EDITORIAL OFFICE London NatureReviews@nature.com The Macmillan Building, 4 Crinan Street, London N1 9XW, UK Tel: +44 (0)20 7843 3620; Fax: +44 (0)20 7843 3629

To subscribe and for more detailed information visit www.nature.com/reviews/drugdisc

chief editor: Peter KirkpatrickSenior editorS: Alexandra Flemming, Charlotte Harrison, Sarah Crunkhorn, Monica Hoyos FlightSenior newS editor: Asher MullardASSiStAnt editor: Man Tsuey Tsecopy editor: Mariam FaruqiSenior copy editorS: Catriona Rodwell, Lucie Woottoncopy editing MAnAger: Lewis PackwoodArt controLLer: Susanne HarrisSenior Art editorS: Vicky Summersby, Patrick Morgan, Kirsten Lee

MAnAging production editor: Judith ShadwellSenior production editor: Simon Fenwickproduction controLLer: Natalie SmithSenior editoriAL ASSiStAnt: Laura CornseditoriAL ASSiStAnt: Ella Linesweb production MAnAger: Deborah AnthonyMArketing MAnAgerS: Tim Redding, Nazly De La RosapubLiShing director: Peter Collins

new york nature@natureny.com Nature Publishing Group, 75 Varick Street, 9th floor, New York, NY 10013–1917, USA Tel: +1 212 726 9200; Fax: +1 212 696 9006

pubLiSher (biophArMA): Melanie Brazil

cuStoMer ServiceS: Feedback@nature.com

Copyright © 2012 Nature Publishing GroupResearch Highlight images courtesy of Getty Images unless otherwise credited.Printed in Wales by Cambrian Printers on acid-free paper.

peter kirkpAtrick ALexAndrA fLeMMing chArLotte hArriSon

MAn tSuey tSeASher MuLLArdSArAh crunkhorn

EDITORS

The number of new drugs approved per billion US dollars invested

in research and development (R&D) has declined progressively

in the past 60 years, despite improvements in scientific and

technological inputs into the R&D process. Given the apparent

lack of impact so far of many solutions to this decline in R&D efficiency,

Scannell and colleagues question whether the underlying problems have

been correctly diagnosed and discuss factors they consider to be responsible,

with the aim of stimulating further systematic analysis. It is anticipated that

the use of biomarkers to match the right drug to the right patient is one

strategy that could reduce the size, failure rates and cost of clinical trials.

In their Perspective article, Kelloff and Sigman highlight the biomarkers

expressed during cancer development and progression, focusing on those

that are most relevant for identifying patients who are likely to respond to

a specific therapy, and those that are most effective for measuring patient

response. The design of biomarker-based cancer clinical trials and the

associated challenges are considered. Accumulating evidence suggests

that indicators of immune system activity in cancer patients can also

be of prognostic value or be used to predict treatment response. This is

discussed by Galluzzi and colleagues, who review the cellular and molecular

mechanisms by which current anticancer agents can activate the immune

system against cancer, and their therapeutic implications. Inappropriate

immune system activation has a key role in chronic inflammatory disorders,

a common feature of which is bone loss. Redlich and Smolen review the

mechanisms mediating bone loss in such disorders, and discuss current and

emerging counteractive therapeutic strategies.

Cancer biomarkers p201

Efficiency of drug R&D p191

GET

TYN

PG

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 169

IN THIS ISSUE

© 2012 Macmillan Publishers Limited. All rights reserved

Advancing the development of tuberculosis therapyAlimuddin Zumla, Richard Hafner, Christian Lienhardt, Michael Hoelscher and Andrew Nunn

Although the development of novel drugs and combination regimens for tuberculosis has accelerated in recent years, the pipeline remains thin and major challenges remain to be addressed in efficiently evaluating newer drugs to improve treatment outcomes, shorten duration of therapy and tackle drug resistance.

Alimuddin Zumla is at the Division of Infection and Immunity, University College London, London NW3 2PF, UK. Richard Hafner is at the National Institute for Allergy and Infectious Diseases, US National Institutes of Health, Bethesda, Maryland 20892, USA. Christian Lienhardt is at the Stop TB Partnership, 20 avenue Appia 1211, World Health Organization, Geneva 27, Switzerland. Michael Hoelscher is at the Division of Infectious Diseases and Tropical Medicine, University of Munich, Leopoldstraße 5, 80802 Munich, Germany. Andrew Nunn is at the Medical Research Council, 222 Euston Road, London NW1 2DA, UK. Correspondence to A.Z. e-mail: a.zumla@ucl.ac.uk doi:10.1038/nrd3694

Tuberculosis (TB) remains one of the most important causes of death from an infectious disease; in 2010 there were 8.8 million incident cases of TB and 1.45 million deaths. Furthermore, multidrug-resistant (MDR)-TB is spreading and poses a major threat to progress in global TB control. Only 1% of patients with MDR-TB are estimated to be on appropriate drug treatment and they have poor treatment outcomes1, highlighting an urgent need for new TB drug regimens that are shorter, more effective and have less toxicity. However, no randomized Phase III trials have been completed in MDR-TB so far, and the implementation of the cur-rent World Health Organization guidelines2, which recommend that patients should be treated for at least 20 months, has resulted in a range of treatment regi-mens that are dependent on laboratory infrastructure for drug susceptibility testing (DST), physician prefer-ence, drug availability and cost. Such individualized treatment approaches have not shown a significant benefit in terms of outcome compared with standardized treatments3.

TB therapy pipelineThere is hope that unmet needs for both drug-susceptible TB and MDR-TB are beginning to be addressed by the acceleration in TB drug discovery, development and evaluation in the past decade, particularly for regi-mens to shorten the duration of treatment and reduce the likelihood of the development of resistance, but the drug pipeline remains thin. At present, ten new or repurposed TB drugs are in clinical trials4. This includes two Phase III trials investigating whether the treatment of drug-susceptible TB can be shortened from the standard 6 months to 4 months by substitution of the repurposed drugs gatifloxacin or moxifloxacin for ethambutol or isoniazid; results are expected in 2013–2014. Another Phase III treatment-shortening

trial in progress is using twice-weekly rifapentine with moxifloxacin during the continuation phase. Phase IIb studies will evaluate the replacement of rifampicin with high-dose rifapentine in the standard first-line regimen.

Two new drugs — bedaquiline (TMC-207), which targets ATP synthesis, and delamanid (OPC-67683), a nitroimidazole derivative that causes intracellular release of lethal reactive nitrogen species — have been tested in Phase II trials in newly diagnosed MDR-TB patients. In both trials, either the investigational drug or a placebo was added to a background regimen, and on the basis of the promising results delamanid is now in a Phase III trial; it is expected bedaquiline will follow shortly. Other compounds in Phase II trials include: the nitroimida-zole derivative PA-824; the repurposed oxazolidinone linezolid, which is being tested for extensively drug-resistant TB (XDR-TB); the novel linezolid analogues PNU-100480 and AZD5847; and SQ109, which is related to ethambutol.

Near-term challenges for drug evaluationFirst, adequate safety evaluations for serious or rare tox-icities for new TB drugs are crucial. Although bedaqui-line and the nitroimidazoles seem promising, the small number and duration of exposures so far means that there is a risk of uncommon, serious toxicities emerging after wider exposure. More new drug candidates are required to raise the probability of developing a novel, short-course and safe ‘universal’ regimen applicable to drug-susceptible TB and all forms of drug-resistant TB. Combining TB drugs — for example, bedaquiline and moxifloxacin — may result in additive cardiac toxicity and requires careful investigation. Drug interactions with bedaquiline, especially with rifamycins, may limit the number of its combinations. New oxazolidinones show promise, but evidence of improved tolerability is

COMMENT

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 171

© 2012 Macmillan Publishers Limited. All rights reserved

missing. Long-term follow-up after marketing is essen-tial, particularly for drugs with very long half-lives and prolonged tissue concentrations.

Second, 14-day early bactericidal activity trials do not necessarily provide evidence of activity, as has been demonstrated with bedaquiline and, historically, with pyrazinamide. They therefore cannot be used to make decisions concerning drug potential and dose establish-ment. Correlation between early bactericidal activity against rapid replicators, and sterilizing activity against slow or non-replicating organisms, is limited for some drugs. Optimal statistical methods are required to quan-tify and compare the relative activity of combinations over time. It is also not known whether any early-phase trial, as currently designed, is likely to give results that are predictive of long-term outcome.

Longer-term needs First, although several companies have committed to develop new TB drugs for little or no returns, some remain reluctant to allow independent evaluation of their drug — for example, by not-for-profit sponsors. This is potentially hindering quicker selection of the optimal combinations for trials, and so greater support for such evaluations from companies is needed.

Second, with increased drug development activity, and given the limited resources, coordination of the DST testing efforts, methodology and database elements is essential for success. Consensus databases are needed for results of DST, genomic data of mutations, phenotypic/culture-based testing, minimum inhibitory concentra-tions and clinical correlations. Greater proficiency is needed in the preclinical testing of drug combinations for the prevention of resistance, as the standard mouse model is probably inadequate.

Third, a greater emphasis on research into the basic biology of non-replicating Mycobacterium tuberculosis — whose persistence and reactivation is crucial in extending the length of current treatment regimens — is needed to develop drugs that target it specifically. The mechanisms of antibiotic tolerance and survival of the bacteria in macrophages remain to be defined, and developing a means of inhibition requires investigation. New ways of using old and new drugs, such as inhala-tion, nanoformulations and the most effective sequenc-ing of drugs that are active against different bacterial subpopulations, also require urgent study. Indeed, we still do not know the optimal doses for rifamycins and quinolones.

Fourth, new approaches for the design of clinical trials and an increased capacity to implement trials to international standards are essential. In general, trials of TB drugs are costly and lengthy because of the extended follow-up time after completing treatment to monitor TB recurrence and longer-term toxicities. The need for multiple trial sites to assess regional variation also adds to the complexities and costs. Furthermore, the current recommended TB treatment regimens have very high success rates for drug-susceptible TB. Such trials utilize non-inferiority designs, which require large numbers of patients who are followed up for at least 12 months

after the completion of treatment. This explains the slow pace and long duration of current trials such as REMox, a three-arm Phase III study in which moxifloxacin is being substituted for two different drugs in the current first-line standard TB therapy, ethambutol and isoniazid, with the treatment duration in the moxifloxacin arms shortened to 4 months compared with 6 months in the standard therapy arm. The trial, which is now in its sixth year, has just completed enrolment and it will be up to another 2 years before we will know the cure rates and the proportion of patients who experience bacteriologi-cal treatment failure or relapse. Alternative clinical trial designs to reduce the number of patients required to identify the shortest, most effective and safest regimens are being developed. A relatively new concept — the adaptive multi-arm multistage (MAMS) trial design5, which has been successfully used in cancer — is under discussion.

Finally, novel and reliable TB-specific biomarkers are needed for the prediction of relapse, sterilizing activity and treatment response to facilitate the testing of new drugs in specific combinations. Biomarkers that allow reliable and nearly real-time assessment of early treat-ment response are particularly needed to take full advan-tage of MAMS-type clinical trial designs, as waiting at least 3 weeks for results from culture-based response tests substantially limits the effectiveness of adaptive designs.

Concluding remarksDespite some notable progress in TB drug development and evaluation, much more is needed to meet the huge challenges presented by the emergence of all forms of drug-resistant TB as well as the convergence of the TB and HIV epidemics. New drugs in novel combinations need more efficient evaluation for safety, efficacy and shorten-ing treatment duration. New biomarkers are required to improve the efficiency of Phase II and III trials using adaptive designs. Coordination and collaboration among drug developers, research funders, national governments and policy makers is essential. In this era of global eco-nomic recession, a major leadership opportunity is arising for countries that have a high prevalence of TB but whose economies are growing, such as China and India.

1. World Health Organization. Global tuberculosis control 2011. WHO website [online], http://www.who.int/tb/publications/global_report/en/index.html (WHO, Geneva, Switzerland, 2011).

2. World Health Organization. Guidelines for the programmatic management of drug-resistant tuberculosis: 2011 update. WHO website [online], http://whqlibdoc.who.int/publications/2011/9789241501583_eng.pdf (WHO, Geneva, Switzerland, 2012).

3. Orenstein, E. W. et al. Treatment outcomes among patients with multidrug-resistant tuberculosis: systematic review and meta-analysis. Lancet Infect. Dis. 9,153–161 (2009).

4. STOP TB Partnership. Working group on new TB drugs discovery portfolio. New TB Drugs website [online], http://www.newtbdrugs.org/pipeline.php (2011).

5. Phillips, P. et al. Innovative trial designs are practical solutions for improving the treatment of tuberculosis. J. Infect. Dis. (in the press).

DisclaimerThe opinions expressed in this article reflect the personal opinions of the authors and do not necessarily reflect the opinions of their institutions or their funders.

Competing financial interestsThe authors declare no competing financial interests.

172 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

C O M M E N T

© 2012 Macmillan Publishers Limited. All rights reserved

Asher Mullard

Protein–protein interaction (PPI) targets have traditionally been shunned by many small-molecule drug developers, despite their therapeutic relevance and untapped abundance, largely because of technological hurdles. A slate of deals, milestone payments and scientific advances, however, suggest that that these challenges are becoming less daunting.

In January, for instance, Boehringer Ingelheim became one of the latest firms to sign up with Forma Therapeutics for access to their approach to blocking PPIs implicated in cancer. Bristol-Myers Squibb, around the same time, extended a drug discovery programme started in 2009 with Ensemble Therapeutics that aims at discovering small-molecule inhibitors of up to eight PPI targets. And last year Shinogi also extended a 1-year-old partnership with Evotec to use fragment-based drug discovery to identify PPI inhibitors.

Increased enthusiasm for PPI inhibitors is not restricted to preclinical programmes though, and companies are advancing a handful of such compounds through clinical development. SARcode, a spin-out of Sunesis Pharmaceuticals, pushed its SAR1118 into Phase III trials for dry eye last year. Top-line pivotal results are expected within months, making it one of the most advanced small-molecule PPI inhibitors. The firm hopes that the drug will reduce T cell-mediated inflammation by blocking the interaction between

intercellular cell adhesion molecule 1 and lymphocyte function-associated antigen 1.

Other agents that are working their way through the clinic include Abbott/Genentech’s navitoclax (formerly ABT263) and Teva’s obatoclax (formerly CEP-41601 and GX-015-070), both of which are in Phase II development as anticancer agents that inhibit the function of the prosurvival BCL-2 family proteins by blocking key PPIs. Roche is testing another two anticancer agents that block the interaction between tumour suppressor p53 and MDM2: the firm plans to move RG7112 into Phase Ib trials later this year, and started Phase I trials of a backup compound, RO5503781, in December.

The emergence of such programmes has been driven by technological and conceptual advances in understanding PPI druggability and the types of molecules that can be used to block interactions (including peptides; see BOX 1), although the hurdles remain substantial compared to other target classes. Nevertheless, Jim Wells, a professor at the University of California San Francisco, USA, and founder of Sunesis, says “we are open to the challenges of inhibiting PPIs because there is no ignoring the fact that the biology here is a gold mine”.

Understanding PPI druggabilityFor most established target classes, inhibitors often bind well-defined

Protein–protein interaction inhibitors get into the grooveDrug developers are getting closer to tapping an unmined gold reserve of protein–protein interaction targets.

▶

A small-molecule macrocycle antagonist (blue) docked into the BCL-XL protein, an anti-apoptotic molecule implicated in the survival of cancer cells. Drug-like small molecules such as macrocycles can bind to BCL-XL, disrupt the protein–protein interaction and accelerate cancer cell death. Image courtesy of Ensemble Therapeutics.

NEWS & ANALYSIS

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 173

© 2012 Macmillan Publishers Limited. All rights reserved

pockets that fit endogenous small molecules, such as ATP in the case of kinases. The traditional aversion to PPIs as targets stemmed, by extension, from the belief that the interfaces between two proteins were too large and featureless for small molecules to be able to act effectively.

But this view has evolved, says Wells. Researchers began challenging the belief in the 1990s with the realization that PPIs have hotspots — small subsets of residues that contribute most of the free energy of binding to both natural partners and small molecules. A second advance, adds Wells, was the recognition that interaction interfaces are dynamic and can be more convoluted in solution than they appear in co-crystal structures. “There are a lot more opportunities for binding than we recognized,” he says.

Mapping methodologies and computational analyses now increasingly shed insight into which targets offer the most appealing hotspots. And protein crystallography has helped chemists, biophysicists and structural biologists to work together to optimize any hits that have cropped up. Because each of these approaches has its limitations, researchers have worked hard on developing experimental tests of druggability as well, says David Fry, a research scientist at Roche. For example, fragment-based screens — which typically use biophysical approaches to detect the weak binding of molecules that are much smaller (Mr = ~150–250) than typical drug candidates — can indicate which targets are viable, because the hit rate correlates with druggability. By spending substantial effort early on to identify the most tractable targets, drug developers have started to make headway (even if one trade-off is that in-depth biological validation of targets tends to come a little later in this space).

Through such starts and stops, drug developers have slowly started to expand the chemical space spanned by their screening libraries. And having initially thought that PPIs are out of bounds, the variety and variability of approachable targets has in fact raised a new challenge: the PPI interfaces and promising small-molecule inhibitors all seem so different from one another that each project requires a fresh start. “It’s a double-edged sword,” says Fry. On one edge, because various chemotypes can bind effectively and because hits don’t need to exactly mimic a target’s natural PPI partner, researchers are realizing that there are many possibly effective scaffolds out there. On the other, the variability means that drug developers cannot fall back on go-to scaffolds that can be used every time as a starting point.

Wells agrees. “Its not like G protein- coupled receptors, where a benzodiazepine core can be generalized to a lot of different systems, or kinases, where a quinazoline works in a lot of cases,” he says. Instead, the field has to take a case-by-case approach.

Yet as more examples of druggable PPIs are reported, many hold out hope that patterns will emerge. “This mega-class of PPIs will get broken into subclasses,” says Fry. A first differentiator, he adds, could be the types of secondary structures that mediate the interactions. Many PPIs recognize an α-helix, for instance, and some groups are already therefore working on developing small-molecule helical mimetic scaffolds with the hope these could have broad utility against PPIs. “There is some progress there, but it’s still pretty early,” says Fry.

Expanding the libraryThe flip side of expanding druggability is to reconsider the kinds of molecules that can be drugs. Perhaps it is no surprise, therefore, that many of the leading PPI development stories involve compounds that traditionalists would view as non-drug-like molecules. Abbott/Genentech’s oral Phase II candidate navitoclax, for example, violates three of Lipinski’s ‘rule of five’ guidelines for oral-drug-likeness (it has a molecular mass of 975 Da, a cLogP of 12, and 11 hydrogen-bond acceptors).

“If you look at the reported PPI inhibitors, they tend to be larger in molecular weight, more three-dimensional… more rigid than would be expected for a ‘typical’ drug,” says Fry. The fact that typical corporate compound libraries have been skewed towards smaller and flatter molecules, he adds, probably underscores the low success rate to date in identifying PPI inhibitors from high-throughput screens. “But people are making efforts to correct this problem,” he adds. “Entities that have decided to go after PPI targets are actively trying to skew their compound libraries by adding more complicated molecules.”

Nick Terrett, CSO at Ensemble, has seen a similar trend from his biotech vantage point. “Most companies now agree that in order to

compete in the PPI area they have to go out and find new molecular areas to work in. This was not true 3–5 years ago,” he says.

As Roche and other pharmaceutical companies have accepted the need to expand the scope of their libraries — and medicinal, formulation and process chemists have started gearing up to the difficulties of working with more complex molecules — a handful of biotech firms have sprung up to meet the demand. Forma Therapeutics, which launched in 2008 to go after difficult oncology targets, has focussed in part on building up a library that could fare better at blocking PPIs. Its screening library, which is approaching one million compounds, includes over 150,000 molecules (with 2–5 stereocentres) that originated from ‘diversity-oriented synthesis’ approaches intended to produce molecules in novel regions of chemical space, as well as an additional library being created that has compounds that are based on protein-mapping and interface analyses. “Our libraries are ever evolving as our understanding of the targets grows,” says Forma’s CEO Steve Tregay.

In addition to the recent partnership with Boehringer Ingelheim, other firms are eager for access to Forma’s library and associated technologies; Forma has inked deals with Janssen Biotech (Johnson & Johnson), Genentech and Eisai in the past year and a half, to the cumulative tune of US$2.5 billion biodollars. Although the company is not exclusively committed to targeting PPIs — it also works on enzymes involved in tumour metabolism and epigenetic modifications — its plans to screen around 30 oncology targets a year may nevertheless open up broad inroads into the PPI cancer space.

Ensemble has taken another approach: a focus on small-molecule macrocycles. Such large, natural-product-like structures have traditionally been very labour-intensive to synthesize, but since launching, the biotech firm has assembled a library of over four million macrocycles. Its ‘DNA-programmed chemistry’ platform uses hybridizing DNA templates to bring reactants together and drive the generation of otherwise-difficult-to-synthesize large ring-like structures,

Box 1 | Peptides offer their own challenges

Biologics, in many ways, are more natural candidates for the inhibition of protein–protein interactions (PPIs). Not only are they more likely to resemble a natural PPI partner, but their large size offers more opportunity to block an interaction and hit various hotspots. Yet a key challenge in developing such therapeutic biologics is their poor cell permeability. Various companies — including Aileron, Compugen, Polyphor, PeptiDream and Bicycle Therapeutics — are nevertheless working on different methods of developing cell-permeable or potential PPI inhibitor peptides. “Peptides are likely to be useful tools for attacking PPIs,” says Roche’s David Fry.

N E W S & A N A LY S I S

174 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

which tend to have a molecular mass ranging from 500 Da to 1,000 Da. “Because these macrocycles are big, they can reach further across the protein interactions and access whatever features might be there,” says Terrett.

In addition to the partnership with Bristol-Myers Squibb, Pfizer has also been working with the firm since 2010 for access to the new chemical space. Michael Taylor, Ensemble’s CEO, adds that the biotech firm has screened roughly 25 PPI targets in the past few years and has identified hits in roughly 50% of these projects. “We’ve been surprised with our ability to hit targets that we thought would be more difficult to hit based on the level of effort that has been spent on them by much larger players,” says Taylor.

The firm’s lead programme is a macrocycle that blocks the interaction between interleukin-17 (IL-17) and its receptor. Although the autoimmune candidate is still in preclinical testing, the programme marks a first publicly disclosed success in developing small molecules against a validated, disease-relevant IL-17–IL-17 receptor PPI. Ensemble hopes to have an oral candidate ready for pre-IND (investigational new drug) regulatory toxicology and related preclinical development later this year.

Beyond the willingness to expand into new chemical space, says Wells, chemists have taken a conceptual advance in regards to PPI drugs by becoming more willing to work with screening hits that have very low binding affinities. Rather than dismissing these as noise, drug hunters will now instead team up with biophysicists and structural biologists to think about how to best advance the weakly

binding compounds. Because a consequence of accepting these low binders is that PPI screens tend to report particularly high rates of false positives, researchers have developed follow-up assays to separate the wheat from the chaff. “We have put a lot of effort into that,” says Fry.

Biological gold mineThe motivation behind all this effort is straightforward: because established drug discovery targets such as G protein-coupled receptors and kinases are already well tapped and competition is fierce, new frontiers are needed. The burgeoning field of interactomics that is uncovering tens of thousands of new PPIs, combined with emerging success stories, suggests that PPIs could be one such fertile ground. Some of the PPI inhibitors that have been reported modulate the activity of targets that also have established active sites, such as kinases, leading some to hope that these may provide a fresh perspective — for instance, in the form of enhanced specificity — for known targets. More exciting, however, is the prospect of altogether new biological avenues of discovery.

One such recent case study centres on bromodomain-containing protein 4 (BRD4), an ‘epigenetic reader’ that recognizes acetylated histones. In late 2010, Bradner and colleagues reported that a small molecule could block the interaction between BRD4 and histones, providing proof-of-principle evidence that PPI inhibition could provide epigenetic control (Nature 468, 1067–1073; 2010). A BRD4–NUT translocation has been linked to an aggressive form of human

squamous carcinoma, and the team was able to generate in vivo data demonstrating the potential therapeutic applications of their compound in this setting. Further studies unveiled its potential in other indications, including MYC-associated cancers, and the compound is now being developed by Tensha Therapeutics. GlaxoSmithKline and others are also working preclinically to develop small-molecule inhibitors of the PPIs between bromodomain-containing proteins and acetylated histones.

And although many of the PPI programmes that have been embraced by industry fall in the oncology space, PPIs will in the long term be important across indications. Researchers recently reported in Nature that a first systematic affinity tagging–purification mass spectrometry analysis of interactions between human and HIV proteins uncovered 497 partnerships, potentially pointing to a range of new treatment pathways (Nature, 21 Dec 2011; doi:10.1038/nature10719). Forma, similarly, is comparing the hotspots of bacterial PPIs with those of human PPIs in the hopes of identifying new classes of anti-infectives. “Just about every protein interacts with another protein, so you would imagine that all therapeutic areas should have these kinds of targets,” says Fry.

That’s not to say that all the challenges of PPI inhibition have been overcome. Key among the field’s needs is more examples of PPI programmes to learn from, and advances in protein co-crystallization techniques to improve optimization approaches. But many are optimistic. “We are in a growth phase,” says Wells. “PPIs are definitely on the rise.”

N E W S & A N A LY S I S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 175

© 2012 Macmillan Publishers Limited. All rights reserved

Dan Jones

A little over 10 years ago, Rino Rappuoli, then a research scientist at Chiron, coined the term ‘reverse vaccinology’ (RV) to describe a way of developing vaccines driven by an explosion in genomics. RV is now poised to deliver on its early promise, with the European Medicines Agency currently reviewing the first RV‑derived vaccine: Novartis’s Bexsero. The vaccine could also become the first vaccine against meningococcus B (MenB), which causes more than 50% of meningococcal meningitis worldwide. A decision on the multicomponent vaccine for Novartis, which acquired Chiron in 2006, is expected imminently.

Traditionally, vaccines have been developed by isolating and purifying antigenic components from the pathogen of interest, which typically has been heat‑killed or chemically inactivated. RV, by contrast, starts with pathogenic genome sequences and then uses bioinformatics tools to predict potentially protective antigenic proteins encoded by the genome that can then be tested in vivo and in vitro. “RV opens up the opportunity to develop vaccines against virtually any type of pathogen,” says Yongqun He, a bioinformatician at the University of Michigan Medical School, Ann Arbour, USA. (The term RV can also be applied to viral vaccine development — where it describes the process of generating vaccines from the known crystallographic structure of broadly neutralizing monoclonal antibodies bound to

viral epitopes — but this approach is not discussed here).

Novartis is not alone in using RV. “It’s one approach that we take in developing vaccines,” says Jim Tartaglia, Head of New Vaccines at Sanofi Pasteur in North America. “RV is particularly useful when you’re working with large bacterial genomes and don’t have a clear lead on what proteins you would use in a vaccine formulation.” Sanofi has used the approach, for example, to develop a protein‑based vaccine for Streptococcus pneumoniae that is currently in Phase I/II trials, as well as other earlier‑stage projects.

The road to BexseroPrior to the advent of RV, MenB vaccines had been held up by two key challenges, says Rappuoli, now Global Head of Vaccines Research at Novartis Vaccines. First, the polysaccharide that formed the basis for effective vaccines in other meningococcal bacteria could not be used for Neisseria meningitidis (the causative agent of MenB) because it is too similar to human molecules and could therefore induce autoimmunity. Second, N. meningitidis strains are highly diverse, making it difficult to develop a universal vaccine with traditional approaches.

RV offered a way around these problems. First, it enabled Rappuoli and his team to identify new antigenic proteins for a vaccine that were not based on bacterial polysaccharides. An initial analysis of the N. meningitidis genome to predict surface‑exposed proteins that could be recognized by components of the immune system generated more than 600 potential antigens, of which roughly 350 could be expressed in Escherichia coli. The team further whittled down the candidate number by immunizing 350 groups of mice with these proteins, whereupon they found that 91 of the surface proteins induced antibodies in vivo. Of these, 29 induced antibodies that killed the bacteria in vitro.

The team was then able to examine the bacterial genomes from 31 MenB strains from around the world to check that the proteins they had selected for further study would provide broad protection — a step that has since become easier thanks to advances in sequencing and bioinformatics over the past decade. “When we started the Bexsero project we were only looking at one genome, but today it is usual to start with many genomes,” says Rappuoli.

Over the following years, the four most immunogenic and conserved antigens were selected and incorporated into Bexsero (formerly known as 4CMenB), which has undergone clinical trials in more than 7,500 infants, toddlers, adolescents and adults. Recently published results from the first Phase III trial of the vaccine, in adolescents, showed

Reverse vaccinology on the cuspAn upcoming decision for Novartis’s Bexsero — the first vaccine against meningococcus B — could substantiate reverse vaccinology.

RV opens up the opportunity to develop vaccines against virtually any type of pathogen.

Nature Reviews Drug Discovery | AOP, published online 10 February 2012; doi:10.1038/nrd3679

NEWS & ANALYSIS

NATURE REVIEWS | DRUG DISCOVERY ADVANCE ONLINE PUBLICATION | 1

© 2012 Macmillan Publishers Limited. All rights reserved

Vaxign, a web‑based platform that searches for antigens from over 70 pathogenic genomes and automates key bioinformatics analyses (J. Biomed. Biotechnol. 2010, 297505; 2010), may attract new researchers.

In addition to improving the in silico steps of RV, efforts are also being directed at streamlining the experimental testing of candidate vaccine ingredients — for example, by pooling antigens prior to immunizing mice (PLoS ONE 5, e11666; 2010).

As these and other hurdles are overcome — and with the possibility that the forerunner of the approach may soon be approved — interest in the broadly acknowledged utility of RV may increase. But Sanjay Gurunathan, Associate Vice President of Clinical Development at Sanofi Pasteur North America, notes that RV is unlikely to ever become the de facto starting point for creating new vaccines. “It’s one tool in our armamentarium, but that doesn’t mean that it’s going to be applicable to all vaccine development projects,” he says. For example, it is less useful when a pathogen’s antigens have already been well described, and when the goal is to optimize known antigens to confer protective immunity. “You have to decide whether to employ RV on a case‑by‑case basis.”

And even when it is used, RV is only ever an opening move. “Fishing out proteins from a genome is just one step — producing and formulating them are also important, and these processes may need to be tweaked in a pathogen‑dependent way,” says Gurunathan. “You have to look at the entire value chain going from gene to registration,” agrees Tartaglia.

that 92–97% of participants had protective serum bactericidal activity to test strains after one dose, versus 99–100% after two or three doses and 29–50% after placebo (Lancet, published online 18 Jan 2012). Another analysis suggested that the vaccine can protect against 77% of more than 800 genetically diverse disease‑causing MenB strains that have been isolated in Europe in recent years.

The future of RVFew RV‑driven projects have made it into the clinic so far, but the approach is nevertheless being applied to a range of pathogens. Novartis has used it to develop vaccines against group B streptococcus and S. pneumoniae, both of which are in Phase I development, and has a group A streptococcus vaccine in the works. A team from Imperial College London is using it to find antigens against Chlamydia pneumoniae, and researchers at the Israel Institute for Biological Research are using it develop a vaccine against Bacillus anthracis.

Yet despite the interest in RV’s potential, some say that the field has advanced only slowly. “There hasn’t been as much progress as I would have hoped,” says Francesco Filippini, a bioinformatician at the University of Padua, Italy. His main concern is that vaccinologists have not substantially improved the initial bioinformatics of candidate protein selection steps.

Better tools are needed for predicting which proteins will be antigenic, agrees Darren Flower, a bioinformatician at Aston University, UK. An empirically based approach, for instance, could draw on the field’s understanding of what kinds

of proteins are known to be antigenic. “We’re trying to pool our knowledge on the characteristics of antigens, and we’ve built a database of antigens to develop quantitative, predictive models of antigenicity that are similar to quantitative structure–activity relationship studies on small molecules.”

Filippini adds that the field would also be well served by moving beyond a focus on antigenicity — that is, how well a protein binds to antibodies or T cell receptors — and towards better ways of predicting which antigens will best induce protective immunity (not all antigens are equally immunogenic). “Immunogenicity has been difficult to predict on the basis of intrinsic properties of proteins, so we need experimental data on which antigens do and do not confer protective immunity [to improve our algorithms],” he says, noting that much of these data have been collected, but are held in commercial development programmes and so are not publicly available.

Another problem that has held up the field is that RV, and its associated sophisticated genomic analyses, have required niche bioinformatics‑intensive expertise. Early tools developed to perform these analyses, like the New Enhanced Reverse Vaccinology Environment (NERVE), failed to draw users because they were too computationally complicated. Newer systems, like He and colleagues’

There hasn’t been as much progress as I would have hoped.

N E W S & A N A LY S I S

2 | ADVANCE ONLINE PUBLICATION www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

PIX

TAL

NEWS IN BRIEF

FDA approves landmark drugsThe FDA started out the year with a busy few weeks, granting landmark approvals for Vertex’s ivacaftor, Genentech’s vismodegib and more.The lowdown: Following on the heels of a bumper crop of approvals in 2011 (Nature Rev. Drug Discov. 11, 91–94; 2012), the US Food and Drug Administration (FDA) started 2012 with a busy few weeks. As Nature Reviews Drug Discovery went to press, the Center for Drug Evaluation and Research had already approved five new molecular entities and one new biologic. The approval of Vertex’s ivacaftor, in particular, may mark a turning point for the treatment of cystic fibrosis; the drug is the first therapeutic to address the underlying cause of the disease (Nature Rev. Drug Discov. 10, 479–480; 2011). Another notable pioneering approval was granted for Genentech’s vismodegib, the first targeted treatment for basal cell carcinoma and the first drug to target the Hedgehog pathway. The agency also approved Pfizer’s axitinib as the fifth kinase inhibitor for the treatment of renal cell carcinoma.

The agency also approved a once-weekly formulation of Amylin’s exenatide, an injectable glucagon-like peptide 1 (GLP1) mimetic that was first approved in 2005 for the treatment of type 2 diabetes. Development of the new formulation has been watched closely by industry, both because of the size of the GLP1 market and because the agency has twice previously refused the therapeutic’s approval (for reasons including the need for more cardiovascular safety data).

The two Phase I/II trials, which will each enrol 12 patients, remain ongoing and are not due to complete until mid-2013.

The boost for stem cell therapy comes only months after Geron — the one-time leader in the field, that had enrolled at least four patients into its human ESC trial — announced its plans to leave the space altogether. Citing financial concerns, Geron has refocused its efforts on its oncology programmes and is looking to out-license its stem cell work.

Safety biomarker gets thumbs up

The FDA’s approval of a diagnostic to guide the usage of Biogen Idec’s natalizumab highlights the potential of biomarkers for addressing post-marketing safety concerns.The lowdown: Biogen Idec’s natalizumab was welcomed as a highly effective therapeutic option for the treatment of multiple sclerosis when it was approved in the United States in 2004, but its association with fatal cases of progressive multifocal leukoencephalopathy led to its withdrawal from the market in 2005. The monoclonal antibody was reintroduced in 2006, with various warnings, once its risks — which are linked to infection or reactivation of the JC virus — became better understood. Now, the FDA has also approved a test for anti-JC-virus antibodies to guide the use of the therapeutic, a move that some analysts predict could double sales of the drug.

Although uptake of the drug in light of the new diagnostic remains to be seen, the approval represents a promising case study for the value of biomarkers in managing drug safety issues that could otherwise scupper candidates. Among the few other similar such examples was the finding that patients with an HLA-B*5701 allele were predisposed to hypersensitivity reactions with ViiV Healthcare’s nucleoside analogue abacavir, leading the agency to recommend genetic screening before using the HIV drug.

well. For one, whereas the study focuses on understanding the role of gap junctions and 2APB in dose-dependent DILI (as induced by acetaminophen), most hepatotoxic drugs induce idiosyncratic DILI. The implications of the study for broader hepatotoxicity therefore remain unclear, he says. The co-formulation of approved or experimental candidates with a DILI inhibitor is also fraught with challenges. “It remains to be seen how applicable this work is,” he concludes.

Stem cells see first light

Human embryonic stem cells can be safely transplanted into patients, shows the first published human data of the new treatment modality.The lowdown: Thirteen years since the discovery of human embryonic stem cells (ESCs), work from Advanced Cell Technology (ACT) provides the first description of their transplantation into humans. As reported in The Lancet, two patients — from separate trials in Stargardt’s macular dystrophy and in dry age-related macular degeneration — were treated with 50,000 human ESC-derived retinal pigment epithelial cells and followed up for 4 months. The study investigators saw no signs of hyperproliferation, abnormal growth, immune-mediated transplant rejection or tumorigenicity during this time, providing evidence that human ESCs can be transplanted safely. A secondary end point suggested that the cells may have improved vision slightly.

A drug for drug-induced liver injury?

Small molecules could potentially reduce the risk of drug-induced liver injury, suggests a study.The lowdown: Hepatotoxicity is one of the most common reasons for abandoning drug candidates or withdrawing agents from the market. Building on emerging data suggesting that liver injury may spread via the gap junctions that connect individual liver cells, a team of investigators set out to examine whether gap junction inhibitors might act as ‘hepatoprotectants’. Reporting in Nature Biotechnology, they first identify the gap junction protein connexin 32 (CX32) as a mediator of drug-induced liver injury (DILI) and then show in mice that the CX32 inhibitory small molecule 2APB seems to reduce liver damage caused by high doses of acetaminophen, both when administered with or after the painkiller. On the basis of these findings, the authors suggest that 2APB or related hepatoprotectants may be useful for the treatment of acetaminophen overdose and for co-formulation with potentially hepatotoxic experimental drug candidates.

Neil Kaplowitz, professor of medicine at the Keck School of Medicine of the University of Southern California, USA, called the study — and the idea that liver injury might spread via intracellular connectivity pathways, in particular — “conceptually very provocative”. But he raised several caveats as

N E W S & A N A LY S I S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 177

© 2012 Macmillan Publishers Limited. All rights reserved

Phase II trial results for GlaxoSmithKline’s firategrast — an oral small-molecule α4β integrin antagonist — published in the Lancet (Lancet 11, 131–139; 2012) indicate that it reduces disease activity in patients with relapsing–remitting multiple sclerosis (RRMS).

Multiple sclerosis is a chronic autoimmune disease characterized by inflammatory demyelination of neurons in the central nervous system. Several disease-modifying agents are approved for RRMS treatment, but most must be injected and their use is limited by side effects, variable efficacy and safety concerns. “We are close to finding treatments that will reduce the inflammatory activity of multiple sclerosis very effectively, but at present these all come with significant risks. A key challenge is identifying ways to reduce these risks while retaining efficacy,” says Alasdair Coles, University of Cambridge, UK.

It is hoped that novel agents, which include firategrast, might help address this challenge (Nature Rev. Drug Discov. 10, 885–887; 2011). Firategrast targets α4β integrins — adhesion molecules expressed on most activated leukocytes — thus reducing inflammatory immune cell trafficking into the nervous system.

Although integrins represent promising therapeutic targets in several diseases, the clinical development of integrin inhibitors, particularly small-molecule drugs, has proved challenging. “In many integrin-dependent diseases the role of integrins is complex. Typically, multiple integrins are involved, making it difficult to identify a specific drug target,” explains Dermot Cox, Royal College of Surgeons, Ireland. “A second problem has been the development of pure antagonists. Previous experience with the platelet integrin αIIbβ3 showed that some oral antagonists intended to prevent platelet aggregation actually acted as agonists at the receptor, leading to increased events,” he adds. α4β integrins are among the few integrins for which pharmacological inhibitors have been successfully developed. “α4β integrins are the target of the monoclonal antibody natalizumab — the most efficacious licensed treatment of the inflammatory phase of multiple sclerosis to date,” notes Coles. However, natalizumab is administered intravenously and has been associated with progressive multifocal leukoencephalo pathy (PML) — a rare, serious brain infection.

The Phase II trial involved 343 RRMS patients who were randomized to receive oral firategrast twice daily (at doses ranging from 150 mg to 1,200 mg) or placebo, for 24 weeks, and were followed for 52 weeks. A statistically significant reduction (49%) in the cumulative number of new gadolinium-enhancing lesions was observed in patients receiving the highest firategrast doses compared to placebo. Firategrast was generally well tolerated, with no cases of PML reported.

The results are promising, although there are some caveats. First, concerning safety, “no case of PML was detected in this study, but that is not surprising as it only enrolled 343 patients, and the incidence of PML is around one in 1,000”, says Cox. “Also, the incidence is related to treatment duration, and in this case it is only 24 weeks,” he adds. With regard to efficacy, “it is encouraging that firategrast is clearly able to reduce cerebral inflammation, but it is disappointing that — even at the highest doses — firategrast did not seem to have the efficacy of natalizumab”, notes Coles. Indeed, natalizumab achieved a 90% decrease in new or active lesions in a comparable Phase II trial. “Perhaps future trials could explore higher doses,” he adds.

Sarah Crunkhorn

T R I A L WAT C H

Integrin antagonist shows promise in MS

N E W S & A N A LY S I S

BIOBUSINESS BRIEFS

178 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

a

Per

cent

age

of p

roje

cts

0102030405060708090

100

Preclinical Early clinical(Phase I–IIa)

Late clinical(Phase IIb–IV)

Mean25th–75th

percentile

b

Respiratory

Oncology

Neurology/CNS

Metabolic/endocrine

Immunology

Cardiovascular

Anti-infective

0 5 10 15

415

8

6

85

6

Number of times mentioned

Figure 1 | Role of personalized medicine approaches in industry: survey responses. a | Percentage of company pipeline relying on biomarker data. b | Number of times respondents mentioned personalized medicines for particular therapeutic areas (see Supplementary information S1 (box) for details). CNS, central nervous system.

Although the potential of personalized medicine based on genomic knowledge has been widely discussed for more than a decade, the pace of implementation has been less rapid than initially hoped. To better understand the current role of personalized medicine and pharmacogenomics in drug development, the Tufts Center for the Study of Drug Development conducted interviews and a follow-up survey of a group of companies about the current role of personalized medicine and pharmacogenomics in their company. Overall, information was collected from 21 companies (9 major biotechnology companies and 12 large pharmaceutical companies) between 2009 and 2010; see Supplementary information S1 (box) for details).

When asked what percentage of their company’s current clinical development pipeline comprised personalized medicines, responses ranged from 12% to 50%. Up to half of a company’s pipeline projects could have associated biomarkers, but substantially fewer projects (10% or less) had identified specific target populations or companion diagnostics. All companies noted that biomarker research is done for compounds before they enter clinical development. However, having a biomarker was not a requirement to move into clinical development for 70% of the companies; in addition, the follow-up survey indicated that across the companies about half of the compounds in early stages of development and a third in late stages use biomarker data in the development process (FIG. 1a). All of the companies explained that the primary intent of biomarker development was to provide more

M A R K E T WAT C H

Industry perspectives on personalized medicine

information about products internally, not for prescribing or monitoring a marketed product. For this reason, the few companies that require associated biomarkers for compounds in development also require personalized medicine end points for all trials, and use these end points for decision making.

These measures of usage across company pipelines varied considerably between therapeutic areas. As indicated in FIG. 1b, oncology was the most cited area in which the companies that were investigated were focusing their personalized medicine research and development (R&D) efforts. The responses indicated that R&D activity was also robust in immunology and neurology, as well as rapidly advancing in other therapeutic areas such as anti-infectives and metabolic/endocrine disorders.

Beyond the scientific problems, respondents considered that utilizing pharmacogenomic data was challenging because of a lack of

regulatory guidance. Although progress has been made — such as the release of draft guidance from the US Food and Drug Administration on pharmacogenomics in early-phase clinical studies — many companies felt that they were unable to use pharmacogenomic data in an approval package until pathways are better defined.

Overall, the respondents were generally optimistic about personalized medicine as a more promising path forward for the industry than the blockbuster model. This approach may enable the industry to overcome its productivity shortfall, so they believe, because targeted medicines are more likely to tip the scales of regulators’ benefit–risk decisions in favour of approval, as well as influence payers’ product value assessment more favourably towards reimbursement. Thus, respondents believed that the impact of personalized medicines on R&D generally, on company portfolios and in the marketplace will be strongly felt for the foreseeable future.

Rachael Zuckerman and Christopher-Paul Milne are at the Tufts Center for the Study of Drug Development,

Tufts University, 192 South Street, Suite 550, Boston, Massachusetts 02111, USA.

e-mails: zuckermanr@gmail.com; Christopher.Milne@tufts.edu

The authors declare no competing financial interests.

N E W S & A N A LY S I S

BIOBUSINESS BRIEFS

178 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

BRA

ND

X

The Supreme Court of Canada has denied a request by Merck that information related to regulatory filings of the asthma drug Singulair (montelukast) should remain confidential.

As part of the regulatory approval process for new drugs, innovator companies are required to disclose certain information — such as that related to ingredients, methods of manufacture, tests of potency, efficacy and safety, as well as details of possible contraindications and side effects — to the regulatory authority. This information is then reviewed by the regulator — the Canadian regulator is called Health Canada — which then releases a publicly available drug product monograph. Not all the information supplied to regulator will be in the monograph.

However, any interested party (and hence potential competitors) can request access to information held by Health Canada under the Canadian Access to Information Act (similar legislation exists in many other countries), which allows public disclosure of selected government information.

The Act has to strike a balance between the often competing objectives of increasing disclosure and protecting the interests of the parties who file the information with the government. So some forms of information are not released: for example, trade secrets; confidential financial, scientific or technical information; and information that would result in financial loss or gain, or would prejudice the competitive position of the party.

Merck asserted that except for the monograph, information related to regulatory filings should remain confidential and not disclosed to any requestors. This dispute between Merck and Health Canada about what information can be released to a requestor has been running for over 10 years, finally ending up at the Supreme Court in February 2012.

The Supreme Court first noted that trade secrets — such as the manufacturing processes of Singulair — should indeed remain undisclosed, before deciding whether or not the remaining

information should be classed as confidential. Merck asserted that even though many of the articles submitted in the regulatory filing were already in the public domain, it was the disclosure of the fact that such articles had been used in the regulatory submission that was important, and this fact should remain confidential. But the Court was not convinced, noting that knowledge of this is generally obtainable — albeit with more effort — by the public, adding that confidentiality should be assessed on a case-by-case basis.

Then the Court looked at the harm-based exemption. Merck said it would suffer harm, because disclosure of information would facilitate competitor drug development and give an incorrect impression of the safety of Singulair. But the Court held that although disclosure may facilitate competitor drug development, Merck had only been able to speculate that this was a possibility, and had not provided enough evidence that it would occur. And with regard to the disclosure of drug safety information, the Court said that “refusing to disclose [information] for fear of public misunderstanding would undermine the fundamental purpose of access to information legislation [which is] to give the public access to information so that they can evaluate it for themselves”.

This ruling does not mean that most regulatory submission data can now be disclosed; rather, there is no blanket rule that such information will not be disclosed. An innovator company must provide evidence that it will suffer harm or that information is confidential (on a case-by-case basis) to prevent the disclosure of such information. In a press release, legal representatives of BIOTECanada, who acted as an intervener in the case, said: “The decision makes the Canadian regulatory system… more consistent with international agreements, such as TRIPS [trade-related aspects of intellectual property rights] and NAFTA [North American Free Trade Agreement], as well as the legislation of other jurisdictions.”

Merck Frosst Canada versus Minister of Health: http://scc.lexum.org/en/2012/2012scc3/2012scc3.html

PATENT WATCH

Merck loses information disclosure dispute

Generic drugs allowed on market for non-patented indications

Several generics companies, including Mylan, Teva and Apotex, have won a court case that could allow them to market generic versions of AstraZeneca’s Crestor for two indications — homozygous familial hypercholesterolaemia and hypertriglyceridaemia — that are not covered by AstraZeneca’s methods of use patents.

In the case before the US Court of Appeals for the Federal Circuit (CAFC), AstraZeneca argued that even though familial hypercholesterolaemia and hypertriglyceridaemia were not covered in patents that describe methods of using

Crestor (US 6858618 and US 7030152), once on the market, generic rosuvastatin would infringe on these patents. Their reasoning was largely based on what they referred to as market realities, asserting that even if a generic drug is formally approved only for unpatented uses, once it is on the market pharmacists and doctors would substitute the generic for all indications.

But the CAFC found this argument unpersuasive, noting that if accepted, this argument would allow a pioneer drug developer to maintain de facto indefinite exclusivity of a pharmaceutical compound by obtaining serial patents for regulatory approved methods of using the compound.

But even though generic rosuvastatin has received approval from the US Food

and Drug Administration for homozygous familial hypercholesterolaemia and hypertriglyceridaemia, the generics companies cannot market their products just yet. They must first win the ongoing litigation that alleges that the patent that claims Crestor itself (US RE37314) — as opposed to the methods patents disputed in this case — is not valid.AstraZeneca versus Apotex et al. http://www.cafc.uscourts.gov/images/stories/opinions-orders/11-1182.pdf

Charlotte HarrisonPATENT ADVISORS

Daniel M. Becker: Dechert, Mountain View, CA, USA.Luke Kempton: Wragge & Co., London, UK.Leslie Meyer-Leon: IP Legal Strategies, Boston, MA, USA.George W. Schlich: Schlich & Co., London, UK. John A. Tessensohn: Shusaku Yamamoto, Osaka, Japan.Philip Webber: Dehns, London, UK.

N E W S & A N A LY S I S

180 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

FcγR

Immunecomplex

RANK

RANKL

FMS

M-CSF

SYK

TNFR

SRC

PI3K

Nature Reviews | Drug Discovery

Spleen tyrosine kinase inhibitors

Spleen tyrosine kinase (SYK) is involved in both inflammation and osteoclast activation, making it a promising target in inflamma-tion-mediated bone loss. Indeed, the SYK inhibitor fostamatinib is in late-stage clinical trials for rheumatoid arthritis.

Table 1 | Recent patent applications related to spleen tyrosine kinase

Patent numbers Asignee Subject

WO 2011086085 AB Science Substituted thiazole and oxazole kinase inhibitors that are potent and selective FLT3 inhibitors or SYK inhibitors

EP 2285366 US 2011112098

Centre national de la recherche scientifique

Molecules, such as the C-13 molecule, that are capable of inhibiting the binding of an antibody or antibody fragment with human SYK

WO 2011144584 WO 2011117160

Roche Novel pyrrolopyrazine derivatives that are useful for treating autoimmune and inflammatory diseases

WO 2011144585 Roche Pyrrolo[2,3-b]pyrazine-7-carboxamide derivatives that inhibit JAK and SYK; useful for treating autoimmune and inflammatory diseases

WO 2011117145 Roche Novel pyrrolopyrazine derivatives that inhibit JAK and SYK, and are useful for treating autoimmune and inflammatory diseases

WO 2011014515 Genomics Institute of the Novartis Research Foundation

2,7-naphthyridin-1-one derivatives that act as SYK inhibitors

WO 2011014795 US 2011053897

Genomics Institute of the Novartis Research Foundation

Compounds that can be used to treat or prevent disorders associated with abnormal SYK activity, such as inflammatory diseases, allergic diseases or cell-proliferative diseases

WO 2011112995 EP 2373318 EP 2373169

Gilead Certain imidazopyridines that are useful for treating disorders that are responsive to SYK inhibition, such as B cell lymphoma, leukaemia, rheumatoid arthritis or COPD

US 2011275655 EP 2376481

GlaxoSmithKline Pyrimidine carboxamide derivatives that act as SYK inhibitors and are useful for treating diseases that result from inappropriate mast cell activation

WO 2011134971 GlaxoSmithKline 7-(1H-pyrazol-4-yl)-1,6-naphthyridine compounds that act as SYK inhibitors; useful for treating diseases resulting from the inappropriate activation of mast and/or basophil cells, macrophages and B cells, as well as related inflammatory responses and tissue damage

WO 2011075515 US 2011075517 US 2011075560

Merck Novel aminopyrimidines that are potent SYK inhibitors and are useful in the treatment and prevention of diseases such as asthma, COPD and rheumatoid arthritis

US 2011124512 EP 2308855

Novartis 2,4-diaminopyrimidine derivatives; useful for treating disorders in which ZAP70 and/or SYK inhibition has a role, or in disorders caused by a malfunction of signalling cascades connected with FAK

EP 2323993 Portola Pharmaceuticals

Inhibitors of SYK and methods of using them to inhibit platelet aggregation and to treat thrombosis or non-Hodgkin’s lymphoma

EP 2321283 Portola Pharmaceuticals

2,6-diamino-pyrimidin-5-yl-carboxamides that act as SYK or JAK inhibitors

US 2012015937 Rigel Pharmaceuticals

Compounds that inhibit protein kinases such as JAK, AXL or SYK

US 2011124512 S. Sarkar et al. (University of Boston)

Methods of treating thrombosis and/or enhancing fibrinolysis by inhibiting the activity of SYK or calpain

WO 2011079051 US 2011152273

Takeda Fused heteroaromatic pyrrolidinones; useful for treating disorders involving the immune system and inflammation, including rheumatoid arthritis, haematological malignancies and epithelial cancers

EP 2371835 JP 2011200238

University of Pennsylvania

Inhibition of SYK expression using small interfering RNA

WO 2011045352 University Hospital Basel

A method of treating cancer — for example, brain tumours — by inhibiting SYK using a specific antibody or a small interfering RNA

JP 2011132263 Vertex Diaminotriazole compounds that inhibit FLT3, JAK3, PDK1 and/or SYK

C-13, methyl 2-{5-[(3-benzyl-4-oxo-2-thioxo-1,3-thiazolidin-5-ylidene)methyl]-2-furyl}-benzoate; COPD, chronic obstructive pulmonary disease; FAK, focal adhesion kinase; FLT3, FMS-related tyrosine kinase 3; JAK, Janus kinase; PDK1, 3-phosphoinositide-dependent protein kinase 1; SYK, spleen tyrosine kinase; ZAP70, ζ-chain associated protein kinase 70.

In their Review on p234 Redlich and Smolen discuss the cellular and signalling pathways underlying, and strategies for therapeutically interfering with, the inflammatory loss of bone. Here in TABLE 1, we summarize patent applications published during the past year related to SYK inhibitors. Data were researched using the SureChem database from Macmillan Publishers.

N E W S & A N A LY S I S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 181

© 2012 Macmillan Publishers Limited. All rights reserved

more drugs. Are you thinking about implementing similar schemes in the European Union?We already have conditional approval pathways in Europe but we are nevertheless collaborating with the US Food and Drug Administration and others on developing new approval approaches as well. The question will be how to implement them.

Two things that have to go with this also may be a bit of a challenge for the industry. First, we would need to be able to apply certain restrictions around prescribing the drugs that are approved via any new pathway because otherwise we are only lowering the evidence standard. Second, we have to be able to ensure that the treatment experience of real-world patients treated under any approval schemes will contribute to future benefit–risk assessments.

Industry is keenly awaiting biosimilar guidance documents that are due to be published later this year. What other influential guidance do you expect to issue? There will be guidance on the evaluation of treatments for bacterial infections that will be introduced in January 2013. And we will also issue a lot of guidance documents over the implementation of the pharmacovigilance legislation.

What role do you see for yourself in encouraging European drug development?We understand that the health of the European drug development sector is not that great at the moment, and we understand that we are competing with other regions. We see ourselves as an enabling agency, and so we do quite a bit to encourage drug development for public health reasons. We think that the scientific advice we give supports the drug development endeavour, as does our commitment to methodological agreements and our work with HTAs.

Overall, my aim is to foster the develop ment of new drugs without compro mising the standards and robustness of decision-making.

What are your priorities in terms of fostering regulatory science?I think the EMA is pretty much on target in terms of regulatory science. Overall, however, we still need to decrease the level of uncertainty around drug approvals, and I think that benefit–risk methodologies could contribute to this. These methodologies rank different events and data, and as the ranking system is explicit they enable increased predictability and a lower margin for interpretation of findings. They might allow us to move away from making decisions by voting, which sometimes provides unpredictable results. These tools have already been tested in a few agencies around Europe, and the results are very encouraging. A broader pilot project will be launched soon.

Increasingly, sponsors also face uncertainty over reimbursement after drugs are approved. How do you plan to work with payers to enable access to new drugs?One thing we are doing is working with health technology assessment (HTA) bodies to streamline our processes. A first step, that has already been done and that has yielded positive results, is that we now work with HTA bodies to provide joint scientific advice. And requests from industry for such joint advice are increasing. The earlier we decide what the requirements are from us for benefit–risk assessment and from the payers, the easier it will be for investors and drug developers.

Regulators and drug developers in the United States are talking about how to enable shorter drug development programmes for

What are your goals as head of the EMA?My goals align with my challenges, and one of my key challenges will be the implementation of new pharmacovigilance legislation. This will be both a short-term challenge, in regards to ensuring that we comply with the requirements of the law, but also a longer-term challenge, in that we don’t want to miss the opportunities that this legislation offers for assessing the real-life benefits of a medicine. Better monitoring of real-life usage data — when patients with concomitant diseases and treatments take the drug — can give us information both on effectiveness and on safety. And although it might show us unexpected adverse effects, it might also unveil unexpected benefits. The history of medicine teaches us that observation is as important as the original trial.

The first deadline for this challenge is in July, which is when companies need to fill in a set of information for their products and when the Pharmacovigilance Risk Assessment Committee will start meeting. I expect initially the new system will require more work both for regulators and for industry, but after it has started and we have developed our routines I think it will enable an easier way of handling risk management.

I think that we also have to really focus on how to enhance and improve post-licensing activities. There will always be some uncertainty about a drug’s safety and efficacy at its time of approval, but there is a lot of room for improvement in the post-marketing arena, both in terms of additional post-marketing requirements and in terms of recovering information about recently approved drugs better and faster.

