacrococcygeal teratoma: a amplification
at 8q and 12p detected by comparative genomic hybridization. Cancer Genet.
Cytogenet. 197, 95–98 (2010).
49. Hong, H. et al. Suppression of induced pluripotent stem cell generation by the
p53-p21 pathway. Nature 460, 1132–1135 (2009).
50. Norris, H.J., Zirkin, H.J. & Benson, W.L. Immature (malignant) teratoma of the
ovary: a clinical and pathologic study of 58 cases. Cancer 37, 2359–2372
(1976).
51. O’Connor, D.M. & Norris, H.J. The influence of grade on the outcome of stage I
ovarian immature (malignant) teratomas and the reproducibility of grading. Int. J.
Gynecol. Pathol. 13, 283–289 (1994).
52. Gonzalez-Crussi, F., Winkler, R.F. & Mirkin, D.L. Sacrococcygeal teratomas in infants
and children: relationship of histology and prognosis in 40 cases. Arch. Pathol.
Lab. Med. 102, 420–425 (1978).
53. Surti, U., Hoffner, L., Chakravarti, A. & Ferrell, R.E. Genetics and biology of human
ovarian teratomas. I. Cytogenetic analysis and mechanism of origin. Am. J. Hum.
Genet. 47, 635–643 (1990).

54. Baker, B.A. et al. DNA content of ovarian immature teratomas and malignant germ
cell tumors. Gynecol. Oncol. 71, 14–18 (1998).
55. Mostert, M. et al. Comparative genomic and in situ hybridization of germ cell tumors
of the infantile testis. Lab. Invest. 80, 1055–1064 (2000).
56. Schneider, D.T. et al. Genetic analysis of childhood germ cell tumors with
comparative genomic hybridization. Klin. Padiatr. 213, 204–211 (2001).
57. Schneider, D.T. et al. Genetic analysis of mediastinal nonseminomatous germ cell
tumors in children and adolescents. Genes Chromosom. Cancer 34, 115–125
(2002).
58. Schneider, D.T. et al. Molecular genetic analysis of central nervous system germ
cell tumors with comparative genomic hybridization. Mod. Pathol. 19, 864–873
(2006).
59. Voorhoeve, P.M. et al. A genetic screen implicates miRNA-372 and miRNA-373 as
oncogenes in testicular germ cell tumors. Cell 124, 1169–1181 (2006).
60. Herszfeld, D. et al. CD30 is a survival factor and a biomarker for transformed human
pluripotent stem cells. Nat. Biotechnol. 24, 351–357 (2006).
61. Werbowetski-Ogilvie, T.E. et al. Characterization of human embryonic stem cells
with features of neoplastic progression. Nat. Biotechnol. 27, 91–97 (2009).
62. Gopalakrishna-Pillai, S. & Iverson, L.E. Astrocytes derived from trisomic human
embryonic stem cells express markers of astrocytic cancer cells and premalignant
stem-like progenitors. BMC Med. Genomics 3, 12 (2010).
63. Furukawa, S. et al. Yolk sac tumor but not seminoma or teratoma is associated with
abnormal epigenetic reprogramming pathway and shows frequent hypermethylation
of various tumor suppressor genes. Cancer Sci. 100, 698–708 (2009).
64. Sauvageau, M. & Sauvageau, G. Polycomb group genes: keeping stem cell activity
in balance. PLoS Biol. 6, e113 (2008).
65. Ohm, J.E. et al. A stem cell-like chromatin pattern may predispose tumor suppressor
genes to DNA hypermethylation and heritable silencing. Nat. Genet. 39, 237–242
(2007).
© 2012 Nature America, Inc. All rights reserved.
66. Netto, G.J. et al. Global DNA hypomethylation in intratubular germ cell neoplasia
and seminoma, but not in nonseminomatous male germ cell tumors. Mod. Pathol.
21, 1337–1344 (2008).
67. Wermann, H. et al. Global DNA methylation in fetal human germ cells and germ
cell tumours: association with differentiation and cisplatin resistance. J. Pathol.
221, 433–442 (2010).
68. Minami, K. et al. DNMT3L is a novel marker and is essential for the growth of
human embryonal carcinoma. Clin. Cancer Res. 16, 2751–2759 (2010).
69. Tang, C. et al. An antibody against SSEA-5 glycan on human pluripotent stem
cells enables removal of teratoma-forming cells. Nat. Biotechnol. 29, 829–834
(2011).
70. Visvader, J.E. & Lindeman, G.J. Cancer stem cells in solid tumours: accumulating
evidence and unresolved questions. Nat. Rev. Cancer 8, 755–768 (2008).
71. Lawrenz, B. et al. Highly sensitive biosafety model for stem-cell-derived grafts.
Cytotherapy 6, 212–222 (2004).
72. Crabbe, A. et al. Effects of MRI contrast agents on the stem cell phenotype.
Cell Transplant. 19, 919–936 (2010).
73. Havranek, P. et al. Sacrococcygeal teratoma in Sweden: a 10-year national
retrospective study. J. Pediatr. Surg. 27, 1447–1450 (1992).
npg
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 857
perspective
74. Lebkowski, J.S. et al. Human embryonic stem cells: culture, differentiation, and
genetic modification for regenerative medicine applications. Cancer J. 7 (suppl. 2),
S83–S93 (2001).
75. Kelly, O.G. et al. Cell-surface markers for the isolation of pancreatic cell types derived
from human embryonic stem cells. Nat. Biotechnol. 29, 750–756 (2011).
76. Roche, S. et al. Comparative analysis of protein expression of three stem cell
populations: models of cytokine delivery system in vivo. Int. J. Pharm. advance
online publication, doi: 10.1016/j.ijpharm.2011.12.041 (20 January 2012).
77. Chambers, I. et al. Functional expression cloning of Nanog, a pluripotency sustaining
factor in embryonic stem cells. Cell 113, 643–655 (2003).
78. Boyer, L.A. et al. Core transcriptional regulatory circuitry in human embryonic stem
cells. Cell 122, 947–956 (2005).
79. Assou, S. et al. A meta-analysis of human embryonic stem cells transcriptome
integrated into a web-based expression atlas. Stem Cells 25, 961–973
(2007).
80. Suh, M.R. et al. Human embryonic stem cells express a unique set of microRNAs.
Dev. Biol. 270, 488–498 (2004).
81. Morin, R.D. et al. Application of massively parallel sequencing to microRNA profiling
and discovery in human embryonic stem cells. Genome Res. 18, 610–621
(2008).
82. Curado, M.P. et al. Cancer Incidence in Five Continents, Vol IX (IARC Scientific
Publications No. 160, Lyon, France, 2007).
83. Sarasquete, M.E. et al. Minimal residual disease monitoring in multiple myeloma:
a comparison between allelic-specific oligonucleotide real-time quantitative
polymerase chain reaction and flow cytometry. Haematologica 90, 1365–1372
(2005).
84. Mason, K.D. & Juneja, S. Go with the flow for monitoring response in myeloma
with minimal residual disease. Leuk. Lymphoma 49, 177–178 (2008).
85. St. Ledger, K. et al. Analytical validation of a highly sensitive microparticle-based
immunoassay for the quantitation of IL-13 in human serum using the Erenna
immunoassay system. J. Immunol. Methods 350, 161–170 (2009).
86. Eriksson, L. et al. An investigation of sensitivity limits in PET scanners.
Nucl. Instrum. Methods Phys. Res. A 580, 836–842 (2007).
87. Schmittgen, T.D. et al. Real-time PCR quantification of precursor and mature
microRNA. Methods 44, 31–38 (2008).
88. Gillis, A.J. et al. Expression and interdependencies of pluripotency factors LIN28,
OCT3/4, NANOG and SOX2 in human testicular germ cells and tumours of the
testis. Int. J. Androl. 34, e160–e174 (2011).
89. Li, J., Schachermeyer, S., Wang, Y., Yin, Y. & Zhong, W. Detection of microRNA by
fluorescence amplification based on cation-exchange in nanocrystals. Anal. Chem.
81, 9723–9729 (2009).
90. Zhou, W.J., Chen, Y. & Corn, R.M. Ultrasensitive microarray detection of short RNA
sequences with enzymatically modified nanoparticles and surface plasmon resonance
imaging measurements. Anal. Chem. 83, 3897–3902 (2011).
91. Cao, F. et al. Noninvasive de novo imaging of human embryonic stem cell-derived
teratoma formation. Cancer Res. 69, 2709–2713 (2009).
92. Su, W. et al. Bioluminescence reporter gene imaging characterize human embryonic
stem cell-derived teratoma formation. J. Cell. Biochem. 112, 840–848 (2011).
 Articles

Multiplexed mass cytometry profiling of cellular states

perturbed by small-molecule regulators
Bernd Bodenmiller1,5,6, Eli R Zunder1,6, Rachel Finck1,6, Tiffany J Chen1–3, Erica S Savig1,4,
Robert V Bruggner1,2, Erin F Simonds1, Sean C Bendall1, Karen Sachs1, Peter O Krutzik1 & Garry P Nolan1

Mass cytometry facilitates high-dimensional, quantitative analysis of the effects of bioactive molecules on human samples at
single-cell resolution, but instruments process only one sample at a time. Here we describe mass-tag cellular barcoding (MCB),
which increases mass cytometry throughput by using n metal ion tags to multiplex up to 2n samples. We used seven tags to
multiplex an entire 96-well plate, and applied MCB to characterize human peripheral blood mononuclear cell (PBMC) signaling
© 2012 Nature America, Inc. All rights reserved.

dynamics and cell-to-cell communication, signaling variability between PBMCs from eight human donors, and the effects of 27
inhibitors on this system. For each inhibitor, we measured 14 phosphorylation sites in 14 PBMC types at 96 conditions, resulting
in 18,816 quantified phosphorylation levels from each multiplexed sample. This high-dimensional, systems-level inquiry allowed
analysis across cell-type and signaling space, reclassified inhibitors and revealed off-target effects. High-content, high-throughput
screening with MCB should be useful for drug discovery, preclinical testing and mechanistic investigation of human disease.

High-throughput in vitro screening has accelerated the discovery of
drug candidates, but paradoxically coincides with a steep decline in
the approval rate for novel molecular entities1,2. The enormous rate
of attrition as drug candidates move through clinical development
can be partly attributed to the disconnect between human physiology
and the in vitro screening regimen, which cannot measure efficacy in
heterogeneous tissues or detect off-target toxicities2–4. If the original
screening regimen more closely reflected human physiology by
using human samples, such as PBMCs or cancer biopsies, efficacy
and toxicity could be identified earlier in the development process.
High-dimensional analysis of cellular signaling networks can pro-
vide a detailed representation of cellular state5,6; it is often presumed
npg
providing a comprehensive view of the cellular state. Third, inter-
cellular communication and emergent systems properties should be
evaluated. And, lastly, measurements should be performed with high
throughput. A screening approach with these features would enable
the identification of compounds with in vivo efficacy against the tar-
geted disease and low toxicity at the earliest stage of drug discovery.
Some methods have attempted to implement these features. Parallel
enzymatic or phage display assays offer exceptional in vitro selectivity
profiling13–17, but do not provide in vivo data. Cellular assays based
on proliferation, apoptosis or expression of reporter proteins approxi-
mate in vivo activity18, but drug selectivity, mechanism of action and

signaling network responses cannot be determined. Gene expression

that additional biologically informative measurements of markers of
pathways would be a desirable outcome for high-throughput screen-
ing. Compounds that target certain signaling molecules can lead to
successful therapeutic outcomes7, but many compounds that target
known oncogenic lesions lack clinical efficacy8. As such, the in vitro
targets of a drug candidate cannot be used to accurately predict effi-
cacy in vivo owing to signaling network complexity and differences
between patients or cell subpopulations of the same patient 6,7,9–12.
Therefore, high-content, single-cell analysis of signaling networks
in human samples during drug development could provide welcome
insight into the manifold effects of drugs on cellular systems.
We propose that an ideal drug-screening approach should, first,
be based on primary human samples, with systemic behavior that
resembles normal physiology and the targeted disease state. Second,
subpopulation-specific, system-wide signaling networks and their
correlation to cell and disease phenotypes should be quantified,
analysis19,20 and liquid chromatography coupled to tandem mass
spectrometry6,21,22 measure thousands of parameters, but lack high
throughput and single-cell resolution23,24. High-throughput micro-
scopy offers deep characterization of single cells23–25, but the limited
number of surface and signaling molecules measured simultaneously
restricts the breadth of analysis.
Fluorescence-based flow cytometry (FBFC) permits measurement
of up to 12 molecules on a single cell simultaneously26–28, allowing
cell subpopulations and their signaling network states to be deter-
mined at the same time29. Drug-screening applications for FBFC
have been implemented by hardware30,31 or by sample multiplexing
with fluorescent cell barcoding (FCB)32,33. With these adaptations,
FBFC has become a powerful tool for drug screening and preclinical
ana­lysis. FBFC falls short of the ideal drug screening method, how-
ever, because the number of simultaneously measured parameters is
limited owing to spectral overlap27, hampering the comprehensive

1Baxter Laboratory in Stem Cell Biology, Department of Microbiology and Immunology, Stanford University, Stanford, California, USA. 2Biomedical Informatics

Program, Stanford University, Stanford, California, USA. 3Department of Computer Science, Stanford University, Stanford, California, USA. 4Cancer Biology Program,
Stanford University, Stanford, California, USA. 5Present address: Institute of Molecular Life Sciences, University of Zurich, Zurich, Switzerland. 6These authors
contributed equally to this work. Correspondence should be addressed to G.P.N. (gnolan@stanford.edu).

Received 30 April; accepted 2 July; published online 19 August 2012; corrected online 23 August 2012; doi:10.1038/nbt.2317

858 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 Articles

Figure 1  Mass-tag cell barcoding. (a) Cells a O

C C
O O
C C
O

were covalently labeled with a bifunctional O N N O SH

SH
SH O N N O M
3+
3+
M
M
3+

compound, maleimido-mono-amide-DOTA M3+ + SH

M
3+
3+ 3+
O N N O SH O N N O M M 3+
SH M
(mDOTA). This compound can be loaded C O SH
SH
C O M
3+
M
3+
O HN O HN S
with a lanthanide(III) isotope ion, and reacts N HS N

covalently with cellular thiol groups through O O

the maleimide moiety. (b) Seven unique

lanthanide isotopes were used to generate b c 10 3
BC1+
128 combinations, enough to barcode each
sample in a 96-well plate. The seven lanthanide La

BC channel 1
BC1+
139 102 BC2+
isotopes, their masses and their locations on Well no. 94
Pr Barcoding Analyte
the 96-well plate are shown. (c) A density dot masses masses 101
141 105
plot of barcoded cells is shown with the y-axis
100
and x-axis plot showing barcoding (BC) channel 1 Nd
146 104 BC
–
BC2+
(lanthanum 139) versus barcoding channel 2
0 1 2 3
(praseodymium 141). Cells positive and Tb 103
10 10 10 10
BC channel 2

Cts
negative for a given channel are indicated. 159
(d) K562 cells were stimulated with orthovanadate, Ho
102 d
placed in a 96-well plate as geometrical patterns 165 1
10
(checkerboard or striped pattern), barcoded,
Tm
analyzed by mass cytometry and subsequently 169 10
0
SLP76-pTyr696
deconvoluted using Boolean gating to validate the m/z
checkerboard
accuracy of the de-barcoding. The two resulting Lu
175
heatmaps of the measured SLP76-Tyr696
© 2012 Nature America, Inc. All rights reserved.

1 • • • • • • • • 128
phosphorylation levels are shown. Combinations
SLP76-pTyr696
analysis of signaling network states within stripes

complex cell populations. Min. Max.

A recent advance in flow cytometry—mass

cytometry—increases the number of parameters that can be mea­
sured, reduces overlap between measurement channels and eliminates
background autofluorescence34,35. For mass cytometry, antibodies are
labeled with isotopically pure metals36 and quantified by inductively
coupled plasma mass spectrometry. Current labeling techniques allow
for 34 parameter measurements35. The large number of parallel mea­
surements per cell makes mass cytometry an ideal method to assay
drug candidates for cellular efficacy and selectivity.
To bring mass cytometry closer to the ideal screening approach, we
have developed MCB, a cell-based multiplexing technique analogous
to FCB, which improves sample throughput, reduces antibody con-
sumption and ensures uniformity of the antibody stain across samples.
In MCB, cells in each sample are tagged with a unique combination
npg
purity are available, and their +3 oxidation state allows chelation into
cell-labeling reagents. For MCB, we used the bifunctional molecule
maleimido-mono-amide-DOTA (mDOTA) for chelation (Fig. 1a).
The DOTA moiety chelates rare earth metal lanthanide(III) ions with
a Kd of ~10−16 and the maleimide moiety rapidly reacts covalently
with cellular thiol groups (Fig. 1a and Supplementary Fig. 1). Using
binary combinations of seven preloaded mDOTA-lanthanide(III)
reagents, 128 (27) combinations are possible, enabling all of the wells
in a 96-well plate to be multiplexed in a single reaction (Fig. 1b).
The MCB protocol is experimentally similar to the FCB protocol
(Supplementary Note 1)32,33, and after measurement, the data set
is deconvolved with Boolean gating on the mDOTA-lanthanide(III)
channels (Fig. 1c).

of mass tags before samples are combined. We arrayed samples in a
96-well format, which we then multiplexed and analyzed on one run
through a mass cytometer. We used MCB to study PBMC signaling
dynamics and cell-to-cell communication, to measure the variability
of PBMC signaling responses between eight human donors and to
define the effects of 27 kinase inhibitors on 14 PBMC subpopula-
tions. In contrast to conventional approaches, which may use a single
molecular readout of a signaling protein and pathway, we measured
the concentration of 14 signaling proteins. This large number of
simultaneously measured parameters enabled the context-specific
classification of inhibitors and cell types. This analysis revealed that
none of the compounds tested was specific for a single cell type, that
inhibitor activity and selectivity were strongly dependent on context,
and that the established topology of hierarchical relationships among
PBMC cell types could be recapitulated based on signaling network
responses alone.
RESULTS
Mass-tag cellular multiplexing
For mass cytometry the lanthanide series of transition metal
elements is used as they are normally not present in biological
samples, a large number of stable isotopes that can be enriched to high
To test the accuracy and robustness of the MCB method, we used
two reference cell samples differing in the abundance of phosphory­
lation on Tyr696 on the SH2 domain–containing 76-kDa leukocyte
protein (SLP76) (Supplementary Fig. 2). We arranged the samples in
checkerboard or striped patterns on 96-well plates for MCB analysis.
After 96-well multiplexing, mass cytometry analysis of Tyr696 phos-
phorylation, and deconvolution, we were able to accurately recover
the patterns (Fig. 1d). The resulting Z-prime values (a measure of the
quality for an assay for high-throughput screening approaches) from
the checkerboard and striped 96-well plates were 0.61 and 0.58, respec-
tively, indicating that MCB is suitably robust for high-throughput
drug screening applications.
Time-course analysis of PBMC signaling
To confirm that MCB allows detection of physiological signaling
events in PBMCs and to assess subpopulation-specific signal­ing
dynamics, we used MCB (Supplementary Fig. 3) to carry out
a 96-sample time-course experiment from 0 to 4 h (Fig. 2a and
Supplementary Results 1). Fourteen signaling nodes and 10 cell-
surface markers were measured over 12 stimulation conditions
from one donor sample (Fig. 2a,b). To identify PBMC populations
(Fig. 2c) and their signaling response, we applied spanning-tree

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 859

 Articles

a d e

BCR/FcR-XL

PMA/Iono.
Reference
IFN-α LPS LPS LPS

GM-CSF
pNF�B

G-CSF
pSTAT1 pSTAT3 pBTK/pITK

IFN-α
pVO4

IFN-γ
IL-12

LPS
IL-3
IL-2 0 min
0 min
1 min
8 time points

5 min
15 min
30 min
60 min
1 min
120 min
240 min
12 conditions

b 5 min

CD20 IgM CD4

15 min
CD33 HLA-DR CD14

CD7 CD3 CD123 30 min

c Dendritic cells CD8+ T cells

60 min
© 2012 Nature America, Inc. All rights reserved.

NK cells
IgM+ B cells 120 min
IgM– B cells
high
HLA-DR
HLA-DRMed
CD14+
HLA-DRlow monocytes CD4+ T cells
240 min
HLA-DRhigh
CD14–
HLA-DRMed
monocytes %
HLA-DRlow
Maximum
Not shown: CD14+ Surfaceneg. & CD14– Surfaceneg.
0 100

Figure 2  PBMC signaling time-course experiment. (a) Twelve conditions and eight different time points were used to capture time-resolved PBMC
signal transduction from 0 to 240 min. (b) The expression and localization of cell surface markers within the SPADE tree is shown. (c) Fourteen
unique PBMC cell types were distinguished by SPADE analysis based on surface marker expression shown in b. (d) The time-resolved response of
the PBMC continuum of subpopulations to IFN-α stimulation by STAT1 phosphorylation, as visualized by SPADE. (e) Time-resolved response of the
PBMC continuum of subpopulations to LPS stimulation by NFκB, STAT3 and BTK/ITK phosphorylation, as visualized by SPADE. Putative intercellular
communication is indicated by black arrows.

progression analysis of density-normalized events (SPADE) 35,36 phosphorylation occurred in T cells and NK cells after 2 h and STAT1
(Fig. 2b–e and Supplementary Note 2). phosphorylation in B cells after 4 h (Fig. 2e and Supplementary
After interferon-alpha (IFN-α) stimulation, STAT1, STAT3 Fig. 9), which is likely attributable to intercellular communication
npg

and STAT5 phosphorylation was induced in most cell types38–40
(Fig. 2c,d and Supplementary Fig. 4); induction peaked at 15 min
and then declined, although elevated STAT1 phosphorylation was
maintained for 4 h in B cells and natural killer (NK) cells (Fig. 2d).
Unlike STAT3 and STAT5, prestimulation and IFN-α–induced phos-
phorylation levels of STAT1 varied widely among cell types, from a
twofold induction in CD14+ monocytes, to a fivefold induction in
other cell types (Fig. 2d). In IFN-α–stimulated T cells, STAT5 phos-
phorylation returned to prestimulation levels after initial activation,
but time-dependent differences in STAT5 induction were observed
in T-cell subtypes (Supplementary Fig. 5).
Time-course analysis by MCB also allowed identification of time-
dependent phenomena such as feedback regulation (Supplementary
Note 3 and Supplementary Figs. 6 and 7) and intercellular com-
munication. Monocytes, which express the lipopolysaccharide (LPS)
receptor toll-like receptor 4 (TLR4)41, responded first to LPS stimula-
tion, with phosphorylation of the canonical LPS pathway members
p38, ERK and NFκB peaking at 15–30 min, followed by S6 phospho-
rylation, which peaked after 2 h (Fig. 2e and Supplementary Fig. 8),
in agreement with previously reported results42. Cells with little
or no LPS receptor expression41, including B cells, T cells and NK
cells, responded to LPS at later time points. STAT3, STAT5 and ITK
through interleukin (IL-6) or other factors, such as TNF-α, which are
known to be released by monocytes after LPS stimulation43. These
results show that the MCB method can be used to identify novel,
dynamic signaling events and intercellular communication on the
network-scale level in complex, heterogeneous cell samples.
Comparison of signaling response in PBMCs from multiple donors
To assess variability in signaling between donors, we interrogated
eight PBMC samples using MCB (Figs. 1 and 3a). As in the ­previous
experiment, 14 signaling nodes and 10 cell-surface markers were
measured over 12 stimulation conditions (Fig. 3a and Supplementary
Fig. 6) and analyzed using SPADE39. Samples were collected 30 min
after stimulation, as the previous time-course experiment revealed
maximal signaling response at this point for most stimulus- and
­phosphorylation-site pairs.
The percentage of each cell type varied between the donor PBMC
samples (Supplementary Table 1). Monocytes ranged from 15%
(donor 7) to 26% (donor 3), T cells from 29% (donor 3) to 51%
(donor 2), and B cells ranged from 5% (donor 2) to 11% (donor 3).
A similar range of cell percentages was also visible for the cell sub-
types (Supplementary Table 1). The relative expression levels of the
surface markers used for immune phenotyping were similar across

860 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 Articles

Figure 3  Signaling response comparison of a b c d f

PBMCs from eight donors. (a) Twelve conditions CD3 CD33 BCR/FcR-XL IFN-α

BCR/FcR-XL
were used to compare signaling response of expression expression pS6 pSTAT5

PMA/Iono.
Reference

GM-CSF
PBMCs from eight different donors after 30-min

G-CSF

IFN-α
pVO4

IFN-γ
IL-12

LPS
IL-3
IL-2
stimulation. (b) The expression of the CD3 cell Donor 1
Donor 1
surface marker within the SPADE tree for all Donor 2

8 donors
Donor 3
donors is shown. (c) The expression of the CD33 Donor 4
Donor 5
Donor 2
cell surface markers within the SPADE tree for Donor 6
Donor 7
all donors. (d) Comparison of the response to Donor 8
Donor 3
12 conditions
30 min BCR/FcR-XL stimulation of the PBMC
continua of subpopulations of the analyzed Donor 4
donors as visualized by SPADE shown by the e Signaling state correlation 1
median phosphorylation levels of S6 protein. 0.9
2 Donor 5
(e) Correlation plot of the fold-change induction 0.8
over all stimuli, phosphorylation site and cell 3 0.7
0.6

Donor
4 Donor 6
type pairs between donors after 30-min 0.5
5
stimulation. (f) As d, but the median of 0.4
6 0.3 Donor 7
phosphorylation on STAT5 after 30 min IFN-α 7 0.2
stimulation is shown. 8 0.1
0 Donor 8
1 2 3 4 5 6 7
Donor %
the donor samples (e.g., CD3, Fig. 3b), except Reference Maximum

for CD33, with donors 3 and 4 showing lower 0 100

expression (Fig. 3c)44.

© 2012 Nature America, Inc. All rights reserved.

Despite differences in cell type abundances, cell signaling in each
cell population was similar across the eight donor samples, includ-
ing S6 phosphorylation after BCR/FcR-XL induction (Fig. 3d and
Supplementary Results 2). Systematic evaluation of signaling
response similarity between donors revealed a high correlation of
fold-change induction for each stimulus, phosphorylation site and
cell type combination between donors (Fig. 3e), ranging from 0.67
(donor 4 versus donor 6) to 0.93 (donor 7 versus donor 8) (Fig. 3e).
Exceptions existed. Contrary to all other donors, phosphorylation
on STAT5 and STAT3 was hardly induced in T cells after INF-α
stimulation in donor 6 (Fig. 3f and Supplementary Fig. 10),
but phosphorylation on STAT1 was induced (Supplementary
Fig. 10), and monocytes of the same donor showed induction of
phosphorylation on STAT3 and STAT5 (Supplementary Fig. 11).
These results show that cellular signaling was largely similar
between individual donor PBMC samples, even when the cell type
abundances varied.
npg
the data, we systematically organized the inhibitor IC50 values and
percent inhibition according by cell type and cell stimulus, and plotted
the data in a two-dimensional layout guided by canonical pathways
(Fig. 5 and Supplementary Results 4–6).
Inhibitor selectivity
To assess inhibitor selectivity, we compared the known targets of each
inhibitor (Supplementary Fig. 6) to its MCB-generated inhibition
finger­print (Fig. 5 and Supplementary Results 4–6). Kinase inhibitors
with a wide range of targets based on their in vitro inhibition profiles,
such as staurosporine13,14,16,17 (Fig. 5b,c, column 22) or the receptor
tyrosine kinase (RTK) inhibitor sunitinib (Sutent)14,16,17 (Fig. 5b,c,
column 24), reduced phosphorylation levels of at least one measured
signaling protein in all cell types under all conditions. In contrast,
other inhibitors showed more selectivity for cell type and stimulus.
Next, we analyzed the signaling network response patterns to deter-
mine the putative selectivity of inhibitors of the JAK-STAT pathway.
Cytokine receptors coupled to the JAK-STAT pathway activate a spe-

Systematic quantification of PBMC response to kinase inhibition
To systematically quantify PBMC response to kinase inhibition, we
applied 96-well MCB to PBMCs treated with an eight-step, fourfold
dose-response titration of 27 unique small-molecule inhibitors, and
used the resulting data to calculate the half-maximal inhibitory con-
centration (IC50) and percent inhibition (Fig. 4a). Twelve stimula-
tion conditions were used for 30 min to maximize signaling-space
coverage (Fig. 4b). Seven channels were used for MCB, 10 for cell-
surface marker quantification to resolve 14 cell types (Fig. 4c) and
14 to quantify protein phosphorylation sites (Fig. 4d), covering
important signaling pathways in all cell types for a network-wide
signaling map (Supplementary Fig. 6). We used a single PBMC donor
sample for all inhibitors to allow comparability, and we tested one
inhibitor, the JAK1/JAK2 inhibitor ruxolitinib (Jakafi), against four
donors to determine inhibitor response variability.
We quantified 18,816 phosphorylation site levels per inhibitor
(12 stimuli × 8 doses × 14 cell types × 14 phosphorylation sites), yield-
ing 2,352 dose-response titrations (14 cell types × 14 phosphorylation
sites × 12 stimuli) for a total of 63,504 dose titrations. The extracted
parameters of all dose response curves, including IC50, fold-change,
percent inhibition values, the corresponding confidence intervals and
Z-prime scores are given in Supplementary Results 3. To visualize
cific set of JAKs (of the four JAK-family kinases: JAKs 1–3, TYK2) upon
ligand binding, which in turn phosphorylate a defined subset of STAT
proteins (STAT1–6)38,39. Seven of the stimuli used (IFN-α, IFN-γ,
G-CSF, GM-CSF, IL-2, IL-3 and IL-12) induce previously reported38,39
JAK-STAT pairs (Supplementary Table 2), allowing the selectivity of
JAK inhibitors to be assessed. These inhibitors included ruxolitinib 45,
clinically approved for treatment of myelofibrosis; ­tofacitinib, a JAK3
inhibitor in phase 3 clinical testing against rheumatoid arthritis;
lestauritinib, a JAK2 and tyrosine kinase inhibitor entering a phase 1
clinical trial; and several research tool compounds, including JAK2
inhibitor III, JAK3 inhibitor VI and pan-JAK inhibitor I.
We observed reduced phosphorylation of various STATs after inhi-
bition with tofacinib compared to control, when we stimulated diverse
cell types with GM-CSF, IL-3, IL-2, G-CSF, IFN-α and IFN-γ, indicat-
ing that tofacinib is a pan-JAK inhibitor (Supplementary Fig. 12a).
Higher IC50 values on JAK2-dependent phosphorylation of STAT5
after IL-3 or GM-CSF stimulation compared with JAK1- and/or
JAK3-dependent phosphorylation after IL-2 or IFN-α stimulation
(Supplementary Fig. 12a,b) suggest that both JAK1 and JAK3 and
to a lesser extent JAK2 or TYK2 are inhibited by tofacinib. This is in
agreement with our in vitro kinase inhibition profile (Supplementary
Table 3), but differs slightly from published in vitro data16.

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 861

 Articles

a b

BCR/FcR-XL

PMA/Iono.
Reference

GM-CSF
Vanadate

G-CSF
ors

IFN-α
IFN-γ
IL-12
ibit

LPS
IL-3
IL-2
inh
27

Per inhibitor

Per well
max. phosphorylation
0 c
Ion count/cell

Percent inhibition
50% max. inhibition

max. inhibition

CD45
DNA
Basal
100 B cells
0
IC50
Un- IgM–
stimulated Cell length Cell length

CD20
[Inhibitor]

CD45
IgM+
d pSTAT1
10
pSlp76/pBLNK
100
pBtk/pItk
100
pPLCγ2 pErk
Surfaceneg.
1,000 CD3 IgM
10
10 100 CD14+
10 CD7

CD45

CD45

CD45

CD45
0 10
0 0

CD45
© 2012 Nature America, Inc. All rights reserved.

0 0 Natural
[µM] [µM] [µM] [µM] [µM]
killer cells
1.51

.0

0 3

0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
.0
0.0610

0.0610
0.0244
0.377
1.51
6.26
25 5
.0
0 3

6.26
0.0244

0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
.0

0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
.0
0.3 7
9

0.0015

0.0015

9
0.0015

0.0015

0.0 15

9
97

CD14–
25

9
0
0.0

0.0
0.0

0.0

0.0

pLat pS6 pNF�B p-p38 pSTAT5 CD14 CD4 IgM

10,000 100 CD7 CD4
100
1,000 10 Per CD14–
10
100 10 cell type

CD45

CD33

CD33
10
10
0 0 0 CD14+
0
−10
[µM] [µM] [µM] [µM] [µM]
CD45
0 3
0.0610
0.0 244

1.51
6.2 6

.0

0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
.0
0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
.0

.0

0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
.0
0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
255
0.377
0.0 15

0.0015

9
0.0015

0.0015

9
0.0015

9
9

9
9
9
0
0.0

0.0

0.0
0.0
0.0

CD14
pAkt pShp2 pZap70/pSyk pSTAT3 CD3 CD3
100 10
10 Low Mid High CD8+ T cells

CD45
CD45
10

CD45
10
0 0
HLA-DR Dendritic CD4+
0 0
Monocytes Low Mid High cells
[µM] [µM] [µM] [µM]
6.26
25 5
.0
1.51

6.26
25 5
.0

0 3
0.0610
0.0244
0 3
0.0610
0.0244
0.377
1.51
6.26
25 5
.0

0.377
1.51
6.26
5
.0
0.0244
0.377
0 3
0.0610

1.51
0.0244
0.377
0 3
0.0610
9

0.0015
0.0015

9
0.0015

9
0.0015

25

HLA-DR
9

9
9

0.0
0.0
0.0

0.0

CD123 CD4

Figure 4  Analysis of PBMC response to kinase inhibition. (a) The effect of 27 inhibitors on PBMC signaling was quantified by MCB, including the IC 50
value and percent inhibition of phosphorylation levels. (b) Experimental set-up for each inhibitor experiment. Twelve stimulation conditions were applied
for 30 min in conjunction with an eight-point, fourfold dilution series of each inhibitor. (c) Gating scheme. Ten cell surface markers were combined to
define 14 cell types. (d) For each cell type, 14 phosphorylation sites covering many immune signaling pathways were quantified by mass cytometry.
Examples of dose-response curves are shown for staurosporine treatment in CD4 + T cells.

Broad inhibitory effects on STAT phosphorylation after GM-CSF, it is bulkier than most ATP-competitive kinase inhibitors and it lacks
npg

IL-3, IL-2, G-CSF, IFN-α and IFN-γ stimulation were also observed for the critical H-bond donor and acceptor pair46.
ruxolitinib (Supplementary Fig. 13), lestauritinib (Supplementary We also analyzed inhibitors of the PI3K-AKT-mTOR-p70S6K
Fig. 14) and the pan-JAK inhibitor I (Supplementary Fig. 15), in agree- signal­ing pathway (Supplementary Note 6, Fig. 5b colored boxes
ment with our in vitro inhibition profile (Supplementary Table 3). and Supplementary Figs. 18–20). Taken together, these results show
Furthermore, lestauritinib and the pan-JAK inhibitor showed size- that MCB can be used to generate a cellular inhibitor ‘fingerprint’ and
able effects on signaling outside the JAK pathways (Supplementary to investigate target selectivity with unprecedented resolution and
Figs. 14 and 15), indicating that these inhibitors affected many signal- throughput in complex cellular mixtures.
ing network nodes.
Detailed inhibition profile analysis of JAK2 inhibitor III Cell type selectivity
(Supplementary Note 4 and Supplementary Fig. 16a,b) and JAK3 We next investigated the cell type selectivity for each inhibitor by
inhibitor VI (Supplementary Note 5 and Supplementary Fig. 17a,b) analyzing the signaling response data from 14 cell types. No inhibi-
indicated inhibition of TYK2 activity by JAK2 inhibitor III and inhibi- tor showed exclusive selectivity for a single cell type, and inhibitors
tion of JAK1/TYK2 activity by JAK3 inhibitor VI, rather than JAK2 with broad pathway activity, such as staurosporine and sunitinib,
inhibition by JAK2 inhibitor III and Jak1-JAK3 inhibition by JAK3 displayed little to no cell type selectivity (Fig. 5b,c, columns 22
inhibitor VI (Supplementary Notes 4,5 and Supplementary Results 5). and 24). In general, the inhibition profile of HLA-DRmid monocytes
Comparison of the JAK2 inhibitor III MCB results with the in vitro differed from those of other cell types (Supplementary Results 4).
kinase assay results were surprising (Supplementary Table 3). The Inhibitors of the Src family kinases (SFKs) and receptor tyrosine
JAK2 inhibitor III did not inhibit JAK-family kinases at concentra- kinase (RTKs)—dasatinib, LCK inhibitor and PP2—inhibited SFK
tions ≤10 µM. This discrepancy between in vitro and in vivo results downstream signaling components in monocytes compared to other
could be due to an allosteric mechanism of inhibition not recapitu- cell types, including SYK, PLCγ2 and BLNK, often independent of
lated in vitro, or additional off-target effects. The JAK2 inhibitor III stimulation (Supplementary Figs. 21 and 22). Many inhibitors in
structure suggests that it is not an ATP-competitive inhibitor, because addition to the JAK inhibitors and sunitinib affected JAK-STAT

862 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 Articles

­signaling in monocytes (Supplementary Figs. 21 and 22), independ- The data also enabled the comparative analysis of cell-signaling-
ent of stimulation conditions, indicating that under the conditions of network responses to inhibition in closely related cell types. Such
our assay, SFK and JAK-STAT signaling pathways are active in mono- responses differ only to a few compounds, including imatinib (Gleevec;
cytes, but inactive in T cells, B cells, dendritic cells and NK cells. Supplementary Note 7 and Supplementary Fig. 23), the c-Jun

a Phosphoprotein placement
SHP

No inhibitor Inhibitor
PLCγ
BTK ZAP70
SLP76
LAT T cells
STAT1
STAT3
STAT5
AKT
S6
ERK p38
NK cells Fold change EC50 Percent inhibition
NFκb
(vs. basal) <6.0*10
–10 ≥1,000
SHP –10
SYK PLCγ 5x 6.0*10 100
BTK BLNK –9 (Basal levels)
LAT B cells 4x 3.0*10

Drug potency
–8
AKT ERK p38 3x 1.5*10
Monocytes
STAT1
STAT3
STAT5 S6 NFκb
Unstim. 2x 7.7*10
–8

–7
1x 3.9*10
SHP 0x 2.0*10
–6
SYK PLCγ –1x –6
BTK –2x 9.9*10

STAT1 AKT
SLP76 LAT

ERK p38
Surface– –3x Inhibitor 5.0*10
–5

–5
STAT3
STAT5 S6 NFκb
Dendritic cells –4x >5.0*10
Non-sigmoidal
0 (No inhib.)
Induction
Stimulation –5x ≤–1,000
Response

b Inhibitor

b.

e
hi
IgM+
III

in

in

IV
in

b
b.

b.

Ru ycin

or

r
I
1

25
.

ni

SB nib

ig

ib
19
ib

b.
.

b
94

hi

hi

b.
ib

n)
as an .

ib

sp
al ge

83

B cells

rti

ib
. b ch ib

on

in
ni
2

01
ib

hi
in

in
in

tin

hi

02
pa

liti
h

am

6
1/

-0

tin
vsold nh

in

au

0
cit
fe

ro
)

69

in

in
in

pt
ss

in

12
60

68
K2

K3
sa

xo
T-

DC

-2
K(

ra

au

w
ni

fa
K

at
(F i

st

2
9

re
o-

k
a

UO
No

AK

PP

SP

VX
BT

Da

H8

Ra

Ro
So

Su
JA

JA

JA

Sy

To
Im

Le
Cr

Lc
IK

St

St
G

Unstim. 1
Stimulation (30 min)

BCR/FcR-XL 2
© 2012 Nature America, Inc. All rights reserved.