AN AUDIENCE WITH…

Guido RasiIn mid-November last year, Guido Rasi began his 5-year mandate as Executive Director of the European Medicines Agency (EMA). Trained as a physician, Rasi has nearly two decades of experience overseeing either research organizations or basic research, including the development of novel drug delivery technologies and preclinical models of carcinoma. More recently, he was Director General of Italy’s Italian Medicines Agency for 3 years, during which time he was also a member of the EMA’s Management Board. Now, as Executive Director of the European regulatory body, his key goal is to ensure that new pharmacovigilance legislation is implemented to best benefit both drug developers and patients, he told Asher Mullard.

N E W S & A N A LY S I S

182 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Years post-launch

Ris

k-ad

just

ed E

BIT

(mill

ions

of U

S$)

Ris

k-ad

just

ed N

PV(m

illio

ns o

f US$

)

400

350

300

250

200

150

100

50

0

-15 1 2 3 4 5 6 7 8 9 10 11 12 13

Base case P&LNet biomarker impact

1,000

900

100

0

$825 $22$92 $939

800

700

Base case25% reduced Phase III costs10% increased salesNPV after use of biomarker

FROM THE ANALYST’S COUCH

Maximizing the value of diagnostics in Alzheimer’s disease drug developmentEric M. Snyder, Jake Olin and Frank S. David

Numerous analyses have emphasized the potential value of biomarkers and diagnostic tests in drug development and commercialization1; however, much less consideration has been given to how companies can realize this value. Here, we examine key strategic issues faced by pharmaceutical companies relating to the development of biomarkers and diagnostics in Alzheimer’s disease (AD). As scientific knowledge of AD biomarkers is still emerging, drug developers, clinicians and diagnostics companies have little relevant expertise, which makes this a particularly challenging disease area for biomarkers (that is, indicators that supplement traditional end points to guide drug development) and commercial diagnostic tests.

Although disease-modifying therapies for AD could dramatically improve the lives of patients and would be expected to rapidly achieve blockbuster status, companies face two daunting challenges. First, during development companies must demonstrate that therapies are disease-modifying and affect the underlying pathology, which is particularly challenging given the lack of qualified biomarkers of disease progression in AD. Second, after launch they must identify the patients in early stages of the disease

(patients with mild cognitive impairment or earlier AD symptoms) who are most likely to benefit from a disease-modifying therapy, which is noteworthy as less than 50% of patients are currently diagnosed this early2. In the context of these challenges, and with multiple efforts underway to develop AD biomarkers in academia and industry3, we examined the approaches of drug companies to define how (and how much) they should invest in AD biomarkers and diagnostic tests to support their drug pipelines.

Selecting the optimal AD analyteThe selection of a diagnostic analyte or modality has a considerable impact on the opportunities for drug companies to realize its value. In AD, biomarker and diagnostics development has focused primarily on two sets of approaches: measuring levels of amyloid-β aggregates or tau proteins in the cerebrospinal fluid (CSF) and brain imaging via magnetic resonance imaging or positron emission tomography (TABLE 1). As biomarkers for use in drug research and development (R&D), these modalities provide more robust end points than current cognitive tests, and can facilitate smaller and shorter clinical trials4.

However, neither CSF analyses nor imaging modalities are optimally suited for use as a commercialized diagnostic outside of clinical trials. CSF measurements require invasive lumbar punctures and brain imaging requires expensive equipment. If these diagnostics are required for patients to use a therapy, they may limit drug use and impede commercial potential.

To overcome this challenge, several companies are pursuing novel modalities that are easier to use. Merck and Neuroptix, for example, have collaborated to develop a device to measure retinal levels of amyloid-β that could be used in an optician’s office setting; and DiaGenic is collaborating with Pfizer and Merz Pharmaceuticals to develop blood-based tools (TABLE 1). These diagnostics still require validation but are less invasive and could be easily adapted to the physician’s office setting.

As companies are seeking to realize the value of biomarkers and/or diagnostics in AD, they may need to invest in multiple approaches, including possibly one for R&D and another for commercialization. For example, Pfizer has facilitated the development of imaging tools through its venture investment in Avid Radiopharmaceuticals, and has also entered a development partnership with DiaGenic for a blood-based biomarker. Pfizer has therefore positioned itself to access an advanced imaging diagnostic in clinical trials and also support the development of a blood-based test that can help physicians diagnose patients once Pfizer’s therapies reach the market.

Structuring a partnershipHow companies access biomarker and/or diagnostic technology further affects the potential value of a new therapeutic. Given the evolving diagnostic landscape in AD, non-exclusive investments in — or partnerships with — early-stage companies may help to catalyse the development of the overall market. Such open approaches may allow several researchers to develop, ▶

Eadie armchair by Donna Wilson, courtesy of www.scp.co.uk

Figure 1 | Profit and loss calculation for a hypothetical AD therapy. The cost of Phase III trials was estimated at US$107.5 million and the probability of reaching the market at 66%, achieving $1 billion in revenue 7 years post-launch. We calculated the earnings before interest (EBIT) and then the net present value (NPV)5 (see Supplementary information S1 (box)). The P&L was then modified to reflect a net biomarker impact scenario. Assuming a 25% reduction in development costs at the Phase III stage, the EBIT increases, causing an increase in NPV by $22 million. Combined with a hypothetical increase in product sales by 10% due to better patient identification, the NPV is increased by a total of $114 million. AD, Alzheimer’s disease; P&L, profit and loss calculation.

N E W S & A N A LY S I S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 183

© 2012 Macmillan Publishers Limited. All rights reserved

ALZHEIMER’S DIAGNOSTICS | MARKET INDICATORS

▶ improve and refine an emerging technology, which could hasten the development of data and generate crucial support needed for its acceptance by regulators, payers and clinicians. Furthermore, in AD particularly, non-exclusive partnerships may mitigate the inherent risk in the fact that much of the research on AD biomarkers is being performed in academia or by early-stage diagnostics companies, and most of the large test developers with R&D and commercial expertise lack experience in AD.

Depending on a drug company’s risk appetite and internal capabilities, however, acquiring a company with proprietary technology or intellectual property may provide a unique competitive advantage for the development of therapeutic agents for AD, and also leave open the option to reap diagnostics revenues. Avid provides an example of how such a strategy can evolve over time. Avid was originally supported in part through several pharmaceutical corporate venture investments and R&D partnerships through the early phases of data generation and regulatory discussions. In 2010, Avid was acquired by Lilly, which hopes to both use and commercialize Avid’s tests in R&D.

Maximizing return on investmentOne of the most crucial issues for drug companies as they commit financial resources to AD biomarkers is how to assess the relative value of these diagnostic tools. We developed

Table 1 | Examples of drug–diagnostic partnerships in Alzheimer’s disease

Drug company Diagnostics company

Rx–Dx partnership structure

Diagnostic modality

Pfizer/Johnson & Johnson

Avid Radio-pharmaceuticals

Venture investment and clinical trial collaboration (bapineuzumab)

PET ligand that binds to amyloid-β

Lilly Avid Radio-pharmaceuticals

Acquisition PET ligand that binds to amyloid-β

Johnson & Johnson GE Healthcare Joint investment PET- and MRI-based tools

AC Immune Bayer Clinical trial (ACI-24) PET ligand that binds to amyloid-β

Lilly C2N Diagnostics R&D grant Radiolabelled production and turnover of amyloid-β and tau in the CSF

Merck Neuroptix Venture investment Measurement of amyloid-β in the retina

Merz Pharmaceuticals

DiaGenic R&D partnership Blood-based PCR test for multiple RNAs affected by AD

Pfizer DiaGenic R&D partnership Blood-based PCR test for multiple RNAs affected by AD

AD, Alzheimer’s disease; CSF, cerebrospinal fluid; MRI, magnetic resonance imaging; PET, positron emission tomography; R&D, research and development; Rx–Dx, drug–diagnostic.

a profit and loss calculation (P&L) for a hypothetical AD therapy entering Phase III clinical trials using historical estimates. In our base case scenario, we assumed that the therapy could achieve US$1 billion in peak sales within 7 years post-launch, in line with current equity research estimates for late-stage therapies such as Pfizer’s bapineuzumab and Lilly’s solanezumab. We then applied estimates of Phase III duration, cost and probability of success5, and calculated the earnings before interest and tax (EBIT) and net present value (NPV) for this product (see FIG. 1 and Supplementary information S1 (box)).

We then modified these assumptions in our P&L to account for a biomarker that could improve R&D efficiency and/or be developed into a commercialized diagnostic in two model cases. In R&D, we calculated the impact of reducing Phase III trial costs by 25%, in line with a recent study that estimated that imaging diagnostics could reduce the size of AD clinical trials by up to 40%4 (see Supplementary information S1 (box)). From a commercial perspective, if a commercial diagnostic can increase the size of the addressable patient population (by improving diagnosis rates), it will increase peak sales. There are clearly additional benefits that could be reaped with biomarkers or diagnostics, such as reduced trial duration or faster progression to peak sales, but even these illustrative calculations demonstrate that biomarkers and diagnostics can substantially increase the value of a

therapeutic agent for AD. A similar analytical approach for defining the potential value of investments in biomarkers or diagnostics is likely to be applicable to other chronic diseases, in which it could help to overcome the challenge of patient diagnosis in R&D and post-launch, respectively.

AD presents an intriguing test case for pharmaceutical companies’ investments in biomarkers and diagnostics. Compared with diseases like cancer and cardiology, in which the scientific, regulatory, clinical and commercial landscapes are more mature, AD requires a more rigorous analysis of risk tolerance, desired and likely amount of return, and internal capabilities to determine the optimal nature and level of investment into biomarkers and/or diagnostics. In this context, pharmaceutical companies with suites of AD programmes may stand to benefit most substantially from these investments, provided they can optimally integrate the technologies and knowledge into their decision-making processes as well as ongoing R&D and commercialization efforts. Importantly, for maximum benefit these investments need to be incorporated into a strategy that encompasses the full range of activities, from biomarker selection to drug commercialization, and they need to be deployed sufficiently early in the clinical development process.

Eric M. Snyder and Jake Olin are at Leerink Swann Consulting LLC, 1251 Avenue of the Americas,

22nd Floor, New York 10020, USA.

Frank S. David is at Leerink Swann Consulting LLC, One Federal St., 23rd Floor, Boston,

Massachusetts 02110, USA.

Correspondence to E.S.  e‑mail: Eric.Snyder@Leerink.Com

doi:10.1038/nrd3535

1. Hurko, O. & Jones, G. K. Valuation of biomarkers. Nature Rev. Drug Discov. 10, 253–254 (2011).

2. Boise, L. et al. Dementia assessment in primary care: results from a study in three managed care systems. J. Gerontol. A Biol. Sci. Med. Sci. 59, M621–M626 (2004).

3. Hampel, H. et al. Biomarkers for Alzheimer’s disease: academic, industry and regulatory perspectives. Nature Rev. Drug Discov. 9, 560–574 (2010).

4. Beckett, L. A. et al. The Alzheimer’s disease neuroimaging initiative: annual change in biomarkers and clinical outcomes. Alzheimers Dement. 6, 257–264 (2010).

5. Dimasi, J. A. et al. Trends in risks associated with new drug development: success rates for investigational drugs. Clin. Pharmacol. Ther. 87, 272–277 (2010).

Competing interests statementThe authors declare competing financial interests: see Web version for details.

SUPPLEMENTARY INFORMATIONSee online article: S1 (box)

ALL LINKS ARE ACTIVE IN THE ONLINE PDF

N E W S & A N A LY S I S

184 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

2-(3,5-dichlorophenyl)-1,3-benzoxazole-6-carboxylic acid; C14H7Cl2NO3; Mr = 308.12

Tafamidis

O

N

HO

O

Cl

Cl

FRESH FROM THE PIPELINE

TafamidisGerard Said, Seden Grippon and Peter Kirkpatrick

In November 2011, tafamidis (Vyndaqel; Pfizer), a small molecule that inhibits the dissociation of transthyretin tetramers, was granted marketing authorization by the European Commission for the treatment of transthyretin amyloidosis in adult patients with stage 1 symptomatic polyneuropathy to delay peripheral neurological impairment.

Transthyretin familial amyloid polyneuropathy (TTR-FAP) is a rare autosomal dominant neurodegenerative disorder. It is characterized by progressive sensory, motor and autonomic impairment, which typically leads to death around a decade after diagnosis1. Cardiac involvement is reported in most cases, and other manifestations can include loss of body weight and cachexia1.

Nerve lesions in TTR-FAP are induced by the deposition of amyloid fibrils, which is due to mutations in TTR — most commonly Val30Met (V30M)1. Treatments are primarily symptomatic, such as analgesics for neuropathic pain, although liver transplantation to reduce the formation of additional amyloid deposits (by removing the main source of TTR) seems to be beneficial in patients with the V30M mutation1.

Basis of discoveryTTR is a 127-amino-acid protein that assembles to form soluble tetramers, which circulate in the blood and cerebrospinal fluid1,2. The normal role of TTR is to transport retinol and thyroxine, although the two thyroxine-binding sites are largely unoccupied in humans1,2.

Pathogenic mutations in TTR, such as the V30M mutation, decrease the stability of TTR tetramers, enhancing their dissociation into monomers2. These monomers undergo partial denaturation to form amyloidogenic intermediates that self-aggregate in the extracellular space into soluble oligomers and protofibrils, ultimately leading to the formation of insoluble amyloid fibrils2.

As with other amyloid diseases, the development of therapeutic strategies for TTR-FAP has focused on reducing the levels of the amyloidogenic peptide — in this case, monomeric TTR2. One such strategy is based on the idea of achieving this goal by stabilizing

the tetrameric form of TTR, the dissociation of which into monomers is the rate-limiting step in the pathogenesis of TTR-FAP2. Following the demonstration that binding of ligands to the thyroxine-binding sites on TTR tetramers could inhibit TTR fibril formation in vitro by stabilizing the tetramer3, structure- guided design and screening of libraries of such potential compounds led to the identification of several classes of kinetic TTR stabilizers2. Optimization of compounds in one such class resulted in tafamidis4 (FIG. 1).

Drug propertiesTafamidis binds highly selectively to TTR in human plasma2,4. It dose-dependently kinetically stabilizes TTR under denaturing conditions (in the presence of 6 M urea) as well as physiological conditions2,4. Importantly, tafamidis kinetically stabilizes a broad spectrum of TTR variants, suggesting that it could be broadly applicable for treating TTR amyloidoses2.

Clinical dataThe efficacy and safety of tafamidis (20 mg orally administered once daily) was evaluated in an 18-month, randomized, double-blind, placebo-controlled trial involving 128 patients with TTR-FAP with the V30M mutation and primarily stage 1 disease5. The primary outcome measures were the Neuropathy Impairment Score of the Lower Limb (NIS-LL; a physician assessment of the neurological exam of the lower limbs) and the Norfolk Quality of Life Diabetic Neuropathy score (a patient-reported outcome that uses a total quality of life (TQOL) score)5. Secondary outcome measures included composite scores of large nerve fibre and small nerve fibre function, and nutritional assessments using the modified body mass index (mBMI; BMI multiplied by serum albumin in g per l)5.

After 18 months of treatment, a greater proportion of the patients in the group receiving tafamidis were NIS-LL responders — defined as a change of less than 2 points in NIS-LL — than in the group receiving placebo5. Although the differences between the groups did not meet the criteria for statistical significance in the pre-specified intent-to-treat population, they did in a pre-specified efficacy evaluable analysis of patients who completed

the 18-month treatment per protocol5. In this analysis, 27 out of 45 (60.0%) of the patients in the tafamidis group were NIS-LL responders, compared with 16 out of 42 (38.1%) of the patients in the placebo group5. The mean change from baseline in TQOL score in these patients was 0.1 in patients in the tafamidis group, compared with 8.9 in the placebo group5. Analysis of the secondary end points showed that tafamidis treatment resulted in less deterioration of neurological function and improved nutritional status (as assessed by mBMI) compared with placebo5.

Of the 91 patients completing the 18-month treatment period, 86 were subsequently enrolled in an open-label extension study, in which they all received once-daily 20 mg tafamidis for a further 12 months5. The rate of change in the NIS-LL score during the 12 months of treatment in this study was similar to that observed in those patients who were randomized and treated with tafamidis in the previous double-blind 18-month period5.

IndicationsTafamidis is approved by the European Commission for the treatment of TTR amyloidosis in adult patients with stage 1 symptomatic polyneuropathy to delay peripheral neurological impairment5. ▶

Figure 1 | Tafamidis. Screening of a library of substituted benzoxazoles using a fibril formation assay led to the identification of tafamidis4. A crystal structure with transthyretin (TTR) demonstrated its association with the thyroxine-binding site4. Tafamidis reaches its EC

50 (the effective

concentration for half-maximal response) for preventing TTR fibril formation at a tafamidis/TTR tetramer ratio of <1, which is consistent with tafamidis effectively stabilizing TTR when it occupies only one of the two thyroxine-binding sites in TTR tetramers2.

N E W S & A N A LY S I S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 185

© 2012 Macmillan Publishers Limited. All rights reserved

Analysing issues for the treatment of FAP is Gerard Said, M.D., Professor, Department of Neurology, Centre Hospitalier Universitaire Pitié-Salpêtrière, Paris, France.

Tafamidis is the first pharmacological treatment available to treat TTR-FAP, and currently the only alternative to liver transplantation. The pivotal randomized placebo-controlled trial of tafamidis was conducted in patients carrying the V30M mutation at stage 1 of the disease, with most of them having symptomatic polyneuropathy and autonomic dysfunction but without walking impairment (see above). During the 18-month trial, patients receiving tafamidis had a slower deterioration of peripheral nerve function, as well as a better quality of life and general condition than patients on placebo. Tafamidis was also well tolerated, with a low incidence of serious adverse events that included urinary tract infection and vomiting, without a trend of increasing incidence over time.

The drug, which inhibits the dissociation of TTR tetramers, is expected to reduce the rate of amyloid formation and deposition in the peripheral nervous system of patients carrying an amyloidogenic mutation. However, amyloid deposits that are already present in target organs when the treatment

is started will not be cleared from these organs, and improvement of clinical manifestations is unlikely to occur. It is therefore important to start tafamidis as early as possible in patients with clinical manifestations of FAP, but major differences exist in the age of onset, clinical presentation and course of the disease among patients carrying the common V30M mutation, especially in non-endemic areas. In Portugal and Brazil, most carriers are symptomatic by 50 years of age. Conversely, in Sweden only 11% of the carriers become symptomatic by 50 years of age6. The rate of clinical deterioration may also vary. In the next few years, we should learn more thanks to the THAOS (Transthyretin Amyloidosis Outcomes Survey) registry, a 10-year global observational study looking at the natural history of patients with TTR amyloidosis and asymptomatic carriers of the genetic defect.

In all patients (carrying the V30M mutation or other mutations) diagnosed at stage 1 of the disease, the choice now is between early liver transplantation and tafamidis. Liver transplantation, which dramatically decreases the level of mutant TTR in the blood, seems to reduce the rate of progression of the neuropathy but does not protect against cardiac involvement, which occurs in about 80% of cases of

TTR-FAP. Cardiac and ocular amyloid depositions progress in patients who have undergone transplants (for example, see REF. 7), but whether these manifestations will be controlled by tafamidis remains to be seen. At present, it seems advisable to start with tafamidis in patients at stage 1 of the disease, with mandatory periodic evaluation. In case of disease progression, the patient should be put on a liver transplantation waiting list, if this has not been already done, as was the case for most patients enrolled in the pivotal trial. In recipients of liver transplants, adjunct therapy with tafamidis also seems advisable.

The availability of a pharmacological treatment for TTR-FAP is a milestone in the field of familial amyloid polyneuropathies. Furthermore, other small-molecule drugs including doxycycline and diflunisal are under clinical evaluation. Gene-based therapy also remains a promising future strategy for TTR amyloidosis, and agents based on small interfering RNA technology and antisense technology are entering clinical evaluation for TTR-FAP.

Gerard Said is at the Department of Neurology, Centre Hospitalier, Universitaire Pitié-Salpêtrière, 75651

Paris Cedex 13, France.

Seden Grippon is at IMS Health, 7 Harewood Avenue, London NW1 6JB, UK.

Peter Kirkpatrick is at Nature Reviews Drug Discovery.

e-mails: grrdsd@gmail.com; SGrippon@imscg.com; p.kirkpatrick@nature.com

doi:10.1038/nrd3675

1. Planté-Bordeneuve, V. & Said, G. Familial amyloid polyneuropathy. Lancet Neurol. 10, 1086–1097 (2011).

2. Johnson, S. M. et al. The transthyretin amyloidoses: from delineating the molecular mechanism of aggregation linked to pathology to a regulatory-agency-approved drug. J. Mol. Biol. 5 Jan 2012 (doi: 10.1016/j.jmb.2011.12.060).

3. Miroy, G. J. et al. Inhibiting transthyretin amyloid fibril formation via protein stabilization. Proc. Natl Acad. Sci. USA 93, 15051–15056 (1996).

4. Razavi, H. et al. Benzoxazoles as transthyretin amyloid fibril inhibitors: synthesis, evaluation, and mechanism of action. Angew. Chem. Int. Ed. 42, 2758–2761 (2003).

5. European Medicines Agency (EMA). European Public Assessment Report. EMA website [online], http://www.ema.europa.eu/docs/en_GB/document_library/EPAR_-_Product_Information/human/002294/WC500117862.pdf (2011).

6. Hellman, U. et al. Heterogeneity of penetrance in familial amyloid polyneuropathy, ATTR Val30Met, in the Swedish population. Amyloid 15, 181–186 (2008).

7. Ohya, Y. et al. Manifestations of transthyretin-related familial amyloidotic polyneuropathy:long-term follow-up of Japanese patients after liver transplantation. Surg. Today 41, 1211–1218 (2011).

Competing financial interestsThe authors declare no competing financial interests.

ANALYSIS | FAMILIAL AMYLOID POLYNEUROPATHY

▶

Box 1 | Market for transthyretin familial amyloid polyneuropathy

Analysing the market for transthyretin familial amyloid polyneuropathy (TTR-FAP) is Seden Grippon, IMS Health, London, UK.

TTR-FAP is a progressive and fatal neurodegenerative genetic disease that is estimated to affect ~5,000–10,000 patients worldwide. Liver transplantation is the only established effective treatment option; however, it has variable success, with the majority of patients continuing to have disease progression of some level, and it is very expensive. Off-label use medications for supportive care include diuretics, analgesics and some calcium channel blockers.

Tafamidis (Vyndaqel; developed by FoldRx, now a subsidiary of Pfizer) became the first drug to be approved for the treatment of TTR-FAP, with its market authorization in the European Union in November 2011. The US Food and Drug Administration issued a refusal to file a letter in April 2011 owing to the application not being sufficiently complete, but it is anticipated that the drug will be resubmitted and approved in the United States in 2012.

Tafamidis has an orphan drug designation for TTR-FAP in the European Union and the United States, which will further strengthen its position in the market for treatments for TTR-FAP, which has been estimated to be worth US$700 million to $1.4 billion annually, based on a treatment cost of ~$140,000 per year (Kozul, M. Think Equity LLC report on Alnylam Pharmaceuticals Inc. 22 Nov 2011). Considering the rapid development timelines and premium orphan pricing, this therapy area is of high potential for other companies as well. Isis Pharmaceuticals has reported that a Phase I trial of an antisense agent targeting the TTR gene, ISIS-TTRRx, has been successfully completed, and a clinical trial in patients is planned for 2012. Alnylam has reported promising initial results in a Phase I trial of a small interfering RNA (siRNA) agent, ALN-TTR-01, that inhibits the expression of the TTR gene. A second-generation siRNA product, ALN-TTR-02, with improved dosing characteristics is due to begin Phase I trials in 2012.

N E W S & A N A LY S I S

186 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Results from a clinical study (ClinicalTrials.gov identifier: NCT01070407) of an investigational adenovirus-based vaccine against hepatitis C virus (HCV) suggest that inducing host T cell responses might be an effective strategy to provide protective immunity against HCV.

HCV infection can be spontane-ously controlled in a proportion of infected individuals, and studies of host genetics and immunology suggest that T cells have an important role in protective immunity against HCV. So a vaccine that induces T cell responses might stimulate immune-mediated control of HCV infection. Indeed, in a previous study, a vaccine based on a segment of DNA coding for the nonstructural region (from NS3 to NS5B) of HCV genotype 1b that was delivered using adenovirus constructs induced T cell responses and suppressed acute viraemia in chimpanzees.

In the current study, published in Science Translational Medicine, Barnes and colleagues investigated whether such an approach might also induce T cell responses in healthy volunteers. To do this, they used two replication-defective vectors based on rare adenoviral serotypes (to try and overcome the formation of neutral-izing antibodies in the volunteers): chimpanzee adenovirus 3 (ChAd3) and human adenovirus 6 (Ad6) vectors, both of which were engineered to express the HCV proteins NS3, NS4 and NS5 of HCV genotype 1b.

After first determining that the vaccine was safe and well tolerated, the authors showed that priming doses of the vaccine induced immune responses (assessed by increases in

interferon-γ production) that peaked at 2–6 weeks and were detectable for 24 weeks after vaccination. Moreover, vaccination induced long-lived central and effector memory T cells that could be detected for 6 months after a boosting dose of the vaccine.

Because a broad T cell response (that is, one that is directed against several target antigens) correlates with better control of the virus, it was encouraging to note that at the highest dose of the vaccine, reactivity against the peptides NS3, NS4A/B, NS5A and NS5B was observed.

The authors next analysed the T cell responses that were induced by vaccine administration. Both vectors induced HCV-specific CD4+ and CD8+ T cells that secreted multiple pro-inflammatory and antiviral cytokines, including interferon-γ, tumour necrosis factor and inter-leukin-2, although the response of CD4+ T cells (which have a key role in defence) was much lower than that of the CD8+ cells.

Next, the authors tested whether the T cell responses induced by geno-type 1b were cross-reactive against other HCV genotypes. Overall, cross-genotype recognition occurred but it was of a lower magnitude; the response to genotype 1a was about half of the response to genotype 1b, and that of genotype 3a was about a fifth of the response to genotype 1b.

Following this, they assessed whether the response induced by a priming dose of the ChAd3 vaccine could be boosted by a dose of the Ad6 vaccine, and vice versa. Although there was some boosting of T cell responses, the overall magnitude did not exceed that

elicited by the priming vaccine; the boosting effect was greatest when ChAd3 was used as the priming vaccine. This suggested that the priming vector induced cross-neutralizing adeno virus antibodies.

The authors note that although priming the T cell response using chimpanzee adenoviral vectors appears to be robust, the use of alternative vectors to boost the response might avoid potential cross-neutralization and so maximize long-term memory responses; trials of such a strategy are underway (EudraCT identifier: 2009-018260-10).

Charlotte Harrison

ORIGINAL RESEARCH PAPER Barnes, E. et al. Novel adenovirus-based vaccines induce broad and sustained T cell responses to HCV in man. Sci. Transl. Med. 4, 115ra1 (2012)

V I R A L D I S E A S E

Steps towards an HCV vaccine

Imag

eSou

rce

R E S E A R C H H I G H L I G H T S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012

Nature Reviews Drug Discovery | AOP, published online 17 February 2012; doi:10.1038/nrd3687

© 2012 Macmillan Publishers Limited. All rights reserved

Treatment options for neuropathic pain, which develops as a result of injury to the nervous system, are limited and have not advanced substantially for many years, in part owing to a lack of understanding of its molecular basis. A recent metabo-lomics study by Patti et al., published in Nature Chemical Biology, impli-cates an endogenous lipid metabolite in neuropathic pain and might provide a new pathway to target with potential analgesics.

In a well-established rat model of neuropathic pain — tibial nerve transection (TNT) — allodynia (pain in response to light touch) persists for at least 9 weeks after the initial transection of the tibial branch of the sciatic nerve, despite the wound appearing healed. Patti and col-leagues took an untargeted mass spectrometry-based metabolomics approach to analyse metabolites from plasma, the tibial nerve, dorsal root ganglia and the rat dorsal horn of TNT rats and control (sham-operated) animals 21 days after surgery. A total of 733 metabolic features showed over a twofold change between the two groups, and 94% of these were derived from the dorsal horn.

Focusing on this site, where the damaged nerve meets the spinal cord,

the authors discovered that sphingo-myelin–ceramide metabolism was markedly affected. In particular, levels of ceramide (d18:1) and several phosphatidylcholines were increased, whereas those of several diacyl-glycerols were decreased, consistent with an increased degradation of sphingomyelin to ceramide. N,N-dimethylsphingosine (DMS) was one of several sphingomyelin–ceramide metabolites that were significantly upregulated.

Having determined the con-centration of DMS present in the dorsal horn 21 days after TNT, the research team then demonstrated that intrathecal injection of a similar concentration of DMS could induce mechanical allodynia in healthy rats within 24 hours. Intrathecal DMS injection also increased the expression of glial fibrillary acidic protein in the spinal cord, which is indicative of astrocyte activa-tion. Among the many substances released by activated astrocytes, interleukin-1β (IL-1β) and monocyte chemoattractant protein 1 (MCP1) are inflammatory mediators with roles in nociceptive responses, and the authors showed that cultured astrocytes released increased levels of both IL-1β and MCP1 following DMS treatment.

These data indicate that sphingo-myelin–ceramide metabolism is altered in the dorsal horn of rats sub-jected to neuropathic pain, resulting in increased levels of the endogenous ceramide catabolite DMS, which can elicit mechanical hypersensitivity in vivo and cytokine release in vitro. They also demonstrate the power of using an untargeted approach to iden-tify potential new drug targets, with future efforts likely to focus on identi-fying the specific enzymes involved in DMS biosynthesis as potential points for therapeutic intervention.

Katrin Legg

ORIGINAL RESEARCH PAPER Patti, G. J. et al. Metabolomics implicates altered sphingolipids in chronic pain of neuropathic origin. Nature Chem. Biol. 22 Jan 2012 (doi:10.1038/nchembio.767)

M E TA B O LO M I C S

Gaining insight into pain

GETTY

R E S E A R C H H I G H L I G H T S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012

© 2012 Macmillan Publishers Limited. All rights reserved

Recent studies have highlighted the role of mitochondrial dysfunction in several neurodegenerative diseases. Now, a study in rat models of Parkinson’s disease highlights the therapeutic benefit of protecting mitochondrial complex I activity with a viral non-coding RNA delivered to the brain using a pep-tide derived from the rabies virus glycoprotein (RVG). This novel therapeutic strategy could have important implications for the treat-ment of Parkinson’s disease and other neurodegenerative diseases involving impaired mitochondrial function.

The non-coding p137 RNA, derived from the human cytomeg-aloviral β2.7 transcript, is expressed during viral infection. It interacts directly with mitochondrial complex I and seems to be essential for prevent-ing cell death and maintaining energy production in infected cells. In this study, Sinclair and colleagues set out to investigate whether p137 RNA could prevent the loss of dopamin-ergic neurons in experimental models of Parkinson’s disease.

To deliver the RNA to the brain, the authors used a previously described RVG derivative containing nine arginine residues that facilitate RNA binding: RVG9R. This peptide binds to acetylcholine receptors, which are exclusively expressed in central nervous system cells, and has been shown to deliver small inter-fering RNA to the brain following

peripheral administration, thus over-coming the need to use more invasive intracerebral delivery procedures.

First, the authors showed that the p137 RNA–RVG9R complex protected cells cultured in vitro from exposure to rotenone, a mito-chondrial complex I inhibitor, and 6-hydroxydopamine (6-OHDA), a neurotoxin that selectively kills dopaminergic neurons and is com-monly used to induce Parkinson’s-like disease in laboratory animals. In light of these results, they admin-istered the complex via intravenous injection to rats 3 days before an acute intranigral 6-OHDA insult. Such pretreatment significantly attenuated the functional deficits induced by the lesion, as assessed in several behavioural tests.

Further experiments showed that in rats treated with p137 RNA–RVG9R, p137 physically interacts with mitochondrial complex I and protects its enzymatic activity in the nigral tissue following the 6-OHDA insult. Importantly, repeated p137 RNA–RVG9R treatment 1–2 days after an intrastriatal 6-OHDA injec-tion was able to attenuate the loss of dopaminergic neurons in the substantia nigra and correct the behavioural deficits induced by the lesion without stimulating a host immune reaction.

The therapeutic complex failed to elicit an increase in T cell infiltration, microglial activation or an antibody response, indicating that the treatment is non-immunogenic.

Although future work should assess the effects of p137 RNA–RVG9R in transgenic models of Parkinson’s disease and determine the pharmacokinetic profile of the agent, this study supports the idea that mitochondrial dysfunction is at the heart of this neurodegenerative disorder and that a non-coding viral RNA could be an efficient and feasible approach to protect these organelles from damage. Moreover, the use of another virus-derived peptide to deliver the RNA transvascularly to damaged neurons could be useful for future clinical applications.

Monica Hoyos Flight

ORIGINAL RESEARCH PAPER Kuan, W. L. et al. A novel neuroprotective therapy for Parkinson’s disease using a viral noncoding RNA that protects mitochondrial Complex I activity. J. Exp. Med. 209, 1–10 (2012)

N E U R O D E G E N E R AT I V E D I S E A S E

Harnessing virus-mediated mitochondrial protection to combat neurodegeneration

R E S E A R C H H I G H L I G H T S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012

Nature Reviews Drug Discovery | AOP, published online 17 February 2012; doi:10.1038/nrd3688

© 2012 Macmillan Publishers Limited. All rights reserved

The transcriptional co-activator PPARγ co-activator 1α (PGC1α) has a central role in the regulation of cellular energy metabolism. Its expression is induced by exercise in muscle, where it mediates various beneficial effects. Now, writing in Nature, Boström and colleagues demonstrate that increased muscle PGC1α expression also positively affects adipose tissue — it stimulates the production and secretion of the novel hormone irisin from the muscle, which activates thermo-genesis in fat, resulting in weight loss and improved glucose homeostasis in obese mice.

Exercise improves metabolic status in obesity and type 2 diabetes, but the underlying molecular mechanisms are poorly understood. Given that muscle PGC1α expression is elevated upon exercise, and transgenic mice with mildly elevated muscle PGC1α

(MCK-PGC1α mice) are resistant to age-related obesity and insulin resistance, Boström and colleagues set out to investigate a possible role of muscle PGC1α in the beneficial metabolic effects of exercise.

First, they analysed the adipose tissue of MCK-PGC1α mice and discovered that mRNA levels of thermogenic genes characteristic of brown fat, including the brown adipocyte marker uncoupling protein 1 (UCP1), were significantly increased in subcutaneous white adipose tissue (WAT) — an effect termed ‘browning’. This thermogenic gene programme was similarly induced when control mice were exposed to wheel running or swimming in warm water.

Next, they treated cultured primary subcutaneous adipocytes with media conditioned by myocytes expressing PGC1α, and found mRNA levels of several brown-fat-specific genes to be increased in the adipocytes. This suggested that the browning they observed in mice may be mediated by a molecule secreted from muscle under the regulation of PGC1α.

To search for such a molecule, they analysed muscle from MCK-PGC1α mice using gene expression arrays and an algorithm that predicts protein secretion, and identified five candidate proteins. Applying these proteins directly to primary white adipocytes during differentiation revealed that one of them — fibronectin type III domain-containing protein 5

(FNDC5) — potently upregulated the expression of UCP1 and other brown fat genes, whereas it downregulated the expression of genes characteristic of WAT development. Importantly, FNDC5 mRNA expression was increased in muscle from mice and humans, following exercise.

Further in vitro studies revealed that FNDC5 undergoes proteolytic cleavage and glycosylation to produce a highly conserved 112-amino-acid secreted polypeptide, which they named irisin. Irisin was detected in mouse and human plasma, and its levels were increased upon exercise.

Finally, they assessed the biologi-cal and therapeutic effects of irisin. Mildly increasing irisin levels in mice by injecting adenoviral vectors expressing FNDC5 induced WAT browning; this resulted in increased energy expenditure, reduced body weight and improved glucose toler-ance in obese, insulin-resistant mice. When anti-FNDC5 antibodies were injected into mice prior to swim-ming, this prevented browning, indicating a requirement for irisin in this exercise-associated effect.

Together, these findings may have implications for the future treatment of metabolic disease. Indeed, Ember Therapeutics, co-founded by the lead author of this study, is currently generating variants of irisin in preparation for clinical trials.

Sarah Crunkhorn

ORIGINAL RESEARCH PAPER Boström, P. et al. A PGC1‑α‑dependent myokine that drives brown‑fat‑like development of white fat and thermogenesis. Nature 481, 463–468 (2012)

M E TA B O L I C D I S E A S E

Exercise hormone fights metabolic disease

GETTY

R E S E A R C H H I G H L I G H T S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012

Nature Reviews Drug Discovery | AOP, published online 17 February 2012; doi:10.1038/nrd3686

© 2012 Macmillan Publishers Limited. All rights reserved

The mixed lineage leukaemia (MLL) gene, which encodes a histone methyltransferase, is frequently translocated in human acute leukae-mias. The interaction of MLL with menin is known to be essential for the oncogenic activity of MLL-fusion proteins; therefore, targeting this interaction could have therapeutic relevance.

Jolanta Grembecka, Tomasz Cierpicki and colleagues have previ-ously characterized the interaction between MLL and menin, which requires the amino terminus of

MLL, and this part of the protein remains intact in all MLL-fusion proteins studied to date. The authors used high-throughput screening to identify compounds that target menin and that suppress its inter-action with MLL. The most potent compound, a thienopyrimidine dubbed MI-1, binds menin and is a competitive inhibitor of the MLL–menin interaction. Structure–activity analyses led to the genera-tion of MI-1 analogues, MI-2 and MI-3. These molecules bind to wild-type menin — but not to menin mutants in which the interaction site with MLL is mutated — and were taken forwards for further testing.

Co-immunoprecipitation experi-ments showed that, compared with a control compound, both MI-2 and MI-3 inhibited the interaction between leukaemic MLL-fusion proteins and menin, and this inhibi-tion promoted the differentiation of leukaemia cells and reduced their ability to form colonies in soft agar. MLL-fusion proteins are thought to promote leukaemogenesis partly by enabling the transcription of HOX genes. The treatment of mouse bone marrow cells that expressed an MLL-fusion protein with either MI-2 or MI-3 downregulated the expression of Hoxa9 and the HOX cofactor Meis1. Further experiments showed that MI-2 reduced the levels of

menin and an MLL–AF9-fusion pro-tein bound at the Hoxa9 promoter, and this was associated with chro-matin condensation. Similar results, including reduced proliferation and increased cell differentiation and apoptosis, were evident in human leukaemia cells with MLL-fusion genes. Interestingly, MI-2 and MI-3 also reduced proliferation in leu-kaemia cells that had high levels of HOX gene expression in the absence of MLL-fusion proteins. The authors speculate that this might be because the interaction between wild-type MLL and menin is prevented on treatment of these cells, and that this interaction is required for HOX gene transcription.

These results indicate that target-ing MLL-driven leukaemia might be possible using drugs that target the MLL–menin interaction. MI-2 and MI-3 should prove useful agents for the development of clinically useful inhibitors and for gaining a better understanding of the function of the MLL–menin interaction in leukaemogenesis.

Nicola McCarthy Chief Editor, Nature Reviews Cancer

This article originally appeared in Nature Rev. Cancer (doi:10.1038/nrc3231).

ORIGINAL RESEARCH PAPER Grembecka, J. et al. Menin–MLL inhibitors reverse oncogenic activity of MLL fusion proteins in leukaemia. Nature Chem. Biol. 29 Jan 2012 (doi:10.1038/nchembio.773)

A N T I C A N C E R D R U G S

Targeting menin

Targeting MLL-driven leukaemia might be possible using drugs that target the MLL–menin interactio.

GETTY

R E S E A R C H H I G H L I G H T S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012

© 2012 Macmillan Publishers Limited. All rights reserved

A N A LG E S I A

Erasing the traces of painOpioids are often given to alleviate pain. Although clinically used doses decrease nociceptive neurotransmission, they do not reverse the synaptic changes — namely long-term potentiation — that underlie the features of chronic pain, such as hyperalgesia. This study showed that a brief high dose of a short-acting opioid agonist reversed plasticity at nociceptive synapses in the rat spinal cord and reversed hyperalgesia in rats. These effects, which involved Ca2+-dependent signalling, lasted well beyond the duration of agonist treatment and suggest that opioids can also erase components of chronic pain.ORIGINAL RESEARCH PAPER Drdla-Schutting, R. et al. Erasure of a spinal memory trace of pain by a brief, high-dose opioid administration. Science 335, 235–238 (2012)

G E N E T I C D I S O R D E R S

A way to prevent heat-induced sudden death?Mutations that affect the muscle Ca2+ release channel ryanodine receptor 1 (RYR1) are associated with life- threatening responses to elevated temperatures. Using mice with a mutant version of Ryr1, Lanner et al. showed that AICAR (5-aminoimidazole-4-carboxamide ribonucleotide) protected mice against excessive heat responses. Rather than working through its known mechanism of targeting AMP-activated protein kinase, AICAR directly inhibited RYR1-mediated Ca2+ leak from the sarcoplasmic reticulum and reduced oxidative or nitrosative stress, which could otherwise lead to sustained muscle contractions.ORIGINAL RESEARCH PAPER Lanner, J. T. et al. AICAR prevents heat-induced sudden death in RyR1 mutant mice independent of AMPK activation. Nature Med. 18, 244–251 (2012)

PA R A S I T E I N F E C T I O N

Understanding drug mechanisms of actionThe mechanism of action of drugs currently used to treat African trypanosomiasis is largely unknown. Alsford et al. used genome-scale RNA interference target sequencing screens in Trypanosoma brucei to identify the genes that contribute to the action of currently used drugs. In addition to known drug transporters, they linked over 50 genes to drug action; for example, a bloodstream-stage-specific invariant surface glycoprotein (ISG75) family was found to mediate the uptake of suramin. These findings could aid the rational design of new therapies for African trypanosomiasis and help to combat drug resistance.ORIGINAL RESEARCH PAPER Alsford, S. et al. High-throughput decoding of antitrypanosomal drug efficacy and resistance. Nature 482, 232–236 (2012)

A N T I C A N C E R D R U G S

Angiogenesis drug does not improve chemotherapyIt is generally thought that anti-angiogenic drugs such as bevacizumab temporarily normalize abnormal tumour vasculature and lead to an improved efficacy of subsequent chemotherapy. However, this study showed that bevacizumab did not improve the delivery of a chemotherapy drug to tumours. Positron emission tomography studies in patients with non-small-cell lung cancer determined that bevacizumab reduced the perfusion and net influx rate of docetaxel to tumours. These effects occurred within 5 hours and persisted after 4 days, highlighting the need for further studies to optimize the scheduling of anti-angiogenic drugs.ORIGINAL RESEARCH PAPER Van der Veldt, A. A. M. et al. Rapid decrease in delivery of chemotherapy to tumors after anti-VEGF therapy: implications for scheduling of anti-angiogenic drugs. Cell 21, 82–91 (2012)

IN BRIEF

R E S E A R C H H I G H L I G H T S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012

© 2012 Macmillan Publishers Limited. All rights reserved

Over the past 60 years, there have been major advances in many of the scientific and technological inputs into drug research and development (R&D). For example, combinatorial chemistry increased the number of drug-like molecules that could be synthesized per chemist per year by perhaps 800-fold during the 1980s and 1990s1–3, and greatly increased the size of chemical libraries4. DNA sequencing has become over a billion times faster since the first genome sequence was determined in the 1970s5–7, aiding the identification of new drug targets. It now takes at least three orders of magni-tude fewer man-hours to calculate three-dimensional protein structure via X-ray crystallography than it did 50 years ago8,9, and databases of three-dimensional protein structure have 300 times more entries than they did 25 years ago9 (see the RCSB Protein Data Bank database website), facilitating the identification of improved lead compounds through structure-guided strategies. High-throughput screening (HTS) has resulted in a tenfold reduction in the cost of testing compound libraries against protein targets

since the mid-1990s10. Added to this are new inventions (such as the entire field of biotechnology, computational drug design and screening, and transgenic mice) and advances in scientific knowledge (such as an understanding of disease mechanisms, new drug targets, biomarkers and surrogate end points).

There have also been substantial efforts to understand and improve the management of the commercial R&D process. Experience has accumulated on why projects overrun11, on the factors that influence financial returns on R&D investment12–17, on project success18 and R&D portfolio manage-ment19–22, on how to reduce costs by outsourcing, and on what is likely to impress or worry the regulatory authorities23.

However, in parallel — as many have discussed — R&D efficiency, measured simply in terms of the number of new drugs brought to market by the global bio-technology and pharmaceutical industries per billion US dollars of R&D spending, has declined fairly steadily24. We call this trend ‘Eroom’s Law’, in contrast to the more

familiar Moore’s Law (‘Eroom’s Law’ is ‘Moore’s Law’ backwards). Moore’s Law is a term that was coined to describe the expo-nential increase in the number of transistors that can be placed at a reasonable cost onto an integrated circuit. This number doubled every 2 years from the 1970s to 2010. The term is used more generally for technolo-gies that improve exponentially over time. The data in FIG. 1a show that the number of new US Food and Drug Administration (FDA)-approved drugs per billion US dol-lars of R&D spending in the drug industry has halved approximately every 9 years since 1950, in inflation-adjusted terms. Part of the contrast between Moore’s Law and Eroom’s Law is related to the complexity and limited current understanding of biological systems versus the relative simplicity and higher level of understanding of solid-state physics25 but, as discussed below, there are other important causes.

Although there are difficulties in making like-for-like comparisons in R&D spending over very long periods, Eroom’s Law has been fairly robust. The number of new drugs introduced per year has been broadly flat over the period since the 1950s, and costs have grown fairly steadily24. The slope of the line, over 10-year periods at least, does not change substantially (FIG. 1b), and assumptions about the delay between R&D investment and drug approval do not have a substantial influence on the overall pattern (FIG. 1c). For more details of the data used for FIG. 1, and the major assumptions made, see Supplementary information S1 (table).

Eroom’s Law indicates that powerful forces have outweighed scientific, technical and managerial improvements over the past 60 years, and/or that some of the improve-ments have been less ‘improving’ than com-monly thought. The more positive anyone is about the past several decades of progress, the more negative they should be about the strength of countervailing forces. If someone is optimistic about the prospects for R&D today, they presumably believe the countervailing forces — whatever they are — are starting to abate, or that there has been a sudden and unprecedented acceleration in scientific, technological or managerial progress that will soon become visible in new drug approvals.

O P I N I O N

Diagnosing the decline in pharmaceutical R&D efficiencyJack W. Scannell, Alex Blanckley, Helen Boldon and Brian Warrington

Abstract | The past 60 years have seen huge advances in many of the scientific, technological and managerial factors that should tend to raise the efficiency of commercial drug research and development (R&D). Yet the number of new drugs approved per billion US dollars spent on R&D has halved roughly every 9 years since 1950, falling around 80‑fold in inflation‑adjusted terms. There have been many proposed solutions to the problem of declining R&D efficiency. However, their apparent lack of impact so far and the contrast between improving inputs and declining output in terms of the number of new drugs make it sensible to ask whether the underlying problems have been correctly diagnosed. Here, we discuss four factors that we consider to be primary causes, which we call the ‘better than the Beatles’ problem; the ‘cautious regulator’ problem; the ‘throw money at it’ tendency; and the ‘basic research–brute force’ bias. Our aim is to provoke a more systematic analysis of the causes of the decline in R&D efficiency.

PERSPECTIVES

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 191

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

b Rate of decline over 10-year periods

Log

(dru

gs p

er b

illio

n U

S$)*

–1.0

–0.5

0

0.5

1.0

1.5

2.0

1950 1960 1970 1980 1990 2000 2010

c Adjusting for 5-year delay in spending impact

Num

ber o

f dru

gs p

er

billi

on U

S$ R

&D

spe

ndin

g*

0

1

10

100

1950 1960 1970 1980 1990 2000 2010

a Overall trend in R&D efficiency (inflation-adjusted)

Num

ber o

f dru

gs p

er b

illio

n U

S$ R

&D

spe

ndin

g*

0.1

1.0

10

100

1950 1960 1970 1980 1990 2000 2010

Drugs per billion US$ R&D spending 5 years previously

Drugs per billion US$R&D spending

FDA tightensregulationpost-thalidomide

First wave ofbiotechnology-derived therapies

FDA clears backlogfollowing PDUFAregulations plus smallbolus of HIV drugs

The magnitude and duration of Eroom’s Law also suggests that a lot of the things that have been proposed to address the R&D pro-ductivity problem are likely, at best, to have a weak effect. Suppose that we found that it cost 80 times more in real terms to extract a tonne of coal from the ground today than it did 60 years ago, despite improvements in mining machinery and in the ability of geologists to find coal deposits. We might expect coal industry experts and executives to provide

explanations along the following lines: “The opencast deposits have been exhausted and the industry is left with thin seams that are a long way below the ground in areas that are prone to flooding and collapse.” Given this analysis, people could probably agree that continued investment would be justified by the realistic prospect of either massive improvements in mining technology or large rises in fuel prices. If neither was likely, it would make financial sense to do less digging.

However, readers of much of what has been written about R&D productivity in the drug industry might be left with the impression that Eroom’s Law can simply be reversed by strategies such as greater man-agement attention to factors such as project costs and speed of implementation26, by reorganizing R&D structures into smaller focused units in some cases27 or larger units with superior economies of scale in others28, by outsourcing to lower-cost countries26, by adjusting management metrics and introducing R&D ‘performance score-cards’29, or by somehow making scientists more ‘entrepreneurial’30,31. In our view, these changes might help at the margins but it feels as though most are not addressing the core of the productivity problem.

There have been serious attempts to describe the countervailing forces or to understand which improvements have been real and which have been illusory. However, such publications have been relatively rare. They include: the FDA’s ‘Critical Path Initiative’23; a series of prescient papers by Horrobin32–34, arguing that bottom-up science has been a disappointing distraction; an article by Ruffolo35 focused mainly on regulatory and organizational barriers; a history of the rise and fall of medical inno-vation in the twentieth century by Le Fanu36; an analysis of the organizational challenges in biotechnology innovation by Pisano37; critiques by Young38 and by Hopkins et al.39, of the view that high-affinity binding of a single target by a lead compound is the best place from which to start the R&D process; an analysis by Pammolli et al.19, looking at changes in the mix of projects in ‘easy’ versus ‘difficult’ therapeutic areas; some broad-ranging work by Munos24; as well as a handful of other publications.

There is also a problem of scope. If we compare the analyses from the FDA23, Garnier27, Horrobin32–34, Ruffolo35, Le Fanu36, Pisano37, Young38 and Pammolli et al.19, there is limited overlap. In many cases, the differ-ent sources blame none of the same counter-vailing forces. This suggests that a more integrated explanation is required.

Seeking such an explanation is important because Eroom’s Law — if it holds — has very unpleasant consequences. Indeed, financial markets already appear to believe in Eroom’s Law, or something similar to it, and the impact is being seen in cost-cutting measures implemented by major drug com-panies. Drug stock prices indicate that inves-tors expect the financial returns on current and future R&D investments to be below the cost of capital at an industry level40, and

Figure 1 | Eroom’s Law in pharmaceutical R&D. a | The number of new drugs approved by the US Food and Drug Administration (FDA) per billion US dollars (inflation‑adjusted) spent on research and development (R&D) has halved roughly every 9 years. b | The rate of decline in the approval of new drugs per billion US dollars spent is fairly similar over different 10‑year periods. c | The pattern is robust to different assumptions about average delay between R&D spending and drug approval. For details of the data and the main assumptions, see Supplementary information S1 (table) and REFS 24,86,87. Note that R&D costs are based on the Pharmaceutical Research and Manufacturers of America (PhRMA) Annual Survey 2011 (REF. 86) and REF. 87. PhRMA is a trade association that does not include all drug and biotechnology companies, so the PhRMA figure understates R&D spending at an industry level. The total industry expenditure since 2004 has been 30–40% higher than the PhRMA members’ total expenditure, which formed the basis of this figure. The new drug count, however, is the total number of new molecular entities and new biologics (applying the same definition as Munos24) approved by the US FDA from all sources, not just PhRMA members. We have estimated real‑term R&D cost inflation figures from REFS 24,87. The overall picture seems to be fairly robust to the precise details of cost and inflation calculations. Panel a is based on a figure that origi‑nally appeared in a Bernstein Research report (The Long View — R&D productivity; 30 Sep 2010). *Adjusted for inflation. PDUFA, Prescription Drug User Fee Act.

P E R S P E C T I V E S

192 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

would prefer less R&D and higher dividends. Investors may well be wrong about this. However, they have less reason to be biased towards optimism about the likelihood of Eroom’s Law being successfully counteracted than those who are working in the industry, or those who sell consulting services to the industry. Shareholders ultimately appoint executives and control resource allocation, so their perceptions matter. It is likely that Pfizer, Merck & Co., AstraZeneca and Eli Lilly will be spending less — in nominal terms — in 2015 than they did in 2011, partly in response to shareholder pressure. Across the top ten large pharmaceutical companies, it seems that nominal R&D spending will be flat until 2015, which represents a decline in real terms. More importantly, the combined effect of declining real-term R&D spending with Eroom’s Law (fewer new drugs per billion US dollars of R&D investment over time) is that there will be fewer new drugs and/or drugs will become inordinately expensive. This will threaten the huge benefits41,42 that follow from the availability of effective and affordable new drugs.

In our view, avoiding such an outcome requires a more systematic analysis of the factors that underlie Eroom’s Law. We think that any serious attempt to explain Eroom’s Law should try to address at least two things: the progressive nature of the decline in the number of new drugs per billion US dollars of R&D spending, and the scale (~80-fold) of the decline. In this article, we make some suggestions. We realize that the industry is heterogeneous, so our gen-eralizations will be wrong in many cases. We appreciate the intellectual effort that has been made by others on analysing the prob-lems of R&D productivity. We note that our chosen measure of R&D efficiency is based on cost per new drug approved. This does not account for the huge variation in the medical and financial value of new drugs. A few breakthrough drugs — for example, a highly effective Alzheimer’s disease treat-ment — could have much greater medical and financial value than a larger number of new drugs that provide only modest incre-mental benefits. We also note that the very long cycle time for drug R&D means that our productivity measure is a lagging indicator; perhaps things have improved, but the result is not yet visible.

However, with the aim of prompting debate and analysis, here we discuss what we consider to be the four primary causes of Eroom’s Law: the ‘better than the Beatles’ problem; the ‘cautious regulator’ problem; the ‘throw money at it’ tendency; and the

‘basic research–brute force’ bias. There may also be some contribution from a fifth factor, termed ‘the low-hanging fruit’ problem, but we consider this to be less important.

Primary causesThe ‘better than the Beatles’ problem. Imagine how hard it would be to achieve commercial success with new pop songs if any new song had to be better than the Beatles, if the entire Beatles catalogue was available for free, and if people did not get bored with old Beatles records. We suggest something similar applies to the discovery and development of new drugs. Yesterday’s blockbuster is today’s generic. An ever-improving back catalogue of approved medicines increases the com-plexity of the development process for new drugs, and raises the evidential hurdles for approval, adoption and reimbursement. It deters R&D in some areas, crowds R&D activity into hard-to-treat diseases and reduces the economic value of as-yet-undiscovered drugs. The problem is progressive and intractable.

Few other industries suffer from this problem. In the example of the coal indus-try noted above, the opencast deposits are mined first. However, the coal is burnt, which increases the value of the coal that is still in the ground. In most intellectual property businesses (for example, fashion or publishing), people get bored with last year’s creations, which maintains demand for novelty. Unfortunately for the drug industry, doctors are not likely to start prescribing branded diabetes drugs because they are bored with generic metformin.

Anti-ulcerants — still a very valuable therapeutic area in terms of revenues — pro-vide an example of the shadow that is cast by successful drugs. A class of anti-acid agents known as potassium-competitive acid block-ers, such as soraprazan (now discontinued), would probably be safe and effective anti-ulcerants, and 15 years ago they could have been blockbusters. The problem today is that there are now two classes of highly effective and safe anti-ulcer drugs on the market: the histamine H2 receptor antagonists (for example, generic ranitidine, which is avail-able over the counter) and the proton pump inhibitors (for example, generic esomepra-zole and several others). Any sensible health-care system would only pay for patients to receive a new branded potassium-competitive acid blocker if they fail to respond to a cheap generic proton pump inhibitor and/or H2 receptor antagonist, but such patients are a very small proportion of the total

population. This general problem applies in diabetes, hypertension, cholesterol management and many other indications.

Pammolli et al.19 have provided a quan-titative illustration of the ‘better than the Beatles’ problem. Their analysis compared R&D projects started between 1990 and 1999 with those started between 2000 and 2004. Attrition rates rose during the latter period. However, the increase could be largely explained by a shift in the mix of R&D projects from commercially crowded therapeutic areas in which historic drug approval probabilities were high (for example, genitourinary drugs and sex hormones) to less crowded areas with lower historical approval probabilities (for example, antineo-plastics and immunomodulatory agents).

There is another related potential cause of Eroom’s Law that has frequently been put forward, termed the ‘low-hanging fruit’ problem, which results from the progressive exploitation of drug targets that are more technically tractable43. To be clear, the ‘low-hanging fruit’ problem argues that the easy-to-pick fruit has gone, whereas the ‘better than the Beatles’ problem argues that the fruit that has been picked reduces the value of the fruit that is left in the tree.