G-CSF 3
GM-CSF 4
IFN-α 5
IFN-γ 6
IL-2 7
IL-3 8
IL-12 9
LPS 10
PMA/Iono. 11
pVO4 12

Column 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

c Inhibitor
b.

e
IFN-α
hi
III

in

rin

IV
in

b
b.

b.

in

or
I
1

25
b.

ni

ig

ib
19

b.
b.

b
94

hi

hi

b.

yc
n)
ib
as an .

ni

sp
al ge

83

rti

ib
. b ch ib

on
hi

in
ni
2

01
ib

hi
hi

in

in
sin

tin

hi

02
pa

liti
m

6
1/

-0

tin
vsold nh

in

au

0
cit
ro
)

fe
69

in

in
in

pt
in

12

68
60
K2

K3

pa
sa

xo
T-

DC

-2
K(

w
as

ra

au

ni

fa
K

at
(F i

st

2
9

re
o-

k
K

UO
No

Ro
AK

PP

VX
SB

SP
BT

Da

H8

Ra

Ru

So

Su
JA

JA

JA

Sy

To
Im

Le
Cr

Lc

St
IK

St
G

CD14– 1
Surf-
Cell type

CD14+ 2
npg

–
CD14
low 3
HLA-DR –
CD14
4
Monocytes

mid
HLA-DR –
CD14
high 5
HLA-DR +
CD14
low 6
HLA-DR +
CD14 7
mid
HLA-DR +
CD14 8
high
HLA-DR
Dendritic 9
CD8+ 10
T cells

CD4+ 11
NK cells 12
IgM– 13
B cells

IgM+ 14
Column 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

Figure 5  Overview of inhibitor impact. (a) A miniaturized signaling network, guided by canonical pathways, including vertical ordering of nodes from
membrane-proximal signaling proteins to nucleus-localized transcription, is used here to depict the effect of a stimulus or inhibitor on each quantified
phosphorylation site after 15-min incubation with the inhibitor and subsequent 30-min cell stimulation. As some antibodies recognize different
proteins in different cell types, three cell type–specific signaling networks are shown. In the absence of inhibitor (“No inhibitor”), the response to each
stimulus relative to the untreated state is represented as fold change by a sized red or black circle (for induction and reduction of phosphorylation
levels, respectively). To visualize the effects of inhibitors (“Inhibitor”), circles were sized inversely to the IC 50 and colored by the percent inhibition
(‘inhibition’). For example, in the presence of ruxolitinib, inhibition of phosphorylation of STAT1 (IC 50 = 23 nM, 93% inhibition) and STAT3 (IC50 = 4 nM,
147% inhibition) was observed (a, “Inhibitor”), whereas without activation of the B cells, no observable effects of ruxolitinib on the quantified signaling
nodes were visible (b, yellow box). Fold-change induction before inhibition and confidence intervals for IC 50 values and percent inhibition are not
visualized, but are given in Supplementary Results 3. (b) The impact of all inhibitors under all stimulation conditions is shown for IgM + B cells.
(c) The impact of all inhibitors on all cell types after 30 min IFN-α stimulation is shown. Sections highlighted by color are detailed in the main text.

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 863

 Articles

PC 2 coordinate PC 1 coordinate Distance in PC space

Min. Max. Min. Max. Min. Max.

a CD14
+ CD14
– c Tofacitinib Staurosporine e
HLA-DR
mid Surface
neg.
Ruxolitinib 1.0 Akti Dasatinib
Monocytes JAK3 inhib.

MCB pairwise distances

Dendritic Sunitinib Dasatinib GO−69
CD14
–
IgM
+ cells 0.9 GO−69 H−89
mid
HLA-DR – Crassin Lck inhib. PP2
B cells IgM H−89 Imatinib
Monocytes Streptonigrin
B cells
JAK(pan) inhib.
0.8 Imatinib Jak1
–
CD14
high
Lestauritinib Jak1 Jak3
HLA-DR + Dasatinib 0.7 Jak3 vs. Lck
Monocytes CD8 VX680 JAK2 inhib.
T cells Lck Rapamycin
0.6 Rapamycin SB−202
+ + Syk inhib. IV SB−202 Sorafenib
CD14 CD14 +
low
HLA-DR
high
CD4 AKTi-1/2 SB-202190 0.5
HLA-DR NK cells Sorafenib Staurosporine
Monocytes Monocytes T cells H89
GDC-0941 Staurosporine Sunitinib
BTK inhib. III
CD14
–
SP600125 DMSO 0.4 Sunitinib Tofacitinib
low
HLA-DR + UO126 0.2 0.4 0.6 0.8 1.0
Monocytes CD14 Rapamycin
neg. Imatinib Go-6983
Surface
IKK inhib. I
Sorafenib Pairwise distances22

d f
+ low

b
CD14 HLA-DR
Monocytes Dasatinib
+ mid
Sunitinib 1.0
CD14 HLA-DR JAK(pan) inhib.
Monocytes PP2 Dasatinib GDC-0941

MCB pairwise distances

+ NK cells GDC-0941 0.9
CD4 –
IgM B cells GDC-0941 Imatinib
T cells Lck inhib.
– Ruxolitinib
CD14 0.8 Imatinib
–
CD14 HLA-DR
low
Surface
neg. Staurosporine Lestauritinib
UO126 Ruxolitinib Sorafenib
Monocytes + AKTi-1/2 vs.
CD8 SB- 0.7 Sorafenib Staurosporine
T cells
CD14
– Rapamycin 202190
high BTK inhib. III Staurosporine Sunitinib
HLA-DR Dendritic cells
IgM
+
JAK3 inhib. IKK inhib. I 0.6
Monocytes + Sunitinib VX680
CD14 B cells Tofacitinib
neg. Imatinib
Surface SP600125 0.5 VX680 Tofacitinib
Syk inhib. IV
© 2012 Nature America, Inc. All rights reserved.

– mid Go-6983 DMSO

CD14 HLA-DR Crassin
Monocytes
Sorafenib
0.4
+
CD14 HLA-DR
high
JAK2 inhib. VX680 Ruxolitinib Streptonigrin H89 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Monocytes
Pairwise distances23

Figure 6  Principal component analysis of cell type and drug response. (a) Cell type PCA across all inhibitors, phosphorylation sites and conditions.
(b) Cell type PCA for streptonigrin across all phosphorylation sites and conditions. (c) Inhibitor PCA across all cell types, phosphorylation sites, and
conditions. (d) Inhibitor PCA for monocytes after IFN-α stimulation across all phosphorylation sites. (e) Pairwise distance correlation plot to show
the agreement between in vivo data generated by MCB and previously generated 16 in vitro kinome inhibition profiles. Distances shown were scaled
as a fraction of the maximum distance. (f) As e, but pairwise distance correlation plot between in vivo data generated by MCB and a different set of
previously generated17 in vitro kinome inhibition profiles.

N-terminal kinase (JNK) inhibitor SP600125 (Supplementary Note 7 inhibitor for each cell type, stimulation and phosphorylation site.
and Supplementary Fig. 24) and the aminoquinone antitumor Likewise, to analyze inhibitor similarity, we created a matrix of IC50
antibiotic, streptonigrin. Streptonigrin induced differential signal- values representing cell type response for each inhibitor, stimulation
ing responses in monocyte subtypes (Supplementary Note 7 and and phosphorylation site. We performed principal component ana­
Supplementary Fig. 25a–d) and in T cells and B-cell subtypes on S6, lysis (PCA) on each of these matrices to ask which cell types or inhibi-
PLCγ2, SLP76/BLNK and STAT5, often independently of the stimu- tors were similar in the reduced-dimensionality space.
lation (Supplementary Fig. 25c,d, columns 10–14, yellow boxes). When the first matrix is used in the PCA to determine cell type
In CD8+ T cells and IgM+ B cells, streptonigrin often had low IC50 similarity, the cell types with related immune functions, such as the
npg

values and strongly inhibited phosphorylation, but we observed only lymphocytes, were grouped closely within principal component space,
a weak inhibition in CD4+ T cells and IgM– B cells (Supplementary as defined by all determined components (encompassing up to 90% of
Fig. 25c,d, columns 10–14, yellow boxes) on the same sites. An excep- the total variance), forming their own cluster (Fig. 6a). A second and
tion was when cells were stimulated with PMA-ionomycin; inhibi- third cluster were formed by the monocyte lineage, with closely related
tion of most signaling molecules was detected in most cell types monocyte subtypes being also closest in principal component space. We
(Supplementary Fig. 25d, row 11, gray and red boxes). Streptonigrin also carried out biclustering of data for all cell types (Supplementary
interferes with cell replication, transcription and cell respiration47, Fig. 26), which showed that at this level of signaling network resolu-
but how this might lead to the observed differences is unclear. An tion, the established endpoint topology of hematopoiesis41 in PBMCs
overview of the cell type profiles of each inhibitor tested is shown in was recapitulated based on signaling network response alone.
Supplementary Results 4. When the input matrix was restricted to the streptonigrin-induced
Overall, these results demonstrate that MCB can be used to reveal signaling responses (Fig. 6b), a distinct picture emerged. Closely related
how different cell types and their underlying signaling network states immune cell types were differentially affected, as CD8+ T cells, IgM+
are uniquely affected by given inhibitors, underscoring the need for B cells, CD14– surfaceneg and NK cells formed their own cluster in
deep signaling profiling for supporting the development of cell type- principal component space, distant from CD4+ T cells and IgM– B cells.
specific compounds. The PCA also captured the differences in inhibitor impact among
the monocyte subtypes compared to the PCA over all conditions, in
Systematic analysis of cell type and inhibitor similarity concordance with analysis from the “Cell type selectivity” section
We next sought to use the large number of signaling molecules quanti- (Supplementary Note 7 and Supplementary Fig. 25).
fied per cell type under many conditions to characterize the cell types In a second analysis, we transposed the matrix to ask which inhibi-
and inhibitors. We analyzed the complete data set and the effect of a tors clustered together in the reduced-dimensionality space defined by
drug in a single cell type and condition. To analyze cell type similarity, all cell type signaling states and conditions (Fig. 6c). The general kinase
we generated a matrix of IC50 values representing the effects of each inhibitors staurosporine, lestauritinib and streptonigrin and the SFK

864 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology


and RTK inhibitors sunitinib, PP2, dasatinib and LCK inhibitor formed
their own cluster in principal component space, reflecting their overall
high impact on signaling networks across cell types and conditions.
By restricting the input matrix to monocytes after IFN-α stimula-
tion across all inhibitors, we addressed the question of which inhibitors
similarly affect JAK-STAT signaling (Fig. 6d). JAK2 inhibitor III, JAK3
inhibitor VI, tofacinib, crassin, SP600125 (a JNK inhibitor) and VX680
inhibitor formed a tight group. As described above, the JAK inhibi-
tors inhibited JAK-STAT signaling after IFN-α stimulation in most
cell types. However, Crassin also displayed a similar profile (Fig. 5c,
row 4, pink box), reproducing a recent finding32. SP600125 was not
specific for JNK inhibition, but also inhibited phosphorylation of
STAT3 and STAT5 across many cell types (Fig. 5c, column 21, yel-
low box; Supplementary Fig. 24 and Supplementary Fig. 27). The
ability of SP600125 to inhibit JAK-STAT signaling was confirmed
by in vitro kinase inhibition assays (Supplementary Table 3). Here
IC50 values of 974 nM, 736 nM, 344 nM and 440 nM were measured
for JAK1, JAK2, JAK3 and TYK2, respectively. VX680, an inhibitor
of Aurora kinases, which is also active against BCR-ABL, FLT3 and
JAK2 (ref. 48), was close to SP600125 in principal component space.
Phosphorylation of STAT3 and STAT5 were inhibited in the presence
© 2012 Nature America, Inc. All rights reserved.
of VX680 after IFN-α stimulation in many cell types (Fig. 5c, column
28, blue box and Supplementary Fig. 28a,b, row 5, green box). This
suggests that VX680 inhibits JAK1 and, even more potently according
to the in vitro data, TYK2 (Supplementary Table 3). However, for
VX680, we observed no or only weak inhibition of phosphorylation
on STAT5 after GM-CSF or IL-3 stimulation, indicating the absence
of JAK2 inhibition, contrary to the in vitro data (Supplementary
Fig. 28b, red boxes).
These results show that PCA allows characterization and identifi­
cation of similar cell type responses to a given inhibitor and that
the inhibitor-induced signaling states were sometimes independent
of cell type and immune function, indicating enormous plasticity
in the cellular signaling network. In addition, PCA allowed rapid
­classification of inhibitors based on their profiles at a given drug
exposure or in a given experimental condition, and suggests novel
specificities for inhibitors SP600125 and VX680.
Comparison of inhibition response in PBMCs from multiple donors
npg
To establish whether the inhibition data sets generated from a single
PBMC donor are generalizable or if there is variability in inhibitor
response between donors, we measured the effects of ruxolitinib on
four of the eight donor samples previously described that best repre-
sent the variability between donors (Fig. 3). The response to inhibi-
tion between donors was similar overall, but also showed marked
differences (Supplementary Fig. 29). Whereas ruxolitinib inhibited
INF-α–stimulated phosphorylation on STAT1 on IgM+ B cells, IgM–
B cells and CD4+ T cells in all donors analyzed (Supplementary
Fig. 29, green boxes), the same site was only inhibited in two out
of four donors in CD8 + T cells (Supplementary Fig. 29, row 10,
red box). Similarly, G-CSF induced phosphorylation on STAT3 in
CD14– HLA-DRmid monocytes was inhibited in all donors except
donor 4 (Supplementary Fig. 29, row 4, blue box). Closer inspec-
tion of these differences in inhibitor response revealed they were
often due to inhibition curves that fall directly above or below the
R2–fold change cutoff used as a threshold for calling a site ‘inhib-
ited’ (Online Methods), and this was often compounded by differ-
ences in the level of pathway activation observed between donors
after stimulation (Fig. 3). We have observed such fluctuations in
human PBMCs particularly in cases of chronic diseases involving
inflammation (G.P.N., data not shown), indicating in part that the
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 865
Articles
differences observed might ­ indicate differing ‘set points’ in cell
subset–specific activation due to prior immune encounters.
Therefore, we conclude that the 27 state-based kinase inhibitor pro-
files are a comprehensive resource describing normal healthy immune
response to kinase inhibition. These results also underscore the need
to measure several donor samples if an inhibitor must be extensively
analyzed, for example, before a clinical trial.
Comparison of in vivo versus in vitro inhibition profiles
Finally, we examined the agreement between data generated by MCB
and previously published in vitro kinase assay data, including one
study with 14 compounds in common with this study16, and one study
with 9 compounds in common with this study17 (Supplementary
Results 7). We compared these data with the matrix of MCB-derived
IC50 values representing inhibitor impact for each cell type, stimula-
tion and phosphorylation. For the MCB data set and the two in vitro
data sets, we plotted the pairwise distances between the compounds
against each other (Fig. 6e,f).
In general, inhibitor relatedness as measured by pairwise distances
was similar in the in vitro and in vivo data sets, resulting in correlation
coefficients of 0.74 and 0.75, respectively. An exception was JAK3
inhibitor VI, for which the distance was greater to most inhibitors in
the in vitro data set compared with the in vivo data set. Nevertheless,
for most of the compounds analyzed, the in vitro inhibition profile
(e.g., Go-6983, rapamycin, GDC-0941, JAK inhibitors) is largely
in agreement with the MCB data, but caution is necessary because
depending upon the stimulation condition, the cell types and the
donor samples used for testing, these data may not be generalizable.
These in vivo and in vitro approaches when used together should
prove highly complementary for the mechanistic investigation of cel-
lular signaling pathways.
DISCUSSION
MCB makes possible high-throughput experiments that are
impractical to do using FBFC or mass cytometry alone. We used
MCB to analyze PBMC signaling dynamics, cell-to-cell communica-
tion and to comprehensively profile small-molecule drug regulators
based on PBMC signaling network response. In these experiments,
18,816 phosphorylation levels were quantified in 1,344 cell popula-
tions from 96 multiplexed samples for each inhibitor. By using n metal
isotopes for binary-encoded MCB, 2n samples can be multiplexed.
This allows >10,000 samples to be multiplexed in a single tube with
15 channels remaining for antibody detection. At this scale, MCB
becomes an attractive technique for high-throughput drug screening
and genome-wide RNA interference knockdown studies.
Despite several limitations (Supplementary Note 8), the approach
presented here allows analyses that span from the systems-level down
to single pathways and molecules. In the experiments described, high-
level compound classification suggested novel molecular targets and
indicated novel mechanisms of action for widely used kinase inhibi-
tors. The ability to identify bio-active compounds such as JAK2 inhib-
itor III, which presumably would not have been identified by in vitro
screening, highlights a key advantage to the in vivo MCB approach.
As MCB enables signaling events to be monitored over time, it pro-
vides an opportunity to study the connectivity of signaling pathways,
the effects of inhibitors on feedback signaling49, and inter­cellular
communication50. Time-resolved, single-cell analysis can reveal
differences between immediate and subsequent indirect, adaptive
effects caused by cross-talk between signaling pathways. Our results
indicate that data from in vivo MCB and in vitro kinome screen-
ing methods16,17 are complementary, suggesting that using both
 Articles
approaches is a potentially useful paradigm for the investigation of
pathway mechanism, connectivity and dynamics.
With high-throughput, systems-level coverage of the most impor-
tant signaling pathways and their interconnections for each cell type,
the cellular signaling states induced by inhibitors could be used as a
metric for preclinical development. Similar MCB analyses done on
defined disease samples could be used to categorize drug effects or
drug combinations, to eventually guide therapeutic strategies based
on discrete knowledge of a patient’s cellular phenotypes and geno-
types. Additionally, the MCB method could be used directly as a tool
for personalized medicine, with the pathway activation and drug
response of a patient’s in vivo or ex vivo tissue samples used to guide
therapy decisions.
Methods
Methods and any associated references are available in the online
version of the paper.
Data availability. All dose response curves can be viewed and all raw
data can be downloaded from http://www.cytobank.org/nolanlab/.
The determined IC50 values, fold changes, percent inhibition
© 2012 Nature America, Inc. All rights reserved.
values, confidence intervals and Z-prime scores are available in
Supplementary Results 2.
Note: Supplementary information is available in the online version of the paper.
Acknowledgments
We would like to thank A. Trejo, M. Clutter, K. Gibbs and G. Behbahani for
their experimental support and discussions, and D. Pe’er and El-ad D. Amir
for their feedback on data analysis. B.B. was supported by fellowships of the
Swiss National Science Foundation (SNF), the European Molecular Biology
Organization (EMBO), and the Marie Curie IOF. E.R.Z. is supported by a
fellowship from National Institute of General Medical Sciences (F32GM093508).
T.J.C. is supported by the Department of Defense (DoD) through the National
Defense Science & Engineering Graduate Fellowship (NDSEG) Program, and the
Stanford Graduate Fellowship in Science and Engineering. S.C.B. is supported by
the Damon Runyon Cancer Research Foundation Fellowship (DRG-2017-09).
G.P.N. is supported by the Rachford and Carlota A. Harris Endowed Professorship
and grants from U19 AI057229, P01 CA034233, HHSN272200700038C,
1R01CA130826, CIRM DR1-01477 and RB2-01592, NCI RFA CA 09-011, NHLBI-
HV-10-05(2), European Commission HEALTH.2010.1.2-1, and the Bill and
npg
Melinda Gates Foundation (GF12141-137101).
AUTHOR CONTRIBUTIONS
B.B. conceived and designed the experiments, performed all PBMC experiments,
MCB multiplexing and mass cytometry analysis, analyzed the data and wrote
the manuscript. E.R.Z. conceived and designed the experiments, developed the
mDOTA reagents, performed all PBMC experiments, MCB multiplexing and mass
cytometry analysis, analyzed the data and wrote the manuscript. R.F. designed
and implemented algorithms and software tools for barcode deconvolution, semi-
automatic cell-type gating and dose-response analysis. T.J.C. performed PCA and
assisted with data analysis. E.S.S. designed and implemented high-density data
visualization. R.V.B. designed and implemented scripts for FCS file processing
and assisted with SPADE analysis. E.F.S. assisted with SPADE analysis, designed
the 96-well barcoding scheme. S.C.B. assisted with antibody labeling and running
the mass cytometer, and high-dimensional analysis of human immune cell
populations. K.S. helped with bioinformatic data analysis. P.O.K. conceived cell
barcoding for mass cytometry. G.P.N. conceived and designed the experiments,
wrote the manuscript. All authors read and approved the final manuscript.
COMPETING FINANCIAL INTERESTS
The authors declare competing financial interests: details are available in the online
version of the paper.
Published online at http://www.nature.com/doifinder/10.1038/nbt.2317. Anal. Chem. 81, 6813–6822 (2009).
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
866 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology
1. Paul, S.M. et al. How to improve R&D productivity: the pharmaceutical industry’s
grand challenge. Nat. Rev. Drug Discov. 9, 203–214 (2010).
2. Nolan, G.P. What’s wrong with drug screening today. Nat. Chem. Biol. 3, 187–191
(2007).
3. Kola, I. & Landis, J. Can the pharmaceutical industry reduce attrition rates?
Nat. Rev. Drug Discov. 3, 711–715 (2004).
4. Kramer, J.A., Sagartz, J.E. & Morris, D.L. The application of discovery toxicology
and pathology towards the design of safer pharmaceutical lead candidates.
Nat. Rev. Drug Discov. 6, 636–649 (2007).
5. Kolch, W. & Pitt, A. Functional proteomics to dissect tyrosine kinase signalling
pathways in cancer. Nat. Rev. Cancer 10, 618–629 (2010).
6. Bodenmiller, B. et al. Phosphoproteomic analysis reveals interconnected system-wide
responses to perturbations of kinases and phosphatases in yeast. Sci. Signal. 3,
rs4 (2010).
7. Zhang, J., Yang, P.L. & Gray, N.S. Targeting cancer with small molecule kinase
inhibitors. Nat. Rev. Cancer 9, 28–39 (2009).
8. Hynes, N.E. & Lane, H.A. ERBB receptors and cancer: the complexity of targeted
inhibitors. Nat. Rev. Cancer 5, 341–354 (2005).
9. Irish, J.M., Kotecha, N. & Nolan, G.P. Mapping normal and cancer cell signalling
networks: towards single-cell proteomics. Nat. Rev. Cancer 6, 146–155 (2006).
10. Knight, Z.A., Lin, H. & Shokat, K.M. Targeting the cancer kinome through
polypharmacology. Nat. Rev. Cancer 10, 130–137 (2010).
11. Arrell, D.K. & Terzic, A. Network systems biology for drug discovery. Clin. Pharmacol.
Ther. 88, 120–125 (2010).
12. Irish, J.M. et al. Single cell profiling of potentiated phospho-protein networks in
cancer cells. Cell 118, 217–228 (2004).
13. Fabian, M.A. et al. A small molecule-kinase interaction map for clinical kinase
inhibitors. Nat. Biotechnol. 23, 329–336 (2005).
14. Karaman, M.W. et al. A quantitative analysis of kinase inhibitor selectivity.
Nat. Biotechnol. 26, 127–132 (2008).
15. Bamborough, P., Drewry, D., Harper, G., Smith, G.K. & Schneider, K. Assessment
of chemical coverage of kinome space and its implications for kinase drug discovery.
J. Med. Chem. 51, 7898–7914 (2008).
16. Anastassiadis, T., Deacon, S.W., Devarajan, K., Ma, H. & Peterson, J.R.
Comprehensive assay of kinase catalytic activity reveals features of kinase inhibitor
selectivity. Nat. Biotechnol. 29, 1039–1045 (2011).
17. Davis, M.I. et al. Comprehensive analysis of kinase inhibitor selectivity.
Nat. Biotechnol. 29, 1046–1051 (2011).
18. Melnick, J.S. et al. An efficient rapid system for profiling the cellular activities of
molecular libraries. Proc. Natl. Acad. Sci. USA 103, 3153–3158 (2006).
19. Evans, W.E. & Relling, M.V. Pharmacogenomics: translating functional genomics
into rational therapeutics. Science 286, 487–491 (1999).
20. Kramer, R. & Cohen, D. Functional genomics to new drug targets. Nat. Rev. Drug
Discov. 3, 965–972 (2004).
21. Bantscheff, M. et al. Quantitative chemical proteomics reveals mechanisms of action
of clinical ABL kinase inhibitors. Nat. Biotechnol. 25, 1035–1044 (2007).
22. Bantscheff, M., Scholten, A. & Heck, A.J. Revealing promiscuous drug-target
interactions by chemical proteomics. Drug Discov. Today 14, 1021–1029
(2009).
23. Perlman, Z.E. et al. Multidimensional drug profiling by automated microscopy.
Science 306, 1194–1198 (2004).
24. Singh, D.K. et al. Patterns of basal signaling heterogeneity can distinguish cellular
populations with different drug sensitivities. Mol. Syst. Biol. 6, 369 (2010).
25. Geva-Zatorsky, N. et al. Protein dynamics in drug combinations: a linear superposition
of individual-drug responses. Cell 140, 643–651 (2010).
26. Hulett, H.R., Bonner, W.A., Barrett, J. & Herzenberg, L.A. Cell sorting: automated
separation of mammalian cells as a function of intracellular fluorescence. Science 166,
747–749 (1969).
27. Perfetto, S.P., Chattopadhyay, P.K. & Roederer, M. Seventeen-colour flow cytometry:
unravelling the immune system. Nat. Rev. Immunol. 4, 648–655 (2004).
28. Chattopadhyay, P.K. et al. Quantum dot semiconductor nanocrystals for
immunophenotyping by polychromatic flow cytometry. Nat. Med. 12, 972–977
(2006).
29. Perez, O.D. & Nolan, G.P. Simultaneous measurement of multiple active kinase
states using polychromatic flow cytometry. Nat. Biotechnol. 20, 155–162 (2002).
30. Young, S.M. et al. High-throughput screening with HyperCyt flow cytometry to detect
small molecule formylpeptide receptor ligands. J. Biomol. Screen. 10, 374–382
(2005).
31. Bartsch, J.W. et al. An investigation of liquid carryover and sample residual for a
high-throughput flow cytometer sample delivery system. Anal. Chem. 76,
3810–3817 (2004).
32. Krutzik, P.O., Crane, J.M., Clutter, M.R. & Nolan, G.P. High-content single-cell drug
screening with phosphospecific flow cytometry. Nat. Chem. Biol. 4, 132–142
(2008).
33. Krutzik, P.O. & Nolan, G.P. Fluorescent cell barcoding in flow cytometry allows
high-throughput drug screening and signaling profiling. Nat. Methods 3, 361–368
(2006).
34. Bandura, D.R. et al. Mass cytometry: technique for real time single cell multitarget
immunoassay based on inductively coupled plasma time-of-flight mass spectrometry.
35. Bendall, S.C. et al. Single-cell mass cytometry of differential immune and drug
responses across a human hematopoietic continuum. Science 332, 687–696
(2011).

36. Lou, X. et al. Polymer-based elemental tags for sensitive bioassays. Angew. Chem.
Int. Edn. Engl. 46, 6111–6114 (2007).
37. Qiu, P. et al. Extracting a cellular hierarchy from high-dimensional cytometry data
with SPADE. Nat. Biotechnol. 29, 886–891 (2011).
38. Shuai, K. & Liu, B. Regulation of JAK-STAT signalling in the immune system.
Nat. Rev. Immunol. 3, 900–911 (2003).
39. Rawlings, J.S., Rosler, K.M. & Harrison, D.A. The JAK/STAT signaling pathway.
J. Cell Sci. 117, 1281–1283 (2004).
40. Platanias, L.C. Mechanisms of type-I- and type-II-interferon-mediated signalling.
Nat. Rev. Immunol. 5, 375–386 (2005).
41. Novershtern, N. et al. Densely interconnected transcriptional circuits control cell
states in human hematopoiesis. Cell 144, 296–309 (2011).
42. Akira, S. & Takeda, K. Toll-like receptor signalling. Nat. Rev. Immunol. 4, 499–511
(2004).
43. Andersson, U. & Matsuda, T. Human interleukin 6 and tumor necrosis factor alpha
production studied at a single-cell level. Eur. J. Immunol. 19, 1157–1160
(1989).
© 2012 Nature America, Inc. All rights reserved.
npg
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 867
Articles
44. Perez-Oliva, A.B. et al. Epitope mapping, expression and post-translational
modifications of two isoforms of CD33 (CD33M and CD33m) on lymphoid and
myeloid human cells. Glycobiology 21, 757–770 (2011).
45. Deisseroth, A. et al. U.s. Food and drug administration approval: ruxolitinib for the
treatment of patients with intermediate and high-risk myelofibrosis. Clin. Cancer
Res. 18, 3212–3217 (2012).
46. Knight, Z.A. & Shokat, K.M. Features of selective kinase inhibitors. Chem. Biol. 12,
621–637 (2005).
47. Lown, J.W. The mechanism of action of quinone antibiotics. Mol. Cell. Biochem. 55,
17–40 (1983).
48. Pardanani, A. JAK2 inhibitor therapy in myeloproliferative disorders: rationale,
preclinical studies and ongoing clinical trials. Leukemia 22, 23–30 (2008).
49. Brandman, O., Ferrell, J.E. Jr, Li, R. & Meyer, T. Interlinked fast and slow
positive feedback loops drive reliable cell decisions. Science 310, 496–498
(2005).
50. Jorgensen, C. et al. Cell-specific information processing in segregating populations
of Eph receptor ephrin-expressing cells. Science 326, 1502–1509 (2009).
 ONLINE METHODS
Kinase inhibitors. All inhibitors and the concentrations used are given in the
Supplementary Methods Table 1.
Preparation of barcoding reagents. Two molar equivalents of maleimido-
mono-amide-DOTA (Macrocyclics) were added to each metal chloride in
20 mM ammonium acetate, pH 6.0. Solutions were then immediately lyophi-
lized and resulting solids were dissolved in DMSO to 10 mM for long-term
storage at −20 °C.
PBMC isolation, culture and stimulation. Human peripheral blood, collected
based on a protocol approved by an internal review board, was obtained from
the Stanford Blood Bank. The samples obtained from healthy donors were
collected in heparin sulfate anticoagulant by leukapheresis and stored at room
temperature for 4–6 h. PBMCs were isolated by Ficoll-Paque density centrifuga-
tion. The isolated PBMCs were resuspended in freezing solution (90% FBS, 10%
DMSO) and stored under liquid nitrogen for future use. For each use, PBMCs
were thawed and then washed twice with room temperature PBMC media
(RPMI, 5% FBS, 2 mM L-glutamine with 100 U/ml penicillin and 100 µg/ml
streptomycin), incubated for 1 h at 37 °C in 5% CO2, and then stimulated as
shown in the Supplementary Methods Table 2 by the addition of IL-2, IL-3,
IL-12, G-CSF, GM-CSF, interferon-α, interferon-γ or LPS at 30 ng/ml, sodium
orthovanadate at 125 µM, phorbol 12-myristate 13-acetate (PMA) at 50 nM,
© 2012 Nature America, Inc. All rights reserved.
ionomycin at 1 µg/ml, or a mixture of anti-IgG, anti-IgM, anti-IgK and anti-IgL
at 10 µg/ml each (BCR/FcR-XL).
Antibodies used for analyses. Metal-labeled antibodies were prepared as
described35. Briefly, antibodies were obtained in carrier protein–free PBS and
then prepared using the MaxPAR antibody conjugation kit (DVS Sciences)
according to the manufacturer’s protocol. After determining the percent yield
by measurement of absorbance at 280 nm, the metal-labeled antibodies were
diluted in Candor PBS Antibody Stabilization solution (Candor Bioscience
GmbH) for long-term storage at 4 °C. Antibodies used in this study are listed
in the Supplementary Methods Table 3.
Accuracy and robustness assessment of MCB. For this analysis K562 cells,
a human myelogenous leukemia cell line, either untreated or treated with
orthovanadate was used. Orthovanadate is a broadly active protein tyrosine
phosphatase inhibitor that increases cellular tyrosine phosphorylation levels.
The induction of SH2 domain-containing leukocyte protein of 76 kDa (SLP76)
phosphorylation on Tyr696 in the orthovanadate-treated cells was observed to
be highly similar in multiplexed samples (Supplementary Fig. 2) compared
npg
to nonmultiplexed ones, indicating that the MCB method does not alter mass
cytometry measurement or introduce artifacts.
Time-course experiment. Approximately 20 million PBMCs were aliquoted
into a 96-well 2-ml block. After resting for 60 min at 37 °C, the PBMCs were
stimulated with agents listed in the Supplementary Methods Table 2 for
0 min, 1 min, 5 min, 15 min, 30 min, 60 min, 120 min and 240 min.
Inhibitor dose-response experiments. Approximately 20 million PBMCs
were aliquoted into a 96-well 2-ml block. After resting for 45 min at
37 °C, the PBMCs were pretreated with the indicated small molecule
kinase inhibitors for 15 min, and then stimulated with agents listed in the
Supplementary Methods Table 2 for 30 min in the presence of the inhibitor.
Small-molecule kinase inhibitors were all dissolved in DMSO before use, and
a 1:1,000 concentration of DMSO was used for all conditions, including the
uninhibited control wells.
PBMC fixation and permeabilization. At the indicated time point after stimu-
lation, 1.6% formaldehyde (final concentration) was added to the PBMC media
and cells were incubated at room temperature for 10 min. The formaldehyde
was then diluted to 0.8% with additional PBMC media, and the fixed cells were
centrifuged at 600g for 5 min at 4°C. After aspirating the supernatant, the cell
pellet was resuspended in ice-cold methanol and transferred immediately to
−80°C for long-term storage.
nature biotechnology
Cell barcoding and antibody staining. PBMC samples in methanol were
brought from –80°C to 4°C on ice, washed once with Cell Staining Media
(CSM, PBS with 0.5% BSA, 0.02% NaN3), and then once with PBS. The cells
were then resuspended in PBS, and DMSO stocks of the barcoding reagent
were added at 1:100 (to 100 nM final concentration). The cells were incubated
at room temperature for 30 min, washed three times with CSM, and then
pooled into a single FACS tube for staining with metal-labeled antibodies
for 1 h at room temperature. A staining volume of 300 µl was used (~1 × 108
cells/ml). After antibody staining, the cells were washed twice with CSM,
and then incubated for 10 min at room temperature (or overnight at 4 °C)
with an iridium-containing intercalator (DVS Sciences) in PBS with 1.6%
formaldehyde. The cells were then washed three times with CSM and once
with PBS, diluted with water to ~106 cells per ml, and filtered through a 70-µm
membrane just before analysis by mass cytometry.
Mass cytometry analysis. Cells were analyzed on a CyTOF mass cyto­meter (DVS
Sciences) at an event rate of ~500 cells per second. The settings of the instrument
and the initial post-processing parameters were described previously34,35. For
each barcoded sample several data files were recorded. These were concatenated
using a script developed in house. The cadmium ion signals with the mass over
charges of 110, 111, 112 and 114 were summed to create a single representative
channel for the CD3-QDot 605 used in the mass cytometry analysis by the Flow
Core package (http://www.bioconductor.org/packages/2.3/bioc/html/flowCore.
html). Before gating of the cell subpopulations and determination of the IC50
values, the data were normalized as described previously35.
In vitro kinase assays. All analyses were performed by Reaction Biology
Corporation, Malvern, Pennsylvania, USA, against active JAK1, JAK2, JAK3
and TYK2. The compounds analyzed are shown in Supplementary Table 3.
All kinase reactions were performed at 10 µM ATP using a 10-step, threefold
serial dilution with 10 µM as the highest compound concentration.
Data analysis. The cell events measured for the PBMC time-course experi-
ment were analyzed using the software tool SPADE as described in the main
text and previously35,37. All time-resolved response curves for all cell types
and stimuli are shown in Supplementary Results 1.
The following summarizes the SPADE algorithm within the context of this
time course data set. First, density-dependent down-sampling of all measured
cell events to a defined target number with equalization of the representation
of rare and abundant cell types was performed. The down-sampled cell events
were then clustered based on expression of ten cell surface markers (CD33,
CD20, CD3, CD4, CD7, CD123, CD14, CD45, IgM and HLA-DR) into pheno-
typically similar agglomerates of cells. Those agglomerates of cells phenotypi-
cally similar in ten dimensions were connected via edges to draw a minimum
spanning tree. Next, an up-sampling step was performed to assign each cell
event from the initial data set to the most representative agglomerate. Finally,
the minimum spanning tree was projected in two dimensions, and circles of
the tree representing cell agglomerates were colored by median intensity level
of a given measured parameter allowing visualization of marker expression
across the entire cellular hierarchy.
The cell events measured for each inhibitor were gated according to the
scheme shown in Figure 4. Each cell type was de-barcoded individually to
account for differences in the distributions of barcode metals due to differing
cell sizes. The de-barcoding was semi-automated for each barcode channel
by creating a boundary at the minimum between the two peaks in the density
estimate and then trimming 2.5% of the cells on each side of that bound-
ary. Subsequently, each cell was sorted into its barcode well according to the
seven-digit binary number assigned. The cells determined to be in the wells
stimulated with BCR-XL then had their IgM gates re-drawn because BCR-XL
masks the IgM epitope and shifts the IgM distribution to lower signal levels.
The dose-response curves were then computed for every combination of
phosphorylation site, modulator and cell type. This was done by fitting the
arcsinh-transformed median signal value S at each dose to the sigmoidal func-
tional form S = Top + (Bottom – Top)/(1+10^(hill*(LogIC 50 –LogDose))).
The fits were calculated using MATLAB’s implementation of a trust region
algorithm using a robust bi-square nonlinear least-squares method with each
doi:10.1038/nbt.2317
 point weighted by the inverse of the standard error of the mean. To deter-
mine which curves showed significant responses, the fitting scheme was first
applied to five control plates of cells that were treated with DMSO but not
inhibitor. Then the false-positive rate was calculated for varying levels of R2
and fold change cutoffs (Supplementary Methods Fig. 1a). An individual
curve was considered a responder if it exhibited a combination of R2 and fold
change that corresponded to a <1% false-positive rate in the inhibitor-free
plates (Supplementary Methods Fig. 1b) for a given analyzed phosphorylation
site (Supplementary Methods Fig. 1c). All dose-response results are shown
Supplementary Results 2–4 or can be viewed at http://www.cytobank.org/
nolanlab/ each curve is shown compared to the reference level and is overlaid
on individual contour plots for each sample with DNA along their hidden x
axes. Empty plots signify samples where there were zero cell counts.
The percent inhibitions reported for all drugs and conditions were those
observed at the highest measured inhibitor concentration, regardless of
whether saturation of inhibition was observed. For each curve, this was com-
puted by dividing the difference between the fitted curve at zero dose and at
the highest inhibitor concentration by the absolute value of the difference
between the fitted curve at zero dose and the reference line (Supplementary
Methods Fig. 1).
Principal component analysis. Principal component analysis (PCA) was used
to visualize the differences between various groups in the data, including all
© 2012 Nature America, Inc. All rights reserved.
cell types, as well as the differences between all inhibitors. Features that were
used for the PCA consisted of all IC50 values. In addition to the overall feature
npg
doi:10.1038/nbt.2317
matrix, PCA was run on data stratified by various subconditions, including
stimulation conditions. The control replicates were combined and averaged
for this analysis. Next, the pairwise Euclidean distance between all pairs of
points (cell types or inhibitors) in PCA space was calculated. This distance
was calculated including those principal components that recovered 90% of
the total variance. K-means clustering was performed using these distances to
determine subgroups of data. A best average silhouette value >100 replicates
of K-means was used to determine the potential number of clusters for each
set of conditions. The final cluster number was determined by inspection
of the silhouette plots. To simplify visualization of the overall relationships
among data points, a minimum spanning tree was created for each cluster.
In order to convey more information, the values of the data points mapped
to PCA space are also represented. The color of the node represents each
node’s location in the first principal component, and the size of the node
represents the location of a node in the second principal component. All
minimum spanning trees were visualized using Cytoscape. The clustergram in
Supplementary Figure 26 was generated using the same pairwise Euclidean
distance matrix described previously.
IC50 values contributing the most to principal components of interest were
found by determining those values most heavily weighted in each principal
component, >2 standard deviations from the mean. All of these methods were
run using MATLAB’s existing toolkits.
Data visualization. All cell density plots and heat maps were created in
Cytobank (http://www.cytobank.org/, Cytobank, Inc.).
nature biotechnology
 Articles

Combinatorial discovery of polymers resistant to

bacterial attachment
Andrew L Hook1,9, Chien-Yi Chang2,9, Jing Yang1,9, Jeni Luckett2, Alan Cockayne2, Steve Atkinson2, Ying Mei3,8,
Roger Bayston4, Derek J Irvine5, Robert Langer3,6,7, Daniel G Anderson3,6,7, Paul Williams2,10, Martyn C Davies1,10 &
Morgan R Alexander1,10

Bacterial attachment and subsequent biofilm formation pose key challenges to the optimal performance of medical devices.
In this study, we determined the attachment of selected bacterial species to hundreds of polymeric materials in a high-
throughput microarray format. Using this method, we identified a group of structurally related materials comprising ester and
© 2012 Nature America, Inc. All rights reserved.

cyclic hydrocarbon moieties that substantially reduced the attachment of pathogenic bacteria (Pseudomonas aeruginosa,
Staphylococcus aureus and Escherichia coli). Coating silicone with these ‘hit’ materials achieved up to a 30-fold (96.7%)
reduction in the surface area covered by bacteria compared with a commercial silver hydrogel coating in vitro, and the same
material coatings were effective at reducing bacterial attachment in vivo in a mouse implant infection model. These polymers
represent a class of materials that reduce the attachment of bacteria that could not have been predicted to have this property
from the current understanding of bacteria-surface interactions.

Healthcare-associated infection is widely acknowledged as the most
frequent adverse event in hospitals. It has been estimated that 80%
of the infections acquired in hospitals involve biofilms1. Biofilms are
surface-associated bacterial communities within which bacteria have
up to 1,000 times higher resistance to antimicrobials and host defenses
compared with their planktonic counterparts2,3. Biofilms present on the
surface of medical devices provide the bacterial inocula for disease and
can also serve as reservoirs of plasmids carrying antibiotic-resistance
genes4,5. Most strategies for reducing biofilm-associated infec-
tions focus on the modification of existing materials that are used
npg
parallel assay format that utilizes polymer microarrays20,21. A micro-
array format has previously proven useful for identifying polymeric
materials that support stem cell attachment and proliferation22–25.
In this study polymer microarrays were adapted for the combinato-
rial development of materials resistant to bacterial attachment and
growth (Fig. 1). In vitro testing has allowed us to identify a novel
class of structurally related polymers that are resistant to bacterial
attachment, outperforming commercially available silver-containing
coatings. In vivo testing of the most promising of our materials has
demonstrated the potential of these materials as coatings to reduce

to manufacture in-dwelling medical devices by the incorporation of device-centered infection.

anti­biotics6,7 or other antimicrobials, such as silver salts, nitrofura-
zone, chlorhexidine, polymerized quaternary ammonium surfactants,
antibacterial peptides and anionic nanoporous hydrogels8–15. These
approaches aim to kill bacterial cells that attach to a material, but we
believe that greater efficacy in preventing biofilm formation might
be achieved by the development of new materials that are inherently
resistant to biofilm formation16. Previous attempts to use an anti-
attachment strategy include poly(ethylene glycol) brushes17 and zwit-
terionic polymers18,19. An important problem that has restricted the
development of materials resistant to bacterial attachment and growth
is the poor understanding of the interactions that bacteria make with
surfaces, which is required for ab initio materials design.
To overcome this constraint, we have developed a high-throughput
approach to study bacterial attachment to hundreds of materials in a
RESULTS
A library of 22 acrylate monomers was selected from those available
commercially to provide wide chemical diversity, including ethylene
glycol chains of various lengths, fluoro-substituted alkanes, ­linear
and cyclic aliphatic, aromatic and amine moieties. To generate a
large combinatorial space, we mixed each of 16 monomers (Fig. 1a,
1–16) as the major component with each of another 6 monomers
(Fig. 1a, A–F) at ratios of 100:0 (homopolymer, 6 repeats), 90:10,
85:15, 80:20, 75:25 and 70:30 to create 576 monomer solutions.
The combined monomer solutions were printed in triplicate as
300-µm diameter spots with a height of ~20 µm onto a poly
(hydroxyl ethylmethacrylate) (pHEMA)-coated microscope slide
where they were photopolymerized by a free radical mechanism

1Laboratory of Biophysics and Surface Analysis, University of Nottingham, Nottingham, UK. 2School of Molecular Medical Sciences, University of Nottingham,

Nottingham, UK. 3Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 4School of Clinical Sciences,
Queen’s Medical Centre, University of Nottingham, Nottingham, UK. 5School of Chemistry, University of Nottingham, Nottingham, UK. 6David H. Koch Institute
for Integrative Cancer Research, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 7Harvard-MIT Division of Health Science Technology,
Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 8Present address: Clemson-MUSC Bioengineering program, Clemson University, Charleston,
South Carolina, USA. 9These authors contributed equally to this work. 10These authors jointly directed this work. Correspondence should be directed to
M.R.A. (morgan.alexander@nottingham.ac.uk).

Received 11 May 2012; accepted 26 June; published online 12 August 2012; doi:10.1038/nbt.2316

868 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 Articles

Figure 1  Schematic of the approach used to

identify hit materials resistant to bacterial
a O O O
(16)
O O
(1) O O (10) O
O
OH O
O O
O
O O O
attachment and scale-up of hit materials. O O O O O

(a) The chemical structures of the monomers. (2) O O (11) O O O

O O O O

(b–f) Outline of the strategy utilized for identifying (3)

O O
(12)
O O O O
O O O O O
hit material composition. (b) Fluorescence n n ≈ 11
O O HO O O

scanning image of the first-generation microarray (4) O (13) (A) OH

O O O O n n=7
after incubation with P. aeruginosa for 72 h. O (B)
O
O
Three replicate arrays were present on each glass O O
O
O
(5) O O O
slide. Scale bar, 10 mm. (c) Intensity map of O
O
OH
(14)
O O
(C)
O O O
the bacterial data from the first-generation array. (6)
O O
O O
O
O O O
3
(d) The feedback loop used to select monomer OH
n
n + m ≈ 2.5
m
O O F O
O
(D) F F
compositions for a second-generation array. The (7)
O O
O
F
O
F F
biological performance of each monomer, the O
OH
O (15)
O O
(E)
O

(8) O O O N
identification of hit material compositions that O O
O O O O O
have synergistic effects and the results from the (9) O O
O
(F)
O O
O m O n
PLS regression of ToF-SIMS spectra and bacterial n≈m≈2
data were used to inform the composition of
Microarray formation Bacterial attachment assay Identify hit compositions Scale up for Produce an optimized
the second-generation array. (e) Schematic
representation of the hit material composition b c for scale up and
in vitro testing
in vivo testing biomedical device

scaled up. (f) An SEM image of the cross-section i

of a silicone catheter coated with a hit polymer h
(thickness, 20–25 µm). Scale bar, 100 µm. e
(g) Confocal images of SYTO17-stained biofilm for
© 2012 Nature America, Inc. All rights reserved.

the three pathogens studied (P. aeruginosa (PA),

S. aureus (SA), UPEC) from coated and uncoated
silicone catheters. Each image is 160 × 160 µm.
(h) Microcomputed tomography scan of a coated f
Feedback loop

Feedback loop
Cross-section
catheter implanted subcutaneously into a murine
model. The catheter has been colored green in d O

O O
n
O Polymer coating

Hit monomers from overall Silicone catheter

this image. (i) A schematic representation of a biological performance
O

catheter coated with the hit composition. O

O

O
OH
O

g
Bacterial
numbers

Synergistic
Uncoated silicone
to form the first-generation combinatorial effects
0 100
polymer microarray (Fig. 1b). Hits from Monomer composition (%)

this first-generation polymer microarray Hit chemical functionalities

from PLS regression
O

O
Coating of hit
were used to formulate the second- O

O
composition

­generation array and so on. UPEC SA PA

Bacterial microarray screening
The pathogens P. aeruginosa, S. aureus and uropathogenic E. coli
(UPEC)26 were transformed with plasmids expressing GFP to facili-
tate a high-throughput screen with a fluorescence intensity readout
npg
estimate of bacterial attachment (Supplementary Fig. 3). In cases
where polymers were formed from monomers that had opposing
influences on bacterial attachment, a linear correlation between
systematic variations in material composition and F was observed

(Fig. 2a). These species were chosen as representatives of both Gram-
positive (S. aureus) and Gram-negative (P. aeruginosa and UPEC)
pathogens that are frequently found in medical device–associated
infections27,28. The polymer microarrays were incubated with a sus-
pension of planktonic bacteria for each of the pathogens separately
for 3 d (72 h) and revealed a wide range of resistance to bacterial
attachment (Fig. 2b–d). To simulate in vivo conditions, such as those
encountered in the urinary tract, we conditioned one microarray with
artificial urine for 72 h before incubation with UPEC and, in another
experiment, we replaced the RPMI-1640 medium with artificial urine
for a 72-h incubation with UPEC (Fig. 2e,f). Biomineralization was
observed on some polymers as opaque deposits after conditioning in
artificial urine and after culture in artificial urine in the absence of
buffering (Supplementary Fig. 1), which stimulated greater bacte-
rial attachment in all cases. The time period chosen provided a suf-
ficiently stringent assay for the identification of a set of ‘hit’ materials
that were resistant to bacterial attachment (Supplementary Fig. 2).
The fluorescence signal (F) (Online Methods, equation (1)) was used
to quantify the amount of bacterial attachment for each bacterial
strain. A linear correlation (R2 = 0.93) between F and surface cover-
age (by area), as determined by confocal measurements of biofilm
after incubation with UPEC, confirmed the utility of F as a reliable
(Supplementary Fig. 4). For some polymers the biological behavior
of the copolymer was superior to that of the homopolymers of the two
constituent monomers. In one such example, the F of P. aeruginosa
(FPA) on the copolymer 13:E (85:15) was 1.0 × 106 ± 0.2 × 106 arbitrary
units (AU), which was lower than the FPA from the homopolymer of
monomer 13 (2.1 × 106 ± 0.2 × 106 AU) and the FPA on the homopoly-
mer of monomer E (29.8 × 106 ± 1.1 × 106 AU) (Supplementary
Fig. 5). This synergistic effect was observed for 11 of the 96 mono-
mer pairs explored in the first-generation array and in all cases led
to greater resistance to bacterial attachment, demonstrating that
screening for hit materials using a copolymer library could produce
copolymers with unexpected properties.
Varying levels of attachment were observed for the three bacte-
rial species to the same polymer surfaces (Supplementary Fig. 6).
This finding is unsurprising given the different surface properties
and macromolecular surface composition of Gram-positive and
Gram-negative bacteria29 and the diverse attachment mechanisms
they use, which can involve surface proteins, flagella, fimbriae30 and
exopolysaccharides31.
To assess the resistance of any given material to attachment by
diverse bacterial species and exposure conditions, and as a guide
to select hit monomers for further study, we developed a composite

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 869

 Articles

Figure 2  Slide inoculation procedure and a Polymer microarray Fluorescence quantified

‘heatmaps’ of bacterial fluorescent intensity GFP-transformed inoculated on each spot
bacteria Grown in LB OD600 = 0.01 Polymer microarray read
determined from first-generation slides by fluorescence scanner
(a) Bacterial attachment assay procedure.
(b–f) Intensity maps of F measured for each
bacterial strain on the first-generation array
after 72-h incubation; P. aeruginosa (b),
S. aureus (c), UPEC (d), UPEC grown in
b (AU)
7
e (AU)
1 6.8 × 10 1 9.3 × 10
7
artificial urine (e) and UPEC grown on an 2
3
2
7 3
artificial urine–conditioned slide (f). 4
5
3.5 × 10 4
5
3.8 × 10
7

(g) Intensity map of the  obtained for each 6

7 1.5 × 10
7 6
7 1.2 × 10
7

material in the array, given as a percentage of 8

9
8
9
6
10 4.3 × 10 10 6
the maximum bacterial numbers for all strains 11 11
2.4 × 10
12 12
at all culture conditions. The major monomers 13 5.4 × 10
5
13 1.5 × 10
5
14 14
are listed on the y-axis whereas the composition 15 15
16 0 16 0
of the minor monomers is shown on the x-axis.
The large shaded area within each outlined area c1 8.0 × 10
7
f 7
1 8.0 × 10
indicates the mean value, and the mean ± 1 s.d. 2 2
3 7 3
unit is presented in the narrow columns to the 4 4.1 × 10 4 3.3 × 10
7
5 5
right (plus) and left (minus) of the mean, n = 3. 6
7 1.7 × 10
7 6
7
7 1.0 × 10
Key to right. A square-root scale is applied to 8
9 6
8
9
5.1 × 10
the intensity indicators. (h) The average  for 10
11
10
11
2.0 × 10
6

12 12
all materials containing a specific monomer, 13 6.4 × 10
5
13 1.3 × 10
5
14 14
ranked from lowest to highest. The major and 15 15
© 2012 Nature America, Inc. All rights reserved.

16 0 16 0
minor monomers were considered separately.
The color next to each monomer is indicative d 7.4 × 10
7
g
1 1 66 (%)
of that monomer’s mean  and is colored by the 2 2
3 3
same intensity scale as in g. 4 3.0 × 10
7
4 34 (%)
5 5
6 6 6
7 9.6 × 10 7 14 (%)
8 8
9 9
bacterial attachment parameter. The F value
6
10 1.9 × 10 10 4 (%)
11 11
12 12
from each strain on each polymer composi- 13
14
1.2 × 10
5 13
14
0.5 (%)

tion, including UPEC incubated on artificial 15

16 0
15
16 0 (%)
urine–conditioned slides and UPEC incu-
(0%)
A(10%)
A(15%)
A(20%)
A(25%)
A(30%)
B(10%)
B(15%)
B(20%)
B(25%)
B(30%)
C(10%)
C(15%)
C(20%)
C(25%)
C(30%)
D(10%)
D(15%)
D(20%)
D(25%)
D(30%)
E(10%)
E(15%)
E(20%)
E(25%)
E(30%)
F(10%)
F(15%)
F(20%)
F(25%)
F(30%)

(0%)
A(10%)
A(15%)
A(20%)
A(25%)
A(30%)
B(10%)
B(15%)
B(20%)
B(25%)
B(30%)
C(10%)
C(15%)
C(20%)
C(25%)
C(30%)
D(10%)
D(15%)
D(20%)
D(25%)
D(30%)
E(10%)
E(15%)
E(20%)
E(25%)
E(30%)
F(10%)
F(15%)
F(20%)
F(25%)
F(30%)
bated in artificial urine, was normalized to
the maximum F on the slide and averaged h B C F D A E

over all three replicates to produce an esti-
mate of bacterial attachment value for all
conditions and pathogens, called iota () and defined in equation (2).
A high value of  indicated high bacterial attachment and therefore a
poor material performance and vice versa. We plotted the value of 
in an intensity map for all first-generation polymer arrays, revealing
which materials were least resistant to attachment by all three bacte-
npg
5 8 15 12 14 9 13 11 2 16 3 4 1 10 6 7
WCA or roughness (Supplementary Fig. 8), for the 496 materials
studied in the first-generation array. ToF-SIMS analysis, coupled with
the chemometrics technique of partial least squares (PLS) regres-
sion, has been demonstrated as a powerful technique for correlating
surface chemistry represented in mass spectra with a univariate data

rial strains (red) and which were most resistant to all three (blue)
(Fig. 2g). To compare the performance of each monomer used to
fabricate the polymers, we determined the average  from each of the
materials containing a given monomer. The monomers were ranked
by their average  value, hence according to their resistance to bacte-
rial attachment (Fig. 2h). Monomers 5 and B produced materials with
the lowest average , that is, the most resistant to attachment, whereas
7 and E produced materials with the highest average .
High-throughput surface characterization of microarrays
To investigate the effect of polymer surface properties on bacterial
attachment, we carried out high-throughput surface characterization
(HT-SC) of the polymer microarray32. Techniques that probe the outer­
most surface of materials were used, including X-ray photoelectron
spectroscopy (XPS) for quantitative elemental and functional analysis,
atomic force microscopy (AFM) for topographical characterization,
time-of-flight secondary ion mass spectrometry (ToF-SIMS) for molec-
ular characterization and water contact angle (WCA) measurement to
probe the surface wettability20,32–34. The influence of each property
on bacterial attachment was assessed for all three bacterial strains
separately. No correlation was identified between bacterial attach-
ment and surface elemental composition (Supplementary Fig. 7),
set such as WCA35,36 or stem cell attachment24,25. It was applied here
to search for correlations between the surface chemistry of the array
and the F values from each strain and to identify important surface
moieties for bacterial attachment.
The PLS regression model produced by this analysis success-
fully predicted the F values for P. aeruginosa and S. aureus from the
ToF-SIMS spectra (Fig. 3a), as evidenced by the linear relationship
between the predicted and experimental F values (Fig. 3b–d), with
an R2 value of 0.68 and 0.76 for the two bacterial species, respectively.
The PLS model for S. aureus gave a good prediction of the bacterial
attachment for all polymers except those containing monomer 1, 6 or
10. No correlation was identified for UPEC (R2 = 0.28). Only 17% of
the materials on the array had a FUPEC value >1% of FUPECmax com-
pared with 97% and 96% for P. aeruginosa and S. aureus, respectively.
The low attachment of this strain on the materials screened made it
difficult to compute a suitable PLS regression model; thus, this lack
of a correlation does not exclude the possibility of a dependence of
UPEC bacterial attachment on surface chemistry.
The successful prediction of attachment by P. aeruginosa and
S. aureus from the ToF-SIMS spectra demonstrates that the attach-
ment of these species is dependent on the polymer surface chem-
istry. More specifically, the influence of each of the hundreds of

870 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 Articles

a F after 72 h 12,000 ToF-SIMS spectra e

Positive Positive Negative Negative

Intensity
Ion RCPA (×107) 7 Ion 7 7
8,000 RCSA (×10 ) RCPA (×10 ) RCSA (×10 )
PLS regression C2H–
4,000 C2H5O+ 8.1 –0.2 –7.8 –5
C3H7O+ 7.4 1.4 C4H– –3.3 –1.6
0 C6H5O– 4.4 26.9 C3H3+ –2.5 –0.8
0 40 80 120 160 200 C6H11O3– 4.4 1.3 C2H3+ –2.3 –1.6
m/z
C5H9+ 3.6 1 C8H15+ –2.2 –1.1

b 40 c 80 d 80
C2H3O+ 3.1 0.1 C4H7+
C3H7+
–2.2 –2.4
C3H3O+ 0.6 2.7 –2.1 –2

Predicted FUPEC (×10 ) (AU)

Predicted FPA (×106) (AU)

Predicted FSA (×106) (AU)

C6H7O+ 0.4 1.8 C4H5+ –1.7 –1.1
30 60 60 C6H6O+ 0.2 2.2 C6H– –1.7 –0.8

6
C2H3O2– 0.1 2.2 CHO2– –1.5 –2.1

20 40 40 C4H5O2– 0.1 2 C4H7O– –1.5 –0.4

C3H3O2– 0 4.5 C2H5+ –1.4 –1
C7H7O+ 0 2.6 C4H3+ –1.1 –0.4
10 20 20
C6H5+ –0.5 5.4 C3H5+ –0.5 –1.5
C7H5O2– 0 –1.4
0 0 0 C7H5O+ 1.4 –1.9
0 10 20 30 40 0 20 40 60 80 0 20 40 60 80
6 6 6
Measured FPA (×10 ) (AU) Measured FSA (×10 ) (AU) Measured FUPEC (×10 ) (AU)

Figure 3  Correlation of the surface chemistries represented in the ToF-SIMS. (a) Schematic depiction of the PLS regression model used to predict the
biological performance of materials by correlating F with the ToF-SIMS spectra. (b–d) The predicted bacterial attachment determined from the PLS regression
model for P. aeruginosa (PA) (R2 = 0.68) (b), S. aureus (SA) (R2 = 0.76) (c), and UPEC (R2 = 0.28) (d). The y = x line is drawn as a guide. Polymers are
grouped according to the major monomer 1 ( ), 2 ( ), 3 ( ), 4 ( ), 5 ( ), 6 ( ), 7 ( ), 8 ( ), 9 ( ), 10 ( ), 11 ( ), 12 ( ), 13 ( ), 14 ( ), 15 ( ), 16 ( ).
© 2012 Nature America, Inc. All rights reserved.

(e) The key ions identified to be important by ToF-SIMS PLS regression analysis for the surface attachment of both P. aeruginosa (PA) and S. aureus (SA).
The regression coefficient (RC) for each ion is also shown from the regression analysis with each bacterium separately. The RCs have been shaded according to
their value (red, positive; blue, negative; white, neutral).

ions in the SIMS spectra on bacterial attachment is quantified by to be associated with reduced bacterial attachment by the PLS regres-
the ­regression ­coefficient, where a positive coefficient indicates that sion ­analysis. The top four hit monomers were selected (15, 5, 8 and B)
the ion in question promotes attachment, whereas a negative one and mixed with each other and monomer 4 at ratios of x:(1 − x) where
indicates that the surface chemistry giving rise to the secondary ion x ­varied from 10 to 90. This array contained 145 different materials
is resistant to bacterial attachment (Fig. 3e). The surface chemical (four replicates of each) plus four materials selected as positive
moieties assigned to secondary ion fragments with the highest PLS controls that in the first-generation array were least resistant to
regression coefficients are shown in Figure 3e for both P. aeruginosa bacterial attachment. Attachment after 3 d of incubation with all
and S. aureus. In general, the presence of hydrocarbon second- three bacterial pathogens, including UPEC incubated with artificial
ary ions in spectra from materials was correlated with high resist- urine–­conditioned slides, was assessed for the array and the  for
ance to bacterial attachment and oxygen-containing ions from each material determined (Fig. 4).
certain pendant groups were observed to be correlated with low
resistance to bacterial attachment for both P. aeruginosa and Scale up from microarray spots
S. aureus. In particular, secondary ions consistent with cyclic carbon The six copolymer compositions (and four corresponding homopoly-
groups (C4H−, C6H−), ester groups (CHO2−), the tertiary butyl moi- mers) with the least bacterial attachment were chosen from the bacte-
npg

ety (C4H7+) and ions from aliphatic groups (C2H3+, C2H5+, C3H7+) rial screen of the second-generation array as hit materials for further
were correlated with lower bacterial attachment for both pathogens. study. Scale-up from microarray spots to 8- to 10-mm dia­meter sam-
Ions from ethylene glycol groups (C2H3O+, C2H3O2−), and hydroxyl- ples (termed coupons) was conducted to investigate the scalability of
containing fragments (C4H5O2−, C6H11O3−) correlated with higher the polymers’ physicochemical properties and biological performance
bacterial attachment. These results point to the influence of moieties (Supplementary Figs. 9 and 10). Initial scale-up experiments included
from particular monomers that are associated with the biological per- monomer 15 (and other triacrylates), but inclusion of this monomer
formance of the resultant polymer. Bacterial attachment is lower in resulted in the formation of brittle materials that were observed to
monomers with the cyclic carbon environments as in monomers 4 crack during the solvent extraction step. This was consistent with the
and B, the relatively higher density of ester groups in materials formed formation of a highly cross-linked polymer and, thus, monomer 15
from the triacrylate monomers 13 and 15, the tertiary butyl group of was removed from subsequent scale-up experiments. The area
monomer 5 and the dimethyl hydrocarbon segment on monomers covered by the bacteria was at least twofold lower on all scaled-up
2, 8 and 12. Bacterial attachment is greater where the ethylene glycol materials compared with the positive control (monomer 7 homopoly-
group is surface expressed as in polymers from monomers 1, 9, 16 mer) (Supplementary Fig. 9a). Confocal microscopy coupled with
and A, and where the hydroxyl group is present as in monomers 6, live and/or dead staining (Supplementary Fig. 10) on the scaled-up
7 and 10. polymer coupons revealed that, in common with the control polymer,
To determine the monomer composition most resistant to a mixture of live and dead bacteria was attached to the materials.
attachment by all three bacterial pathogens, we formulated a To simulate the more challenging in vivo conditioning by urine that
­second-­generation array that focused on the constituents from the is likely to occur on urinary catheters, P. aeruginosa, S. aureus and
first-generation array most resistant to bacterial attachment (Fig. 1d) UPEC were incubated with polymer coupons preconditioned with
but with greater variation in composition for each monomer pair. artificial urine. For all three strains a greater than threefold reduction
Monomer 4 was included as it contained the cyclic hydrocarbon moi- in the area of the sample covered by bacteria was observed for this
eties identified by the ToF-SIMS ions, C4H− and C6H−, and revealed assay compared with the positive control (monomer 7 homopolymer)

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 871

 Articles

Figure 4   determined for all materials represented in the second-generation O O

O O
array. Five monomers from the first-generation array were mixed pairwise O
at 14 different ratios (15, 5, 8 and B were mixed with each other and 15 O

monomer 4 at ratios of x:(1 − x) where x varied from 10 to 90). The content 100 90 85 80 75 70 60 55 45 40 30 25 20 15 10
of the major monomer is indicated along the x-axis: (a) 15, (b) 8, (c) 4, a 8
4
96.2%

(d) B and (e) 5 (Fig. 1a). The co-monomer is indicated in the y-axis. B
5
(f) The  determined for high-bacterial-attachment polymers included in the
second-generation array from the first-generation array. A square-root scale O O

37.6%
is applied to the color bar to highlight the low  measured for this array. The 8
O O

mean of each value is shown as the large shaded area within each outlined b 15
4
rectangle and the mean ± 1 s.d. unit is presented in the narrow columns to B
the right (plus) and left (minus) of the mean, n = 4. 5
O
O
O 14.4%

(Supplementary Fig. 9b), suggesting that these materials are resist- c 15

4

ant to bacterial attachment in the presence of urine components. 8

B
ToF-SIMS of the polymer coupons indicates that the aliphatic pendant 5
5.2%
O
group on monomers 4 and 8 was surface enriched compared with the O
O

equivalent microarray spots, which suggests that in some instances d B

O

15
scaling up from a microarray spot to a coupon modifies the mate- 8
rial surface chemistry and biological performance (Supplementary 4
5
1.5%

Figs. 9 and 11).

O
© 2012 Nature America, Inc. All rights reserved.

Development of medical device coatings e 15

5 O

8 0.0%
To establish the potential of these hit materials as effective coatings 4
for medical devices, we dip-coated oxygen plasma activated silicone f B

catheters with monomer solution followed by UV curing (Fig. 1e) and
compared bacterial attachment with silicone catheters and a commer-
cially available, state-of-the-art, silver-­containing coating. The four
homopolymers of the hit monomers, six hit copolymer formulations
and polymers from monomers 6, 7 and 1:E (70:30) (positive controls)
were coated along the luminal and abluminal surfaces of the catheters
(Fig. 1f,g). Scanning electron microscopy (SEM) was used to verify
the presence of the coatings (Supplementary Fig. 12). The coated
catheters were incubated with P. aeruginosa, S. aureus or UPEC for
72 h. Representative confocal images from the coated catheters after
bacterial incubation are shown in Figure 5a and the area of biofilm
coverage was quantified for each species. The corresponding  val-
ues, normalized to the biofilm area coverage as measured on silicone
alone, are shown in Figure 5b.
For all three pathogens, the coated catheters had a substantially
npg
1(70%)E(30%)
7(70%)E(30%)
6(70%)E(30%) –0.6%
pathogens. The proportion of the surface covered for this coating was
2.3% ± 1.3%, 1.0% ± 0.4% and 1.5% ± 0.7% for P. aeruginosa, S. aureus
and UPEC, respectively. This amounted to a 12-fold reduction in the
composite parameter  compared with an uncoated silicone catheter,
and sevenfold lower than the BactiGuard-coated Bardex catheter.
There are several other different strategies that can be found in the
literature to reduce bacterial attachment to surfaces that are not yet
commercially available; for example, zwitterionic coatings function
by preventing attachment18,19. Comparison with the performance of
our hits to these literature reports is difficult as experimental meth-
odologies vary widely. Attachment of P. aeruginosa (PA01) after 3 h
incubation on a poly(sulfobetaine methacrylate)-grafted surface has

lower bacterial coverage than the silicone elastomer. Several of the hit
acrylate materials were superior to the silver-hydrogel (BactiGuard)-
coated latex catheter (Bardex), which has been shown to be clinically
effective at reducing catheter-associated asymptomatic bacteriuria
when compared with standard latex catheters28. It should be noted
that silver-hydrogel catheters are thought to be resistant to bacte-
rial attachment as a consequence of the toxicity of silver ions to the
bacteria, whereas our hit polymer coatings prevent attachment. The
lowest area coverage for P. aeruginosa was achieved on the homopoly-
mer formed from monomer 4, which had 1.2% ± 0.5% of the surface
populated with bacteria. This represents a 28-fold reduction compared
with the silicone catheter, and a 17-fold reduction compared with the
BactiGuard-coated Bardex catheter. For S. aureus the best performing
copolymer was formed from monomers B:5 (70:30) and resulted in
a bacterial surface area coverage of 0.5% ± 0.3% , which represents a
67-fold reduction compared to silicone and a 30-fold reduction com-
pared to BactiGuard. For UPEC the best performing copolymer was
B:4 (90:10), which resulted in 1.2% ± 0.6% of the area populated with
bacteria, representing a ninefold reduction compared to silicone and
a sixfold reduction compared to BactiGuard. The homopolymer of B
produced the catheter coating with the lowest  of 2.0% ± 1.0%, and
was the best material at preventing the attachment of all three different
been reported to be lower by 25-fold compared to that on glass19,
and in an incubation time closer to ours, a poly(carboxybetaine
methacrylate)-grafted surface had 11-fold lower attachment of
P. aeruginosa after 96-h bacterial exposure compared to glass18.
Mechanism for reduction of bacterial coverage
The absence of a correlation between bacterial attachment and the
contact angle or roughness of the materials studied (Supplementary
Fig. 8) suggests that the interaction of bacteria with the meth­acrylate
or acrylate library cannot be explained simply by hydrophobic inter-
actions or roughness only, as was previously invoked to explain bac-
terial performance on self-assembled monolayers, stainless steel and
certain polymers37. It should be noted that the bacterial cell incuba-
tion experiments were not conducted under flow conditions, thus,
this assay did not assess any advantage gained from the shielding from
shear that increased roughness could provide.
The ability to predict bacterial attachment from the chemistry of
the materials as represented by the ToF-SIMS spectra (Fig. 3b–d)
confirms that the bacteria-material interaction is dependent on sur-
face chemistry. The PLS regression highlighted the ethylene glycol
moieties and hydroxyl groups employed in these libraries for pro-
moting the attachment of P. aeruginosa and S. aureus. The greater

872 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 Articles

a P. aeruginosa
Figure 5  Proportion of surface covered by
bacteria on catheters coated with hit polymers
after 72 h incubation with planktonic bacteria.
S. aureus (a) Confocal microscopy images of P. aeruginosa,
S. aureus and UPEC stained with SYTO17
UPEC growing on coated catheters, uncoated silicone
and silicone treated with media without bacteria
as a controls. Each image is 160 × 160 µm.
b 80
(b) Quantification of bacterial area coverage
for P. aeruginosa (blue), S. aureus (red), UPEC
(green) and  (purple) from confocal images of
each sample. Area coverage was normalized to
60
the coverage on silicone. The error bars represent
mean ± 1 s.d. unit, n = 5. The composition of
Biofilm coverage (%)

the positive control was specific to the bacterial

strain used: P. aeruginosa = 6(100%), S. aureus =
40 7(100%), UPEC = 1(70%)E(30%). The
fluorescence signal observed on the media-
only controls is due to auto-fluorescence of the
substrate and translates to a small erroneous
20 (<0.5%) estimate of area of surface covered
with bacteria.
© 2012 Nature America, Inc. All rights reserved.

0 appear to be important for reduced bacterial

Media only
(Silicone)
Media only
(Bacti-Guard)

Silicone

(Bacti-Guard)

Positive control

B(100%)

5(100%)

8(100%)

B(100%)

B(70%)5(30%)

B(90%)4(10%)

B(60%)4(40%)

B(60%)8(40%)

B(25%)8(75%)

B(40%)4(60%)
attachment. Antibacterial behavior has previ-
ously been reported for zwitterionic materi-
als19 where disparate chemical properties are
presented in close proximity on the molecu-
lar scale analogous to the weak amphiphiles
attachment associated with surface hydroxyl groups suggests a role identified in this study. To further investigate the role of weakly
for hydrogen bonding, which may be through an interaction with amphiphilic acrylate materials in preventing bacterial attachment,
the lipopolysaccharides, lipoteichoic acids or exopolysaccharides we produced a third polymer microarray that contained homopoly-
present on the bacterial cell surface31. For materials with resistance mers of 15 methacrylate or acrylate monomers that had aliphatic,
to bacterial attachment, the PLS regression analysis identified hydro- cyclic or aromatic pendant groups. F for P. aeruginosa, S. aureus and
phobic moieties such as aromatic and aliphatic carbon groups when UPEC and  for each material is shown in Supplementary Figure 14.
accompanied by the weakly polar ester groups. By comparison with The relatively high attachment of P. aeruginosa provided the greatest
polystyrene, a purely hydrophobic material that is well known to discrimination between the materials. The six materials with the
support bacterial attachment (Supplementary Fig. 13)38, the role of the lowest FPA all contained cyclic or aromatic hydrocarbon groups.
ester group and the weakly amphiphilic structure of these hit polymers In contrast, the six materials with the highest FPA all contained linear
npg

Figure 6  In vivo performance of hit polymer.

Catheters dip-coated with hit polymer 4
a d
Uncoated
radiance (p/s/cm2/sr) (×105)

copolymerized with DEGMA were implanted 1.0 Coated

Normalized average

subcutaneously in mice. Mice were inoculated

1.E+10
after 1 d with S. aureus Xen29, which were
Cfu/mg tissue day 4

1.E+09
injected into the center of the tube. (a) The
1.E+08
bioluminescence at the infection site was 0.1
1.E+07
measured on the day of inoculation (day 0) and
1.E+06
for the next 4 d. Error bars, ± 1 s.d. unit; n = 9
1.E+05
(three separate experiments, each comprising
0.01 1.E+04
three control catheter mice and three polymer-
Spleen
Kidney
Catheter

Surrounding
tissue

Day 0 Day 1 Day 2 Day 3 Day 4

coated catheter mice). The difference in
bioluminescence between coated and uncoated
samples from day 1 to 4 was confirmed to 6

99.5% confidence (t-test). Bioluminescence 4

3
b c
was normalized to the output at day 0 and 2

all measurements have had the background 10

6
8
luminescence subtracted, measured from an 6
4
implanted catheter prior to inoculation. (b,c) 3
Luminescence images with overlaid brightfield 2

images of mice implanted with both uncoated

Uncoated Coated Uncoated Coated p/s/cm2/sr
(left) and coated (right) catheter segments on
day 0 (b) and day 4 (c). (d) At day 4 the mice were euthanized, the insertion site plus the kidneys and spleen harvested, and the number of bacteria on the
tissue determined. Error bars, ± 1 s.d. unit; n = 3 (taken from one group of three). The difference in cfu counts between coated and uncoated samples was
confirmed to 99% confidence for the catheter and surrounding tissue, to 90% confidence for the kidneys and to 75% confidence for the spleen (t-test).

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 873

 Articles
aliphatic carbon pendant groups, suggesting that the presence of ring
structures is a determining factor for preventing bacterial attachment
to methacrylate/acrylate polymers (Supplementary Fig. 14).
Confocal microscopy did not detect a significant number of either
dead or live bacteria on the hit materials after 72 h incubation with
planktonic bacteria (Fig. 5a), indicating that the mechanism behind
the low attachment is prevention of bacterial attachment rather than a
mechanism involving killing. Consistent with this conclusion, growth
curves, which quantify bacterial numbers in the presence of the mate-
rials, showed no inhibition by the hit polymers for the bacterial strains
used (Supplementary Fig. 15). In addition live and/or dead staining
of the very low coverage of UPEC biofilm that were found revealed
both live and dead cells present within the biofilm, which is typical of
normal biofilms (Supplementary Fig. 10). Furthermore, there is no
evidence of cytotoxicity to human cells of the polymers as the materials
have been shown to support the culture of delicate embryonic stem
cell lines24.
The resistance of the lead acrylates discovered in this study to attach-
ment of bacteria likely depends on the ability of bacterial cells to sense
and respond to their immediate environment. This may be a conse-
quence of the individual cells sensing the nature of the polymer surface
© 2012 Nature America, Inc. All rights reserved.
by their cell envelope–associated sensory proteins or by specific surface
structures such as flagella and pili involved in near-surface movement39.
It is also possible that the bacterial population collectively responds
to the polymer through quorum sensing (­bacterial cell-to-cell com-
munication) mechanisms40, such that the lack of bacterial attachment
occurs through this population-­dependent decision-­making process.
In both scenarios the mechanism may be more complex than simple
physicochemical interactions between the bacteria and surface. Specific
examples include the Rcs sensor kinase, which controls the expression
of a number of E. coli genes in response to growth on a solid surface41.
Furthermore, production of the P. ­aeruginosa ­exopolysaccharides Pel
and Psl are both under quorum-sensing control42. Thus, it is highly
likely that bacterial responses to surfaces are more sophisticated than
is currently appreciated.
In vivo assessment of hit material
The in vivo environment is more realistic than in vitro assessment of
candidate materials with regards to eventual clinical application. We
npg
therefore carried out a subcutaneous foreign body infection model
to test the efficacy of one of the hit materials (Fig. 1h). A copolymer
of monomer 4 and di(ethylene glycol) methyl ether methacrylate
(DEGMA) amenable to the dip-coating methodology was prepared
by catalytic chain transfer polymerization (measured properties of
the resultant polymer are shown in Supplementary Table 1). The
DEGMA was included to tune the polymer’s mechanical properties
for in vivo use without compromising the polymer’s resistance to bac-
terial attachment (Supplementary Figs. 16 and 17) and to exploit the
ability of the oligo(ethylene glycol) moiety to reduce protein fouling43
in the high protein–containing environment.
Both control silicone and dip-coated silicone catheters were
implanted subcutaneously into mice and inoculated with biolumi-
nescent S. aureus Xen29 by injection into the catheter lumen after 1 d.
Immediately following inoculation similar bioluminescence was
observed for both coated and uncoated catheters (Fig. 6). After 1 d,
we observed more than a tenfold reduction in bioluminescence on the
coated catheters compared with the uncoated silicone, a difference
that persisted for 4 d (Fig. 6a–c). Bioluminescence requires the bac-
teria to be respiring aerobically, thus, it will not detect bacteria that
are viable but are either dormant or growing anerobically. To inves-
tigate whether bacterial numbers on coated catheters were reduced,
874 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology
we euthanized the mice on day 4 and quantified the numbers of
bacteria at the infection sites on the catheter and in the surround-
ing tissues, including kidneys and spleen. Bacterial numbers were
reduced by nearly two orders of magnitude on the coated catheter
compared to the uncoated catheter, and an order of magnitude reduc-
tion in bacterial numbers was observed in the tissue surrounding the
implant, and in the kidneys and the spleen, suggesting a reduced
amount of systemic bacteria (Fig. 6e). The reduction in biolumines-
cence on the coated catheter in the live animal demonstrates that the
coating successfully reduced bacterial in vivo survival in the presence
of the host defenses compared to the silicone control. We interpret
the persistence of bacteria on the silicone catheter as indicative of
bacteria that were able to attach to the silicone and form biofilms,
thereby avoiding clearance by the host defenses. These observations
suggest that the performance of the hit polymer identified in vitro
translates to the challenging in vivo environment. Sufficiently small
diameter (2.7 mm), silver-containing, hydrogel-coated catheters
were not commercially available for the mouse model used in this
study, preventing a direct comparison of the in vivo performance
of the polymer in this animal study with the same controls as the
in vitro experiments.
DISCUSSION
We have developed a method for the discovery of polymeric materials
resistant to bacterial attachment using high-throughput surface char-
acterization and chemometrics, which has identified simple chemical
moieties that reduce bacterial adhesion to surfaces. The new class of
materials discovered, which could not have been predicted from the
current understanding of bacteria-material interactions, are resistant
to bacterial attachment. An in vitro comparison after a 3-d incubation
revealed up to 67-fold less bacterial attachment and biofilm forma-
tion compared with a commercial uncoated medical grade silicone,
and a 30-fold lower bacterial attachment and biofilm formation than
a commercial silver hydrogel coating. The resistance of one of these
coatings to bacterial attachment was demonstrated in vivo using a
mouse foreign-body infection model, which confirms the potential
of the hit materials as coatings for indwelling biomedical devices. In
contrast to the use of coatings that incorporate molecules that are
toxic to and kill pathogens, the anti-attachment mechanism described
here does not place selective evolutionary pressure on organisms to
develop antimicrobial resistance.
By further developing this high-throughput screening approach
to include other bacterial species (including freshly isolated clinical
strains and pathogens specific to particular niches and different growth
conditions), new polymeric materials may be found that are resistant
to colonization by combinations of pathogens for medical devices,
water purification systems, food preparation surfaces and utensils, or
any other surface where bacterial adhesion is problematic.
Methods
Methods and any associated references are available in the online
version of the paper.
Note: Supplementary information is available in the online version of the paper.
Acknowledgments
Funding from the Wellcome Trust (grant no. 085245 and support from N. Shepherd)
and the Medical Research Council UK (for the in vivo work; grant no. G0802525)
is gratefully acknowledged. M. Alexander gratefully acknowledges the Royal
Society for the provision of his Wolfson Research Merit Award. Assistance with
ToF-SIMS measurements from D. Scurr is kindly acknowledged. Assistance with
the preparation of polymer for in vivo studies by E. Eaves, N. Nguyen and J. Li is
kindly acknowledged.