In our opinion, the ‘low-hanging fruit’ problem is less important than the ‘better than the Beatles’ problem. First, estimates of the number of potential drug targets44,45 versus the number of drugged targets46 sug-gest that many decades-worth of new targets remain if the industry continues to exploit four or five new targets each year. It is also becoming clear that many drugs may derive their therapeutic benefit from interactions with multiple proteins rather than a single target. These drugs are ‘magic shotguns’ rather than ‘magic bullets’47. If this turns out to be more generally true, then worrying about the ‘low-hanging fruit’ problem would be similar to worrying that a shortage of notes is threatening the future of music composition. In our view, the ‘low-hanging fruit’ explanation is sometimes tautological as ‘technically easy’ tends to be equated with ‘already discovered’48. Indeed, investigation of the history of drug discovery suggests that there was little easy or obvious about some of the great discoveries of the 1940s and 1950s, such as the anti-inflammatory effects of corticosteroids, the psychiatric effects of imipramine or lithium, or the antibacterial properties of penicillin36,49–51.

The ‘cautious regulator’ problem. Progressive lowering of the risk tolerance of drug regu-latory agencies obviously raises the bar for

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 193

© 2012 Macmillan Publishers Limited. All rights reserved

the introduction of new drugs, and could substantially increase the associated costs of R&D52. Each real or perceived sin by the industry, or genuine drug misfortune, leads to a tightening of the regulatory ratchet, and the ratchet is rarely loosened, even if it seems as though this could be achieved without causing significant risk to drug safety. For example, the Ames test for mutagenicity may be a vestigial regulatory requirement; it prob-ably adds little to drug safety but kills some drug candidates. Furthermore, for most of the past 60 years large and sophisticated drug companies may not have been disappointed to see the regulatory ratchet tighten because it reduced competition.

It also seems that the concern that drug companies could cheat the system in some way has led the cautious regulator to apply an audit-based approach to regulatory documentation, as the more demanding the reporting requirements are, the harder it is to cheat without leaving some kind of error or inconsistency in what is reported. The scale of reporting was summarized recently by the Chief Scientific Officer of Novo Nordisk in the company’s third quarter 2011 results conference call with respect to the submis-sion to the FDA of data on two new insulin therapies: “If printed and stacked, the many million pages of documentation, with a total of 9 million electronic links, [would] exceed the height of [the] Empire State Building.”

The impact of the ‘cautious regulator’ problem on Eroom’s Law is apparent in FIG. 1. First, it shows R&D efficiency dipping in the early 1960s following the 1962 Kefauver Harris Amendment to the Federal Food, Drug, and Cosmetic Act, which was introduced in the wake of the thalidomide drug safety disaster. For the first time, medi-cines had to demonstrate efficacy, and the safety hurdles were also raised. This reduced financial returns on R&D for a decade or so12,14, before rising drug prices outstripped R&D cost inflation and increased financial returns in the 1970s15. Interestingly, FIG. 1 also shows a rise in R&D efficiency in the mid to late 1990s, which is likely to be due to two regulatory factors: primarily the clearing of a regulatory backlog at the FDA following the implementation of the 1992 Prescription Drug User Fee Act (PDUFA), but also a small contribution from the rapid development and approval of several HIV drugs. In the case of HIV drugs, organized and politically astute lobbying effectively lowered the normal regulatory hurdles53.

The ‘cautious regulator’ problem fol-lows, in part, from the ‘better than the Beatles’ problem, as the regulator is more

risk-tolerant when few good treatment options exist; or, put another way, the avail-ability of safe and effective drugs to treat a given disease raises the regulatory bar for other drugs for the same indication. Although the ‘cautious regulator’ problem is tractable in principle, it is hard to see the regulatory environment relaxing to any great extent. Society may be right to prefer a tougher regulator, even if it means more costly R&D. Drug safety matters. And although the 1950s and 1960s may be viewed by some as a golden age in terms of thera-peutic innovation36,48,54, it seems unlikely that the severe adverse outcomes for many patients taking part in clinical trials during this period36 would be acceptable today.

The ‘throw money at it’ tendency. The ‘throw money at it’ tendency is the tendency to add human resources and other resources to R&D, which — until recent years — has generally led to a rise in R&D spending in major companies, and for the industry overall. It is probably due to several factors, including good returns on investment in R&D for most of the past 60 years, as well as a poorly understood and stochastic innova-tion process that has long pay-off periods. In addition, intense competition between marketed drugs (where being second or third to launch is often worth less than being first) provides a rationale for investing additional resources to be the first to launch. There may also be a bias in large companies to equate professional success with the size of one’s budget.

Unfortunately for people working in R&D today, tackling the ‘throw money at it’ tendency looks feasible. Investors and many senior executives think that a lot of costs can be cut from R&D without reducing output substantially. Whether this is correct remains to be seen, although if so, it may be the single strategy most likely to counteract Eroom’s Law in the short term. The risk, however, is that the lack of under-standing of factors affecting return on R&D investment that contributed to relatively indiscriminate spending during the good times could mean that cost cutting is simi-larly indiscriminate. Costs may go down, without resulting in a substantial increase in productivity.

The ‘basic research–brute force’ bias. The ‘basic research–brute force’ bias is the ten-dency to overestimate the ability of advances in basic research (particularly in molecular biology) and brute force screening methods (embodied in the first few steps of the

standard discovery and preclinical research process) to increase the probability that a molecule will be safe and effective in clinical trials (FIG. 2). We suspect that this has been the intellectual basis for a move away from older and perhaps more productive methods for identifying drug candidates32–34. It should be noted that many of our com-ments are more relevant to small-molecule drugs, although the data used for FIG. 1 also include biologics.

FIGURE 2 illustrates the standard model of most drug R&D. It is — effectively — a serial search, filter and selection process. Scientific and technical advances have, superficially at least, increased the breadth of the funnel (that is, more potential targets have been identified, and more drug-like molecules have been synthesized). They have improved the filtering efficiency by several orders of magnitude (for example, HTS versus testing in expensive and low-throughput animal models). They should also have increased the quality of filtering and selection (for example, the use of pathway analysis for tar-get selection, the use of transgenic mice for target validation and the use of accumulated experience to favour molecules that would be likely to have good ADMET (absorption, distribution, metabolism, excretion and toxicology) characteristics).

The cumulative effect of improvements in these early steps should have been a higher signal-to-noise ratio among drug candidates entering clinical trials; that is, the candidates chosen should have had a greater likelihood of successfully demonstrating effective-ness and safety in these trials. This, in turn, should have increased R&D efficiency, given that most of the costs of new drug development are related to the costs of failed projects22. Yet the probability that a small-molecule drug successfully completes clini-cal trials has remained more or less constant for 50 years21, and overall R&D efficiency has declined24.

So how can some parts of a process improve dramatically, yet important meas-ures of overall performance remain flat or decline? There are several possible explana-tions, but it seems reasonable to wonder whether companies industrialized the wrong set of activities34,36,38. At first sight, R&D was more efficient several decades ago (FIG. 1), when many research activities that are today regarded as critical (for example, the derivation of genomics-based drug targets and HTS) had not been invented, and when other activities (for example, clinical science, animal-based screens and iterative medicinal chemistry) dominated.

P E R S P E C T I V E S

194 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Decline in approved drugs per billion US$ spent on R&D

Nature Reviews | Drug Discovery

Targetidentification

Targetvalidation

Huge apparent improvements in efficiency and quality in many research inputs:• Approximate Moore’s Law improvements in many cases• Qualitative improvements in other cases

Small changes in success ofmolecules entering clinicaltrials over the past 50 years

Eroom’s Law:increase in cost perapproved molecule

Targetto hit

Hit tolead

Leadoptimization Preclinical Clinical trials

(Phase I, II, III)

There have been several interesting critiques of modern research33,48,55, but here we highlight two potential problems. First, much of the pharmaceutical industry’s R&D is now based on the idea that high-affinity binding to a single biological target linked to a disease will lead to medical benefit in humans39. However, if the causal link between single targets and disease states is weaker than commonly thought38,56, or if drugs rarely act on a single target, one can understand why the molecules that have been delivered by this research strategy into clinical development may not necessarily be more likely to succeed than those in earlier periods.

Indeed, drug-like small molecules tend to bind promiscuously, and this sometimes turns out to have an important role in their efficacy47,57 as well as their so-called off-target effects39. Targets are parts of complex networks leading to unpredictable effects58, and biological systems show a high degree of redundancy, which could blunt the effects of highly targeted drugs56,57. Perhaps this helps to explain why the R&D process was more cost-effective several decades ago (FIG. 2), when expensive labour-intensive animal models — rather than cheap automated molecular assays — formed the basis of initial drug screening36,49–51,59.

More recent analysis also points to a similar conclusion. More first-in-class small-molecule drugs approved between 1999 and 2008 were discovered using phenotypic assays than using target-based assays60. Drugs approved during this period would have been discovered when screening activity was dominated by the target-based approach, so one might have expected more target-based discoveries. Perhaps

target-based approaches are efficient for pursuing validated therapeutic hypotheses but are less effective in the search for innova-tive drugs that have a better chance of clearing the ‘better than the Beatles’ barrier.

The second potential problem follows from the nature of chemical space and a shift from iterative medicinal chemistry coupled with parallel assays (pre-1990s) to serial filtering that begins with HTS of a static compound library against a target. Directed iteration — even if each cycle is slow — may be a much more efficient way of searching a large and high-dimensional chemical space than fast HTS of a predefined collection of compounds (BOX 1).

As an aside, biologics have had a higher success rate than small molecules once they leave research and enter clinical trials. There was an approximately 32% approval rate for biologics versus an approximately 13% approval rate for small-molecule drugs first tested in humans between 1993 and 2004 (REF. 21). This may not be surprising for copies or close analogues of endogenous signalling molecules (for example, insulins, erythropoietins or growth hormones) or for agents that replace dysfunctional proteins (for example, clotting factors, lysosomal enzymes, and so on). The high rates of success in clinical trials of monoclonal anti bodies (and related fusion proteins) is perhaps more notable61. One might expect them to suffer from the same kind of prob-lems with single-target efficacy as small mol-ecules (albeit with fewer off-target effects). However, they have opened up new sets of therapeutic targets, which may suffer less from the ‘better than the Beatles’ problem. Perhaps their success is also a function of their limited target set — either cell surface

proteins or protein-based extracellular signalling molecules. In both cases, the chain of causality between target binding and therapeutic effect is relatively short. Out of 34 monoclonal antibodies or other targeted biologics (such as fusion proteins or aptamers) that have been approved by the FDA, 13 target white blood cell-specific antigens (for example, CD20) and are used for haematological cancers or immunosup-pression; three target receptors in the human epidermal growth factor receptor family and are used in oncology; seven target tumour necrosis factor or interleukins and are used for immunomodulation in autoimmune diseases; and four target vascular endothelial growth factor variants and are used in oncol-ogy or ophthalmology.

In our view, there are several reasons why the ‘basic research–brute force’ bias has come to dominate drug research. First, by the early 1980s there was already a sense that the pace of pharmaceutical innovation was slowing. The ‘cautious regulator’ prob-lem was an obvious drag52,54,62. The ‘better than the Beatles’ problem was starting to emerge, with complaints that new drugs had only modest incremental benefit over what was already available62. There were concerns about the ‘low-hanging fruit’ problem, with a growing sense that the industry had started to run out of good animal models to screen drugs for still poorly treated diseases52,62.

Second, the ‘basic research–brute force’ bias matched the scientific zeitgeist48, par-ticularly as the older approaches for early-stage drug R&D seemed to be yielding less. What might be called ‘molecular reduction-ism’ has become the dominant stream in biology in general, and not just in the drug

Figure 2 | How can some parts of the R&D process improve, yet the overall efficiency decline? Dramatic improvements in brute force screening methods and basic science should have tended to increase the efficiency of the research process (more leads tested against more tar‑gets, at a lower cost; shown in gold) and raised its quality (better targets as disease pathways and mechanisms are understood, better leads that avoid old mistakes surrounding ADMET (absorption, distribution, metab‑olism, excretion and toxicity) characteristics, and so on). This, in turn,

should have increased the probability that molecules would succeed in the clinic (shown in red), which in turn should have increased overall efficiency, as research and development (R&D) costs are dominated by the cost of failure. However, the probability that a small molecule successfully completes clinical trials has remained more or less constant for 50 years21, whereas overall R&D efficiency has declined24. One pos‑sible explanation for this is that much of the industry industrialized and ‘optimized’ the wrong set of R&D activities.

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 195

© 2012 Macmillan Publishers Limited. All rights reserved

industry33,34,55: “Since the 1970s, nearly all avenues of biomedical research have led to the gene”63. Genetics and molecular biology are seen as providing the ‘best’ and most fundamental ways of understanding biologi-cal systems, and subsequently intervening in them64. The intellectual challenges of reductionism and its necessary synthesis (the ‘-omics’) appear to be more attractive to many biomedical scientists than the messy empiricism of the older approaches.

Third, the ‘basic research–brute force’ bias matched the inclination of many com-mercial managers, management consult-ants and investors. The old model, based on iterative medicinal chemistry, animal-based screening and clinical science was seen as “too dependent on either inefficient trench-warfare type of slog or the unpre-dictable emergence of seemingly capricious geniuses like James Black, Paul Janssen, Daniel Bovet, Gertrude Elion, or Gerald Hitchings”33. Automation, systematization and process measurement have worked in other industries. Why let a team of chemists and biologists go on a trial and error-based search of indeterminable duration, when one could quickly and efficiently screen mil-lions of leads against a genomics-derived target, and then simply repeat the same industrial process for the next target, and the next? In the early 1990s, few companies thought they could thrive or survive without moving towards a drug discovery process based on HTS and the products of the human genome.

Here, we are reminded of a debate25 about improving clinical trial efficiency, triggered by an editorial by Andy Grove65, the former Chief Executive of Intel — a man with personal experience of Moore’s Law. Grove noted the “disappointing output” of R&D in the drug industry and made suggestions to radically change clinical trials by making more use of electronic health data65. Some biomedical scientists probably find Grove’s intervention irritating, given the simplicity and predictability of semiconductor physics versus “biology’s mysteries”25. However, shareholders and taxpayers have been persuaded to fund a lot of R&D because biomedical scientists (and drug industry executives) have told them that — thanks to molecular reductionism — it would soon become more predictable63, more productive and less mysterious.

We think that the ‘basic research–brute force’ bias is supported by survivor bias among R&D projects. This makes drug discovery and development sound more pro-spectively rational than it really is. Nearly all

Box 1 | Directions in small-molecule drug discovery

The 1990s saw a major shift in small-molecule drug discovery strategies, from iterative low-throughput in vivo screening and medicinal chemistry optimization to target-based high-throughput screening (HTS) of large compound libraries. At first sight, the former is slow and expensive in terms of the number of compounds that can be tested, whereas the latter is fast and cheap59. However, the topography of chemical space and the nature of industrialized drug discovery may conspire to make the second approach less productive. The problem is not necessarily HTS per se (the pros and cons of which are actively debated79); rather, it may be the research processes that new technologies helped to cement.

First, real-world compound libraries for HTS cover infinitesimally small and somewhat redundant regions of chemical space, which is vast; it has been suggested that there could be between 1026 and 1062 (REFS 80,81) chemotypes that would comply with the Lipinski guidelines for oral drugs82, and each chemotype has a large number of potential derivatives. By contrast, a typical corporate screening collection for HTS contains around 106 chemical entities and perhaps 103 chemotypes. Furthermore, mergers have revealed that different companies’ compound libraries often substantially overlap. This is not surprising: companies generated their libraries in similar ways, as they used clustered sets of molecules from similar historical campaigns; there is a limited set of commercially available reagents; and a relatively small number of reactions are amenable to high-throughput automated synthesis.

Second, it has proved to be difficult to design systems that reward people for producing ‘good’ hits and leads rather than ‘more’ hits and leads. Collections are biased towards developable compounds with acceptable ADME (absorption, distribution, metabolism and excretion) characteristics. Companies want measurable developability benchmarks. There are few immediate prizes for chemical or biological novelty. The pre-selection and pre-design of screening collections means that the lead structures are largely foreseen. It provides no easy way to jump from local chemical optima to something better.

Third, the process to whittle down a few thousand HTS hits into a couple of qualified leads has been dominated by molecules that win on potency measures. Selection is based on serial assays, with most molecules failing at each step. There is no practical way to view the full biological profile of all hits at an early stage. Hits with merely adequate target potency but with other potentially attractive features (such as good ADME, other interesting biological properties, and so on) could be thrown away. This further focuses the search process on small parts of screening collections. It may even focus the search process on a suboptimal part of the screening collection. Recent research suggests that there is a negative correlation between in vitro potency and desirable ADME and toxicology83. Given these features of HTS in the real world, we should expect different drug companies to produce similar molecules for a given target. We should also expect these molecules to reflect local optima within the screening collections, rather than global optima from the much larger chemical universe.

Before the 1990s, however, the standard approach for small-molecule drug discovery involved synthesizing and screening a relatively small number of compounds. There would be a few tens of molecules (often fewer) in active assessment at any one time, and perhaps 1,000 molecules synthesized by a team of chemists during a 5-year project. The search usually started with a molecule that was known, or suspected, to have promising pharmacology but perhaps with poor ADME characteristics: adrenaline led to the development of beta blockers, and histamine lead to the development of cimetidine. Phenomenological screening was also used, to a small extent, to provide starting points. Each molecule was then assessed in a range of concurrent assays (often in vivo59, considering potency, ADME, toxicity, selectivity and so on). Molecules were then modified (or discarded) depending on the results of the assays. The cycle was repeated, with the biological results being used to establish structure–activity relationships for each assay and thus advance the structures of lead compounds through the chemical space until one or two compounds met the multiple criteria necessary for progression into clinical trials. Unlike the screening case, after a few iterations one had compounds specifically customized to a particular target, with structures that would not have been foreseen at the start of the process. This approach prevented trial compounds from being confined to minor local optima. It facilitated what Sir James Black called “obliquity”84 — the art of looking for one thing and finding something else. It made it less likely that competitors had identical drugs. Remarkably, the search for blockbuster drugs using this method was often achieved with fewer than 1,000 compounds.

This is a profoundly different search strategy to the one that was industrialized, but one that may be more efficient when there is a very large number of items arranged in a high-dimensional space, as is the case with drug-like molecules (see Supplementary information S2 (box)). This is because it is possible to traverse large regions of a high-dimensional space with a small number of steps85, whereas any static, predefined compound library will cover only a tiny part of the chemical space. Perhaps this is part of the explanation of the pre-1990s productivity? These kinds of arguments are not lost on the drug industry. Efforts are underway to try to combine some of the obvious advantages of HTS with the advantages of small teams dedicated to a broader exploration of the biological profiles of a set of evolving lead compounds. The idea is to analyse several structure–activity relationships in parallel (for example, potency at the target, potency at likely toxicity sites, potency in cellular assays, in vivo ADME) to direct rapid, sometimes automated, iterative chemistry.

P E R S P E C T I V E S

196 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

drugs are sold with a biological story that sounds like molecular reductionism and that sometimes, but not always, turns out to be true: for example, “drug x works by binding receptor a, which influences pathway b, which adjusts physiological process c, which alleviates disease d.” Such stories get confused with prediction because we hear very little about the vast majority of the other projects that were also initiated on the basis of high-affinity binding of a plausible candidate to a plausible target, and that had similarly plausible biological stories until the point at which they failed in development for unexpected reasons.

It would be interesting to see how well prospective estimates of plausibility cor-related with subsequent attrition. This point is illustrated by the anticancer drug iniparib. Attendees of the 2010 meeting of the American Society of Clinical Oncology (ASCO), or readers of the New England Journal of Medicine66, could have been forgiven for believing that iniparib had a spectacular effect on metastatic breast can-cer in a Phase II trial because it inhibited a specific target, poly(ADP-ribose) polymer-ase 1 (which is involved in DNA repair), and therefore potentiated chemotherapy. However, the following year, Phase III trial results presented at the 2011 ASCO meeting indicated that iniparib did not work very well in breast cancer67, and it did not seem to inhibit poly(ADP-ribose) polymerase 1 very much either68.

Fortunately, the ‘basic research–brute force’ issue is tractable in several ways. First, in a handful of therapeutic areas the research process does appear to be delivering better systems-level insights, better targets (or sets of targets) and better candidate drugs. Oncology is the most obvious example. It is hard to look at the genesis of drugs like crizotinib69, vemurafenib70 or vismodegib71 and think that one is simply looking at ran-dom survivors. Furthermore, in oncology the regulator is less cautious and the back catalogue of approved drugs is far from ‘Beatle-esque’. One or two other disease areas with simple genetics may perhaps resemble oncology. Second, more emphasis could be put on iterative approaches, on animal-based screening or even on early proof of clinical efficacy in humans, and less on the predic-tive power of high-affinity binding to the target of a molecule from a static library. Novartis is one company that is emphasizing proof-of-concept trials for drugs in rare diseases for which there is a high unmet need and a compelling match between the drug’s mode of action and the disease.

Only if there is success here does the company invest in more expensive trials in more com-mon diseases in which the mode of action may be more speculative, or in which the risk–benefit profile may be less clear. Third, in some therapeutic areas people could just stop believing in the current predictive ability of ‘basic research–brute force’ screening approaches, and resist the temptation to put molecules into clinical trials without having more compelling evidence of the validity of the underlying therapeutic hypothesis.

There is, of course, no way of going back in time to see how well more recent R&D approaches would have worked in the 1940s and 1950s. It is possible that research has become much better at delivering the right molecules into the clinic but that the improvements have been swamped by the ‘better than the Beatles’ problem, the ‘low-hanging fruit’ problem and the ‘cautious regulator’ problem.

Ironically however, if the industry really has been doing the right things, the ultimate prognosis may be bleaker. One can think of the opportunities for R&D in terms of a Venn diagram: as science and technology improve, some sets grow (for example, the set of drug-gable targets, the set of drug-like molecules and the set of drugged targets), whereas other sets shrink (for example, the set of economi-cally exploitable and still untreated diseases, or the set of acceptable off-target effects). It is obvious that R&D productivity could decline despite improvements in the inputs if the intersection that contained commercially attractive and approvable drug candidates shrunk. This idea is illustrated in FIG. 3, in which the notional set of validated targets grows between 1970 and 2010, but it does not grow fast enough to offset the growth in the set of targets that would either worry a cautious regulator or fail the ‘better than the Beatles’ test.

Finally, we note that it would be easier to improve the signal-to-noise ratio of drugs that enter clinical trials if: first, there was a detailed understanding of why drugs fail in the clinic; second, this led to the discov-ery of a small number of common failure modes; and third, this knowledge could be used to change the early stages of the R&D process. If it is impractical to carry out retro spective analyses on the precise molec-ular mechanisms of clinical trial failure, or if such retrospective analyses show that trials fail for many rare and idiosyncratic reasons, or if cycle times are so long that the lessons are obsolete by the time they are learned, then incremental improvement will be more difficult. Both the regulators23 and

the industry18 are interested in the analysis of failure but it receives less scrutiny than one might expect given its dominant role in the costs of R&D.

Secondary symptomsThe four proposed primary causes of Eroom’s Law discussed above have given rise to several ‘symptoms’ that tend to further increase costs, particularly the costs of clini-cal development. Some of these symptoms are highlighted below.

The narrow clinical search problem. The narrow clinical search problem is the shift from an approach that looked broadly for therapeutic potential in biologically active agents to one that seeks precise effects from molecules designed with a single drug target in mind. In the 1950s and 1960s, initial screening was typically performed in animals, not in vitro or in silico, and drug candidates were given in early stages of the development process to a range of physi-cians. Discovery involved, to an extent, the ability of physicians to spot patterns through careful clinical observation, especially in therapeutic areas in which symptomatic improvements are readily observable, such as psychiatry36,49–51. This is sometimes dis-missed as serendipity but the approach made it likely that new therapeutic effects would be detected. Even recently, it appears that many — perhaps most — new therapeutic uses of drugs have been discovered by moti-vated and observant clinicians working with patients in the real world72. Some drug com-panies, particularly smaller and mid-sized firms, recognize this opportunity and are active repositioners of existing drugs.

However, the ‘cautious regulator’ prob-lem and the ‘basic research–brute force’ bias have pushed most of the drug industry towards a narrow clinical search strategy. If a drug has an effect but this is not the precise effect that the trial designers antici-pated, then the trial fails. Opportunities for serendipity are actively engineered out of the system. Perhaps it is too risky to let bright doctors with large numbers of patients make broad clinical observations, or to let creative scientists rummage around in rich clinical data sets, in case they find something unexpected, which has to be explained to the cautious regulator who then kills the project. Modern multicentre trials tend to spread the patients so thinly that a doctor who did want to look for pat-terns might miss them. In Phase II trials — perhaps the best opportunity to spot new things — the average number of patients

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 197

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Validated targets

All potential drug targets

1970 2010

Targets that failthe ‘cautiousregulator’ test

Targets that failthe ‘better thanthe Beatles’ test

per multicentre trial site is now very small: between five and ten patients in oncology, central nervous system and respiratory disease trials73.

The big clinical trial problem. The first randomized controlled trial, published in 1948, recruited 109 patients and randomized 107 of them74. Between 1987 and 2001, the number of patients per pivotal trial for anti-hypertensive agents rose from around 200 to around 450 (REF. 75). Between 1993 and 2006, the average number of patients across the pivotal trials for a new oral antidiabetic drug rose from around 900 to over 4,000 (REF. 76). The first pivotal trial for Merck’s simvastatin (a cholesterol-lowering agent), published in 1994, recruited around 4,400 patients77. A pivotal trial for Merck’s anacetrapib, an investigational cholesterol-modulating agent intended to be used on top of drugs like simvastatin, is currently recruiting around 30,000 patients.

This expansion is a consequence of sev-eral factors. First, the ‘better than the Beatles’ problem increases trial size. Everything else being equal, clinical trial size should be inversely proportional to the square of the effect size. If the effect size halves, the trial has to recruit four times as many patients to have the same statistical power. The problem is that treatment effects on top of an already effective treatment are usually smaller than treatment effects versus placebo. Furthermore, Phase III trials have become a messy mixture of science, regulation, public relations and marketing. Trying to satisfy these multiple constraints tends to inflate their size and cost.

The best clinical trial to show efficacy would be something relatively small in a homogeneous patient sample recruited from as few centres as possible — the medical equivalent of a well-controlled experiment. But this tends to make the cautious regulator uneasy given variation in practice patterns and patients. What about rare side effects (the FDA has recently required post-marketing trials for long-acting bronchodilators in around 53,000 patients)? Small trials also make for bad marketing and, in the world of evidence-based medicine, poor market access. It is better to involve the senior doctors at the major centres. The number of principal investigators per drug in clinical trials has doubled over the past decade73. The consequence of this is multicentre trials that add noise and heterogeneity, and are therefore bigger and more expensive.

The multiple clinical trial problem. The ‘better than the Beatles’ problem has increased the complexity of medical practice. In some areas, where once there were only one or two treatment options, there is now a rich back catalogue. For example, the treatment of patients with type 2 diabetes was once a choice of insulin or diet and exercise, but can now involve a combination of drugs from around ten different drug classes: biguanides, thiazolidinediones, sul-fonylureas, meglitinides, alpha-glucosidase inhibitors, dipeptidyl peptidase 4 inhibitors, glucagon-like peptide 1 analogues, amylin analogues, long-acting and short-acting insulin analogues, as well as various human insulins and insulin mixes. Treatment for patients with colon cancer was once a choice

between surgical resection or palliative care, but now the National Comprehensive Cancer Network’s colon cancer treatment guidelines contain up to 100 pages of detailed treatment algorithms.

The cautious regulator is less prepared to assume that the safety and efficacy of new drugs can be generalized across such hetero-geneous and fragmented patient popula-tions. Cost-sensitive health-care funders are also wary. This means narrower indications and more clinical trials per drug. The first long-acting insulin analogue, glargine, was approved by the FDA in 1999 following three pivotal Phase III trials. The newest long-acting insulin analogue, degludec, was filed for regulatory approval in 2011 following 12 pivotal trials (and, as mentioned above, an Empire State Building’s worth of documen-tation). Some successful drugs in complex therapeutic areas appear to demand, over their life cycle, dozens of Phase III trials78.

The long cycle time problem. In the 1950s and 1960s, cycle times were remarkably short by modern standards. The regula-tor was less cautious and there was less molecular reductionism before agents were screened for efficacy in animal models and in patients. This sped up innovation. The first antidepressant, imipramine, was synthe-sized in around 1951. It was screened almost immediately in rats, and tested personally by a few scientists at the drug company Geigy51. It was then tested without much success in various patient groups in 1952, tested again in 1953, found to be problematic in patients with psychosis in 1954 and tried yet again in 1955 before it was identified as an anti-depressant in 1956. It completed preclinical development and had not one but three clin-ical cycles within 5 or 6 years. In 2005–2006, the typical period of time in clinical develop-ment for a new drug was over 9 years21. The biggest increase in development times came between the 1960s and the 1980s21.

An idea: the CDDOThis article is intended to provoke further analysis of the forces that have counter-vailed scientific, technical and managerial improvements over the past 60 years. We have avoided cures, partly because the ratio of published cures to diagnoses is already too high. We do, however, have one idea, which might also be viewed as a thought experiment.

We suggest that all large drug companies introduce a new board level role, which we call the Chief Dead Drug Officer (CDDO). This role would be focused on drug failure

Figure 3 | Venn diagram illustrating hypothetical headwinds to R&D efficiency. Research and development (R&D) efficiency could decline if scientific, technical and managerial improvements are offset by other factors. For example, R&D efficiency could be limited by the supply of validated targets that could be drugged without failing the ‘cautious regulator’ test and/or the ‘better than the Beatles’ test. In this hypothetical illustration, the increase in the number of validated targets between 1970 and 2010 is outweighed by increasing regulatory caution and an improving catalogue of approved drugs.

P E R S P E C T I V E S

198 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

at all stages of R&D, and the CDDO would have a fixed time — for example, 18 months — from appointment to compose a detailed report that aims to explain the causes of Eroom’s Law. This report would be submitted to the board of the company, included in the company’s annual report to shareholders, and would also be submitted for publication in a scientific journal and sent to organizations such as the FDA and the US National Institutes of Health. The remuneration for the role would be structured in such a way as to provide a strong incentive to provide an accurate forecast of the future R&D productivity of the company and the industry overall. For example, perhaps the salary could be relatively modest, but the CDDO could be eligible for an enormous bonus if their projections after a 10-year period are no more than 10% too optimistic or no more than 30% too pessimistic.

We like the idea for several reasons. First, the CDDO has no incentive to be irrationally optimistic. Second, R&D costs are dominated by the cost of failure73. Most molecules fail. Most research scientists spend most of their time on products that fail. It seems fitting that someone on the board should focus on the products that consume most of the R&D organization’s time, energy and money. Third, an expertise in drug failure should qualify the CDDO to produce a good explanation of Eroom’s Law.

The CDDO’s report should aim to explain the scale of the change in productivity. It should set out the major factors responsible for the progressive decline, and rank them in order of importance. It should consider how the relative importance of these factors has changed over time. Perhaps changes at the FDA dominated from 1960 to 1970, but something else dominates now? The analysis should compare different therapeutic areas. It should assess the extent to which the different factors are tractable. There should be some effort to quantify the ‘better than the Beatles’ problem and the ‘low-hanging fruit’ problem, as well as the potential value of underexploited drug targets. Attention should be given to the regulatory ratchet. Which requirements are most costly and least valuable? Which requirements might the regulator be persuaded to drop? What proportion of R&D cost is a direct consequence of the ‘throw money at it’ tendency? In which therapeutic areas are molecular reductionism and brute force screening methods a distraction, and in which are they genuinely helpful? What explains the difference between these

therapeutic areas? Perhaps the CDDO could quantify their analysis with a series of Venn diagrams like those in FIG. 3, to identify which sets and intersections have grown, and by how much, and which sets and intersections have shrunk. There should also be an attempt to measure the veracity of previous diagnostic and forecasting exercises. What has been the accuracy of internal forecasts on drug approvability and commercial success? Has this changed over time? What have been the most common kinds of error?

If the CDDOs provide a good explanation that is consistent with the idea that the countervailing forces will abate, or will be overcome, then all is well and good. If the explanation is unconvincing, or identifies forces that appear to be intractable, then the problems are obvious. At least it would advance the debate on how to balance the property rights of shareholders and the financial responsibilities of company boards with the wider benefits of safe, effective and affordable new drugs.

The prognosis for Eroom’s LawJust as we wanted to avoid proposing cures, we do not want to say too much about the prognosis for Eroom’s Law. However, it might appear strange if we said nothing.

Despite the durability of the trend in FIG. 1, we would be surprised if Eroom’s Law holds at an industry level over the next 5–7 years. Our view follows from two some-what mechanical factors, in addition to one more interesting reason.

Turning to the first of the mechanical factors, the amount spent on R&D is not going to increase. The ‘throw money at it’ tendency is being tackled by most compa-nies, with varying degrees of intensity. The second mechanical factor is the cumber-some biosimilar approval pathway that is emerging in the United States. Every aspect of the biosimilar production process can be scrutinized by the originator’s lawyers, and this raises the prospect of endless blocking litigation. Consequently, developers of biosimilar products anticipate to get at least some of these products approved via the standard new biologics approval pathway (the FDA’s biologics license application (BLA) process). These products will be approved as though they were novel agents, so they will inflate the number of novel approvals at very low R&D costs.

Turning to the interesting reason, we suspect that the signal-to-noise ratio may be improving among the compounds being developed for oncology indications. One or

two other therapeutic areas may be similar in this respect. Perhaps there are hints of this in the FDA’s new drug approvals in 2011. These totalled 30 overall, the most since 2004, although Munos24 has shown that the distribution of new drugs approved by the FDA per year resembles the output of a Poisson process, so we do not want to over-interpret one good year (if new drug approvals did follow a Poisson process with a mean number of 26 from 1980 to 2010, we would expect 30 drugs to be approved by chance alone around once every 5 years). Looking in more depth at the nature of the 30 new drugs, eight were anticancer agents (brentuximab vedotin, vandetanib, crizotinib, ipilimumab, asparaginase, vemu-rafenib, ruxolitinib and abiraterone acetate). A focus on rare and poorly treated diseases is also visible in the 2011 total; 11 of the 30 new drugs were orphan drugs, and the orphan drugs included seven of the eight new anticancer agents. Orphan drugs are less prone to many of the factors dis-cussed above, including the ‘better than the Beatles’ problem, the ‘cautious regulator’ problem and the big clinical trial problem.

Flat to declining R&D costs, as well as a bolus of oncology drugs, more orphan drugs and ‘biosimilars as BLAs’, might put an end to Eroom’s Law at an industry level. Whether this improves things enough to provide decent financial returns on the industry’s R&D investment is a different question. Financial markets don’t think so. Industry executives do. It would be interesting to see what CDDOs think.

Jack W Scannell, Alex Blanckley and Helen Boldon are at Sanford C. Bernstein Limited, 50 Berkeley Street,

Mayfair Place, London W1J 8SB, UK.

Brian Warrington is at Phoenix IP Ventures, 45 The Drive, Hertford, Hertfordshire SG14 3DE, UK.

Correspondence to J.W.S.  e-mail: Jack.Scannell@Bernstein.com

doi:10.1038/nrd3681

1. Hogan, J. C. Combinatorial chemistry in drug discovery. Nature Biotech. 15, 328–330 (1997).

2. Geysen, H. M., Schoenen, F., Wagner, D. & Wagner, R. Combinatorial compound libraries for drug discovery: an ongoing challenge. Nature Rev. Drug Discov. 2, 222–230 (2003).

3. [No authors listed.] Combinatorial chemistry. Nature Biotech. 18, IT50–IT52 (2000).

4. Dolle, R. E. Historical overview of chemical library design. Methods Mol. Biol. 685, 3–25 (2011).

5. Sanger, F. Sequences, sequences, and sequences. Annu. Rev. Biochem. 57, 1–28 (1988).

6. Sanger, F. et al. Nucleotide sequence of bacteriophage phi X174 DNA. Nature 265, 687–695 (1977).

7. Meldrum, C., Doyle, M. A. & Tothill, R. W. Next-generation sequencing for cancer diagnostics: a practical perspective. Clin. Biochem. Rev. 32, 177–195 (2011).  

8. Joachimiak, A. High-throughput crystallography for structural genomics. Curr. Opin. Struct. Biol. 19, 573–584 (2009).

9. Van Brunt, J. Protein architecture: designing from the ground up. Nature Biotech. 4, 277–283 (1986).

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 199

© 2012 Macmillan Publishers Limited. All rights reserved

10. Mayr, L. M. & Fuerst, P. The future of high-throughput screening. J. Biomol. Screen. 13, 443–448 (2008).

11. Schnee, J. E. Development cost: determinants and overruns. J. Bus. 45, 347–374 (1972).

12. Baily, M. N. Research and development costs and returns: the U.S. pharmaceutical industry. J. Polit. Econ. 80, 70–85 (1972).

13. Comanor, W. Research and technical change in the pharmaceutical industry. Rev. Econ. Stat. 47, 182–190 (1965).

14. Grabowski, H. G., Vernon, J. M. & Thomas, L. G. Estimating the effects of regulation on innovation: an international comparative analysis of the pharma-ceutical industry. J. Law Econ. 21, 133–165 (1978).

15. Grabowski, H. & Vernon, J. A new look at the returns and risks to pharmaceutical R&D. Manage. Sci. 36, 804–821 (1990).

16. Jensen, E. J. Research expenditures and the discovery of new drugs. J. Ind. Econ. 36, 83–95 (1987).

17. Joglekar, P. & Paterson, M. L. A closer look at the returns and risks of pharmaceutical R&D. J. Health Econ. 5, 153–177 (1986).

18. Elias, T., Gordian, M., Singh, N. & Zemmel, R. Why products fail in Phase III. In Vivo 24, 49–56 (2006).

19. Pammolli, F., Magazzini, L. & Riccaboni, M. The productivity crisis in pharmaceutical R&D. Nature Rev. Drug Discov. 10, 428–438 (2011).

20. Kola, I. & Landis, J. Can the pharmaceutical industry reduce attrition rates? Nature Rev. Drug Discov. 3, 711–715 (2004).

21. DiMasi, J. A., Feldman, L., Seckler, A. & Wilson, A. Trends in risks associated with new drug development: success rates for investigational drugs. Clin. Pharmacol. Ther. 87, 272–277 (2010).

22. Paul, S. M. et al. How to improve R&D productivity: the pharmaceutical industry’s grand challenge. Nature Rev. Drug Discov. 9, 203–214 (2010).

23. US Food and Drug Administration. Innovation or Stagnation: Challenge and Opportunity on the Critical Path to New Medical Products. FDA website [online], http://www.fda.gov/ScienceResearch/SpecialTopics/CriticalPathInitiative/CriticalPathOpportunitiesReports/ ucm077262.htm (2004).

24. Munos, B. Lessons from 60 years of pharmaceutical innovation. Nature Rev. Drug Discov. 8, 959–968 (2010).

25. Borhani, D. W. & Butts, J. A. Rethinking clinical trials: biology’s mysteries. Science 334, 1346–1347 (2011).

26. David, E., Tramontin, T. & Zemmel, R. Pharmaceutical R&D: the road to positive returns. Nature Rev. Drug Discov. 8, 609–610 (2009).

27. Garnier, J. P. Rebuilding the R&D engine in big pharma. Harv. Bus. Rev. 86, 68–79 (2008).

28. Agarwal, S. et al. Unlocking the value in big pharma. McKinsey Quarterly 2, 65–73 (2001).

29. Ruffolo, R. R. Engineering success: Wyeth redefines its research & development organisation. Drug Discovery World website [online], http://www.ddw-online.com/s/business/p148328/engineering-sucess:-wyeth-redefines-its-research-&-development-organisation-fall-05.html (2005).

30. Douglas, F. L., Narayanan, V. K., Mitchell, L. & Litan, R. E. The case for entrepreneurship in R&D in the pharmaceutical industry. Nature Rev. Drug Discov. 9, 683–689 (2010).

31. Zhong, X. & Moseley, G. B. Mission possible: managing innovation in drug discovery. Nature Biotech. 25, 945–946 (2007).

32. Horrobin, D. Realism in drug discovery — could Cassandra be right? Nature Biotech. 19, 1099–1100 (2001).

33. Horrobin, D. F. Innovation in the pharmaceutical industry. J. R. Soc. Med. 93, 341–345 (2000).

34. Horrobin, D. F. Modern biomedical research: an internally self-consistent universe with little contact with medical reality? Nature Rev. Drug Discov. 2, 151–154 (2003).

35. Ruffolo, R. R. Why has R&D productivity declined in the pharmaceutical industry? Expert Opin. Drug Discov. 1 99–102 (2006).

36. Le Fanu, J. The Rise and Fall of Modern Medicine (Little Brown, London, 1999).

37. Pisano, G. Science Business: The Promise, the Reality, and the Future of Biotech. (Harvard Business School Press, Boston, 2006).

38. Young, M. P. Prediction v Attrition. Drug Discovery World website [online], http://www.ddw-online.com/s/business/p92811/prediction-v-attrition-fall-08.html (2008).

39. Hopkins, A. L., Mason, J. S. & Overington, J. P. Can we rationally design promiscuous drugs? Curr. Opin. Struct. Biol. 16, 127–136 (2006).

40. Tollman, P., Morieux, Y., Murphy, J. K. & Schulze, U. Identifying R&D outliers. Nature Rev. Drug Discov. 10, 653–654 (2011).

41. Ford, E. S. et al. Explaining the decrease in U.S. deaths from coronary disease, 1980–2000. N. Engl. J. Med. 356, 2388–2398 (2007).

42. Lichtenberg, F. The impact of drug launches on longevity: evidence from longitudinal disease-level data from 52 countries, 1982–2001. Int. J. Health Care Finance Econ. 5, 47–73 (2005).

43. Schnee, J. E. R&D strategy in the U.S. pharmaceutical industry. Res. Policy 8, 364–382 (1979).

44. Hopkins, A. L. & Groom, C. R. The druggable genome. Nature Rev. Drug Discov. 1, 727–730 (2002).

45. Russ, A. P. & Lampel, S. The druggable genome: an update. Drug Discov. Today 10, 1607–1610 (2005).

46. Overington, J. P., Al-Lazikani, B. & Hopkins, A. L. How many drug targets are there? Nature Rev. Drug Discov. 5, 993–996 (2006).

47. Roth, B. L., Sheffer, D. L. & Kroeze, W. K. Magic shotguns versus magic bullets: selectively non-selective drugs for mood disorders and schizophrenia. Nature Rev. Drug Discov. 3, 353–359 (2004).

48. Wurtman, R. J. & Bettiker, R. L. The slowing of treatment discovery, 1965–1995. Nature Med. 1, 1122–1125 (1995).

49. Healy, D. The Psychopharmacologists: Volume 2 93–118 (Hodder Arnold, London, 1999).

50. Healy, D. The Psychopharmacologists: Volume 2 259–264 (Hodder Arnold, London, 1999).

51. Healy, D. The Antidepressant Era (Harvard University Press, Cambridge, Massachusetts, 1997).

52. Weatherall, M. An end to the search for new drugs? Nature 296, 387–390 (1982).

53. Richard, J. & Wurtman, M. D. What went right: why is HIV a treatable infection? Nature Med. 3, 714–717 (1997).

54. [No authors listed.] A dearth of new drugs. Nature 283, 609 (1980).

55. Persson, C. G., Erjefält, J. S., Uller, L., Andersson, M. & Greiff, L. Unbalanced research. Trends Pharmacol. Sci. 22, 538–541 (2001).

56. Ainsworth, C. Networking for new drugs. Nature Med. 17, 1166–1168 (2011).

57. Denome, S. A., Elf, P. K., Henderson, T. A., Nelson, D. E. & Young, K. D. Escherichia coli mutants lacking all possible combinations of eight penicillin binding proteins: viability, characteristics, and implications for peptidoglycan synthesis. J. Bacteriol. 181, 3981–3993 (1999).

58. Keith, C. T., Borisy, A. A. & Stockwell, B. R. Multicomponent therapeutics for networked systems. Nature Rev. Drug Discov. 4, 71–78 (2005).

59. Lombardino, J. G. & Lowe, J. A. The role of the medicinal chemist in drug discovery — then and now. Nature Rev. Drug Discov. 3, 853–862 (2004).

60. Swinney, D. C. & Anthony, J. How were new medicines discovered? Nature Rev. Drug Discov. 10, 507–519 (2011).

61. Reichert, J. M. Probabilities of success for antibody therapeutics. mAbs 1, 387–389 (2009).

62. Steward, F. & Wibberly, G. Drug innovation — what’s slowing it down? Nature 284, 118–120 (1980).

63. Collins, F. S. Medical and societal consequences of the Human Genome Project. N. Engl. J. Med. 341, 28–37 (1999).

64. Rees, J. Post-genome integrative biology: so that’s what they call clinical science. Clin. Med. 1, 393–400 (2001).

65. Grove, A. Rethinking clinical trials. Science 333, 1679 (2011).

66. O’Shaughnessy, J. et al. Iniparib plus chemotherapy in metastatic triple-negative breast cancer. N. Engl. J. Med. 364, 205–214 (2011).

67. O’Shaughnessy, J. et al. A randomized Phase III study of iniparib (BSI-201) in combination with gemcitabine/carboplatin (G/C) in metastatic triple-negative breast cancer (TNBC). J. Clin. Oncol. 29, Abstr. 1007 (2011).

68. Guha, M. PARP inhibitors stumble in breast cancer. Nature Biotech. 29, 373–374 (2011).

69. Soda, M. et al. Identification of the transforming EML4–ALK fusion gene in non-small-cell lung cancer. Nature 448, 561–566 (2007).

70. Chapman, P. B. et al. Improved survival with vemurafenib in melanoma with BRAF V600E mutation. N. Engl. J. Med. 364, 2507–2516 (2011).

71. [No authors listed.] Regulatory watch: leading hedgehog inhibitor submitted for approval as skin cancer drug. Nature Rev. Drug Discov. 10, 802–803 (2011).

72. DeMonaco, H. J., Ali, A. & von Hippel, E. The major role of clinicians in the discovery of off-label drug therapies. Pharmacotherapy 26, 323–332 (2006).

73. Mathieu, M. P. (ed.) Parexel’s Bio/Pharmaceutical R&D Statistical Sourcebook 2010/2011 163–261 (Barnett International, Needham, Massachusetts, 2010).

74. Marshall, G. et al. Streptomycin treatment of pulmonary tuberculosis. BMJ 30, 769–782 (1948).

75. MacNeil, J. S. H. Changes in the characteristics of approved New Drug Applications for antihypertensives. Thesis, Massachusetts Institute of Technology (2007).

76. Lin, H. S. Changes in the characteristics of new drug applications for the treatment and prevention of diabetes mellitus. Thesis, Massachusetts Institute of Technology (2007).

77. Scandinavian Simvastatin Survival Study Group. Randomised trial of cholesterol lowering in 4444 patients with coronary heart disease: the Scandinavian simvastatin survival study (4S). Lancet 344, 1383–1389 (1994).

78. Munos, B. How to avert biopharma’s R&D crisis. In Vivo 29, 2011800050 (2011).

79. Macarron, R. et al. Impact of high-throughput screening in biomedical research. Nature Rev. Drug Discov. 10, 188–195 (2011).

80. Bohacek, R. S., McMartin, C. & Guida, W. C. The art and practice of structure-based drug design: a molecular modeling perspective. Med. Res. Rev. 16, 3–50 (1996).

81. Brown, D. Future pathways for combinatorial chemistry. Mol. Divers. 2, 217–222 (1996).

82. Lipinski, C. A., Lombardo, F., Dominy, B. W. & Feeney, P. J. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev. 46, 3–26 (2001).

83. Gleeson, M. P., Hersey, A., Montanari, D. & Overington, J. Probing the links between in vitro potency, ADMET and physicochemical parameters. Nature Rev. Drug Discov. 10, 197–208 (2011).

84. Kay, J. Obliquity: Why our goals are best achieved indirectly (Profile Books, London, 2010).

85. Watts, D. J. & Strogatz, S. H. Collective dynamics of ‘small-world’ networks. Nature 393, 440–442 (1998).

86. Pharmaceutical Research and Manufacturers of America. Pharmaceutical Industry Profile 2011. PhRMA website [online], http://www.phrma.org/sites/default/files/159/phrma_profile_2011_final.pdf (Washington DC, PhRMA, April 2011).

87. Congress of the United States: Congressional Budget Office. Research and Development in the Pharmaceutical Industry. Congressional Budget Office (CBO) website [online], http://www.cbo.gov/ftpdocs/76xx/doc7615/10-02-DrugR-D.pdf (October 2006).

AcknowledgementsW. Bains, T. Curtis, B. Charlton, M. Young, O. Imasogie, G. Porges, and B. Munos were generous with their time and ideas during various stages in the genesis of this article.

DisclaimerThe information provided herein was prepared by Sanford C. Bernstein & Co. LLC and Brian Warrington. It is not invest-ment research, although it may refer to a Bernstein research report or the views of a Bernstein research analyst. This com-munication does not constitute a complete fundamental analysis of any companies mentioned. Unless indicated, all views expressed herein are the views of the authors and may differ from or conflict with those of the Bernstein Research Department. This article does not constitute investment advice or recommendations and is not a solicitation or an offer to purchase or sell securities. The information contained herein is only good as of the date and time of the publication; we do not undertake to advise of any changes in the opinions or information contained herein.

Competing interests statementThe authors declare competing financial interests: see Web version for details.

FURTHER INFORMATIONRCSB Protein Data Bank database: http://www.rcsb.org/pdb/statistics/holdings.do

SUPPLEMENTARY INFORMATIONSee online article: S1 (table) | S2 (box)

ALL LINKS ARE ACTIVE IN THE ONLINE PDF

P E R S P E C T I V E S

200 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

B I O M A R K E R S — O P I N I O N

Cancer biomarkers: selecting the right drug for the right patientGary J. Kelloff and Caroline C. Sigman

Abstract | This Perspective highlights biomarkers that are expressed as a con sequence of cancer development and progression. We focus on those bio markers that are most relevant for identifying patients who are likely to respond to a given therapy, as well as those biomarkers that are most effective for measuring patient response to therapy. These two measures are necessary for selecting the right drug for the right patient, regardless of whether the setting is in drug development or in the post‑approval use of the drug for patients with cancer. We also discuss the innovative designs of clinical trials and methodologies that are used to validate and qualify biomarkers for use in specific contexts. Furthermore, we look ahead to the promises and challenges in the field of cancer biomarkers.

Despite extraordinary advances in our understanding of the biology that underlies the development and progression of cancer as well as potential molecular targets for its treatment1–4, more than 90% of all new oncology drugs that enter clinical develop-ment do not obtain marketing approval5. Many drugs fail in late stages of develop-ment — often in Phase III trials — because of inadequate activity, lack of strategies for combating resistance to these drugs, unexpected safety issues or difficulties in determining efficacy because of reasons that include confounded outcomes of clinical trials. Moreover, an increased understanding of cancer biology has shown that cancers are heterogeneous diseases6,7, which suggests that there is a high likelihood that effective cancer treatments will need to address patient-specific molecular defects and aspects of the tumour microenvironment.

Together, these factors contribute to the high costs and slow pace of oncology drug development, and clearly highlight the need for faster, more cost-effective strategies for evaluating cancer drugs and more effectively defining the patients who will benefit from treatment. Recent initiatives from the US Food and Drug Administration (FDA) have acknowledged the role of well-characterized biomarkers in drug development strategies. In particular, the FDA has called for the use of analytically validated biomarkers that have strong evidence of being fit for purpose to evaluate patient response to therapy, potential toxicity and drug resistance

(see the FDA Critical Path Initiative on the FDA website, and the report entitled “Draft Guidance for Industry: Qualification Process for Drug Development Tools” on the FDA website). The FDA has also recognized that diagnostic tests that are used to measure biomarkers can have a role in the selection of patients who are likely to respond to specific therapies (see the report entitled “Drug-Diagnostic Co-Development Concept Paper” on the FDA website).

With the intense work proceeding in target discovery and validation, more cancers are being characterized for which targeted therapy has been or will be developed. Such targeted drugs are directed at defects in cellular signalling that may occur in only a small subset of patients and may therefore be effective only — or mostly — in cancer patients with these defects in cellular signal-ling, thus confirming the concept and promise of applying personalized medi-cine in oncology. The defects, and thus the selected patient subsets, are characterized by the presence of distinct biomarkers.

Well-known examples of targeted drugs and their respective biomarkers are: trastuzumab (Herceptin; Roche/Genentech), which is approved for the treatment of indi-viduals with breast cancer characterized by HER2 (also known as ERBB2) amplification and overexpression8; and imatinib (Gleevec; Novartis), which is approved for the treat-ment of chronic myeloid leukaemia and acts by inhibiting the product of the BCR–ABL fusion gene that is present in patients with chronic myeloid leukaemia9. Vemurafenib

(Zelboraf; Daiichi Sankyo/Roche) — a BRAF kinase inhibitor — has been approved by the FDA for the treatment of the >50% of patients with metastatic melanoma who have the BRAFV600E mutation in their tumours10–12. Crizotinib (Xalkori; Pfizer), an inhibitor of anaplastic lymphoma kinase (ALK), has also been approved by the FDA for the treatment of patients with non-small-cell lung cancer (NSCLC) who carry an ALK gene rearrangement. The drug is particularly efficacious against NSCLC in most patients with tumours that express the echinoderm microtubule-associated protein-like 4 (EML4)–ALK fusion protein found in approximately 5–7% of patients with NSCLC13–16. In addition, inhibitors of epidermal growth factor receptor (EGFR) have been approved for the treatment of NSCLC in patients with characteristics (for example, not smoking) that are known to be strongly correlated with the presence of EGFR mutations17.

In this Perspective, we highlight the sources, function and evaluation of biomarkers that are useful in cancer drug development and patient care. However, we focus on the use of biomarkers in the advancement of personalized medicine for patients with cancer; this involves identifying which patients are likely to respond to — or be resistant to — a given treatment (for example, through the use of companion diagnostics), and measuring the activity of treatments (that is, the efficacy). The uses and development of strategies for cancer biomarkers have been comprehensively discussed in the literature (reviewed in

REFS 18–22).

Cancer biology creates biomarkers There are many ways to characterize bio-markers; here, we focus on cancer biology because of the logical construct it creates to focus the search for — and on which to prioritize — biomarkers at the molecular, cellular and tissue level.

During cancer development and progres-sion, the normal function of cells, tissues and organs is disrupted. These changes occur as a result of genomic changes, which include mutations, genome rearrangements, amplifications and deletions, as well as epigenetic modulation of gene expression. Clinical cancer — the end result of this chaotic process — is characterized by unregu-lated cellular proliferation as well as cellular and clonal heterogeneity. Consequently, initial effective therapeutic intervention at defined molecular target sites is usually transient and followed by resistance to the therapy.

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 201

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Rate of mutagenesis

Exposure dose(mutagens and mitogens)

Rate of DNA lesions

Rate of mitogenesis

Rate of clonal evolution Cancer

Rate of variant clone formation

Rate of clonal expansion

Rate of cell death

Rate of cell death

Figure 1 | Rate of cancer development and progression. Cancer devel‑opment and progression with its attendant biomarkers proceeds at a rate of clonal evolution ultimately determined by the rates of variant clone cell formation and expansion, which in turn is driven by exogenous

mutagens and mitogens (the exposure dose), and countered by cell death. The variant clones that have a growth advantage over normal cells result in clinical cancer with its corresponding causative and resulting biomarkers.

These events are intrinsic properties of neoplastic cells and tissues, and have been described by Hanahan and Weinberg1,2 as acquired characteristics of cancer. These hallmarks are not only brought about by genetic changes in tumour cells but also by changes in the tumour microenviron-ment and the influences of the tumour microenvironment on the tumour. Cancer progression is also an evolutionary process in which there is clonal selection for the altered cells that are able to evade immune surveillance and thrive in the stressful con-ditions created by the disruption to normal cellular function23. The following five fea-tures of cells that undergo clonal selection have been identified as stress phenotypes of cancer: DNA damage, oxidative stress, mitotic stress, pro-apoptotic stress and metabolic stress3.

All these features and the molecular events that precede them could be con-sidered as sources of cancer biomarkers and targets for therapeutic intervention. Furthermore, the complexity of identifying the specific factors that lead to neoplastic progression in cells and tissues is evidenced by the intricate network of protein signalling pathways encoded by the human genome. With the many possible genetic variants taken into account, the approximately 23,000 genes encoded by the human genome are the basis for more than one million potential protein–protein inter actions24, some of which could be characteristics of neoplasia. Advances in genomics, molecular and cellular biology, and tissue pathology (particularly imaging

of the pathology) are providing the tools for identifying and evaluating biomarkers of cancer progression.

Molecular progression. Cancer is inherently a disease of genetic progression. This progression is observed in specific molecular and more general genotypic damage that is associated with increasingly severe histopathological phenotypes25–29. Early crucial steps include the inactivation of tumour suppressor genes such as the genes encoding adenomatous polyposis coli (APC)27,29,30, BRCA1, BRCA2 (REF. 31) or phosphatase and tensin homolog (PTEN)32–34, and the activation of oncogenes such as RAS and PIK3CA35 (the gene encoding phosphoinositide 3-kinase). In some cases, such as cancer of the oesophagus, the progression of cancer in high-risk tissues has been correlated with the appearance of a cluster of genetic defects such as point mutations and loss of heterozygosity in P16 and P53 tumour suppressor genes, as well as P16 promoter methylation36,37. Baseline levels of these molecular targets and changes in their level following treatment are potential biomarkers of cancer response or progression.

Polyak and colleagues6,38,39 have described intratumoural genetic hetero-geneity in the progression of breast cancer, and noted that the degree of heterogeneity is associated with known breast cancer subtypes and clinical outcomes. This sug-gests the utility of analysing subpopula-tions of tumour cells as biomarkers for identifying the most effective therapies

and distinguishing between indolent and aggressive forms of disease. Moreover, the complexity and heterogeneity of cancer progression suggests that single and small panels of molecular biomarkers may not always be able to characterize a tumour. This is being addressed by the development of gene and protein expression arrays, which may be used as biomarkers to classify the disease subtype of an individual patient. These arrays include molecular profiles that define breast cancer subtypes40,41 and pro-files that identify patients with lymphoma who will respond to therapy42.