Author contributions
M.R.A., M.C.D., P.W., R.L. and D.G.A. conceived the high-throughput strategy.
Y.M., A.L.H. and J.Y. prepared the microarrays. J.Y. coated catheters with hit polymers.
J.Y. and A.L.H. performed the HT-SC. C.-Y.C. performed the biological assays with
support from S.A. D.J.I. prepared polymer for in vivo assays. J.L. and A.C. performed
the in vivo assays. A.L.H., C.-Y.C., J.Y., M.R.A., M.C.D., P.W., R.B., D.G.A. and R.L.
contributed to the analysis of data and the writing of the manuscript.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Published online at http://www.nature.com/doifinder/10.1038/nbt.2316. cell adhesion revealed by high-throughput surface characterization of combinatorial
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
1. Davies, D. Understanding biofilm resistance to antibacterial agents. Nat. Rev. Drug
Discov. 2, 114–122 (2003).
2. Costerton, J.W., Stewart, P.S. & Greenberg, E.P. Bacterial biofilms: a common cause
of persistent infections. Science 284, 1318–1322 (1999).
3. Smith, A.W. Biofilms and antibiotic therapy: is there a role for combating bacterial
resistance by the use of novel drug delivery systems? Adv. Drug Deliv. Rev. 57,
1539–1550 (2005).
4. Danese, P.N. Antibiofilm approaches: prevention of catheter colonization. Chem.
Biol. 9, 873–880 (2002).
5. Krol, J.E. et al. Increased transfer of a multidrug resistance plasmid in Escherichia coli
biofilms at the air-liquid interface. Appl. Environ. Microbiol. 77, 5079–5088 (2011).
© 2012 Nature America, Inc. All rights reserved.
6. Darouiche, R.O. et al. A comparison of two antimicrobial-impregnated central venous
catheters. N. Engl. J. Med. 340, 1–8 (1999).
7. Raad, I. et al. Central venous catheters coated with minocycline and rifampin for
the prevention of catheter-related colonization and bloodstream infections. A
randomized, double-blind trial. Ann. Intern. Med. 127, 267–274 (1997).
8. Yorganci, K., Krepel, C., Weigelt, J.A. & Edmiston, C.E. Activity of antibacterial
impregnated central venous catheters against Klebsiella pneumoniae. Intensive Care
Med. 28, 438–442 (2002).
9. Jaeger, K. et al. Efficacy of a benzalkonium chloride-impregnated central venous
catheter to prevent catheter-associated infection in cancer patients. Chemotherapy
47, 50–55 (2001).
10. Greenfeld, J.I. et al. Decreased bacterial adherence and biofilm formation on
chlorhexidine and silver sulfadiazine-impregnated central venous catheters
implanted in swine. Crit. Care Med. 23, 894–900 (1995).
11. Guay, D.R. An update on the role of nitrofurans in the management of urinary tract
infections. Drugs 61, 353–364 (2001).
12. Caillier, L. et al. Synthesis and antimicrobial properties of polymerizable quaternary
ammoniums. Eur. J. Med. Chem. 44, 3201–3208 (2009).
13. Darouiche, R.O., Mansouri, M.D., Gawande, P.V. & Madhyastha, S. Efficacy of
combination of chlorhexidine and protamine sulphate against device-associated
pathogens. J. Antimicrob. Chemother. 61, 651–657 (2008).
14. Li, P. et al. A polycationic antimicrobial and biocompatible hydrogel with microbe
membrane suctioning ability. Nat. Mater. 10, 149–156 (2011).
15. Costa, F., Carvalho, I.F., Montelaro, R.C., Gomes, P. & Martins, M.C.L. Covalent
npg
immobilization of antimicrobial peptides (AMPs) onto biomaterial surfaces. Acta
Biomater. 7, 1431–1440 (2011).
16. Monds, R.D. & O’Toole, G.A. The developmental model of microbial biofilms: ten
years of a paradigm up for review. Trends Microbiol. 17, 73–87 (2009).
17. Holmes, P.F. et al. Surface-modified nanoparticles as a new, versatile, and
mechanically robust nonadhesive coating: suppression of protein adsorption and
bacterial adhesion. J. Biomed. Mater. Res. A 91A, 824–833 (2009).
18. Cheng, G. et al. Zwitterionic carboxybetaine polymer surfaces and their resistance
to long-term biofilm formation. Biomaterials 30, 5234–5240 (2009).
19. Cheng, G., Zhang, Z., Chen, S.F., Bryers, J.D. & Jiang, S.Y. Inhibition of bacterial adhesion
and biofilm formation on zwitterionic surfaces. Biomaterials 28, 4192–4199 (2007).
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 875
Articles
20. Hook, A.L. et al. High-throughput methods applied in biomaterial development and
discovery. Biomaterials 31, 187–198 (2010).
21. Pernagallo, S., Wu, M., Gallagher, M.P. & Bradley, M. Colonising new frontiers-
microarrays reveal biofilm modulating polymers. J. Mater. Chem. 21, 96–101
(2011).
22. Anderson, D.G., Levenberg, S. & Langer, R. Nanoliter-scale synthesis of arrayed
biomaterials and application to human embryonic stem cells. Nat. Biotechnol. 22,
863–866 (2004).
23. Mei, Y. et al. Mapping the interactions among biomaterials, adsorbed proteins, and
human embryonic stem cells. Adv. Mater. 21, 2781–2786 (2009).
24. Mei, Y. et al. Combinatorial development of biomaterials for clonal growth of human
pluripotent stem cells. Nat. Mater. 9, 768–778 (2010).
25. Yang, J. et al. Polymer surface functionalities that control human embryoid body
material microarrays. Biomaterials 31, 8827–8838 (2010).
26. Berger, H., Hacker, J., Juarez, A., Hughes, C. & Goebel, W. Cloning of the
chromosomal determinants encoding hemolysin production and mannose-resistant
hemagglutination in Escherichia coli. J. Bacteriol. 152, 1241–1247 (1982).
27. Hidron, A.I. et al. Antimicrobial-resistant pathogens associated with healthcare-
associated infections: annual summary of data reported to the national healthcare
safety network at the Centers for Disease Control and Prevention, 2006–2007.
Infect. Control Hosp. Epidemiol. 29, 996–1011 (2008).
28. Schumm, K. & Lam, T.B. Types of urethral catheters for management of short-term
voiding problems in hospitalised adults. Cochrane DB. Syst. Rev. 16, CD004013
(2008).
29. Katsikogianni, M. & Missirlis, Y.F. Concise review of mechanisms of bacterial
adhesion to biomaterials and of techniques used in estimating bacteria-material
interactions. Eur. Cell. Mater. 8, 37–57 (2004).
30. O’Toole, G., Kaplan, H.B. & Kolter, R. Biofilm formation as microbial development.
Annu. Rev. Microbiol. 54, 49–79 (2000).
31. Flemming, H.C. & Wingender, J. The biofilm matrix. Nat. Rev. Microbiol. 8,
623–633 (2010).
32. Urquhart, A.J. et al. High-throughput surface characterisation of a combinatorial
material library. Adv. Mater. (Deerfield Beach Fla.) 19, 2486–2491 (2007).
33. Davies, M.C. et al. High-throughput surface characterization: a review of a new tool
for screening prospective biomedical material arrays. J. Drug Target. 18, 741–751
(2010).
34. Hook, A.L. et al. Polymers with hydro-responsive topography identified using
high-throughput AFM of an acrylate microarray. Soft Matter 7, 7194–7197
(2011).
35. Taylor, M. et al. Partial least squares regression as a powerful tool for investigating
large combinatorial polymer libraries. Surf. Interface Anal. 41, 127–135 (2009).
36. Urquhart, A.J. et al. TOF-SIMS analysis of a 576 micropatterned copolymer array to
reveal surface moieties that control wettability. Anal. Chem. 80, 135–142 (2008).
37. Dalton, H.M. & March, P.E. Molecular genetics of bacterial adhesion and biofouling.
in Handbook of Bacterial Adhesion: Principles, Methods, and Applications (eds.
An, Y.H. & Friedman, R.J.) 43–51 (Humana Press, 2000).
38. Ong, Y.L., Razatos, A., Georgiou, G. & Sharma, M.M. Adhesion forces between
E. coli bacteria and biomaterial surfaces. Langmuir 15, 2719–2725 (1999).
39. Conrad, J.C.C.J.C. et al. Flagella and pili-mediated near-surface single-cell motility
mechanisms in P. aeruginosa. Biophys. J. 100, 1608–1616 (2011).
40. Williams, P., Winzer, K., Chan, W.C. & Camara, M. Look who’s talking: communication
and quorum sensing in the bacterial world. Philos. Trans. R. Soc. Lond. B. Biol.
Sci. 362, 1119–1134 (2007).
41. Ferrieres, L. & Clarke, D.J. The RcsC sensor kinase is required for normal biofilm
formation in Escherichia coli K-12 and controls the expression of a regulon in
response to growth on a solid surface. Mol. Microbiol. 50, 1665–1682 (2003).
42. Gilbert, K.B., Kim, T.H., Gupta, R., Greenberg, E.P. & Schuster, M. Global position
analysis of the Pseudomonas aeruginosa quorum-sensing transcription factor LasR.
Mol. Microbiol. 73, 1072–1085 (2009).
43. Harder, P., Grunze, M., Dahint, R., Whitesides, G.M. & Laibinis, P.E. Molecular
conformation in oligo(ethylene glycol)-terminated self-assembled monolayers on gold
and silver surfaces determines their ability to resist protein adsorption. J. Phys.
Chem. B. 102, 426–436 (1998).
 ONLINE METHODS
Polymer array synthesis. Polymer microarrays were synthesized using meth-
ods previously described22. Monomers were purchased from Aldrich, Scientific
Polymers and Polysciences, and printed onto epoxy-coated slides (Xenopore),
dip-coated into 4% (w/v) pHEMA (Aldrich) using 946MP6B pins (ArrayIt)
and a Pixsys 5500 robot (Cartesian) or a XYZ3200 dispensing workstation
(Biodot). The arrays were dried at <50 mTorr for at least 7 d.
High-throughput surface characterization. Arrays were characterized by
AFM, WCA, XPS and ToF-SIMS. ToF-SIMS measurements were conducted
on an ION-ToF IV instrument operated using a monoisotopic Bi3+ primary
ion source operated at 25 kV and in ‘bunched mode’. A 1 pA primary ion beam
was rastered, and both positive and negative secondary ions were collected
from a 100 × 100 µm area. Ion masses were determined using a high resolu-
tion Time-of-Flight analyzer. The typical mass resolution (at m/z 41) was
just over 6,000. XPS was carried out on a Kratos Axis Ultra instrument using
monochromated Al Ka radiation (1486.6eV), 15 mA emission current, 10 kV
anode potential and a charge-compensating electron flood. High-resolution
core levels were acquired at a pass energy of 20 eV. WCA was measured
using the sessile drop method on an automated Krüss DSA 100 instrument44.
A water drop with a volume of ~400 picoliter was used. AFM measurements
were taken using a Nanoscope 3000A instrument in tapping mode. Silicon tips
with a resonant frequency of ~300 kHz and a force constant of 40 N/m were
© 2012 Nature America, Inc. All rights reserved.
used (Tap300Al, Budget Sensors) for dry state and silicon nitride tips with a
resonant frequency of 7.5 kHz were used for fluid measurements. 5 µm regions
of the polymer were taken and the root mean square (RMS) roughness was
measured across this region. The ToF-SIMS spectra data were analyzed using
principle component analysis (PCA), and the correlation between ToF-SIMS
spectra and bacterial adhesion was analyzed using partial least-squares (PLS)
regression35. Both multivariate analysis methods were carried out using the
Eigenvector PLS_Toolbox 3.5.
Scale-up of materials. Selected compositions were scaled up to 10-mm pol-
ymer coupons. These were prepared by casting 5 µl of monomer solution
(75% (v/v) monomer, 25% (v/v) DMF and 1% (w/v) 2,2-dimethoxy-2-phenyl
acetophenone) onto epoxy-functionalized slides (Xenopore) dip-coated with
4% (w/v) pHEMA in ethanol. For growth inhibition studies, 40 µl of monomer
solution was pipetted into a well of a 96-well microwell plate. Samples were
irradiated with UV (365 nm) for 10 min to initiate polymerization with O2 <
2,000 p.p.m. The samples were dried at <50 mTorr for at least 7 d. To produce
coated catheters, 4-cm long silicone lengths were cut from a silicone Foley uri-
nary catheters (Bard, outer diameter 7.3 mm –in vitro or 2.7 mm -in vivo). The
npg
inside and outside surface was oxygen plasma treated for 5 min at 50 W. For
in vitro use, plasma-treated catheters were immediately immersed in monomer
solution for 10 s and blotted to remove excessive monomer solution before
photopolymerization using UV (365 nm) for 1 min, with O2 < 2,000 p.p.m. For
in vivo studies, plasma-activated catheters were dip-coated with a 20% poly-
mer solution in dichloromethane. The samples were then dried at <50 mTorr
for at least 7 d. Polymer for in vivo studies was prepared by catalytic chain
transfer polymerization. The resultant polymer solution was used for coating
silicone catheters without further purification. Uncoated silicone catheters
and BactiGuard (silver containing hydrogel)-coated latex catheters (Bardex)
were used as controls. SEM imaging of coated catheters was conducted on a
Jeol 6060LV variable pressure SEM. Samples were gold-coated before imaging
using a Leica EM SCD005 sputter coater.
Bacterial growth conditions. Three different bacterial species, P. aeruginosa
PAO1, S. aureus 8325-4 and UPEC were routinely grown on either LB (Luria-
Bertani, Oxoid, UK) agar plates at 37 °C or in broth at 37 °C with 200 r.p.m.
shaking. Three constitutively GFP-expressing plasmids, pGFP45, pSB2019 and
pSB2020 (ref. 46), were transformed into P. aeruginosa PA01, S. aureus 8325-
4 and UPEC, respectively, and maintained by adding appropriate antibiotics
to the culture media. RPMI-1640 chemically defined medium (Sigma, UK)
and artificial urine47 were used in biofilm experiment for standardizing the
conditions and mimicking CAUTI, respectively. For the in vivo study S. aureus
Xen29 (Caliper) were routinely grown on Tryptic soya Broth (Oxoid, UK) at
37 °C until an OD of 0.8, washed twice in phosphate buffered saline (PBS,
nature biotechnology
Oxoid, UK) and stored in aliquots of 1 × 109 c.f.u. in PBS/20% glycerol. These
were then thawed and diluted with PBS before injection into the lumen of
the catheter47.
Prior to incubation with the bacteria, the microarray slides were washed
in distilled H2O for 10 min, air-dried and UV sterilized. Artificial urine–
­conditioned slides were incubated for 72 h at 37 °C in 15 ml of artificial urine
with 5% CO2. Subsequently, slides were washed three times in RPMI-1640
medium or artificial urine. Bacteria were grown on polymer slides under
­similar conditions to those previously described48,49. Briefly, UV-sterilized pol-
ymer slides were incubated in 15 ml medium inoculated with diluted (OD 600 =
0.01) GFP-tagged bacteria from overnight cultures grown at 37 °C with
60 r.p.m. shaking for 24 h or 72 h. As growth medium controls, the slides were
also incubated without bacteria. At the desired time points, the slides were
removed and washed three times with 15 ml PBS at room temperature for
5 min. After rinsing with distilled H2O to remove salts and air dried, the fluo-
rescent images from the slides incubated in medium only and medium con-
taining bacteria were acquired using a GenePix Autoloader 4200AL Scanner
(Molecular Devices, US) with a 488 nm excitation laser and a blue emission
filter (510–560 nm). The total fluorescence intensity from polymer spots was
acquired using GenePix Pro 6 software (Molecular Devices, US). A similar
bacterial assay was also applied to scaled-up coupons and 4-cm sections of
coated catheters. After washing with distilled H2O, the coupons or catheters
were stained with 20 µM SYTO17 dye (Invitrogen, UK) at room temperature
for 30 min. After air drying, the samples were examined using a Carl Zeiss
LSM 700 Laser Scanning Microscope with ZEN 2009 imaging software (Carl
Zeiss, Germany). The area covered by bacteria on the surface was analyzed
using open source ImageJ 1.44 software (National Institutes of Health, US).
The viability of bacteria attached to polymer surfaces was assessed by live/dead
staining. Briefly bacteria were stained with 10 µM SYTO 9 green-fluorescent
dye for live bacteria and 60 µM propidium iodide red-fluorescent dye for cell
membrane damaged (dead) bacteria. After staining at room temperature for
30 min, the samples were rinsed with distilled H 2O, air dried and observed
using Laser Scanning Confocal Microscopy. To evaluate the growth inhibitory
properties of the hit polymers, polymer-coated wells were UV sterilized for
15 min and inoculated with bacteria (OD600 = 0.01) from overnight cultures.
The OD was monitored by an Infinite 200 microplate reader (Tecan, UK) at
37 °C every 30 min for 24 h to obtain the respective growth curves.
The fluorescence signal (F) from each bacterial pathogen was determined
using equation (1), where F is the fluorescence intensity measured per unit
area by the laser scanner after incubation with bacteria, and Fcontrol is the
fluorescence intensity measured per unit area by the laser scanner measured
on a control slide consisting of a replica array that was incubated in media for
72 h without bacteria. For polymers where F was below the limit of detection,
F was made to equal 0 (Supplementary Discussion).
F = Ftest − Fcontrol (1)
The bacterial performance () was determined using equation (2), where
the subscript to the F indicates the bacterial pathogen and the F max is the
maximum fluorescence signal measured on any spot on the array for a given
pathogen. Artificial urine is abbreviated to ‘au’. Note that  reported for
second-generation arrays and scaled-up samples did not include results of
UPEC in artificial urine.
i = (F PA / F PA max + F SA / F SA max
+ F UPEC /F UPEC max + F UPEC inau /F UPEC inau max
(2)
+ F UPEC onau conditioned /F UPEC onau conditionedmax )
÷ 5 × 100
Catalytic chain transfer polymerization. Polymer for in vivo studies was
prepared by catalytic chain transfer polymerization: a degassed (N2) solu-
tion of 15% (v/v) ethylene glycol dicyclopentenyl ether acrylate, 5% (v/v)
di(ethylene glycol) methyl ether methacrylate (DEGMA), 0.5% (w/w)
2,2′-azobis-(2,4-dimethyl-4-methoxyvaleronitrile) and 2,000 p.p.m. tetra­
phenylcobaloxime boron fluoride (bis[(difluoroboryl) diphenylglyoxima
to]cobalt(II)) in DCM was placed in a Schlenk flask at 35 °C for 24 h.
The vessel was then placed into an ice bath for 5 min and exposed to air
doi:10.1038/nbt.2316
 to terminate the polymerization. For analysis, the DCM was removed at
reduced pressure at room temperature.
Polymer characterization. 1H and 13C nuclear magnetic resonance (NMR)
spectra were acquired in deuterated chloroform on both Bruker DPX (300
MHz) and AV (400 MHz) instruments. Differential scanning calorimetry
(DSC) measurements were conducted on a DSC Q2000 (TA Instruments).
Samples were placed in a Tzero Hermetic Pan with a pressed lid. Heat flow was
measured against a sealed empty pan. Sample was cycled between −40 °C and
90 °C 3 times at 10 °C/min. Number-average molecular weight (Mn), weight-
average molecular weight (Mw), and dispersity (PD) were obtained by gel
permeation chromatography (GPC) with a fitted IR detector. Polymer samples
(7 mg/ml) dissolved in tetrahydrofuran were flowed (1 ml/min) through
a PLgel 5 mm Guard column (Polymer Laboratories) and two PLgel 5 mm
MIXED-C columns (Polymer Laboratories) at 40 °C. Fatigue testing was
carried out on a Mayes Servohydraulic testing machine with a set of Instron
wedge grips with vee jaws. The sample length between the two grips was 6 cm.
The sample was bent 500 times and the total distance of movement between
the two grips during bending was 1cm.
Murine catheter implant model. To evaluate the in vivo resistance of polymer
materials to bacterial attachment, the real time, non-invasive catheter foreign
body implant model was used50. A 1-cm coated catheter segment was inserted
© 2012 Nature America, Inc. All rights reserved.
subcutaneously into female BALB/c mice, 19–22 g (Charles River). The animals
were allowed to recover with regular observation for 24 h before anesthesia
with isoflurane and injection of 1 × 105 S. aureus Xen29 (Caliper Life Sciences
Inc) in 50 µl PBS into the lumen of the catheter using a 31 gauge needle and
syringe. Mice were imaged for up to 5 min before (to assess background lumi-
nescence) and 10 min after initial bacterial inoculation, and then every 24 h,
using an IVIS spectrum camera (Caliper). EMLA cream was applied daily after
imaging and weight and clinical condition of animals recorded.
The total photon emissions from the catheter implantation site were quan-
tified by using the living image software package (Xenogeny Corp.), over a
4-d period. At day 4 the mice were humanely killed and the catheter and
surrounding tissue removed, the mouse kidneys and spleen harvested, and
the number of S. aureus Xun 29 colony-forming units (cfu) present deter-
mined using standard procedures50. The cfu counts were normalized to the
mass of tissue taken. Reported bioluminescence values had the background
npg
doi:10.1038/nbt.2316
luminescence measured from uninoculated inserted catheters subtracted.
In each experiment, three mice were implanted with control or coated cath-
eters and the experiment repeated on three occasions with data from n = 9 for
each group pooled for statistical analysis.
In pilot studies, Nano-ct, computed tomographic scans were used to con-
firm the subcutaneous localization of the catheter and were acquired using a
Nanoect/CT (Bioscan). Briefly, after catheter insertion, mice were ­anesthetized
using isoflurane and scanned at 45 kVP, using high-resolution parameters. The
resulting images were reconstructed and analyzed using invivoscope software.
Surgical details of murine catheter implant model. All animal work was
approved following local ethical review at Nottingham University and per-
formed under Home Office license authority. Female BALB/c mice, 19–22 g
(Charles River) were housed in IVCs under a 12 h light cycle, with food and
water ad libitum. Briefly, the experimental procedure was carried out as fol-
lows. One hour before catheter implantation, analgesic (Rymadil (Pfizer)
2.5 mg/kg) was administered by subcutaneous injection. Animals were anes-
thetized using isoflurane, hair on one flank removed by shaving and the shaved
area sterilized with hydrex clear (Ecolab). A 10-mm incision was made in the
flank, a subcutaneous pocket created and 1-cm segment of catheter (TReleflex
scale Fr8) inserted. The incision was then closed with Vetbond (3MM) and
EMLA cream (AstraZeneca) applied to the implantation site.
44. Taylor, M., Urquhart, A.J., Zelzer, M., Davies, M.C. & Alexander, M.R. Picoliter
water contact angle measurement on polymers. Langmuir 23, 6875–6878
(2007).
45. Messina, M. Gene Regulation in Pseudomonas Aeruginosa: from Environmental
Signals to Responses via Global Post-Transcriptional Control and Intracellular
Messaging. PhD thesis, Univ. Nottingham (2010).
46. Qazi, S.N.A., Rees, C.E.D., Mellits, K.H. & Hill, P.J. Development of GFP vectors
for expression in Listeria monocytogenes and other low G+C gram positive bacteria.
Microb. Ecol. 41, 301–309 (2001).
47. Brooks, T. & Keevil, C.W. A simple artificial urine for the growth of urinary pathogens.
Lett. Appl. Microbiol. 24, 203–206 (1997).
48. Diggle, S.P. et al. The galactophilic lectin, LecA, contributes to biofilm development
in Pseudomonas aeruginosa. Environ. Microbiol. 8, 1095–1104 (2006).
49. Johansson, E.M.V. et al. Inhibition and dispersion of Pseudomonas aeruginosa
biofilms by glycopeptide dendrimers targeting the fucose-specific lectin LecB. Chem.
Biol. 15, 1249–1257 (2008).
50. Kuklin, N.A. et al. Real-time monitoring of bacterial infection in vivo: Development
of bioluminescent staphylococcal foreign-body and deep-thigh-wound mouse
infection models. Antimicrob. Agents Chemother. 47, 2740–2748 (2003).
nature biotechnology
 letters

Directed differentiation of human pluripotent stem

cells into mature airway epithelia expressing functional
CFTR protein
Amy P Wong1,2, Christine E Bear3, Stephanie Chin3, Peter Pasceri1, Tadeo O Thompson2, Ling-Jun Huan3,
Felix Ratjen4,5, James Ellis1,2,6 & Janet Rossant1,6,7

Cystic fibrosis (CF) is a fatal genetic disease caused by Because lung forms from definitive endoderm, we first differenti-
mutations in the CFTR (cystic fibrosis transmembrane ated hESCs toward definitive endoderm using a previously described
© 2012 Nature America, Inc. All rights reserved.

conductance regulator) gene, which regulates chloride and
water transport across all epithelia and affects multiple
organs, including the lungs. Here we report an in vitro directed
differentiation protocol for generating functional CFTR-
expressing airway epithelia from human embryonic stem cells.
Carefully timed treatment by exogenous growth factors that
mimic endoderm developmental pathways in vivo followed by
air-liquid interface culture results in maturation of patches of
tight junction–coupled differentiated airway epithelial cells
that demonstrate active CFTR transport function. As a proof
of concept, treatment of CF patient induced pluripotent stem
cell–derived epithelial cells with a small-molecule compound
to correct for the common CF processing mutation resulted
in enhanced plasma membrane localization of mature CFTR
protein. Our study provides a method for generating patient-
specific airway epithelial cells for disease modeling and
in vitro drug testing.
npg
method7, based on developmental pathways of endoderm formation at
gastrulation6. Treatment with activin-A and WNT3A for 4 d was suf-
ficient to induce a large percentage of cells into definitive endoderm as
determined by co-expression of CXCR4 and cKIT, and the endoderm
transcription factors FOXA2 and SOX17 (Supplementary Fig. 1).
After gastrulation, anterior-posterior signals pattern the primitive
gut tube into distinct regions6. The developing cardiac mesoderm
patterns the anterior ventral foregut endoderm through secretion
of fibroblast growth factor-2 (FGF2)8. High concentration of FGF2
induces an NKX2.1 expression domain typical of the early lung endo-
derm9. Sonic hedgehog (SHH) signaling promotes embryonic lung
growth and suppresses pancreatic development10,11. Thus, to promote
anterior foregut identity and specify lung cell fate, definitive endo-
derm cells were treated with FGF2 and SHH (Supplementary Fig. 2).
After 5 d of exposure, 85% of the cells expressed the pan-endoderm
transcription factor FOXA2. The majority (78%) of the cells co-
expressed the pan-epithelial marker EpCAM and the transcription

factor NKX2.1. Upregulation of genes associated with anterior foregut

Efforts to differentiate human embryonic stem cells (hESCs) into
lung epithelia have generated cells that express distal airway epithelial
pheno­types expressing surfactant protein-C1–3. These reports relied
on spontaneous mixed-lineage embryoid bodies to differentiate hESCs
directly into lung endoderm with low efficiency and generated mostly
cells expressing distal alveolar markers. A recent study showed a step-
wise generation of lung endoderm progenitors from human induced
pluripotent stem cells (iPSCs) but failed to generate mature proximal
and distal lung epithelial phenotypes4. To date, no studies have been
able to generate proximal conducting airway epithelia with functional
polarized CFTR. Therefore, we developed a method to recapitulate the
sequential processes that progressively restrict progenitor cells from
endoderm to proximal lineage-specific lung epithelia5,6 and were able
to generate functional proximal conducting airway epithelia express-
ing CFTR from human pluripotent stem cells.
endoderm transcription factors—SOX2 and NKX2.1—as well as
pharyngeal endoderm FOXG and thyroid TG and PAX9 was observed
by means of RT-PCR. The posterior hindgut gene CDX2 was not
detected. NKX2.1 is a good marker for lung endoderm but is also
expressed in the thyroid and forebrain12. Although other thyroid
genes were seen, notably, the ectoderm PAX6 was also not detected,
which excludes the possibility of forebrain-derived NKX2.1 expres-
sion. The definitive endoderm gene SOX17 was downregulated,
but some expression of transcription factors indicative of liver
(HNF4) and pancreas (NKX6.1 and PDX) was detected, suggesting
that other endoderm lineages were also present. Nonetheless, FGF2
and SHH can efficiently induce anterior foregut derivatives from
definitive endoderm.
Gain- and loss-of-function studies in mouse embryonic lung organ
cultures combined with studies in null and transgenic mouse lines

1Program in Developmental & Stem Cell Biology, Hospital for Sick Children, Toronto, Ontario, Canada. 2Ontario Human Induced Pluripotent Stem Cell Facility,
Toronto, Ontario, Canada. 3Program in Molecular Structure and Function, Hospital for Sick Children, Toronto, Ontario, Canada. 4Program in Physiology & Experimental
Medicine, Hospital for Sick Children, Toronto, Ontario, Canada. 5Division of Respiratory Medicine, Hospital for Sick Children, Toronto, Ontario, Canada. 6Department
of Molecular Genetics, University of Toronto, Toronto, Ontario, Canada. 7Department of Obstetrics and Gynaecology, University of Toronto, Toronto, Ontario, Canada.
Correspondence should be addressed to J.R. (janet.rossant@sickkids.ca) or J.E. (jellis@sickkids.ca).

Received 22 May; accepted 13 July; published online 26 August 2012; doi:10.1038/nbt.2328

876 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 letters

a hPSCs
Endoderm
progenitors
Anterior foregut
progenitors
Early proximal lung
progenitors
b Group Treatment

A Undifferentiated PSCs

B Anterior foregut endoderm progenitor cells

D0 D4 D9 D15
C + FGF7 (50 ng/ml) + FGF10 (50 ng/ml)
FOXJ1 (ciliated cells)
SOX9 (distal tip cells)
Activin A (100 ng/ml) FGF2 (500 ng/ml) Days 9–14: D + FGF7 (50 ng/ml) + FGF10 (50 ng/ml) + BMP4 (5 ng/ml)
Trp63 and KRT14
Wnt3a (25 ng/ml) SHH (50 ng/ml) FGF7 (50 ng/ml)
(basal cells)
FGF10 (50 ng/ml) E + FGF7 (50 ng/ml) + FGF10 (50 ng/ml) + BMP4 (10 ng/ml)
SCGB1A1 (Clara cells)
BMP4
SFTPC (type II cells)
CFTR (chloride channel) F + FGF7 (50 ng/ml) + FGF10 (50 ng/ml) + BMP4 (50 ng/ml)
c

Relative expression

Relative expression

Relative expression
Relative expression
KRT5 TRP63 FOXJ1 SOX17
1.5 1.0 1.0 10.0

(to adult lung)

(to adult lung)

(to adult lung)

(to adult lung)
0.8 0.8 5.0
1.0 0.6 0.6
Figure 1  Low concentration of BMP4 upregulates 0.5 0.4 0.4 0.2
0.2 0.2 0.1
genes associated with early proximal lung
progenitors. (a,b) Differentiation to immature

A
B
C
D
E
Adtal l F
tl g
g

A
B
C
D
E
Ad al lu F
ul ng
g

A
B
C
D
E
Adtal l F
tl g
g

A
B
C
D
E
Ad al l F
ul ung
g
ul un
un

un

ul un
un

un
lung endoderm cells. (c) Expression of airway

tl

tl
t

t
Fe

Fe

Fe

Fe
genes such as KRT5 and TRP63 (basal cell

Relative expression
Relative expression

Relative expression

Relative expression
marker), FOXJ1 and SOX17 (ciliated cell markers), 5.0
NKX2.1 1.5
CFTR
3.0
FOXA2
1.0
MUC5AC

(to adult lung)

(to adult lung)

(to adult lung)

(to adult lung)

NKX2.1 (earliest marker of the lung endoderm), 4.0 1.0
2.0
0.8
3.0 0.5 0.6
CFTR and the pan-endoderm marker FOXA2 was 2.0 0.2 0.4
1.0
upregulated with 10 ng/ml of BMP4 (treatment D). 1.0 0.1 0.2
The airway marker MUC5AC (goblet cell marker)
A
B
C
D
E
Ad al l F
tl g
g

A
B
C
D
E
Adtal l F
tl g
g

A
B
C
D
E
Ad al lu F
ul ng
g

A
B
C
D
E
Ad al l F
ul ung
g
was not detected. In addition, the type II alveolar
© 2012 Nature America, Inc. All rights reserved.

ul un
un

ul un
un

un

un
tl

tl
cell marker surfactant protein-C (SFTPC) and
t

t
Fe

Fe

Fe

Fe
the Clara cell marker SCGB1A1 were not detected.

Relative expression
Relative expression

1.0 SFTPC Relative expression 1.0 SCGB1A1 25 SOX9

SOX9, a marker of the distal tip progenitors found

(to adult lung)

(to adult lung)

0.8 (to adult lung) 0.8 20

15
in the developing embryonic lung, was also detected. 0.6 0.6 10
0.4 0.4 5.0
Genes were normalized to the housekeeping gene 2.5
0.2 0.2
β-ACTIN and expressed relative to adult tissue
positive control RNA. Error bars, mean ± s.e.m.
A
B
C
D
E
Ad al l F
ul ung
g

A
B
C
D
E
Adtal l F
tl g
g

A
B
C
D
E
Ad al l F
ul ung
g
un

ul un
un

un
tl

tl
(n = 3 experiments).
t

t
Fe

Fe

have identified specific growth factors important for lung develop-
ment from the NKX2.1-expressing endoderm5, including FGF and
BMP4. FGF10 is a key growth factor expressed by the mesenchyme
at the earliest stage of lung development and stimulates lung bud
outgrowth and organogenesis13. FGF7, also expressed by the mesen-
chyme, is mainly involved in stimulating fluid secretion in the lungs
but has a role in epithelial cell growth14. We found that the combina-
tion of FGF7 and FGF10 at 50 ng/ml augmented mRNA expression
of the transcription factor genes NKX2.1 and FOXA2 compared to
npg
Fe
not detected. SOX9, a distal tip marker found in the developing lung
bud, was also detected. Therefore, a combination of FGF7, FGF10
and low concentrations of BMP4 can induce upregulation of some
conducting airway cell lineages from anterior foregut endoderm.
To further promote airway differentiation, we added FGF18, which
enhances proximal but not distal airway formation and plays a role
in increasing the size of the conducting airways following matura-
tion of the epithelium19 (Fig. 2a,b). This resulted in further upreg-
ulation of airway genes KRT5, TRP63, FOXJ1, SOX17, MUC5AC

definitive endoderm levels (Supplementary Fig. 3). As the lung bud
actively branches, the bud stalk (future conducting airways) matures
in an environment where signals that drive bud tip outgrowth are
reduced or repressed15. High concentrations of BMP4 stimulate a
distal cell fate whereas low concentrations promote a proximal cell
fate16. Retinoic acid signaling upregulates FGF10 in the developing
lung mesenchyme to stimulate lung bud outgrowth17 but it also plays
a key role in inducing alveolar epithelial cell fate and surfactant pro-
duction during the later phase of lung development18. Therefore, to
enhance proximal airway fate and suppress distal cell fate, we incu-
bated cultures with varying levels of BMP4 and in the absence of
retinoic acid. The hESC-derived anterior foregut cells were exposed
to FGF10, FGF7 and varying concentrations of BMP4 to determine
the optimal concentration of BMP4 that would induce proximal cell
fate (Fig. 1a,b). RT-PCR showed that low concentrations of BMP4
induced upregulated gene expression of airway cell genes such as
KRT5 and TRP63 (also known as TP63; P63, basal cell), FOXJ1 and
SOX17 (glial cell), NKX2.1, CFTR and pan-endoderm marker FOXA2
(Fig. 1c). Another airway marker, MUC5AC (goblet cell marker), was
not detected. In addition, the distal type II alveolar cell marker SFTPC
(encoding surfactant protein-C) and Clara cell secretory marker
SCGB1A1 (more abundant in the smaller conducting airways) were
and CFTR, lower levels of SCGB1A1 and little detection of SFTPC
(Fig. 2c). NKX2.1 and FOXA2 and the distal tip progenitor SOX9
were downregulated. Relative to their adult tissue counterparts,
other endoderm lineage markers—AFP (liver), PDX1, TG and PAX9
(pharyngeal endoderm)—were not detected at this stage. Moreover,
flow cytometric quantification revealed cells that expressed pan-
cytokeratin (panKRT, 33%) CFTR (30%), FOXJ1 (36%) and NKX2.1
(32%) (Fig. 2d), suggesting that at least one-third of the cells in the
culture were of the ciliated CFTR-expressing airway phenotype. Over
50% of the cells were P63+, suggesting that the vast majority of the
cells were potentially basal cell progenitors, previously shown to
give rise to other proximal airway lineages20. Significant cell pro-
liferation, as measured by BrdU incorporation, occurred during
the first 14 d of differentiation and declined by the late proximal
specification stage (D19, Fig. 2e,f). The increased expression of
proximal lineage markers on day 19 is consistent with the reduc-
tion in cell proliferation as differentiation progressed. Collectively,
our data support a mechanism for airway development in which
FGF7, FGF10, BMP4 and FGF18 are required to promote airway
lineage development.
Notably, this method of directed differentiation was broadly
applicable to several pluripotent cell lines with varying efficiencies

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 877

 letters

a Endoderm Anterior foregut Immature proximal

c
KRT5 TRP63 FOXJ1 SOX17

Relative expression

Relative expression

Relative expression

Relative expression
hPSC progenitors progenitors Proximal lung cells
1.5 1.0 1.0 10.0
**

(to adult lung)

(to adult lung)

(to adult lung)

(to adult lung)

specification 7.5
0.8 * 0.8 5.0
1.0 0.6 0.6 2.5
1.0
D0 D4 D9 D15 D20 0.5 0.4 0.4 ***
0.2 0.2 5.0

Activin A (100 ng/ml) FGF2 (500 ng/ml) Days 9–14: Days 15–19:

tl g
g

tl g
g
tl g
g
Wnt3a (25 ng/ml) SHH (50 ng/ml)

tl g
g
G

A
B
G
H
Adtal l I

A
B
G
H
Adtal l I
H

A
B
A
B

Adtal l I
Adtal l I

ul un
un

ul un
un
ul un
un

ul un
un
FGF7 (50 ng/ml) FGF7 (10 ng/ml)
FGF10 (50 ng/ml) FGF10 (10 ng/ml)

Fe

Fe
Fe

Fe
BMP4 (5 ng/ml) FGF18
SCGB1A1

Relative expression
Relative expression

Relative expression
Relative expression
b
MUC5AC CFTR SFTPC
1.0 2.0 1.0 1.0

(to adult lung)

(to adult lung)

(to adult lung)

(to adult lung)
0.8 1.5 0.8 0.8
0.6 0.6 0.6
1.0
Group Treatment 0.4 ** * 0.4 0.4
0.2 0.5 0.2 0.2
A Undifferentiated PSCs

tl g
g
tl g
g

tl g
g

tl g
g
G

G
H
H

H
A
B

A
B

A
B

A
B
Adtal l I
Adtal l I

Adtal l I

Adtal l I
ul un
un
ul un
un

ul un
un

ul un
un
B Anterior foregut endoderm progenitor cells

Fe
Fe

Fe

Fe
G FGF7 + FGF10 (10 ng/ml) + FGF18 (5 ng/ml)

Relative expression

Relative expression

Relative expression
NKX2.1 FOXA2 SOX9 AFP

Relative expression
5.0 3.0 25 1.0

(to adult lung)

(to adult lung)

(to adult lung)

(to fetal liver)

4.0 20 0.8
H FGF7 + FGF10 (10 ng/ml) + FGF18 (10 ng/ml) 2.0 15
3.0 10 0.6
2.0 5.0 0.2
1.0
I FGF7 + FGF10 (10 ng/ml) + FGF18 (50 ng/ml) 1.0 2.5 0.1

tl g
g

tl g
g

tl g
g

er
G

G
H

H
A
B

A
B

A
B

A
B
Adtal l I

Adtal l I

Adtal l I

ta I
ul un
un

ul un
un

ul un
un

iv
ll
Fe
Fe

Fe

Fe
Figure 2  FGF18 promotes proximal airway
Relative expression

Relative expression

Relative expression
epithelia formation. (a,b) Differentiation to 1.0
PDX1
1.0
TG
1.0
PAX9
(to pancreas)

0.8 0.8 0.8

(to thyroid)

(to thyroid)
proximal lung cells. (c) Expression of proximal 0.6 0.6 0.6
0.2 0.2
lung cell genes KRT5, TRP63, FOXJ1, SOX17, 0.2
© 2012 Nature America, Inc. All rights reserved.