Another approach for developing biomarkers to understand genome-based cancer progression could involve analysing the changes that occur along specific signal-ling pathways. Vogelstein and colleagues43 identified 12 pathways that were altered in a small group of pancreatic cancers; these pathways may provide biomarkers that could be useful in the development of molecularly targeted therapies. However, the authors stated that this approach is difficult, as many key pathways overlap and it is not always clear which pathways are crucial for the control of cancer progression. In addition, much of the information about the signalling pathways has not come from human cancer tissue; rather, it has come from different cell types and animal models in which the relevance of the pathways may be different from that in human tumours.

The challenge is to identify those muta-tions that are the drivers of cancer (that is, driver mutations) and those that are passenger mutations. For example, driver mutations do

P E R S P E C T I V E S

202 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Table 1 | Biomarkers derived from cancer development and progression

Category* Examples

Response biomarkers‡

Pharmacodynamic biomarkers (biological effects produced by a drug that may or may not be directly related to neoplastic processes)

• Effect on molecular target (for example, inhibition of EGFR, HER2, MET or VEGFR)

• Induction of enzyme activity relevant to drug toxicity (for example, CYP1A1 or CYP1A2 activity)

• Functional (and molecular) imaging of drug interaction at target tissue

Efficacy and potential surrogate end point biomarkers

• Anatomical imaging (for example, MRI or CT)• CTCs, endothelial cells or tumour DNA• Functional imaging (for example, FDG–PET, DCE‑MRI or DW‑MRI)• Genomic and proteomic expression profiles (for example,

reverse‑phase protein microarrays)• Quantitative pathology (for example, computer‑assisted image

analysis) or cytology proliferation biomarkers (for example, PCNA, KI67 or FLT)

Disease characterization§

Prognostic biomarkers (correlated with clinical outcome but not directly related to specific mechanisms of cancer progression)

• DNA methylation, gene expression profiles (for example, HER2, OncoType DX or MammaPrint), CTCs, CEA, AFP, CA125 (measured by Rustin Response Criteria) and PSA (for example, PSA kinetics)

Predictive biomarkers (predict tumour response to specific targeted drug interventions)

• Expression of molecular targets and pathways associated with cancer progression (for example, HER2, EGFR, MET, PI3K–AKT pathway, EML4–ALK, BCR–ABL and mutated BRAFV600E)

• Expression of molecular targets and pathways associated with resistance to cancer therapy (for example, ERCC1 or EGFRT790M mutation)

Risk biomarkers (describe risks of cancer or cancer progression)

• Genetic predisposition (for example, mutations in APC, BRCA1, BRCA2, MLH1 or MSH2, as well as Li‑Fraumeni syndrome, ataxia telangiectasia or PTEN loss)

• Environmental and lifestyle factors (for example, HPV or HBV infection, or tobacco use)

• Multifactorial risk model (for example, Gail risk model for breast cancer risk)

• Genetic polymorphisms (for example, polymorphisms in CYP1A1, GSTM1, GSTP1 or SRD5A2)

AFP, alpha‑fetoprotein; ALK, anaplastic lymphoma kinase; APC, adenomatous polyposis coli; CA125, cancer antigen 125; CEA, carcinoembryonic antigen; CT, computed tomography; CTC, circulating tumour cell; CYP1A1, cytochrome P450 1A1; DCE, dynamic contrast‑enhanced; DW, diffusion‑weighted; EGFR, epidermal growth factor receptor; EML4, echinoderm microtubule‑associated protein‑like 4; ERCC1, DNA excision repair cross‑complementation protein 1; FDG–PET,18F‑fluorodeoxy glucose positron emission tomography; FLT, 18F‑3′‑fluoro‑3′‑deoxy‑l‑thymidine; GSTM1, glutathione S‑transferase μ1; GSTP1, glutathione S‑transferase π1; HBV, hepatitis B virus; HPV, human papilloma virus, MLH1, MutL protein homolog 1; MRI, magnetic resonance imaging; MSH2, MutS protein homolog 2; PCNA, proliferating cell nuclear antigen; PI3K, phosphoinositide 3‑kinase; PSA, prostate‑specific antigen; PTEN, phosphatase and tensin homolog; SRD5A2, steroid‑5‑alpha‑reductase 2; VEGFR, vascular endothelial growth factor receptor. *Many of the biomarkers listed, as well as some that are not listed, fit into multiple categories depending on the context in which they are used. ‡Measurements that are associated with a response or lack of response to a therapy; a response can be defined using any recognized clinical end point. §Measurements available at the time of diagnosis or treatment that are associated with disease progression.

not necessarily occur at high frequencies in tumours and so may be difficult to detect without examining many tumours44. For example, mutations in the gene encoding isocitrate dehydrogenase 1, which leads to a build-up of 2-hydroxyglutarate, have been found at low levels in colorectal cancer and other cancers but were only identified as potential driver mutations after The Cancer Genome Atlas study of 500 glioblastomas found that these mutations occurred at a frequency of 12% in these tumours45. Despite rapid advances in the capacity and sensitivity of sequencing technologies, limitations still exist in the detection of the mutational spectrum in cancer cells. Specifically, even the most advanced current techniques only detect mutations that occur within the first 12 of the 30 generations of clonal expansion that are required to form a clinically detectable tumour46.

Cell and tissue progression. Experimental and epidemiological studies of cancer development and progression show that cancers are characterized by cellular prolif-eration47. Mutagenesis can damage the cell and destroy the control of normal growth, resulting in increased proliferation in addition to loss of cell death and maturation pathways. Generally, the rate of cancer development is defined as the rate at which clonal variants that grow faster than surrounding cells appear during neoplasia, adding to the overall proliferation rate48,49 (FIG. 1). The proliferation rate of cancer cells in the total cell population determines the rate at which genomic variants are pro-duced, because each transition through the cell cycle converts damaged DNA lesions into mutations and also subjects the genome to a greater sensitivity to mutagens49,50. This results in an initial upregulation of the kinetic cycle that involves increased produc-tion of cancer cells, genetic variants and fast-growing cellular variants. The driving force of the cycle is entropic — that is, a selection pressure exists towards increasing disorder and heterogeneity as the controls for maintaining homeostasis are lost.

The cellular processes described above result in changes in cell phenotypes — such as DNA ploidy as well as nuclear and nucleolar morphometry — that can be measured quantitatively (that is, they could be considered as biomarkers) even before they are detected in the tissue phenotype. Biomarkers such as morphometric changes (including nuclear size and shape, and clus-tering) can be quantitatively detected via well-advanced techniques such as optical

imaging51,52 and high-resolution micro-scopy. These technologies could be valu-able in the development of cancer drugs, in which subtle but statistically significant effects may be seen in treated samples compared with controls, and in which the number of samples available for evaluation may be limited.

Changes in cell metabolism also provide opportunities for identifying biomarkers of cancer progression. The high relevance of 18F-fluorodeoxyglucose positron emission

tomography (FDG–PET) imaging for measuring cancer biology, progression and response to cancer therapy relies on the Warburg phenomenon53,54. Growing tumours require high amounts of carbon and other elements as well as energy to produce progeny cells, and this leads to a substantial upregulation in glucose transporters, which creates an influx of glucose and FDG into cells, resulting in a detectable image within a darker background54. This phenomenon, which occurs in almost all cancers,

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 203

© 2012 Macmillan Publishers Limited. All rights reserved

Box 1 | Biomarker validation and qualification

The qualification of biomarkers as tools for efficient drug development was first described in the US Food and Drug Administration (FDA) Critical Path Initiative, which was published in 2004 as the strategy of the FDA’s Center for Drug Evaluation and Research (CDER) for modernizing the development, evaluation and manufacture of drugs and other medical products. The CDER defines the qualification of biomarkers as “… a conclusion that within a stated context of use, the results of assessment with a biomarker can be relied upon to have a specific interpretation in drug development and regulatory review” — that is, the biomarker is ‘fit for purpose’. The CDER has provided a guidance document on the qualification process (titled “Draft Guidance for Industry: Qualification Process for Drug Development Tools”).Some of the features of this process are as follows:• Biomarkers are measured using specific devices (for example, assays or instruments), and if

they are marketed for managing patients in clinical practice, they are reviewed by the FDA’s Center for Devices and Radiological Health (CDRH) for the ability to analytically measure the biomarker.

• Biomarkers that are being considered for qualification are independent of the specific device that performs the measurement, and devices that measure the biomarker are expected to yield equivalent results.

• Biomarker qualification cannot be achieved without analytical and clinical validation of at least one device to measure the biomarker. Analytical validity may be defined as the ability of an assay to accurately and reliably measure the analyte of interest in the laboratory and in specimens that are representative of the population of interest114,115. Clinical validation requires the detection or prediction of the associated disorder of interest in specimens from targeted patient groups. Standards for collecting, analysing and reporting these biomarker data114,116 are summarized in BOX 3.

• The clearance of a measurement device by the CDRH does not imply that the biomarker has been demonstrated to have a qualified use in drug development, and additional data are needed from studies designed to establish qualification.

• Qualification data include consensus methods and evidence that correlates biomarker measurements with sufficient clinical outcomes. The lack of standardization of protocols for biomarker specimens is a limiting factor in acquiring these data.

• Biomarker qualification is intended to provide some degree of generalizability for the use of the biomarker as a drug development tool across multiple clinical disorders, drugs or drug classes, and for benefiting patients. Historically, the acceptance of biomarker measurements has usually been specific to a sponsor and a drug.

• Once qualified, a biomarker can be used by drug developers for its qualified ‘context of use’ in regulatory submissions without the need for further review by the CDER.

• The term ‘context of use’ used by the FDA describes the setting (or settings) in which the biomarker is qualified, and boundaries within which the available data justify its use. The context of use may expand as new data are obtained.Biomarker qualification also enables collaboration among stakeholders, reduces costs

for individual stakeholders and provides biomarkers that are useful to multiple parties.

accounts for the high utility of FDG–PET imaging in the diagnosis, staging and management of patients with cancer over the past 20 years53. More recently, there has been an accumulation of FDG–PET-computed tomography (CT) data, which shows that a semi-quantitative FDG–PET measurement — known as a standard uptake value (SUV) — can serve as a reliable biomarker of efficacy in response to standard55 and targeted56–58 therapy.

Drug development and patient careIn the early stages of cancer drug develop-ment, biomarkers are used to: identify and validate therapeutic targets; screen and optimize candidate drugs; provide proof

of concept for drugs and disease models; and enhance a mechanistic understanding of drugs or drug combinations. In the later stages of cancer drug development and in the management of patients with cancer, biomarkers can be used to: identify opti-mal target populations of patients; predict the efficacy of the drug and the patient’s response, resistance and toxicity; and rapidly distinguish between non-responders and patients who respond to therapeutic intervention18.

Biomarkers can be defined based on their use in general and specific contexts, both in drug development and in the management of patients with cancer. It is useful to distin-guish biomarkers that can be used for disease

characterization from biomarkers that can be used to measure response to therapy. Generally, biomarkers that are measured without regard to therapy (or before therapy) are used to characterize the disease; these biomarkers are referred to as prognostic, predictive and risk biomarkers. Conversely, biomarkers that are measured during therapy are used to measure response to treatment (such as pharmacodynamics, efficacy and as surrogate end points). Definitions and specific examples of these biomarkers are described below, and many examples are shown in TABLE 1.

Disease characterization (prognostic, predictive and risk biomarkers). The pres-ence of prognostic biomarkers is highly correlated with clinical outcome but may not be implicated in specific mechanisms of cancer development and progression. Some of these biomarkers indicate prognosis in general, such as gene expression patterns59 that provide molecular signatures predicting the likelihood of cancer recurrence and the need for chemotherapy in patients with oestrogen receptor-positive breast cancer60,61. Other biomarkers indicate the likelihood of response to chemotherapy in general; examples of such biomarkers include FDG–PET avidity (which is meas-ured before therapy by SUV) and molecular signatures that can identify patients with B cell lymphoma who are likely or unlikely to respond to standard therapy62. FDG–PET avidity is a well-recognized biomarker of cancer prognosis, in which higher SUVs are correlated with — or are prognostic for — aggressive forms of disease in lung cancer, lymphoma and other cancers53. These prog-nostic biomarkers are useful for managing patients with cancer; they are also being increasingly used in clinical trials to define patient populations and to select patients who are likely to respond to or benefit from chemotherapy while excluding patients who are not likely to respond63.

Predictive biomarkers (which are often early genetic events in cancer) generally reflect specific mechanisms of cancer pro-gression, correlate with clinical outcomes and predict tumour response to specific drug interventions that target the mecha-nism of cancer progression. Perhaps the best examples of predictive biomarkers are companion diagnostics such as those measuring HER2, BCR–ABL, EGFR, mutated BRAF and ALK (described above), which identify patients who are likely to respond to drugs interfering with these specific targets.

P E R S P E C T I V E S

204 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Box 2 | Determining that a biomarker assay is ready for clinical studies

The US Food and Drug Administration (FDA), the US Centers for Disease Control (CDC) and the Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group have published the following criteria for the analytical and clinical validation of biomarker assays:• “Class II Special Controls Guidance Document: Drug Metabolizing Enzyme Genotyping

System” (see the guidance document on the FDA website)

• “Guidance on Pharmacogenetic Tests and Genetic Tests for Heritable Markers” (see the FDA website)

• “Drug–Diagnostic Co-Development Concept Paper” (see the FDA website)

• “In Vitro Diagnostic Devices: Guidance for the Preparation of 510(k) Submissions” (see the guidance document on the FDA website)

• “Interactive Review for Medical Device Submissions: 510(k)s, Original PMAs, PMA Supplements, Original BLAs, and BLA Supplements”115 (see the guidance document on the FDA website)

Briefly, a detailed characterization of the assay platform, including the methodology for detecting the analytes of interest and a description of the methodology and specimen preparation, is needed in the documentation that supports analytical validation, and each assay should have standard controls that produce results in the informative range.

The assay should measure a biologically and clinically meaningful end point (that is, it should correlate with clinical outcomes). Clinical validity requires integration of analytical validity with predictive values of positive and negative tests that take disease prevalence, penetrance, expressivity and other genetic or environmental factors into account.

Analytical validityPre-analytical, analytical and post-analytical criteria for determining validity are available.

The assay is well characterized. Quantitative assays are preferred over qualitative assays; quality control is established; intra- and interlaboratory assay precision, the range of values for patient specimens and failure rates are known; and consistent results are obtained among multiple laboratories.

Assay data are from adequate sources. The most reliable sources are: collaborative studies with large panels of well-characterized samples, summary data obtained from well-designed external proficiency testing and interlaboratory comparison programmes. The least reliable sources are: unpublished or non-peer reviewed research, and studies by clinical laboratories or assay manufacturers on the performance of the basic methodology conducted on a different target.

Parameters for assessing the internal validity of the studies are used. The assay platform and test procedures are described, including the reproducibility of test results, quality control and comparison to a gold standard. Samples are representative of the study population. Blinded testing and interpretation have been performed. Data analysis includes point estimates of sensitivity and specificity with 95% confidence intervals as well as sample size and/or power calculations. Studies are graded as convincing, adequate or inadequate to provide estimates of analytical sensitivity and specificity using intended sample types from representative populations.

Clinical validityEvaluatable points. Evaluatable points include: disease prevalence in the study population; genotype and/or phenotype relationships; genetic, environmental and other modifiers of disease; positive and negative predictive values; methods to resolve clinical false-positive results; and assay validation on all patient populations.

Study designs for the clinical evaluation of tests. Study designs are ranked by reliability (from highest to lowest), and include: longitudinal cohort or validated clinical decision studies; case-control studies; cross-sectional or unvalidated clinical decision studies; case series; unpublished research; clinical laboratory and manufacturer studies; and expert opinion.

Parameters for assessing the internal validity of studies. The disorder (or phenotype) and clinical outcomes of interest are well described, with prevalence estimates as well as verified cases and controls. The study design, tests and methodology, as well as the patient population, have been described. The handling of indeterminate results and outliers is described. Blind testing and interpretation are performed. Data analysis includes possible biases, as well as point estimates of clinical sensitivity and specificity with 95% confidence intervals provided, along with estimates of positive and negative predictive power. Studies are graded as convincing, adequate or inadequate based on their design and conduct in representative patient populations that measure the strength of association between a biomarker and a specific, well-defined disease or phenotype.

Of note, biomarkers that predict the response of the tumour to specific drugs may also be prognostic for patients. For example, the overexpression or ampli-fication of HER2 in breast cancer indicates both a poor prognosis and that the disease will respond to trastuzumab64. By contrast, the presence of certain EGFR mutations in patients with NSCLC indicates a good prognosis but also predicts response to EGFR inhibitors65. In addition, in cancer progression models genetic events lead to loss of function (for example, tumour sup-pressors) or gain of function (for example, by gene amplification). In the case of BRCA1 mutations, the loss of function is in DNA repair mechanisms, resulting in sensitivity to poly(ADP–ribose) polymer-ase inhibitors66,67. Predictive biomarkers may also identify patients who are likely to be resistant to specific treatments. An example is the presence of activated KRAS, which is correlated with resistance to EGFR inhibitors in patients with tumours that overexpress EGFR68.

Many current drug development strategies use predictive biomarkers. For example, studies have been designed that evaluate new drugs targeting specific genes or muta-tions by identifying patient populations carrying these mutations (and/or genes) or gene expression patterns (see below).

Risk biomarkers, which may identify clinical cohorts of patients for intervention, measure the risks of cancer, cancer progres-sion and recurrence. They include measure-ments of carcinogen exposure69, genetic predisposition69–71, pharmaco genomic para meters72, previous disease or precursor lesion25,73, and multifactorial risk models (such as the Gail risk model for breast cancer risk)74.

Although it has not been covered in this article, another important use of biomarkers is in the prediction of drug toxicity, which is often measured by the presence of aberrant drug-metabolizing enzymes such as polymorphisms in the gene encoding the UDP-glucuronosyltransferase 1A1 enzyme — the presence of which can identify patients who are likely to suffer severe toxicity (diarrhoea or leukopaenia) when treated with irinotecan75.

Response to therapy (pharmacodynamic, efficacy and surrogate end point biomarkers). Pharmacodynamic biomarkers measure the effects produced by a drug that may or may not be directly related to cancer progression. Some of these biomarkers are measured in blood, saliva, urine or hair follicles; many are not meaningful unless they are measured

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 205

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Baseline measurement with analytically validated imaging assay

a Develop evidence that response measured by imaging-based biomarker correlates with clinical benefit

Treatment with approved chemotherapeutic drug (standard therapy)

Predetermined response level achieved

Continue treatment to clinical end point (or end points) (for example, OS, DFS, PFS or OR) by conventional measurement

Baseline measurement with analytically validated imaging assay

b Further evidence of development (for example, with different drugs that have different mechanisms of action in a specific target organ (or organs)) to demonstrate clinical benefit for the evaluation of new therapies

Randomize to treatment with new therapy versus standard therapy

Predetermined response level achieved

Continue treatment to clinical end point (or end points) or carry out confirmatory trial with clinical end points

If response is met for predetermined percentage of patients, this may support claims of clinical benefit and accelerate approval of new therapy

Figure 2 | Qualification of imaging-based biomarkers as measures of clinical benefit. A two‑phase clinical strategy is described for developing evidence to qualify an imaging‑based biomarker to measure response to therapy. The first phase (part a) is carried out to demonstrate that the bio‑marker measurement correlates with a clinical end point, and the second phase (part b) is carried out to explore the specificity of the correlation across drugs with different mechanisms of action and different cancer target organs. DFS, disease‑free survival; OR, overall response; OS, overall survival; PFS, progression‑free survival.

in the target tissue (or tissues) of the drug. Although some of these biomarkers can be used for many mechanistically distinct drug classes (for example, biomarkers of cellular proliferation or cellular death), most reflect the anticancer activity of only a few drug classes (for example, oestrogen levels for drugs that affect hormone synthesis). Often, these biomarkers are measures of the reduced expression or activity of a molecular target in response to a mechanism-based therapy, such as reduced EGFR tyrosine kinase activity in response to EGFR inhibitors. Biomarkers of drug-specific toxicities, such as induction of cytochrome P450 enzymes, are also pharmacodynamic biomarkers76.

Although biomarkers that are based on the effects of a drug are useful end points for preclinical and early-stage clinical pharma-cology studies, and for correlating therapeu-tic activity to drug dose and administration, they are not usually adequate as measures of anticancer activity. Nevertheless, newer tech-niques for tissue-based pharmaco dynamic measurements — including molecular and functional imaging — will allow the development of biomarkers that are better measurements of drug effect and that will be useful in preclinical and early clinical phases of oncology drug development. For example, such biomarkers could be used to select the drug dose and dose regimen, identify the

best duration of treatment, identify possible toxicities and compare potencies of several drug candidates.

Efficacy biomarkers measure the effects of cancer therapy on tumours; these bio-markers are specifically modulated by drug treatment, and their modulation correlates at least to some degree with clinical status and benefit. Typically, these biomarkers are used as end points in Phase II trials.

Historically, measurements of tumour response by anatomical imaging (CT or mag-netic resonance imaging) have been incorpo-rated into specific criteria (for example, the World Health Organization tumour response criteria and the Response Evaluation Criteria for Solid Tumours (RECIST)) for defining a complete response or partial response, as well as for defining a progressive form of the disease and a stable form of the disease (in addition to evaluating drug efficacy)77. However, these anatomical measurements may miss some aspects of tumour response, such as the formation of necrotic tissue and the slowing of tumour metabolism, particu-larly for molecularly targeted therapies.

Functional imaging (for example, FDG–PET) and functional imaging combined with anatomical imaging (for example, FDG–PET/CT) are integrative biomarkers that cap-ture both tumour size and viability. To date, no cancer drug has been approved based on a functional imaging end point. Other exam-ples of efficacy biomarkers include complete pathological response in locally advanced breast cancer63 and change in number of cir-culating tumour cells (CTCs) from baseline to post-treatment in advanced prostate78 and breast cancers79. As described below, there are ongoing efforts to qualify FDG–PET/CT imaging as a biomarker for measuring drug efficacy in lymphoma, lung cancer and other cancers, and to qualify CTCs as a biomarker of drug efficacy in prostate cancer.

Biomarkers can serve as surrogate end points in clinical trials to support the regula-tory approval of drugs. When used in this manner, biomarkers can measure drug efficacy, and there is a strong correlation between the biomarker and the clinical outcomes80. Given the complexity of cancer — and with the exception of biomarkers for haematological cancers — the evolution of biomarkers that can be used as surrogate end points has been slower for oncological drugs than for drugs developed for other diseases (for example, AIDS or cardiovascular disease). The consequence of this is the high cost of conducting clinical trials on oncolog-ical drugs that provide sufficient efficacy and safety data to support regulatory approval.

P E R S P E C T I V E S

206 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Standard of care: group A or B

Random selectionto group A or B

Outcomes

Biomarker-guidedselection to group A or B

Outcomes

Box 3 | Biomarkers in clinical trials of new drugs and of widely used drugs

In the development of new drugs, the clinical effects of the drug are evaluated rather than the biomarker classifier (that is, the biomarker-based measurement used to characterize the patients’ disease); therefore, definitive trials often randomize patients to biomarker-positive and biomarker-negative arms. This establishes the efficacy spectrum of the drug and establishes the eventual claims that will be made for the use of the drug. However, if the biology of the new drug and its specificity for a biomarker classifier is robust, it is sometimes efficient to initially evaluate only biomarker-positive patients in Phase Ib and Phase II proof-of-concept studies. By contrast, in clinical trials of approved therapies, especially to advance personalized medicine, the biomarker classifier (rather than the drug) is evaluated against known clinical outcomes (see figure below).

Note that standardized imaging, circulating tumour cells or any initially qualified response biomarker will result in an efficient measure of response to therapies and correlative clinical outcomes with both new drugs and approved therapies.

Using imaging-based measurements of cancer progression, incremental progress has been made in the evaluation of solid cancers and in the delay of cancer progres-sion (examples include hormonal therapy for the treatment of breast cancer and newer molecularly targeted therapies such as vemurafenib and crizotinib). Based on scientific and public health considerations, there is high interest in the oncological drug research and development community to define the developmental pathway that is required to use surrogate end points for sup-porting drug approvals. Intensive ongoing activities in industry and in the public sector will provide the data and examples needed for definitive guidance on this subject81.

Biomarker qualificationThe FDA Critical Path Initiative highlighted the need for — and the utility of — clearly defined, well-characterized biomarkers as tools for drug development. Key com-ponents of the process of validation and qualification of these biomarker tools are summarized in BOX 1, and the factors that determine whether a biomarker is ready for use in clinical studies are described in BOX 2. Although the FDA does not prescribe what data are required for qualifying a biomarker, general guidelines include: the need for data from multiple clinical studies; strong corre-lation of biomarker expression with clinical status, risk or outcomes (for example, overall survival or long-term disease-free survival); and a high correlation of biomarker expres-sion with the clinical status of the patient or end point of the clinical trial.

The qualification of biomarkers for use in the development of oncology drugs is best understood by examples. In this section, we summarize ongoing efforts to qualify CTCs and two imaging methodologies — volumetric CT and FDG–PET/CT imaging — as response biomarkers to measure drug efficacy, and potentially as surrogate end points that will support the regulatory approval of oncology drugs. Several examples described below are related to the design of clinical trials that can be used to evaluate biomarkers, and can lead to the qualification of the biomarker for classifying disease as well as predicting response to therapy.

CTCs. Because of their presence in cancer, and the information they contain about the molecular and phenotypic characteristics of the tumours from which they originate, CTCs have attracted interest as biomarkers. Enumeration of CTCs — as measured by the CellSearch assay — has been cleared

by the FDA as a correlate for poor prognosis in patients with breast, prostate, colon and lung cancers. There are ongoing efforts to qualify the measurement of CTCs for deter-mining patient response and drug efficacy in metastatic, castration-resistant prostate can-cer78. Although results are promising, issues related to sensitivity and enumeration with-out functional characterization of the cells may limit the generalizability of the assay. Newer technologies — such as fibre-optic scanning, filtration and electrophoretic sepa-ration methods, and microfluidic techniques82 — will allow better definition of the charac-teristics of CTCs (for example, by identifying cells that are undergoing or have undergone epithelial to mesenchymal transition)4.

Imaging-based biomarkers. Imaging-based biomarkers are used for staging, re-staging and monitoring the treatment of patients. Anatomical imaging that uses CT and uni-dimensional measurements of tumour size has been indispensable in establishing the RECIST criteria as a standard for assessing response to therapy77, but for many reasons there has been great interest in carrying out translational research to improve these cri-teria. Such reasons include the complexity of the biology and heterogeneity of tumour masses, the precision of measurement that is needed, the possibility of misclassification (especially near the cut-off points) and the considerable interest in targeted cytostatic therapy, which requires improved measure-ments of stable disease. The analysis and validation of data collected from studies that use volumetric CT is one of the ongoing

initiatives to improve on existing methods, and if this work shows that volumetric CT provides advantages over unidimensional measurements it may be incorporated into an improved version of the RECIST criteria83,84.

PET imaging and PET/CT imaging using18F-FDG has been evaluated in several studies, and evidence suggests that PET/CT imaging offers diagnostic advantages over other imaging modalities for detecting the presence of major cancers, including lym-phoma, lung cancer, head and neck cancer, oesophageal cancer, colorectal cancer, breast cancer and pancreatic cancer as well as gastro-intestinal stromal tumours53. After a patient has received treatment, FDG–PET imaging is used to detect recurrent or residual disease, or to determine the extent of a known recur-rence, and this method is approved by the FDA for this purpose. Based on available data, it is also anticipated that FDG–PET imaging will be more useful than standard anatomical measurements for detecting the activity of molecularly targeted drugs53,57,58.

Despite the large body of data providing a rationale for the use of FDG–PET imaging as a biomarker for measuring the efficacy of many drugs and in many cancers, chal-lenges remain. Most data generated using FDG–PET imaging are from qualitative assays conducted without standard protocols for acquiring and reporting data. The use of semi-quantitative SUVs in FDG–PET studies could counteract variations between protocols. Decrement of SUVs in early stages of treatment will probably enable the evaluation of response to therapy in Phase II and Phase III studies. Indeed, retrospective

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 207

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Patients with NSCLC(Asian non-smokers;Asian light smokers)

Registration Randomization OutcomeStratification

EGFR mutation

Gefitinib

Biomarker testing for EGFR mutation status

Paclitaxel andcarboplatin

Primary end point:progression-freesurvival

+

EGFR mutation –

EGFR mutation +

EGFR mutation –

Box 4 | Clinical trial design with analysis stratified by biomarker status

The Iressa Pan-Asia Study (IPASS)95,96 (see figure) analysed clinical trial results according to patient biomarker status.• A randomized Phase III study comparing gefitinib (Iressa; AstraZeneca) with standard paclitaxel

and carboplatin therapy was carried out in 1,217 Asian patients with non-small-cell lung cancer (NSCLC) who were non-smokers (defined as <100 cigarettes during their lifetime) or former light smokers (defined as individuals who stopped smoking ≥15 years previously, and smoked ≤ 3,650 packs of cigarettes during their lifetime) and had not received prior chemotherapy.

• Patients received either gefitinib (250 mg per day orally) or standard paclitaxel and carboplatin (given by infusion every 3 weeks for up to six cycles) until one of the following occurred: the disease progressed (the primary end point); unacceptable toxicity was reached; the patient or physician requested stopping; there was non-compliance with protocol; or once six cycles of chemotherapy were completed.

• Tumour samples from consenting patients were assessed for biomarkers and particularly for mutations in the epidermal growth factor receptor (EGFR) gene. Patients were considered to be positive for an EGFR mutation if 1 out of 29 mutations was detected. Patients were not randomized on the basis of biomarker status, but the clinical outcomes were.

• The study showed that gefitinib was superior (24.9% of patients had progression-free survival at 12 months) to standard therapy (6.7%) in the total study population.

• In the 261 patients who were positive for EGFR mutations, progression-free survival or overall survival in those patients who were treated with gefitinib was significantly higher than in those patients who were treated with standard therapy (hazard ratio = 0.74, P <0.001); the converse was observed in 176 patients with no EGFR mutations (hazard ratio = 2.85, P <0.001).

studies in patients with lung cancer have shown that an SUV decrement of ≥25% after two cycles of standard cytotoxic chemo-therapy correlates with better survival55, and recent studies using EGFR inhibitors have demonstrated a correlation between response to therapy and clinical benefit57,58. Additional studies will be required to dem-onstrate whether these effects are specifically correlated with drug treatment, and are con-sequently useful for discriminating between effective and ineffective treatments (FIG. 2).

Biomarker-based study designsNew oncology drugs that have been tested in clinical trials have a low success rate and long development times (>10 years)5; this has prompted interest in clinical trial strate-gies that will evaluate drug efficacy and safety in a more definitive, quick and effi-cient manner85. Biomarkers that characterize disease and biomarkers that characterize

response have equal importance in attaining this goal. The multiple parameters that define subsets of cancer and predict the efficacy of drugs in these subsets suggest that the development of complex biomarker-based measurements or molecular signatures such as gene expression patterns could be valuable for defining groups of patients who are best suited to specific clinical trials.

Here, we discuss innovative biomarker-based clinical trial designs and method-ologies for defining, evaluating and using biomarker classifiers (that is, the biomarker-based measurements used to characterize the patients’ disease). Although most clinical trial designs discussed in this article have been used to evaluate single-molecule bio-markers, they could also be used for the evaluation of complex molecular signatures. These clinical trial designs include: the ran-domization or analysis of patients who have been stratified by the presence or absence

of a biomarker (or biomarkers); enrichment designs and biomarker-based assignment of specific drug therapy; and adaptive clinical trial designs. Newer enabling technologies such as next-generation gene sequencing aug-ment the ability to define and use complex biomarker classifiers. The trial designs presented in this article exemplify this active field, and illustrate the diversity and merit of methodologies that vary depending on the knowledge of the drug or biomarker.

During the development of new drugs, clinical trials that use biomarkers to select patients are designed to determine the effect of the drug on the patient rather than the effect of the drug on the biomarker, and so definitive trials are often stratified (for randomization of patients and/or analysis of data) based on the presence or absence of the biomarker. These studies are intended to establish the range of efficacy of the drug and to support eventual claims made for its use. However, if knowledge of the biology of the new drug suggests that its efficacy is strongly correlated with the presence of the biomarker, then enriching Phase II proof-of-concept studies (and possibly registration studies) with patients who are biomarker-positive may be an efficient way of evaluating the drug. When biomarkers are used to advance per-sonalized medicine, approved drugs are used to evaluate how the biomarker changes in response to the drug (BOX 3).

Statistical guidelines for determining the sample size can be used to weigh the advantages of stratified versus enrichment designs for biomarker-based studies86,87. These guidelines are based on anticipated response to treatment and/or the clinical outcome of patients who are biomarker-positive as well as the percentage of patients in the study population who are anticipated to be biomarker-positive88. Adaptive designs provide an alternative and potentially more efficient and cost-effective approach for evaluating both new drugs and biomarker classifiers. In adaptive studies, data that accu-mulate during the study are reviewed on an interim basis and used to change features of the study to reflect what is learned during the study. The study designs highlighted in this article are prospective clinical trials involving biomarker measurements that do not drive therapy (such as the MARVEL and IPASS clinical trials) as well as biomarker measure-ments that do drive therapy (such as the I-SPY 2 and BATTLE trials).

Retrospective analysis of biomarkers in patient samples from studies in which well-annotated samples have been col-lected prospectively may also be useful in

P E R S P E C T I V E S

208 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Biopsy sample,blood sample

Studypatients

MRI MRI, biopsy sample

Paclitaxel +/– new drug C, D or E

MRI, biopsy sample

HER2

Surgery

12 weekly cycles

Adapt Doxorubicin and cyclophosphamide

Doxorubicin and cyclophosphamide

4 weeklycycles

4 weeklycycles

+

MRI, blood sample

Tissuesample

HER2 –

Paclitaxel and trastuzumab +/– new drug A, B or C

Randomization

Randomization

Box 5 | Adaptive randomization

The I-SPY Trial 2 (I-SPY2)63 (see figure) is using an adaptive clinical trial design for the simultaneous testing of matched diagnostics and therapeutics.• The I-SPY 2 trial, which began in 2010 and is anticipated to accrue 720 patients over 4 years,

is a randomized Phase II adaptive trial design in the setting of neoadjuvant treatment for locally advanced breast cancer.

• The primary objective of the trial is to determine whether experimental drug regimens added to standard therapy increase the probability of achieving a pathological complete response, when compared with standard neoadjuvant therapy, for each prospectively identified molecular biomarker signature. The trial also aims to predict the probability of a successful Phase III trial for each drug regimen and molecular signature combination.

• The initial molecular signatures are composed of combinations of hormone receptor, HER2 (also known as ERBB2) and MammaPrint61 expression; 6–10 drug regimens representing different drug classes will be evaluated against approximately ten molecular signatures representing the disease most relevant for therapeutic intervention; the primary end point is the pathological complete response; the volume of the tumour, measured by magnetic resonance imaging (MRI), will also be evaluated as a response biomarker.

• One arm will receive standard neoadjuvant chemotherapy, beginning with weekly doses of paclitaxel (plus trastuzumab for patients who are HER2-positive) followed by doxorubicin and cyclophosphamide chemotherapy. The other arm will receive standard therapy with up to five new drug regimens evaluated simultaneously during the paclitaxel treatment period.Initially, randomization to new drug regimens is balanced. As results become available,

predictive indexes are built for each drug regimen and molecular signature combination, and randomization is adapted towards assigning new patients to drug regimens that are predicted to benefit them based on their molecular signatures. Drug regimens that do not sufficiently benefit patients with any of the signatures can be identified by testing them in approximately 60 patients and then dropping them from the study. Drug regimen and molecular signature combinations that are likely to be successful in Phase III trials can be identified by testing them in ≤120 patients.

developing evidence for the clinical utility of a biomarker. The specific design, utility, advantages and disadvantages of innovative study designs have been discussed in the literature (see REFS 22,86,87,89–94).

Randomization or analysis stratified by bio-marker status. These designs, which may be used in definitive Phase III trials, evaluate the correlation between the patient’s response to drug treatment (or clinical outcome) and bio-marker status in both biomarker-positive and biomarker-negative patients. This is achieved either by randomizing patients according to their biomarker status or by carrying out a pre-planned analysis of study results using the biomarker status of the patients. The MARVEL trial87 (Supplementary information S1 (box)), for example, randomized patients according to their biomarker status; although this design has high merit at certain stages of drug devel-opment, the MARVEL trial was terminated early because the biomarker-specific drug activity was established in other trials.

Randomization of patients according to biomarker status has merit in comparisons of efficacy and utility. Because biomarker status is measured at the baseline, a balanced number of biomarker-positive and biomarker-negative patients can be accrued, and the numbers of biomarker-positive and biomarker-negative cases are known. However, the results of the clinical trial may not be produced before interest in the study drugs has waned or until assays of the bio-marker have evolved to reflect new technol-ogy and an updated understanding of biology. In addition, because of the high numbers of patients that could be required, and the possi-ble difficulty in accruing patients (especially if the biomarker has relatively low prevalence), using this strategy with more than two or three treatment regimens is not practical.

In the alternative stratification design (IPASS95,96; BOX 4), participants are rand-omized to different treatments without regard to biomarker status, and biomarker stratification is applied during the analysis of the results. This alternative design provides the opportunity for exploratory analyses of a range of stratifications according to various measurements of biomarker status to improve the definition of the biomarker. A disadvan-tage of this design, however, is that there is no guarantee of balanced populations for analysis.

Enrichment designs and biomarker-based assignment of specific drug therapy. Enrichment studies are primarily useful in clinical trials of drugs that use companion diagnostics87,97. In these trials, the biomarker

status of all patients who are enrolled in the clinical trial is usually assessed but only biomarker-positive patients are randomized and studied. Enrichment designs can be particularly useful in Phase II studies if a correlation between the test biomarker and drug activity has been established in preclinical studies. If the evidence is par-ticularly strong (for example, in the case of HER2 expression, which is associated with response to trastuzumab in breast cancers), registration studies and accelerated approval in biomarker-positive patients may be pos-sible. It should be noted that when data show a correlation between a biomarker and drug activity, it can be ethically problematic to

randomize biomarker-positive patients to non-targeted standard therapy.

Enrichment studies enable a rapid accu-mulation of efficacy data; despite this, there are complications related to not obtaining information on biomarker-negative patients. That is, nothing is learned about the effects of the drug in biomarker-negative patients, some of whom might benefit from the drug. This suggests the merit of running a subset of the trial in all patients before restricting randomization to biomarker-positive patients. A further consideration for enrichment trials that use companion diag-nostics is that predictive biomarkers have a prognostic component (for example, many

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 209

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

All patients

RegistrationRandomization

Outcome Stratification

Analysis for best biomarkerclassifier

Tissue collectedfor biomarker analysis

Standardtherapy

Newtherapy

Overallsurvival

Box 6 | Molecular signature design

Clinical trials that use adaptive analysis allow for the simultaneous testing of matched diagnostics and therapeutics (see figure).• These are pivotal Phase III clinical trials — standard-of-care versus a new therapy — in which the

primary end point is overall survival.

• Tissue samples from patients are collected at baseline but not analysed until the trial is near completion.

• The ‘most likely’ biomarker classifiers (that is, the biomarker-based measurements used to characterize the patients’ disease) are chosen in early stages of drug development, and analytical validation of the corresponding assays is well developed when the trial is commenced.

• At completion of the trial, tissue samples are analysed for a set of biomarkers using all possible biomarker combinations to derive a classifier composed of the biomarker combination that correlates best with overall survival.

• A P = 0.05 probability that the difference in survival between patients who were treated with the new therapy versus those on standard therapy is not due to chance is required to demonstrate that the new therapy is more efficacious. For analysis of the results, the P value required is split between P = 0.01 for analysis of results in all patients regardless of the classifier and, if the analysis in all patients does not show the new therapy to be more efficacious, P = 0.04 for the analysis of the results in patients with the classifier, which is identified by a ‘training set’ and validated by a test set of patients.

• The goal of such trials is to achieve regulatory registration of the new drug in at least the cohort of patients identified by the subset classifier.

oestrogen receptor-positive breast cancers have favourable prognoses but they may also be specifically predicted to be responsive to antihormone therapy98), and so randomi-zation to different therapies by biomarker status is needed to understand the degree to which the biomarker can predict a response to specific treatments.

When the correlation between biomarker-positive status and response to therapy is compelling, and sufficient data from biomarker-negative patients exist, biomarker-based assignment of drug classes becomes the preferred trial design. This is because the slow accrual of patients is the primary rate-limiting step in the completion of Phase II and III clinical trials, and accrual is expected to be particularly challenging for the poten-tially small patient populations identified by enrichment studies based on the molecular profiles of tumours of individual patients. Studies are being carried out to address this issue.

For example, in the second phase of the BATTLE study99–101, eligible patients (that is, patients with stage IIIB, stage IV or

advanced incurable NSCLC, who received at least one prior treatment) were enrolled into an umbrella clinical trial in which molecular biomarker analysis was performed on their tumours. Based on their specific tumour biomarkers, patients were assigned to one of four Phase II trials that was most likely to provide benefit. The Lung Cancer Mutation Consortium has taken this concept one step further102. It has analysed specific driver mutations in >1,000 patients with advanced lung cancer, with the intent of developing a library of mutational patterns and associated validated assays for matching patients with therapies that may be beneficial in lung cancers caused by these mutations.

Adaptive designs. Clinical trials with adaptive designs can be considered to be an advanced form of enrichment design (reviewed in REFS 89,103; see the report entitled “Guidance for Industry: Adaptive Design Clinical Trials for Drugs and Biologics” on the FDA website). Changes to the number of patients, treatment alloca-tion and randomization methods, eligibility

criteria, schedule of patient evaluations, clinical trial end points and the number of interim analyses, as well as the selection of treatment regimens, can be made during studies with adaptive designs. The major advantage of such trials is that they allow investigators to incorporate knowledge that was not available at the start of the study into the design of the trial, and it is widely believed that adaptive designs may be more efficient than traditional designs, requiring fewer patients and shorter times to produce meaningful information about study drugs, end points and biomarkers.

Many adaptive designs are based on Bayesian statistics89 (see the report entitled “Guidance for Industry: Adaptive Design Clinical Trials for Drugs and Biologics” on the FDA website); that is, they determine the probabilities of parameters taking on given values. These probabilities can be recalculated continually as new data on study parameters become available, and then be used to optimize the information obtained from the study89.

For example, the I-SPY 2 trial63 (BOX 5) is a prototype adaptive randomization design that takes patients’ biomarker profiles into account. This is an intervention clinical trial of neoadjuvant treatment of locally advanced breast cancer that compares several drugs; patients are classified initially on the basis of values of selected sets of biomarkers, and then randomized to the various drug treat-ments. As efficacy data become available, they are used to recalculate the probabilities of the drugs being active in the different biomarker profiles, and the new probabilities in turn are used to modify the randomiza-tion scheme so that newly accrued patients receive treatments that are likely to be more effective based on the accumulating response data. The design leads to a more efficient evaluation of the drugs, and it is anticipated that the subsequent Phase III trials in these targeted populations will be considerably smaller, quicker and less costly than standard Phase III trials.

A second example of Bayesian adaptive randomization is a pilot trial in the BATTLE programme99,101 (Supplementary information S1 (box)). In this study, biopsy samples were taken at the baseline from patients with lung cancer to evaluate the molecular signatures of their cancers, and the patients were ini-tially randomized to one of four molecularly targeted drug treatments, without regard to molecular events shown in the biopsy samples. After fewer than 100 patients were evaluated, their results were used to adjust randomization so that new patients were

P E R S P E C T I V E S

210 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Table 2 | Opportunities and challenges for biomarker-based drug development

Opportunities Challenges

The use of high‑content screening assays, next‑generation sequencing and low‑cost genome sequencing

Matching data to outcomes; management of the high volume of data generated

Increasing the importance of cytogenetic, epigenetic and proteomic data

A requirement for the development of further standards

Co‑development of diagnostics and therapeutics

More efficient coordination of research and development

Clarification of regulatory pathways

Regulatory approvals based on molecular disease versus target organ disease

International harmonization and standardization

Economic and cultural differences

Multisector cooperation Precompetitive and intellectual property issues

Increasing interest and clarity in the clinical utility of biomarkers

Access to complete data sets that are associated with clinical outcomes

Increasing the efficiency and sophistication of clinical trial designs

Mitigating the risk of false positives and false negatives

There are increasing numbers of molecular targets and analytically validated assays

Clearance of the assay by the US Food and Drug Administration versus access to the assay facilited by Clinical Laboratory Improvement Amendments; coverage of the assay by the patient’s health insurance policy

randomized to treatment based on the results of baseline biopsy samples and the results of previously treated patients.

Another example of a clinical trial design for developing drugs and biomarkers is the adaptive signature design88,104,105. This design has the benefit of qualifying a biomarker (for which an analytically validated assay exists) at the same time as a Phase III drug approval trial is being run. In this design, a Phase III randomized, controlled clinical trial of an investigational drug is carried out in a defined population of patients. If the results of the study do not demonstrate a statistically significant benefit of the drug in the whole study population, then the study population is divided into two groups: a signature validation group and a signature development group. Using the signature development group, a single biomarker signature is developed that best identifies those patients who respond to the test drug better than to the control therapy. The signature validation group then compares the outcomes of patients who have a positive biomarker signature on the test drug with those who have a positive biomarker signature on the control therapy. A variation of this design is shown in BOX 6. This trial design is promising as it utilizes an end point that has unequivocal clinical benefit and could probably result in the regulatory approval of a new drug in a molecularly defined cohort of individuals.

A final example of adaptive designs in clinical trials is response-adaptive designs, wherein the therapy that individual patients receive is changed during the trial based on real-time measures (taken at early stages in the trial) of response or non-response to the initial drug therapy. Unlike the other studies described in this section, response biomarkers rather than predictive biomarkers are evaluated in this adaptive clinical trial design. These designs are frequently used when alternative, effective and approved (or experimental) drugs (or drug regimens) are available and there is also high confi-dence in the validity of the response bio-marker. The best current examples of this design are the clinical trials for Hodgkin’s lymphoma (ClinicalTrials.gov identifier: NCT00822120) and diffuse large B cell lymphoma106, in which assessments of early response are provided by functional and anatomical imaging. Examples of variations in response-adaptive designs (for example, randomized discontinuation design) have been discussed in the literature (reviewed in REFS 97,104,107,108).

Lots of promise, lots of challengesThe promise of biomarker science and its successful application resides in the numer-ous scientific disciplines and projects that contribute to cancer research. As cancer is a genetic disease, the success achieved to date

from the pilot projects run by The Cancer Genome Atlas study that classify the disease in the brain45,109, ovary110 and lung — as well as new work by this group that is underway to sequence more than 20 types of cancer — is likely to identify key driver mutations111. These efforts are anticipated to better enable the identification and treatment of molecu-larly based subsets of disease. Technological advances will also contribute to the success of The Cancer Genome Atlas projects. These advances include: next-generation gene sequencing, with its high rate of output; the high degree of coordination among studies, including the development of standards and methods; new bioinformatics capabilities for handling large quantities of data; and collab-oration among multidisciplinary scientists.

Disease severity and progression rates brought about by genetic changes depend on factors such as the type and stage of the tumour as well as the tumour microenviron-ment. In addition, the analysis of the epigenome has emerged as an area of high research interest and promise, as a result of confounding the predictions of tumour progression and response to treatment that are based solely on the primary genetic sequence of the tumour112. The inherent heterogeneity of a growing tumour mass, which is derived from many years of clonal evolution, generates challenges for genome-based prediction of tumour behaviour as well as for the selection of an appropriate therapy. This complexity, which has been verified using genome sequencing data111, creates obstacles both for drug development and for the management of patients with cancer.

These challenges are being met in part using several new technologies that are designed to provide integrative measure-ments of the biological phenotype of cancer. These technologies include single-cell phosphoprotein analysis, technologies that enumerate and characterize CTCs, and adoptions and improvements in molecular imaging methods. This exciting research has a profound impact on the progress of biomarker-based research; namely, it enables an improved selection of patients who are then matched to specific therapies as well as an improved ability to measure patient responses to therapy in early stages of their treatment.

The net result of improved biomarker-based science is the observation of larger effects and an improved ability to detect response to therapy in the arms of randomized trials that contain biomarker-targeted therapy compared with clinical trial arms that use conventional therapy.

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 211

© 2012 Macmillan Publishers Limited. All rights reserved

Glossary

Adaptive clinical trial designsDefined by the US Food and Drug Administration as study designs that include prospectively planned opportunities for the modification of one or more specified aspects of the study design and hypotheses based on analyses of data (usually interim data).

Analytical and clinical validationThe process of assessing the biomarker assay and measuring of its performance characteristics, and determining the range of conditions (including clinical settings) under which the assay will give reproducible and accurate data.

Biomarker qualificationThe evidentiary process of linking a biomarker with biological processes and clinical end points. Qualification refers to the verification that the biomarker is ‘fit for purpose’.

BiomarkersFactors that are objectively measured and evaluated as indicators of normal biological or pathological processes, or are pharmacological responses to therapeutic intervention.

BRAFV600E mutationA mutation in the BRAF gene leading to valine being substituted by glutamate at codon 600; found in human cancers such as papillary thyroid carcinoma, colorectal cancer, melanoma, non-small-cell lung cancer and hairy cell leukaemia.

Clinical Laboratory Improvement Amendments(CLIA). US legislation defining quality assurance practices in clinical laboratories, and requiring them to measure performance at each step of the testing process from the beginning to the test result.

Companion diagnosticsIn vitro diagnostic devices (assays) that provide information that is essential for the safe and effective use of a corresponding therapeutic product. The use of a companion diagnostic with a particular therapeutic product is stipulated in the instructions for use in the labelling of both the diagnostic and the corresponding therapeutic product, as well as in the labelling of any generic equivalents of the therapeutic product.

DNA ploidyA term used to describe the number of chromosome sets or deviation from the normal number of chromosomes in a cell.

Driver mutationsMutations that are causally associated with cancer development. Their presence is not required for maintenance of the final cancer (although it often is) but it must have been selected at some point during cancer development.

Epigenetic modulationHeritable changes in gene expression or cellular phenotype caused by mechanisms of interaction with the genome other than changes in the underlying DNA sequence. Examples of such changes include DNA methylation and histone deacetylation, both of which serve to suppress gene expression without altering the sequence of the silenced genes.

Epithelial to mesenchymal transitionA cellular transition associated with cancer metastases in which an epithelial cell loses polarity and intercellular adhesion, and degrades basement membrane components to become a migratory mesenchymal cell.

Fit for purposeA term used in the context of biomarker qualification. A biomarker is qualified on the basis of data that support its use in specific contexts — that is, evidence linking the biomarker to specific biology and clinical end points. The biomarker is considered to be ‘fit for purpose’ or qualified for use in those settings in which the supporting evidence is sufficiently robust.

Gail risk modelAn algorithm formulated by Mitchell Gail (US National Cancer Institute) that uses personal and family history to estimate a woman’s absolute risk of developing breast cancer.

Investigational new drug applicationAn application that is submitted to a regulatory agency before a drug can be studied in humans. This application contains experimental data on: how, where and by whom the new studies will be conducted; the chemical structure of the drug; the drug’s mechanism of action and metabolism; any toxic effects; and how the compound is manufactured.

Next-generation gene sequencingNew technologies for high-throughput and high-speed sequencing, and potentially cost-effective sequencing of the human genome, epigenome and transcriptome (the part of the genome transcribed into RNA).

Passenger mutationsMutations that are not selected during cancer development and do not directly contribute to cancer development. Cells that acquire driver mutations already carry biologically inert somatic passenger mutations, which are included in the clonal expansion that follows and will be present in all cells of the resulting cancer.

Standard uptake value(SUV). A value that quantifies positron emission tomography imaging; calculated as a ratio of tissue radioactivity concentration at a specific time and injected dose at the time of injection, divided by body weight.

Umbrella clinical trialA clinical trial protocol into which patients are enrolled, their biomarker status is obtained and then they are assigned to treatment in one of a group of clinical trials of targeted drugs. Patients are assigned to the drug that is most likely to benefit them based on their biomarker status.

Warburg phenomenonThe observation that cancer cells, unlike normal cells, preferentially utilize the glycolytic pathway rather than the Kreb’s cycle to metabolize glucose.

This has and will continue to increase the number of drugs that are approved for oncology indications based on smaller and faster clinical trials. An additional

challenge in the development of oncol-ogy drugs is the plasticity of cancer, which results in resistance to specific therapies in individual patients and requires

extensive biomarker-based research as well as improved drug combination strategies to overcome mechanisms of drug resistance.

The development of biomarkers has been considerably enhanced by initiatives from the FDA that seek to clarify regulatory science, such as the FDA Critical Path Initiative and documents that provide guidance on the qualification of biomarkers as drug development tools (see the report entitled “Draft Guidance for Industry: Qualification Process for Drug Development Tools” on the FDA website). This report separates the definitions and data requirements for the analytical and clinical validation of assays that are used for measuring biomarkers from those for used for the clinical qualifica-tion of biomarkers. Thus, biomarker-based research and the regulatory strategy can be focused on the linkage of the biomarker data to important clinical outcomes, providing a path for evidentiary standards for use in specific contexts to be developed and agreed upon.

One practical consequence of the FDA initiatives is the ability to develop a com panion diagnostic that is linked to the investigational new drug application of a specific drug, whereby the use of the diagnostic is limited to the specific drug. This requires a different set of rigorous data than for standard diagnostics. One challenge provided by companion diagnostics is the estimated 3-year timeframe required for the co-development of the companion diagnostic with the drug, as well as the attendant expense and risk of developing a failed drug. Another challenge is that the biomarker can be easily marketed via Clinical Laboratory Improvement Amendments (CLIA)-certified laboratories before adequate clinical utility has been established, which can result in potential misuse of the biomarker in patient management, even though the biomarker may be analytically valid.

The alternative path to market is the clearance of the biomarker by the FDA, and although the data required to achieve this are more rigorous this does not mean that the test has clinical utility. Conversely, tests that have not received clearance by the FDA may have clinical utility (for example, EGFR mutations in lung cancer). An additional confounding factor is the reimbursement, from health-care providers, of these assays without solid evidence of their clinical utility. With the increasing costs of health care, there is intense interest in better coordinating health insurance coverage with the use of approved assays and more efficacious targeted drugs113.

P E R S P E C T I V E S

212 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Within the development and uses of the biomarkers outlined in this article, there is a key role for public–private partnerships. These include: partnerships among academic institutions, the US National Cancer Institute and international govern-ment institutions to perform research that results in better clinical data and more effec-tive treatment and management of patients; partnerships between the FDA and other regulatory agencies to seek and analyse the best data and evidence on which to build guidance and regulatory policy; and partner-ships between the diagnostics and devices industry to develop assays that enable better characterization of patients and allow the industry to build improved business models (including for companion diagnostics) in which the value of their products can be realized.

The pharmaceutical and biotechnology industries have an important role in defining the criteria needed to determine the early responses of tumours to therapy, and they have an important role in designing more efficient drug development processes around biomarkers. Patients and patient groups must advocate receiving treatment and manage-ment that is more effective, and they must act as an educational force for the potential value of personalized treatment to be recognized. The role of payers and providers of health care is to obtain better data related to the effectiveness and clinical utility of biomarkers. Finally, the conveners of public–private partnerships have an indispensable role in the governance and funding of these partner-ships. In summary, the promise of biomarker research, which clarifies the opportunities and challenges that will define future efforts in this area, is now being realized (TABLE 2).

Gary J. Kelloff is at the National Cancer Institute, Cancer Imaging Program, EPN Rm 6058, 6130

Executive Blvd, Bethesda, Maryland 20852, USA.

Caroline C. Sigman is at CCS Associates, 1923 Landings Drive, Mountain View,

California 94043, USA.

e-mails: kelloffg@mail.nih.gov; csigman@ccsainc.com

doi:10.1038/nrd3651 Published online 10 February 2012

1. Hanahan, D. & Weinberg, R. A. The hallmarks of cancer. Cell 100, 57–70 (2000).

2. Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646–674 (2011).

3. Luo, J., Solimini, N. L. & Elledge, S. J. Principles of cancer therapy: oncogene and non-oncogene addiction. Cell 136, 823–837 (2009).

4. Polyak, K. & Weinberg, R. A. Transitions between epithelial and mesenchymal states: acquisition of malignant and stem cell traits. Nature Rev. Cancer 9, 265–273 (2009).

5. Paul, S. M. et al. How to improve R&D productivity: the pharmaceutical industry’s grand challenge. Nature Rev. Drug Discov. 9, 203–214 (2010).

6. Polyak, K., Shipitsin, M., Campbell-Marrotta, L., Bloushtain-Qimron, N. & Park, S. Y. Breast tumor heterogeneity: causes and consequences. Breast Cancer Res. 11 (Suppl. 1), 18 (2009).

7. Polyak, K. & Kalluri, R. The role of the microenvironment in mammary gland development and cancer. Cold Spring Harb. Perspect. Biol. 2, a003244 (2010).

8. Pegram, M. D., Pauletti, G. & Slamon, D. J. HER-2/neu as a predictive marker of response to breast cancer therapy. Breast Cancer Res. Treat 52, 65–77 (1998).

9. Druker, B. J. Perspectives on the development of imatinib and the future of cancer research. Nature Med. 15, 1149–1152 (2009).

10. Vultur, A., Villanueva, J. & Herlyn, M. Targeting BRAF in advanced melanoma: a first step toward manageable disease. Clin. Cancer Res. 17, 1658–1663 (2011).