0.1 0.1 0.1

MUC5AC and CFTR were upregulated with
10 ng/ml of FGF18 (treatment H). Lower
as

d
G

G
H

H
A
B

A
B

A
B
nc I

I
oi

oi
re

yr

yr
d

Th

Th
levels of the Clara cell marker SCGB1A1,
Pa

and no significant detection of SFTPC were 100 100 100 100 100

Percent of max

Percent of max
Percent of max

Percent of max

Percent of max
80 80 80 80 80
detected. The transcription factors 60
32.2
60
30.9
60
36
60
32.1
60
53.9

NKX2.1 and FOXA2 that regulate SCGB1A1 40 40 40 40 40

and SFTPC expression as well as the distal 20 20 20 20 20
0 0 0 0 0
tip progenitor marker SOX9 were downregulated 2 3
010 10 10 10
4 5
0102 103 104 105 2 3
0 10 10 10 10
4 5 2 3
010 10 10 10
4 5 2 3
010 10 10 10
4 5

(compared to B). Other endoderm lineage panKRT CFTR FOXJ1 NKX2.1 P63

markers AFP (liver), PDX1 (pancreas), TG

(thyroid) and PAX9 (pharyngeal endoderm)
Definitive Endoderm
(D4) e Anterior foregut
endoderm (D9)
Proximal specification
(D14)
Proximal specification
(D19) f
100 60

BrdU positive
100 100 100
Percent of max

Percent of max

Percent of max

Percent of max
were not detected. (d) Representative flow

Percent
80 80 80 80 40
histograms of cells from treatment H reveal 60 60 60 49.1 60 20
40 47.9 40 42.8 40 40 28.9
approximately one-third of the cells are 20 20 20 20
0

epithelial (panKRT+), CFTR+, FOXJ1+ and

ox AF 4)

ox y ( )
e 4)

)
Pr arl D9

19
(D

. l D1
0 0 0 0

(D
.e E(
E
2 3 4 5 2 3 4 5 2 3 4 5 2 3 4 5
NKX2.1+. A majority of the cells express 010 10 10 10 010 10 10 10 0 10 10 10 10 0 10 10 10 10

at
BrdU BrdU BrdU BrdU
P63 (a homolog of p53 that initiates the

Pr
stratified program in epithelial cells). Gray solid
histograms represent respective isotype controls. (e) Representative histograms of BrdU incorporation at different stages of differentiation. (f) Average
cell proliferation at different stages of differentiation from four independent lines. Genes were normalized to the housekeeping gene β-ACTIN and
npg

expressed relative to adult lung positive control RNA. Error bars, mean ± s.e.m. ((n = 3 experiments). *P < 0.01, **P < 0.001, ***P < 0.05 compared
to treatment B. Numbers in brackets correspond to timepoint (day) during differentiation.

(Supplementary Fig. 4). During the early stages of differentiation, (an airway progenitor cell marker20) was observed (Supplementary
upregulated expression of the early lung marker NKX2.1 was vari- Fig. 6), suggesting that the putative P63+ basal progenitor cells may
able between the hESC lines. Similarly, differentiation of several wild- have differentiated into other airway epithelial lineages, as has been
type iPSC and CF iPSC lines also showed varying efficiencies in lung previously shown21. In addition, the majority of the cells expressed
marker expression (Supplementary Fig. 5). This suggests that opti- other conducting airway epithelia markers: acetylated tubulin
mization of the differentiation protocol may be required to reflect TUBA1A (cilia), MUC1 (Goblet cell), KRT14 (basal epithelia) and
line-to-line variability. the pan-endoderm marker FOXA2. A smaller percentage (15%) of
To further develop and mature the cells toward functional airway cells expressed the transcription factor NKX2.1, which in later stages
epithelium in vitro, we employed commercially available media that of lung development regulates Clara cell and type II alveolar cell dif-
support the growth and differentiation of primary bronchial epithelial ferentiation22. No cells positive for the pancreatic ductal epithelia
cells ex vivo along with air-liquid interface (ALI) growth condition to marker HPD1 were found, excluding the possibility that pancreatic
mimic the post-natal airway epithelial niche in vivo and promote dif- ductal epithelial cells were the source of CFTR. Proximal airway epi-
ferentiation, maturation and polarization of the epithelium (Fig. 3a). thelium was established as indicated by protein expression of mucin 16
To achieve ALI, the cells were exposed to air on the apical side of a tran- (a marker of the tracheal epithelium23) and cytokeratin 16 (ref. 24)
swell membrane while the basolateral side was exposed to media. After (Supplementary Fig. 7). Gene expression analysis by real-time
5 weeks of ALI, flow cytometry revealed that at least 50% of the CFTR+ RT-PCR showed upregulation of proximal airway lineage genes
cells coexpressed panKRT, FOXJ1 and LHS28 (basal bodies of cilia), (SOX17, FOXJ1, MUC5AC, TRP63, KRT5, ARG2, SOX2, CFTR,
indicating an enrichment of cells typical of the ciliated epithelium KRT16, MUC16, NGFR) comparable or higher than levels in total
(Fig. 3b). A greatly reduced percentage (5%) of cells expressing P63 adult lung or tracheal tissue (Fig. 3c). Clara cell marker SCGB1A1

878 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 letters

a hPSC
Endoderm
progenitors
Anterior foregut
progenitors Immature lung cells
Lung epithelial cells
lineage maturation
b
0.899 42.8
105

104
D0 D4 D9 D15 D20 D26 D40+

CFTR
Air lift
103
Activin A (100 ng/ml) FGF2 (500 ng/ml) Days 9–14: BEGM B-ALI
Wnt3a (25 ng/ml) SHH (50 ng/ml) FGF7 (50 ng/ml) FGF18 (10 ng/ml)
FGF10 (50 ng/ml) 0
BMP4 (5 ng/ml)
22.3 34
Days 15–19:
FGF7 (10 ng/ml) 0 103 104 105
FGF10 (10 ng/ml) panKRT
FGF18 (10 ng/ml)
c SOX17 (ciliated cell) FOXJ1 (ciliated cell) MUC5AC (goblet cell) TRP63 (basal cell progenitors)
10
5 10.4 14.6
P < 0.001 P < 0.05
1.5 5.0 1.5
Relative expression

Relative expression
15.0
Relative expression

Relative expression
P < 0.01
4.0 104
1.0 10.0 1.0
3.0

CFTR
0.5 2.0 0.5 3
5.0 10
1.0
0 0 0 0
5 wk T AL FL E 5 wk T AL FL E 5 wk T AL FL E 5 wk T AL FL E 102
ALI ALI ALI ALI 0
50.2 24.8
KRT5 (Basal cell) ARG2 (goblet cell) SOX2 (proximal lineages) CFTR
0 102 103
4
20.0 10 105
Relative expression

5.0 4.0
Relative expression

Relative expression

Relative expression
200 P < 0.05 FOXJ1
15.0 4.0 3.0
100
3.0
5.0 10.0 2.0
2.0 5 23.6 19.2
10
© 2012 Nature America, Inc. All rights reserved.

2.5 5.0 1.0 1.0

0 0 0 0 4
5 wk T AL FL E 5 wk T AL Thy E 5 wk T AL FL E 5 wk T AL FL E 10
ALI ALI ALI ALI

CFTR
KRT16 (tracheal epithelium) MUC16 (tracheal epithelium) NGFR (basal cell progenitors) 3
10
20.0
Relative expression

200 100
Relative expression

Relative expression

P < 0.01 P < 0.01

150 75 15.0
0
100 50 10.0
2 2 41.4 15.8
1 1 5.0 3 4 5
0 10 10 10
0 0 0 LHS28
5 wk T S FL E 5 wk T AL FL E 5 wk T AL Thy E
ALI ALI ALI

d Merged Z01 PanCK g

Apical

h
Apical

e
npg

Merged Z01 CFTR

i Apical
j Apical

Figure 3  Air liquid interface induces airway epithelial cell differentiation and promotes apical expression of CFTR. (a) Schematic of differentiation
protocol to generate mature airway epithelium using air liquid interface to induce polarization and apical expression of CFTR as observed in mature
airway epithelium. (b) The percentage of CFTR+, panKRT+ and FOXJ1+ cells was higher after 3 weeks of ALI. (c) Gene expression levels of airway
markers FOXJ1, MUC5AC, KRT5, TRP63 and SOX17 were significantly higher or comparable to adult lung tissue. Noticeably, CFTR was also
upregulated. The Clara cell marker SCGB1A1 and type II alveolar cell marker SFTPC were not significantly upregulated. Genes were normalized
to β-ACTIN and expressed relative to adult lung tissue positive control RNA. Error bars, mean ± s.e.m (n = 3). *P < 0.01, **P < 0.001 compared
to adult lung tissue. (d) To confirm the cells are epithelial, co-staining for ZO1 and pan-cytokeratin marker (panKRT) confirmed co-localization of
the two proteins suggesting establishment of a tight epithelium. (e) Maximal intensity projections of Z-stack confocal images of a 5-week-old ALI
culture of hESC CA1 line-derived epithelia co-express the tight junction–associated protein ZO1 and CFTR (clone L12B4). (f) The X-Z planar view
of the epithelium show apical localization of the CFTR protein. (g) H&E staining show cilia on some cells. (h) Higher magnification shows ciliated
cells (green) and apical localization of CFTR (orange; white arrowheads point to apical CFTR). (i) Low magnification of an ALI transwell cross-section
stained for cilia (beta IV tubulin, green) and CFTR (orange) shows nonuniform growth of cells with areas of pseudostratified cells and areas of sparse
monolayer cells. (j) High magnification of a culture stained for MUC5AC (green) on the surface of the cells (arrowheads). Scale bars: 60 µm (d–f);
50 µm (g,j); 25 µm (h); 90 µm (i).

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 879

 letters

and alveolar epithelial cell SFTPC, PDPN, P2X7 and the transcrip-
tion factor genes NKX2.1, FOXA2, FOXA1 and SOX9 were only
observed at very low levels (Supplementary Fig. 8). Low levels
of the thyroid TG and PAX9, liver HNF4 and AFP, and pancreas
PDX1 genes were also detected (Supplementary Fig. 9), suggesting
that the maturation of the lung lineages remained heterogeneous
with other endoderm lineages also present. With the exception of
PITX3, the other esophageal markers DLX3 and MUC2 were not
detected. The stratified epithelial marker mainly expressed in the
skin, KRT15, was also detected. However, no gene expression of the
neuronal lineage PAX6 and the forebrain gene FOXG1 was observed.
Immunofluorescence staining and confocal analysis of 5-week-old
ALI cultures showed contiguous patches of epithelial cells typified by
membrane expression of Zona Occludin-1 (ZO1), a protein associ-
ated with tight junctions, and co-staining with pan-KRT (Fig. 3d)
and CFTR (Fig. 3e). Reconstructed confocal stacked images showed
apical plasma membrane localization of CFTR (green) (Fig. 3f).
Stained sections of the cultures showed ciliated cells (Fig. 3g, H&E),
confirmed with antibody staining for cilia (beta IV tubulin, green)
and apically localized CFTR (orange) (Fig. 3h high, Fig. 3i low mag-
nification, respectively). In addition, the cultures stained positive for
© 2012 Nature America, Inc. All rights reserved.
Directed differentiation of CF mutant iPSCs into airway epithelial
cells holds great promise for disease modeling and drug discovery.
The most common CF mutation (~70% of cases25) is caused by a
phenyl­alanine deletion at position 508 (F508del). We generated
CF-iPSC lines from three F508del subjects by reprogramming primary
human fibroblasts using retroviruses containing the four pluripotency
factors (OCT4, KLF4, C-MYC and SOX2) as previously described26.
Genotype analysis was performed to confirm that the reprogrammed
cells carried the F508del homozygous mutation (Supplementary
Fig. 10). CF-iPSC lines expressed pluripotency markers TRA1-81,
TRA1-60 and NANOG, highly expressed endogenous pluripotency
genes OCT4, SOX2, KLF4 and C-MYC, and did not express the repro-
gramming retroviral transgenes, a hallmark of pluripotent stem cells
(Supplementary Fig. 11). In vitro embryoid body formation and
in vivo teratoma assays revealed that the lines generate cell types of all
three germ layers (Supplementary Fig. 12a). Expression of the other
pluripotency markers TERC and TERT and genes associated with full
reprogramming DNMT3B and REX1 were detected (Supplementary
Fig. 12b). Therefore CF-iPSCs generated by retroviral-mediated
reprogramming characteristically resembles hESCs and are func-
tionally pluripotent.

MUC5AC on the surface of the cells, indicative of mucin produc- To sample the functional expression of CFTR in patches of epithe-
tion (Fig. 3j). These findings suggest air-liquid interface can induce lium, we used a modification of the iodide efflux27 method for detecting
maturation and polarization of a ciliated large airway epithelium with regulated CFTR channel activity. The expression of the mature,
proper localization of the CFTR protein. complex, glycosylated form of CFTR, ‘Band C’, which represents the

Figure 4  Establishment of functional CFTR a CF b c

hESC-derived airway epithelia and correction hESC iPSC Caco-2
70 40
(kD)
lodide concentration (µM)

of CF phenotype in CF-iPSC-derived epithelial 30

Peak iodide efflux post

CFTR 60
cells with a small-molecule compound C18. 50 20

agonist (µM)
170 Band C hESCs
(a) Western blot shows Band C appearing 10
40 CF iPSCs
(~170 kDa) in 6-week-old hESC-derived epithelial Band B 2
30
cells (the CA1 cell line) indicative of the complex
20
glycosylated functional form of CFTR. Caco-2 1
cells were used as a positive control. The 10
0 0
antibody used in this western blot (mAb #660) Calnexin
Agonist – + – +
–2 –1 0 1 2 3 4 5
recognizes a peptide within NBD1 of CFTR. Time (min)
Calnexin was probed for loading control.
(b) Representative iodide flux graph shows
d R-domain CFTR/DAPI ZO1/DAPI Merged
npg

cyclic AMP agonist induced CFTR activity in

Human ESCs

a differentiated hESC-derived 6-week-old ALI e

culture but not in CF-iPSC-derived culture. DMSO C18
(c) The response to stimulation of CFTR in
two hESC lines, H9 (red) and CA1 (green) and Band C
control Caco-2 cell (black, three cultures).
Each line represents a different responsive Band B
culture (that is, showing an increase in efflux
CF iPSC line 2

rate within 1–2 min of stimulation). Four H9

cultures were responsive from a total of 13
cultures and 3 CA1 cultures were responsive 1.5 f
of 16 studied. The H9 cell line could be C18
differentiated to exhibit relatively robust 1.0 DMSO
C/B ratio

responses. (d) Representative photomicrographs

CF iPSC C18-treated

of hESC (CA1 line), CF-iPSC GM00997 Line 2 0.5

treated with either DMSO (control) or C18
(10 µM) and co-stained for tight junction– 0

associated protein ZO1 and CFTR. This antibody

recognizes an epitope in the R-domain. Plasma
membrane localization of CFTR was observed
(white arrows) after 2 d of treatment with C18 in
the CF-iPSC-derived epithelial cells but not in DMSO controls. Scale bar, 22 µm. (e) Cropped western blot shows the accumulation of Band C
(mature complex glycosylated form) in C18-corrected cells whereas the predominant form of the mutant protein in uncorrected cells is Band B (core-
glycosylated, ER-retained protein). The antibody (mAb #450) recognizes the CFTR peptide: 698-705). For full-length blots, see Supplementary Figure 14.
(f) Ratio of C (complex glycosylated) to B of F508del-CFTR protein in C18-treated cultures (n = 4) versus DMSO treated cultures (n = 3). Error bars,
mean ± s.e.m.

880 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology


plasma membrane–localized functional protein, was detectable by
western blot analysis in hESC-derived, but not in CF-iPSC–derived,
cultures (Fig. 4a). We then tested whether the in vitro differenti-
ated epithelium from normal hESCs and CF-iPSCs (Supplementary
Fig. 13) possessed cyclic AMP–regulated CFTR anion channel activity
using a halide efflux assay. As expected from similar studies in epithe-
lial cells endogenously expressing CFTR in their apical membranes,
such as Caco-2 cells, a human colorectal adenocarcinoma epithelial
cell line, cAMP agonists stimulated peak iodide efflux within 1–2 min
in differentiated cultures (Fig. 4b). There was heterogeneity with
respect to the functional expression of CFTR in different cell
lines, with two of the four hESC lines exhibiting cAMP-activated
efflux. Of these two hESC cell lines, one exhibited robust responses
(Fig. 4c) similar to levels of control Caco-2 cells and the other, modest
responses (green symbols). There was heterogeneity with respect to
the efficacy of the differentiation protocol, with two of four differenti-
ation trials leading to responsive cultures in the first cell line and three
of six in the second cell line. The reasons for this variability have yet
to be fully defined but likely reflect different proportions of polarized
CFTR-expressing epithelial cells in each culture. Nonetheless, this
stepwise, directed differentiation overcomes the previously encoun-
© 2012 Nature America, Inc. All rights reserved.
tered barriers to generating mature functional CFTR-expressing
proximal airway epithelia.
In the F508del CF mutation, the mutant CFTR protein does not
fold properly in the endoplasmic reticulum, preventing it from being
properly trafficked to the plasma membrane. Instead, the mutant pro-
tein is rapidly targeted for degradation28. Recent studies have shown
that small molecules called ‘corrector’ compounds are effective in
partially rescuing the trafficking defect of the major mutant29. As a
proof of concept experiment to determine whether CF-iPSC-derived
epithelial cells may be used to evaluate CF corrector compounds,
we tested the effect of C18, an active analog of the small molecule
VX-809 (currently in phase 2 clinical trials), in promoting plasma
membrane localization of F508del-CFTR in CF-iPSC–derived epi-
thelial cells (Fig. 4d). Notably, no surface-localized F508del-CFTR
was detected in control DMSO-treated CF-iPSC cultures, but cultures
treated for 24 h with C18 (10 µM) exhibited patches of cells expressing
CFTR on their cell surface. Although we could not observe substantial
changes in cAMP-regulated iodide efflux from C18-treated cells (data
npg
not shown), we did observe a trend toward a change in the Band C
to Band B ratio of F508del-CFTR protein in C18-treated cultures
versus DMSO-treated cultures from one F508del-CFTR proband
(Fig. 4e,f). Overall, CF-iPSC–derived airway cells may provide a
renewable source of patient-specific cells to identify new or validate
existing CF therapeutic drugs.
To our knowledge there has been no previous report demonstrat-
ing that human pluripotent stem cells can be directed to differentiate
in vitro into CFTR-functional conducting airway epithelium. Although
the differentiation protocol generates heterogeneous endoderm line-
ages, a great majority of the cells express airway epithelia markers,
with establishment of CFTR function observed in a third of the
cultures. Further refinement by isolating the cells using positive and
negative selection or cell-surface marker identification of lung pro-
genitor populations would improve the purity of lung epithelial cells.
This study also serves as a proof of concept that CF-iPSC–derived
epithelial cells may be used to validate existing, or identify new, thera-
peutic modulators of CFTR activity. This can be done in a patient-
specific manner, taking into account the genetic modifiers 30 that
underlie the heterogeneity in F508del CF pathologies. Patient-specific
iPSC-derived airway epithelial cells hold future promise of regenera-
tive medicine approaches to treat serious lung diseases.
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 881
letters
Methods
Methods and any associated references are available in the online
version of the paper.
Note: Supplementary information is available in the online version of the paper.
Acknowledgments
We would like to thank S. Yamanaka (CiRA, Kyoto, Japan) and A. Nagy (Samuel
Lunenfeld Research Institute, Toronto, Ontario, Canada) for providing human
iPSC lines (201B7, 253G1 and PB-4Fout, respectively). We would also like to thank
R. Bridges (Rosalind Franklin University, Chicago) who provided the C18 analog.
This work was funded by an Emerging Team grant from the Canadian Institutes of
Health Research (GPG-102171) to C.E.B., F.R., J.E. and J.R. This work was supported
(in part) by an Ontario Ministry of Economic Development and Innovation (MEDI)
grant. A.P.W. was a recipient of the MEDI Post-doctoral Award. Monoclonal CFTR
antibodies #450 and #660 were courtesy of J.R. Riordan (University of North Carolina,
Chapel Hill, North Carolina, USA). CA1, CA2 hESC were obtained from A. Nagy
(Mount Sinai Hospital, Toronto, Canada). H9 hESC were obtained from The WiCell
Research Institute (Wisconsin, USA).
AUTHOR CONTRIBUTIONS
A.P.W., J.R., J.E., C.E.B. and F.R. conceived the study and experimental design.
A.P.W. performed and analyzed experiments and wrote the manuscript. C.E.B.,
P.P., T.O.T., L.-J.H., S.C. and F.R. provided reagents, conceptual and/or technical
support in generating iPSC lines, doing the teratoma assay and making iodide
efflux measurements. All authors edited and approved the final manuscript.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Published online at http://www.nature.com/doifinder/10.1038/nbt.2328.
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
1. Van Haute, L., De Block, G., Liebaers, I., Sermon, K. & De Rycke, M. Generation
of lung epithelial-like tissue from human embryonic stem cells. Respir. Res. 10,
105 (2009).
2. Samadikuchaksaraei, A. et al. Derivation of distal airway epithelium from human
embryonic stem cells. Tissue Eng. 12, 867–875 (2006).
3. Wang, D., Haviland, D.L., Burns, A.R., Zsigmond, E. & Wetsel, R.A. A pure
population of lung alveolar epithelial type II cells derived from human embryonic
stem cells. Proc. Natl. Acad. Sci. USA 104, 4449–4454 (2007).
4. Mou, H. et al. Generation of multipotent lung and airway progenitors from mouse
ESCs and patient-specific cystic fibrosis iPSCs. Cell Stem Cell 10, 385–397
(2012).
5. Kimura, J. & Deutsch, G.H. Key mechanisms of early lung development. Pediatr.
Dev. Pathol. 10, 335–347 (2007).
6. Zorn, A.M. & Wells, J.M. Vertebrate endoderm development and organ formation.
Annu. Rev. Cell Dev. Biol. 25, 221–251 (2009).
7. D’Amour, K.A. et al. Efficient differentiation of human embryonic stem cells to
definitive endoderm. Nat. Biotechnol. 23, 1534–1541 (2005).
8. Ameri, J. et al. FGF2 specifies hESC-derived definitive endoderm into foregut/
midgut cell lineages in a concentration-dependent manner. Stem Cells 28, 45–56
(2010).
9. Serls, A.E., Doherty, S., Parvatiyar, P., Wells, J.M. & Deutsch, G.H. Different
thresholds of fibroblast growth factors pattern the ventral foregut into liver and lung.
Development 132, 35–47 (2005).
10. Bellusci, S. et al. Involvement of Sonic hedgehog (Shh) in mouse embryonic lung
growth and morphogenesis. Development 124, 53–63 (1997).
11. Kim, S.K. & Melton, D.A. Pancreas development is promoted by cyclopamine, a
hedgehog signaling inhibitor. Proc. Natl. Acad. Sci. USA 95, 13036–13041
(1998).
12. Pan, Q. et al. In vivo characterization of the Nkx2.1 promoter/enhancer elements
in transgenic mice. Gene 331, 73–82 (2004).
13. Bellusci, S., Grindley, J., Emoto, H., Itoh, N. & Hogan, B.L. Fibroblast growth
factor 10 (FGF10) and branching morphogenesis in the embryonic mouse lung.
Development 124, 4867–4878 (1997).
14. Shiratori, M. et al. Keratinocyte growth factor and embryonic rat lung morphogenesis.
Am. J. Respir. Cell Mol. Biol. 15, 328–338 (1996).
15. Morrisey, E.E. & Hogan, B.L. Preparing for the first breath: genetic and cellular
mechanisms in lung development. Dev. Cell 18, 8–23 (2010).
16. Weaver, M., Yingling, J.M., Dunn, N.R., Bellusci, S. & Hogan, B.L. Bmp signaling
regulates proximal-distal differentiation of endoderm in mouse lung development.
Development 126, 4005–4015 (1999).
17. Desai, T.J. et al. Distinct roles for retinoic acid receptors alpha and beta in early
lung morphogenesis. Dev. Biol. 291, 12–24 (2006).
 letters
18. Metzler, M.D. & Snyder, J.M. Retinoic acid differentially regulates expression of
surfactant-associated proteins in human fetal lung. Endocrinology 133, 1990–1998
(1993).
19. Whitsett, J.A. et al. Fibroblast growth factor 18 influences proximal programming
during lung morphogenesis. J. Biol. Chem. 277, 22743–22749 (2002).
20. Rock, J.R. et al. Basal cells as stem cells of the mouse trachea and human airway
epithelium. Proc. Natl. Acad. Sci. USA 106, 12771–12775 (2009).
21. Hong, K.U., Reynolds, S.D., Watkins, S., Fuchs, E. & Stripp, B.R. In vivo differentiation
potential of tracheal basal cells: evidence for multipotent and unipotent subpopulations.
Am. J. Physiol. Lung Cell. Mol. Physiol. 286, L643–L649 (2004).
22. Minoo, P. et al. Physical and functional interactions between homeodomain NKX2.1
and winged helix/forkhead FOXA1 in lung epithelial cells. Mol. Cell. Biol. 27,
2155–2165 (2007).
23. Davies, J.R., Kirkham, S., Svitacheva, N., Thornton, D.J. & Carlstedt, I. MUC16 is
produced in tracheal surface epithelium and submucosal glands and is present in
secretions from normal human airway and cultured bronchial epithelial cells.
Int. J. Biochem. Cell Biol. 39, 1943–1954 (2007).
© 2012 Nature America, Inc. All rights reserved.
npg
882 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology
24. Huang, T.H. et al. Control of keratin gene expression by vitamin A in tracheobronchial
epithelial cells. Am. J. Respir. Cell Mol. Biol. 10, 192–201 (1994).
25. Ratjen, F. & Doring, G. Cystic fibrosis. Lancet 361, 681–689 (2003).
26. Hotta, A. et al. EOS lentiviral vector selection system for human induced pluripotent
stem cells. Nat. Protoc. 4, 1828–1844 (2009).
27. Kim Chiaw, P., Wellhauser, L., Huan, L.J., Ramjeesingh, M. & Bear, C.E. A chemical
corrector modifies the channel function of F508del-CFTR. Mol. Pharmacol. 78,
411–418 (2010).
28. Lewis, H.A. et al. Impact of the deltaF508 mutation in first nucleotide-binding
domain of human cystic fibrosis transmembrane conductance regulator on domain
folding and structure. J. Biol. Chem. 280, 1346–1353 (2005).
29. Van Goor, F. et al. Correction of the F508del-CFTR protein processing defect
in vitro by the investigational drug VX-809. Proc. Natl. Acad. Sci. USA 108,
18843–18848 (2011).
30. Wright, F.A. et al. Genome-wide association and linkage identify modifier loci of
lung disease severity in cystic fibrosis at 11p13 and 20q13.2. Nat. Genet. 43,
539–546 (2011).
 ONLINE METHODS
Maintenance of pluripotent stem cells. Human ESCs and iPSCs were main-
tained on mitotically inactivated mouse embryonic fibroblast feeders in
Knockout DMEM (GIBCO) with 15% serum replacement (GIBCO), Glutamax
(Invitrogen), penicillin/streptomycin (GIBCO), 1 mM nonessential amino acids
(GIBCO), 0.5 mM mercaptoethanol and 10 ng/ml FGF2 (Peprotech). CA1, CA2
hESCs were obtained from A. Nagy (Mount Sinai Hospital, Toronto, Canada).
H9 hESC were obtained from The WiCell Research Institute (Wisconsin, USA).
The hESC were cultured under the approval of the Canadian Institutes of
Health Research Stem Cell Oversight Committee. Culture conditions for hESCs
are identical to those for hiPSCs. To harvest cells for differentiation, hESC and
iPSC colonies were dissociated with 0.25% trypsin (Invitrogen), washed with
culture media and centrifuged to pellet the cells.
Differentiation of human ESC and iPSC into definitive endoderm.
Differentiation into definitive endoderm was done as previously described7.
Briefly, pluripotent stem cells were harvested, gently triturated into single-cell
suspensions and seeded onto transwells (0.4 µm pore size, Corning) pre-coated
with human placental collagen type IV, which has previously been shown to
support airway epithelial cell growth31. The cells were immediately treated with
100 ng/ml activin-A and 25 ng/ml WNT3A (R&D Systems) for 4 consecutive
days in Endoderm Differentiation Media consisting of serum-free Knockout
DMEM (Invitrogen) with Glutamax (Invitrogen), penicillin/streptomycin
© 2012 Nature America, Inc. All rights reserved.
(GIBCO), 1 mM nonessential amino acids (GIBCO) and 0.5 mM mercaptoeth-
anol. Subsequent differentiation steps were performed on the transwells.
Differentiation of definitive endoderm into anterior foregut endoderm
progenitors. For anterior foregut endoderm differentiation and especially
embryonic lung progenitors, definitive endoderm cells were treated with
500 ng/ml FGF2 (Preprotech) and 50 ng/ml Sonic hedgehog (SHH, Cedarlane)
for 5 d in Endoderm Differentiation Media. Extended culture with FGF2 and
SHH did not significantly augment the number of NKX2.1+ cells generated.
Directed differentiation of foregut endoderm into mature lung cell fates.
The cells were treated with 50 ng/ml FGF10, 50 ng/ml KGF (FGF7) and
5 ng/ml BMP4 (all R&D systems) for 5 d followed by 10 ng/ml FGF10,
10 ng/ml FGF7 and 10 ng/ml FGF18 (Sigma-Aldrich) for an additional 5 d. To
differentiate the cells into mature airway epithelial cells, the cells were cultured
in Bronchial Epithelial Growth Media (BEGM, Lonza) supplemented with
FGF18 (10 ng/ml) for 10 d followed by Bronchial-Air Liquid interface (B-ALI,
Lonza) media for an additional ≥15 d. The cells were ‘air-lifted’ and B-ALI
media was only added to the bottom but not the top of the transwell.
npg
Generation and characterization of human iPSC lines. Human skin fibro­
blasts (GM00997, GM04320) were obtained from the Coriell Cell repository
(Coriell Institute for Medical Research) and HSC patient fibroblast from the
Hospital for Sick Children (Toronto, Canada) with informed consent. These
fibroblasts were isolated from the donor skin biopsy as previously described26.
By 4 weeks of reprogramming obvious human ESC-like EGFP+ colony num-
bers were enriched under puromycin selection before picking and expansion.
CF-iPSC lines with ESC-like morphology were further assessed for pluri­
potency marker expression (NANOG, TRA1-81, TRA1-60) by flow cytometry
and immunofluorescence, and real-time qPCR to examine upregulation of
endogenous pluripotency genes (OCT4, SOX2, C-MYC and KLF4) and down-
regulation of the exogenous retroviral transgenes. In addition, gene expres-
sion of other pluripotency markers DNMT3B, REX1, TERC, TERT were also
assessed. Karyotype analysis was done to determine genetic stability. Finally,
iPSC lines were subjected to in vitro embryoid body and in vivo teratoma assays
for functional tests of pluripotency as previously described26.
Teratoma formation assay. Teratoma formation experiments were done in
NOD/SCID immunodeficient mice as previously described26. Human ESCs
(lines CA1, CA2 and H9) and mouse embryonic fibroblasts were used as posi-
tive and negative controls for experiments, respectively. All procedures using
animals have been approved by the SickKids Animal Care Committee under
the auspices of The Canadian Council on Animal Care.
doi:10.1038/nbt.2328
Quantitative Real-time PCR. Total RNA was prepared using the RNeasy Kit
(Qiagen). RNA was reverse-transcribed for first-strand cDNA using Superscript II
(Invitrogen) according to the manufacturer’s protocol. Quantitative real-time PCR
(SYBR green detection method; Applied Biosystems, Foster City, CA) was per-
formed for amplification of the genes listed in Supplementary Table 1. Real-time
PCR (45 cycles of amplification) was performed on the LightCycler 480 System
(Roche). Gene expression was normalized to the housekeeping gene β-ACTIN and
expressed relative to a positive control sample. Denaturing curves for each gene were
used to confirm DNA product and eliminate the possibility of pseudogene amplifi-
cation or primer-dimers. All experiments were done in triplicate with at least three
separate differentiation cultures. Primer sequences are listed in Supplementary
Table 1 and positive control tissue RNA listed in Supplementary Table 2.
Immunofluorescence. Transwells were fixed with fresh paraformaldehyde
(4%) for 1 h at room temperature. For CFTR staining, transwells were fixed
with ice-cold methanol (100%) in −20 °C for 10 min. For cytoplasmic or nuclear
stains, cells were permeabilized and blocked with a solution containing 0.25%
Triton X-100 (Invitrogen), 2% BSA and 5–10% normal goat or donkey serum.
Primary antibodies used are listed in Supplementary Table 3. Secondary anti-
bodies include goat anti-rabbit, mouse or rat (IgG), or donkey anti-goat IgG
Alexa Fluor 488 and 555 (Molecular Probes). Nuclei were counterstained with
DAPI (Invitrogen). Stains were visualized with the Confocal Digital Imaging
System (Nikon) and analyzed with Volocity Software (PerkinElmer). Images
were digitally processed using Adobe Photoshop CS5 (Adobe) in accordance
with Nature Publishing Guidelines by altering only contrast and brightness.
Flow cytometry. Flow cytometry staining was performed as per manufac-
turer’s protocol. For intracellular staining, cells were permeabilized with
permeabilization (PERM) buffer containing saponin (BD Biosciences). For
non-intracellular flow, the cells were resuspended and stained in FACS buffer
containing 0.2% BSA. Primary antibodies used are listed in Supplementary
Table 2. Secondary antibodies used include goat anti-mouse (IgG or IgM) or
goat anti-rabbit (IgG) Alexa Fluor 488, 647 (Molecular Probes) or PE-Cy7
(BD Biosciences). Non-immune reactive isotypes were used as staining con-
trols. Data acquisition was performed using the LSRII flow cytometer (BD
Biosciences) and analyzed with Flowjo software (Tree Star Inc.).
Iodide efflux assay. Iodide loading and washing. Iodide flux across the apical
side of epithelial cells in ALI cultures reports CFTR channel activity since CFTR
is permeable to iodide. Cells were loaded with 850 µl of NaI solution (3.0 mM
KNO3, 2.0 mM Ca(NO3)2, 11 mM glucose, 20 mM HEPES, 136 mM NaI) from
the bottom of transwells at 37 °C for 1 h for iodide uptake in cells. Resultant
iodide solution was washed out 10 times with 4 ml of washing solution contain-
ing nitrate (3.0 mM KNO3, 2.0 mM Ca(NO3)2, 11 mM glucose, 20 mM Hepes,
136 mM NaNO3) and 100 µM amiloride, an epithelial sodium channel (ENaC)-
specific inhibitor was added. cAMP-stimulated halide flux. The following time
course used 350 µl of respective solution, which was added to the top of tran-
swells and then transferred to a 96-well plate well at each one minute time point.
Time course for each culture involved: 3 min wash with washing solution and
8 min cAMP-stimulated halide flux with cAMP agonists, forskolin (10 µM) and
3-isobutyl-1-methylxanthine (100 µM, IBMX), and a CFTR potentiator, genis-
tein (50 µM) in washing solution. Vehicle dimethyl sulfoxide (DMSO) served as
the negative control. CFTR channel activity was detected as a change in iodide
flux 1–2 min after addition of cAMP agonists. As it has been known that Caco-2
can recapitulate properties of fully differentiated epithelia, which include api-
cal expression of CFTR and CFTR-mediated vectorial transepithelial chloride
flux at 3–5 d post-confluency32, cells were grown to 5 d post-confluency
and then used for the functional discontinuous iodide efflux assay. Caco-2
cells were split and plated on 12-well dish (with each well ~3.8 cm2 in area)
coated with collagen IV before measurement. Measurement and calibration.
The halide-selective microelectrode (Lazar Research Laboratories, Los Angeles,
CA) was used to measure the absolute iodide electrode potential (mV) value.
Readings were recorded using the Digidata 1320A Data Acquisition System
with the Clampex 8.1 program. A calibration curve was created by measuring
the mV values of nitrate solutions containing 2 µM to 1 mM iodide to convert
mV values to iodide concentration (µM).
nature biotechnology
 Immunoblotting. Cells were solubilized in 1% sodium dodecyl sulfate (SDS)
and sample protein run on 6% SDS gels for SDS-PAGE. Protein samples were
transferred to nitrocellulose paper. Primary CFTR antibody (MAB1660 (R&D)
or antibody #450 courtesy of J.R. Riordan) and goat anti-mouse Ig HRP
secondary antibody were used for immunoblotting. Immunoblot was exposed
to enhanced chemiluminescence (ECL).
Proliferation assay. BrdU incorporation was assessed by flow cytometry using
the BD BrdU flow kit (catalog # 552598) and performed using manufacturer’s
protocol. Cells were incubated with BrdU for 24 h before analysis.
© 2012 Nature America, Inc. All rights reserved.
npg
nature biotechnology
Statistical analysis. Unless otherwise specified, for statistical analysis,
unpaired t-tests were performed. When more than two groups were compared,
one-way ANOVA was used followed by Dunnett’s post-test if significance was
observed. Results were expressed as mean ± s.e.m.
31. Trinh, N.T., Prive, A., Maille, E., Noel, J. & Brochiero, E. EGF and K+ channel
activity control normal and cystic fibrosis bronchial epithelia repair. Am. J. Physiol.
Lung Cell. Mol. Physiol. 295, L866–L880 (2008).
32. Sood, R. et al. Regulation of CFTR expression and function during differentiation
of intestinal epithelial cells. EMBO J. 11, 2487–2494 (1992).
doi:10.1038/nbt.2328
 letters

Polyethyleneimine is a potent mucosal adjuvant for

viral glycoprotein antigens
Frank Wegmann1, Kate H Gartlan1, Ali M Harandi2, Sarah A Brinckmann1, Margherita Coccia1,
William R Hillson1, Wai Ling Kok3, Suzanne Cole3, Ling-Pei Ho3, Teresa Lambe4, Manoj Puthia5,
Catharina Svanborg5, Erin M Scherer6, George Krashias1, Adam Williams7, Joseph N Blattman6,
Philip D Greenberg6, Richard A Flavell7, Amin E Moghaddam1, Neil C Sheppard1,8,9 & Quentin J Sattentau1,9

Protection against mucosally transmitted infections probably and reactogenicity5–8. Receptor agonist–based adjuvants, such as
requires immunity at the site of pathogen entry1, yet there CpG (a toll-like receptor (TLR) 9 ligand) and poly(I:C) (a TLR3
© 2012 Nature America, Inc. All rights reserved.

are no mucosal adjuvant formulations licensed for human
use. Polyethyleneimine (PEI) represents a family of organic
polycations used as nucleic acid transfection reagents in vitro
and DNA vaccine delivery vehicles in vivo 2,3. Here we show
that diverse PEI forms have potent mucosal adjuvant activity
for viral subunit glycoprotein antigens. A single intranasal
administration of influenza hemagglutinin or herpes simplex
virus type-2 (HSV-2) glycoprotein D with PEI elicited robust
antibody-mediated protection from an otherwise lethal
infection, and was superior to existing experimental mucosal
adjuvants. PEI formed nanoscale complexes with antigen,
which were taken up by antigen-presenting cells in vitro and
in vivo, promoted dendritic cell trafficking to draining lymph
nodes and induced non-proinflammatory cytokine responses.
PEI adjuvanticity required release of host double-stranded DNA
that triggered Irf3-dependent signaling. PEI therefore merits
further investigation as a mucosal adjuvant for human use.
npg
ligand), have well-defined mechanisms of action9. Other adjuvants,
however, including aluminum salts, do not require TLR signaling
for immune activation, implying the existence of distinct adjuvant-
­elicited immune activation pathways10–12.
PEI is a cationic polymer, synthesized in various forms including
linear or branched and high or low molecular weight species, that
is widely used as a gene transduction agent. PEI forms complexes
with nucleic acids through electrostatic interactions forming nano-
scale ‘polyplexes’. These polyplexes bind cell surface heparan sulfate
proteoglycans and enter cells through endocytosis3. During endo-
somal acidification, buffering by PEI results in osmotic rupture of
the endosome, allowing entry of the polyplexes directly into the
cytoplasm3. In addition to their use as gene transduction agents, PEI
has been used to increase the immunogenicity of DNA vaccines13,14.
However, it is unclear from these studies whether PEI acts solely by
protecting DNA and enhancing its cellular targeting and uptake, or
whether PEI has intrinsic immune-activating properties in vivo.

Although immune responses can be elicited at mucosal surfaces by
systemic immunization, the compartmentalization of the immune
system means that this may be suboptimal for protection from
mucosal pathogens1. Despite extensive efforts, no adjuvanted mucosal
vaccine targeting major mucosally transmitted infectious diseases,
such as influenza A, HSV-2 or HIV-1, has been licensed for use in
humans4. Currently available experimental mucosal adjuvants include
cholera holotoxin, the cholera toxin B subunit (CTB), the Escherichia
coli–derived heat-labile enterotoxin, α-galactosylceramide (αGalCer,
a CD1 ligand), CpG-containing oligodeoxynucleotides (CpG) and
the double-stranded RNA analog polyinosinic-polycytidylic acid
(poly(I:C)). Despite showing promise in terms of immune-stimulating
­activity, these adjuvants have yet to be licensed for human use, in large
part because of safety concerns linked to local or systemic ­toxicity
As PEI can efficiently deliver cargo into cells that express heparan
sulfate proteoglycans, such as antigen-presenting cells (APC)15,16 we
hypothesized that it might also function as a mucosal adjuvant for
glycoprotein antigens. To address this, we immunized mice twice
intranasally with a weakly immunogenic, endotoxin-free, experi-
mental viral vaccine antigen—recombinant envelope glycoprotein
gp140 derived from the HIV-1CN54 isolate—alone, with the known
mucosal adjuvant CTB, or complexed with various PEI forms. We
found that antigen-specific IgG titers in the serum were ~100-fold
higher when gp140 was administered with PEI than alone and up
to six-fold higher than those elicited by CTB (Fig. 1a). As HIV-1 is
transmitted predominantly through the genital mucosae, we analyzed
IgA levels in vaginal lavages. In these experiments, linear PEI forms
were not more potent than CTB, whereas branched PEI forms of
750 kD (B750) and 25 kD (B25) gave titers of antigen-specific mucosal

1The Sir William Dunn School of Pathology, University of Oxford, Oxford, UK. 2Department of Microbiology and Immunology, Institute of Biomedicine, Sahlgrenska

Academy, University of Gothenburg, Gothenburg, Sweden. 3The Medical Research Council Human Immunology Unit, Weatherall Institute of Molecular Medicine,
The John Radcliffe Hospital, Oxford, UK. 4The Jenner Institute, The University of Oxford, Oxford, UK. 5The Division of Microbiology, Immunology and Glycobiology,
Lund University, Lund, Sweden. 6The Department of Immunology, University of Washington, Seattle, Washington, USA. 7Yale School of Medicine, New Haven,
Connecticut, USA. 8Present address: Vaccine Research, Pfizer Inc., San Diego, California, USA. 9These authors contributed equally to this work. Correspondence
should be addressed to Q.J.S. (quentin.sattentau@path.ox.ac.uk).

Received 3 April; accepted 19 July; published online 26 August 2012; doi:10.1038/nbt.2344

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 883

 letters

a b c d e f g

Log lgG1/IgG2a titer ratio

* * ** * 3

Log denatured IgG titer

5 3 4 1,000 1.3 n.s. 25

normalized to total IgA

Log lgG endpoint titer

**
Log lgG endpoint titer

Log lgA endpoint titer

* * n.s. ***

Log native IgG titer

IC50 [mM NH4SCN]

CLN/spleen ratio
** ** ** 800 * * * 2 20
4 1.2
2 3 n.s.
600 1 15
3 1.1
n.s. 400 0 10
1 2
2 200 1.0 –1 5
1 0 1 0 0.9 –2 0

+ e
+ e

gp + P one

1 + A

0 pG
gp 0 + EI- L40
0 I- 60

I-B 0

B
gp + I- 40

14 + EI B

0 pG

0 pG

gp 0 + TB
0 I- 60

I-B 0

25

gp 0 + TB
14 + EI B

0 pG
25

gp 0 + TB

gp 0 + TB

I
I

I
14 + C T
I

I
14 + T

14 + C T
14 + C T

PE
PE

PE
gp 140 40 lon
gp 140 40 lon

PE

PE
PE 5
PE 5

gp 140 LG

14 CT
gp 40 + P CT
gp 40 + P CT

gp 0 C
gp 40 + C

gp 40 + C
gp 40 C
14 PE L1
+ B7
PE -L
14 PE L1
+ B7

14 C
14 C

14 C
C
14 C
0 al
P -
a

14 +
a

+
+

+
+

gp +
gp gp1 40

gp 140
gp gp1 40

14 0

gp 140

gp 40

0
0
gp p14

14
1
1

1
gp

gp
gp

1
gp

gp

gp
g
0

1
1

h i Control
j k 100 Control
100 4
∆ body weight (%)

100 HA alone gD

Percent survival
Percent survival

80

Disease score
HA + CTB 3 gD + CT
90 HA + αGalCer 60 gD + αGalCer
HA + PEI 2 gD + PEI
50 40
80 1 gD + PEI (single dose)
20 gD (single dose)
0
70 0 0
0 2 4 6 8 10 0 2 4 6 8 10 0 10 20 5 10 15 20
Days after challenge Days after challenge Days after challenge Days after challenge

Figure 1  Mucosal adjuvant activity of PEI and protection from disease by mucosal pathogens. (a,b) Immunoglobulin titers for BALB/c mice primed
and boosted once with 10 µg gp140 plus CTB or a linear (L) or branched (B) PEI: gp140-specific serum IgG (a) and vaginal IgA (b) 4 weeks after
© 2012 Nature America, Inc. All rights reserved.

booster immunization (ANOVA with Bonferroni correction; n = 6). (c) gp140-specific serum IgG on day 21 after primary immunization (ANOVA with
Bonferroni correction; n = 5). (d) gp140-specific serum IgG avidity index (2 weeks after booster) measured by chaotropic ELISA using NH 4SCN (ANOVA
followed by Dunnet’s Test; n = 5). (e) Ratio of serum IgG endpoint titers against native versus SDS/DTT-denatured gp140 (2 weeks after booster; ANOVA
with Bonferroni correction; n = 5). (f) Serum IgG1/IgG2a titer ratio (2 weeks after booster; Kruskal-Wallis followed by Dunn’s test; n = 5). Data are
representative of two experiments. (g) Ratio of local (CLN) proliferation versus systemic (spleen) proliferation as determined by 3H-thymidine incorporation
(unpaired t-test; n = 6) of BALB/c mice immunized intranasally with three doses of gp140 alone or with CTB or PEI. (h,i) Weight loss (h) and
Kaplan-Meier survival plots (i) for mice after highly pathogenic live influenza A/PR8 virus challenge 3 weeks following immunization. Data are representative
of three independent experiments (n = 4–5). (j,k) Disease score (j) and Kaplan-Meier survival plots (k) for immunized mice after highly pathogenic vaginal
HSV-2 challenge. C57Bl/6 mice, n = 7. All data are presented as mean of replicates from individual experiments ± s.e.m.; n.s., not significant (P > 0.05);
*, P < 0.05; **, P < 0.01; ***, P < 0.001. CT, cholera holotoxin; CTB, cholera toxin B subunit; HA, hemagglutinin; gD, glycoprotein D.

IgA more than tenfold higher than those elicited by CTB (P < 0.05, affinity maturation by PEI. Moreover, gp140-specific antibodies
ANOVA with Bonferroni correction, Fig. 1b). As PEI-B750 and PEI- induced by PEI preferentially recognized gp140 in its native confor-
B25 had similar levels of adjuvant activity and PEI-B25 is the best- mation (Fig. 1e), suggesting that PEI may protect the ­immunizing
studied form in terms of gene delivery17, we further characterized the antigen from denaturation and degradation in vivo, possibly by
safety and adjuvanticity of PEI-B25. a mechanism analogous to the compaction and chaperoning
When PEI is used as a DNA transfection reagent, high levels of DNA. By contrast with CpG, which elicited an IgG2a-biased
are known to be cytotoxic in vitro, but little is known about the response, PEI, cholera holotoxin and CTB elicited a dominant IgG1
npg

toxi­city of PEI in vivo18. We found that a single intranasal dose of isotype (Fig. 1f) consistent with the bias toward T helper type 2 (TH2)
gp140 with either 10 or 20 µg PEI-B25 was safe and equivalently immunity previously reported for CTB20.
immuno­genic (Supplementary Fig. 1a–c). Similarly safe were cholera To further investigate a role for PEI in TH2-type immune biasing, we
holotoxin, CTB and poly(lactic-co-glycolic acid) nanoparticles used antigen to in vitro restimulate lymphocytes derived from spleen or
(PLGA, Supplementary Fig. 1d,e) used at, or above, standard doses nasal epithelium–draining cervical lymph node (CLN). Proliferation
(Supplementary Table 1). By contrast, 20 µg CpG promoted signi­ of splenocytes and CLN cells from mice that received antigen-PEI
ficant (P < 0.05, ANOVA followed by Dunnett’s test) weight loss. or antigen-CTB was significantly increased (P < 0.01 and P < 0.001,
Despite this, neither PEI nor CpG obviously modified the nasal epi- respectively; ANOVA followed by Dunnett’s test) compared to anti-
thelium when compared with PBS alone (Supplementary Fig. 1f), gen alone (Supplementary Fig. 2b), demonstrating that adjuvanted
although further investigation is required to rule out any localized intranasal immunization induced both local and systemic T-cell activa-
neurological effects19. tion. However, the ratio of CLN cells to splenocyte proliferation was
We compared the adjuvanticity of PEI with other prominent significantly (P < 0.01, two-tailed t-test) greater in PEI-immunized
experimental mucosal adjuvants. Twenty-one days after a single mice than in CTB-immunized mice, suggesting that PEI acts more
intranasal gp140 administration, we observed serum antigen- locally than CTB (Fig. 1g). Splenocyte and CLN cell cytokine release
s­pecific IgG endpoint titers of ~103 for PEI, CTB and CpG, which profiles confirmed a TH2 bias. CTB and PEI induced similar levels of
were substantially greater than the titers induced by gp140 alone or interleukin (IL)-4, IL-5 and IL-13, but PEI induced lower levels of the
when formulated with PGLA (Fig. 1c). After a booster immuniza- pro-inflammatory cytokines IL-17, IFN-γ and TNF-α (Supplementary
tion, we observed a similar pattern of responses with higher end- Fig. 2c,d). As previously reported21, CTB induced high levels of the
point serum IgG ­titers (Supplementary Fig. 2a). Notably, 2 weeks pro-inflammatory cytokine IL-17. Comparison of ratios of IL-17/IL-4,
after the booster PEI induced significantly (P < 0.05, ANOVA fol- IL-17/IL-2 and IL-2/IL-4 revealed that PEI induced significantly
lowed by Dunnett’s test) higher avidity antigen-specific IgG than (P < 0.01, Kruskal-Wallis followed by Dunn’s test) lower IL-17 than CTB
cholera holotoxin, CTB or CpG (Fig. 1d), implying enhanced both in spleen and local lymph nodes (Supplementary Fig. 2e–j).

884 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology


We analyzed species-related PEI activity in rabbits immunized
intranasally with HIV-1 gp140 alone or with equivalent doses of
PEI or CTB. PEI increased antigen-specific serum IgG responses by
16-fold more than a safe dose of CTB, which gave titers ­statistically
­indistinguishable from gp140 alone (Supplementary Fig. 3a).
Vaginal IgG titers were lower than serum titers but confirmed the
trend with three-fold greater responses in the PEI group compared
to CTB, whereas vaginal antigen-specific IgA was not detected
(Supplementary Fig. 3b).
We next sought to determine the efficacy of PEI as an adjuvant
in a model of respiratory infectious disease. After priming with
purified influenza A/PR8 hemagglutinin alone, formulated in PEI
or in other mucosal adjuvants at standard doses (Supplementary
Table 1), we challenged mice intranasally with a pathogenic dose
of influenza A/PR8. Mice that received no antigen, antigen alone
or antigen in αGalCer had severe (15–25%) weight loss, whereas
mice that received antigen in CTB only exhibited moderate (~10%)
weight loss (Fig. 1h). Notably, mice immunized with antigen in PEI
were completely protected from weight loss, showing significantly
(P < 0.05, Kruskal-Wallis followed by Dunn’s test) enhanced protec-
tion compared to CTB or αGalCer (Supplementary Fig. 4a), mirrored
© 2012 Nature America, Inc. All rights reserved.
by complete protection from an otherwise lethal influenza
A/PR8 challenge (Fig. 1i). Of note, PEI-elicited hemagglutinin-
specifc IgG showed an enhanced ratio of native to denatured antigen
binding compared to other groups (Supplementary Fig. 4b), sug-
gesting that PEI-induced antibodies would preferentially bind native
viral hemagglutinin. Similar ­titers of antigen-specific serum IgG1,
IgG2a and IgA were elicited by all adjuvants, and IgA titer correlated
significantly (P < 0.05, Spearman-correlation of log-transformed
titers) with reduced disease (Supplementary Fig. 4c,d). Similarly,
titers of antigen-­specific IgA but not IgG in the bronchoalveolar lavage
correlated with protection from disease (Supplementary Fig. 4e–g).
We also compared PEI and CTB in a more conventional for-
mat using prime and booster immunizations to increase adaptive
immune responses. Again, PEI provided complete protection from
disease, whereas CTB-treated mice showed a trend toward weight
loss (Supplementary Fig. 5a,b). Antigen-specific IgG levels in the
serum were significantly (P < 0.01, ANOVA followed by Dunnett’s
test) higher for PEI than for CTB after the prime but were equiva-
npg
lent after the booster (Supplementary Fig. 5c). In this format, both
serum IgG1 and IgA titers correlated with protection from disease
(Supplementary Fig. 5d,e), and median antibody avidity was tenfold
higher for PEI than CTB (Supplementary Fig. 5f). We conclude from
these experiments that the superior protection by PEI from influenza-
mediated disease is likely to reflect increased avidity and/or recogni-
tion of native antigen by antibodies elicited by this adjuvant.
To investigate whether intranasal administration of antigen in
PEI induced protective responses at a distant mucosal surface, we
adopted an established mouse model of mucosal immunization
with recombinant HSV-2 glycoprotein D followed by vaginal chal-
lenge with an otherwise lethal dose of HSV-2 (ref. 22). Mice were
primed and boosted twice intranasally with glycoprotein D alone,
in cholera holotoxin, αGalCer or in PEI. We also challenged mice
primed but not boosted with glycoprotein D alone or glycoprotein
D in PEI. Mice receiving three immunizations of glycoprotein D in
αGalCer, cholera holotoxin and PEI were completely protected from
disease and terminal weight loss, whereas mock-immunized mice or
those receiving glycoprotein D alone were not protected (Fig. 1j,k).
Notably, mice receiving a single dose of glycoprotein D in PEI were
largely protected from disease and 80% protected from terminal
weight loss, by contrast with glycoprotein D alone (Fig. 1j,k and
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 885
letters
P < 0.05, ANOVA with Bonferroni correction, Supplementary
Fig. 6a). Before challenge, antigen-­specific serum IgG titers were
significantly (P < 0.001, ANOVA with Bonferroni correction) greater
in mice receiving adjuvanted glycoprotein D compared to mice
­receiving glycoprotein D alone (Supplementary Fig. 6b). Antigen-
specific vaginal IgA was undetectable.
The robust adaptive immunity induced by PEI suggested antigen
targeting to APC either directly, or indirectly by targeting to mucosal
epithelial cells followed by cell death and APC uptake, or both.
As polycationic PEI forms complexes with polyanionic nucleic acids
promoting cellular uptake, we hypothesized that the same might hold
for glycoprotein antigens with a net negative surface charge, such as
gp140 (ref. 23). To investigate this, we performed surface plasmon
resonance binding assays. Soluble gp140 interacted with immobilized
PEI weakly but with a relatively slow off-rate, whereas immobilized
gp140 interacted strongly with soluble PEI with fast on-rates and slow
off-rates (Fig. 2a). Use of PEI as an adjuvant will better mimic the first
condition in which antigen binds polymeric PEI in solution. To fur-
ther explore PEI-gp140 interactions, we carried out scanning electron
microscopic analysis of PEI and gp140 alone, combined, or PEI com-
plexed with DNA. We confirmed previous studies demonstrating that
PEI alone appears as an amorphous matrix whereas PEI and DNA form
globular polyplexes of 100–200 nm24 (Fig. 2b). By contrast, gp140 and
PEI formed a distinct species of particle with an average diameter of
750 nm (Fig. 2b,c), approximately two- to tenfold larger than particles
formed between ovalbumin (OVA) and PEI25.
To test whether gp140-PEI particles were targeted for APC uptake
in vitro, we treated bone marrow–derived dendritic cells (BMDCs)
or an epithelial cell line with fluorochrome-labeled gp140 (gp140 647)
or gp140647-PEI, and analyzed the cells by confocal microscopy and
flow cytometry. Addition of gp140647 alone gave diffuse low-level
BMDC uptake, whereas gp140647 with PEI yielded robust intra­
cellular uptake of particulate complexes (Fig. 2d–f). Epithelial cells
were also efficiently targeted by gp140647-PEI (Fig. 2g). To investigate
PEI-antigen uptake and cellular recruitment in vivo, we inoculated
mice intranasally with PEI-gp140647 and evaluated antigen uptake
of disaggregated nasal-associated lymphoid tissue (NALT) and
draining CLN by flow cytometry. At 4 h after intranasal administra-
tion, PEI significantly enhanced gp140647 uptake by epithelial cells
(P < 0.001) and microfold cells (P < 0.01, two-tailed t-test) that
express high levels of heparan sulfate proteoglycans targeted by PEI-
DNA complexes3 (Fig. 2h). This PEI-mediated antigen targeting was
accompanied by modest transient recruitment of neutrophils into the
NALT and CLN (Fig. 2i and Supplementary Fig. 7c,d), observed for
many adjuvants and routes of immunization11,26. At 24 h after admin-
istration, gp140647 was primarily found associated with dendritic
cells in draining CLN, and PEI mediated significant recruitment of
dendritic cells and B cells (P < 0.001 and P < 0.01, respectively,
two-tailed t-test) to CLN (Fig. 2j,k and Supplementary Fig. 7a,b).
As PEI targets cargo to the cytoplasm3, we investigated its abil-
ity to enhance antigen cross-presentation in vitro and in vivo using
an OVA model system. PEI-OVA-exposed BMDC significantly
(P < 0.05, ANOVA followed by Dunnett’s test) enhanced the response
of B3Z T cells, a T-cell hybridoma activated by recognition of
H-2Kb in association with OVA(257–264) peptide (Supplementary
Fig. 8a), consistent with increased cross-priming in the presence
of PEI in vitro observed previously25. However, there was no
increase in the frequency of CLN gp140-specific CTLs induced
in vivo by PEI after intranasal immunization (Supplementary
Fig. 8b). This finding is consistent with the lack of IFNγ secre-
tion detected in gp140-­restimulated lymphocytes from immunized
 letters

a b c d WGA555 gp140647 Merge + DAPI

6,000
PEI gp140
Response units

gp140647 alone
1,250
4,000
1,000

Diameter (nm)
2,000
750

0 500
0 200 400 600 250
PEI-gp140 PEI-DNA
Time (s)

gp140647 + PEI
PEI over immobilized BSA 0
PEI over immobilized gp140

A
14

N
I-D
BSA over immobilized PEI

p
I-g

PE
gp140 over immobilized PEI

PE
gp140 over empty flow cell
BSA over empty flow cell

e x-y y-z f g ***

h
150 400 2,000 PBS + Ag 4 h
*** PEI PEI + Ag 4 h

gp140+ cell count

gp140647 (GMFI)

gp140647 (GMFI)
300 PBS + Ag 1,500
100

4 h NALT
PEI + Ag
200 1,000
***
50 ***
BMDC

100 500
n.s.
*

LA4
0 0 0
gp140-A647 gp140-A647
S S

I+ g
Ag

S S
PE Ag
Ag

ia ls
on lls

tro ac
ils

B C
T lls
lls

l
ta
PE A
PB PB

PB PB

el el

D
M l ce

ph

ce
ce
© 2012 Nature America, Inc. All rights reserved.

N o/m

To
+

+
I+

ith M c
x-z

eu
Ep
Figure 2  Interactions between 400 i
* 100 PBS + Ag 24 h 800 16j k
3

PBS + Ag 4 h PBS + Ag 24 h
gp140 cell count × 10

PEI + Ag 24 h PEI + Ag 24 h
antigen and PEI, and between PEI + Ag 4 h 80
cell count (×10 )

12

cell count × 106

cell count × 103
3

200 600
immune cells and PEI-antigen ***

24 h CLN
24 h CLN
4 h NALT

24 h CLN
60 8
complexes. (a) Surface plasmon 20 400 **
15 40 3
resonance with surface-immobilized 10 * 200 2
+

20
protein or PEI detects direct 5 1
biophysical interaction between 0 0 0 0
N al c s
tro lls
on ils
ac

B C
T ls
lls

ac

ils

lls

lls

tro c
ils

lls

lls

l
ta

ta

ta
PEI and bovine serum albumin
l

eu /ma
el el

D
eu e
M ph

ce
ce

ph

ce

ce

ph

ce

ce
m

To

To

To
ith M c

o/

o/

tro

o
B

T
on

on
(BSA) or PEI and gp140. Data are
i

eu
M

M
N

N
Ep

representative of three independent

experiments. (b) Scanning electron micrographs of PEI alone, gp140 alone or PEI complexed with gp140 or DNA. Scale bar, 500 nm. (c) PEI-antigen
complex formation. Quantification of PEI-gp140 or PEI-DNA particle sizes determined by counting of randomly chosen fields imaged by SEM.
(d) BMDC uptake of gp140 647-PEI complexes. gp140 647 (2 µg/ml) complexed or not with 0.1 µg/ml PEI. Cells were surface stained with Alexa
Fluor 555–labeled wheat germ agglutinin (WGA 555) and nuclear stained with Hoechst 34580. Laser scanning confocal micrographs of BMDCs
incubated with gp140647 alone (top panel) or gp140 647 mixed with PEI (bottom panel). Scale bars, 20 µm. (e) Three-dimensional reconstructions
of a single BMDC exposed to gp140 647 complexed with PEI. Scale bar, 10 µm. (f,g) Flow cytometric quantification of gp140 647 uptake by BMDCs
(f) or LA-4 mouse lung epithelial cells (g) after overnight exposure to antigen with or without PEI. Ag, antigen; GMFI, geometric mean fluorescence
intensity (n = 3, ANOVA followed by Dunnett’s test). (h–k) Flow cytometric identification of uptake of gp140 647 by NALT (h) or CLN (j) cells, or cell
npg

recruitment into NALT (i) or CLN (k), 4 h (h,i) or 24 h (j,k) after intranasal immunization of BALB/c mice with 10 µg gp140647 ± 20 µg PEI
(n = 5, two-tailed t-test; all other comparisons between control and test groups not significant). Leukocyte populations were identified with
appropriate labeling and gates. Data are representative of two or more independent experiments and are presented as mean values of replicates from
one experiment ± s.e.m. *, P < 0.05; **, P < 0.01; ***, P = 0.001. DC, dendritic cells; mono, monocytes; mac, macrophages.

mice (Supplementary Fig. 2c,d) and probably reflects insufficient cells, resulting in activation of other cells in an Irf3-dependent path-
PEI-induced T helper type 1 (TH1)-biased help. way28. As PEI exhibits varying degrees of cellular toxicity in vitro18, we
A major function of adjuvants is activation of innate immune hypothesized that this might liberate cellular dsDNA leading to acti-
responses that trigger adaptive immunity. We therefore studied a vation of intracellular dsDNA sensors. Indeed, PEI induced apoptosis
well-established mode of adjuvant mechanism, direct activation and cell death in the lung epithelial cell line LA-4 (Supplementary
of APCs resulting in increased expression of costimulatory mol- Fig. 10a,b). Although alum-released dsDNA triggers cytoplasmic
ecules and cytokine secretion27, by comparing immature BMDCs DNA sensors28, it is not clear how released host dsDNA crosses the
incubated in vitro with PEI to those activated by the TLR4-ligand plasma membrane to engage these sensors. We tested the targeting
lipopolysaccharide (LPS). LPS induced predicted upregulation by alum and PEI of synthetic dsDNA (poly(dA:dT)) to the cytoplasm
of maturation markers CD40, CD86 and major histocompat- of B16-blue type-I interferon reporter cells. Poly(dA:dT) combined
ibility (MHC) class-II, whereas PEI did not increase expression with alum or PEI induced significant (P < 0.001, two-way ANOVA)
of these markers above control values, indicating the absence of activation of reporter cells, with PEI-induced activation about four-
direct maturation signals (Supplementary Fig. 9a,b). Consistent fold higher than that with alum (Fig. 3a,b).
with this result, PEI failed to activate two TLR reporter cell lines To investigate DNA release in vivo, we administered PEI or alum
(Supplementary Fig. 9c,d) and other reporter lines specific for intraperitoneally and assayed peritoneal lavage for free dsDNA.
individual TLRs (data not shown). Both PEI and alum mediated release of highly significant (P < 0.001,
It was recently demonstrated that the aluminum-based adjuvant, ANOVA followed by Dunnett’s test) amounts of dsDNA into the lavage
alum, triggers the release of double-stranded (dsDNA) from dying (Fig. 3c). This prompted us to carry out intranasal immunization of

886 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology

 letters

a Alum 0 µg/ml b PEI 0 µg/ml c d Control e WT Ag alone

WT Ag + PEI
Alum 1.5 µg/ml PEI 1.5 µg/ml DNAse
Alum 4.4 µg/ml PEI 3 µg/ml
4 *** 4
* WT Ag + CT

Log lgG endpoint titer

1.0 1.0
Alum 13.3 µg/ml PEI 6 µg/ml *** *** Irf3–/– Ag alone

per lavage (ng)

A650 (triplicate ± s.d.)
A650 (triplicate ± s.d.)

n.s.

Log dsDNA
0.8 Alum 40 µg/ml 0.8 PEI 12 µg/ml 3 3 n.s. Irf3–/– Ag + PEI
*** –/–
Irf3 Ag + CT
0.6 0.6

Log IgG1 endpoint titer

2 2 6 IN immunization
0.4 0.4 ******
*** 5
0.2 *** *** 0.2
1 1
4
0 0

I
um

er
tro

PE

PE
on

C
0 500 1,500 3,000 0 500 1,500 3,000

C
on

Al
3

al
al

αG
0
C

14
Poly(dA:dT) (ng/ml) Poly(dA:dT) (ng/ml)

14

14

+
gp
gp

gp
2

0
14
gp
Figure 3  PEI-induced host dsDNA release and cytoplasmic recognition by means of Irf3-dependent 1
signaling. (a,b) B16-blue type-I interferon reporter cells were exposed to synthetic dsDNA (poly(dA:dT)) lacking 0 10 20 30 40 50
TLR9-activating CpG motifs, in the presence or absence of the indicated concentrations of alhydrogel (alum, a) or PEI (b) Days after prime
immunization
(parallel experiment; two-way ANOVA). (c) BALB/c mice (n = 5) were intraperitoneally injected with PEI (65 µg) or
alhydrogel (alum, 1 mg) and dsDNA in peritoneal lavage was detected 24 h after injection by fluorescence-based dsDNA f WT
assay (ANOVA followed by Dunnett’s test). (d) BALB/c mice (n = 5) were intranasally immunized with 10 µg gp140 alone 6 Irf3–/–

Log IgG1 endpoint titer

or co-formulated with 20 µg PEI, 2 µg cholera holotoxin or 2 µg αGalCer and treated intranasally at 2.5 h and 18 h after **
5
immunization with 0.5 mg DNase I or vehicle control and antigen-specific serum IgG was determined by ELISA 3 weeks n.s.
4
after immunization. Data are representative of two or more independent experiments; ANOVA with Bonferroni correction.
3
(e,f) C57BL/6 WT control or Irf3−/− mice were intranasally immunized with gp140 alone or co-formulated with 20 µg PEI
or 2 µg cholera holotoxin at day 0, 22 and 36 and antigen-specific serum IgG1 was determined by ELISA (gp140 alone 2
n = 3, other groups n = 6–7). IN, intranasal. (e) Time course of gp140-specific serum IgG1 responses. (f) Day 35 serum 1
© 2012 Nature America, Inc. All rights reserved.

IgG1 responses of immunized mice; ANOVA with Bonferroni correction. Data are representative of two or more independent

T
PE
on

C
experiments and are presented as mean values of replicates from one experiment ± s.e.m.; n.s., not significant (P > 0.05);

+
al

0
0

14
14

14
*, P < 0.05; **, P < 0.01; ***, P < 0.001.

gp
gp

gp
mice with PEI and gp140 followed by intranasal DNase-I treatment28 equivalent in wild-type and Nlrp3−/− mice (Supplementary Fig. 11e).
to evaluate the effect of degrading released DNA on events trigger- However, Nlrp3−/− mice immunized with antigen and PEI showed
ing adaptive immunity. DNAse treatment significantly (P < 0.05, a >20-fold increased level of specific IgG2c compared to wild type
ANOVA with Bonferroni correction) reduced PEI-induced, but not (P < 0.05, ANOVA with Bonferroni correction Supplementary
αGalCer- or cholera holotoxin-induced, antigen-specific antibody Fig. 11f,g), whereas no significant isotype bias was observed with
responses, demonstrating that DNase activity has an adjuvant-specific cholera holotoxin adjuvantation (Supplementary Fig. 11h–j). These
downmodulating effect (Fig. 3d). We examined the possible role of data show that Nlrp3 inflammasome function is not required for
Irf3 in the DNA-mediated adjuvant effect of PEI by intranasal admin- overall PEI adjuvant activity, but suggest that inflammasome activa-
istration of PEI and gp140 to Irf3−/− mice followed by assay of antigen- tion biases adaptive immunity toward a TH2 response.
specific IgG1. Similar IgG1 responses to antigen alone were observed This study describes both the innate and adaptive immune-
in both groups, indicating that Irf3−/− mice were not generally less ­activating activities of PEI driving in vivo mucosal adjuvanticity.
sensitive to intranasal immunization. By contrast, PEI activity was Innate immune pathways are activated by the release of ­intracellular
reduced by 41-fold (P < 0.01, ANOVA with Bonferroni correction) dsDNA that acts as a damage-associated molecular pattern trigger-
in the Irf3−/− mice compared to wild-type controls, demonstrating ing the Irf3 interferon pathway through cytoplasmic DNA sensors.
npg

strong dependence upon this pathway for PEI adjuvanticity (Fig. 3e,f). This mode of innate immune activation has recently been proposed
We observed a non-significant (P > 0.05, ANOVA with Bonferroni for the TLR-independent adjuvant alum28. Another innate immune
correction) reduction (sevenfold) in IgG1 titer for cholera holotoxin, pathway activated by PEI is the inflammasome, potentially either
suggesting a weak or absent Irf3 requirement for this adjuvant. through the lysosomal destabilizing activity of PEI 3, or through
The reported lysosomal disruption by PEI implicit in its gene release of other damage-associated molecular patterns, such as uric
transduction activity3 suggested that PEI might trigger the Nlrp3 acid. We also show that PEI triggers an influx of APC to the site of
inflammasome. Because the release of IL-1β is a hallmark of inflam- immunization and associates with antigen to form nanoparticles
masome activation that has been implicated in the adjuvant activity that are efficiently taken up by APC. Thus, PEI acts at multiple
of alum29, we assayed IL-1β in lavage 4 h and 24 h after intraperito- levels to deliver adjuvant activity for glycoprotein antigens
neal PEI administration. At both time points, IL-1β levels were ele- (Supplementary Fig. 12).
vated (Supplementary Fig. 11a). Treatment of bone marrow–derived Several studies have demonstrated increased immunogenicity of
macrophages (BMM) with LPS and alum as a positive control or LPS vaccine DNA when formulated with PEI14,31. However, the DNA
and PEI led to increased release of IL-1β and IL-18 (Supplementary delivery activity of PEI predicts neither that PEI will have intrinsic,
Fig. 11b,c). To confirm inflammasome assembly, we assayed innate immune-activating properties, nor that PEI will function
caspase-1 activation29,30 in lysates of BMM that were unstimulated, as an adjuvant for non-DNA antigens. Although one recent study
PEI-treated, LPS-primed or LPS primed with PEI or ATP. Relative demonstrated NF-κB activation after in vitro treatment of BMDCs
to untreated cells or cells treated with LPS alone, both PEI alone and with PEI14, this does not imply adjuvant activity in vivo.
PEI with LPS induced caspase-1 cleavage (Supplementary Fig. 11d). The minimal induction of proinflammatory cytokines and modest
We investigated inflammasome activation in the adjuvant activity of local infiltration of neutrophils by PEI suggests that this adjuvant
PEI by intranasally immunizing wild-type and Nlrp3 −/− mice with is unlikely to elicit inflammation-induced tissue damage at sensi-
PEI and gp140. PEI with antigen induced higher antigen-specific tive mucosal surfaces and would be free of concerns relating to the
total serum IgG titers compared with antigen alone that were consequences of high-level mucosal IL-17 induction, such as