11. Chapman, P. B. et al. Improved survival with vemurafenib in melanoma with BRAF V600E mutation. N. Engl. J. Med. 364, 2507–2516 (2011).

12. US Food and Drug Administration. FDA labeling information — Zelboraf. FDA website [online], http://www.accessdata.fda.gov/drugsatfda_docs/label/2011/202429s000lbl.pdf (2011).

13. Soda, M. et al. Identification of the transforming EML4–ALK fusion gene in non-small-cell lung cancer. Nature 448, 561–566 (2007).

14. Kwak, E. L. et al. Anaplastic lymphoma kinase inhibition in non-small-cell lung cancer. N. Engl. J. Med. 363, 1693–1703 (2010).

15. Cui, J. J. et al. Structure based drug design of crizotinib (PF-02341066), a potent and selective dual inhibitor of mesenchymal-epithelial transition factor (c-MET) kinase and anaplastic lymphoma kinase (ALK). J. Med. Chem. 54, 6342–6363 (2011).

16. US Food and Drug Administration. FDA labeling information — Xalkori. FDA website [online], http://www.accessdata.fda.gov/drugsatfda_docs/label/2011/202570s000lbl.pdf (2011). 

17. Dancey, J. E. Epidermal growth factor receptor inhibitors in non-small cell lung cancer. Drugs 67, 1125–1138 (2007).

18. Park, J. W. et al. Rationale for biomarkers and surrogate end points in mechanism-driven oncology drug development. Clin. Cancer Res. 10, 3885–3896 (2004).

19. Gutman, S. & Kessler, L. G. The US Food and Drug Administration perspective on cancer biomarker development. Nature Rev. Cancer 6, 565–571 (2006).

20. Amur, S., Frueh, F. W., Lesko, L. J. & Huang, S. M. Integration and use of biomarkers in drug development, regulation and clinical practice: a US regulatory perspective. Biomark. Med. 2, 305–311 (2008).

21. Majewski, I. J. & Bernards, R. Taming the dragon: genomic biomarkers to individualize the treatment of cancer. Nature Med. 17, 304–312 (2011).

22. Dancey, J. E. et al. Guidelines for the development and incorporation of biomarker studies in early clinical trials of novel agents. Clin. Cancer Res. 16, 1745–1755 (2010).

23. Merlo, L. M., Pepper, J. W., Reid, B. J. & Maley, C. C. Cancer as an evolutionary and ecological process. Nature Rev. Cancer 6, 924–935 (2006).

24. Anderson, L. Candidate-based proteomics in the search for biomarkers of cardiovascular disease. J. Physiol. 563, 23–60 (2005).

25. Kelloff, G. J. et al. Progress in chemoprevention drug development: the promise of molecular biomarkers for prevention of intraepithelial neoplasia and cancer — a plan to move forward. Clin. Cancer Res. 12, 3661–3697 (2006).

26. Sidransky, D. Emerging molecular markers of cancer. Nature Rev. Cancer 2, 210–219 (2002).

27. Vogelstein, B. et al. Genetic alterations during colorectal-tumor development. N. Engl. J. Med. 319, 525–532 (1988).

28. Le, Q. T. & Giaccia, A. J. Therapeutic exploitation of the physiological and molecular genetic alterations in head and neck cancer. Clin. Cancer Res. 9, 4287–4295 (2003).

29. Ilyas, M., Straub, J., Tomlinson, I. P. & Bodmer, W. F. Genetic pathways in colorectal and other cancers. Eur. J. Cancer 35, 1986–2002 (1999).

30. Fearon, E. R. & Vogelstein, B. A genetic model for colorectal tumorigenesis. Cell 61, 759–767 (1990).

31. Dillon, D. A., Howe, C. L., Bosari, S. & Costa, J. The molecular biology of breast cancer: accelerating clinical applications. Crit. Rev. Oncog. 9, 125–140 (1998).

32. Leslie, N. R. & Downes, C. P. PTEN function: how normal cells control it and tumour cells lose it. Biochem. J. 382, 1–11 (2004).

33. Parsons, R. Human cancer, PTEN and the PI-3 kinase pathway. Semin. Cell Dev. Biol. 15, 171–176 (2004).

34. Sansal, I. & Sellers, W. R. The biology and clinical relevance of the PTEN tumor suppressor pathway. J. Clin. Oncol. 22, 2954–2963 (2004).

35. Vande Woude, G. F. et al. Reanalysis of cancer drugs: old drugs, new tricks. Clin. Cancer Res. 10, 3897–3907 (2004).

36. Maley, C. C. et al. Selectively advantageous mutations and hitchhikers in neoplasms: p16 lesions are selected in Barrett’s esophagus. Cancer Res. 64, 3414–3427 (2004).

37. Reid, B. J., Blount, P. L. & Rabinovitch, P. S. Biomarkers in Barrett’s esophagus. Gastrointest. Endosc. Clin. N. Am. 13, 369–397 (2003).

38. Polyak, K. & Garber, J. Targeting the missing links for cancer therapy. Nature Med. 17, 283–284 (2011).

39. Polyak, K. Molecular markers for the diagnosis and management of ductal carcinoma in situ. J. Natl Cancer Inst. Monogr. 2010, 210–213 (2010).

40. Peppercorn, J., Perou, C. M. & Carey, L. A. Molecular subtypes in breast cancer evaluation and management: divide and conquer. Cancer Invest. 26, 1–10 (2008).

41. Nielsen, T. O. et al. A comparison of PAM50 intrinsic subtyping with immunohistochemistry and clinical prognostic factors in tamoxifen-treated estrogen receptor-positive breast cancer. Clin. Cancer Res. 16, 5222–5232 (2010).

42. Lenz, G. et al. Stromal gene signatures in large-B-cell lymphomas. N. Engl. J. Med. 359, 2313–2323 (2008).

43. Jones, S. et al. Core signaling pathways in human pancreatic cancers revealed by global genomic analyses. Science 321, 1801–1806 (2008).

44. Bozic, I. et al. Accumulation of driver and passenger mutations during tumor progression. Proc. Natl Acad. Sci. USA 107, 18545–18550 (2010).

45. Parsons, D. W. et al. An integrated genomic analysis of human glioblastoma multiforme. Science 321, 1807–1812 (2008).

46. Salk, J. J., Fox, E. J. & Loeb, L. A. Mutational heterogeneity in human cancers: origin and consequences. Annu. Rev. Pathol. 5, 51–75 (2010).

47. Kelloff, G. J. et al. Perspectives on surrogate end points in the development of drugs that reduce the risk of cancer. Cancer Epidemiol. Biomarkers Prev. 9, 127–137 (2000).

48. Boone, C. W., Kelloff, G. J. & Freedman, L. S. Intraepithelial and postinvasive neoplasia as a stochastic continuum of clonal evolution, and its relationship to mechanisms of chemopreventive drug action. J. Cell Biochem. 53 (Suppl. 17G), 14–25 (1993).

49. Boone, C. W., Kelloff, G. J. & Steele, V. E. The natural history of intraepithelial neoplasia: relevance to the search for intermediate endpoint biomarkers. J. Cell Biochem. 50 (Suppl. 16G), 23–26 (1992).

50. Boone, C. W., Kelloff, G. J. & Steele, V. E. Natural history of intraepithelial neoplasia in humans with implications for cancer chemoprevention strategy. Cancer Res. 52, 1651–1659 (1992).

51. Schwarz, R. A. et al. Noninvasive evaluation of oral lesions using depth-sensitive optical spectroscopy. Cancer 115, 1669–1679 (2009).

52. Thekkek, N. & Richards-Kortum, R. Optical imaging for cervical cancer detection: solutions for a continuing global problem. Nature Rev. Cancer 8, 725–731 (2008).

53. Kelloff, G. J. et al. Progress and promise of FDG-PET imaging for cancer patient management and oncologic drug development. Clin. Cancer Res. 11, 2785–2808 (2005).

54. Vander Heiden, M. G., Cantley, L. C. & Thompson, C. B. Understanding the Warburg effect: the metabolic requirements of cell proliferation. Science 324, 1029–1033 (2009).

55. Weber, W. A. et al. Positron emission tomography in non-small-cell lung cancer: prediction of response to chemotherapy by quantitative assessment of glucose use. J. Clin. Oncol. 21, 2651–2657 (2003).

P E R S P E C T I V E S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 213

© 2012 Macmillan Publishers Limited. All rights reserved

56. Aukema, T. S. et al. Is 18F-FDG PET/CT useful for the early prediction of histopathologic response to neoadjuvant erlotinib in patients with non-small cell lung cancer? J. Nucl. Med. 51, 1344–1348 (2010).

57. Zander, T. et al. Early prediction of nonprogression in advanced non-small-cell lung cancer treated with erlotinib by using [18F]fluorodeoxyglucose and [18F]fluorothymidine positron emission tomography. J. Clin. Oncol. 29, 1701–1708 (2011).

58. Mileshkin, L. et al. Changes in 18F-fluorodeoxyglucose and 18F-fluorodeoxythymidine positron emission tomography imaging in patients with non-small cell lung cancer treated with erlotinib. Clin. Cancer Res. 17, 3304–3315 (2011).

59. Galanina, N., Bossuyt, V. & Harris, L. N. Molecular predictors of response to therapy for breast cancer. Cancer J. 17, 96–103 (2011).

60. Tang, G. et al. Comparison of the prognostic and predictive utilities of the 21-gene Recurrence Score assay and Adjuvant! for women with node-negative, ER-positive breast cancer: results from NSABP B-14 and NSABP B-20. Breast Cancer Res. Treat. 127, 133–142 (2011).

61. Glas, A. M. et al. Converting a breast cancer microarray signature into a high-throughput diagnostic test. BMC Genomics 7, 278 (2006).

62. Alizadeh, A. A. et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503–511 (2000).

63. Barker, A. D. et al. I-SPY 2: an adaptive breast cancer trial design in the setting of neoadjuvant chemotherapy. Clin. Pharmacol. Ther. 86, 97–100 (2009).

64. Slamon, D. J. et al. Use of chemotherapy plus a monoclonal antibody against HER2 for metastatic breast cancer that overexpresses HER2. N. Engl. J. Med. 344, 783–792 (2001).

65. Pao, W. et al. EGF receptor gene mutations are common in lung cancers from “never smokers” and are associated with sensitivity of tumors to gefitinib and erlotinib. Proc. Natl Acad. Sci. USA 101, 13306–13311 (2004).

66. Audeh, M. W. et al. Oral poly(ADP-ribose) polymerase inhibitor olaparib in patients with BRCA1 or BRCA2 mutations and recurrent ovarian cancer: a proof-of-concept trial. Lancet 376, 245–251 (2010).

67. Tutt, A. et al. Oral poly(ADP-ribose) polymerase inhibitor olaparib in patients with BRCA1 or BRCA2 mutations and advanced breast cancer: a proof-of-concept trial. Lancet 376, 235–244 (2010).

68. Molinari, F. et al. Increased detection sensitivity for KRAS mutations enhances the prediction of anti-EGFR monoclonal antibody resistance in metastatic colorectal cancer. Clin. Cancer Res. 17, 4901–4914 (2011).

69. Hulka, B. S. & Wilcosky, T. Biological markers in epidemiologic research. Arch. Environ. Health 43, 83–89 (1988).

70. Frank, R. & Hargreaves, R. Clinical biomarkers in drug discovery and development. Nature Rev. Drug Discov. 2, 566–580 (2003).

71. Fearon, E. R. Human cancer syndromes: clues to the origin and nature of cancer. Science 278, 1043–1050 (1997).

72. Lesko, L. J. & Woodcock, J. Pharmacogenomic-guided drug development: regulatory perspective. Pharmacogenomics J. 2, 20–24 (2002).

73. Kelloff, G. J. & Sigman, C. C. New science-based endpoints to accelerate oncology drug development. Eur. J. Cancer 41, 491–501 (2005).

74. Gail, M. H. & Mai, P. L. Comparing breast cancer risk assessment models. J. Natl Cancer Inst. 102, 665–668 (2010).

75. Lee, S. Y. & McLeod, H. L. Pharmacogenetic tests in cancer chemotherapy: what physicians should know for clinical application. J. Pathol. 223, 15–27 (2011).

76. Tan, D. S. et al. Biomarker-driven early clinical trials in oncology: a paradigm shift in drug development. Cancer J. 15, 406–420 (2009).

77. Eisenhauer, E. A. et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur. J. Cancer 45, 228–247 (2009).

78. Danila, D. C., Fleisher, M. & Scher, H. I. Circulating tumor cells as biomarkers in prostate cancer. Clin. Cancer Res. 17, 3903–3912 (2011).

79. Swaby, R. F. & Cristofanilli, M. Circulating tumor cells in breast cancer: a tool whose time has come of age. BMC Med. 9, 43 (2011).

80. Wagner, J. A., Williams, S. A. & Webster, C. J. Biomarkers and surrogate end points for fit-for-purpose development and regulatory evaluation of new drugs. Clin. Pharmacol. Ther. 81, 104–107 (2007).

81. Johnson, J. R. et al. Accelerated approval of oncology products: the food and drug administration experience. J. Natl Cancer Inst. 103, 636–644 (2011).

82. Danila, D. C., Pantel, K., Fleisher, M. & Scher, H. I. Circulating tumors cells as biomarkers: progress toward biomarker qualification. Cancer J. 17, 438–450 (2011).

83. Tran, L. N. et al. Comparison of treatment response classifications between unidimensional, bidimensional, and volumetric measurements of metastatic lung lesions on chest computed tomography. Acad. Radiol. 11, 1355–1360 (2004).

84. Zhao, B., Schwartz, L. H. & Larson, S. M. Imaging surrogates of tumor response to therapy: anatomic and functional biomarkers. J. Nucl. Med. 50, 239–249 (2009).

85. Beckman, R. A., Clark, J. & Chen, C. Integrating predictive biomarkers and classifiers into oncology clinical development programmes. Nature Rev. Drug Discov. 10, 735–748 (2011).

86. Maitournam, A. & Simon, R. On the efficiency of targeted clinical trials. Stat. Med. 24, 329–339 (2005).

87. Freidlin, B., McShane, L. M. & Korn, E. L. Randomized clinical trials with biomarkers: design issues. J. Natl Cancer Inst. 102, 152–160 (2010).

88. Freidlin, B., Jiang, W. & Simon, R. The cross-validated adaptive signature design. Clin. Cancer Res. 16, 691–698 (2010).

89. Berry, D. A. Bayesian clinical trials. Nature Rev. Drug Discov. 5, 27–36 (2006).

90. Simon, R. Validation of pharmacogenomic biomarker classifiers for treatment selection. Cancer Biomark. 2, 89–96 (2006).

91. Simon, R. Development and evaluation of therapeutically relevant predictive classifiers using gene expression profiling. J. Natl Cancer Inst. 98, 1169–1171 (2006).

92. Simon, R. Development and validation of biomarker classifiers for treatment selection. J. Stat. Plan. Inference 138, 308–320 (2008).

93. Hoering, A., Leblanc, M. & Crowley, J. J. Randomized Phase III clinical trial designs for targeted agents. Clin. Cancer Res. 14, 4358–4367 (2008).

94. Orloff, J. et al. The future of drug development: advancing clinical trial design. Nature Rev. Drug Discov. 8, 949–957 (2009).

95. Fukuoka, M. et al. Biomarker analyses and final overall survival results from a Phase III, randomized, open-label, first-line study of gefitinib versus carboplatin/paclitaxel in clinically selected patients with advanced non-small-cell lung cancer in Asia (IPASS). J. Clin. Oncol. 29, 2866–2874 (2011).

96. Mok, T. S. et al. Gefitinib or carboplatin-paclitaxel in pulmonary adenocarcinoma. N. Engl. J. Med. 361, 947–957 (2009).

97. Temple, R. J. Enrichment designs: efficiency in development of cancer treatments. J. Clin. Oncol. 23, 4838–4839 (2005).

98. Clark, G. M. in Diseases of the Breast (eds Harris, J. R., Lippman, M. E., Morrow, M. & Osborne, C. K.) 489–514 (Lippincott Williams & Wilkins, Philadelphia, 2000).

99. Rubin, E. H., Anderson, K. M. & Gause, C. K. The BATTLE Trial: a bold step toward improving the efficiency of biomarker-based drug development. Cancer Discov. 1, 17–20 (2011).

100. Gold, K. A. et al. The BATTLE to personalize lung cancer prevention through reverse migration. Cancer Prev. Res. (Phila.) 4, 962–972 (2011).

101. Kim, E. S. et al. The BATTLE Trial: personalizing therapy for lung cancer. Cancer Discov. 1, 45–53 (2011).

102. Kris, M. G. et al. Identification of driver mutations in tumor speciments from 1,000 patients with lung adenocarcinoma: the NCI’s Lung Cancer Mutation Consortium. J. Clin. Oncol. 29, abstract CRA7506 (2011).

103. Berry, D. A. Adaptive clinical trials in oncology. Nature Rev. Clin. Oncol. 8 Nov 2011 (doi:10.1038/nrclinonc.2011.165).

104. Freidlin, B. & Simon, R. Adaptive signature design: an adaptive clinical trial design for generating and prospectively testing a gene expression signature for sensitive patients. Clin. Cancer Res. 11, 7872–7878 (2005).

105. Freidlin, B. & Korn, E. L. Biomarker-adaptive clinical trial designs. Pharmacogenomics 11, 1679–1682 (2010).

106. Horning, S. J. et al. Interim positron emission tomography scans in diffuse large B-cell lymphoma: an independent expert nuclear medicine evaluation of the Eastern Cooperative Oncology Group E3404 study. Blood 115, 775–777 (2010).

107. Seymour, L. et al. The design of Phase II clinical trials testing cancer therapeutics: consensus recommendations from the clinical trial design task force of the National Cancer Institute Investigational Drug Steering Committee. Clin. Cancer Res. 16, 1764–1769 (2010).

108. Ratain, M. J. & Sargent, D. J. Optimising the design of Phase II oncology trials: the importance of randomisation. Eur. J. Cancer 45, 275–280 (2009).

109. Cancer Genome Atlas Research Network. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455, 1061–1068 (2008).

110. Cancer Genome Atlas Research Network. Integrated genomic analyses of ovarian carcinoma. Nature 474, 609–615 (2011).

111. Chin, L., Andersen, J. N. & Futreal, P. A. Cancer genomics: from discovery science to personalized medicine. Nature Med. 17, 297–303 (2011).

112. Boumber, Y. & Issa, J. P. Epigenetics in cancer: what’s the future? Oncology 25, 220–226, 228 (2011).

113. Schulman, K. A. & Tunis, S. R. A policy approach to the development of molecular diagnostic tests. Nature Biotech. 28, 1157–1159 (2010).

114. Lee, J. W. et al. Fit-for-purpose method development and validation for successful biomarker measurement. Pharm. Res. 23, 312–328 (2006).

115. Teutsch, S. M. et al. The Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Initiative: methods of the EGAPP Working Group. Genet. Med. 11, 3–14 (2009).

116. Pepe, M. S., Feng, Z., Janes, H., Bossuyt, P. M. & Potter, J. D. Pivotal evaluation of the accuracy of a biomarker used for classification or prediction: standards for study design. J. Natl Cancer Inst. 100, 1432–1438 (2008).

Competing interests statementThe authors declare no competing financial interests.

FURTHER INFORMATIONClinicalTrials.gov website: http://clinicaltrials.govFDA website — Class II Special Controls Guidance Document: http://www.fda.gov/MedicalDevices/DeviceRegulation andGuidance/GuidanceDocuments/ucm077933.htmFDA website — Critical Path Initiative: http://www.fda.gov/ScienceResearch/SpecialTopics/CriticalPathInitiative/CriticalPathOpportunitiesReports/ucm077262.htmFDA website — Draft Guidance for Industry: Qualification Process for Drug Development Tools: http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatory Information/Guidances/UCM230597.pdfFDA website — Drug–diagnostic Co-Development Concept Paper: http://www.fda.gov/downloads/Drugs/ScienceResearch/ResearchAreas/Pharmacogenetics/UCM116689.pdfFDA website — Guidance for Industry: Adaptive Design Clinical Trials for Drugs and Biologics: http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatory Information/Guidances/UCM201790.pdfFDA website — Guidance on Pharmacogenetic Tests and Genetic Tests for Heritable Markers: http://www.fda.gov/ MedicalDevices/DeviceRegulationandGuidance/GuidanceDocuments/ucm077862.htmFDA website — Interactive Review for Medical Device Submissions: 510(k)s, Original PMAs, PMA Supplements, Original BLAs, and BLA Supplements: http://www.fda.gov/MedicalDevices/DeviceRegulationandGuidance/GuidanceDocuments/ucm089402.htmFDA website — In Vitro Diagnostic Devices: Guidance for the Preparation of 510(k) Submissions: http://www.fda.gov/MedicalDevices/DeviceRegulationandGuidance/GuidanceDocuments/ucm094533.htm

SUPPLEMENTARY INFORMATIONSee online article: S1 (box)

ALL LINKS ARE ACTIVE IN THE ONLINE PDF

P E R S P E C T I V E S

214 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

In 1979–1980, several laboratories identified and char-acterized viral genes encoding products that would suffice to immortalize and transform cultured mam-malian cells1,2. The discovery that these viral oncogenes have closely related homologues in the human genome — proto-oncogenes — and the identification of other cancer-related genes, namely oncosuppressor genes and genome stability genes, consolidated the perhaps sim-plistic notion that cancer constitutes a cell-autonomous genetic disease. Thus, throughout the past century, most — if not all — therapeutic approaches for cancer were aimed at selectively eradicating neoplastic (as opposed to normal) cells3.

In the early years of anticancer therapy, tumour-specific markers were not available; therefore, conventional anti-neoplastic agents (including DNA-damaging chemicals, anthracyclines, antimetabolites, mitotic spindle poisons and other agents) were selected for their ability to kill cells with elevated proliferation rates — one of the most common properties of tumour cells4. However, as several cell types continuously proliferate in physiological condi-tions, the use of conventional chemotherapeutics is often associated with severe side effects, including myelo-paenia, mucositis (linked to gastrointestinal toxicity) and alopecia.

In 1960, Nowell and Hungerford5 first described the Philadelphia chromosome — t(9;22)(q34;q11) — an abnormality specifically associated with chronic myelo-genous leukaemia (CML)6. The subsequent discovery that the fusion protein breakpoint cluster region (BCR)–ABL, which is encoded by the Philadelphia chromosome,

is the sole aetiological determinant of CML7,8 paved the way for the field of ‘targeted’ anticancer therapy. This culminated in 2001 with the approval of the BCR–ABL inhibitor imatinib mesylate (Gleevec/Glivec; Novartis) for the treatment of CML9. Imatinib has revolutionized the treatment of CML, improving the 5-year survival rates of patients with CML from approximately 30% to 90–95%9. Of note, imatinib also efficiently inhibits tyrosine kinases other than BCR–ABL, including KIT and platelet-derived growth factor receptor (PDGFR); a pharmacological pro-file that led to the approval of imatinib by the US Food and Drug Administration (FDA) for the treatment of tumours with activating KIT mutations and rearrangements in the gene encoding the β-subunit of PDGFR10,11.

Since the advent of imatinib, dozens of other targeted chemotherapeutics have been developed and entered the clinical routine with promising results, including erlotinib (Tarceva; Roche/Genentech)12 and lapatinib (Tykerb; GlaxoSmithKline)13 — two tyrosine kinase inhibitors that block epidermal growth factor receptor (EGFR)- and HER2 (also known as ERBB2)-mediated signalling, respectively. Thus, with the exception of the ever-growing class of therapeutic monoclonal antibodies, most of which elicit an anticancer effect via the immune system13, virtually all conventional and targeted antineo-plastic agents have been developed based on their ability to directly interact with — and hence inhibit the growth of or kill — cancer cells.

Nonetheless, during the past two decades, it has become clear that the interaction between tumours and their microenvironment (including stromal, endothelial

1INSERM U848, Institut Gustave Roussy, Pavillon de Recherche 1, 39 rue Camille Desmoulins, F‑94805 Villejuif, France.2Institut Gustave Roussy, F‑94805 Villejuif, France. 3Université Paris-Sud, Paris 11, F‑94276 Le Kremlin‑Bicêtre, France. 4Center of Clinical Investigations in Biotherapies of Cancer (CICBT 507), Institut Gustave Roussy, F‑94805 Villejuif, France.5Metabolomics Platform, Institut Gustave Roussy, F‑94805 Villejuif, France. 6Centre de Recherche des Cordeliers, F‑75006 Paris, France.7Pôle de Biologie, Hôpital Européen Georges Pompidou, AP‑HP, F‑75008 Paris, France.8Université Paris Descartes, Sorbonne Paris Cité, F‑75006 Paris, France.Correspondence to: G.K.  e‑mail: kroemer@orange.frdoi:10.1038/nrd3626 Published online 3 February 2012

The secret ally: immunostimulation by anticancer drugsLorenzo Galluzzi1,2,3, Laura Senovilla1,2,3, Laurence Zitvogel1,2,3,4 and Guido Kroemer1,5,6,7,8

Abstract | It has recently become clear that the tumour microenvironment, and in particular the immune system, has a crucial role in modulating tumour progression and response to therapy. Indicators of an ongoing immune response, such as the composition of the intratumoural immune infiltrate, as well as polymorphisms in genes encoding immune modulators, have been correlated with therapeutic outcome. Moreover, several anticancer agents — including classical chemotherapeutics and targeted compounds — stimulate tumour-specific immune responses either by inducing the immunogenic death of tumour cells or by engaging immune effector mechanisms. Here, we discuss the molecular and cellular circuitries whereby cytotoxic agents can activate the immune system against cancer, and their therapeutic implications.

Proto-oncogenesGenes that contribute to tumorigenesis via a gain-of-function mutation or overexpression.

REVIEWS

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 215

© 2012 Macmillan Publishers Limited. All rights reserved

Oncosuppressor genesGenes that normally interfere with tumorigenesis. Deletion or loss-of-function mutations in these genes are permissive for the development of cancer.

Genome stability genesGenes that control cell cycle progression or DNA repair mechanisms, thereby supervising the maintenance of genome stability.

Therapeutic monoclonal antibodiesA novel class of highly specific therapeutic agents that are used in various medical fields and often function by activating immune effector mechanisms.

Personalized medicineA medical approach whereby therapies are tailored to each patient’s pathological and genetic features to minimize side effects and improve efficacy.

Merkel cell carcinomaA considerably rare human skin cancer of viral origin. Notably, the incidence of this cancer is particularly high, and the course of disease particularly aggressive, in immunocompromised individuals.

and immune cells) is crucial not only during onco-genesis and tumour progression but also in the context of anticancer therapies14. On one hand, this has led to the development of chemotherapeutics that specifically target trophic tumour–stroma interactions — for example, bevaci zumab (Avastin; Roche/Genentech) is a mono-clonal antibody that neutralizes vascular endothelial growth factor (VEGF), thereby inhibiting neoangio-genesis15. On the other hand, it has recently been discov-ered that, beyond their cytotoxic properties, numerous anticancer agents have the capacity to stimulate the innate and acquired immune system, and in some instances even to evoke long-term protective memory T cell responses, thus facilitating tumour eradication16.

Here, we discuss the immune parameters that may predict the responses of cancer patients to therapy, and we analyse the molecular and cellular mechanisms by which conventional chemotherapeutics as well as targeted agents can activate the immune system against cancer.

Predictive impact of immune parametersDuring the past two decades, oncologists have become increasingly interested in personalized medicine, and this has led to the identification and clinical exploitation of several molecular biomarkers. Thus, prognostic and/or predictive value has been demonstrated for a range of biomarkers, including gene signatures, protein expres-sion levels and/or the amounts of specific circulating factors17. Accumulating evidence now suggests that indicators of immune system activity in patients with cancer can also be of prognostic value and/or be used to predict the therapeutic response to specific treatments18 (TABLE 1).

These biomarkers can be grouped into two main categories: local indicators, which mainly reflect the physical engagement of immune effector cells against cancer (for example, tumour infiltration by cytotoxic T lymphocytes (CTLs)); and systemic indicators, which can represent the generation of an antitumour humoral or cellular response (for example, levels of tumour antigen-specific circulating antibodies or the frequency of antigen-specific T cell precursors). Systemic indicators can also reveal the propensity of the host immune sys-tem to favourably respond to malignancies (for example, single nucleotide polymorphisms (SNPs) in genes coding for modulators of innate or adaptive immunity). Although local biomarkers can only be assessed at tumour sites (that is, the primary location of the tumour, invaded lymph nodes or distant metastases), their systemic counterparts can be measured using laboratory tests on peripheral blood samples.

Local immune biomarkers. The local expression of com-ponents and regulators of the extracellular matrix, the anti-angiogenic factor thrombospondin and genes of the monocytic lineage (for example, the genes encod-ing CCAAT/enhancer binding protein-α and colony stimulating factor 2 receptor-α, which are indicative of phagocytic infiltration) is prognostically favourable in patients with large B cell lymphoma, irrespective of the presence of rituximab (Rituxan; Roche/Genentech)

— a monoclonal antibody that targets CD20 — in the chemotherapeutic cocktail19. Along similar lines, genes related to extracellular matrix remodelling, particularly the gene encoding osteonectin (also known as SPARC), were found to be overexpressed in metastatic lesions from patients with melanoma who responded favourably to dacarbazine (an alkylating agent used for the treat-ment of multiple neoplasms), and this correlated with improved long-term survival20.

Hierarchical clustering of gene expression data from 58 patients with breast cancer bearing the HER2 ampli-fication has recently enabled the generation of a prog-nostic predictor whose performance is not influenced by HER2 status. Such a HER2-derived prognostic predic-tor includes molecules involved in tumour invasion and metastasis as well as components of the immune system such as the C-type lectin CD69 and the T cell surface glycoprotein CD3 δ- and ζ-chains (CD3D and CD247, respectively)21. Spectral clustering of micro array data also revealed that the elevated expression of genes involved in both the innate and adaptive immune responses is associated with increased overall and progression-free survival in patients with head and neck squamous cell carcinoma22.

Along similar lines, an immunity-related gene sig-nature was found to predict the response of patients with oestrogen receptor-negative breast cancer who were treated with anthracyclines or taxanes23,24. However, statistical significance — with a possible prognostic influence of molecular subtypes — was not observed in all clinical cohorts included in the study. Patients with Merkel cell carcinoma who have tumours that display a CD8+ cell-associated gene signature (including genes encoding components of cytotoxic granules, cytokines, chemokines and the CD8 α-chain) were found to pro-gress relatively slowly; this led to the discovery that a robust intratumoural (but not peritumoural) infiltration of cytotoxic CD8+ lymphocytes constitutes an independ-ent predictor of survival in Merkel cell carcinoma25.

Similarly, a high number of tumour-infiltrating lym-phocytes (including T cells and B cells) and macrophages has been associated with low tumour stage and favour-able prognosis in patients with small-cell lung carcinoma who are treated with first-line surgery26. Such a prognos-tic value for macrophagic tumour infiltration has not been confirmed by subsequent studies, demonstrating that elevated levels of intratumoural CD68+ and CD163+ macrophages negatively affect the course of disease (by favouring angiogenesis at primary tumour sites, stimu-lating tumour cell invasiveness and de facto suppressing antitumour immune responses)27 in malignancies as diverse as Hodgkin’s lymphoma28, renal cell carcinoma (RCC)29, leiomyosarcoma30 as well as hepatocellular31, lung32 and prostate cancer33.

Intriguingly, the presence of CD14+ macrophages (which are known to contribute to T cell activation) at the tumour-invasive margin has recently been associated with limited metastatic potential and favourable prognosis in patients with colorectal carcinoma34. It remains unclear, however, whether these observations should be ascribed to the functional profile of CD14+ macrophages, their

R E V I E W S

216 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Table 1 | Examples of immune parameters affecting chemotherapeutic responses

Parameters Cancers Notes Refs

Local indicators

‘Germinal-centre B cell’ and stromal gene signatures

Large B cell lymphoma •Include CEBPA, CSF2RA, THBS1 and genes encoding EM regulators

•Favourable response irrespective of the presence of rituximab in the therapeutic regimen

19

CD16+ myeloid cell infiltration

Colorectal cancer •Correlates with improved OS even after adjustment for other prognostic factors

46

CD20+ cell infiltration HNSCC •Associated with improved PFS in low-risk patients•Negatively influences PFS in high-risk patients

49

Cytotoxic cell infiltration

Hodgkin’s lymphoma •Includes TIA1+ and granzyme B+ cells•Negatively influences OS

47

Granzyme B+ cell infiltration

Ovarian cancer •Administered alone or in combination with low levels of T

reg cells

•Correlates with improved PFS and OS in patients receiving neoadjuvant therapy

48

HER2-derived prognostic predictor

Patients with HER2-overexpressing breast cancer

•Includes CD69, CD3D, CD247 and genes encoding regulators of metastasis

•Prognostic performance is not influenced by HER2 status

21

High CD8+/Treg

cell ratio Breast cancer •Correlates with histological response, PFS and OS in patients receiving neoadjuvant therapy

•Outperforms classical predictive factors in multivariate analyses

35,43

High T reg cell levels

and/or low CD8+/T

reg cell ratio

Gastric cancer, hepatocellular cancer, lung cancer, metastatic renal cancer

•Negatively affect prognosis 39–41, 123

Immune-related gene expression module

Breast cancer •Includes CD3G, STAT1 and several genes encoding cytokine receptors

•Prognostic value in untreated ER-negative and HER2-negative patients as well as HER2-positive patients

•Associated with pCR in anthracycline- and taxane-treated patients

23,24

Intratumoral CD8+ cell infiltration

Merkel cell carcinoma •Independent predictor of survival 25

Macrophages at the invasive margin

Colorectal cancer •Associated with reduced metastatic potential and improved prognosis

34

Macrophagic infiltration

Hepatocellular cancer, Hodgkin’s lymphoma, leiomyosarcoma, lung adenocarcinoma, prostate cancer, renal cell carcinoma

•Associated with increased angiogenesis, elevated metastatic potential and worsened OS

27–33

NK cell infiltration Hodgkin’s lymphoma •Associated with improved OS 47

Spectral clustering of V2 and V3 vectors

HNSCC •Include genes encoding regulators of the EM–cell interaction as well as innate and adaptive immunity

•Associated with increased OS and PFS

22

Stromal and immune response-related gene signatures

Metastatic melanoma •Include SPARC•Upregulated in lesions that responded to dacarbazine

and linked with improved OS

20

TH2 response Pancreatic cancer •Features GATA3 expression by immune infiltrates

•Correlates with reduced OS37

TIL density Colorectal carcinoma, liver metastasis

•Includes CD3+, CD8+ and granzyme B+ cells•Prognostic marker for pCR and DFS

44

TILs (T and B cells) and macrophages

Small-cell lung carcinoma

•Associated with reduced tumour size and low tumour stage

•High intratumoural CD8+ cells and macrophages are predictive of improved OS in patients treated with first-line surgery

26

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 217

© 2012 Macmillan Publishers Limited. All rights reserved

FOXP3Forkhead box P3. A transcription factor from the FOX protein family that underlies the capacity of regulatory T cells to negatively regulate immune responses and favour self-tolerance (including in cancer).

T helper type 2 (TH2) responseThe production — by CD4+ T helper lymphocytes — of cytokines including interleukin-4 (IL-4), IL-5 and IL-13 (which promote atopy), as well as IL-10 (which exerts immunosuppressive functions).

selective localization at the tumour invasive front or the peculiar pathophysiological features of colorectal carci-noma, as discussed below.

In patients with breast cancer who are treated with neoadjuvant chemotherapy, a high ratio between CD8+ and FOXP3+ lymphocytes (immunosuppressive regula-tory T (Treg) cells) correlates with improved histologi-cal response, as well as relapse-free and overall survival, and actually outperforms classical predictive factors in multivariate analyses35,36. Similarly, the polarization of the immune infiltrate towards GATA-binding factor 3

expression — reflecting a T helper type 2 (TH2) response that is often associated with immunosuppression — appears to negatively affect survival in pancreatic cancer37. Such a TH2 response has been ascribed to the conditioning of dendritic cells (DCs) via thymic stromal lymphopoietin secreted by cancer-associated fibroblasts37,38, which sug-gests that the tumour stroma has a crucial role in the shaping of anticancer immune responses.

High levels of tumour-infiltrating FOXP3+ Treg cells negatively affect prognosis in several other neoplasms, including RCC as well as hepatocellular39, gastric40 and

Table 1 (cont.) | Examples of immune parameters affecting chemotherapeutic responses

Parameters Cancers Notes Refs

Systemic indicators

ALK-specific antibodies

Anaplastic large cell lymphoma

•High antibody titres inversely correlate with tumour stage and the amount of circulating tumour cells

51

AZGP1-specific antibodies

Lung adenocarcinoma •High antibody titres correlate with improved OS in patients with stage I–II disease

52

CEA-specific antibodies

Colorectal cancer •High antibody titres constitute an independent favourable predictor of PFS

50

MCV-specific antibodies

Merkel cell carcinoma •High antibody titres correlate with improved PFS 54

Mucin 1-specific antibodies

Pancreatic cancer •High antibody titres are associated with improved OS, even after multivariate analysis for tumour stage, patient age and gender

53

Fcγ receptor genotypes

Multiple cancer types •Various SNPs influence the response to monoclonal antibody-based therapies

62,64, 65

IFNγ secretion by NK cells

Gastrointestinal stromal tumour

•Correlates with improved OS in imatinib-treated patients

66

IFNγ secretion by PBMCs challenged with PSA peptides

Metastatic castration-resistant prostate cancer

•Correlates with a trend towards improved OS in patients treated with a PSA-based vaccine

67

IL10 promoter genotype

Lymphoid neoplasms •The –3575A/A genotype is associated with improved OS

59

IL-16, IL-19, LILRA4, KLRC4 and CD5 genotypes

Chronic lymphocytic leukaemia

•SNPs in these genes all have prognostic value independently of IgV

H mutations

58

IL-2, IL-8, IL-12B and IL-1RN genotypes

Follicular lymphoma •A score of the number of deleterious genotypes in these genes is strongly associated with OS

56

IL4 promoter genotype

Renal cell carcinoma •The heterozygous–589T, –33T/–589C, –33C genotype is a prognostic risk factor for RCC

57

IL4R genotype Diffuse large B cell lymphoma

•The I75V mutation is associated with worse OS and PFS

61

IL6 promoter genotype

ER-positive, node-positive breast cancer

•At least one copy of the haplotype –597G, –572G, –373(10A/11T), –174G is associated with worsened OS

60

NCR3 genotype Gastrointestinal stromal tumour

•The TC and CC genotypes at position 3,790 correlate with reduced OS in imatinib-treated patients

55

ALK, anaplastic lymphoma kinase; AZGP1, zinc-binding α2-glycoprotein 1; CD3G, T cell surface glycoprotein CD3 g-chain; CD5, T cell surface glycoprotein CD5; CD69, early activation antigen CD69; CEA, carcinoembryonic antigen; CEBPA, CCAAT/enhancer binding protein α; CSF2RA, colony stimulating factor 2 receptor α; DFS, disease-free survival; EM, extracellular matrix; ER, oestrogen receptor; GATA3, GATA binding protein 3; HNSCC, head and neck squamous cell carcinoma; IFNγ, interferon-γ; IgV

H, immunoglobulin heavy-chain variable region; IL-1RN, IL-1 receptor antagonist; IL-2, interleukin-2; KLRC4, NK cell lectin-like

receptor subfamily C member 4; LILRA4, leukocyte immunoglobulin-like receptor subfamily A member 4; MCV, Merkel cell polyoma virus; NCR3, natural cytotoxicity triggering receptor 3; NK, natural killer; OS, overall survival; PBMC, peripheral blood mononuclear cell; pCR, pathological complete response; PFS, progression-free survival; PSA, prostate-specific antigen; SNP, single nucleotide polymorphism; SPARC, secreted protein acidic and rich in cysteine; STAT1, signal transducer and activator of transcription 1; T

H2, T helper type 2; THBS1, thrombospondin 1; TIA1, T cell-restricted intracellular antigen 1; TIL,

tumour-infiltrating lymphocyte; Treg

, regulatory T cell (forkhead box P3 (FOXP3)+ immunoregulatory lymphocyte).

R E V I E W S

218 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Activating NK receptor p30(NKp30). Natural killer cell activating receptor encoded by the natural cytotoxicity triggering receptor 3 (NCR3) gene. The main endogenous ligand of NKp30 is the B7 homolog B7H6, which is selectively expressed on several distinct cancer cell types.

lung cancer41 but not colorectal carcinoma42,43. This apparent paradox has been explained with the hypoth-esis that Treg cells in the intestine might suppress the pro-tumorigenic inflammation driven by the local microbial flora43. In line with this notion, the levels of FOXP3+ lymphocytes at the invasive margin of colo-rectal cancer liver metastases reportedly do not cor-relate with the therapeutic outcome, whereas a high density of tumour-infiltrating lymphocytes (including CD3+, CD8+ and granzyme B+ cells) constitutes a prog-nostic biomarker for response to chemotherapy and progression-free survival44.

Intriguingly, the overlap between the antigenic reper-toire of effector memory T cells and that of Treg cells in patients with colorectal cancer appears to be very limited, and only T cell responses against antigens that are recog-nized by Treg cells appear to increase following Treg cell depletion45. These findings either suggest that only a few antigenic determinants are crucial for T cell-mediated immunity (at least in the colorectal carcinoma setting) or that Treg cell-independent mechanisms regulate effector T cells whose specificity is not represented in the Treg cell antigenic repertoire.

In patients with colorectal cancer, high levels of CD16+ myeloid cell infiltrates have been associated with improved survival, even after adjusting for known prog-nostic factors including extension of the primary tumour (pT), involvement of regional lymph nodes (pN), tumour growth and grade, vascular invasion, tumour growth and patient age46. Along similar lines, immuno-histochemical studies have revealed that infiltration by natural killer (NK) cells constitutes a positive prog-nostic factor in patients with Hodgkin’s lymphoma47, whereas high levels of cytotoxic T cell-restricted intra-cellular antigen 1 (TIA1)+ cells and granzyme B+ cells were shown to negatively influence overall survival47. This latter observation has not been confirmed in more recent studies, which have shown that high levels of granzyme B+ cells (alone or combined with low amounts of Treg cells) correlate with improved progression-free and overall survival in patients with ovarian carcinoma who are subjected to neoadjuvant chemotherapy48. The reasons for this apparent discrepancy remain to be elu-cidated but may be linked to tumour-intrinsic features, tumour grade or therapeutic approach. In support of this hypothesis, it has been shown that intense CD20+ cell infiltration is associated with improved disease-free sur-vival in patients with low-risk squamous cell carcinoma of the oro- and hypopharynx, whereas the converse is true in high-risk patients49.

Irrespective of these unresolved issues, it appears that tumour infiltration by high levels of immune effector cells (including CD8+ lymphocytes, CD16+ myeloid cells and NK cells) and low amounts of immunosuppressive cells is frequently associated with improved patient survival.

Systemic immune biomarkers. High levels of circulating antibodies against tumour-specific antigens often corre-late with improved survival. This has been demonstrated for antibodies targeting the carcinoembryonic antigen in patients with colorectal cancer50, anaplastic lymphoma

kinase in patients with anaplastic large cell lymphoma51, zinc-binding α2-glycoprotein 1 in patients with lung adenocarcinoma52, and mucin 1 in patients with pan-creatic cancer53. On a slightly different note, antibodies targeting the Merkel cell polyoma virus, the aetiological determinant of Merkel cell carcinoma, have identified a subset of patients with improved progression-free sur-vival54. It remains unclear whether this reflects a role for the Merkel cell polyoma virus during tumour progres-sion or simply the continuous expression of viral anti-gens by tumour cells.

SNPs that affect the expression levels or pattern of modulators of innate and cognate immunity also consti-tute prognostic and/or predictive biomarkers. An SNP at position 3,790 in the 3′ untranslated region of the gene encoding activating NK receptor p30 (NKp30; also known as NCR3) genetically predisposes individuals to predomi-nantly express the immunosuppressive NKp30c isoform over the immunostimulatory NKp30a and NKp30b iso-forms. This correlates with impaired NKp30-dependent cytokine release, defective DC–NK crosstalk and reduced survival of patients with gastrointestinal stro-mal tumours (GISTs) who are treated with imatinib, irrespective of KIT mutational status55. SNPs in genes encoding interleukin-2 (IL-2), IL-8, IL-12B and the IL-1 receptor antagonist have been shown to predict survival in a cohort of 278 patients with follicular lymphoma56, whereas the heterozygous IL4 genotype -589T/C-33T/C was found to constitute an independent prognostic risk factor in patients with RCC57.

Along similar lines, SNPs influencing the production of IL-16, IL-19, leukocyte immunoglobulin-like receptor subfamily A member 4 (LILRA4; a cell-surface receptor of plasmacytoid DCs), NK cell lectin-like receptor subfam-ily C member 4 (KLRC4; an NK cell receptor) and CD5 (a marker of T and B1 lymphocytes) have prognostic value in patients with chronic lymphocytic leukaemia58, and the IL10 -3575A/A genotype has been linked to improved survival in lymphoid neoplasms59. Host genetic variants in IL6 and IL4R (the gene encoding the IL-4 receptor) predict outcome in patients with oestrogen receptor-positive, node-positive breast cancer and large B cell lym-phoma, respectively60,61. Finally, polymorphisms in the genes encoding Fcγ receptors (FcγRs) on immune cells have been shown to affect the response of several neo-plasms to monoclonal antibody-based regimens62–64, thus constituting prominent predictive biomarkers. The sig-nificance of these polymorphisms is further augmented by the recent discovery that FcγRs on tumour-associated leukocytes can provide a signalling scaffold that promotes cancer cell death in response to the activation of death receptors by monoclonal antibodies65.

Thus, a growing body of evidence suggests that indicators of host immune performance can be used as prognostic and/or predictive biomarkers in oncol-ogy. In addition to the levels of tumour antigen-specific circulating autoantibodies and SNPs in genes that medi-ate immune responses, other factors can be assessed to obtain prognostic and/or predictive hints. For instance, the ability of circulating NK cells to produce interferon-γ (IFNγ) in the context of a DC–NK crosstalk reportedly

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 219

© 2012 Macmillan Publishers Limited. All rights reserved

Metronomic chemotherapyThe chronic administration (at regular intervals) of chemotherapy at low, minimally toxic doses, with no prolonged drug-free periods.

Myeloid-derived suppressor cells(MDSCs). A heterogeneous population of cells that are defined by their myeloid origin, immature state and their ability to potently suppress T cell responses.

Signal transducer and activator of transcription(STAT). The mammalian genome encodes at least seven distinct members of the STAT family of transcription factors, which regulate multiple aspects of cell growth, survival and differentiation.

T helper type 1 (TH1) antitumour responseThe production — by CD4+ T helper lymphocytes — of cytokines, mainly interferon-γ and lymphotoxin-α, which exert immunostimulatory functions.

correlates with improved survival in patients with GISTs who receive imatinib66. Similarly, IFNγ secretion by peripheral blood mononuclear cells challenged with a prostate-specific antigen peptide has been associated with a trend towards enhanced survival in patients with metastatic castration-resistant prostate cancer who are treated with a viral prostate-specific antigen vaccine67. Together, these observations suggest that innate and adaptive immunity have an important role in both tumour progression and response to therapy. The mech-anisms by which conventional and targeted anticancer agents activate the immune system are discussed below.

Immunomodulation by conventional agentsMost — if not all — conventional anticancer agents that have been used in the clinic throughout the past cen-tury were selected for their ability to target one of the six classical hallmarks of tumour cells, as formulated by Hanahan and Weinberg4 in 2000 (BOX 1). The majority of these compounds, when used at clinically useful doses, are endowed with intrinsic immunosuppressive proper-ties owing to the fact that they preferentially kill rapidly proliferating cells. However, along with the understand-ing that the tumour stroma has a crucial role in the response of cancer cells to therapy, and with the contin-uously increasing interest in metronomic chemotherapy, several off-target and cancer cell-exogenous immune mechanisms have been discovered that contribute to the efficacy of conventional anticancer agents (FIG. 1).

Gemcitabine, a nucleoside analogue that is used for the treatment of various carcinomas and some forms of non-Hodgkin’s lymphoma, has been shown to increase the expression of class I human leukocyte antigen (HLA) on malignant cells68, enhance the cross-presentation of

tumour antigens to CD8+ T cells69,70 and selectively kill myeloid-derived suppressor cells (MDSCs), both in vitro and in vivo71,72, thus facilitating T cell-dependent anti-cancer immunity68–72. Similar effects on class I HLA expression have also been observed with oxaliplatin (a platinum-based compound that is currently used to treat colorectal and pancreatic cancer) and cyclophosphamide (an alkylating agent that is often used in combination reg-imens to treat some forms of lymphoma, leukaemia and solid tumours)68. Recent data suggest that oxaliplatin and other platinum-based compounds (including cisplatin and carboplatin) not only stimulate class I HLA expres-sion but also inhibit signal transducer and activator of tran-scription 6 (STAT6)-regulated expression of programmed death ligand 2 (PDL2), thus limiting immunosuppression by both DCs and tumour cells73.

Although high-dose cyclophosphamide exerts immuno suppressive effects and is used in this regard to treat autoimmune diseases74, low-dose cyclophos-phamide selectively suppresses inhibitory cell subsets including MDSCs and Treg cells (which facilitate IFNγ-mediated anticancer immunity)75–77, favours the dif-ferentiation of IL-17-producing CD4+ T helper cells (sustaining a T helper type 1 (TH1) antitumour response)78, restores T cell and NK cell effector functions in patients with end-stage cancer76, stimulates the expression of type I IFNs, leads to the preferential expansion of CD8α+ DCs (the main DC subset implicated in cross-presentation)79 and inhibits the generation of immuno-suppressive cytokines (including IL-4, IL-10 and IL-13) while stimulating antitumour innate immunity80.

Of note, the combination of cyclophosphamide and an agonist antibody targeting the co-stimulatory recep-tor OX40 (also known as TNFSF4), which is expressed by Treg cells and activated CD4+ T lymphocytes, can induce cell death in intratumoural (but not peripheral) Treg cells, leading to CD8+ T cell influx and antitumour immunity against otherwise poorly immunogenic mela-noma B16 cells81. This suggests that cyclophosphamide may counter act immunosuppressive leukocytes more efficiently within the tumour than in the periphery but does not explain why circulating MDSC levels were sig-nificantly increased and correlated with clinical stage and metastatic tumour burden in 17 patients with early-stage breast cancer who received cyclophosphamide-containing chemotherapy82. It is possible that patients with breast cancer who are treated with cyclophospha-mide experience only a temporary reduction in numbers of Treg cells, which is followed by a rebound that does not seem to affect the elicitation of stable antitumour immune responses83.

Although these results have not yet been validated in clinical studies, preclinical data indicate that the thera-peutic efficacy of anthracyclines, including doxoru-bicin and daunorubicin, relies — to a large extent — on immune mechanisms. Doxorubicin, which is currently used for the treatment of several cancers, including hae-matological malignancies, many types of carcinoma and soft tissue sarcomas, reportedly enhances the prolifera-tion of tumour antigen-specific CD8+ T cells in tumour-draining lymph nodes and promotes tumour infiltration

Box 1 | The hallmarks of cancer revisited

Inaseminalpaperdatingbackto2000,HanahanandWeinberg4enumeratedsixhallmarksthatcharacterizemost—ifnotall—humancancers:limitlessproliferativepotential;self-sufficiencyingrowthsignals;insensitivitytoantigrowthsignals;evasionofapoptosis;sustainedangiogenesis;andtissueinvasionandmetastasis.Thisformulationwassimpleandrelativelyaccuratewhenitwasproposedin2000,andhasindeeddominatedthecancerresearchsceneoverthepastdecade,constituting(inAugust2011)themostcitedCellpaper.Theadventofthetwenty-firstcenturyhaswitnessedanintenseexperimentaleffortdevotedtotheelucidationofthecellularandmolecularcircuitriesunderlyingoncogenesisandtumourprogression,whichhasledtotheidentificationofanadditionalfourfeaturesofcancer:alteredmetabolism211;escapefromimmunosurveillance212;chromosomaldefectsandgeneticinstability213;andinflammation214.AnintegratedviewofthesehallmarksisprovidedinREF. 215.Inthisreview,HanahanandWeinberg215donotgivemuchcredittoother

alterationsthathaverecentlybeenassociatedwiththemalignantphenotype,suchasdefectsinthemolecularmachineryforautophagy216,217.However,theydoemphasizethefactthattumourscannolongerbeviewedascell-intrinsicgeneticdiseases215.Thispresumablyconstitutesthemostprominentchangeinperspectiveregardingcancerthathasoccurredthroughoutthepastdecade:itisnowclearthatneoplasmsoriginate,progressandrespondtotherapyfromwithinacomplexmicroenvironmentwithwhichtheyconstantlyinteract197,218.Targetingthecrosstalkbetweencancercellsandtheirmicroenvironment,whichincludessolublefactors,componentsoftheextracellularmatrixaswellasstromal,endothelialandimmunecells,constitutesthebasisofseveralanticancerstrategiesthatarecurrentlybeinginvestigated—inparticularimmunochemotherapy.

R E V I E W S

220 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Oxaliplatin

Cisplatin

Doxorubicin

Mitoxantrone

Daunorubicin

Methotrexate

Gemcitabine

Cyclophosphamide

Docetaxel

Paclitaxel

Vincristine

T cell

T cell

γδ

T cellIL-17 T cell

Treg cell

FOXP3+

MDSC

DC

DC

DC

IL-4, IL-10, IL-13T cell

IL-2

T cell

PDL2 ICD

M6PR

MCP1 MCP1

CD8+ CD8+

IFNγ

T cell

CD4+

IL-17

Type I IFNs

IFNγ

P-STAT6

ApoptoticMDSC

P-STAT6 CRT

MHC-I

MHC-I

MHC-I

Tumour

OX40A cell surface glycoprotein of the tumour necrosis factor receptor superfamily that, upon binding to its ligand (OX40L, which is expressed on activated antigen-presenting cells), delivers a co-stimulatory signal that is essential for the long-term survival of CD4+ T cells.

gδ T cellsA small subset of T cells that possess an invariant T cell receptor on their surface and operate at the boundary between innate and adaptive immunity.

first by IL-17-secreting gδ T cells, then by activated IFNγ-secreting CD8+ T cells84,85. In mice lacking either γδ T cell receptor chains or IL-17A, as well as following the abrogation of IL-17-elicited signalling cascades with suitable monoclonal antibodies, doxorubicin failed to induce tumour infiltration by CTLs, and its therapeutic efficacy was highly compromised84,85. This demonstrates that IL-17-producing γδ T cells are required for the full-blown therapeutic potential of doxorubicin.

Along similar lines, daunorubicin, which is now being used to treat acute myeloid leukaemia and neuro-blastoma, reportedly exacerbates antigen expression by cancer cells, thus promoting IL-2 and IFNγ synthesis by local T cells and facilitating the development of a tumour-specific immune response86. Of note, oxaliplatin (but not the chemically related platinum-based compounds cisplatin and carboplatin), cyclophosphamide and anthracyclines such as doxorubicin and mitoxantrone have been shown to trigger immunogenic cell death79,87–89 (BOX 2), constituting yet another mechanism by which these conventional chemotherapeutics can stimulate antitumour immunity.

Taxanes are known to bind with a high affinity to microtubules and interfere with mitotic spindle func-tions, leading to mitosis abortion and often resulting in cell death following mitotic catastrophe90. Nevertheless, accumulating data indicate that at least part of the thera-peutic success of microtubule inhibitors might be derived from cancer cell-extrinsic immune mechanisms. Thus, paclitaxel — the taxane that is currently most widely used in the clinic — has been reported to specifically impair cytokine production and viability in FOXP3+ Treg cells but not in FOXP3–CD4+ effector cells, independently of Toll-like receptor 4 signalling91. In addition, similarly to cispla-tin and doxorubicin, paclitaxel reportedly has the capacity to increase the permeability of tumour cells to granzyme B, thereby rendering them susceptible to CTL-mediated lysis even if they do not express the antigen recognized by CTLs92. This bystander effect perhaps explains why small numbers of CTLs can mediate potent anticancer responses when combined with some types of chemotherapy.

Docetaxel, which is used for the treatment of anthra-cycline-resistant breast and lung cancers, has been shown to decrease splenic MDSC levels in mice with

Figure 1 | Mechanisms through which conventional chemotherapeutics affect the immune system. Conventional chemotherapeutics can stimulate the immune system against cancer by: directly activating CD4+, CD8+ or γδ T cells, leading to the production of interleukin-2 (IL-2), interferon-γ (IFNγ) and IL-17; facilitating the maturation or activation of dendritic cells (DCs); inhibiting or depleting immunosuppressive myeloid-derived suppressor cells (MDSCs) and regulatory T (T

reg) cells; inhibiting signal transducer and activator of transcription 6 (STAT6) phosphorylation and

programmed death ligand 2 (PDL2)-mediated immunosuppression; triggering immunogenic cell death (ICD); upregulating class I major histocompatibility complex (MHC-I) molecules on cancer cells; increasing the permeability of tumour cells to granzyme B via the upregulation of the mannose-6-phosphate receptor (M6PR); inhibiting IL-4, IL-10 and IL-13 production; stimulating the release of monocyte chemoattractant protein 1 (MCP1); or upregulating the expression of type I IFNs. CRT, calreticulin; FOXP3, transcription factor forkhead box P3; P-STAT6, phosphorylated STAT6.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 221

© 2012 Macmillan Publishers Limited. All rights reserved

4T1-NeuMurine breast cancer cells (syngenic to BALB/c mice) that are transduced to express the rat Neu gene (the rat orthologue of human HER2; also known as ERBB2).