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 887

 letters
inflammation and autoimmunity32. The ability of PEI to induce pro-
tective immune responses to mucosal viral infection implies that the
immune responses of neither the TH1 nor the TH17 subset of helper
T cells are an absolute requirement.
In line with a major role for antibodies as a dominant protective
effector mechanism, we identified IgA and IgG as correlates of pro-
tection (Supplementary Figs. 4 and 5). The enhanced avidity and
recognition of native antigen after PEI-adjuvantation are very encour-
aging, particularly for highly conformational antigens such as those
tested here. Finally, the generalizable adjuvant activity of different
PEIs suggests that potency and safety may be further optimized both
for DNA and non-nucleic acid antigen delivery.
Methods
Methods and any associated references are available in the online
version of the paper.
Note: Supplementary information is available in the online version of the paper.
Acknowledgments
This study was supported by grants from The MRC UK, The EU Network of
Excellence ‘Europrise’, The International AIDS Vaccine Initiative Neutralizing
© 2012 Nature America, Inc. All rights reserved.
Antibody Consortium (IAVI), The Bill and Melinda Gates Foundation
Collaboration for Vaccine Design (CAVD), Dormeur Investment Service Ltd.,
the FP7-funded High Impact Project Advanced Immunization Technologies
(ADITEC) and The EuroNanoMed-European Commission funded iNanoDCs.
Q.J.S. is a Jenner Institute Investigator and a James Martin Senior Fellow.
Author Contributions
Q.J.S. conceived the study and N.C.S. provided initial proof of concept. F.W.,
K.H.G., S.A.B., N.C.S., A.M.H., M.C., W.R.H., W.L.K., S.C., T.L., M.P., E.M.S., G.K.,
A.W. and A.E.M. designed and performed experiments. Q.J.S., F.W. and K.H.G.
wrote the paper. A.M.H., L.-P.H., C.S., J.N.B., P.D.G., R.A.F., A.E.M. and N.C.S.
supplied reagents and made editorial suggestions.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Published online at http://www.nature.com/doifinder/10.1038/nbt.2344.
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
1. Chen, K. & Cerutti, A. Vaccination strategies to promote mucosal antibody responses.
Immunity 33, 479–491 (2010).
npg
2. Gunther, M. et al. Polyethylenimines for RNAi-mediated gene targeting in vivo and
siRNA delivery to the lung. Eur. J. Pharm. Biopharm. 77, 438–439 (2011).
3. Lungwitz, U., Breunig, M., Blunk, T. & Gopferich, A. Polyethylenimine-based
non-viral gene delivery systems. Eur. J. Pharm. Biopharm. 60, 247–266 (2005).
4. Lycke, N. Recent progress in mucosal vaccine development: potential and limitations.
Nat. Rev. Immunol. 12, 592–605 (2012).
5. Fujihashi, K., Koga, T., van Ginkel, F.W., Hagiwara, Y. & McGhee, J.R. A dilemma
for mucosal vaccination: efficacy versus toxicity using enterotoxin-based adjuvants.
Vaccine 20, 2431–2438 (2002).
6. Lewis, D.J. et al. Transient facial nerve paralysis (Bell’s palsy) following intranasal
delivery of a genetically detoxified mutant of Escherichia coli heat labile toxin. PLoS
ONE 4, e6999 (2009).
888 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology
7. Sogaard, O.S. et al. Improving the immunogenicity of pneumococcal conjugate
vaccine in HIV-infected adults with a toll-like receptor 9 agonist adjuvant: a
randomized, controlled trial. Clin. Infect. Dis. 51, 42–50 (2010).
8. Heikenwalder, M. et al. Lymphoid follicle destruction and immunosuppression after
repeated CpG oligodeoxynucleotide administration. Nat. Med. 10, 187–192
(2004).
9. Kasturi, S.P. et al. Programming the magnitude and persistence of antibody
responses with innate immunity. Nature 470, 543–547 (2011).
10. De Gregorio, E., D’Oro, U. & Wack, A. Immunology of TLR-independent vaccine
adjuvants. Curr. Opin. Immunol. 21, 339–345 (2009).
11. Lambrecht, B.N., Kool, M., Willart, M.A. & Hammad, H. Mechanism of action of
clinically approved adjuvants. Curr. Opin. Immunol. 21, 23–29 (2009).
12. Gavin, A.L. et al. Adjuvant-enhanced antibody responses in the absence of toll-like
receptor signaling. Science 314, 1936–1938 (2006).
13. Hu, K. et al. An ocular mucosal administration of nanoparticles containing DNA
vaccine pRSC-gD-IL-21 confers protection against mucosal challenge with herpes
simplex virus type 1 in mice. Vaccine 29, 1455–1462 (2011).
14. Ma, Y.F. & Yang, Y.W. Delivery of DNA-based cancer vaccine with polyethylenimine.
Eur. J. Pharm. Sci. 40, 75–83 (2010).
15. Orr, G. et al. Syndecan-1 mediates the coupling of positively charged submicrometer
amorphous silica particles with actin filaments across the alveolar epithelial cell
membrane. Toxicol. Appl. Pharmacol. 236, 210–220 (2009).
16. Wegrowski, Y. et al. Cell surface proteoglycan expression during maturation of human
monocytes-derived dendritic cells and macrophages. Clin. Exp. Immunol. 144,
485–493 (2006).
17. Park, K. Luciferin liposomes for enhanced in vivo bioluminescence. J. Control.
Release 141, 109 (2009).
18. Hunter, A.C. Molecular hurdles in polyfectin design and mechanistic background
to polycation induced cytotoxicity. Adv. Drug Deliv. Rev. 58, 1523–1531
(2006).
19. Mutsch, M. et al. Use of the inactivated intranasal influenza vaccine and the risk
of Bell’s palsy in Switzerland. N. Engl. J. Med. 350, 896–903 (2004).
20. Yamamoto, S. et al. A nontoxic mutant of cholera toxin elicits Th2-type responses
for enhanced mucosal immunity. Proc. Natl. Acad. Sci. USA 94, 5267–5272
(1997).
21. Lee, J.B., Jang, J.E., Song, M.K. & Chang, J. Intranasal delivery of cholera toxin
induces Th17-dominated T-cell response to bystander antigens. PLoS One 4, e5190
(2009).
22. Lindqvist, M., Persson, J., Thorn, K. & Harandi, A.M. The mucosal adjuvant effect
of alpha-galactosylceramide for induction of protective immunity to sexually
transmitted viral infection. J. Immunol. 182, 6435–6443 (2009).
23. Kong, L. et al. Expression-system-dependent modulation of HIV-1 envelope
glycoprotein antigenicity and immunogenicity. J. Mol. Biol. 403, 131–147 (2010).
24. Rudolph, C. et al. Methodological optimization of polyethylenimine (PEI)-based
gene delivery to the lungs of mice via aerosol application. J. Gene Med. 7, 59–66
(2005).
25. Chen, J. et al. Improved antigen cross-presentation by polyethyleneimine-based
nanoparticles. Int. J. Nanomedicine 6, 77–84 (2011).
26. Calabro, S. et al. Vaccine adjuvants alum and MF59 induce rapid recruitment of
neutrophils and monocytes that participate in antigen transport to draining lymph
nodes. Vaccine 29, 1812–1823 (2011).
27. Manicassamy, S. & Pulendran, B. Modulation of adaptive immunity with Toll-like
receptors. Semin. Immunol. 21, 185–193 (2009).
28. Marichal, T. et al. DNA released from dying host cells mediates aluminum adjuvant
activity. Nat. Med. 17, 996–1002 (2011).
29. Eisenbarth, S.C., Colegio, O.R., O’Connor, W., Sutterwala, F.S. & Flavell, R.A. Crucial
role for the Nalp3 inflammasome in the immunostimulatory properties of aluminium
adjuvants. Nature 453, 1122–1126 (2008).
30. Hornung, V. et al. Silica crystals and aluminum salts activate the NALP3 inflammasome
through phagosomal destabilization. Nat. Immunol. 9, 847–856 (2008).
31. Torrieri-Dramard, L. et al. Intranasal DNA vaccination induces potent mucosal and
systemic immune responses and cross-protective immunity against influenza viruses.
Mol. Ther. 19, 602–611 (2011).
32. Littman, D.R. & Rudensky, A.Y. Th17 and regulatory T cells in mediating and
restraining inflammation. Cell 140, 845–858 (2010).
 ONLINE METHODS
Antigens and adjuvants. CN54 gp140 was manufactured to good manu-
facturing practice specification by Polymun Scientific, Vienna, Austria33.
Hemagglutinin was derived by detergent extraction from purified, egg-
grown influenza A virus (A/Puerto Rico/8 (A/PR/8), H1N1), as previously
described34. Hemagglutinin was tested for the presence of endotoxin at
the clinical bio-manufacturing facility (University of Oxford, UK) and was
<0.5 EU/mL. CHO K1-expressed HSV-2 glycoprotein D was purified using
HiTrap Chelating HP column (GE Healthcare)35. Fluorescent labeling of CN54
gp140 (gp140647) was done in-house using an Alexa Fluor 647 protein labeling
kit (Invitrogen). The following adjuvants were used: ‘standard’ polyethylene-
imine, branched, 25 kDa (PEI or PEI-B25, Sigma); polyethyleneimine, branched,
750 kDa (PEI-B750, Sigma); polyethyleneimine, linear, 40 kDa (PEI-L40,
Polymer Chemistry Innovations, Inc.); polyethyleneimine, linear, 160 kDa
(PEI-L160, Polymer Chemistry Innovations, Inc.); cholera toxin (cholera
holotoxin, azide-free holotoxin, List biological laboratories); cholera toxin
B subunit (CTB, Sigma), dialyzed against PBS and concentrated as required;
immunostimulatory CpG 1826 (Invivogen); the synthetic α-galactosylcera­
mide analog KRN7000 (αGalCer, Enzo) or α-galactosylceramide (Alexis
Biochemicals); poly(lactic-co-glycolic acid) nanoparticles (PLGA, 245 nm,
Corpuscular); polyinosinic:polycytidylic acid (poly(I:C), Invivogen); lipopol-
ysaccharide (LPS, Invivogen). Antigen-PEI complexes were formed at least
2 h before immunization by adding the antigen into a pre-diluted PEI solution
© 2012 Nature America, Inc. All rights reserved.
followed by immediate mixing. To surface-immobilize antigen on PLGA par-
ticles, 60 µg gp140 and 30 µL PLGA particles at 200 mg/mL were coincubated
in a volume of 60 µL overnight at 4 °C and used for immunization. To deter-
mine PLGA surface immobilization of gp140, washed particles were lysed with
100 µL 0.2 M NaOH/5% SDS for 1 h at RT and protein contents of supernatant
and lysate was determined by BCA assay. 60% of the total protein input was
immobilized on PLGA particles.
Mice, immunizations and virus challenge. C57BL/6 Irf3−/− mice were derived
from RBRC strain 00858 (T. Taniguchi, University of Tokyo, Japan); C57BL/6
Nlrp3−/− mice were generated by Sutterwala et al. at the Yale University School
of Medicine36. Mice were immunized intranasally by instilling maximally
12.5 µL vaccine formulation at a time into both nostrils of lightly anaesthetized
mice. For experiments presented in Figure 1a,b, BALB/c mice (n = 6) were
primed and boosted once intranasally with 10 µg HIV-1 gp140 combined with
10 µg CTB or 10 µg of linear (L) or branched (B) PEI at different molecular
weights (kDa) as indicated. Data in Figure 1c–f are derived from BALB/c mice
(n = 5) that were primed and boosted once intranasally with 10 µg HIV-1 gp140
combined with 20 µg CTB, CpG, PEI, 2 µg cholera holotoxin or 1 mg PLGA
npg
nanoparticles. Vaginal mucosal lavage samples were collected by pipetting
2 × 50 µL of sterile PBS in and out of the vagina several times using sterile
blunted micropipette tips33. Mice were euthanized and bronchoalveolar ­lavage
was collected by inserting a lavage tube into the trachea of the mouse and flush-
ing the lungs once with 1.5 mL ice-cold PBS. DNase I treatment was performed
2.5 h and 18 h post intranasal immunization by instilling 1,000 units DNase
I (Roche) in 20 µL HBSS into both nostrils of lightly anaesthetized mice.
Influenza challenge was done as described37. Briefly, mouse-adapted human
influenza A virus (A/PR/8 (H1N1) was diluted to 0.05 U/mL ­hemagglutinin in
sterile, endotoxin-free PBS immediately before challenge. Immunized C57Bl/6
mice were lightly anesthetized and inoculated intranasally with 40 µL of the
diluted virus. Following challenge, mice were monitored for body weight
for 19 d. Mice losing >30% of their original weight were euthanized.
Intraperitoneal immunizations of mice were performed with antigen and/or
adjuvant in 0.5 mL PBS. Mice were euthanized 4 h or 24 h after immuniza-
tion and peritoneal lavage was collected by injecting and recovering 1× 2 mL
ice-cold PBS from the peritoneal cavity. Supernatants were used for cytokine/
chemokine assays or for detection of free dsDNA using a Quant-iT PicoGreen
kit (Invitrogen). For rabbit immunogenicity studies, female New Zealand
white rabbits, 10–12 weeks old, were immunized three times intranasally at
weeks 0, 3 and 12 with antigen alone or combined with adjuvant. For immu-
nizations, rabbits were sedated and anesthetized38, nasal and vaginal swab
samples were collected as described earlier38. The above experiments were per-
formed in accordance with UK national guidelines and were authorized by the
UK Home Office and the Oxford local institutional ethical review board.
doi:10.1038/nbt.2344
Vaginal HSV-2 challenge was performed as described35. Briefly, female C57Bl/6
mice were injected subcutaneously with 3 mg of Depo-Provera (DP) (Pharmacia)
in PBS, and 6 d later, the DP-treated mice were anesthetized and challenged
by intravaginal administration of 9 × 104 plaque forming units (PFU) of
HSV-2 strain 333 in 10 µL of Hanks’ balanced salt solution (HBSS). Mice were
examined daily for vaginal inflammation, neurological illness and survival
after HSV-2 challenge. HSV-2 challenge experiments were performed at the
Experimental Biomedicine animal facility, Sahlgrenska Academy at University
of Gothenburg, Sweden, with the approval from the Ethical Committee for
Laboratory Animals in Gothenburg, Sweden.
ELISA. Vaginal lavage and serum samples were cleared by centrifugation and
stored at −20 °C until analyzed. Antigen-coated and blocked ELISA plates
were incubated with serial dilutions of samples. For rabbit total IgG assays,
plates were coated with capture antibody (Serotec 403001) and blocked with
10% goat serum in WB. For denatured ELISAs, the antigen was denatured in
2% (w/v) SDS, 100 mM DTT for 5 min at 95 °C and coated on ELISA plates.
Bound antibodies were detected with the appropriate secondary reagents: anti-
mouse IgG-HRP (STAR120P, Serotec, Oxford, UK); anti-mouse IgA (STAR85,
Serotec) conjugated to biotin; anti-mouse IgG1-HRP and IgG2a-HRP (BD
Biosciences, Oxford, UK); anti-rabbit IgG-HRP (Sigma-Aldrich); TMB sub-
strate (Thermo Fisher Scientific, Rockford, IL); and OD values were read at
450 nm. ELISA data were analyzed by subtracting the plate background from
all readings and by fitting a sigmoidal dose response curve to the data for each
sample (Prism software, Graphpad). The endpoint titer was determined from
the maximum dilution of sample at which the OD450 signal was greater than
the defined threshold, which was always greater than two s.d. above back-
ground. The content of glycoprotein D-specific IgA and IgG antibodies were
determined in sera and vaginal swab samples, collected 3 weeks after the final
immunization, by ELISA-based analysis as described earlier35.
The avidity index was determined by diluting immune sera to a concen-
tration that corresponds to OD = 0.5 in an endpoint ELISA, and allowing
antibodies to react overnight with antigen-coated plates. After washing,
­various concentrations (63 mM–4 M) of the chaotropic reagent ammonium
thiocyanate were added to the wells. After 20 min, the plates were washed
and the remaining antibody detected. The avidity index is the concentration
of chaotrope required to reduce the optical density (OD) by 50% compared
to that in wells with no chaotrope.
T-cell restimulation and proliferation. For gp140-specific responses, spleno-
cytes or CLN cells were cultured in the presence or absence of 10 µg/mL
antigen for 48 h (splenocytes) or 72 h (CLN). Cells were pulsed with 1 µCi
of [3H] thymidine per well (Perkin-Elmer) for the last 8–12 h of incubation.
For glycoprotein D-specific responses, splenocytes or cells from mediastinal
or subiliac lymph nodes were stimulated with 3 µg/mL glycoprotein D for
96 h and [3H] thymidine-pulsed for the last 6 h of incubation. Cellular DNA
was harvested and assayed by liquid scintillation counting. Data are expressed
as counts per minute (c.p.m.). Cytokine profiling was performed on superna-
tants 48 h after antigen-restimulation by Bio-Plex cytokine assay (Bio-Rad,
Hercules, CA) according to the manufacturer’s instructions.
Surface plasmon resonance. Experiments were conducted using a Biacore
3000 biosensor (GE Healthcare) in 0.01 M HEPES, 0.15 M NaCl, 3 mM EDTA,
0.005% surfactant p20 (pH 7.4) at 25 °C. PEI, gp140 or BSA were immobilized
by amine coupling on separate CM5 sensor chip flow cells to produce 4,000,
17,000 or 11,000 RU, respectively. Samples of soluble PEI (50 µg/mL), gp140
(50 µg/mL) or BSA (50 µg/mL) were injected for 5 min at a flow rate of 5 µL
per minute. An activated and ethanolamine-blocked empty flow cell was used
as negative control.
Scanning electron microscopy (SEM). For SEM, 12 µg PEI was used alone
or in combination with 4 µg gp140 (Polymun) or 18.8 µg (N:P = 6) salmon
sperm DNA (Invitrogen) in 1 mM HEPES. Samples were fixed with 2.5%
glutaraldehyde and aliquots deposited on clean 13 mm silane (3 aminopro-
pyltriethoxysilane)-coated coverslips. The samples were kept overnight at
4 °C to allow the material to attach to the glass after which they were carefully
rinsed several times with distilled H2O, dehydrated through a series of ethanols
nature biotechnology
 and critically point-dried. After sputter coating with gold the samples were
examined in a JOEL JSM 6390 scanning electron microscope.
In vitro antigen uptake assay. Fluorescently labeled gp140647 antigen
(2 µg/mL) was pre-incubated with PEI (0.1 µg/mL) or PBS for 3 h at 4 °C with
shaking. Antigen, PEI or PEI-antigen polyplexes were then co-incubated with
either BMDC or LA4 epithelial cells for 4 h at 37 °C in 24-well plates. Cells were
then repeatedly washed in PBS and harvested by EDTA (5 mM) treatment.
Antigen uptake was determined by flow cytometry in duplicate wells and data
pooled from three independent experiments. In parallel assays, BMDCs were
adhered to coverslips before 4 h incubation with either gp140647 (2 µg/mL) or
gp140647 pre-incubated with PEI (0.1 µg/mL). Hoechst 34580 dye (2 µg/mL,
Molecular Probes) and WGA555 (wheat germ agglutinin, 5 µg/mL, Molecular
Probes) staining was used in the final 10 min of incubation to label the nucleus
and cell surface, respectively. Antigen or PEI-antigen uptake was then visual-
ized by confocal microscopy via Olympus FV1000, IX81, 60× N/A (1.35) and
FluoView software (Olympus).
In vitro apoptosis/cytotoxicity assay. The epithelial-derived cell line LA4
was co-incubated with increasing doses of PEI (0.01–10 µg/mL) for 18 h at
37 °C. Cells were then harvested by EDTA (5 mM) treatment followed by
cell scraping and stained for viability using the LIVE/DEAD viability dye
(Molecular Probes) followed by Annexin V staining (Molecular Probes) in
© 2012 Nature America, Inc. All rights reserved.
appropriate buffers. Cell death/apoptosis was examined by flow cytometry
via FACSCalibur (BD) and FlowJo software (Tree Star Inc).
In vivo recruitment and antigen uptake assay. Mice were immunized
intranasally with PEI (20 µg/mL), gp140647 (10 µg/mL) or PEI/gp140647
polyplexes (2:1) and 4 or 24 h later NALT tissue and CLNs removed for
analysis. Single-cell suspensions were prepared in PBS-EDTA (2 mM) by
gentle teasing and then cells were passed through a sieve to remove aggre-
gates. Fc Receptors were bound by Fc Block (BD) and cells were then stained
for flow cytometry and B cells (CD45+ CD11b– CD19+ MHC-II+), T cells
(CD45+ CD11b– MHC-II+ CD3+), monocytes (CD45+ CD11b+ Ly6C2+ F4/
80int), macrophages (CD45+ CD11b+ F4/80hi CD11c– Ly6Clo), neutrophils
(CD45+ CD11b+ Ly6C+ F4/80–), dendritic cells (CD45+ CD11c+ F4/80–/+
MHC-IIhi) epithelial cells (CD45– E-cadherin+ Microfold–) and M cells (CD45–
E-­cadherin+/− Microfold+) examined by CyAn ADP analysis (Beckman
Coulter, USA) and FlowJo software (Tree Star Inc.). Gp140647-positive gates
were based on the ‘no antigen’ control group and corrected for nonspecific flu-
orescence. Absolute cell numbers were determined in parallel by flow cytom-
etry using Liquid Counting Beads (BD). The following antibodies were used:
npg
CD11b (M1/70), CD19 (ID3), CD3 (145-2C11), Ly6C (AL-21), F4/80 (CI:
A3-1), CD11c (HL3), MHC-II (2G9), CD16/CD32 (2.4G2), CD45 (30-F11),
E-cadherin (S-17), anti-Microfold mAb (NKM 16-2-4).
In vitro stimulation of BMDCs and BMMs. BMDCs were generated as
described earlier39. For maturation, cells were incubated overnight with
1 µg/mL LPS (Sigma) or 10 µg/mL PEI (25 kDa, Sigma) and subsequently
analyzed by flow cytometry using the following antibodies: MHC-II (2G9),
CD16/CD32 (2.4G2), CD40 (3/23), CD80 (RM80), CD86 (GL1). BMMs
were derived from bone marrow cells cultured for 7 d in R10 medium
nature biotechnology
s­ upplemented with 15% L-929 cell-conditioned medium. BMM were
LPS-­stimulated overnight or left unstimulated, followed by incubation with
15 µg/mL Alhydrogel (Alum), 20 µg/mL PEI or 4 h or 10 µg/mL ATP for
30 min. Secretion of IL-1β and IL-18 were measured by ELISA (R&D Systems
and Bender Medsystems, respectively). Mature caspase-1 was detected in
cell lysates separated by gel electrophoresis and subjected to western blot
analysis using rabbit anti-caspase-1 (M-20, Santa Cruz Biotechnology) for
visualization of Caspase-1 p10 (apparent MW 17 kDa as verified by blocking
with M-20 peptide, data not shown).
In vitro cross-presentation assay. Splenic dendritic cells were isolated by
digestion, density gradient centrifugation, followed by positive selection
using αCD11c-microbeads (Miltenyi Biotec, Cologne, Germany) as per the
manufacturer’s instructions. Dendritic cell were pulsed with OVA (50 µg/mL)
alone, OVA (50 µg/mL) pre-complexed with PEI (0.002–0.2 µg/mL), OVA
(50 µg/mL) in the presence of CpG (1 µg/mL) or SIINFEKL peptide (20 µg/
mL) for 3 h at 37 °C. dendritic cell were washed and 5 × 104 cells co-incubated
(1:1) with the lacZ-inducible SIINFEKL-specific CD8+ T-cell hybridoma B3Z,
for 24 h at 37 °C. Hybridoma responses were assessed by intracellular β-galac-
tosidase activity and read by spectrophotometer (OD 595nm/635nm).
In vivo CD8+ T-cell induction. C57Bl/6 mice were immunized four times
intranasally with gp140 (10 µg) or gp140 + PEI (20 µg). Two weeks after
the final booster, mice were euthanized and CLN removed and single-cell
suspensions prepared. Lymphocytes were restimulated with medium, gp140
(50 µg/mL) or concanavalin A (1 µg/mL) at 37 °C for 4 h before Brefeldin
A (10 µg/mL) added and cells incubated for a further 18 h. Cells were then
washed and fixed in 4% paraformaldehyde before surface staining for CD8α
(53-6.7), CD4 (GK1.5) and CD3ε (145-2C11). After further washing, cells were
permeabilized in Perm/Wash buffer (BD) and stained for intracellular IFNγ
(XMG1.2) and examined by CyAn ADP analysis (Beckman Coulter, USA) and
FlowJo software (Tree Star Inc.). Absolute cell numbers were determined in
parallel by flow cytometry using Liquid Counting Beads (BD).
Reporter cell assays. B16-blue IFNα/β-cells, THP1-blue CD14 cells and
C3H/WT MEF cells (Invivogen) were cultured and used according to the
supplier’s instructions.
33. Wegmann, F. et al. A novel strategy for inducing enhanced mucosal HIV-1 antibody
responses in an anti-inflammatory environment. PLoS ONE 6, e15861 (2011).
34. Skehel, J.J. & Waterfield, M.D. Studies on the primary structure of the influenza
virus hemagglutinin. Proc. Natl. Acad. Sci. USA 72, 93–97 (1975).
35. Del Campo, J. et al. Intranasal immunization with a proteoliposome-derived
cochleate containing recombinant gD protein confers protective immunity against
genital herpes in mice. Vaccine 28, 1193–1200 (2010).
36. Sutterwala, F.S. et al. Critical role for NALP3/CIAS1/Cryopyrin in innate and adaptive
immunity through its regulation of caspase-1. Immunity 24, 317–327 (2006).
37. Krashias, G. et al. Potent adaptive immune responses induced against HIV-1 gp140
and influenza virus HA by a polyanionic carbomer. Vaccine 28, 2482–2489 (2010).
38. Gomez Roman, V.R. et al. Development of standard operating procedures to obtain
longitudinal vaginal specimens from nulliparous rabbits as part of HIV vaccine
mucosal immunogenicity studies. J. Immunol. Methods 363, 29–41 (2010).
39. Inaba, K. et al. Generation of large numbers of dendritic cells from mouse bone
marrow cultures supplemented with granulocyte/macrophage colony-stimulating
factor. J. Exp. Med. 176, 1693–1702 (1992).
doi:10.1038/nbt.2344

Engineering phosphorus metabolism in plants to
produce a dual fertilization and weed control system
Damar Lizbeth López-Arredondo & Luis Herrera-Estrella
High crop yields depend on the continuous input of
orthophosphate (PO4−3)-based fertilizers and herbicides1,2.
Two major challenges for agriculture are that phosphorus
is a nonrenewable resource and that weeds have developed
broad herbicide resistance3–5. One strategy to overcome both
problems is to engineer plants to outcompete weeds and
© 2012 Nature America, Inc. All rights reserved.
microorganisms for limiting resources, thereby reducing the
requirement for both fertilizers and herbicides. Plants and most
microorganisms are unable to metabolize phosphite (PO3−3),
so we developed a dual fertilization and weed control system
by generating transgenic plants that can use phosphite as a
sole phosphorus source. Under greenhouse conditions, these
transgenic plants require 30–50% less phosphorus input when
fertilized with phosphite to achieve similar productivity to that
obtained by the same plants using orthophosphate fertilizer
and, when in competition with weeds, accumulate 2–10 times
greater biomass than when fertilized with orthophosphate.
Drought, poor soil fertility and aggressive weeds pose major con-
straints to meeting the increasing demand for global food produc-
tion. Starting with the green revolution in the 1960s, higher yields
have been accompanied by a steady increase in the use of fertiliz-
ers and herbicides1,2. Fertilizers based on orthophosphate attract
npg
special attention because in addition to an essential role in central
metabolic processes for all living organisms6, phosphorus is a non-
renewable resource (with a rapidly increasing price) that may only
last from an estimated 70–200 years if current use is maintained3,4.
Orthophosphate availability is a key determinant of plant productivity
as it is the only chemical form of phosphorus that can be assimilated
by plants to meet their nutritional requirements. However, ortho-
phosphate availability is limiting for plant growth in about 67% of
cultivated soils7. Low orthophosphate availability is mainly due to
high reactivity of orthophosphate with soil components and rapid
conversion by soil bacteria of orthophosphate to organic forms that
cannot be taken up by plants. Owing to both of these factors, as little
as 20–30% of the orthophosphate that is applied as fertilizer is actually
used by cultivated plants8,9. The inefficient utilization of orthophos-
phate present in fertilizer is further aggravated by the competition of
weeds with crops for soil resources2. Moreover, because many weeds
have developed herbicide resistance, excessive applications of both
orthophosphate fertilizers and herbicides are nowadays routine,
Laboratorio Nacional de Genómica para la Biodiversidad, Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Nacional, Irapuato, Guanajuato,
México. Correspondence should be addressed to L.H.-E. (lherrera@langebio.cinvestav.mx).
Received 7 May; accepted 1 August; published online 26 August 2012; doi:10.1038/nbt.2346
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 889
letters
which not only increases production costs and food prices, but
also poses serious environmental problems2,5,10. Orthophosphate-
agricultural run-off can increase orthophosphate concentrations by
two- to threefold in rivers and oceans, thereby promoting the forma-
tion of toxic algae blooms that are a major ecological problem11.
With orthophosphate reserves rapidly declining, research has focused
on the development of alternative fertilization schemes to reduce
orthophosphate application, cost-effective technologies for phos-
phorus recycling and the design of more effective weed-management
systems3,6,12. Because orthophosphate cannot be substituted in plant
nutrition, relatively little attention has been given to the use of other
chemical forms of phosphorus to formulate effective and potentially
less environmentally hazardous fertilizers. Phosphite, a reduced form
of phosphorus, was proposed as a promising alternative fertilizer after
the Second World War, owing to its distinct chemical and biochemi-
cal properties compared with orthophosphate12,13, including higher
solubility, lower reactivity with soil components and the inability of
most microorganisms to use it as a phosphorus source14–16. However,
plants cannot metabolize phosphite, limiting its use as a fertilizer17–23.
As phosphite inhibits plant growth, its use as a broad-spectrum,
nonselective, weed-management system has been proposed 13,24.
The reported beneficial effects of phosphite on productivity of
plants could be due to inhibition of weed growth, and other well-
­documented properties such as reducing the incidence of diseases
caused by Phythophthora species (phosphite inhibits the growth of
oomycetes and activates plant defense mechanisms)21,24,25.
A few bacterial strains that are capable of oxidizing phosphite
to produce orthophosphate have been described previously 16. In
Pseudomonas stutzeri WM88, the ptxD gene encodes a phosphite-
specific oxido­reductase that oxidizes phosphite using NAD+ as a
cofactor, to produce orthophosphate and NADH as products26,27.
To test whether a pathway to convert phosphite into orthophosphate
could be engineered into plants, we produced transgenic Arabidopsis
lines expressing the ptxD coding sequence (ptxDAt lines) under the
control of the constitutive CaMV 35S promoter (35SøPTXD) (Fig. 1
and Supplementary Fig. 1). To confirm that 35S:PTXD plants are
capable of using phosphite as a phosphorus source, we germinated
seeds from transgenic lines and nontransformed siblings in media
lacking orthophosphate and supplemented with phosphite. As previ-
ously reported17–23, control seedlings are unable to metabolize phos-
phite, displaying restricted development and achieving a maximum
 letters
Figure 1  Transgenic Arabidopsis plants that express a phosphite
oxidoreductase can use phosphite as a phosphorus source. (a) Comparative
growth of seedlings from ptxDAt-3, 5 and 7 lines, and WT (Col-0) in solid
media containing 1 mM phosphate (Pi) or 1 mM phosphite (Phi).
(b) Transgenic Arabidopsis lines and WT control grown in vertical plates
containing 1 mM phosphite. (Phi) (c) Transgenic and WT plants grown in a
sterilized mixture of sand and vermiculite (left to right) not fertilized
(left-hand panel, control) or fertilized with either orthophosphate (middle
panel) or phosphite (right-hand panel) as phosphorus source (phosphite
and orthophosphate applied at 40 mg kg−1). Inset: top view of WT plants
with phosphite. Plants are representative of two independent experiments
(n = 8–10 plants/treatment each). Scale bars, 5 cm.
size of 3–6 mm in length before their growth is completely arrested.
In contrast, transgenic seedlings harboring the 35S:PTXD construct
can use phosphite as a phosphorus source to sustain growth similar
to that observed for transgenic and control plants grown in media
containing orthophosphate (Fig. 1a). ptxDAt lines cultivated in
vertically oriented agar plates displayed vigorous growth with well-
developed root systems, whereas control seedlings were arrested at
the cotyledonary stage, developing short primary roots with little or
no lateral root formation (Fig. 1b). These results show that transgenic
© 2012 Nature America, Inc. All rights reserved.
plants expressing ptxD are unaffected by the growth-inhibitory effect
of phosphite and can use phosphite as a phosphorus source.
To determine whether PTXD Arabidopsis plants could efficiently use
phosphite as a phosphorus source under greenhouse conditions, ptxDAt
and control seedlings were transferred to pots containing a sterilized
mixture of sand and vermiculite supplemented with orthophosphate or
phosphite as phosphorus sources. ptxDAt lines fertilized with phosphite
clearly showed vigorous growth, produced biomass and accumulated
phosphorus at levels indistinguishable from those of control and trans-
genic plants grown with orthophosphate (Fig. 1c and Supplementary
Fig. 2), whereas control plants grown with phosphite as the sole phos-
phorus source displayed more restricted growth than those grown in
unfertilized substrate and survived for a maximum of 20 d (Fig. 1c).
These results confirm that the addition of phosphite further reduces
plant growth when orthophosphate availability is low22,23 and suggest
the possibility of controlling the growth of plants that are unable to
metabolize phosphite, such as weeds, under these conditions.
To determine whether expression of the ptxD gene could enable
npg
other plant species to use phosphite as a phosphorus source, we pro-
duced transgenic tobacco plants harboring the 35SøPTXD construct
(ptxDNt lines) (Supplementary Fig. 3). Transgenic and control seed-
lings were transferred to plastic bags containing a sterilized mixture
of sand and vermiculite amended with either phosphite or ortho-
phosphate as the phosphorus source. Both WT and transgenic plants
showed considerably poorer growth in the unfertilized substrate com-
pared with plants fertilized with orthophosphate (Fig. 2a). Upon sup-
plementation of the growth substrate with phosphite, the development
of WT plants was completely arrested and the plants died a few days
after transplantation, whereas ptxDNt plants had vigorous growth,
normal shoot and root pheno­types, and performed quantitatively as
well as WT and ptxDNt plants fertilized with orthophosphate (Fig. 2a
and Supplementary Fig. 4). As orthophosphate is essential for CO2
fixation in the chloroplast, and plants suffering from orthophosphate
starvation have reduced levels of photosynthesis28, we measured the
rate of photosynthesis in transgenic and control plants under dif-
ferent phosphorus treatments. When fertilized with either ortho-
phosphate or phosphite, ptxDNt plants had rates of photosynthesis
similar to those observed for control plants grown in orthophos-
phate (Supplementary Fig. 4), suggesting that expression of ptxD or
phosphite fertilization has no measurable effect on photosynthesis
890 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology
a Pi Phi b Phi
ptxDAt-
WT ptxDAt-3 WT ptxDAt-3 7 5 3 WT
ptxDAt-7 ptxDAt-5 ptxDAt-7 ptxDAt-5
T2
c No P Pi Phi
WT ptxDAt-3 ptxDAt-5 WT ptxDAt-3 ptxDAt-5 WT ptxDAt-3 ptxDAt-5
activity in these plants. Although phosphite is not toxic to animals
or humans according to the US Food and Drug Administration, it
was important to determine whether phosphite was still present in
ptxDNt plants. Phosphite content in leaves, flowers and fruits of
ptxDNt plants fertilized with phosphite was below detection levels
(< 0.1 nmoles g−1) and only orthophosphate was present, suggesting
that most, if not all, phosphite had been oxidized to orthophosphate
in adult tobacco plants (Supplementary Fig. 5). Phosphite accumu-
lation and its effect on photosynthesis and development in control
plants fertilized with phosphite could not be evaluated because these
plants died soon after transplantation.
A crucial step to validate the potential of our strategy was to deter-
mine whether the system works in soils in which microorganisms are
present. Therefore, we performed greenhouse experiments using a non-
sterile, low orthophosphate, alkaline soil obtained directly from a field
of the central agricultural region of Mexico (Guanajuato State), that
contained cations, organic matter and native soil microflora (Online
Methods). In these experiments, unfertilized and orthophosphate-
fertilized ptxDNt and WT plants had similar phenotypes (Fig. 2b).
When fertilized with phosphite, germination of WT seeds was delayed
and seedling growth was compromised, causing death 8–15 d after
germination. In contrast, ptxDNt plants germinated normally and
seedlings showed vigorous growth (Fig. 2b). In the different phosphite
fertilization treatments, capsule, seed and total biomass of ptxDNt
plants was 15–25% higher than that produced by plants fertilized
with the same amount of orthophosphate (Fig. 2c). Moreover, total
biomass and seed production of ptxDNt plants fertilized with
30 mg kg−1 of phosphite was similar to that produced by plants fer-
tilized with 60 mg kg−1 of orthophosphate. ptxDNt plants fertilized
with 60 mg kg−1 of phosphite produced 16% and 20% higher total
­biomass and seed production, respectively, than plants fertilized with
the same amount of orthophosphate (Fig. 2c). Similar results were
obtained when WT and ptxDNt plants were grown in a low ortho-
phosphate, nonsterile acidic soil, obtained from a field in the north-
west agricultural region of Mexico (Sinaloa State) (Online Methods
and Supplementary Fig. 6). Together, these results suggest that the
native microbial activity in these agricultural soils is insufficient
either to convert phosphite into sufficient orthophosphate to support
the growth of WT plants or to eliminate the plant growth inhibitory
activity of the supplied phosphite.
In addition to not being metabolized by WT plants, phosphite also
inhibits the growth of plants and weeds by interfering with the ­signaling
pathways that induce the rescue system activated in orthophosphate-
starved plants22,23. Therefore, cultivation of transgenic plants express-
ing the ptxD gene combined with phosphite as a ­phosphorus fertilizer
 letters

Figure 2  Effect of phosphite fertilization on a No P Phosphate Phosphite

WT and transgenic tobacco plants in a low- WT ptxDNt-36 WT ptxDNt-36
phosphate nonsterile soil. (a) 65-d-old WT and
ptxDNt-36 plants fertilized with orthophosphate

WT
or phosphite in a sterile sand and vermiculite
mixture. (b) 4-month-old ptxDNt-80 and WT

ptxDNt-36
plants grown under different orthophosphate
(left) or phosphite (right) fertilization regimes
in the nonsterile alkaline agricultural soil 2 20 40 80 20 40 80 20 40 80 20 40 80 mg kg
-1

(Online Methods). (c) Total biomass (leaves

plus stem biomass), capsule and seed biomass b c 80
70
*
produced by 4-month-old ptxDNt-80 (black *

Total biomass
60

(g per plant)
bars) and WT (white bars) plants grown under 50 *
40
different orthophosphate (Pi) or phosphite (Phi) 30
fertilization regimes in the nonsterile alkaline 20
agricultural soil 2. Treatment with no addition 10
0

WT
of P fertilizer (No P) was used as a control. mg kg–1

0
10
30
60
10
30
60
Data represent the average of two independent
Pi Phi
experiments (n = 10 plants/treatment/line). *,
P < 0.005, compared to corresponding control. 30 *
Error bars, mean ± s.e.m. 25 *

(g per plant)
20

Capsules
15
could serve as a weed control system in soils 10
© 2012 Nature America, Inc. All rights reserved.

with low orthophosphate availability. To assess 5

this possibility, we first tested whether the grass 0
weed false brome (Brachypodium distachyon) mg kg–1

0
10
30
60
10
30
60
and the broad-leaved weed tall morning-glory Pi Phi
ptxDNt-80

20
(Ipomoea purpurea) were able to use phosphite *
as a phosphorus source. We found that neither 15 *

(g per plant)
of these two weeds is able to sustain growth

Seed
10
*
in soil fertilized with phosphite (Fig. 3a and 5
Supplementary Fig. 7). Then, we carried out
0
greenhouse growth competition experiments
mg kg–1

0
10
30
60
10
30
60
between ptxDNt plants and B. distachyon using 0 10 30 60 10 30 60
–1 Pi Phi
a nonsterile alkaline soil obtained directly from mg kg

a field of the northeast agricultural region of

Mexico (Coahuila State), fertilized with either phosphite or orthophos-
phate. Both unfertilized B. distachyon and tobacco plants grew poorly in
comparison to the same plants fertilized with orthophosphate (Fig. 3b).
Supplementation with orthophosphate resulted in weeds growing
significantly faster (P < 0.005) than tobacco plants (Fig. 3b). In con-
npg
the efficacy of this technology. A possible solution to this problem
would be to apply phosphite as a foliar fertilizer rather than incorpo-
rating it into the soil. Foliar application of phosphite was shown to be
effective in providing phosphorus to support the growth of ptxDNt
plants and to inhibit the growth of WT plants (Supplementary Fig. 11).