SNARESoluble NSF (N-ethylmaleimide- sensitive factor) attachment protein (SNAP) receptor.

4T1-Neu cancers, leading to increased antitumour cyto-toxic responses93. Although the molecular mechanisms underlying these observations have not yet been fully elucidated, docetaxel may deplete MDSCs by inhibiting STAT3-regulated gene transactivation93. Of note, simi-larly to other conventional chemotherapeutics including (but not limited to) paclitaxel and doxorubicin, doce-taxel reportedly stimulates the secretion of monocyte chemoattractant protein 1 (MCP1) by tumour cells94–96, which facilitates the recruitment of macrophages and the establishment of an immunosuppressive stroma97.

Other microtubule inhibitors that affect the immune system during cancer chemotherapy include vincristine (which is currently used for the treatment of non-Hodgkin’s lymphoma) and vinorelbine (which is currently used to treat breast cancer and non-small-cell lung cancer). Vincristine, in combination with doxorubicin and gluco-corticoids, has been shown to increase the abundance of specific DC subsets (that is, CD83+ DCs and type I DCs) in patients with multiple myeloma98 and — similarly to paclitaxel and doxorubicin — it has been shown to stim-ulate DC-mediated antigen presentation99. Conversely, cancer cells succumbing to vinorelbine treatment appear to negatively affect antitumour immunity by promoting the bystander death of immune cells100.

The prototypic folate antagonist methotrexate was originally developed as an anticancer agent owing to its ability to inhibit the generation of tetrahydrofolate, which is an essential precursor for the synthesis of nucleo-tides101. As they preferentially target rapidly dividing

cells, methotrexate and the second-generation folate antagonist pemetrexed (Alimta; Lilly) exert consistent immunosuppressive effects. Thus, methotrexate has become a mainstay in the treatment of several auto-immune disorders, particularly rheumatoid arthritis103. Methotrexate is also used for the treatment of multiple haematological and solid tumours, whereas pemetrexed is now approved by the FDA for the therapy of pleural mesothelioma and non-small-cell lung cancer102. Recent data suggest that low concentrations of methotrexate may boost the maturation of DCs and their ability to stimulate T cells104, further confirming the notion that metronomic chemotherapy may have immunostimula-tory rather than immunosuppressive effects.

Taken together, these observations indicate that, beyond the fact that some chemotherapeutics have immunosuppressive effects, the therapeutic efficacy of several conventional anticancer agents relies — at least in part — on cancer cell-extrinsic molecular and cellular cascades that stimulate antitumour immunity (TABLE 2).

Immunomodulation by targeted agentsAlthough targeted anticancer agents are much more specific than conventional chemotherapeutics, they are not fully devoid of unwanted side effects. For instance, tyrosine kinase inhibitors including the EGFR-specific small-molecule compounds erlotinib and gefitinib (Iressa; AstraZeneca) elicit a broad spectrum of adverse effects on skin and hair105, whereas several monoclonal antibodies mediate cardiac toxicity106. Intriguingly, the therapeutic efficacy of several targeted agents appears to rely — at least partially — on off-target mechanisms, some of which are mediated by the immune system (FIG. 2).

In multiple instances, erlotinib has been shown to exert consistent antitumour effects in patients with acute myeloid leukaemia, a type of cancer in which EGFR is not expressed107,108, in line with the fact that erlotinib and gefitinib affect several cancer cell-intrinsic signal-ling pathways109–111. More recently, EGFR inhibition (by erlotinib or by the monoclonal antibody cetuximab (Erbitux; Bristol-Myers Squibb/Lilly/Merck Serono)) has been associated with increased expression of class I and class II major histocompatibility complex (MHC) molecules, possibly promoting cancer-directed immune and/or inflammatory responses112.

Moreover, an anti-murine EGFR antibody report-edly mediates anticancer effects that are superior to the anticancer effects of AG1478, a chemical inhibitor of EGFR, as a result of the induction of immunogenic cell death and the consequent elicitation of tumour-specific CD4+ and CD8+ T cell responses113. Of note, erlotinib reportedly impairs T cell responses by inhibiting sig-nal transduction by AKT or RAF and/or extracellular signal-regulated kinase (ERK)114. Taken together, these observations suggest that EGFR-targeted antibodies (but not relatively unspecific chemicals such as erlotinib) may efficiently activate the immune system against EGFR-expressing malignancies.

Most patients with CML receiving imatinib treat-ment who are in remission exhibit robust antileukaemic immune responses, which are frequently mediated by

Box 2 | Mechanisms of immunogenic cell death

Thenotionofimmunogeniccelldeath(ICD)reflectsthecapacityofcancercellssuccumbingtospecificlethalstimuli(forexample, anthracyclines,oxaliplatinorγ-radiation)toserveasatherapeuticvaccineandstimulateanantitumourimmuneresponse219,220.RecentworkfromourlaboratorieshasestablishedthemoleculardeterminantsofICD,whichoftenmanifestswithmorphologicalandbiochemicalfeaturesofapoptosis221,222.Thus,ICDappearstorelyonthreecardinalevents203:theexposureoftheendoplasmicreticulumchaperoneproteincalreticulinatthesurfaceofdyingcells,thusenhancingtheuptakeoftumourantigensbydendriticcells89;thereleaseofthenon-histonechromatin-bindingnuclearproteinhigh-mobilitygroupprotein B1(HMGB1),whichstimulatesantigenprocessingandpresentationtoT cells223;andthesecretionofATP,whichleadstotheactivationoftheinflammasomeandtheproductionofpro-inflammatorycytokines224.DuringICD,calreticulinexposurefollowstheactivationofacomplexsignalling

pathwaythatisinitiatedbyendoplasmicreticulumstress,transducedbypro-apoptoticproteins(includingcaspase 8andtheBCL-2familymembersBAXandBCL-2antagonist/killer 1(BAK))andexecutedbythemachineryforSNARE-dependentexocytosis225.ItiscurrentlyunclearwhichprecisemolecularcascadesleadtothereleaseofHMGB1(whichappearstooccurviaacaspase-dependentmechanism)223andtothesecretionofATP226(whichmayinvolvethecellularmachineryforautophagy227orpannexin1channels228).Instancesofcelldeaththatfailoneoftheprocesseslistedabove(thatis, calreticulin

exposure,HMGB1releaseandATPsecretion)arenon-immunogenic.However,ithasrecentlybeenshownthattheimmunogenicityofcelldeathcanberestoredwithappropriatepharmacologicalinterventions.Thus,althoughcisplatinalonedoesnotinducethepre-apoptoticexposureofcalreticulin,itcandosoincombinationwiththeendoplasmicreticulumstressorthapsigargin,de factorenderingcisplatin-inducedcelldeathimmunogenic88.Thishaspromisingimplicationsforthedevelopmentofnovelimmunogenicanticancerregimens.

R E V I E W S

222 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

CD20+CD5+sIgM+ B lymphocytesB lymphocytes expressing at their surface CD5, CD20 and immunoglobulin M (IgM) molecules.

tumour necrosis factor (TNF)-secreting CD4+ T cells115. In addition, a subset of imatinib-treated patients with CML develop CD20+CD5+sIgM+ B lymphocytes that pro-duce anti-carbohydrate antibodies with antitumour activity, and high levels of these lymphocytes have been correlated with cytogenetic or molecular remission116.

In patients with GISTs, the therapeutic outcome of imatinib administration is influenced by various immune parameters, including the level of IFNγ pro-duced by circulating NK cells and the expression pattern of the activating NK receptor NKp30 (REFS 55,66). Moreover, at clinically relevant concentrations, imatinib

Table 2 | Examples of immune system effects of conventional anticancer chemotherapeutics

Agent Indications Notes Refs

Carboplatin Ovarian carcinoma, lung cancer, head and neck cancer

•Inhibits PDL2 expression 73

Cisplatin Sarcoma, lymphoma, germ cell tumour, lung cancer, ovarian cancer

•Inhibits PDL2 expression•Increases the permeability of tumour cells to granzyme B

73,92

Cyclophos - phamide

Lymphoma, leukaemia, solid tumours

•Immunosuppressive at high doses•Increases class I HLA expression on cancer cells•Selectively inhibits T

reg cells and MDSCs, perhaps with a

preferential activity on intratumoural populations•Favours the differentiation of IL-17-producing CD4+ cells•Stimulates the expansion of CD8α+ DCs•Restores T cell and NK cell functions•Inhibits IL-4, IL-10 and IL-13 production•Induces immunogenic cell death

68,74–81

Daunorubicin AML, neuroblastoma •Exacerbates antigen expression by cancer cells, thus promoting IL-2 and IFNγ synthesis by local T cells

86

Docetaxel NSCLC, breast cancer, ovarian cancer, prostate cancer

•Depletes MDSCs, possibly by inhibiting STAT3-dependent gene expression

•Induces MCP1 expression on tumour cells, in turn driving the establishment of an immunosuppressive stroma

93,96

Doxorubicin Several solid and haematopoietic tumours

•Favours the proliferation of tumour-specific CD8+ T cells•Promotes tumour infiltration by IL-17-secreting γδ T cells and

activated IFNγ-secreting CD8+ T cells•Induces immunogenic cell death•Stimulates antigen presentation by DCs•Increases the permeability of tumour cells to granzyme B•Induces MCP1 expression on tumour cells, in turn driving the

establishment of an immunosuppressive stroma

84,85,89, 92,94,99

Gemcitabine NSCLC, pancreatic cancer, bladder cancer, breast cancer

•Increases class I HLA expression•Enhances tumour antigen cross-presentation•Selectively kills MDSCs

68–72

Methotrexate Several solid and haematopoietic tumours

•Immunosuppressive at high doses•Stimulates DC maturation and function

102,104

Mitoxantrone Breast cancer, AML, non-Hodgkin’s lymphoma

•Induces immunogenic cell death 89

Oxaliplatin Colorectal cancer •Increases class I HLA expression•Inhibits PDL2 expression•Induces immunogenic cell death

68,73, 87,88

Paclitaxel Kaposi’s sarcoma, lung cancer, ovarian cancer, breast cancer, head and neck cancer

•Specifically impairs cytokine production and viability in FOXP3+ T

reg cells but not in FOXP– CD4+ effector cells

•Stimulates antigen presentation by DCs•Increases the permeability of tumour cells to granzyme B•Induces MCP1 expression on tumour cells, in turn driving the

establishment of an immunosuppressive stroma

91,92, 95,99

Vincristine Non-Hodgkin’s lymphoma

•Increases the abundance of CD83+ DCs and type I DCs•Stimulates antigen presentation by DCs

98,99

Vinorelbine Breast cancer •Vinorelbine-treated tumour cells facilitate the bystander death of immune cells

100

AML, acute myeloid leukaemia; DC, dendritic cell; HLA, human leukocyte antigen; IFNγ, interferon-γ; IL-2, interleukin-2; MCP1, monocyte chemoattractant protein 1; MDSC, myeloid-derived suppressor cell; NK, natural killer; NSCLC, non-small-cell lung cancer; PDL2, programmed death ligand 2; STAT3, signal transducer and activator of transcription 3; T

reg, regulatory T cell

(forkhead box P3 (FOXP3)+ immunoregulatory lymphocyte).

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 223

© 2012 Macmillan Publishers Limited. All rights reserved

FOXP3+

Nature Reviews | Drug Discovery

B cell

CD20+

CD5+

sIgM+

B cell

FOXP3+

Erlotinib Cetuximab Imatinib Nilotinib

Trastuzumab

Histone deacetylase inhibitorsNitrogen bisphosphonates Bevacizumab Rituximab Cetuximab

Dasatinib Sunitinib Sorafenib Bortezomib ABT-737

T cell

T cell

T cell

T cell

Treg cell

MDSC

DC

DC

IL-10

T cell

CD8+

T cell

TH1

T cell

CD8+

IFNγ

TNF

α-CHO antibodies

T cell

CD4+

T cell

Mφ

CD4+

IFNγ

AKT

RAF

ERK

MHC-IIMHC-I

Tumour cell

IDO

STAT3

STAT5

Treg cellNK cell

NK cell

PDL1

ADCCADCPCDC

Indoleamine 2,3-dioxygenaseAn enzyme that catalyses the degradation of the essential amino acid l-tryptophan to N-formylkynurenine, which is an immunosuppressive metabolite.

has been shown to inhibit STAT3 and STAT5 signal-ling, decrease the frequency of Treg cells and impair their immunosuppressive function in vivo117.

The molecular mechanisms underlying the immuno-genic effects of imatinib have just begun to emerge. Thus, recent data obtained from a murine model of spontaneous GIST as well as from a GIST patient cohort suggest that imatinib stimulates anticancer immunity by reducing the expression of the immunosuppressive enzyme indoleamine 2,3-dioxygenase by tumour cells118. Thus, even though imatinib was suggested to inhibit T cell proliferation and cytokine secretion in vitro119–121 as well as the generation of antigen-specific memory CD8+ T cells in vivo122, it seems that at least part of its therapeutic efficacy results from the elicitation of an antitumour immune response.

Other tyrosine kinase inhibitors that may activate the immune system against cancer include sunitinib (Sutent; Pfizer), which is currently used for the treat-ment of RCC, and sorafenib (Nexavar; Bayer), which is now used to treat RCC and advanced hepatocellular carcinoma. Both agents have been associated with decreased levels of infiltrating Treg cells and MDSCs in patients with RCC123–127, and facilitate the develop-ment of antitumour TH1 responses125. Of note, pre-conditioning with sunitinib has been ascribed with immuno stimulatory functions even in murine models of protozoan infection, increasing the frequency of IFNγ+ cells and IFNγ+TNF+CD4+ cells, stimulating the activity of splenic macrophages and synergizing with a prototypic immune-dependent chemotherapeutic (sodium stibogluconate) to eradicate pathogens128.

Figure 2 | Mechanisms through which targeted anticancer agents affect the immune system. Targeted anticancer agents can facilitate anticancer immune responses by: directly activating CD4+ or CD8+ T cells, leading to the production of tumour necrosis factor (TNF) or interferon-γ (IFNγ); stimulating natural killer (NK) and dendritic cell (DC) functions as well as the DC–NK crosstalk; inhibiting or depleting immunosuppressive myeloid-derived suppressor cells (MDSCs) and regulatory T (T

reg) cells; upregulating class I or class II major histocompatibility complex (MHC) molecules on cancer

cells; inducing antibody-dependent cellular cytotoxicity (ADCC), antibody-dependent cellular phagocytosis (ACDP) or complement-dependent cytotoxicity (CDC); inhibiting indoleamine 2,3-dioxygenase (IDO); or stimulating CD20+CD5+sIgM+ B lymphocytes that produce anti-carbohydrate (α-CHO) antibodies with anticancer activity. Conversely, targeted anticancer agents can inhibit immune responses by functionally impairing or depleting B cells, CD4+ and CD8+ T lymphocytes, NK cells or DCs, by facilitating the recruitment of macrophages (MΦ) or by activating them to emit immunosuppressive signals. CD20+CD5+sIgM+; B lymphocytes expressing at their surface CD5, CD20 and immunoglobulin M (IgM) molecules; ERK, extracellular signal-regulated kinase; FOXP3, transcription factor forkhead box P3 (FOXP3); IL-10, interleukin-10; PDL1, programmed death ligand 1; sIgM, surface immunoglobulin M; STAT3, signal transducer and activator of transcription 3; T

H1, T helper cell type 1.

R E V I E W S

224 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Phosphoinositide 3-kinases(PI3Ks). A family of kinases that encompasses three distinct classes of lipid kinases that can be discriminated based on primary structure, substrate specificity and regulation. PI3Ks regulate an extraordinarily diverse group of cellular functions.

BH3 mimeticA small, plasma membrane- permeant chemical that is structurally similar to the BCL-2 homology 3 (BH3) domain of pro-apoptotic BCL-2 family members.

Antibody-dependent cellular cytotoxicityThe process whereby an effector cell of the innate immune system (for example, a natural killer cell) kills a target cell that has been bound by specific antibodies.

Antibody-dependent cellular phagocytosisThe process whereby a phagocytic cell of the innate immune system (for example, a macrophage) takes up and degrades a target cell that has been bound by specific antibodies.

Complement-dependent cytotoxicityThe activation of the complement cascade against a target cell that has been bound by specific antibodies, leading to the formation of a membrane attack complex and cell lysis.

CD40A co-stimulatory molecule (the receptor for the CD40 ligand; also known as CD154) that is expressed by antigen- presenting cells and required for their activation.

Inhibitor of apoptosis proteinsA family of functionally and structurally related proteins that inhibit caspase activation but also exert E3 ubiquitin ligase activity following death receptor ligation.

By contrast, sorafenib reportedly inhibits an array of DC functions, including cytokine secretion, the expression of co-stimulatory molecules and T cell activation129.

Along similar lines, dasatinib (Sprycel; Bristol-Myers Squibb), a tyrosine kinase inhibitor that is approved for the treatment of imatinib-refractory leukaemia, inhibits Treg cells as well as antigen-specific cytotoxic CD8+ lymphocytes130–132 and blocks multiple T cell func-tions in vitro and in vivo133–135, perhaps by inhibiting lymphocyte-specific protein tyrosine kinase (LCK)136, overall favouring immunosuppression. The broad-spectrum tyrosine kinase inhibitor nilotinib (Tasigna; Novartis), which is currently used to treat some forms of CML, inhibits antigen-specific CD8+ T cell prolif-eration137 at pharmacological concentrations and, simi-larly to dasatinib, it reduces NK cell cytotoxicity and IFNγ generation138. Bortezomib (Velcade, Millennium Pharmaceuticals), a moderately specific inhibitor of proteasomal degradation that is currently used to treat multiple myeloma, reportedly suppresses DC139 and CD4+ cell functions140, and decreases CD4+ and CD8+ cell counts141,142, yet spares Treg cells143.

Owing to the crucial role of phosphoinositide 3-kinases (PI3Ks) in multiple aspects of cell biology, the PI3K network constitutes an attractive target for anticancer therapy144, yet most — if not all — PI3K inhibitors (for example, wortmannin) inhibit immune cells and exert consistent immunosuppressive effects145. In a murine model of pancreatic cancer, the BH3 mimetic ABT-737 (which exerts anticancer effects by inhibiting several members of the BCL-2 protein family) reduces the abun-dance of specific lymphocyte and DC subpopulations, limits the persistence of memory B cells and inhibits the development of cytotoxic T cell responses146. Taken together, these results illustrate the notion that the immu-nological side effects of targeted anticancer agents are not always therapeutically beneficial.

Therapeutic monoclonal antibodies constitute another broad category of targeted anticancer agents. Although some (such as rituximab) exert anticancer effects by activating innate immune effector mechanisms such as antibody-dependent cellular cytotoxicity147, antibody-dependent cellular phagocytosis148 or complement-dependent cytotoxicity149, others — such as the anti-EGFR agents cetuximab and panitumumab (Vectibix; Amgen) — were designed and developed to operate as highly selective inhibitors of cancer-related signalling cascades. Recent data indicate that therapeutic antibodies from both of these categories also engage adaptive immune mechanisms against cancer. For instance, cetuximab favours not only NK cell-mediated antibody-dependent cellular cytotoxicity150 but also the stimulation of tumour antigen-specific CD8+ cells by DCs151.

In addition to inhibiting HER2-transduced signals, trastuzumab (Herceptin; Roche/Genentech), a HER2- targeting antibody approved for the treatment of advanced breast cancer, favours the generation of HER2- specific cytotoxic CD8+ lymphocytes152 and supports tumour infiltration by NK cells153. Bevacizumab, a VEGF-specific antibody that is currently used for the treatment of metastatic cancers, reduces the percentage

of circulating Treg cells in patients with colorectal car-cinoma154, repletes B cell and T cell compartments in patients with metastatic colorectal cancer155, favours the differentiation of DCs156 and facilitates tumour infil-tration by lymphocytes157. Of note, it has recently been shown that cetuximab-coated EGFR-expressing colo-rectal carcinoma cells can be recognized by CD163+

macrophages, leading to the FcγR-dependent generation of membrane-bound (for example, PDL1) and soluble (for example, IL-10) immunosuppressive signals158. These results suggest that, at least in some instances, therapeu-tic antibodies can engage unwanted immunosuppressive mechanisms.

Among therapeutic antibodies, so-called immuno-stimulatory antibodies — that is, targeted therapeutics that promote antitumour immunity by directly modu-lating immune functions — deserve special mention147. These include: ipilimumab (Yervoy; Bristol-Myers Squibb), which is an FDA-approved molecule that antagonizes cytotoxic T lymphocyte antigen 4 (CTLA4) on the surface of helper T cells, thus inhibiting the development of immune tolerance159; MDX-1106, which specifically targets programmed death protein 1 (PD1), a transmembrane receptor that mediates immuno-suppressive functions in activated T cells160; several anti-bodies that block T cell immunoglobulin and mucin domain-containing protein 3 (TIM3), a transmembrane protein that identifies antigen-specific dysfunctional CD8+ cells161; and many antibodies that activate CD40, in turn stimulating antigen-presenting cells for tumour-specific T cell priming and activation162, alone or in combination with other anticancer interventions (including the con-ventional chemotherapeutics gemcitabine and dacar-bazine as well as the targeted agent AZD8055). Although a detailed description of these agents is beyond the scope of this Review, it is worth noting that these immunostimula-tory antibodies reportedly reverse immune tolerance and induce consistent antitumour responses163–168.

Various other targeted anticancer agents have been shown to exert immunostimulatory functions. For instance, nitrogen bisphosphonates (such as zoledro-nate, which is commonly prescribed for osteoporosis) have been shown to trigger inflammatory and/or innate immunity169, delay the progression of breast cancer170, and facilitate the expansion of patient-derived DCs for the generation of efficient vaccines against hepatocellu-lar carcinoma171. Chemical agents antagonizing inhibitor of apoptosis proteins, which operate at the crossroad between cell death induction and nuclear factor-κB signalling, not only facilitate the demise of cancer cells via tumour cell-intrinsic mechanisms172 but also enhance co-stimulation, thus lowering the threshold for the activation of the immune system in response to physiological stimuli and hence augmenting the efficacy of both prophylactic and therapeutic antitumour vac-cines173. Similarly, besides facilitating tumour cell death, several histone deacetylase inhibitors (such as vorinostat (Zolinza; Merck), sodium butyrate and MS-275) report-edly increase the expression of NK-activating receptor ligands on the surface of cancer cells, thereby facilitating tumour cell recognition by NK cells174.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 225

© 2012 Macmillan Publishers Limited. All rights reserved

Histone deacetylase inhibitorsA class of compounds that interfere with the function of histone deacetylases, thereby affecting the epigenetic regulation of transcription.

Exosome-based vaccinesAnticancer vaccines that are based on 30–90 nm vesicles originating from late endosomes. These vesicles are secreted by professional antigen-presenting cells and, because of their composition, they exert profound immunostimulatory functions.

Wilms’ tumour 1A transcription factor encoded by the Wilms’ tumour 1 (WT1) gene that has an essential role in the normal development of the urogenital system.

NY-ESO-1A highly immunogenic cancer–testis antigen that is widely used in clinical cancer vaccine trials.

In summary, it appears that targeted anticancer agents often affect the immune system, thereby exerting unwanted immunosuppressive functions or stimulating therapy-relevant anticancer immune responses (TABLE 3).

ImmunochemotherapyA growing body of literature suggests that many anti-cancer compounds, including conventional chemo-therapeutics and targeted agents, cooperate with immunostimulatory interventions including vaccina-tion protocols and the administration of immunogenic cytokines to achieve efficient anticancer immunity. Here, we present promising immunochemotherapeutic strat-egies while deliberately overlooking the large number of studies that are based on purely immunomodulatory approaches (reviewed in detail in REF. 175).

At non-immunosuppressive doses doxorubicin and cyclophosphamide have been shown to enhance the effi-cacy of a HER2-positive allogeneic vaccine (that secretes granulocyte–macrophage colony-stimulating factor) in patients with metastatic breast cancer176. In addition, cyclophosphamide reportedly increases the efficiency of exosome-based vaccines, an effect that has been associated with limited Treg cell-mediated immunosuppression and enhanced secondary (but not primary) cytotoxic anti-tumour responses77. A similar synergy was observed when docetaxel was coupled to a recombinant viral vac-cine in patients with metastatic androgen-independent prostate cancer177,178, and when imatinib and oxaliplatin were administered in combination with DC vaccination in a murine model of imatinib-resistant lymphoma and in patients with colorectal cancer, respectively117,179.

Moreover, imatinib effectively and safely increased the efficacy of Wilms’ tumour 1 peptide vaccination in a patient with imatinib-resistant CML180, and oxaliplatin has recently been shown to synergize with an adenoviral system for the liver-specific, inducible expression of IL-12 in the elimination of liver metastasis in mice181, whereas similar results have not been observed with nucleoside analogues such as gemcitabine and 5-fluorouracil. This phenomenon was paralleled by a shift in the tumour-immune microenvironment, including increased levels of CD8+ cells and reduced levels of MDSCs181, which may have been linked with the ability of oxaliplatin to trigger immunogenic cell death87.

In support of this notion, it has recently been shown that once they are loaded on DCs, apoptotic and necrotic bodies from tumour cells that are prone to undergo immunogenic cell death in vitro induce clinical and immunological responses in patients with indolent non-Hodgkin’s lymphoma; however, these responses are not observed if tumour cells fail to acquire the hallmarks of immunogenic cell death as they die182. This demon-strates that the crosstalk between dying tumour cells and components of the immune system, both in vivo (during cancer chemotherapy) and in vitro (in the context of the generation of autologous DC vaccines) is a crucial deter-minant for the therapeutic outcome of anticancer therapy.

As melanoma appears to be a privileged setting for anticancer vaccines, considerable efforts have been made in developing efficient strategies for vaccination.

The most recent of these efforts involve combination regimens with conventional and targeted anticancer agents, which have demonstrated promising results183. In disease-free melanoma patients the administration of dacarbazine before peptide vaccination increased the number of antitumour long-lasting effector memory CD8+ T cells184, possibly owing to a widening of the tumour antigen-specific T cell receptor repertoire185. Temozolomide, another alkylating agent that is used to treat brain tumours and melanoma, has recently been shown to increase the rate of immunological responders among patients with stage IV melanoma undergoing vaccination with the telomerase peptide vaccine GV1001 (REF. 186). In mice, CD8+ responses elicited by HLA-A2-restricted peptides from auto-immunogenic cancer–testis antigen  1 (NY-ESO-1; also known as CTAG1) can be improved by the administration of gem-citabine187, in line with the immunostimulatory effects observed in patients treated with this nucleoside ana-logue (as mentioned above)68,71,187.

Anticancer vaccination strategies based on the recombinant NY-ESO-1 protein, NY-ESO-1-pulsed DCs or NY-ESO-1-reactive lymphocytes have been extensively studied in the past decade, yielding prom-ising results188–190. However, recent data indicate that in some instances peptide immunization may lead to antigen-specific T cell apoptosis in lymph nodes and hence cause therapeutic failure, which can be prevented using unmethylated CpG sequences as an adjuvant191. Appropriate adjuvants that protect effector T cells from activation-induced death may therefore constitute a prime requirement for the success of anticancer vaccines, and CpG oligonucleotides have already been shown in several instances to preserve — if not ameliorate — the efficacy of immunotherapy and immunochemotherapy in both preclinical and clinical settings192–194. As a final example, the targeted agent temsirolimus — a non-immunosuppressive analogue of rapamycin that is used for RCC therapy — enhanced IFNγ secretion and cyto-toxic responses by T cells in mice treated with a heat shock protein-based antitumour vaccine195.

Together, these observations suggest that conven-tional chemotherapeutics or targeted anticancer agents with immunostimulatory functions may be beneficial in the development of cancer immunochemotherapy protocols.

Concluding remarksThe molecular and cellular pathways whereby anti-tumour immunity (endogenous or elicited by immuno-stimulatory interventions) is enhanced by chemotherapy are complex and poorly understood. It has been sug-gested that immunosuppressive Treg cells and MDSCs may be more sensitive to conventional chemotherapeu-tics than effector T cells, possibly because of their higher proliferation rates, resulting in the stimulation of anti-tumour immunity196. Although this appears to be plau-sible in some instances, it is unlikely to account for the immunostimulatory effects of all conventional chemo-therapeutics. Moreover, it surely does not explain how targeted anticancer agents enhance immune functions,

R E V I E W S

226 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Table 3 | Examples of effects of targeted therapeutic agents on the immune system

Agents Indications Notes Refs

CD40-specific antibodies

Experimental drugs •Immunomodulatory antibodies that activate CD40 147,162, 166,167

TIM3-specific antibodies

Experimental drugs •Immunomodulatory antibodies that target TIM3 147,161,163

Bevacizumab Colorectal cancer, lung cancer, kidney cancer

•Depletes circulating Treg

cells•Repletes B and T cell compartments•Favours the differentiation of DCs•Facilitates tumour infiltration by lymphocytes

154–157,229

BH3 mimetics Experimental drugs •Deplete specific DC and T cell populations•Limit the persistence of memory B cells•Inhibit cytotoxic T cell responses

146

Bortezomib Multiple myeloma •Suppresses DC and CD4+ cell functions•Decreases CD4+ and CD8+ cell counts•Does not affect T

reg cells

139–143

Cetuximab Colorectal cancer, head and neck cancer

•Activates NK cell-mediated ADCC•Stimulates the activation of antigen-specific CD8+ cells by DCs•Increases class I and II MHC expression•Stimulates the activation of CD163+ macrophages, leading to

immunosuppression

112,150, 151,158

Dasatinib Imatinib-refractory chronic myelogenous leukaemia

•Inhibits Treg

cells as well as antigen-specific CD8+ T cell functions•Blocks several T cell functions, in vitro and in vivo, perhaps by inhibiting

LCK activity•Reduces NK cell-mediated cytotoxicity and IFNγ secretion

130–136, 138

Erlotinib NSCLC, pancreatic cancer

•Increases class I and II MHC expression•Impairs T cell responses by inhibiting RAF-, ERK- and AKT-transduced signals

112,114

HDAC inhibitors Cutaneous T cell lymphoma

•Upregulate activating NK cell receptor ligands on the surface of tumour cells 174

IAP inhibitors Experimental drugs •Enhance co-stimulation, thereby lowering the threshold for the activation of anticancer immunity

173

Imatinib Chronic myelogenous leukaemia, GIST

•Stimulates the production of carbohydrate-specific antibodies with antitumour activity

•Stimulates the development of antileukaemic TNF-secreting CD4+ T cells•Activates NKp30-mediated NK cell functions•Interferes with immunosuppressive functions of T

reg cells while inhibiting

STAT3 and STAT5 activity•Limits IDO expression by tumour cells•Inhibits T cell proliferation and cytokine secretion in vitro•Inhibits antigen-specific CD8+ T cells in vivo

55,66, 115–122

Ipilimumab Melanoma •Immunomodulatory antibody that targets CTLA4 147,159, 165,168

MDX-1106 Experimental drug •Immunomodulatory antibody that targets PD1 147,160,161, 163,164

Nilotinib Chronic myelogenous leukaemia

•Inhibits antigen-specific CD8+ T cell proliferation•Reduces NK cell-mediated cytotoxicity and IFNγ secretion

137,138

Nitrogen bisphosphonates

Osteoporosis •Trigger inflammation and/or innate immunity•Delay the progression of breast cancer•Facilitate the expansion of patient-derived DCs

169–171

PI3K inhibitors Experimental drugs •Exert profound immunosuppressive effects 145

Sorafenib Renal cell carcinoma, advanced hepato cellular carcinoma

•Limits infiltration by Treg 

cells and MDSCs•Inhibits cytokine secretion, expression of co-stimulatory molecules and T cell

activation by DCs

126,129

Sunitinib Renal cell carcinoma •Limits infiltration by Treg

cells and MDSCs while inhibiting STAT3 activity•Increases the frequency of IFNγ+ and IFNγ+TNF+CD4+ cells, and stimulates the

activity of splenic macrophages

123–125, 127,128

Trastuzumab Advanced-stage breast cancer

•Favours the generation of HER2-specific CD8+ cells•Stimulates tumour infiltration by NK cells

152,153

ADCC, antibody-dependent cellular cytotoxicity; BH3, BCL-2 homology 3; CTLA4, cytotoxic T lymphocyte antigen 4; DC, dendritic cell; ERK, extracellular signal-regulated kinase; GIST, gastrointestinal stromal tumour; HDAC, histone deacetylase; IAP, inhibitor of apoptosis protein; IDO, indoleamine 2,3-dioxygenase; IFNγ, interferon-γ; LCK, lymphocyte-specific protein tyrosine kinase; MDSC, myeloid-derived suppressor cell; MHC, major histocompatibility complex; NK, natural killer; NKp30, activating NK receptor p30; NSCLC, non-small-cell lung cancer; PD1, programmed death protein 1; PI3K, phosphoinositide 3-kinase; STAT3, signal transducer and activator of transcription 3; TIM3, T cell immunoglobulin and mucin domain-containing molecule 3; TNF, tumour necrosis factor; T

reg, regulatory T cell

(forkhead box P3 (FOXP3)+ immunoregulatory lymphocyte).

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 227

© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

Mature DC

Tumour-specificeffector memory T cells

CD8+

CD8+

CD8+

CD8+

CRT

ICD

Tumour cell

Tumourantigens

MHC-I

VaccinesPhagosome

HMGB1

ATP

Immature DC

Elicitation of long-lasting protective T cell responses

Therapeutic success

AnthracyclinesCyclophosphamideOxaliplatin

TLR4

P2RX7

Chemotherapy andimmunotherapy

Othermedications

ER stress

Non-responders Responders

ER stressorsER stress

HMGB1 release HMGB1 release

TLR4 signalling

TLR4 agonists

TLR4 signalling

ATP release ATP release

P2RX7P2RX7 agonists

P2RX7

Ecto-ATPase inhibitors

CRT exposure rCRT CRT exposure

Antigen uptakeby DCs

Antigen uptakeby DCs

Patient selection

Therapeutic regimens

Dosage Druginteractions

Order ofadministration

a

b

c

CRTR

a phenomenon that presumably relies on private (com-pound-specific) off-target signalling pathways. Of note, disrupted STAT signalling has been associated in several instances with decreased immunosuppression by

Treg cells, MDSCs and DCs73,117,127, resulting in efficient antitumour immunity. The STAT family of transcription factors, notably STAT3, which controls the expression of several immunosuppressive molecules including PDL1

Figure 3 | Requirements for successful immunochemotherapy. Three elements appear to be necessary for the therapeutic success of immunochemotherapy. First, long-lasting protective T cell responses must be elicited. This is achieved following the induction of immunogenic cell death (ICD) by specific chemotherapeutic agents or after the administration of efficient anticancer vaccines. Second, patients who have the potential to benefit from immunochemotherapy must be properly selected. Such patients are proficient in the signalling pathways that link ICD inducers and anticancer vaccines to the elicitation of a long-lasting protective T cell response. Of note, defects in some of the molecular mechanisms that underlie the productive crosstalk between dying cancer cells and the immune system can be corrected by exogenous interventions, de facto widening the fraction of patients that might successfully respond to immunochemotherapy. Third, appropriate therapeutic regimens must be designed. With respect to this, both the dosage and the order of administration of chemotherapeutic agents relative to immunotherapeutic interventions have been shown to influence the success of immunochemotherapy. In addition, drug interactions — which are particularly frequent among patients with cancer — should be carefully evaluated. CRT, calreticulin; CRTR, CRT receptor; DC, dendritic cell; ER, endoplasmic reticulum; HMGB1, high-mobility group protein B1; MHC, major histocompatibility complex; P2RX7, P2 purinergic receptor 7; rCRT, recombinant CRT; TLR4, Toll-like receptor 4.

R E V I E W S

228 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Fibroblast activation protein-αA transmembrane serine protease that is induced on reactive stromal fibroblasts in epithelial cancers, sarcomas and granulation tissue, and may be involved in tumour invasion, tissue remodelling and wound repair.

Oncogene addictionThe phenomenon whereby cancer cells depend on the continuous hyperactivation of one or more oncogenes for their survival.

and PDL2, therefore appears to be a therapeutic target for immunostimulation197. However, STAT3-specific inhibitors such as Stattic have only recently become available198 and have not yet been tested in this setting.

Several studies have recently investigated the impli-cations of the crosstalk between non-immune compo-nents of the tumour stroma and the immune system, with encouraging results. For instance, it has been shown that a small fraction of stromal cells expressing fibroblast activation protein-α mediate consistent immunosuppres-sive functions, and depletion of this protein results in the rapid necrosis of cancer cells via IFNγ- and TNF-dependent immune mechanisms199.

Along similar lines, the secretion of the chemokine CC motif ligand 21 by tumour cells has been identified as a central event in the generation of an immunotoler-ant lymphoid-like stroma, which features an impaired cytokine milieu and the accumulation of immunosup-pressive cell populations200. This phenomenon appears to be mediated by the chemokine CC motif receptor 7 (CCR7) on stromal but not tumour cells, and immuno-competence is restored to normal levels in Ccr7–/– mice or following the blockage of CCR7 by specific anti-bodies200. Together with the observation that aggressive cancers can be eradicated by the concomitant targeting of the tumour stroma by CD4+ and CD8+ T cells201, these results reinforce the notion that both the cellular and humoral components of the tumour cell microenviron-ment constitute promising targets for the development of successful immunochemotherapeutic strategies.

Irrespective of these considerations, three aspects appear to be crucial for the design and clinical implemen-tation of efficient immunochemotherapy regimens (FIG. 3). The first is the elicitation of long-term protec-tive T cell responses. The effects of several chemothera-peutics on immunosuppressive cell populations such as Treg cells and MDSCs are indeed reversible, implying that they progressively disappear at the end of treat-ment, whereas immunogenic cell death inducers and vaccination protocols result in the generation of long-lasting effector memory CD8+ T cells, which persist for years after therapy and have the potential to prevent relapse202.

The second consideration is patient selection. The success of immunochemotherapy is based on the pro-ductive crosstalk between dying cancer cells and the host immune system, and this is mediated by several molecular and functional determinants203. For instance, there appear to be at least three conditions that are nec-essary for cell death to be immunogenic (BOX 2). Thus, only a fraction of patients can productively respond to immunochemotherapy, raising the need for patient stratification based on reliable and routinely assessable biomarkers. Recent preclinical data suggest that defects in some of the determinants that underlie a therapeuti-cally favourable crosstalk between dying tumour cells and the immune system can be corrected88,89. Although this has not yet been formally confirmed in the clinic, the fraction of patients that would eventually benefit from immunochemotherapy might therefore be larger than expected. Intriguingly, the immune system has

been found to contribute to tumour regression even in models of interrupted oncogene addiction, which has always been considered as a merely cancer cell-intrinsic response204, perhaps implying that there may be an overlap between the patient populations that benefit from immunochemotherapy and those that respond to chemotherapy.

The third consideration regards therapeutic regi-mens. Recent studies suggest that not only the dosage176 but also the order of administration of chemo- and immunotherapy has a crucial role in determining the patient’s response92,205–207. There are contrasting reports in the literature; some studies indicate that chemo-therapy should be given first205,206, whereas others suggest that immunotherapy should be given first92,207. Different tumour types and/or different immunochemotherapy regimens might therefore behave differently with respect to this. Moreover, patients with cancer are particularly at risk for drug interactions, as they often receive a wide panel of medications — as part of the anticancer ther-apy itself, for managing its side effects or for treating coexisting pathologies (especially in elderly patients)208. Some of these interactions may impinge on the outcome of immunochemotherapy and hence require careful consideration. For instance, glucocorticoids — which are often given to patients with cancer for reducing inflammation-related pain — are potent immunosup-pressors209. Conversely, histamine blockers — which are normally administered to manage the gastrointestinal side effects of anticancer therapy — reportedly enhance anticancer immune functions210.

As it stands, one of the major obstacles in the develop-ment of efficient immunochemotherapeutic protocols is the consistent gap in our knowledge of the mechanisms whereby conventional chemotherapeutics and targeted anticancer agents influence the immune system. These drugs are known to exert widespread effects on a pleth-ora of distinct cell types, yet only some of the underlying molecular cascades have been precisely characterized. Additional studies will have to fill in this gap, thereby rendering the overall picture less nebulous. Irrespective of these considerations, it is becoming clear that actively mobilizing our increasingly less secret ally — the immune system — will be beneficial for the development of future anticancer therapies.

Until now, the development of anticancer drugs has been based on two phases: a first phase of preclinical investigation (in vitro and in vivo, on human tumours xenotransplanted in immunodeficient mice), which ignores any possible contribution of the immune system to therapeutic efficacy; and a second phase of clinical development, which features a deceptively small rate of successes. It appears plausible that, so far, chemothera-peutic agents that activate the immune system against cancer have been selected based on the empirical cri-terion of efficacy rather than on the rational seeking of immunostimulatory effects. If the paradigm that we propose in this article holds true, the combination of immunological and oncological expertise will be advantageous in the development of novel anticancer strategies.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 229

© 2012 Macmillan Publishers Limited. All rights reserved

1. Czernilofsky, A. P. et al. Nucleotide sequence of an avian sarcoma virus oncogene (src) and proposed amino acid sequence for gene product. Nature 287, 198–203 (1980).

2. Roussel, M. et al. Three new types of viral oncogene of cellular origin specific for haematopoietic cell transformation. Nature 281, 452–455 (1979).

3. Chabner, B. A. & Roberts, T. G. Jr. Chemotherapy and the war on cancer. Nature Rev. Cancer 5, 65–72 (2005).

4. Hanahan, D. & Weinberg, R. A. The hallmarks of cancer. Cell 100, 57–70 (2000).

5. Nowell, P. & Hungerford, D. A minute chromosome in chronic granulocytic leukemia. Science 132, 1497 (1960).

6. Rowley, J. D. A new consistent chromosomal abnormality in chronic myelogenous leukaemia identified by quinacrine fluorescence and Giemsa staining. Nature 243, 290–293 (1973).

7. Stam, K. et al. Evidence of a new chimeric bcr/c-abl mRNA in patients with chronic myelocytic leukemia and the Philadelphia chromosome. N. Engl. J. Med. 313, 1429–1433 (1985).

8. Ben-Neriah, Y., Daley, G. Q., Mes-Masson, A. M., Witte, O. N. & Baltimore, D. The chronic myelogenous leukemia-specific P210 protein is the product of the bcr/abl hybrid gene. Science 233, 212–214 (1986).The studies in REF. 7 and REF. 8 reported the molecular characterization of the transcript and protein products of the BCR–ABL gene, laying the foundations of targeted anticancer therapy.

9. Capdeville, R., Buchdunger, E., Zimmermann, J. & Matter, A. Glivec (STI571, imatinib), a rationally developed, targeted anticancer drug. Nature Rev. Drug Discov. 1, 493–502 (2002).

10. Apperley, J. F. et al. Response to imatinib mesylate in patients with chronic myeloproliferative diseases with rearrangements of the platelet-derived growth factor receptor β. N. Engl. J. Med. 347, 481–487 (2002).

11. Demetri, G. D. et al. Efficacy and safety of imatinib mesylate in advanced gastrointestinal stromal tumors. N. Engl. J. Med. 347, 472–480 (2002).

12. Dowell, J., Minna, J. D. & Kirkpatrick, P. Erlotinib hydrochloride. Nature Rev. Drug Discov. 4, 13–14 (2005).

13. Moy, B., Kirkpatrick, P., Kar, S. & Goss, P. Lapatinib. Nature Rev. Drug Discov. 6, 431–432 (2007).

14. Blankenstein, T. The role of tumor stroma in the interaction between tumor and immune system. Curr. Opin. Immunol. 17, 180–186 (2005).

15. Ferrara, N., Hillan, K. J., Gerber, H. P. & Novotny, W. Discovery and development of bevacizumab, an anti-VEGF antibody for treating cancer. Nature Rev. Drug Discov. 3, 391–400 (2004).

16. Zitvogel, L., Apetoh, L., Ghiringhelli, F. & Kroemer, G. Immunological aspects of cancer chemotherapy. Nature Rev. Immunol. 8, 59–73 (2008).

17. Schilsky, R. L. Personalized medicine in oncology: the future is now. Nature Rev. Drug Discov. 9, 363–366 (2010).

18. Zitvogel, L., Kepp, O. & Kroemer, G. Immune parameters affecting the efficacy of chemotherapeutic regimens. Nature Rev. Clin. Oncol. 8, 151–160 (2011).

19. Lenz, G. et al. Stromal gene signatures in large-B-cell lymphomas. N. Engl. J. Med. 359, 2313–2323 (2008).

20. Nardin, A. et al. Dacarbazine promotes stromal remodeling and lymphocyte infiltration in cutaneous melanoma lesions. J. Invest. Dermatol. 131, 1896–1905 (2011).

21. Staaf, J. et al. Identification of subtypes in human epidermal growth factor receptor 2-positive breast cancer reveals a gene signature prognostic of outcome. J. Clin. Oncol. 28, 1813–1820 (2010).

22. Thurlow, J. K. et al. Spectral clustering of microarray data elucidates the roles of microenvironment remodeling and immune responses in survival of head and neck squamous cell carcinoma. J. Clin. Oncol. 28, 2881–2888 (2010).

23. Desmedt, C. et al. Biological processes associated with breast cancer clinical outcome depend on the molecular subtypes. Clin. Cancer Res. 14, 5158–5165 (2008).

24. Desmedt, C. et al. Multifactorial approach to predicting resistance to anthracyclines. J. Clin. Oncol. 29, 1578–1586 (2011).

25. Paulson, K. G. et al. Transcriptome-wide studies of Merkel cell carcinoma and validation of intratumoral CD8+ lymphocyte invasion as an independent predictor of survival. J. Clin. Oncol. 29, 1539–1546 (2011).

26. Eerola, A. K., Soini, Y. & Paakko, P. A high number of tumor-infiltrating lymphocytes are associated with a small tumor size, low tumor stage, and a favorable prognosis in operated small cell lung carcinoma. Clin. Cancer Res. 6, 1875–1881 (2000).

27. Qian, B. Z. & Pollard, J. W. Macrophage diversity enhances tumor progression and metastasis. Cell 141, 39–51 (2010).

28. Kamper, P. et al. Tumor-infiltrating macrophages correlate with adverse prognosis and Epstein–Barr virus status in classical Hodgkin’s lymphoma. Haematologica 96, 269–276 (2011).

29. Komohara, Y. et al. Macrophage infiltration and its prognostic relevance in clear cell renal cell carcinoma. Cancer Sci. 102, 1424–1431 (2011).

30. Lee, C. H. et al. Prognostic significance of macrophage infiltration in leiomyosarcomas. Clin. Cancer Res. 14, 1423–1430 (2008).

31. Ding, T. et al. High tumor-infiltrating macrophage density predicts poor prognosis in patients with primary hepatocellular carcinoma after resection. Hum. Pathol. 40, 381–389 (2009).

32. Zhang, B. C. et al. Tumor-associated macrophages infiltration is associated with peritumoral lymphangiogenesis and poor prognosis in lung adenocarcinoma. Med. Oncol. 28, 1447–1452 (2010).

33. Nonomura, N. et al. Infiltration of tumour-associated macrophages in prostate biopsy specimens is predictive of disease progression after hormonal therapy for prostate cancer. BJU Int. 107, 1918–1922 (2011).

34. Kinouchi, M. et al. Infiltration of CD14-positive macrophages at the invasive front indicates a favorable prognosis in colorectal cancer patients with lymph node metastasis. Hepatogastroenterology 58, 352–358 (2011).

35. Ladoire, S. et al. Pathologic complete response to neoadjuvant chemotherapy of breast carcinoma is associated with the disappearance of tumor-infiltrating Foxp3+ regulatory T cells. Clin. Cancer Res. 14, 2413–2420 (2008).

36. Ladoire, S. et al. In situ immune response after neoadjuvant chemotherapy for breast cancer predicts survival. J. Pathol. 224, 389–400 (2011).

37. De Monte, L. et al. Intratumor T helper type 2 cell infiltrate correlates with cancer-associated fibroblast thymic stromal lymphopoietin production and reduced survival in pancreatic cancer. J. Exp. Med. 208, 469–478 (2011).

38. Pedroza-Gonzalez, A. et al. Thymic stromal lymphopoietin fosters human breast tumor growth by promoting type 2 inflammation. J. Exp. Med. 208, 479–490 (2011).

39. Fu, J. et al. Increased regulatory T cells correlate with CD8 T-cell impairment and poor survival in hepatocellular carcinoma patients. Gastroenterology 132, 2328–2339 (2007).

40. Shen, Z. et al. Higher intratumoral infiltrated Foxp3+ Treg numbers and Foxp3+/CD8+ ratio are associated with adverse prognosis in resectable gastric cancer. J. Cancer Res. Clin. Oncol. 136, 1585–1595 (2010).

41. Petersen, R. P. et al. Tumor infiltrating Foxp3+ regulatory T-cells are associated with recurrence in pathologic stage I NSCLC patients. Cancer 107, 2866–2872 (2006).

42. Correale, P. et al. Regulatory (FoxP3+) T-cell tumor infiltration is a favorable prognostic factor in advanced colon cancer patients undergoing chemo or chemo-immunotherapy. J. Immunother. 33, 435–441 (2010).

43. Ladoire, S., Martin, F. & Ghiringhelli, F. Prognostic role of FOXP3+ regulatory T cells infiltrating human carcinomas: the paradox of colorectal cancer. Cancer Immunol. Immunother. 60, 909–918 (2011).

44. Halama, N. et al. Localization and density of immune cells in the invasive margin of human colorectal cancer liver metastases are prognostic for response to chemotherapy. Cancer Res. 71, 5670–5677 (2011).

45. Bonertz, A. et al. Antigen-specific Tregs control T cell responses against a limited repertoire of tumor antigens in patients with colorectal carcinoma. J. Clin. Invest. 119, 3311–3321 (2009).

46. Sconocchia, G. et al. Tumor infiltration by FcγRIII (CD16)+ myeloid cells is associated with improved survival in patients with colorectal carcinoma. Int. J. Cancer 128, 2663–2672 (2011).

47. Alvaro-Naranjo, T. et al. Tumor-infiltrating cells as a prognostic factor in Hodgkin’s lymphoma: a quantitative tissue microarray study in a large retrospective cohort of 267 patients. Leuk. Lymphoma 46, 1581–1591 (2005).

48. Polcher, M. et al. Foxp3+ cell infiltration and granzyme B+/Foxp3+ cell ratio are associated with outcome in neoadjuvant chemotherapy-treated ovarian carcinoma. Cancer Immunol. Immunother. 59, 909–919 (2010).

49. Distel, L. V. et al. Tumour infiltrating lymphocytes in squamous cell carcinoma of the oro- and hypopharynx: prognostic impact may depend on type of treatment and stage of disease. Oral Oncol. 45, e167–e174 (2009).

50. Wu, X. J. et al. Circulating antibodies to carcinoembryonic antigen related to improved recurrence-free survival of patients with colorectal carcinoma. J. Int. Med. Res. 39, 838–845 (2011).

51. Ait-Tahar, K. et al. Correlation of the autoantibody response to the ALK oncoantigen in pediatric anaplastic lymphoma kinase-positive anaplastic large cell lymphoma with tumor dissemination and relapse risk. Blood 115, 3314–3319 (2010).

52. Albertus, D. L. et al. AZGP1 autoantibody predicts survival and histone deacetylase inhibitors increase expression in lung adenocarcinoma. J. Thorac. Oncol. 3, 1236–1244 (2008).

53. Hamanaka, Y. et al. Circulating anti-MUC1 IgG antibodies as a favorable prognostic factor for pancreatic cancer. Int. J. Cancer 103, 97–100 (2003).

54. Touze, A. et al. High levels of antibodies against Merkel cell polyomavirus identify a subset of patients with Merkel cell carcinoma with better clinical outcome. J. Clin. Oncol. 29, 1612–1619 (2011).

55. Delahaye, N. F. et al. Alternatively spliced NKp30 isoforms affect the prognosis of gastrointestinal stromal tumors. Nature Med. 17, 700–707 (2011).The outcome of imatinib treatment in patients with GISTs was found to be influenced by the expression pattern of alternatively spliced NKp30 isoforms, thus unveiling an immune mechanism underlying at least part of the therapeutic efficacy of imatinib.

56. Cerhan, J. R. et al. Prognostic significance of host immune gene polymorphisms in follicular lymphoma survival. Blood 109, 5439–5446 (2007).

57. Kleinrath, T., Gassner, C., Lackner, P., Thurnher, M. & Ramoner, R. Interleukin-4 promoter polymorphisms: a genetic prognostic factor for survival in metastatic renal cell carcinoma. J. Clin. Oncol. 25, 845–851 (2007).

58. Sellick, G. S. et al. Scan of 977 nonsynonymous SNPs in CLL4 trial patients for the identification of genetic variants influencing prognosis. Blood 111, 1625–1633 (2008).

59. Domingo-Domenech, E. et al. Impact of interleukin-10 polymorphisms (–1082 and –3575) on the survival of patients with lymphoid neoplasms. Haematologica 92, 1475–1481 (2007).

60. DeMichele, A. et al. Host genetic variants in the interleukin-6 promoter predict poor outcome in patients with estrogen receptor-positive, node-positive breast cancer. Cancer Res. 69, 4184–4191 (2009).

61. Schoof, N. et al. Favorable impact of the interleukin-4 receptor allelic variant I75 on the survival of diffuse large B-cell lymphoma patients demonstrated in a large prospective clinical trial. Ann. Oncol. 20, 1548–1554 (2009).

62. Bibeau, F. et al. Impact of FcγRIIa–FcγRIIIa polymorphisms and KRAS mutations on the clinical outcome of patients with metastatic colorectal cancer treated with cetuximab plus irinotecan. J. Clin. Oncol. 27, 1122–1129 (2009).

63. Ferris, R. L., Jaffee, E. M. & Ferrone, S. Tumor antigen-targeted, monoclonal antibody-based immuno therapy: clinical response, cellular immunity, and immuno-escape. J. Clin. Oncol. 28, 4390–4399 (2010).

64. Wang, B., Kokhaei, P., Mellstedt, H. & Liljefors, M. FcγR polymorphisms and clinical outcome in colorectal cancer patients receiving passive or active antibody treatment. Int. J. Oncol. 37, 1599–1606 (2010).

65. Wilson, N. S. et al. An Fcγ receptor-dependent mechanism drives antibody-mediated target-receptor signaling in cancer cells. Cancer Cell 19, 101–113 (2011).Fcγ receptors on tumour-associated leukocytes were shown to provide a dynamic platform that facilitates the monoclonal antibody-dependent activation of death receptor 5 on tumour cells, and hence their death.

66. Menard, C. et al. Natural killer cell IFN-γ levels predict long-term survival with imatinib mesylate therapy in gastrointestinal stromal tumor-bearing patients. Cancer Res. 69, 3563–3569 (2009).

R E V I E W S

230 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

67. Gulley, J. L. et al. Immunologic and prognostic factors associated with overall survival employing a poxviral-based PSA vaccine in metastatic castrate-resistant prostate cancer. Cancer Immunol. Immunother. 59, 663–674 (2010).

68. Liu, W. M., Fowler, D. W., Smith, P. & Dalgleish, A. G. Pre-treatment with chemotherapy can enhance the antigenicity and immunogenicity of tumours by promoting adaptive immune responses. Br. J. Cancer 102, 115–123 (2010).

69. Nowak, A. K. et al. Induction of tumor cell apoptosis in vivo increases tumor antigen cross-presentation, cross-priming rather than cross-tolerizing host tumor-specific CD8 T cells. J. Immunol. 170, 4905–4913 (2003).

70. Nowak, A. K., Robinson, B. W. & Lake, R. A. Synergy between chemotherapy and immunotherapy in the treatment of established murine solid tumors. Cancer Res. 63, 4490–4496 (2003).

71. Mundy-Bosse, B. L. et al. Myeloid-derived suppressor cell inhibition of the IFN response in tumor-bearing mice. Cancer Res. 71, 5101–5110 (2011).

72. Vincent, J. et al. 5-fluorouracil selectively kills tumor-associated myeloid-derived suppressor cells resulting in enhanced T cell-dependent antitumor immunity. Cancer Res. 70, 3052–3061 (2010).

73. Lesterhuis, W. J. et al. Platinum-based drugs disrupt STAT6-mediated suppression of immune responses against cancer in humans and mice. J. Clin. Invest. 121, 3100–3108 (2011).

74. Weiner, H. L. & Cohen, J. A. Treatment of multiple sclerosis with cyclophosphamide: critical review of clinical and immunologic effects. Mult. Scler. 8, 142–154 (2002).

75. Medina-Echeverz, J. et al. Successful colon cancer eradication after chemoimmunotherapy is associated with profound phenotypic change of intratumoral myeloid cells. J. Immunol. 186, 807–815 (2011).

76. Ghiringhelli, F. et al. Metronomic cyclophosphamide regimen selectively depletes CD4+CD25+ regulatory T cells and restores T and NK effector functions in end stage cancer patients. Cancer Immunol. Immunother. 56, 641–648 (2007).

77. Taieb, J. et al. Chemoimmunotherapy of tumors: cyclophosphamide synergizes with exosome based vaccines. J. Immunol. 176, 2722–2729 (2006).

78. Viaud, S. et al. Cyclophosphamide induces differentiation of Th17 cells in cancer patients. Cancer Res. 71, 661–665 (2011).

79. Schiavoni, G. et al. Cyclophosphamide synergizes with type I interferons through systemic dendritic cell reactivation and induction of immunogenic tumor apoptosis. Cancer Res. 71, 768–778 (2011).

80. Guerriero, J. L. et al. DNA alkylating therapy induces tumor regression through an HMGB1-mediated activation of innate immunity. J. Immunol. 186, 3517–3526 (2011).

81. Hirschhorn-Cymerman, D. et al. OX40 engagement and chemotherapy combination provides potent antitumor immunity with concomitant regulatory T cell apoptosis. J. Exp. Med. 206, 1103–1116 (2009).

82. Diaz-Montero, C. M. et al. Increased circulating myeloid-derived suppressor cells correlate with clinical cancer stage, metastatic tumor burden, and doxorubicin-cyclophosphamide chemotherapy. Cancer Immunol. Immunother. 58, 49–59 (2009).

83. Ge, Y. et al. Metronomic cyclophosphamide treatment in metastasized breast cancer patients: immunological effects and clinical outcome. Cancer Immunol. Immunother. 14 Sep 2011(doi:10.1007/s00262-011-1106-3).

84. Ma, Y. et al. Contribution of IL-17-producing γδ T cells to the efficacy of anticancer chemotherapy. J. Exp. Med. 208, 491–503 (2011).

85. Mattarollo, S. R. et al. Pivotal role of innate and adaptive immunity in anthracycline chemotherapy of established tumors. Cancer Res. 71, 4809–4820 (2011).The studies in REF. 84 and REF. 85 unravel the cellular dynamics and molecular determinants underlying the immune infiltration of experimental breast adenocarcinomas and fibrosarcomas in response to anthracycline-based chemotherapy, highlighting a crucial early role for IL-17-producing γδ T cells.

86. Haggerty, T. J. et al. Topoisomerase inhibitors modulate expression of melanocytic antigens and enhance T cell recognition of tumor cells. Cancer Immunol. Immunother. 60, 133–144 (2011).

87. Tesniere, A. et al. Immunogenic death of colon cancer cells treated with oxaliplatin. Oncogene 29, 482–491 (2010).

88. Martins, I. et al. Restoration of the immunogenicity of cisplatin-induced cancer cell death by endoplasmic reticulum stress. Oncogene 30, 1147–1158 (2011).The release of ATP by dying cancer cells, which is required for immunogenic cell death, relies on the cellular machinery for autophagy, as demonstrated in human and murine genetic models of autophagy deficiency in vitro and in vivo.

89. Obeid, M. et al. Calreticulin exposure dictates the immunogenicity of cancer cell death. Nature Med. 13, 54–61 (2007).This was the first demonstration that apoptotic cell death can occur in an immunogenic fashion, provided that the endoplasmic reticulum chaperone protein calreticulin is exposed on the surface of dying cells.

90. Vitale, I., Galluzzi, L., Castedo, M. & Kroemer, G. Mitotic catastrophe: a mechanism for avoiding genomic instability. Nature Rev. Mol. Cell Biol. 12, 385–392 (2011).

91. Zhu, Y., Liu, N., Xiong, S. D., Zheng, Y. J. & Chu, Y. W. CD4+Foxp3+ regulatory T-cell impairment by paclitaxel is independent of toll-like receptor 4. Scand. J. Immunol. 73, 301–308 (2011).

92. Ramakrishnan, R. et al. Chemotherapy enhances tumor cell susceptibility to CTL-mediated killing during cancer immunotherapy in mice. J. Clin. Invest. 120, 1111–1124 (2010).

93. Kodumudi, K. N. et al. A novel chemoimmuno-modulating property of docetaxel: suppression of myeloid-derived suppressor cells in tumor bearers. Clin. Cancer Res. 16, 4583–4594 (2010).

94. Niiya, M. et al. Induction of TNF-α, uPA, IL-8 and MCP-1 by doxorubicin in human lung carcinoma cells. Cancer Chemother. Pharmacol. 52, 391–398 (2003).

95. Geller, M. A., Bui-Nguyen, T. M., Rogers, L. M. & Ramakrishnan, S. Chemotherapy induces macrophage chemoattractant protein-1 production in ovarian cancer. Int. J. Gynecol. Cancer 20, 918–925 (2010).

96. Qian, D. Z. et al. CCL2 is induced by chemotherapy and protects prostate cancer cells from docetaxel-induced cytotoxicity. Prostate 70, 433–442 (2010).

97. Fujimoto, H. et al. Stromal MCP-1 in mammary tumors induces tumor-associated macrophage infiltration and contributes to tumor progression. Int. J. Cancer 125, 1276–1284 (2009).

98. Kovarova, L. et al. Dendritic cell counts and their subsets during treatment of multiple myeloma. Neoplasma 54, 297–303 (2007).

99. Shurin, G. V., Tourkova, I. L., Kaneno, R. & Shurin, M. R. Chemotherapeutic agents in noncytotoxic concentrations increase antigen presentation by dendritic cells via an IL-12-dependent mechanism. J. Immunol. 183, 137–144 (2009).

100. Thomas-Schoemann, A. et al. Bystander effect of vinorelbine alters antitumor immune response. Int. J. Cancer 129, 1511–1518 (2011).

101. Purcell, W. T. & Ettinger, D. S. Novel antifolate drugs. Curr. Oncol. Rep. 5, 114–125 (2003).

102. Gibbs, D. & Jackman, A. Pemetrexed disodium. Nature Rev. Drug Discov. 5, S16–S17 (2005).

103. Cronstein, B. N. Low-dose methotrexate: a mainstay in the treatment of rheumatoid arthritis. Pharmacol. Rev. 57, 163–172 (2005).

104. Kaneno, R., Shurin, G. V., Tourkova, I. L. & Shurin, M. R. Chemomodulation of human dendritic cell function by antineoplastic agents in low noncytotoxic concentrations. J. Transl. Med. 7, 58 (2009).

105. Hartmann, J. T., Haap, M., Kopp, H. G. & Lipp, H. P. Tyrosine kinase inhibitors — a review on pharmacology, metabolism and side effects. Curr. Drug Metab. 10, 470–481 (2009).

106. Monsuez, J. J., Charniot, J. C., Vignat, N. & Artigou, J. Y. Cardiac side-effects of cancer chemotherapy. Int. J. Cardiol. 144, 3–15 (2010).

107. Chan, G. & Pilichowska, M. Complete remission in a patient with acute myelogenous leukemia treated with erlotinib for non small-cell lung cancer. Blood 110, 1079–1080 (2007).

108. Pitini, V., Arrigo, C. & Altavilla, G. Erlotinib in a patient with acute myelogenous leukemia and concomitant non-small-cell lung cancer. J. Clin. Oncol. 26, 3645–3646 (2008).In the studies published in reference 107 and reference 108, patients who were simultaneously affected by lung cancer and leukaemia were treated with erlotinib or gefitinib for the first condition and

experienced complete leukaemic remission; this demonstrated the existence of therapeutic off-target mechanisms ignited by EGFR inhibitors.

109. Boehrer, S. et al. Erlotinib exhibits antineoplastic off-target effects in AML and MDS: a preclinical study. Blood 111, 2170–2180 (2008).

110. Boehrer, S. et al. Erlotinib antagonizes constitutive activation of SRC family kinases and mTOR in acute myeloid leukemia. Cell Cycle 10, 3168–3175 (2011).

111. Stegmaier, K. et al. Gefitinib induces myeloid differentiation of acute myeloid leukemia. Blood 106, 2841–2848 (2005).

112. Pollack, B. P., Sapkota, B. & Cartee, T. V. Epidermal growth factor receptor inhibition augments the expression of MHC class I and II genes. Clin. Cancer Res. 17, 4400–4413 (2011).

113. Garrido, G. et al. Induction of immunogenic apoptosis by blockade of epidermal growth factor receptor activation with a specific antibody. J. Immunol. 187, 4954–4966 (2011).

114. Luo, Q. et al. Erlotinib inhibits T-cell-mediated immune response via down-regulation of the c-Raf/ERK cascade and Akt signaling pathway. Toxicol. Appl. Pharmacol. 251, 130–136 (2011).

115. Chen, C. I., Maecker, H. T. & Lee, P. P. Development and dynamics of robust T-cell responses to CML under imatinib treatment. Blood 111, 5342–5349 (2008).

116. Catellani, S., Pierri, I., Gobbi, M., Poggi, A. & Zocchi, M. R. Imatinib treatment induces CD5+ B lympho cytes and IgM natural antibodies with anti-leukemic reactivity in patients with chronic myelogenous leukemia. PLoS ONE 6, e18925 (2011).

117. Larmonier, N. et al. Imatinib mesylate inhibits CD4+CD25+ regulatory T cell activity and enhances active immunotherapy against BCR-ABL– tumors. J. Immunol. 181, 6955–6963 (2008).

118. Balachandran, V. P. et al. Imatinib potentiates antitumor T cell responses in gastrointestinal stromal tumor through the inhibition of Ido. Nature Med. 17, 1094–1100 (2011).This article demonstrates that the immuno stimulatory properties of imatinib (at least in part) result from a shift in the ratio between effector T cells and Treg cells; this ratio is secondary to the imatinib-mediated down regulation of indoleamine 2,3-dioxygenase in cancer cells.

119. Gao, H. et al. Imatinib mesylate suppresses cytokine synthesis by activated CD4 T cells of patients with chronic myelogenous leukemia. Leukemia 19, 1905–1911 (2005).

120. Seggewiss, R. et al. Imatinib inhibits T-cell receptor-mediated T-cell proliferation and activation in a dose-dependent manner. Blood 105, 2473–2479 (2005).

121. Leder, C., Ortler, S., Seggewiss, R., Einsele, H. & Wiendl, H. Modulation of T-effector function by imatinib at the level of cytokine secretion. Exp. Hematol. 35, 1266–1271 (2007).

122. Sinai, P. et al. Imatinib mesylate inhibits antigen-specific memory CD8 T cell responses in vivo. J. Immunol. 178, 2028–2037 (2007).

123. Adotevi, O. et al. A decrease of regulatory T cells correlates with overall survival after sunitinib-based antiangiogenic therapy in metastatic renal cancer patients. J. Immunother. 33, 991–998 (2010).

124. Ko, J. S. et al. Sunitinib mediates reversal of myeloid-derived suppressor cell accumulation in renal cell carcinoma patients. Clin. Cancer Res. 15, 2148–2157 (2009).

125. Finke, J. H. et al. Sunitinib reverses type-1 immune suppression and decreases T-regulatory cells in renal cell carcinoma patients. Clin. Cancer Res. 14, 6674–6682 (2008).

126. Desar, I. M. et al. Sorafenib reduces the percentage of tumour infiltrating regulatory T cells in renal cell carcinoma patients. Int. J. Cancer 129, 507–512 (2011).

127. Xin, H. et al. Sunitinib inhibition of Stat3 induces renal cell carcinoma tumor cell apoptosis and reduces immunosuppressive cells. Cancer Res. 69, 2506–2513 (2009).

128. Dalton, J. E. et al. Inhibition of receptor tyrosine kinases restores immunocompetence and improves immune-dependent chemotherapy against experimental leishmaniasis in mice. J. Clin. Invest. 120, 1204–1216 (2010).

129. Hipp, M. M. et al. Sorafenib, but not sunitinib, affects function of dendritic cells and induction of primary immune responses. Blood 111, 5610–5620 (2008).

130. Fei, F. et al. Dasatinib inhibits the proliferation and function of CD4+CD25+ regulatory T cells. Br. J. Haematol. 144, 195–205 (2009).

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 231

© 2012 Macmillan Publishers Limited. All rights reserved

131. Fei, F. et al. Dasatinib exerts an immunosuppressive effect on CD8+ T cells specific for viral and leukemia antigens. Exp. Hematol. 36, 1297–1308 (2008).

132. Weichsel, R. et al. Profound inhibition of antigen-specific T-cell effector functions by dasatinib. Clin. Cancer Res. 14, 2484–2491 (2008).

133. Schade, A. E. et al. Dasatinib, a small-molecule protein tyrosine kinase inhibitor, inhibits T-cell activation and proliferation. Blood 111, 1366–1377 (2008).

134. Fraser, C. K. et al. Dasatinib inhibits recombinant viral antigen-specific murine CD4+ and CD8+ T-cell responses and NK-cell cytolytic activity in vitro and in vivo. Exp. Hematol. 37, 256–265 (2009).

135. Fraser, C. K. et al. Dasatinib inhibits the secretion of TNF-α following TLR stimulation in vitro and in vivo. Exp. Hematol. 37, 1435–1444 (2009).

136. Blake, S., Hughes, T. P., Mayrhofer, G. & Lyons, A. B. The Src/ABL kinase inhibitor dasatinib (BMS-354825) inhibits function of normal human T-lymphocytes in vitro. Clin. Immunol. 127, 330–339 (2008).

137. Chen, J. et al. Nilotinib hampers the proliferation and function of CD8+ T lymphocytes through inhibition of T cell receptor signalling. J. Cell. Mol. Med. 12, 2107–2118 (2008).

138. Salih, J. et al. The BCR/ABL-inhibitors imatinib, nilotinib and dasatinib differentially affect NK cell reactivity. Int. J. Cancer 127, 2119–2128 (2010).

139. Nencioni, A. et al. Proteasome inhibitor bortezomib modulates TLR4-induced dendritic cell activation. Blood 108, 551–558 (2006).

140. Berges, C. et al. Proteasome inhibition suppresses essential immune functions of human CD4+ T cells. Immunology 124, 234–246 (2008).

141. Basler, M., Lauer, C., Beck, U. & Groettrup, M. The proteasome inhibitor bortezomib enhances the susceptibility to viral infection. J. Immunol. 183, 6145–6150 (2009).

142. Heider, U. et al. Decrease in CD4+ T-cell counts in patients with multiple myeloma treated with bortezomib. Clin. Lymphoma Myeloma Leuk. 10, 134–137 (2010).

143. Blanco, B. et al. Treatment with bortezomib of human CD4+ T cells preserves natural regulatory T cells and allows the emergence of a distinct suppressor T-cell population. Haematologica 94, 975–983 (2009).

144. Liu, P., Cheng, H., Roberts, T. M. & Zhao, J. J. Targeting the phosphoinositide 3-kinase pathway in cancer. Nature Rev. Drug Discov. 8, 627–644 (2009).

145. Koyasu, S. The role of PI3K in immune cells. Nature Immunol. 4, 313–319 (2003).

146. Carrington, E. M. et al. BH3 mimetics antagonizing restricted prosurvival Bcl-2 proteins represent another class of selective immune modulatory drugs. Proc. Natl Acad. Sci. USA 107, 10967–10971 (2010).

147. Weiner, L. M., Surana, R. & Wang, S. Monoclonal antibodies: versatile platforms for cancer immuno-therapy. Nature Rev. Immunol. 10, 317–327 (2010).This article summarizes recent advances in the development and use of immunostimulatory monoclonal antibodies for anticancer therapy.

148. Winiarska, M., Glodkowska-Mrowka, E., Bil, J. & Golab, J. Molecular mechanisms of the antitumor effects of anti-CD20 antibodies. Front. Biosci. 16, 277–306 (2011).

149. Zipfel, P. F. & Skerka, C. Complement regulators and inhibitory proteins. Nature Rev. Immunol. 9, 729–740 (2009).

150. Marechal, R. et al. Putative contribution of CD56 positive cells in cetuximab treatment efficacy in first-line metastatic colorectal cancer patients. BMC Cancer 10, 340 (2010).

151. Banerjee, D. et al. Enhanced T-cell responses to glioma cells coated with the anti-EGF receptor antibody and targeted to activating FcγRs on human dendritic cells. J. Immunother. 31, 113–120 (2008).

152. zum Büschenfelde, C. M., Hermann, C., Schmidt, B., Peschel, C. & Bernhard, H. Antihuman epidermal growth factor receptor 2 (HER2) monoclonal antibody trastuzumab enhances cytolytic activity of class I-restricted HER2-specific T lymphocytes against HER2-overexpressing tumor cells. Cancer Res. 62, 2244–2247 (2002).

153. Arnould, L. et al. Trastuzumab-based treatment of HER2-positive breast cancer: an antibody-dependent cellular cytotoxicity mechanism? Br. J. Cancer 94, 259–267 (2006).

154. Wada, J. et al. The contribution of vascular endothelial growth factor to the induction of regulatory T-cells in malignant effusions. Anticancer Res. 29, 881–888 (2009).

155. Manzoni, M. et al. Immunological effects of bevacizumab-based treatment in metastatic colorectal cancer. Oncology 79, 187–196 (2010).

156. Osada, T. et al. The effect of anti-VEGF therapy on immature myeloid cell and dendritic cells in cancer patients. Cancer Immunol. Immunother. 57, 1115–1124 (2008).

157. Shrimali, R. K. et al. Antiangiogenic agents can increase lymphocyte infiltration into tumor and enhance the effectiveness of adoptive immunotherapy of cancer. Cancer Res. 70, 6171–6180 (2010).

158. Pander, J. et al. Activation of tumor-promoting type 2 macrophages by EGFR-targeting antibody cetuximab. Clin. Cancer Res. 17, 5668–5673 (2011).

159. Hodi, F. S. et al. Improved survival with ipilimumab in patients with metastatic melanoma. N. Engl. J. Med. 363, 711–723 (2010).

160. Kline, J. & Gajewski, T. F. Clinical development of mAbs to block the PD1 pathway as an immuno-therapy for cancer. Curr. Opin. Investig. Drugs 11, 1354–1359 (2010).

161. Fourcade, J. et al. Upregulation of Tim-3 and PD-1 expression is associated with tumor antigen-specific CD8+ T cell dysfunction in melanoma patients. J. Exp. Med. 207, 2175–2186 (2010).

162. Bennett, S. R. et al. Help for cytotoxic-T-cell responses is mediated by CD40 signalling. Nature 393, 478–480 (1998).

163. Sakuishi, K. et al. Targeting Tim-3 and PD-1 pathways to reverse T cell exhaustion and restore anti-tumor immunity. J. Exp. Med. 207, 2187–2194 (2010).

164. Wang, W. et al. PD1 blockade reverses the suppression of melanoma antigen-specific CTL by CD4+CD25Hi regulatory T cells. Int. Immunol. 21, 1065–1077 (2009).

165. Yuan, J. et al. CTLA-4 blockade enhances polyfunctional NY-ESO-1 specific T cell responses in metastatic melanoma patients with clinical benefit. Proc. Natl Acad. Sci. USA 105, 20410–20415 (2008).

166. Beatty, G. L. et al. CD40 agonists alter tumor stroma and show efficacy against pancreatic carcinoma in mice and humans. Science 331, 1612–1616 (2011).

167. Jiang, Q. et al. mTOR kinase inhibitor AZD8055 enhances the immunotherapeutic activity of an agonist CD40 antibody in cancer treatment. Cancer Res. 71, 4074–4084 (2011).

168. Robert, C. et al. Ipilimumab plus dacarbazine for previously untreated metastatic melanoma. N. Engl. J. Med. 364, 2517–2526 (2011).

169. Norton, J. T., Hayashi, T., Crain, B., Corr, M. & Carson, D. A. Role of IL-1 receptor-associated kinase-M (IRAK-M) in priming of immune and inflammatory responses by nitrogen bisphosphonates. Proc. Natl Acad. Sci. USA 108, 11163–11168 (2011).

170. Rack, B. et al. Effect of zoledronate on persisting isolated tumour cells in patients with early breast cancer. Anticancer Res. 30, 1807–1813 (2010).

171. Cabillic, F. et al. Aminobisphosphonate-pretreated dendritic cells trigger successful Vγ9Vδ2 T cell amplification for immunotherapy in advanced cancer patients. Cancer Immunol. Immunother. 59, 1611–1619 (2010).

172. Gyrd-Hansen, M. & Meier, P. IAPs: from caspase inhibitors to modulators of NF-κB, inflammation and cancer. Nature Rev. Cancer 10, 561–574 (2010).

173. Dougan, M. et al. IAP inhibitors enhance co-stimulation to promote tumor immunity. J. Exp. Med. 207, 2195–2206 (2010).

174. Schmudde, M. et al. Histone deacetylase inhibitors sensitize tumour cells for cytotoxic effects of natural killer cells. Cancer Lett. 272, 110–121 (2008).

175. Lesterhuis, W. J., Haanen, J. B. & Punt, C. J. Cancer immunotherapy — revisited. Nature Rev. Drug Discov. 10, 591–600 (2011).

176. Emens, L. A. et al. Timed sequential treatment with cyclophosphamide, doxorubicin, and an allogeneic granulocyte–macrophage colony-stimulating factor-secreting breast tumor vaccine: a chemotherapy dose-ranging factorial study of safety and immune activation. J. Clin. Oncol. 27, 5911–5918 (2009).

177. Arlen, P. M. et al. A randomized Phase II study of concurrent docetaxel plus vaccine versus vaccine alone in metastatic androgen-independent prostate cancer. Clin. Cancer Res. 12, 1260–1269 (2006).

178. Garnett, C. T., Schlom, J. & Hodge, J. W. Combination of docetaxel and recombinant vaccine enhances T-cell responses and antitumor activity: effects of docetaxel on immune enhancement. Clin. Cancer Res. 14, 3536–3544 (2008).

179. Lesterhuis, W. J. et al. A pilot study on the immunogenicity of dendritic cell vaccination during adjuvant oxaliplatin/capecitabine chemotherapy in colon cancer patients. Br. J. Cancer 103, 1415–1421 (2010).

180. Narita, M. et al. WT1 peptide vaccination in combination with imatinib therapy for a patient with CML in the chronic phase. Int. J. Med. Sci. 7, 72–81 (2010).

181. Gonzalez-Aparicio, M. et al. Oxaliplatin in combination with liver-specific expression of interleukin 12 reduces the immunosuppressive microenvironment of tumours and eradicates metastatic colorectal cancer in mice. Gut 60, 341–349 (2011).

182. Zappasodi, R. et al. Improved clinical outcome in indolent B-cell lymphoma patients vaccinated with autologous tumor cells experiencing immunogenic death. Cancer Res. 70, 9062–9072 (2010).This article provides proof of principle that the propensity of tumour cells to undergo immunogenic cell death in vitro can influence the therapeutic outcome of DC-based anticancer vaccines.

183. Rosenberg, S. A., Restifo, N. P., Yang, J. C., Morgan, R. A. & Dudley, M. E. Adoptive cell transfer: a clinical path to effective cancer immunotherapy. Nature Rev. Cancer 8, 299–308 (2008).

184. Nistico, P. et al. Chemotherapy enhances vaccine-induced antitumor immunity in melanoma patients. Int. J. Cancer 124, 130–139 (2009).

185. Palermo, B. et al. Dacarbazine treatment before peptide vaccination enlarges T-cell repertoire diversity of melan-A-specific, tumor-reactive CTL in melanoma patients. Cancer Res. 70, 7084–7092 (2010).

186. Kyte, J. A. et al. Telomerase peptide vaccination combined with temozolomide: a clinical trial in stage IV melanoma patients. Clin. Cancer Res. 17, 4568–4580 (2011).

187. Rettig, L. et al. Gemcitabine depletes regulatory T-cells in human and mice and enhances triggering of vaccine-specific cytotoxic T-cells. Int. J. Cancer 129, 832–838 (2011).

188. Adams, S. et al. Immunization of malignant melanoma patients with full-length NY-ESO-1 protein using TLR7 agonist imiquimod as vaccine adjuvant. J. Immunol. 181, 776–784 (2008).

189. Davis, I. D. et al. Blood dendritic cells generated with Flt3 ligand and CD40 ligand prime CD8+ T cells efficiently in cancer patients. J. Immunother. 29, 499–511 (2006).

190. Robbins, P. F. et al. Tumor regression in patients with metastatic synovial cell sarcoma and melanoma using genetically engineered lymphocytes reactive with NY-ESO-1. J. Clin. Oncol. 29, 917–924 (2011).

191. Muraoka, D. et al. Peptide vaccine induces enhanced tumor growth associated with apoptosis induction in CD8+ T cells. J. Immunol. 185, 3768–3776 (2010).

192. Bourquin, C., Schreiber, S., Beck, S., Hartmann, G. & Endres, S. Immunotherapy with dendritic cells and CpG oligonucleotides can be combined with chemotherapy without loss of efficacy in a mouse model of colon cancer. Int. J. Cancer 118, 2790–2795 (2006).

193. Brody, J. D. et al. In situ vaccination with a TLR9 agonist induces systemic lymphoma regression: a Phase I/II study. J. Clin. Oncol. 28, 4324–4332 (2010).

194. Zoglmeier, C. et al. CpG blocks immunosuppression by myeloid-derived suppressor cells in tumor-bearing mice. Clin. Cancer Res. 17, 1765–1775 (2011).

195. Wang, Y., Wang, X. Y., Subjeck, J. R., Shrikant, P. A. & Kim, H. L. Temsirolimus, an mTOR inhibitor, enhances anti-tumour effects of heat shock protein cancer vaccines. Br. J. Cancer 104, 643–652 (2011).

196. Yang, X. F. Factors regulating apoptosis and homeostasis of CD4+CD25highFOXP3+ regulatory T cells are new therapeutic targets. Front. Biosci. 13, 1472–1499 (2008).

197. Yu, H., Kortylewski, M. & Pardoll, D. Crosstalk between cancer and immune cells: role of STAT3 in the tumour microenvironment. Nature Rev. Immunol. 7, 41–51 (2007).

198. Schust, J., Sperl, B., Hollis, A., Mayer, T. U. & Berg, T. Stattic: a small-molecule inhibitor of STAT3 activation and dimerization. Chem. Biol. 13, 1235–1242 (2006).

199. Kraman, M. et al. Suppression of antitumor immunity by stromal cells expressing fibroblast activation protein-α. Science 330, 827–830 (2010).

R E V I E W S

232 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

CORRIGENDUM

2011 in reflectionAsher MullardNature Reviews Drug Discovery 11, 6–8 (2012) | doi:10.1038/nrd3643

DaclatasvirwasoriginallymisidentifiedasanNS5Aproteaseinhibitor,whenitisanNS5Areplicationcomplexinhibitor.Thishasbeencorrectedintheonlineversionofthearticle.

ERRATUM

Interfacial inhibitors: targeting macromolecular complexesYves Pommier and Christophe MarchandNature Reviews Drug Discovery 11, 25–36 (2012) | doi:10.1038/nrd3404

InTable1,theinformationlistedinthe‘Substrates’columnfor‘Etoposideandteniposide’and‘Mitoxantrone’isincorrect;‘TOP1–DNAcomplex’shouldbe‘TOP2–DNAcomplex’inbothinstances.Thishasbeencorrectedintheonlineversionofthearticle.

200. Shields, J. D., Kourtis, I. C., Tomei, A. A., Roberts, J. M. & Swartz, M. A. Induction of lymphoidlike stroma and immune escape by tumors that express the chemokine CCL21. Science 328, 749–752 (2010).

201. Schietinger, A., Philip, M., Liu, R. B., Schreiber, K. & Schreiber, H. Bystander killing of cancer requires the cooperation of CD4+ and CD8+ T cells during the effector phase. J. Exp. Med. 207, 2469–2477 (2010).

202. Zitvogel, L., Kepp, O. & Kroemer, G. Decoding cell death signals in inflammation and immunity. Cell 140, 798–804 (2010).

203. Kepp, O. et al. Molecular determinants of immunogenic cell death elicited by anticancer chemotherapy. Cancer Metastasis Rev. 30, 61–69 (2011).

204. Rakhra, K. et al. CD4+ T cells contribute to the remodeling of the microenvironment required for sustained tumor regression upon oncogene inactivation. Cancer Cell 18, 485–498 (2010).

205. Chakraborty, M. et al. The use of chelated radionuclide (samarium-153-ethyl enediaminetetra-methylenephosphonate) to modulate phenotype of tumor cells and enhance T cell-mediated killing. Clin. Cancer Res. 14, 4241–4249 (2008).

206. Lynch, T. et al. Phase II trial of ipilimumab (IPI) and paclitaxel/carboplatin (P/C) in first-line stage IIIb/IV non-small cell lung cancer (NSCLC). J. Clin. Oncol. (Meeting Abstracts) 28, 7531 (2010).

207. Gabrilovich, D. I. Combination of chemotherapy and immunotherapy for cancer: a paradigm revisited. Lancet Oncol. 8, 2–3 (2007).

208. Scripture, C. D. & Figg, W. D. Drug interactions in cancer therapy. Nature Rev. Cancer 6, 546–558 (2006).

209. Marx, J. How the glucocorticoids suppress immunity. Science 270, 232–233 (1995).

210. Kubecova, M., Kolostova, K., Pinterova, D., Kacprzak, G. & Bobek, V. Cimetidine: an anticancer drug? Eur. J. Pharm. Sci. 42, 439–444 (2011).

211. Kroemer, G. & Pouyssegur, J. Tumor cell metabolism: cancer’s Achilles’ heel. Cancer Cell 13, 472–482 (2008).

212. Schreiber, R. D., Old, L. J. & Smyth, M. J. Cancer immunoediting: integrating immunity’s roles in cancer suppression and promotion. Science 331, 1565–1570 (2011).

213. Fukasawa, K. Oncogenes and tumour suppressors take on centrosomes. Nature Rev. Cancer 7, 911–924 (2007).

214. Vakkila, J. & Lotze, M. T. Inflammation and necrosis promote tumour growth. Nature Rev. Immunol. 4, 641–648 (2004).

215. Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646–674 (2011).

216. Morselli, E. et al. Oncosuppressive functions of autophagy. Antioxid. Redox Signal. 14, 2251–2269 (2011).

217. Mathew, R., Karantza-Wadsworth, V. & White, E. Role of autophagy in cancer. Nature Rev. Cancer 7, 961–967 (2007).

218. Mueller, M. M. & Fusenig, N. E. Friends or foes — bipolar effects of the tumour stroma in cancer. Nature Rev. Cancer 4, 839–849 (2004).

219. Green, D. R., Ferguson, T., Zitvogel, L. & Kroemer, G. Immunogenic and tolerogenic cell death. Nature Rev. Immunol. 9, 353–363 (2009).

220. Kepp, O., Tesniere, A., Zitvogel, L. & Kroemer, G. The immunogenicity of tumor cell death. Curr. Opin. Oncol. 21, 71–76 (2009).

221. Galluzzi, L. et al. Molecular definitions of cell death subroutines: recommendations of the Nomenclature Committee on Cell Death 2012. Cell Death Differ. 15 Jul 2011 (doi:10.1038/cdd.2011.96).This review contains up-to-date recommendations for the functional classification of cell death subroutines, as formulated by the Nomenclature Committee on Cell Death in 2012.

222. Kroemer, G. et al. Classification of cell death: recommendations of the Nomenclature Committee on Cell Death 2009. Cell Death Differ. 16, 3–11 (2009).This review contains recommendations for the morphological classification of cell death subroutines, as formulated by the Nomenclature Committee on Cell Death in 2009.

223. Apetoh, L. et al. Toll-like receptor 4-dependent contribution of the immune system to anticancer chemotherapy and radiotherapy. Nature Med. 13, 1050–1059 (2007).

224. Ghiringhelli, F. et al. Activation of the NLRP3 inflammasome in dendritic cells induces IL-1β-dependent adaptive immunity against tumors. Nature Med. 15, 1170–1178 (2009).

225. Panaretakis, T. et al. Mechanisms of pre-apoptotic calreticulin exposure in immunogenic cell death. EMBO J. 28, 578–590 (2009).

226. Martins, I. et al. Chemotherapy induces ATP release from tumor cells. Cell Cycle 8, 3723–3728 (2009).

227. Michaud, M. et al. Autophagy-dependent anticancer immune responses induced by chemotherapeutic agents. Science (in the press).

228. Chekeni, F. B. et al. Pannexin 1 channels mediate ‘find-me’ signal release and membrane permeability during apoptosis. Nature 467, 863–867 (2010).

229. Zhang, J. et al. VEGF blockade inhibits lymphocyte recruitment and ameliorates immune-mediated vascular remodeling. Circ. Res. 107, 408–417 (2010).

AcknowledgementsThe authors are supported by grants from the Ligue contre le Cancer (the French National League against Cancer), the AXA Chair for Longevity Research, Cancéropôle Ile-de-France, Institut National du Cancer (the French National Cancer Institute), the Bettencourt-Schueller Foundation, The Fondation de France, the Fondation pour la Recherche Médicale (the Foundation for Medical Research), the Agence National de la Recherche (the National Agency for Research) and the European Commission (APO-SYS, ArtForce, ChemoRes. Death-Train) as well as the LabEx Immuno-Oncology.

Competing interests statementThe authors declare no competing financial interests.

FURTHER INFORMATIONGuido Kroemer’s homepage: http://www.kroemerlab.com

ALL LINKS ARE ACTIVE IN THE ONLINE PDF

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 233

© 2012 Macmillan Publishers Limited. All rights reserved

The major functions of the skeleton include: provision of structural support for the body, protection of inner organs and accommodation of the haematopoietic sys-tem. Bone, the unique skeletal component, undergoes continuous modelling and remodelling — a process brought about by two major cell types: the bone-forming osteoblasts and the bone-degrading osteoclasts. The development and activation of these cell types are tightly regulated processes, and a complex network of signalling pathways are involved in mediating their effects1–4.

Bone remodelling starts with the induction of a resorption lacuna by activated osteoclasts (FIG. 1), which is followed by the activation of osteoblasts that fill the lacuna with new bone matrix. However, the net balance in the remodelling process can be disrupted by factors that interfere with osteoblast function or osteoclast activity. Once osteoclasts reach such a level of hyperactivity, or osteoblasts are so insufficiently active that the resorp-tion lacunae cannot be filled by new matrix, a net loss of bone tissue ensues — known as osteopaenia — which ultimately leads to osteoporosis and increased bone fragility4,5. This imbalance can be of metabolic origin (such as osteomalacia) or endocrine origin (such as hyperparathyroidism or postmenopausal osteoporosis)6,7, the latter affecting more than 50% of women over 60 years of age. Oestrogen deficiency is the most common cause of osteoclast hyperactivity that leads to high fracture risk, and is associated with increased mortality5,8,9.

The process of remodelling is necessary not only for adapting bony strength to growth and changes in mechan-ical load but also for repairing mechanical damage. Bone remodelling occurs continuously throughout life and

involves cortical bone as well as the inner bony web — the trabecular bone4. It is only during childhood, when bone grows and is modelled10, that bone formation can occur independently of bone resorption. Bone also constitutes the body’s calcium depot and therefore bone remodelling allows for maintenance of blood calcium levels. When calcium is needed — for instance, during lactation, with decreased physical exercise or with reduced mobility (for example, in elderly individuals) — bone resorption will prevail, ultimately leading to osteoporosis11–13. Conversely, as a consequence of intensive exercise with its repetitive mechanical effects, bone formation exceeds bone resorp-tion and is accompanied by an increase in bone mass5,13,14.

An imbalance between bone formation and bone resorption is also linked to various diseases. For example, many chronic inflammatory conditions are associated with systemic osteoporosis and increased fracture rates. Indeed, inflammation has been shown to lead to excessive bone resorption as well as impaired bone formation15,16.

Inflammation is characterized by the activation of several cell populations of the innate and adaptive immune system that produce inflammatory cytokines. These messenger molecules not only perpetuate inflam-mation but in turn also activate bone degradation and inhibit bone-building mechanisms. Indeed, it has been shown that the degree of the inflammatory response is linked to the extent of local and systemic bone loss15,17. Although therapeutic intervention can effectively reduce inflammation, it frequently does not eliminate it and so some bone loss continues. Indeed, synthetic disease-modifying drugs, such as methotrexate in rheumatoid arthritis or azathioprine in inflammatory bowel disease,

Division of Rheumatology, Department of Medicine 3, Medical University of Vienna, Währinger Gürtel 18‑20, A‑1090 Vienna, Austria.e‑mails: kurt.redlich@meduniwien.ac.at; josef.smolen@wienkav.at; josef.smolen@ meduniwien.ac.atdoi:10.1038/nrd3669

Inflammatory bone loss: pathogenesis and therapeutic interventionKurt Redlich and Josef S. Smolen

Abstract | Bone is a tissue undergoing continuous building and degradation. This remodelling is a tightly regulated process that can be disturbed by many factors, particularly hormonal changes. Chronic inflammation can also perturb bone metabolism and promote increased bone loss. Inflammatory diseases can arise all over the body, including in the musculoskeletal system (for example, rheumatoid arthritis), the intestine (for example, inflammatory bowel disease), the oral cavity (for example, periodontitis) and the lung (for example, cystic fibrosis). Wherever inflammatory diseases occur, systemic effects on bone will ensue, as well as increased fracture risk. Here, we discuss the cellular and signalling pathways underlying, and strategies for therapeutically interfering with, the inflammatory loss of bone.

R E V I E W S

234 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Osteoclast

Osteoblast precursor

Resorptionpit

Filled resorption pit

Endothelial cell,blood vessel

Mesenchymalstem cell

Osteoclast precursor

Osteoblasts

Osteocytes

?

Bone resorption Bone formation

Haematopoieticstem cell

Nature Reviews | Drug Discovery

and even targeted biological therapies accomplish a state of persistent remission in only a small proportion of patients. Consequently, inflammatory osteoporosis represents a significant unmet and often insufficiently recognized medical need.

Therefore, in this Review we provide an overview of the cells and signalling pathways involved in the control of bone remodelling, and present the mechanisms mediating inflammatory bone loss. We also discuss the therapeutic potential and limitations of agents that may be used to treat osteoporosis in chronic inflammatory diseases, as well as novel approaches that are currently under investigation.

Bone physiologyBone formation and resorption are complex processes involving various cell types and signalling pathways. These are considered in detail below.

Osteoblasts. Osteoblasts develop from mesodermal pro-genitor cells and, when fully differentiated, produce bone matrix (FIG. 1). For the evolution of this highly specialized cell lineage, mesenchymal progenitor cells need to express runt-related transcription factor 2 (RUNX2). Activation of RUNX2 allows mesenchymal progenitor cells to proliferate and differentiate into preosteoblasts and subsequently into mature osteoblasts18. The activation of this master regula-tor of osteoblastogenesis is induced via the SMAD proteins SMAD1, SMAD5 and SMAD8. SMADs are phosphoryl-ated primarily after the interaction of bone morphogenetic proteins (BMPs) with their receptors, and are essential not only for activation of RUNX2 but also for other

osteoblast-typical genes19 (FIG. 2a). Furthermore, other growth factors — such as transforming growth factor-β and fibroblast growth factor — as well as hormones, like parathyroid hormone (PTH), also affect RUNX2 and osteoblastogenesis20,21.

Importantly, RUNX2 in turn interacts with several binding partners and co-modulators. Indeed, post-translational events appear to partly regulate RUNX2 expression. RUNX2 can be phosphorylated and activated by mitogen-activated protein kinases (MAPKs), which in turn are induced by fibroblast growth factor and PTH as well as BMPs. Moreover, RUNX2 expression is enhanced by protein–protein interactions between AP1 (induced by PTH) and SMADs (induced by BMPs); it is also enhanced by protein–protein interactions between phosphorylated RUNX2 and other nuclear factors. Overall, RUNX2 appears to orchestrate multiple osteoblastogenic sig-nals and thus also amplifies its own gene activation and transcriptional activity22–24.

Another pivotal system in the differentiation and activation of osteoblasts is the WNT–Frizzled–β-catenin signalling pathway2,25. WNT proteins constitute a large number of evolutionarily highly conserved glycoproteins. WNTs are secreted by various cells, especially in regions where there is a mix of different cell populations, includ-ing bone marrow stromal cells and haematopoietic stem cells26,27. They bind to a receptor complex composed of the Frizzled receptor and associated low-density lipoprotein receptor-related proteins (LRPs; for example, LRP4, LRP5 and LRP6)28, and Frizzled then signals via β-catenin.

The WNT pathway can be inhibited by various mol-ecules. One of them is Dickkopf-related protein 1 (DKK1), which is secreted in some normal tissues such as the spleen and skin29,30 but readily secreted by cytokine-activated or malignant cells31,32. DKK1 binds to LRP4, LRP5 or LRP6 by utilizing a co-receptor, Kremen 1 (also known as the Dickkopf receptor) or Kremen 2, to form a ternary com-plex that is endocytosed, thus eliminating LRP from the cell surface and preventing it from binding to WNT33. Another WNT signalling inhibitor is sclerostin, which is a mainly osteocyte-derived cytokine. Like DKK1, sclerostin binds to LRP4, LRP5 or LRP6 but does not require the co-receptor Kremen for this interaction; however, it is possible that another as yet unknown co-receptor may be needed for this activity34. These inhibitors of the WNT signalling pathway constrain osteoblast function. Following engage-ment of its receptor, PTH also uses the β-catenin pathway for osteoblast activation35.

Osterix is another essential molecule that is involved in osteoblast differentiation. Osterix-deficient animals do not form bone despite having normal levels of RUNX2 (REF. 36), indicating that osterix is activated downstream of RUNX2. Osterix is regulated primarily by BMP2 and acts via nuclear factor of activated T cells, cytoplasmic 1 (NFATC1; also known as NFAT2)36,37.

Although in many cells nuclear factor-κB (NF-κB) pathways are centrally involved in cell development and function, the NF-κB pathway appears to have no role in osteoblast differentiation; rather, NF-κB activation sup-presses osteoblast activities38,39, and this is discussed in more detail below in the context of inflammatory bone loss.

Figure 1 | Maintenance of bone structure. Within the bony tissue, osteoblasts and osteoclasts perform their activities to build, degrade and thus constantly remodel bone. Osteoclasts begin degrading bone, and the resorption pits are partly filled by new bone matrix produced by osteoblasts, which is subsequently mineralized. In addition, the scheme shows osteocytes within bone, interacting with each other as well as with osteoclasts and osteoblasts via their dendrite-like extensions, which reside within tiny canaliculi spreading throughout the bone.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 235

© 2012 Macmillan Publishers Limited. All rights reserved

TGFβ

TGFβRIGF FGFR

BMPR

Frizzled

LRP5/LRP6

PTHR

PTHWNT

BMPFGF

OSM

OSMR

JAK

STAT

PKAβ-catenin

MAPKKs

JNKp38 RUNX2

STAT

STAT STAT

PKC

AP1

a

Gene transcription

Nucleus CytoplasmRUNX2, OC, PTHR, SOST, OSX, RANKL, OPG

ERK

gp130

Cell membrane

SMADs

JAK

LRP5/LRP6

TGFβRIGF FGFR

BMPRTNFR

TNF

Frizzled PTHR

WNT

IL-6

gp130IL6R

DKK1

SOST

SMURF1/ SMURF2

b

Gene transcription

NucleusCytoplasm

RUNX2, OC, PTHR, OSX, OPG SOST, RANKL

Nature Reviews | Drug Discovery

TNF, IL-1, IL-6

STAT STAT

STAT STAT

MAPKKs

p38 JNK ERK

AP1

JAK JAK

Figure 2 | Signalling in osteoblasts in health and during inflammation. a | Under healthy conditions, various receptors and downstream signalling pathways can be activated in osteoblasts. The most common ligands, receptors and signal transduction moieties are depicted. Essential pathways involve: bone morphogenetic proteins (BMPs) and their receptors, which utilize SMAD proteins for direct or transcriptional activation of runt-related transcription factor 2 (RUNX2) and subsequent downstream cellular events; the WNT–Frizzled pathway, which uses β-catenin for further activities; as well as parathyroid hormone (PTH) and its receptor signalling mediated by protein kinase A (PKA). Additional pathways are also shown. b | Under inflammatory conditions the release of pro-inflammatory cytokines such as tumour necrosis factor (TNF) and interleukin-6 (IL-6) leads to the inhibition of osteoblasts. This occurs partly via inhibition of mitogen-activated protein kinase (MAPK) activities by activated signal transducers and activators of transcription (STATs), partly via activation of nuclear factor-κB (not depicted) and partly via the effects of SMAD ubiquitylation regulatory factor 1 (SMURF1) and SMURF2. The pro-inflammatory cytokines also upregulate Dickkopf-related protein 1 (DKK1) and sclerostin (SOST), which inhibit the WNT–Frizzled pathway, whereas many other osteoblast gene products are downregulated. AP1, activator protein 1; BMPR, BMP receptor; ERK, extracellular signal-regulated kinase; FGF, fibroblast growth factor; FGFR, FGF receptor; gp130, glycoprotein 130; IGF, insulin-like growth factor; IL-6R, IL-6 receptor; JAK, Janus kinase; JNK, JUN N-terminal kinase; LRP5, low-density lipoprotein receptor-related protein 5; MAPKK, MAPK kinase; OC, osteocalcin; OPG, osteoprotegerin; OSM, oncostatin M; OSMR, OSM receptor; OSX, osterix; p38, p38 MAPK; PTHR, PTH receptor; RANKL, receptor activator of NF-κB ligand; TGFβ, transforming growth factor-β; TGFβR, TGFβ receptor; TNFR, TNF receptor.

R E V I E W S

236 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Receptor activator of NF-κB ligand(RANKL). The ligand for receptor activator of NF-κB (RANK). RANKL is expressed on osteoblasts and other mesenchymal cells, as well as on non-mesenchymal cells such as T- and B lymphocytes. It is also secreted as a soluble molecule. Furthermore, mesenchymal cells can produce osteoprotegerin, which is a decoy receptor of RANKL that prevents the binding of RANKL to RANK. A monoclonal antibody against RANKL, denosumab, is licensed for the treatment of osteoporosis.

Glycoprotein 130 (gp130). A transmembrane protein forming the signalling subunit of the interleukin-6 (IL-6) receptor family. When IL-6 or other IL-6 family members bind to their specific receptors (α-chain), they can interact with the gp130 protein (β-chain). Subsequently, this triplet dimerizes to form a hexamer. The gp130 co-receptor activates signal transduction via Janus kinases.

Sclerosteosis A rare, hereditary bone disease in which bones grow abnormally to produce a high bone mass phenotype; it is due to mutations in the gene encoding sclerostin, which is produced primarily by osteocytes and inhibits osteoblast function. Receptor activator of NF-κB (RANK). A transmembrane protein belonging to the tumour necrosis factor receptor family and expressed on the surface of cells of haemato poietic origin. RANK is the pivotal cytokine receptor for osteoclastogenesis and leads to the activation of the transcription factor activator protein 1, which involves mainly heterodimers of the JUN and FOS families of proteins.

The expression of osteocalcin, the PTH receptor, sclerostin and receptor activator of NF-κB ligand (RANKL; also known as TNFSF11), among other genes, is char-acteristic of the osteoblast lineage1,40; RANKL is a pivotal cytokine that leads to osteoclast differentiation. Osteoblasts can also produce osteoprotegerin (also known as TNFRSF11B), which is a decoy receptor molecule that naturally binds to RANKL41 to inhibit osteoclast activation and thus protect against bone loss.

Ultimately, osteoblast activation results in the pro-duction of matrix proteins (such as collagen 1) and regu-lators of matrix mineralization (such as osteopontin and osteonectin)42.

Some osteoblasts develop into lining cells that cover bone surfaces4,43,44, but most develop into osteocytes (FIG. 1). Osteocytes are terminally differentiated osteo-blast lineage cells that constitute the most abundant bone cell population and have long been regarded as being inert. However, although they are trapped within the matrix in lacunae, they form a widespread net-work throughout bone, allowing for mechanosensing and mechanotransduction45. Their importance is fur-ther revealed by the fact that they produce and secrete various molecules including sclerostin, which inhibits osteoblast function and activates osteoclasts44,46. Thus, osteocytes have an important role in the regulation of bone remodelling, and influence both osteoblast and osteoclast function.

Interestingly, osteocytes also express glycoprotein 130 (gp130), which is a receptor subunit that is capable of intracellular signalling and is needed for the cellular action of several cytokines. These cytokines’ cognate receptors (α-chains) cannot induce the activation of intracellular signal transduction by themselves; the α-chain–cytokine dimer can recruit gp130 (the β-chain) to form a trimolecular complex or even a hexamolecular moiety (consisting of two trimolecular complexes of the cytokine, the cognate cytokine receptor and gp130)47,48. Cellular gp130 can bind to the complex of a cytokine and a soluble (circulating) specific α-chain. Therefore, even if they do not express on their membrane the α-chain that specifically binds to the respective cytokine, cells carrying gp130 are highly responsive to cytokines that use gp130 as a co-receptor, such as interleukin-6 (IL-6), oncostatin M, leukaemia inhibitory factor and other cytokines48–51. Consequently, in contrast to the inhibi-tory or neutralizing action of most circulating cytokine receptor molecules on their ligands, circulating α-chains of the IL-6 receptor family act as agonists for cytokines of the IL-6 family, thus enhancing their activity.

Once osteoblasts have produced a bone matrix they may die by apoptosis, which is a major mechanism regu-lating their number and overall function. This pathway is partly mediated by BMP2 and also influenced by hormo-nal factors52–54. Likewise, the control of osteocyte number may occur via pro-apoptotic mechanisms55.

Osteoblast hyperactivity usually leads to exagger-ated bone formation and increased bone mass, such as in sclerosteosis, which is associated with a defect in the sclerostin gene56,57. Conversely, osteoblasts are also the major cell population within bone that initiates

degradation of bony tissue by activating osteoclasts. This can be demonstrated by the mechanism by which PTH leads to bone loss in hyperparathyroidism: PTH binds to its receptor (found only on osteoblasts), which in turn activates RANKL expression and concomitantly downregulates osteoprotegerin expression, thus enhanc-ing osteoclast activation and ultimately leading to bone loss. This mechanism further demonstrates the close coupling between bone formation and bone resorption.

Osteoclasts. Osteoclasts are derived from haematopoietic stem cells, a development that involves an initial step of differentiation toward monocytes and macrophages, requiring macrophage colony-stimulating factor (M-CSF) and its receptor FMS (FIG. 3). Through the activation of the transcription factor extracellular sig-nal-regulated kinase and the anti-apoptotic serine/threo-nine kinase AKT, FMS signalling leads to the induction of cyclin D and, subsequently, the proliferation and survival of osteoclast progenitor cells3.

The next step in the differentiation cascade involves the activation of the receptor for RANKL, receptor activator of NF-κB (RANK; also known as TNFRSF11A)58, which is a member of the tumour necrosis factor (TNF) receptor family. RANKL is expressed on the membrane of osteo-blasts, stromal cells and other cells, and also exists in a soluble form41,58. When RANKL binds to RANK it acti-vates signal transduction pathways involving the adaptor protein TNF receptor-associated factor 6. Subsequently, several kinases such as p38 MAPK (also known as MAPK14) and JUN N-terminal kinase 1 (also known as MAPK8) are activated, which lead to induction of transcription via the various hetero- and (occasionally) homodimers of the AP1 family of proteins. This family includes the molecules FOS, FOSB, FOS-related antigen 1 (FRA1), FRA2, JUN, JUNB and JUND, as well as activat-ing transcription factor and MAF59,60. AP1 regulates the differentiation and proliferation, as well as apoptosis, of various cell types61. TNF receptor-associated factor 6 can also activate IL-1 receptor-associated kinase 1 (IRAK1) or IRAK3, and thus the NF-κB pathway.

Furthermore, signals mediated by receptors using immunoreceptor tyrosine-based activation motifs (ITAMs) recruit spleen tyrosine kinase (SYK), which in turn induces NFATC1 via phospholipase Cγ activation. NFATC1, which is also partly activated by FOS-mediated signalling, is an important transcription factor for osteo-clast generation62,63. Genes transcribed during this pro-cess include the genes encoding the calcitonin receptor, tartrate-resistant acid phosphatase, as well as H+-ATPase, matrix metalloproteinase 13 and cathepsin K, which enable the acidification and degradation of the bony matrix64–66.

M-CSF-, RANKL- and FOS-deficient animals lack osteoclasts and consequently have an osteopetrotic phe-notype67–70. Moreover, monocyte chemoattractant pro-tein 1 (also known as CCL2) and its major receptor, the CC chemokine receptor 2 (CCR2), are also involved in osteoclastogenesis as they increase the expression of RANK71,72. Although osteoblasts are the major source of RANKL expression, various other cells can express this pivotal cytokine, including other mesenchymal cells

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 237

© 2012 Macmillan Publishers Limited. All rights reserved

IL-1RFcγR

Immunecomplex

RANK

RANKL

FMS

CCR2

M-CSF

JAK

STAT

SYK

RAS

Kinases

JNKp38

NF-κB

STAT

STAT STAT

NFAT

Gene transcription

NucleusCytoplasm

Ruffled membrane

TRAP, CTR, CSK, TNF, IL-6, IL-1, MMP13, MMP9, CA2

ERK

PLCγMAPK

TNFR

SRC

PI3K

AKT

mTOR

4EBP/S6K

PPARγ HIF1α

IKK

Anti-IL-6; anti-IL-6R

Anti-TNF

IL-1RA, anti-IL-1

T cell inhibition

B cell depletion

CCR inhibition

MCP1

Denosumab

OPG

AP1

IL-6

gp130IL6R

JAK JAK

Cytoplasm

Odanacatib

Nature Reviews | Drug Discovery

H+ H+ H+Cl– Cl– Cl–

H+ H+ H+Cl– Cl– Cl–

CSK

CSK

CSK

CSK

CSK

CSK

Figure 3 | Signalling in osteoclasts in health and during inflammation. Macrophage colony-stimulating factor (M-CSF), receptor activator of NF-κB ligand (RANKL) and their respective receptors are essential for the differentiation and activation of osteoclasts. Their major signal transduction paths are indicated. Two of several additional receptors affecting osteoclastogenesis are also shown: Fc receptor γ-chain (FcγR), which utilizes spleen tyrosine kinase (SYK) for signalling; and CC chemokine receptor 2 (CCR2), the receptor for monocyte chemoattractant protein 1 (MCP1) and some other chemo kines, which uses the pathway mediated by Janus kinases (JAKs) and signal transducers and activators of transcription (STATs). The activation of these receptors leads to the transcription of genes encoding a variety of molecules that are characteristic of osteoclasts, some of which are indicated in the figure. In the presence of RANKL, various pro-inflammatory cytokines can amplify osteoclastogenesis. Among these cytokines, tumour necrosis factor (TNF), interleukin-6 (IL-6) and IL-1 can lead to a massive upregulation of osteoclast and inhibition of osteoblast activities. Furthermore, as osteoclasts are derived from monocyte and macrophage precursor cells that express FcγR, the presence of immune complexes may enhance osteoclast activation. TNF signals via nuclear factor-κB (NF-κB) and mitogen-activated protein kinases (MAPKs), and the IL-6 receptor (IL-6R) uses the JAK–STAT pathway. The activation of osteoclasts via the inflammatory mechanisms mentioned leads to exaggerated systemic and, where pertinent, local bone loss. Agents that can inhibit these ligands, receptors, or signalling or effector molecules are either licensed or in development. Depicted are various therapeutic options, many of which are in clinical use. Anti-TNF strategies, anti-IL-6R strategies and IL-1R antagonist (IL-1RA) treatment all inhibit inflammatory events by blocking cytokine activity; B cell depletion with CD20-specific antibodies depletes (RANKL-expressing) activated B cells and also reduces the generation of autoantibodies and immune complexes (although such effects are also seen with other effective therapies); and inhibition of T cell co-stimulation might reduce RANKL-expressing activated T cells. Other compounds that may interfere with inflammatory osteoclastogenesis include denosumab (an anti-RANKL agent), rapamycin (a mammalian target of rapamycin (mTOR) inhibitor), odanacatib (a cathepsin K (CSK) inhibitor), as well as JAK and SYK inhibitors. Bisphosphonates (not depicted) primarily effectuate osteoclast apoptosis. However, the extent to which some of these agents inhibit osteoclast activation or differentiation has not yet been fully evaluated. 4EBP, eukaryotic translation initiation factor 4E-binding protein; AP1, activator protein 1; CA2, carbonic anhydrase 2; CTR, calcitonin receptor; ERK, extracellular signal-regulated kinase; HIF1α, hypoxia-inducible factor 1α; IKK, IκB kinase; JNK, JUN N-terminal kinase; MMP9, matrix metalloproteinase 9; NFAT, nuclear factor of activated T cells; OPG, osteoprotegerin; p38, p38 MAPK; PI3K, phosphoinositide 3-kinase; PLCγ, phospholipase Cγ; PPARγ, peroxisome proliferator-activated receptor-γ; S6K, ribosomal protein S6 kinase; TNFR, TNF receptor; TRAP, tartrate-resistant acid phosphatase.

R E V I E W S

238 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

MicroRNAs(miRNAs). Short ribonucleic acid molecules that comprise about 22 nucleotides; they bind to complementary sequences of mRNA molecules, leading to translational repression or degradation of the mRNA. Cellular functions can be inhibited or activated by miRNAs.

like fibroblasts and non-mesenchymal cells like T cells, especially when activated73 (FIG. 4). Thus, activated T lym-phocytes can induce bone loss, be it local or widespread; indeed, TNF-expressing T cells may even have an impor-tant role in osteoporosis induced by oestrogen deficiency74.

Finally, osteoclastogenesis is also regulated by micro-RNAs (miRNAs), which are a large family of single-stranded, short (~20 nucleotide-long) RNAs that bind to complementary sequences of target mRNA transcripts and cause translational repression or gene silencing. miRNAs regulate cell differentiation, proliferation and death, as well as organogenesis and haematopoiesis. miRNAs also affect the activity of osteoclasts and osteo-blasts, and miR-223 appears to have a prominent role in osteoclast differentiation and function75–78.

Inflammatory bone lossSystemic osteoporosis and increased fracture rates occur in several chronic inflammatory conditions. Among them are several rheumatological diseases, such as rheu-matoid arthritis79–82, systemic lupus erythematosus83, axial spondyl arthritis and psoriatic arthritis84, as well as inflammatory bowel disease (IBD)15,85,86, celiac disease87,88, cystic fibrosis89, chronic obstructive pulmonary disease (COPD)90 and periodontitis91. Supporting the observations in patients with these disorders, systemic bone loss is also seen in experimental models of arthritis and colitis92,93.