trast, fertilization with phosphite resulted in limited growth of weed
plants and vigorous growth of ptxDNt plants (Fig. 3b). Upon addition of
120 mg kg−1 of phosphite to the growth substrate, the biomass of ptxDNt
plants increased 1,000%, whereas that of the grass weed decreased 50%
compared with supplementation with orthophosphate fertilization
(Fig. 3c). Similar results were obtained in competition experiments
between transgenic tobacco and three other agronomically impor-
tant weeds, namely tall morning-glory, Alexander grass (Brachiaria
plantaginea) and smooth pigweed (Amaranthus hybridus)
(Supplementary Figs. 8–10), for which resistance to herbicides has
already been reported2,5. In competition experiments using the alkaline
agricultural soil from Coahuila State, weeds were not killed by phos-
phite fertilization, which is in line with the proposed strategies of an
agro-ecological agriculture, in which the use of weed control systems
aims to allow cultivated plants to attain optimal production without
killing weeds.
Evaluation of transgenic tobacco plants in agricultural soils showed
that the presence of native micro-flora in the soil does not interfere
with the effectiveness of phosphite as a weed control and phosphorus-
fertilization system. However, it is possible that bacteria that are able to
metabolize reduced forms of phosphorus could be enriched in the soil
through repeated use of phosphite as a fertilizer, ­potentially ­reducing
The design of foliar fertilization schemes using phosphite could pre-
vent or delay the enrichment of soil bacteria capable of metaboliz-
ing phosphite or reduce the potential impact of such bacteria on the
efficacy of this technology.
The transgenic system reported here exploits the chemical and
biological properties of phosphite to increase the capacity of cul-
tivated plants to compete with weeds and soil microorganisms for
limiting resources such as nutrients and water, potentially providing
significant advantages over current orthophosphate fertilizer systems.
Although the system remains to be tested under field conditions, our
results suggest that in soils with low orthophosphate availability, the
use of PTXD transgenic plants in conjunction with phosphite ferti-
lization might reduce production costs and energy consumption by:
(i) reducing the amount of phosphorus fertilizer needed to achieve
optimal plant productivity, (ii) achieving fertilization and weed con-
trol with a single treatment, and (iii) reducing or eliminating the cost
of additional herbicides. The results obtained by incorporating phos-
phite into the soil, or by foliar application, suggest that phosphite
could be used as both a pre- and post-emergence weed control agent.
A potential additional benefit of the phosphite-based system is that
in contrast to resistance to herbicides that can be achieved by differ-
ent mechanisms, such as those preventing uptake, sequestering the

nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 891

 letters

a No P Phosphate Phosphite b No P Phosphate Phosphite

Figure 3  Effectiveness of the PTXD/phosphite system for weed control in a

nonsterile soil. (a,b) Comparative growth of the grass weed Brachypodium
distachyon alone (a) or in competition with ptxDNt-80 transgenic tobacco
plants (b) in the nonsterile alkaline soil 2 (Online Methods), respectively,
© 2012 Nature America, Inc. All rights reserved.

treated with no addition of a P source (No P) as a control or amended with

120 mg kg−1 of orthophosphate (phosphate) or phosphite 40 d after c 2.5
germination. (c) Quantification of biomass production of B. distachyon *
(white bars) and transgenic tobacco plants (black bars) of the plants shown 2.0

Biomass (g per plant)

in b under orthophosphate (phosphate, Pi) or phosphite (Phi) 80 or
120 mg kg−1 regimens. The data represent the average and standard error 1.5
of two independent experiments with three replicates per treatment.
*, P < 0.005, compared to corresponding control. Error bars, mean ± s.e.m. 1.0 *
0.5 *
herbicide in specific subcellular compartments or making the target *
enzyme resistant to the herbicide5, the only effective mechanisms by 0
0 80 120 80 120
–1
mg kg
which weeds could escape phosphite control would be the acquisition Pi Phi
of the capacity to metabolize phosphite.
It will be crucial to assess any potential hazards due to genetic mod- orthophosphate fertilization, and of the potential negative or posi-
ification. Phosphite is classified as an organic fertilizer and is already tive environmental impacts of this technology. Such data would also
extensively used in agriculture in several countries, and in 1997 it inform the potential for the evolution of phosphite-metabolizing
was approved by the US Environmental Protection Agency for use as weeds and microorganisms, which could interfere with the efficacy
a fungicide in many food and non-food crops because it is innocu- of large-scale phosphite fertilizer application. In summary, the pro-
ous to humans and animals. Moreover, the products of the reaction, duction of transgenic crop plants able to utilize phosphite, together
npg

orthophosphate and NADH, are also innocuous and present in all
living cells. Additionally, because phosphite is not naturally present
in soils, the potential escape of PTXD-encoding transgenes into wild
species should have no effect on genetic diversity or survival outside
agricultural systems. Another potential advantage of the phosphite/
PTXD fertilization scheme is that its use could decrease the environ-
mental impact of current orthophosphate-based fertilization systems
on aquatic ecosystems. In contrast to orthophosphate runoff that
even at micromolar concentrations promotes massive algal growth
that leads to oxygen depletion in rivers, lakes and oceans, causing
the death of other organisms such as fish10, phosphite is not metabo-
lized by algae29 and therefore cannot promote toxic algae blooms.
Moreover, at the micromolar range, in which orthophosphate is
present in contaminated rivers or oceans, phosphite has little or no
toxicity to algae and other aquatic organisms, making a negative effect
on aquatic biodiversity unlikely.
Although under greenhouse conditions this technology was effec-
tive in reducing fertilizer use and in reducing the growth of the tested
weeds, field experiments in multiple testing sites with a variety of
soil chemistries will be required to validate its commercial viability.
These field studies should include a careful evaluation of the amount
of leaching of phosphite into runoff in comparison with traditional
with the application of phosphite as a source of phosphorus, might
potentially become an effective phosphorus-fertilization and weed
control scheme in the almost 67% of cultivated land with low ortho-
phosphate availability.
Methods
Methods and any associated references are available in the online
version of the paper.
Accession codes. Sequence data from this article can be found in the
Arabidopsis Information Resource or GenBank/EMBL database under
the following accession numbers: ACT2 (At3g18780), ACTINNT
(GQ339768), PTXD (AF061070).
Note: Supplementary information is available in the online version of the paper.
Acknowledgments
We thank O. Martínez for advice in statistical analysis; M.A. Leyva, A. Vera and
J.L. Cabrera for technical support; K. Wrobel and K. Wrobel for the phosphite
detection method; V. Limones for assistance in Tobacco transformation and
C. Castro and E. Alva in Arabidopsis greenhouse experiments. We are grateful
to A. Estrada, J.D. Frier and Universidad Autónoma Chapingo for providing
Brachypodium distachyon, Amaranthus hybridus and Ipomoea purpurea and
Brachiaria plantaginea seeds, respectively; and C. Morales for assistance in

892 VOLUME 30  NUMBER 9  SEPTEMBER 2012  nature biotechnology


photosynthesis measurements. We thank S. Gillmore and V. Albert for critical
reading this manuscript. This work was supported in part by grant from the
Howard Hughes Medical Institute (grant 55005946) to L.H.-E. D.L.L.-A. is
indebted to CONACyT, México, for a PhD fellowship (no. 207308). pWM302
was a gift from W.W. Metcalf.
AUTHOR CONTRIBUTIONS
D.L.L.-A. and L.H.-E. designed experiments, analyzed data and wrote the paper.
D.L.L.-A. performed experiments.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Published online at http://www.nature.com/doifinder/10.1038/nbt.2346. starvation response in Brassica nigra seedlings. Plant Physiol. 110, 105–110
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
1. Tilman, D. et al. Agricultural sustainability and intensive production practices.
Nature 418, 671–677 (2002).
2. Gianessi, L.P. & Reigner, N.P. The value of herbicides in US crop production. Weed
Technol. 21, 559–566 (2007).
3. Dawson, C.J. & Hilton, J. Fertiliser availability in a resource-limited world: production
and recycling of nitrogen and phosphorus. Food Policy 36, S14–S22 (2011).
4. Cordell, D., Drangert, J.-O. & White, S. The story of phosphorus: global food security
© 2012 Nature America, Inc. All rights reserved.
and food for thought. Glob. Environ. Change 19, 292–305 (2009).
5. Powles, S.B. & Yu, Q. Evolution in action: plants resistant to herbicides. Annu.
Rev. Plant Biol. 61, 317–347 (2010).
6. Gilbert, N. The disappearing nutrient. Nature 461, 716–718 (2009).
7. Cakmak, I. Plant nutrition research: priorities to meet human needs for food in
sutainable ways. Plant Soil 247, 3–24 (2002).
8. Hinsinger, P. Bioavailability of soil inorganic P in the rhizosphere as affected by
root-induced chemical changes: a review. Plant Soil 237, 173–195 (2001).
9. Syers, J.K., Johnston, A.E. & Curtis, D. Efficiency of soil and fertilizer phosphorus
use: reconciling changing concepts of soil phosphorus behavior with agronomic
information. Fertilizer and Plant Nutrition Bulletin 18 (Food and Agriculture
Organization of the United Nations, 2008).
10. Carpenter, S.R. Phosphorus control is critical to mitigating eutrophication.
Proc. Natl. Acad. Sci. USA 105, 11039–11040 (2008).
11. Downing, J.A., Watson, S.B. & McCauley, E. Predicting cyanobacterial dominance in
lakes. Rapid communication. Can. J. Fish. Aquat. Sci. 58, 1905–1908 (2001).
12. Abelson, P.H. A potential phosphate crisis. Science 283, 2015 (1999).
npg
nature biotechnology  VOLUME 30  NUMBER 9  SEPTEMBER 2012 893
letters
13. Ruthbaum, H.P. & Baille, W.J.H. The use of red phosphorus as a fertilizer. Part 4.
Phosphite and phosphate retention in soil. N. Z. J. Sci. 7, 446–451 (1964).
14. Morton, S. et al. Analysis of reduced phosphorus in samples of environmental
interest. Environ. Sci. Technol. 39, 4369–4376 (2005).
15. Pasek, M. Rethinking early earth phosphorus geochemistry. Proc. Natl. Acad. Sci.
USA 105, 853–858 (2008).
16. White, A.K. & Metcalf, W.W. Microbial metabolism of reduced phosphorus
compounds. Annu. Rev. Microbiol. 61, 379–400 (2007).
17. Schroetter, S., Angeles-Wedler, D., Kreuzig, R. & Schnug, E. Effects of phosphite
on phosphorus supply and growth of corn (Zea mays). Landbauforschung Völkenrode
56, 87–99 (2006).
18. Ouimette, D.G. & Coffey, M.D. Phosphonate levels in avocado (Persea amercana)
seedlings and soil following treatment with Fosetyl-Al or potassium phosphonate.
Plant Dis. 73, 212–215 (1989).
19. Carswell, C. et al. The fungicide phosphonate disrupts the phosphate-
(1996).
20. Carswell, M.C., Grant, B.R. & Plaxton, W.C. Disruption of the phosphate-starvation
response of oilseed rape suspension cells by the fungicide Phosphonate. Planta
203, 67–74 (1997).
21. Förster, H. et al. Effect of phosphite on tomato and pepper plants and on
susceptibility of pepper to Phytophthora root and crown rot in hydroponic culture.
Plant Dis. 82, 1165–1170 (1998).
22. Ticconi, C.A., Delatorre, C.A. & Abel, S. Attenuation of phosphate starvation
responses by phosphite in Arabidopsis. Plant Physiol. 127, 963–972 (2001).
23. Varadarajan, D.K. et al. Phosphite, an analogue of phosphate, suppresses the
coordinated expression of genes under phosphate starvation. Plant Physiol. 129,
1232–1240 (2002).
24. McDonald, A.E., Grant, B.R. & Plaxton, W.C. Phosphite (phosphorus acid): its
relevance in the environment and agriculture and influence on plant phosphate
starvation response. J. Plant Nutr. 24, 1505–1519 (2001).
25. Saindrenan, P., Barchietto, T. & Gilbert, B. Modification of the phosphite induced
resistance response in leaves of cowpea infected with Phytophthora cryptogea by
α-aminooxyacetate. Plant Sci. 58, 245–252 (1988).
26. Costas, A.M.G., White, A.K. & Metcalf, W.W. Purification and characterization of a
novel phosphorus-oxidizing enzyme from Pseudomonas stutzeri WM88. J. Biol.
Chem. 276, 17429–17436 (2001).
27. Metcalf, W.W. & Wolfe, R.S. Molecular genetic analysis of phosphite and
hypophosphite oxidation by Pseudomonas stutzeri WM88. J. Bacteriol. 180,
5547–5558 (1998).
28. Rao, I.M., Arulanantham, A.R. & Terry, N. Leaf phosphate status, photosynthesis
and carbon partitioning in sugar beet. II. Diurnal changes in sugar phosphates,
adenylates, and nicotinamide nucleotides. Plant Physiol. 90, 820–826 (1989).
29. Lee, T.-M. The effects of phosphite on phosphate starvation responses of Ulva
lactuca (Ulvales, Chlorophyta). J. Phycol. 41, 975–982 (2005).
 ONLINE METHODS
Design of 35S:ptxD. To obtain transgenic PTXD plants, the complete coding
sequence of ptxD gene from P. stutzeri WM88 was amplified from pWM302
(kind gift of W.W. Metcalf 8), and placed under the control of CaMV35S pro-
moter in vector pB7WG2D (Gateway Technology) using the following primers
and standard conditions:
PTXDFWB1 5′- GGGGACAAGTTTGTACAAAAAAGCAGGCTAAATGC­
TGCCGAAACTCGTTATAACTC-3′
PTXDRVB2 5′- GGGGACCACTTTGTACAAGAAAGCTGGGTATCAAC­
ATGCGGCAGGCTC-3′
The ptxD amplification fragment was subjected to two sequential site-
specific recombination reactions, using pDONR221 donor and pB7WG2D
destination vectors. Vectors were subjected to restriction analysis and DNA
sequencing to confirm the presence of the expected sequences.
Generation and selection of transgenic plants. pB7WG2D:ptxD was intro-
duced into Agrobacterium tumefaciens strain GV2260 and used to produce
ptxDAt lines following a modified floral dip transformation protocol30.
To select transgenic plants, seeds were sown on growth medium (MS salts,
5 g L−1 sucrose, 10 g L−1 agar, pH 5.7) supplemented with 20 mg L−1 phos-
phynotricin. Thirty independent Arabidopsis lines were transferred to
soil (perlite/vermiculite/Canadian peat moss; 1:1:1) and self-pollinated. T2
lines with a 3:1 segregation (χ2 test, P < 0.05) for phosphinotricin resistance
© 2012 Nature America, Inc. All rights reserved.
were selected and homozygous T3 seeds stocks obtained. Fifteen transgenic
lines were assayed for their capacity to utilize phosphite as a unique phos­
phorus source; all were found to have this capacity. Presence and expression
of 35SøPTXD construct were analyzed in seven of these lines by Southern
blot hybridizations31 and qRT-PCR analyses32, respectively. Two homozygous
lines with high- (ptxDAt-3 and 4) and two with low-expression levels
(ptxDAt-5 and 7) were selected for further characterization.
To transform Nicotiana tabacum L. cv Xanthi, the leaf-disk method33 was
used. After co-cultivation, explants were transferred to selection medium
(MS salts, 30 g L−1 sucrose, 1 g L−1 N6-benzyladenine, 2.5 g L−1 gelrite, pH 5.7)
supplemented with 5 mg L−1 phosphynotricin and subcultured in fresh medium
every 2–3 weeks. Regenerated shoots were cultured on selective medium
supplemented with 10 mg L−1 of indolacetic acid to induce root formation.
Growth of eight lines randomly selected lines in phosphite-containing media
confirmed that all were capable of using phosphite as a phosphorus source.
Presence and expression of the 35SøPTXD construct was confirmed in 80 and
7 randomly selected lines by genomic PCR and qRT-PCR32, respectively.
PCR and Southern blot analysis. For genomic PCR analysis, 200 ng of total
npg
genomic DNA were used to amplify the complete pxtD coding region using
the following primers and standard conditions:
PTXDFW 5′- ATGCTGCCGAAACTCGTTATAACTC -3′
PTXDRV 5′- TCAACATGCGGCAGGCTC -3′
For Southern blot hybridization analysis31, 15 µg of total DNA was digested
with EcoRI or EcoRV restriction enzyme (Invitrogen). A probe corresponding
to the first 400 bp of ptxD gene generated by PCR using PTXDRT primers was
used for hybridization experiments.
PTXDRTFW 5′- ATGCTGCCGAAACTCGTTATAACTC -3′
PTXDRTRV 5′- CTGCAAGCGATCAGCCATG -3′
Real time-PCR. Total RNA was isolated and Real-time qPCR determination
of ptxD transcripts (PTXDRT primers) was performed in an ABI PRISM
7500 thermocycler (Applied Biosystems) using the Arabidopsis Actin 2 (ACT2
primers) and tobacco ActinNt (ACTINNT primers) transcript levels for
normalization. Expression levels were obtained from at least three replicates.
ACT2FW 5′- ATATGGCATCATACCTTCTACAACGAGCTTCGTG-3′
ACT2RV 5′- ATATGGCATCATACCTTCTACAACGAGCTTCGTG-3′
ACTINNTFW5 5′- ATATGGCATCATACCTTCTACAACGAGCTTCGTG-´3
ACTINNTRV 5′- ATCCAACACAATACCAGTTGTACGACCACTAG-´3
Greenhouse experiments. For Arabidopsis greenhouse experiments, pots
were filled in with 0.3 kg of a sterile mixture of sand and vermiculite (1:1)
and mechanically mixed with either phosphate (KH2PO4) or phosphite
nature biotechnology
(NaH2PO3.5H2O) at 10, 20 and 40 mg kg−1. For tobacco experiments, 3 kg of
sterile sand and vermiculite mixture or a nonsterile agricultural soil (acidic
soil (from the northwest agricultural region of Mexico, Mocorito, Sinaloa):
pH 5.1, organic matter 1.4%, phosphorus (Bray) 5 mg kg−1, N 5 mg kg−1,
K 101.65 mg kg−1, Ca 3,607.2 mg kg−1, Mg 498.252 mg kg−1, Fe 7 mg kg−1,
Mn 26 mg kg−1; alkaline soil 1 (from the northeast agricultural region of
Mexico, Arteaga, Coahuila): pH 8.2, organic matter 0.39%, phosphorus (Bray)
3.62 mg kg−1, N 5.85 mg kg−1, K 64.20 mg kg−1, Ca 9,437.5 mg kg−1, Mg 404.75 mg kg−1,
Fe 3.32 mg kg−1, Mn 3.60 mg kg−1; alkaline soil 2 (from the central agricultural
region of Mexico, Celaya, Guanajuato): pH 8.7, organic matter 1.22%, phos-
phorus (Bray) 4.26 mg kg−1, N 17.56 mg kg−1, K 38.7 mg kg−1, Ca 14,000 mg
kg−1, Mg 712.5 mg kg−1, Fe 2.24 mg kg−1, Mn 1.66 mg kg−1) were weighed in
black plastic bags amended with different amounts of phosphorus (0, 10, 20,
30, 40 60 or 80 mg kg−1) as Pi (KH2PO4) or phosphite (KH2PO3) according
to each experiment. One plant per pot and 8–10 plants per line per treatment
were included. Pots were arranged following a completely random design in
the greenhouse. Arabidopsis aerial parts were cut off at the soil surface 36 d
after germination (dag), whereas tobacco aerial parts from alkaline and acidic
soil were harvested 4 months and 70 dag, respectively, and then dried in a
vacuum oven at 70–72 °C for biomass quantification. In both cases, experi-
ments with alkaline and acidic soils, capsules and seeds were collected by plant
4 months after germination. Photosynthesis rate was determined using a
portable Li-6200 photosynthesis system (Li-Cor, Lincoln, NE, USA) on fully
expanded, young leaves. Leaf area was calculated using the software ImageJ34.
To test the ability of weeds to use phosphite as a phosphorus source,
5 seeds of I. purpurea or B. distachyon were sown in 2 kg of nonsterile or sterile
sand and vermiculite mixture or 5 kg of nonsterile alkaline soil 2 amended
with phosphate or phosphite as phosphorus source. Weeds grown in sand and
vermiculite mixture were harvested for biomass quantification 20 dag and
that grown in alkaline soil 2 was harvested 37 dag. For growth competition
experiments, different proportions of seeds from ptxDNt-80 and the follow-
ing weeds were used (transgenic:weed): Ipomoea purpurea (13:30), Brachiaria
plantaginea (13:60), Amaranthus hybridus (5:15), Brachypodium distachyon
(50:50). Seeds were sown in a tray containing a nonsterile mixture of sand and
vermiculite (2 kg for experiments with Amaranthus hybridus and 6 kg for the
rest) or the alkaline agricultural soil (5 kg) as indicated for each experiment.
The source of phosphorus was provided trough either irrigation with a solution
(1 mM) or amendment in the soil with 60, 80 or 120 mg kg−1 of phosphate or
phosphite, as indicated in each experiment. Plants were harvested for biomass
quantification at 54 dag for Brachiaria and Brachypodium and at 40 dag for
Ipomoea and A. hybridus experiments using sand and vermiculite mixture. For
experiments with alkaline soil 2 Brachiaria and Brachypodium was harvested
at 57 or 70 dag, respectively, and for experiments with alkaline soil 1 both
was harvested at 50 dag. For foliar fertilization experiments, tobacco plants
growing in the alkaline soil 2 were sprayed with a 100 mM Pi (KH 2PO4) or
phosphite (KH2PO3) solution with Tween 20 0.1% (Sigma) every 15 d after
20 d old; deionized water was used as a control treatment. All parameters
were subjected to statistical analysis using ANOVA and Tukey tests (P < 0.05).
All plants were photographed when indicated for each experiment and photo­
graphs were adjusted in brightness and contrast when needed.
Phosphite and orthophosphate determination. Tissue from tobacco plants
was collected and lyophilized. Samples (40–50 mg) were extracted with 0.5 ml
of 10 mM EDTA pH 8.0, during 30 min in ultrasonic bath. After centrifuga-
tion (10,000g, 10 min), the supernatant was collected, cleaned up (Supelclean
LC-18 SPE tubes 3 ml, Supelco) and filtered (IC Acrodisc filter, 0.2 µm,
Sigma-Aldrich). Processed samples were analyzed in an Agilent series 1050
liquid chromatographic system equipped with a quaternary pump, a col-
umn oven and ChemStation (Agilent Technologies, Palo Alto, CA, USA).
The chromatographic column was Luna SAX18 (250 × 4.6 mm, 5 µm) from
Phenomenex. An ICP-MS platform (model 7500ce Agilent Technologies,
Tokyo, Japan) was used with a MiraMist Teflon nebulizer. The Peltier-cooled
chamber was operated at 2 °C. Tuning procedure was performed daily using
diluted Agilent solution (Li, Y, Tl, Ce, 1µg L−1 each). The HLPC-ICP-MC
instrument operating conditions were done using a 10 mM potassium phthal­
ate pH 4.0 mobile phase, isocratic elution, room temperature, 1.2 mL min–1
doi:10.1038/nbt.2346
 flow and 50 µL injection volume. The ICP-MS detection procedure was
performed using 1,500 W forward power, 0.9 l min−1 nebulizer gas flow,
0.1 l min−1 make-up gas, 100 ms dwell time, 3.5 mL min−1 He in the
collision/reaction cell, platinum sample and skimmer cones, 8-mm sample
depth, 31P channel monitored and time-resolved analysis acquisition mode.
To control polyatomic interferences from 15N 16O +, 14N 16O 1H + and
12C1H 16O+ ions potentially occurring at m/z = 31, the octopole reaction/
3
collision cell was used in kinetic energy discrimination mode. The gas flow
rate was selected to provide the highest possible signal to noise ratio, as
described elsewhere35. Commercial standards of two phosphorus species
were used for external calibration (between 0.2–2.5 nmoles of phosphate
and phosphite on column). The recovery of procedure, evaluated by standard
addition experiments, was 94.2% and 98.6% for Pi and phosphite, respec-
tively. Total phosphorus content was determined using vanadate-molybdate
colorimetric method36.
© 2012 Nature America, Inc. All rights reserved.
npg
doi:10.1038/nbt.2346
30. Martinez-Trujillo, M. et al. Improving transformation efficiency of Arabidopsis
thaliana by modifying the floral dip method. Plant Mol. Biol. Rep. 22, 63–70
(2004).
31. Sambrook, J. & Russell, D.W. Molecular Cloning: A Laboratory Manual (Cold Spring
Harbor Laboratory, 2001).
32. Livak, K.J. & Schmittgen, T.D. Analysis of relative gene expression data using real-
time quantitative PCR and the 2[-Delta Delta C (T)] method. Methods 25, 402–408
(2001).
33. Horsch, R.B. et al. A simple and general method for transferring genes into plants.
Science 227, 1229–1231 (1985).
34. Abramoff, M.D., Magelhaes, P.J. & Ram, S.J. Image processing with ImageJ.
Biophotonics Int. 11, 36–42 (2004).
35. Wrobel, K. et al. Phosphorus and osmium as elemental tags for the determination
of global DNA methylation- a novel application of high performance liquid
chromatography inductively coupled plasma mass spectrometry in epigenetic studies.
J. Chromatogr. B Analyt. Technol. Biomed. Life Sci. 878, 609–614 (2010).
36. Hesse, P.R. Soil phosphorus: its measurements and its uptake by plants. Aust. J.
Soil Res. 35, 227–239 (1971).
nature biotechnology
 careers and recruitment
Theory to practice: real-world case-based learning for
management degrees
Maria Theodosiou, Jean-Philippe Rennard & Arsia Amir-Aslani
For scientists looking to break into the biotech industry, scientific knowledge and technical skills are
only a partial requirement for success.
B iotech is a science-driven industry that
requires business leaders with technical
© 2012 Nature America, Inc. All rights reserved.
training. Success in the biotech sector hinges
only partially on scientific and technical skills.
In addition to the aforementioned skills, busi-
ness knowledge, management of people and
resources as well as teamwork are equally
essential for a positive outcome, yet the latter
are not skills that are part of a traditional sci-
ence PhD curriculum.
Today’s scientists are badly prepared
for the job market outside of academia.
Knowledgeable in their chosen discipline, they
often lack basic knowledge and understanding
of the business world, which makes them less
marketable as job candidates. Previously, we
highlighted the rise of professional master’s
degrees, which aim to create the next genera-
tion of leaders in biotech1. Here, we focus on
the unique educational approaches for training
npg
dual-skilled business leaders.
The complex environment of today’s bio-
tech industry, with the challenges of devel-
oping revolutionary drug therapies and
increasing R&D productivity, as well as inte-
grating scientific advances such as pharma-
cogenomics and high-throughput sequencing,
necessitates the right education to create
tomorrow’s leaders. Programs designed to
meet these challenges must remain flexible
and adaptable to industry’s ever-changing
needs. Traditional classroom learning is
no longer enough; practical experience is
required to educate and inform students.
The educational experience should be about
more than memorizing facts and concepts; it
Maria Theodosiou, Jean-Philippe Rennard and
Arsia Amir-Aslani are at the Grenoble School of
Management, Grenoble, France.
e-mail: arsia.amir-aslani@grenoble-em.com
894 volume 30 number 9 September 2012 nature biotechnology
should be about applying a body of knowledge
in real-life situations.
Real cases as teaching tools
The Chinese philosopher Confucius stated,
“I hear and I forget; I see and I remember; I
do and I understand.” This is the idea behind
case-based learning, where examples from real
businesses are used as part of the educational
experience. The cases used can be either writ-
ten case studies or ‘live’ cases involving a bio-
tech company. Case-based learning provides
students with a global perspective on the chal-
lenges they face in their future employment.
The projects allow students to apply classroom
learning and their own experiences to solve
problems where situations are in most cases
both dynamic and highly unstructured.
Written case-based learning lends itself to the
teaching of a variety of courses: strategic man-
agement, biotech entrepreneurship and business
development. The technique can vary widely
between these courses in terms of instruction,
but the end goal is the same—students applying
their knowledge to solve a real-life problem—
albeit in these cases a problem that someone
else has dealt with before them. These courses
are typically taught by experienced industry
professionals who lend their own perspective
and expertise. The discussion of each case is
shaped both by the instructor and by the per-
sonal experiences and perspectives of the stu-
dents. The goal is to immerse the students in
the learning situation, allowing them to learn
not only from the instructor but also from their
peers. The role of the instructor is to stimulate
the imagination of the students and challenge
their current way of thinking; at the same time
the students gain knowledge and apply concepts
directly to a problem.
The written case studies used for teaching
are usually summaries of a business situation
or challenge faced by a particular person or
company in the recent past. These studies,
although useful for giving students practice
in applying concepts, have their limitations.
As the studies are summaries, only the infor-
mation deemed relevant by the author is
included in the write-up, which can affect the
interpretation of the situation. Although the
facts are typically plainly stated, some infor-
mation may be missing. Therefore, students
cannot be aware of all the factors that caused
the protagonist to take the particular course of
action described in the case.
However, although the outcome of these
written case studies is already known, which
can be an impediment to solving the case
with a fresh perspective, it is not the issue at
the heart of this learning experience. Rather,
the point is to learn to properly analyze the
context of a case, apply the concepts learned
in class and arrive at a solution with the infor-
mation provided. There is no right or wrong
answer. Frequently it is through the discus-
sion of the written case with the instructor
and their peers that students learn. They learn
not only how to apply business concepts but
also how to set aside their own prejudices to
solve a real business problem. The final deci-
sion made by the company described in the
study is not important, nor is its final outcome
in certain cases. What is important is the
process the students go through to determine
what they would do if they found themselves
facing a similar type of problem in their com-
panies.
Case-based learning is not restricted to writ-
ten case studies, but can also incorporate real-
world cases. In this type of learning, students
are separated into groups and spend months
working on a real-life innovation project,
which is typically an issue faced by an actual
biotech company where students take on the

role of consultants to help find a solution to the
particular issue. In addition to learning about
a particular scientific issue addressed by the
company, students learn how to work as part of
a team, manage conflict and people, and solve
problems as they arise. The goal is to render
students independent, able to apply concepts
learned in class and search for information on
their own, as they will have to do in their future
employment.
Keeping in mind that most of the projects
that serve as the live-case part of the curricu-
lum are solicited from start-ups, students learn
how to manage a project with limited resources
at their disposal. In addition, the projects are
at various stages of development, and most
are unstructured. Students are called upon to
work with scientists who frequently have no
idea what potential their discovery has in the
commercial market, or how to shape a product
and bring it to market.
© 2012 Nature America, Inc. All rights reserved.
What students learn quickly in their innova-
tion project is that dealing with real-life situ-
ations is neither easy nor simple. Whereas the
projects are intended to integrate academic and
technical coursework, the learning experience
goes beyond meeting these objectives. The stu-
dents learn how to function as part of a team,
how to structure and solve problems and how
to be flexible. Biotech is a constantly evolving
sector, and through the innovation project stu-
dents get first-hand experience of its volatility.
Learning the language of business
Working as part of a team, where each person
contributes to the whole, may be difficult for
scientists who tend to work independently.
Students pursuing the business aspects of sci-
ence learn to evaluate projects and solve prob-
npg
lems differently from pure scientists because
for them science is only one half of the equa-
tion for success—the other is commercial via-
bility. Business is a whole new language that
these students must learn in order to function
efficiently and bridge the communication gap
between science and business.
Students learn how to interact with the com-
pany’s management team. Often, these manag-
ers are scientists with no business background,
and the students can help to develop the proj-
ect’s commercial market potential. On other
occasions, management may consist of busi-
ness managers who have limited understand-
ing of the science and therefore of the potential
of their product. Here, the utility of dual skills
in business and science is a benefit.
The tasks the students are asked to perform
in the context of the innovation project are
varied as no two projects are the same in their
nature biotechnology volume 30 number 9 september 2012 895
careers and recruitment
context and requirements. Students may be
required to tackle studies on market research,
feasibility, intellectual property or finance.
By conducting market research, the stu-
dents try to answer the central question asked
by a company: how to improve current market
offerings through their product. To accom-
plish this, students must employ their com-
petitive intelligence skills; that is, analyze the
current market offerings and identify needs
that are not currently addressed, and thus
determine how the new product will address
these needs. Another component of competi-
tive intelligence is to identify current as well as
putative future competitors. In addition, they
must identify and contact clients and experts
to discuss the putative product and how it will
be able to address market needs. Through this
information, they can help the company define
their product in a way that corresponds to the
needs of the market. Although this task is by no
Students pursuing the business
aspects of science learn to
evaluate projects and solve
problems differently from pure
scientists because for them
science is only one half of the
equation for success—the other
is commercial viability.
means simple, it can be further complicated by
confidentiality agreements the students must
perform under, so as not to unintentionally
give information to competitors, especially if
the company is working in a small niche area.
Only by understanding both the science of the
technology and the business aspects can the
students succeed.
In the process of assessing the putative prod-
uct, the students may have to identify putative
partners for the company. Establishing part-
nerships is one way for a start-up to share risk
and acquire the necessary competencies to
bring their product to market. In the case of
established partnerships or alliances, the stu-
dents can undertake cost and feasibility stud-
ies for the project. This necessitates contact
with the partner companies to determine the
timeline in which they expect certain tasks to
be completed as well as the financial burden,
both in terms of people and materials. This is
an interesting endeavor where the interplay
between the partners can be observed—which
tasks can be done concurrently and which
tasks must await the completion of another,
aspects that are essential to project manage-
ment. The feasibility of the whole project as
well as potential issues with the technology
are usually observed here. If the information
obtained is incomplete or lacking, it can pose
difficulties in establishing the final cost, and
therefore the financial feasibility of the project.
In the end, all the students can do is provide
recommendations based on their knowledge,
information provided and their findings. It is
then up to the management of the company to
decide whether to accept or reject these recom-
mendations.
Depending on the stage of the project, the
students may work with the company to raise
money by approaching venture capitalists or
partners. This can give students a close-up view
of the difficulties encountered by companies
when raising capital. Whereas academics live
with the principle “publish or perish,” in indus-
try the prevailing principle is “secure funding
or die.” The ability to convince investors of the
utility and necessity of the product is vital for
the entrepreneur.
Conclusions
Real life–based projects serve to challenge stu-
dents in their thinking and capabilities. Not
only do they learn how to work as part of a
team, where everyone must do their part in a
timely fashion, but also how to manage people.
The ultimate fate of a company lies solely in the
hands of its management team, and through
this type of project students can have a signifi-
cant role in the success or failure of a product.
Such projects are dynamic and can change at
any time based on new information. Through
the real-world case study, students learn to be
flexible and adaptable to change. Students not
only learn by doing but also have the opportu-
nity to reflect and determine how to respond
better to a similar situation in the future.
Both written case studies and real-life
innovation projects have a role in educating
tomorrow’s biotech leaders. But although the
outcome of the written case studies may be
known and the author provides, in his or her
judgment, all the necessary information, in a
live case the data may be flawed or incomplete,
the parameters constantly shifting, the stakes
higher and the solutions unknown. Students
with dual skills in science and business can,
however, meet the challenge.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
1. Theodosiou, M., Rennard, J.P. & Amir-Aslani, A. Nat.
Biotechnol. 30, 367–368 (2012).
 people

Sophiris Bio (San Diego) has named Randall E. Woods (left) CEO.
Woods brings almost 40 years of industry experience to Sophiris,
including past roles as CEO at Sequel Pharmaceuticals, NovaCardia
and Corvas International, and president of Boehringer Mannheim’s
US pharmaceutical operations. He currently serves on the board of
Arena Pharmaceuticals and is chairman of the board for Sorbent
Therapeutics.
“Randy’s blend of executive experience in biotech research and
development, pharmaceutical sales and marketing, coupled with
successful fundraising and value creation makes him an ideal fit for leading Sophiris into
the next phase of growth, beginning with advancing PRX302 into pivotal trials later this
year,” says Lars Ekman, executive chairman and president of Sophiris.
© 2012 Nature America, Inc. All rights reserved.
position of vice president, clinical development
at Cytori Therapeutics.
Epistem (Manchester, UK) co-founder Chris
Potten passed away on August 6 at the age of
71. One of the world’s most influential figures
in epithelial biology, Potten pioneered adult
stem cell research in this area beginning in
the early 1970s, leading a research team at the
Paterson Institute For Cancer Research at the
Christie Hospital, Manchester, until retiring in
2000. He was recognized for his work with a Life
Fellowship from the Cancer Research Campaign

(now Cancer Research UK), in addition to the

Pharmacyclics (Sunnyvale, CA, USA) has
announced the appointment of Joshua T.
Brumm as executive vice president of finance.
He was most recently CFO and senior vice
president for ZELTIQ Aesthetics. Before join-
ing ZELTIQ, Brumm served as director of
finance at Proteolix.
Dirk J. Evers has been named senior vice
president, informatics and Kevin V. Shianna
senior vice president, sequencing operations
at New York Genome Center (New York).
Evers joins NYGC from Illumina, where
he led the company’s computational biol-
ogy efforts in the UK. Shianna was most
recently an assistant professor at the School
of Medicine, director of operations for the
npg
Nile Therapeutics (San Mateo, CA, USA) has
promoted Darlene Horton to president and
CEO, and appointed her as a director of the
company. She previously served as Nile’s chief
medical officer after joining the company from
Itero Biopharmaceuticals.
Christine Larson has been named vice presi-
dent and CFO, a newly created position, at
Dynavax Technologies (Berkeley, CA, USA).
She brings nearly 25 years of experience as a
financial professional, most recently leading
financial and strategic planning, accounting,
and treasury and investor relations for PDL
Biopharma.
Ivana Magovcevic-Liebisch has been appointed
International Marie Curie and Weiss medals for
services to radiation research. He is survived by
his second wife, Carol, and three sons.
Seth M. Shaw has been appointed CEO
and chairman of the board of directors of
Immunovative (Montreal, Quebec, Canada),
replacing Antonio Treminio. Shaw is the
founder of Novastar Resources and served as
the founding CFO of Physician Therapeutics
in 2004.
ImmunoCellular Therapeutics (Woodland
Hills, CA, USA) president and CEO Manish
Singh has resigned, with company founder,
CSO and chairman John S. Yu taking on the
roles on an interim basis until a permanent suc-

Center for Human Genome Variation and
founding director for the Genomic Analysis
Facility at Duke University.
DARA BioSciences (Raleigh, NC, USA) has
appointed Timothy J. Heady to its board of
directors. Heady retired in 2011 as CEO of
UnitedHealthcare Pharmacy. Prior to join-
ing UnitedHealthcare in 2001, he was a senior
executive with Searle/Pharmacia where he
managed the customer operations unit.
Neuren Pharmaceuticals (Sydney) has
announced that Joseph Horrigan has joined
the company as vice president of clinical devel-
opment and medical affairs. Horrigan most
recently served as assistant vice president and
head of medical research at Autism Speaks,
the autism science and advocacy organiza-
tion. Previously, he coordinated pediatric drug
development in the neurosciences medicines
development center at GlaxoSmithKline.
to the newly created position of executive vice
president, COO of Dyax (Burlington, MA,
USA). She joined the company in 2001 and has
previously served as vice president of intellec-
tual property, executive vice president and gen-
eral counsel, and most recently executive vice
president and chief business officer. Dyax also
announced the promotion of Burt Adelman,
chief medical officer, to executive vice president.
Endocyte (West Lafayette, IN, USA) has named
David Meek to the newly created position of
chief commercial officer. Meek brings more
than 20 years of pharmaceutical industry expe-
rience, coming to Endocyte from Novartis
Pharma where he was most recently the head of
northern and central Europe oncology.
Cardio3 BioSciences (Mont-Saint-Guibert,
Belgium) has announced the appointment of
Alexander Milstein as vice president of clini-
cal development. Previously, Milstein held the
cessor is selected. Singh was appointed president
and CEO in February 2008.
ADVENTRX Pharmaceuticals (San Diego)
has named Santosh J. Vetticaden chief medi-
cal officer and senior vice president. Vetticaden
has more than 25 years of experience in drug
development and medicine, most recently as
senior vice president and chief medical and
development officer at Cubist Pharmaceuticals.
Before joining Cubist, he served in the same
capability at Maxygen.
Coronado Biosciences (Burlington, MA,
USA) has announced the election of Harlan F.
Weisman to its board of directors. Weisman
brings over 20 years of experience as a senior
healthcare executive. He was previously chief
science and technology officer for the Johnson &
Johnson (J&J) medical devices and diagnostics
group as well as former chairman, pharmaceu-
tical R&D at J&J.

896 volume 30 number 9 September 2012 nature biotechnology