In periodontitis, both local and systemic bone loss has been reported. Periodontitis is one of the most common chronic disorders in humans and is elicited by subgingi-val infection with various bacterial species. It has been associated with an increased risk of conditions linked to chronic inflammation, including coronary arterioscle-rosis94 and systemic osteoporosis91. In addition, alveo-lar bone loss occurs95; this local bone loss appears to be at least partly mediated by T cells that are activated by local bacteria to overexpress RANKL and stimulate local osteoclastogenesis96,97.

In rheumatoid arthritis two types of local events occur in addition to systemic bone changes: erosive changes and juxta-articular bone loss (BOX 1).

By contrast, in all other chronic inflammatory diseases not occurring in proximity to bone, including IBD and COPD, the bone loss observed is only systemic in nature.

Effects of inflammation on bone cellsInflammation — the body’s response to injury to limit and repair damage — can be elicited by several infectious and non-infectious pathogenic stimuli. It involves the activa-tion of cell populations of the innate and adaptive immune response that produce soluble molecules, which together combat the cause of harm and/or amplify the defence response. In the course of the inflammatory response, an array of cytokines are activated, including interferons, interleukins, chemokines, and so on. Many of these cytokines affect the differentiation and function of osteo-clasts and osteoblasts. The cytokines activated during the course of the inflammatory response have profound effects on the differentiation and activity of osteoblasts and osteoclasts, and are therefore considered to be the mediators of inflammation-associated osteoporosis.

Osteoclasts. Pro-inflammatory cytokines such as TNF, IL-1 and IL-6 all affect osteoclastogenesis. All three cytokines can be produced by an array of cells, and activate a multitude of other cell families by autocrine, paracrine and endocrine mechanisms involving differ-ent — although partly redundant — signal transduction pathways101–108. These three messenger molecules also induce other cytokines, acute phase reactants, non-cytokine inflammatory mediators and proteases, many of which can degrade various tissues109–119. Importantly, they can also amplify osteoclast function and inhibit osteoblast function.

As mentioned above, although osteoclast differentia-tion and activation is pivotally dependent on the presence of M-CSF and RANKL, osteoclastogenesis is enhanced in the presence of TNF, IL-1 or IL-6 (REFS 120–125) (FIG. 3). This is partly a consequence of the induction of RANKL in target cells, but these pro-inflammatory cytokines also appear to induce the differentiation and activation of osteoclasts from the preosteoclast level independently of RANKL123. In addition, under just permissive (subosteo-clastogenic) concentrations of RANKL, TNF can induce the differentiation of monocytes and macrophages to the stage of preosteoclasts58,126. The osteoclastogenic activity of TNF is mediated by the p55 TNF receptor and may be partly counteracted by the activation of the p75 TNF receptor127–129.

Another important pathway involved in osteoclast activation is the induction of SYK, which is mediated by ITAM proteins such as the macrophage fusion receptor or the Fc receptor γ-chain (FcγR; the Fc receptor for immunoglobulin G)130–132. FcγRs may have a particular role in the context of inflammation, as they are activated by immune complexes that can be present in patients with rheumatoid factor-positive rheumatoid arthritis or in patients with systemic lupus erythematosus133–135. Indeed, mice lacking FcγR develop mild osteopetrosis136, thus affirming the role of this receptor in osteoclasto-genesis. Rheumatoid arthritis patients with autoanti-bodies, in contrast to patients who are negative for rheumatoid factor or anti-citrullinated protein antibod-ies (ACPAs), have higher disease activity and more joint damage137. This is presumably mediated by immune complexes that increase TNF production either via the binding of the Fc IgG to the FcγR or via the binding of the antigen component to Toll-like receptors138,139.

It is not completely clear which of the pro-inflam-matory cytokines serves as the major mediator of sys-temic osteoporosis. Although TNF, IL-1 and IL-6 can all amplify osteoclast differentiation and activation, some evidence suggests that IL-6 may have a major role in the development of systemic osteoporosis.

First, IL-6 is the most abundant cytokine in the circulation and has endocrine activity104. Second, in con-trast to other soluble receptors, the soluble IL-6 receptor (IL-6R) — which is also present in high concentrations in the circulation — serves as an agonist of IL-6, and trans-signalling of the soluble IL-6–IL-6R complex via the abundantly present cellular signalling moiety gp130 can dramatically amplify the cytokine’s activity47,105. Third, IL-6 is induced by TNF and IL-1; TNF can also

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 239

© 2012 Macmillan Publishers Limited. All rights reserved

Osteoclast

Fibroblast-likesynoviocyte

Osteoblasts

Bone resorption

Anti-IL-17

Anti-TNF; IL-1RA;anti-IL-6; anti-IL-6R

Anti-DKK1;Anti-SOST

B celldepletion orinhibition

JAK inhibition;SYK inhibition

Inhibition of T cellco-stimulation

DKK1SOST

Ag presentation;co-stimulation

Macrophage

Immunecomplex

Pathogen(LPS)

IL-17IFNγ

IL-6TNFIL-1

RANKL

RANKL

RANK

Preosteoclasts

IL-6IL-1βTNF

IL-6R

TNFR

TNFR

IL-17

CD4+ T cell

B cell

Autoantibodies

Ag

C′

C′R

TLR

Nature Reviews | Drug Discovery

Bone formation

Intermittent PTH

C C

CC, MC

B

BB

ST

ST

STJS

Bisphosphonates;odanacatib; denosumab

RANK

IL-6R

FcR C′R

activate IL-1 but IL-6 can only induce IL-1 activation and not TNF activation103,109,120,140. Thus, IL-6 appears to mediate the effects of both of these other cytokines on osteoclastogenesis, but not vice versa. Fourth, IL-6 can be activated in stress situations irrespective of inflam-mation103,109,118. Fifth, IL-6 mediates the activation of the hypothalamic–pituitary–adrenal axis by IL-1 and TNF141, which leads to the release of glucocorticoids that are also capable of inducing bone loss142,143. Last, oes-trogen deficiency, which occurs following menopause,

causes upregulation of IL-6, whereas IL-6-deficient animals are protected from bone loss induced by oestro-gen deficiency6,144,145.

As oestrogens decrease RANKL expression and increase osteoprotegerin production in osteoblasts, a deficiency in oestrogen will prevent this effect, resulting in an increase in bone resorption over bone formation6. Of interest in this context, deficiency of CCR2, another receptor molecule involved in inflammatory pathways, also protects from oestrogen-dependent osteoporosis71.

Figure 4 | Summary of local and systemic bone events in chronic inflammation, and therapeutic options. B cells, T cells and macrophages can activate osteoclasts in all inflammatory diseases via receptor activator of NF-κB ligand (RANKL) expression (especially in B cells and T cells) and RANK activation (on macrophages), and via secretion of osteoclastogenic cytokines (by all three cell types). In rheumatoid arthritis local activation of fibroblasts also has a role, and in all chronic inflammatory diseases endothelial cells and hypervascularity may have additional effects (not depicted). Many of the cytokines involved are indicated, as are licensed therapies and approaches that are currently under development for the treatment of bone loss. These act indirectly by interfering with the inflammatory processes, or directly by acting on osteoclasts and/or osteoblasts. Although the general figure is schematic in nature, the inserts on the top right show stains of synovial tissue from an animal with experimental inflammatory destructive arthritis with bony and partly subchondral (right side of insert) erosions; similar observations can be made in rheumatoid arthritis patients. At the top, a stain (for tartrate-resistant acid phosphatase) of an inflamed synovial membrane and erosions is shown. The dashed arrows depict a mature, active osteoclast in an erosion; the solid arrows point to preosteoclasts, all arising outside the bone in the hyperinflamed synovial membrane. Note that these cells are on the exosteal side of bone and partly originate in the inflamed synovial tissue (that is, the preosteoclasts and osteoclasts). The lower insert shows a Goldner’s trichrome stain of an erosion in destructive arthritis. The solid arrows point towards osteoblasts, the dashed arrow towards an osteoclast; in destructive arthritis, osteoblasts are indeed present but are usually not capable of sufficiently repairing the destruction owing to their inhibition by inflammatory cytokines and the exaggerated bone-degrading activity of osteoclasts. Ag, antigen; B, bone; C, cartilage; C′, complement; C′R, complement receptor; DKK1, Dickkopf-related protein 1; FcR, Fc receptor; IFNγ, interferon-γ; IL-1RA, interleukin-1 receptor antagonist; JAK, Janus kinase; JS, joint space; LPS, lipopolysaccharide; MC, mineralized cartilage (which is also degraded by osteoclasts); PTH, parathyroid hormone; SOST, sclerostin; ST, inflamed synovial tissue (synovitis); SYK, spleen tyrosine kinase; TLR, Toll-like receptor; TNF, tumour necrosis factor.

R E V I E W S

240 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

T helper 17 (TH17). A recently recognized, unique T helper (TH) cell subset that distinguishes itself from the traditional TH1 and TH2 subsets by producing interleukin-17 (IL-17). TH17 cells appear to have a pivotal role in the pathogenesis of chronic inflammatory and autoimmune diseases, and IL-17 is a strong inducer of RANKL and can thus activate osteoclastogenesis.

Bone mineral density (BMD). An estimate of the amount of mineral matter, expressed as g per cm2 of bone. Its extent is expressed as a t-score (the number of standard deviations above or below the mean of gender- and ethnicity-matched healthy adults of 30 years of age) and z-score (the same as above but matched for the same age). A t-score lower than –2.5 standard deviations reflects osteoporosis.

In summary, IL-6 appears to have an important role in mediating inflammatory osteoporosis, and it may also be involved in pathways leading to osteoporosis that are not elicited by inflammation. Additional cytokines are likely to be involved, including other molecules that utilize the gp130 co-receptor, such as IL-11 and leukae-mia inhibitory factor51,146,147. TNF and IL-1 also appear to have roles beyond IL-6 activation, especially as arthritis and bone loss can occur in IL-6-deficient animals100,148.

It has also been suggested that T helper 17 (TH17) cells have a paramount role in inflammation and auto-immunity98,99. They produce the cytokine IL-17, which represents one of the strongest stimulators of RANKL (the key factor in osteoclast differentiation and activa-tion) in fibroblasts and other cell populations100.

Osteoblasts. The occurrence of osteoporosis in chronic inflammatory diseases questions the role of osteoblasts in maintaining the balance in bone remodelling in these disorders. Interestingly, osteoblasts are present at the site of local erosions in rheumatoid arthritis92 (FIG. 4) but their number and activity is apparently too low to antagonize the exaggerated osteoclast action. Indeed, osteoblast function has been found to be severely impaired in dis-eases such as rheumatoid arthritis owing to the role of

pro-inflammatory cytokines (FIG. 2b). TNF, for example, inhibits osteoblast differentiation on several levels via the p55 TNF receptor, including via inhibition of the differentiation factor RUNX2 (REFS 24,149–153), which is mediated in part by inducing RUNX2 ubiquitylation. However, other pro-inflammatory cytokines such as IL-1 and IL-6 also inhibit osteoblastogenesis154–156, and among the gp130 cytokines only oncostatin M appears to have osteoblast-promoting activities49,50. Many cytokines also activate the NF-κB pathway, which has a negative effect on osteoblast function, presumably via inhibition of JUN N-terminal kinase 1 and thus of the pivotal transcription factor AP1 (REF. 39).

The negative effects of pro-inflammatory cytokines on osteoblasts are further mediated by DKK1, which is an inhibitor of the WNT signalling pathway and is induced by TNF32 (FIG. 2b). Sclerostin represents another inhibitor of osteoblast activation by interfering with the WNT signalling pathway but also by binding to and antagonizing BMPs157–159.

In summary, inflammatory bone loss is a result of the upregulation (and thus hyperactivity) of osteoclasts as well as the downregulation (and thus hypoactivity) of osteoblasts, which leads to a profound net reduction in bone mass.

Treating systemic bone loss Eliminating the underlying cause of disease has proved to be effective in treating osteoporosis in several chronic inflammatory disorders. For example, maintaining a gluten-free diet in celiac disease leads to a rapid recovery of bone mass160,161. Tetracyclines appear to have efficacy in the treatment of systemic and local bone loss in patients with periodontitis, although this may be partly medi-ated by non-antimicrobial effects of tetracyclines162. Moreover, in cystic fibrosis, in which recurrent infections appear to be responsible for the development of systemic bone loss89, rigorous anti-infective therapy leads to an improvement in bone mineral density (BMD)163. However, the causes of most chronic inflammatory disorders are unknown, and therapeutic strategies that are designed to specifically target the inflammatory process must there-fore be used (FIG. 4).

As detailed above, pro-inflammatory cytokines are not only operative in the final pathways of inflamma-tion but they also affect osteoclast and osteoblast activity, leading to systemic bone loss. Therefore, reducing and ideally abrogating inflammation remains an important therapeutic goal.

In a prototypical inflammatory disease like rheu-matoid arthritis, treatment with a synthetic disease-modifying antirheumatic drug such as methotrexate is associated with an improvement in BMD164 but any residual disease activity may allow bone loss to continue. Inflammation is also frequently combated with gluco-corticoids, especially in COPD, rheumatoid arthritis or IBD. Importantly, although these agents are highly effective at reducing or blocking the inflammatory pro-cess, they are also major mediators of osteoporosis as they heavily interfere with bone remodelling165,166. Thus, appropriate measures must be taken to prevent bone

Box 1 | The paradigm of rheumatoid arthritis

In addition to the systemic changes to bone that are similar to those seen in all inflammatory diseases, local events occur in rheumatoid arthritis. The associated inflammation involves all compartments of the joints, such as the joint capsule with its innermost layer, the synovial membrane, cartilage and juxta-articular bone. Erosions are a cardinal sign of rheumatoid arthritis and can occur very early in the course of the disease242–244. They are a consequence of the differentiation and activation of osteoclasts from macrophages within the aggressive synovial tissue (named ‘pannus’), and of the subsequent invasion of subchondral bone. These changes can also be seen in experimental models of arthritis82,245. The enormous osteoclastogenic drive in rheumatoid arthritis is presumably due to the massive overexpression of receptor activator of NF-κB ligand on synovial fibroblasts and infiltrating lymphocytes82,246–248, as well as the high levels of the pro-inflammatory cytokines tumour necrosis factor and interleukin-6 (REFS 232,249), which can amplify osteoclast differentiation and activation, and inhibit osteoblast function. When assessing joints by magnetic resonance imaging, the bone marrow oedema readily seen directly below or near the cartilage surface constitutes an inflammatory infiltrate consisting — in part — of activated B cells. This apparently happens in an attempt to shield off the bone marrow cavity after minute cortical breaks have occurred, which are indicative of a very early erosive phase250,251.

Another site of bone loss that is even part of the 1987 classification criteria for rheumatoid arthritis252 is periarticular demineralization, which constitutes a consequence of osteoclast activation in bone areas adjacent to the joint but not directly below the cartilage layer and thus not easily accessible to the synovial tissue253. As it occurs earlier than systemic osteoporosis, it is likely to be due to local cytokine production and subsequent osteoclast activation. This could ensue in one of the following ways: via little canals that exist throughout compact bone in a perpendicular manner and lengthwise, which partly connect compact bone with the outer side of bone (known as Haversian and Volkmann’s canals); via other diffusion pathways; or via microinvasion of pannus tissue, thus constituting the radiographic surrogate of the bone marrow oedema seen upon magnetic resonance imaging.

The only other disease that shares some of the local features seen in rheumatoid arthritis is psoriatic arthritis. However, in psoriatic arthritis the cytokine levels in joints are only about half of those observed in rheumatoid arthritis249; presumably, therefore, the local changes, erosions and juxta-articular bone loss are usually much less pronounced254.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 241

© 2012 Macmillan Publishers Limited. All rights reserved

loss when glucocorticoids are used in the treatment of inflammation167,168. Vitamin D and calcium supplemen-tation should accompany all therapies for osteoporosis.

Targeted therapies that are licensed for several chronic inflammatory conditions as well as the direct treatment of bone loss are listed in TABLE 1, and agents that are currently in clinical trials are listed in TABLE 2. All biologics licensed for the treatment of rheumatoid arthritis inhibit local bone erosions, and some have also been shown to improve systemic bone loss. These and other promising agents are discussed below.

Cytokine inhibition. Inhibition of TNF, especially when combined with methotrexate, is a highly effec-tive anti-inflammatory approach that dramatically reduces or even halts the progression of local bone loss (erosions), and also prevents systemic loss of bone in rheumatoid arthritis and respective experimental mod-els of arthritis92,164,169. Likewise, inhibiting TNF in IBD leads to an improvement in BMD170,171. Several TNF blockers are currently in use, including: the recep-tor construct etanercept (Enbrel; Amgen/Pfizer); the chimeric monoclonal antibody infliximab (Remicade; Centocor Ortho Biotech); the human monoclonal anti-bodies adalimumab (Humira; Abbott) and golimumab (Simponi; Centocor Ortho Biotech); and the pegylated Fab′ fragment of a humanized monoclonal antibody, certolizumab pegol (Cimzia; UCB Pharma). All five agents exhibit similar efficacy on joint damage and, in studies so far, also on systemic osteoporosis in rheuma-toid arthritis. They are all also effective in the treatment of psoriatic arthritis, axial spondylarthritis and psoriasis. By contrast, in IBD only the antibodies interfere with the disease; etanercept does not.

As discussed above, IL-6 is another pro-inflammatory cytokine that is involved in the activation of osteo-clasts124,125. Tocilizumab (Actemra; Roche) is a human-ized IL-6R antibody that retards joint damage in rheumatoid arthritis172 and, at least in high doses, may alleviate the symptoms of Crohn’s disease173. It is also likely to inhibit inflammation-associated osteoporosis. This can be assumed based on the protective effects of IL-6 deficiency on bone loss in ovariectomized animals. Another monoclonal antibody targeting IL-6R (sari-lumab) and several monoclonal antibodies against IL-6 itself are being investigated, and some are currently in or have completed Phase II trials; these include olokizumab, sirukumab and ALD518/BMS-945429 (REFS 174–176). Their effect on bone loss will probably be assessed during their clinical development.

Anakinra (Kineret; Amgen) is a recombinant IL-1R antagonist that is clinically only weakly effective in rheu-matoid arthritis177. However, experimental models sug-gest that its anti-inflammatory and osteoclast-inhibiting effects may improve systemic bone loss178.

Finally, as mentioned previously, IL-17 is one of the most active molecules involved in osteoclastogenesis. IL-17-specific antibodies have shown clinical efficacy in rheumatoid arthritis179 and may have relevance in IBD, and it is therefore conceivable that these effects may translate into inhibition of systemic bone loss. Indeed,

indirectly supporting this, IL-17 levels were dramati-cally reduced and the generation of experimental arthritis and bony damage inhibited in animals deficient in miR-155 (REF. 180); miR-155 has an important role in immunoinflammatory responses — it is upregulated by lipopolysaccharide and highly expressed in the synovial membrane of patients with rheumatoid arthritis180,181.

Inhibition of B‑ and T cells. Although pro-inflammatory cytokine activation has been suggested to constitute the final pathway in the cascade leading to chronic inflamma-tory disorders, several events take place upstream of their activation that involve T- and B cells. Abatacept (Orencia; Bristol-Myers Squibb), a cytotoxic T lymphocyte-associ-ated antigen 4 (CTLA4)–Ig T cell co-stimulation inhibi-tor, and rituximab (Rituxan/MabThera; Biogen Idec/Genentech/Roche), a CD20-specific antibody that depletes B cells, are both clinically effective and inhibit the progres-sion of joint damage182. Although there is a lack of detailed data on their effects on systemic bone loss, this is likely to be a consequence of their overall anti-inflammatory effects. However, activated T- and B cells can promote systemic osteoclastogenesis by various means (especially through RANKL expression; for more details see above), and blocking these pathways — which are presumed to be upstream of inflammatory cytokine activation — may exert beneficial effects on bone damage183–185. The efficacy of these compounds in psoriatic arthritis has not been well studied.

Inhibition of osteoclastogenic pathways. Several inhibi-tors of signal transduction mechanisms have been tested in the treatment of chronic inflammatory diseases. For example, inhibiting Janus kinases (JAKs) with tofacitinib and inhibiting SYK with fostamatinib has been proven to be efficacious in rheumatoid arthritis186,187. As the JAK–STAT (signal transducer and activator of transcription) pathway is upregulated and SYK is activated in inflamma-tory diseases, via the action of IL-6 and the engagement of FcγRs by immune complexes, respectively, inhibiting these signalling molecules may also positively affect systemic inflammatory bone loss. Although there is a lack of direct evidence, the ability of tofacitinib to reduce the progression of structural damage in rheumatoid arthri-tis186 indicates that this local anti-osteoclastogenic effect may also extend to systemic bone loss.

Another pathway that activates osteoclasts involves AKT and the serine/threonine protein kinase mamma-lian target of rapamycin (mTOR). Rapamycin and its analogues are used as immunomodulatory agents, par-ticularly in patients undergoing renal transplant. This treatment also appears to have some efficacy in rheuma-toid arthritis and Crohn’s disease, as well as in respective experimental models188,189. Moreover, recent evidence suggests that inhibition of mTOR may be highly effica-cious in reducing osteoclast activity and improving bone quality190. mTOR mediates its effects by translational control, including that mediated by the transcription factor hypoxia-inducible factor 1α; however, it may also directly exert transcriptional control mechanisms such as on peroxisome proliferator-activated receptor-γ191–193.

R E V I E W S

242 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

Proteasome A very large intracellular protein complex that enzymatically degrades damaged proteins but is also involved in the regulation of intracellular proteins. Proteasomes degrade bone morphogenetic proteins (BMPs), and proteasome inhibitors may therefore enhance the ability of BMPs to activate osteoblasts.

The transcription factors hypoxia-inducible factor 1α and peroxisome proliferator-activated receptor-γ have important roles in bone homeostasis in general and osteoclastogenesis in particular194. Moreover, it appears that TNF and other cytokines may mediate their effects in part via mTOR activation195.

Inhibition of the proteasome may represent another promising approach for the treatment of systemic bone loss. However, the currently available proteasome inhibi-tor bortezomib (Velcade; Millennium Pharmaceuticals), which is a highly effective agent for the treatment of refractory multiple myeloma197, is associated with neu-rotoxicity. Osteoblast-specific proteasome inhibition appears to mediate its effects by inhibiting the degradation of BMP2, making it more readily available to stimulate osteoblasts196. Bortezomib counteracts osteolysis induced in multiple myeloma, and appears to exert this effect not only by affecting the tumour but also by inter fering with osteoclastogenesis and osteoblast function198–200. Moreover, proteasome inhibition appears to be effec-tive in improving signs and symptoms of experimental colitis and in models of arthritis201–203.

Blocking osteoclast activity. Various compounds can interfere directly with osteoclast activity without having any known effect on inflammation.

Traditionally, bisphosphonates — equivalents of inorganic pyrophosphates with high affinity for cal-cium phosphate — have demonstrated high efficacy in the treatment of osteoporosis204. Indeed, the more recently developed aminobisphosphonates alen-dronate (Fosamax; Merck), ibandronate (Boniva; GlaxoSmithKline), risedronate (Actonel; Sanofi) and zoledronate (Zometa; Novartis) are successfully used for combating postmenopausal osteoporosis, and some have also shown efficacy in treating osteoporosis associated with inflammatory disorders80,205,206. The efficacy of bis-phosphonates in inflammatory bone loss can also be seen in an experimental model of lipopolysaccharide-mediated inflammation207. Interestingly, the data obtained in this model indicate that bisphosphonates inhibit the expres-sion of FOS and NFATC1 in bone marrow-derived macrophages. This suggests that bisphosphonates have osteoclast-inhibiting effects beyond the direct induc-tion of apoptosis, which is regarded as the main action of these compounds.

Common adverse events of bisphosphonates include oesophageal irritation when they are administered via the oral route, and an increase in acute reactions (such as fever or myalgia) when they are administered intra-venously, especially at the first application (in up to 30% of cases). Rare but serious side effects include impairment

Table 1 | Currently available agents to target inflammation and/or inflammatory bone loss

Agents (trade name; company)

Type of molecule

Targets Diseases Mode of action Refs

Bisphosphonates (many) Small, chemical, oral

Osteoclasts Osteoporosis Anti-osteoclast 15

Denosumab (Prolia/Xgeva; Amgen)

Human mAb RANKL Osteoporosis Anti-osteoclast 212

Adalimumab* (Humira; Abbott)

Human mAb TNF Rheumatoid arthritis, psoriatic arthritis, ankylosing spondylitis, juvenile idiopathic arthritis, inflammatory bowel disease

Anti-inflammatory 230

Certolizumab* (Cimzia; UCB Pharma)

Humanized Fab

Etanercept* (Enbrel; Amgen/Pfizer)

Receptor construct

Golimumab* (Simponi; Centocor Ortho Biotech)

Human mAb

Infliximab* (Remicade; Centocor Ortho Biotech)

Chimeric mAb

Abatacept (Orencia; Bristol-Myers Squibb)

Receptor construct

CD80 and/or CD86 (co-stimulation)

Rheumatoid arthritis Anti-inflammatory 183

Rituximab (Rituxan/MabThera; Biogen Idec/Genentech/Roche)

Chimeric mAb

CD20 (B cell) Rheumatoid arthritis, B cell lymphoma, (multiple sclerosis)‡

Anti-inflammatory 184

Tocilizumab (Actemra; Roche)

Humanized mAb

IL-6R Rheumatoid arthritis, juvenile idiopathic arthritis, Castleman’s disease

Anti-inflammatory 172

Belimumab (Benlysta; GlaxoSmithKline)

Human mAb BLYS Systemic lupus erythematosus

B cell inhibition 255

BLYS, B lymphocyte stimulator; IL-6R, interleukin-6 receptor; mAb, monoclonal antibody; RANKL, receptor activator of NF-κB ligand; TNF, tumour necrosis factor. *Not all agents may be licensed in all countries for all indications, and etanercept is not effective in inflammatory bowel diseases. ‡Anti-CD20 is not licensed for multiple sclerosis but has shown significant efficacy in clinical trials.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 243

© 2012 Macmillan Publishers Limited. All rights reserved

ExostosesBony overgrowths on a bony surface, frequently seen as spurs. This abnormal or exaggerated bone formation can occur in the context of an inflammatory response that involves bone, and may cause pain or discomfort.

of renal function, osteonecrosis of the jaw and untypical fractures. Osteonecrosis of the jaw occurs especially in patients who undergo invasive dental procedures; its incidence ranges between 1 in 10,000 and 1 in 250,000 (REF. 208). The pathogenesis of osteonecrosis of the jaw is unknown but was thought to be related to the use of bisphosphonates. The increased rate of insufficiency fractures, particularly in the subtrochanteric and pelvic regions, may be a consequence of reduced bone turnover208.

Powerful results in the treatment of postmenopau-sal osteoporosis have been obtained with denosumab (Prolia/Xgeva; Amgen), a monoclonal antibody spe-cific for RANKL209,210 that inhibits the most important pathway of osteoclast differentiation and activation: the RANK pathway. A statistically significant increase in BMD at the lumbar spine (~4%) and trochanter region (~2%) has been seen with denosumab compared with placebo after 1 year of therapy in patients with inflam-matory bone loss associated with rheumatoid arthritis, even though disease activity was not affected at all by denosumab therapy nor was progression of cartilage damage211. Thus, denosumab inhibits systemic (and local) bone loss independently of inflammatory disease activity. It may constitute a highly effective means of treating systemic osteoporosis in chronic inflammatory diseases, even if the local disease process continues to be active, and may be particularly useful in patients who are glucocorticoid-dependent and therefore suffer from additional glucocorticoid-induced bone loss212.

The adverse event profile of denosumab includes an increase in eczema and possibly also infections such as cellulitis213. Interestingly, osteonecrosis of the jaw has also been observed in patients undergoing denosumab ther-apy and is thus not just a complication of bisphosphonate treatment214; this suggests that the interference with bone homoeostasis, rather than the specific mode of action of bisphosphonates, may be responsible for this major and debilitating safety issue. Therefore, it is possible that with a more widespread and longer-term use of denosumab, atypical femoral and pelvic fractures may also become apparent. Data from registries or post-marketing surveillance may reveal other rare adverse events.

Finally, another promising target for the treatment of inflammatory osteoporosis is cathepsin K. A small chemical compound, odanacatib, is a potent inhibitor of this enzyme and has already been successfully applied in Phase III trials in postmenopaousal osteoporosis262.

Enhancing osteoblast activity. Various therapeutic measures may be taken to activate osteoblasts but these are either not yet fully established for inflammatory bone loss or they are still in the developmental stage.

Intermittent PTH application is a major stimulus for osteoblastogenesis (FIG. 2a; 4). Applying PTH in a TNF-dependent experimental model of arthritis, especially in combination with TNF blockade, reversed systemic bone loss92. Indeed, even the accrued local erosions were filled and thus local bone destruction reversed; this is an effect that is usually not seen with any other type of treatment, as local bone changes can be halted but are

rarely reversed in rheumatoid arthritis. The PTH ana-logue teriparatide (Forteo; Lilly) may constitute an effec-tive way to interfere with inflammatory bone loss215 but respective trials will be needed.

As detailed above, molecules like DKK1 and scle-rostin are major inhibitors of osteoblast function, and act by interfering with the WNT signalling pathway. Importantly, both of these proteins are induced by pro-inflammatory cytokines such as TNF32. TNF appears to stimulate the overexpression of sclerostin by osteoblasts themselves, thus leading to autocrine inhibition of osteo-blast activity216. Monoclonal antibodies specific for DKK1 may protect against systemic bone loss under inflam-matory conditions217. Clinical trials using a humanized monoclonal antibody of DKK1 — PF-04840082 — are underway, and preclinical modelling has been used to predict the dose to be studied in early-phase clinical trials218.

Several pharmaceutical companies are currently also developing monoclonal antibodies specific for sclerostin, and inhibiting sclerostin may represent yet another option for interfering with inflammatory systemic bone loss. Recently, the results of a Phase II trial of a sclerostin-specific monoclonal antibody — AMG785/CDP7851 — in postmenopausal osteoporosis were announced in a press release (see the 21 April 2011 press release on the Amgen website). The primary end point was met with a statistically significant increase in lumbar spine BMD at 12 months in the active arm compared with the placebo arm, and the antibody was reported to have compared positively with the two active comparators, teriparatide and alendronate219–221.

There is currently an insufficient knowledge of the adverse event profiles of these future therapies. However, one of the reasons why bone formation is suppressed dur-ing joint inflammation might be to prevent the formation of aberrant bone in and around the joint. In this context, it is important to bear in mind that DKK1-knockout mice develop exostoses during inflammation, anti-DKK1 ther-apy leads to osteophyte formation in experimental arthri-tis, and high bone mass mutations can also be afflicted by exostoses32,222. Thus, exostoses are a potential side effect of such therapy. Heterotopic ossifications have long been known to occur after joint replacement surgery, and are treated using NSAIDs223. They have also been reported to occur following the treatment of spinal fusion sur-gery with BMP2, a therapy that was licensed several years ago224. Finally, sclerosteosis is associated with some adverse effects on bone, such as obstruction of foramina by skeletal overgrowth225. Although this occurs during a lifetime of constitutive sclerostin deficiency and is thus less likely to occur with short-term sclerostin blockade in adulthood, it is a possible adverse effect that needs to be addressed in clinical trials.

Local bone loss therapyLocal demineralization. Although periarticular osteo-porosis is a characteristic feature of rheumatoid arthri-tis, it is the most ignored of the three bone loss aspects of the disease; the other two are articular and systemic bone loss. However, its presence does not necessarily

R E V I E W S

244 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

indicate an increased fracture risk; rather, it constitutes a surrogate of the inflammatory process in the adjacent joints. Consequently, interfering with this process should restore the periarticular bony structure; indeed, this has been proven to be the case with TNF inhibition that is combined with methotrexate treatment226.

In addition, in periodontitis there is substantial local loss of alveolar bone, which is the major cause of tooth loss in this disease227. Effective therapy will improve both local and systemic bone loss97,162.

Articular bone loss: erosions. Erosions are a major char-acteristic of rheumatoid arthritis but they can also occur — albeit usually to a lesser extent — in other diseases such as psoriatic arthritis228. Radiographic scores are the gold standard for assessing joint damage in rheu-matoid arthritis and psoriatic arthritis229, and the total scores usually comprise the erosion score and the joint space narrowing score as a surrogate of cartilage dam-age. Effective treatment will halt the progression of joint damage, either by reaching remission with chemical

Table 2 | Potential future therapeutics to target inflammatory bone loss

Agents (trade name; company)

Targets Disease approved for

Current stage of development

Mode of action Refs

Tofacitinib (Pfizer) JAK1 and JAK3

NA Phase III completed (RA)

Anti-inflammatory 186

Fostamatinib (Rigel/AstraZeneca)

SYK NA Phase II/III (RA, SLE)

Anti-inflammatory 187

Sarilumab (Regeneron/Sanofi)

IL-6R NA Phase IIb (RA) Anti-inflammatory

Anti-IL-6: sirukumab (Janssen), olokizumab (UCB Pharma) and BMS-945429 (Alder Biopharmaceuticals/Bristol-Myers Squibb)

IL-6 NA Phase IIb (for inflammatory diseases)

Anti-inflammatory 174–176, 256

Anti-IL-17: secukinumab (Novartis) and LY2439821 (Lilly)

IL-17 NA Phase IIb (for inflammatory diseases)

Anti-inflammatory 179

Anti-sclerostin (Amgen/UCB Pharma)

Sclerostin NA Phase II Interference with osteoblast inhibition

219–221

Anti-DKK1: BHQ880 (Novartis) and PF-04840082 (Pfizer)

DKK1 NA Phase II Interference with osteoblast inhibition

217, 218, 261

Odanacatib (Merck Sharp & Dohme)

Cathepsin K Osteoporosis Phase III Interference with the major bone-degrading osteoclastic enzyme cathepsin K

262, 263

PTH: teriparatide (Forteo; Lilly)

Osteoblast Osteoporosis Licensed Repair of bone loss 92

Antagomirs miRNA NA Inhibition of miRNAs that induce inflammatory responses or inhibit bone formation

234

mTOR inhibitors: sirolimus (Rapamune; Pfizer) and everolimus (Afinitor; Novartis)

mTOR Transplantation Licensed for transplantation

Anti-osteoclast 189

Ustekinumab (Stelara; Janssen)

IL-12 and/or IL-23

Psoriasis Licensed for psoriasis

Anti-inflammatory 257

Ofatumumab (GlaxoSmithKline)

CD20 NA Phase II/III for other diseases

B cell depletion 258

Atacicept (Merck Serono)

BLYS NA Phase II/III for other diseases

B cell inhibition 259, 260

Bortezomib (Millennium) and similar molecules

Proteasome Bone loss Licensed in multiple myeloma

Interference with degradation of BMPs and other molecules leading to increase in osteoblast function and/or decrease in osteoclast function

196–203

BLYS, B lymphocyte stimulator; BMP, bone morphogenetic protein; DKK1, Dickkopf-related protein 1; IL-6, interleukin-6; IL-6R, IL-6 receptor; JAK1, Janus kinase 1; miRNA, microRNA; mTOR, mammalian target of rapamycin; NA, not applicable (currently in development); PTH, parathyroid hormone; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus; SYK, spleen tyrosine kinase.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 245

© 2012 Macmillan Publishers Limited. All rights reserved

disease-modifying antirheumatic drugs such as metho-trexate, sulphasalazine or leflunomide, or by using TNF blockers in combination with methotrexate230. However, other biological agents such as abatacept, rituximab and tocilizumab also retard or halt the progression of bony joint destruction and cartilage damage. From 6 months of treatment onwards, there appear to be no differences between the various targeted therapeutic principles.

TNF blockers plus methotrexate have been repeatedly shown to inhibit joint damage in rheumatoid arthritis, even if the activity of the disease continues to be high231 (as assessed by composite measures, joint counts or acute-phase reactant levels). Presumably, this inhibition of joint damage is due to a reduction in biologically active TNF below the threshold that is needed to activate osteoclasts, which may be higher than that required to induce inflam-mation232. Consequently, TNF blockers plus methotrexate might not only interfere with the loss of bone after halting disease activity but may also reduce systemic osteoclast activity even if patients continue to have an active form of the disease. A similar observation was recently made fol-lowing the inhibition of IL-6 pathways with tocilizumab233.

As mentioned above, among other new principles, interference with miRNAs may become a useful thera-peutic approach181. Indeed, one particular miRNA — miR-155 — may have an important role in the devel-opment of chronic arthritis, and its absence is associated with a dramatic reduction in the inflammatory as well as the destructive response, including abrogation of osteo-clast generation180. Therapeutic inactivation of miRNAs is currently a highly attractive topic234.

Directly inhibiting osteoclasts (rather than indirectly inhibiting them by interfering with the inflammatory pathways) may also be an effective approach for retarding or halting the progression of joint damage in rheumatoid arthritis. This approach may be particularly indicated to prevent an increase in bony joint destruction in patients in whom inflammation cannot be sufficiently controlled by other therapies. The use of bisphosphonates has been shown to inhibit erosions, reduce periarticular bone loss and increase systemic trabecular bone mass, with an overall massive reduction in osteoclast numbers, in exper-imental models of arthritis235,236. A recent study of zole-dronate in patients with early-stage rheumatoid arthritis suggested that the use of this bisphosphonate may be a viable approach to halt the erosive process237.

Regarding the inhibition of RANKL, experimental data from animal models indicate that it has an excellent effect on all aspects of bone loss92,238,239. Denosumab has also been studied in rheumatoid arthritis and has dem-onstrated substantial efficacy on bony erosions, both radiographically and by magnetic resonance imaging211. However, joint space narrowing (a surrogate of cartilage damage) continued to progress somewhat with deno-sumab treatment. As cartilage damage may have a more important effect on physical function than bony dam-age240, RANKL inhibition does not appear to be sufficient as a therapy for joint damage (for its systemic effects on bone, see above). Thus, if used in rheumatoid arthritis, RANKL inhibition would have to be combined with drugs that inhibit cartilage degradation. Methotrexate,

which was combined with denosumab in the study described above, was not sufficient for this purpose.

Joint damage in rheumatoid arthritis is regarded as irre-versible, as there is little evidence for substantial repair even with the best therapies241. This is due to the low activity of osteoblasts in rheumatoid arthritis, which apparently do not resume appropriate bone production even if osteoclast function is inhibited by highly effective anti-inflammatory therapies such as biological agents. Therefore, stimulation of osteoblasts appears to be of additional importance for future therapeutic approaches. Indeed, in experimental models, PTH in combination with TNF inhibition was highly efficacious in halting the progression of joint dam-age and in repairing pre-existing joint destruction92. Future studies will determine whether blocking inhibitory mol-ecules such as DKK1 and sclerostin (see above) will allow osteoblasts to resume their local bone repair activity.

SummaryInflammation is a cardinal and highly neglected cause of both local and systemic bone loss, which may ultimately lead to disability and mortality. The negative bone mass balance that occurs in the course of chronic inflamma-tory diseases such as IBD, periodontitis, rheumatoid arthritis and other inflammatory diseases is mostly medi-ated by cytokines that activate osteoclasts while simulta-neously impeding osteoblast function.

Inhibition of inflammation-induced bone loss can be achieved in several ways. Most importantly, interfer-ing with inflammation is paramount in combating local bony damage and systemic osteoporosis. However, cur-rent therapies lead to remission of inflammation in only a relatively small proportion of patients suffering from these diseases, and smouldering inflammation usually continues to exert negative consequences on systemic and (when pertinent) local bone. Therefore, the development of novel therapies is vital; these include treatments that focus on reducing osteoclast activity and/or enhancing osteoblast functionality to restore the appropriate bal-ance between bone degradation and rebuilding of bone mass. One such agent is denosumab, a RANKL inhibi-tor, which has recently been licensed for the treatment of postmenopausal osteoporosis and bone metastases, and targets osteoclast activation. Another possibility is the use of bisphosphonates, which also inhibit osteoclasts but less potently than denosumab. Intermittent use of PTH, which activates osteoblasts, is yet another approach for achieving balance in the modelling–remodelling process.

Future therapies that are currently being evaluated target various molecules such as the inhibitors of osteo-blast activation DKK1 and sclerostin, as well as molecules that are involved in the signalling of both inflammatory and bone cells such as JAKs and SYK. The challenge will be to find the right combination of anti-inflammatory and bone mass-restoring principles and to overcome the limitations of potential adverse events, such as infec-tions, that might occur with some of these combinations. However, the armament for combating inflammation-induced bone loss is increasing, and therefore inflam-matory systemic osteoporosis and irreversibility of local joint damage may soon become features of the past.

R E V I E W S

246 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

1. Sims, N. A. & Gooi, J. H. Bone remodeling: multiple cellular interactions required for coupling of bone formation and resorption. Semin. Cell Dev. Biol. 19, 444–451 (2008).

2. Bodine, P. V. & Komm, B. S. Wnt signaling and osteoblastogenesis. Rev. Endocr. Metab. Disord. 7, 33–39 (2006).

3. Yavropoulou, M. P. & Yovos, J. G. Osteoclastogenesis — current knowledge and future perspectives. J. Musculoskelet. Neuronal. Interact. 8, 204–216 (2008).

4. Eriksen, E. F. Cellular mechanisms of bone remodeling. Rev. Endocr. Metab. Disord. 11, 219–227 (2010).

5. Raisz, L. G. & Rodan, G. A. Pathogenesis of osteoporosis. Endocrinol. Metab. Clin. North Am. 32, 15–24 (2003).

6. Frenkel, B. et al. Regulation of adult bone turnover by sex steroids. J. Cell Physiol. 224, 305–310 (2010).

7. Mosekilde, L. Primary hyperparathyroidism and the skeleton. Clin. Endocrinol. (Oxf.) 69, 1–19 (2008).

8. Bliuc, D. et al. Mortality risk associated with low-trauma osteoporotic fracture and subsequent fracture in men and women. JAMA 301, 513–521 (2009).

9. Johnell, O. et al. Mortality after osteoporotic fractures. Osteoporos. Int. 15, 38–42 (2004).

10. Grabowski, P. Physiology of bone. Endocr. Dev. 16, 32–48 (2009).

11. Holm, K. & Hedricks, C. Immobility and bone loss in the aging adult. Crit. Care Nurs. Q. 12, 46–51 (1989).

12. Michalakis, K., Peitsidis, P. & Ilias, I. Pregnancy- and lactation-associated osteoporosis: a narrative mini-review. Endocr. Regul. 45, 43–47 (2011).

13. Howe, T. E. et al. Exercise for preventing and treating osteoporosis in postmenopausal women. Cochrane Database. Syst. Rev. CD000333 (2011).

14. Sinaki, M. et al. The role of exercise in the treatment of osteoporosis. Curr. Osteoporos. Rep. 8, 138–144 (2010).

15. Mundy, G. R. Osteoporosis and inflammation. Nutr. Rev. 65, S147–S151 (2007).

16. Romas, E. & Gillespie, M. T. Inflammation-induced bone loss: can it be prevented? Rheum. Dis. Clin. North Am. 32, 759–773 (2006).

17. Smolen, J. S. et al. Radiographic changes in rheumatoid arthritis patients attaining different disease activity states with methotrexate monotherapy and infliximab plus methotrexate: the impacts of remission and TNF-blockade. Ann. Rheum. Dis. 68, 823–827 (2009).

18. Gaur, T. et al. Canonical WNT signaling promotes osteogenesis by directly stimulating Runx2 gene expression. J. Biol. Chem. 280, 33132–33140 (2005).

19. Derynck, R. & Zhang, Y. E. Smad-dependent and Smad-independent pathways in TGF-β family signalling. Nature 425, 577–584 (2003).This study provides an authoritative review of the signal transduction pathways mediated by the transforming growth factor-β protein family, to which BMPs belong, which utilize SMADs as intracellular effectors of transcriptional regulation.

20. Qin, L. et al. Gene expression profiles and transcription factors involved in parathyroid hormone signaling in osteoblasts revealed by microarray and bioinformatics. J. Biol. Chem. 278, 19723–19731 (2003).

21. Bodine, P. V., Seestaller-Wehr, L., Kharode, Y. P., Bex, F. J. & Komm, B. S. Bone anabolic effects of parathyroid hormone are blunted by deletion of the Wnt antagonist secreted frizzled-related protein-1. J. Cell Physiol. 210, 352–357 (2007).

22. Gazzerro, E. & Canalis, E. Bone morphogenetic proteins and their antagonists. Rev. Endocr. Metab. Disord. 7, 51–65 (2006).

23. Franceschi, R. T. & Xiao, G. Regulation of the osteoblast-specific transcription factor, Runx2: responsiveness to multiple signal transduction pathways. J. Cell Biochem. 88, 446–454 (2003).

24. Zaidi, M. Skeletal remodeling in health and disease. Nature Med. 13, 791–801 (2007).

25. Lian, J. B. et al. Networks and hubs for the transcriptional control of osteoblastogenesis. Rev. Endocr. Metab. Disord. 7, 1–16 (2006).

26. Yamane, T. et al. Wnt signaling regulates hemopoiesis through stromal cells. J. Immunol. 167, 765–772 (2001).

27. Van Den Berg, D. J., Sharma, A. K., Bruno, E. & Hoffman, R. Role of members of the Wnt gene family in human hematopoiesis. Blood 92, 3189–3202 (1998).

28. van Amerongen, R. & Nusse, R. Towards an integrated view of Wnt signaling in development. Development 136, 3205–3214 (2009).

29. Fedi, P. et al. Isolation and biochemical characterization of the human Dkk-1 homologue, a novel inhibitor of mammalian Wnt signaling. J. Biol. Chem. 274, 19465–19472 (1999).

30. Kwack, M. H. et al. Dihydrotestosterone-inducible Dickkopf 1 from balding dermal papilla cells causes apoptosis in follicular keratinocytes. J. Invest. Dermatol. 128, 262–269 (2008).

31. Tian, E. et al. The role of the Wnt-signaling antagonist DKK1 in the development of osteolytic lesions in multiple myeloma. N. Engl. J. Med. 349, 2483–2494 (2003).

32. Diarra, D. et al. Dickkopf-1 is a master regulator of joint remodeling. Nature Med. 13, 156–163 (2007).

33. Mao, B. et al. Kremen proteins are Dickkopf receptors that regulate Wnt/β-catenin signalling. Nature 417, 664–667 (2002).

34. Veverka, V. et al. Characterization of the structural features and interactions of sclerostin: molecular insight into a key regulator of Wnt-mediated bone formation. J. Biol. Chem. 284, 10890–10900 (2009).

35. Romero, G. et al. Parathyroid hormone receptor directly interacts with Dishevelled to regulate β-catenin signaling and osteoclastogenesis. J. Biol. Chem. 285, 14756–14763 (2010).

36. Nakashima, K. et al. The novel zinc finger-containing transcription factor osterix is required for osteoblast differentiation and bone formation. Cell 108, 17–29 (2002).This publication provides the first description of the role of osterix in osteoblast differentiation.

37. Koga, T. et al. NFAT and osterix cooperatively regulate bone formation. Nature Med. 11, 880–885 (2005).

38. Chang, J. et al. Inhibition of osteoblastic bone formation by nuclear factor-κB. Nature Med. 15, 682–689 (2009).

39. Krum, S. A., Chang, J., Miranda-Carboni, G. & Wang, C. Y. Novel functions for NFκB: inhibition of bone formation. Nature Rev. Rheumatol. 6, 607–611 (2010).

40. Gooi, J. H. et al. Calcitonin impairs the anabolic effect of PTH in young rats and stimulates expression of sclerostin by osteocytes. Bone 46, 1486–1497 (2010).

41. Yasuda, H. et al. Osteoclast differentiation factor is a ligand for osteoprotegerin/osteoclastogenesis-inhibitory factor and is identical to TRANCE/RANKL. Proc. Natl Acad. Sci. USA 95, 3597–3602 (1998).This was one of the initial studies describing RANKL and its pivotal osteoclastogenic role.

42. Murshed, M., Harmey, D., Millan, J. L., McKee, M. D. & Karsenty, G. Unique coexpression in osteoblasts of broadly expressed genes accounts for the spatial restriction of ECM mineralization to bone. Genes Dev. 19, 1093–1104 (2005).

43. Bonewald, L. F. Osteocytes as dynamic multifunctional cells. Ann. NY Acad. Sci. 1116, 281–290 (2007).

44. Bonewald, L. F. & Johnson, M. L. Osteocytes, mechanosensing and Wnt signaling. Bone 42, 606–615 (2008).

45. Schneider, P., Meier, M., Wepf, R. & Muller, R. Towards quantitative 3D imaging of the osteocyte lacuno–canalicular network. Bone 47, 848–858 (2010).

46. van Bezooijen, R. L., ten Dijke, P., Papapoulos, S. E. & Lowik, C. W. SOST/sclerostin, an osteocyte-derived negative regulator of bone formation. Cytokine Growth Factor Rev. 16, 319–327 (2005).

47. Murakami, M. et al. IL-6-induced homodimerization of gp130 and associated activation of a tyrosine kinase. Science 260, 1808–1810 (1993).This paper clarifies the complexity of IL-6R molecules and signalling.

48. Rose-John, S., Scheller, J., Elson, G. & Jones, S. A. Interleukin-6 biology is coordinated by membrane-bound and soluble receptors: role in inflammation and cancer. J. Leukoc. Biol. 80, 227–236 (2006).

49. Walker, E. C. et al. Oncostatin M promotes bone formation independently of resorption when signaling through leukemia inhibitory factor receptor in mice. J. Clin. Invest. 120, 582–592 (2010).

50. Malaval, L., Liu, F., Vernallis, A. B. & Aubin, J. E. GP130/OSMR is the only LIF/IL-6 family receptor complex to promote osteoblast differentiation of calvaria progenitors. J. Cell Physiol. 204, 585–593 (2005).

51. Sims, N. A. & Walsh, N. C. gp130 cytokines and bone remodelling in health and disease. BMB Rep. 43, 513–523 (2010).

52. Manolagas, S. C. Birth and death of bone cells: basic regulatory mechanisms and implications for the pathogenesis and treatment of osteoporosis. Endocr. Rev. 21, 115–137 (2000).

53. Hay, E., Lemonnier, J., Fromigue, O., Guenou, H. & Marie, P. J. Bone morphogenetic protein receptor IB signaling mediates apoptosis independently of differentiation in osteoblastic cells. J. Biol. Chem. 279, 1650–1658 (2004).

54. Bradford, P. G., Gerace, K. V., Roland, R. L. & Chrzan, B. G. Estrogen regulation of apoptosis in osteoblasts. Physiol. Behav. 99, 181–185 (2010).

55. Moriishi, T. et al. Overexpression of Bcl2 in osteoblasts inhibits osteoblast differentiation and induces osteocyte apoptosis. PLoS ONE 6, e27487 (2011).

56. de Vernejoul, M. C. & Kornak, U. Heritable sclerosing bone disorders: presentation and new molecular mechanisms. Ann. NY Acad. Sci. 1192, 269–277 (2010).

57. Li, X. et al. Targeted deletion of the sclerostin gene in mice results in increased bone formation and bone strength. J. Bone Miner. Res. 23, 860–869 (2008).

58. Teitelbaum, S. L. Bone resorption by osteoclasts. Science 289, 1504–1508 (2000).

59. Wagner, E. F. Bone development and inflammatory disease is regulated by AP-1 (Fos/Jun). Ann. Rheum. Dis. 69 (Suppl. 1), 86–88 (2010).

60. Schonthaler, H. B., Guinea-Viniegra, J. & Wagner, E. F. Targeting inflammation by modulating the Jun/AP-1 pathway. Ann. Rheum. Dis. 70 (Suppl. 1), 109–112 (2011).

61. Shaulian, E. & Karin, M. AP-1 as a regulator of cell life and death. Nature Cell Biol. 4, E131–E136 (2002).

62. Koga, T. et al. Costimulatory signals mediated by the ITAM motif cooperate with RANKL for bone homeostasis. Nature 428, 758–763 (2004).

63. Takayanagi, H. et al. Induction and activation of the transcription factor NFATc1 (NFAT2) integrate RANKL signaling in terminal differentiation of osteoclasts. Dev. Cell 3, 889–901 (2002).

64. Delaisse, J. M. et al. Matrix metalloproteinases (MMP) and cathepsin K contribute differently to osteoclastic activities. Microsc. Res. Tech. 61, 504–513 (2003).

65. Supanchart, C. & Kornak, U. Ion channels and transporters in osteoclasts. Arch. Biochem. Biophys. 473, 161–165 (2008).

66. Goldring, S. R., Roelke, M. S., Petrison, K. K. & Bhan, A. K. Human giant cell tumors of bone identification and characterization of cell types. J. Clin. Invest. 79, 483–491 (1987).

67. Dougall, W. C. et al. RANK is essential for osteoclast and lymph node development. Genes Dev. 13, 2412–2424 (1999).

68. Hofbauer, L. C. et al. The roles of osteoprotegerin and osteoprotegerin ligand in the paracrine regulation of bone resorption. J. Bone Miner. Res. 15, 2–12 (2000).

69. Wiktor-Jedrzejczak, W. et al. Total absence of colony-stimulating factor 1 in the macrophage-deficient osteopetrotic (op/op) mouse. Proc. Natl Acad. Sci. USA 87, 4828–4832 (1990).

70. Grigoriadis, A. E. et al. c-Fos: a key regulator of osteoclast-macrophage lineage determination and bone remodeling. Science 266, 443–448 (1994).This study reveals the essential role of the transcription factor AP1, and particularly its component FOS, in osteoclastogenesis.

71. Binder, N. B. et al. Estrogen-dependent and C-C chemokine receptor-2-dependent pathways determine osteoclast behavior in osteoporosis. Nature Med. 15, 417–424 (2009).

72. Kim, M. S., Day, C. J. & Morrison, N. A. MCP-1 is induced by receptor activator of nuclear factor-κB ligand, promotes human osteoclast fusion, and rescues granulocyte macrophage colony-stimulating factor suppression of osteoclast formation. J. Biol. Chem. 280, 16163–16169 (2005).

73. Kong, Y. Y. et al. Activated T cells regulate bone loss and joint destruction in adjuvant arthritis through osteoprotegerin ligand. Nature 402, 304–309 (1999).This publication links the adaptive immune system to osteoclast activation and inflammatory bone destruction.

74. Roggia, C. et al. Up-regulation of TNF-producing T cells in the bone marrow: a key mechanism by which estrogen deficiency induces bone loss in vivo. Proc. Natl Acad. Sci. USA 98, 13960–13965 (2001).This study provides evidence of the role of TNF in osteoporosis induced by oestrogen deficiency.

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 247

© 2012 Macmillan Publishers Limited. All rights reserved

75. Kapinas, K. & Delany, A. M. MicroRNA biogenesis and regulation of bone remodeling. Arthritis Res. Ther. 13, 220 (2011).

76. Sugatani, T., Vacher, J. & Hruska, K. A. A microRNA expression signature of osteoclastogenesis. Blood 117, 3648–3657 (2011).

77. Kapinas, K., Kessler, C., Ricks, T., Gronowicz, G. & Delany, A. M. miR-29 modulates Wnt signaling in human osteoblasts through a positive feedback loop. J. Biol. Chem. 285, 25221–25231 (2010).

78. Sugatani, T. & Hruska, K. A. Impaired micro-RNA pathways diminish osteoclast differentiation and function. J. Biol. Chem. 284, 4667–4678 (2009).

79. Gough, A. K., Lilley, J., Eyre, S., Holder R. L. & Emery P. Generalised bone loss in patients with early rheumatoid arthritis. Lancet 344, 23–27 (1994).

80. Romas, E. Bone loss in inflammatory arthritis: mechanisms and therapeutic approaches with bisphosphonates. Best Pract. Res. Clin. Rheumatol. 19, 1065–1079 (2005).

81. Roldan, J. F., del, Rincón, I. & Escalante, A. Loss of cortical bone from the metacarpal diaphysis in patients with rheumatoid arthritis: independent effects of systemic inflammation and glucocorticoids. J. Rheumatol. 33, 508–516 (2006).

82. Gravallese, E. M. et al. Identification of cell types responsible for bone resorption in rheumatoid arthritis and juvenile rheumatoid arthritis. Am. J. Pathol. 152, 943–951 (1998).This study reveals the role of synovial-derived osteoclasts in local bone damage (erosions) in patients with rheumatoid arthritis.

83. Garcia-Carrasco, M. et al. Osteoporosis in patients with systemic lupus erythematosus. Isr. Med. Assoc. J. 11, 486–491 (2009).

84. Grisar, J. et al. Ankylosing spondylitis, psoriatic arthritis, and reactive arthritis show increased bone resorption, but differ with regard to bone formation. J. Rheumatol. 29, 1430–1436 (2002).

85. Ali, T., Lam, D., Bronze, M. S. & Humphrey, M. B. Osteoporosis in inflammatory bowel disease. Am. J. Med. 122, 599–604 (2009).

86. Paganelli, M. et al. Inflammation is the main determinant of low bone mineral density in pediatric inflammatory bowel disease. Inflamm. Bowel Dis. 13, 416–423 (2007).

87. Bianchi, M. L. & Bardella, M. T. Bone in celiac disease. Osteoporos. Int. 19, 1705–1716 (2008).

88. Cashman, K. D. Altered bone metabolism in inflammatory disease: role for nutrition. Proc. Nutr. Soc. 67, 196–205 (2008).

89. Shead, E. F., Haworth, C. S., Barker, H., Bilton, D. & Compston, J. E. Osteoclast function, bone turnover and inflammatory cytokines during infective exacerbations of cystic fibrosis. J. Cyst. Fibros. 9, 93–98 (2010).

90. Dam, T. T., Harrison, S., Fink, H. A., Ramsdell, J. & Barrett-Connor, E. Bone mineral density and fractures in older men with chronic obstructive pulmonary disease or asthma. Osteoporos. Int. 21, 1341–1349 (2010).

91. Yoshihara, A., Seida, Y., Hanada, N. & Miyazaki, H. A longitudinal study of the relationship between periodontal disease and bone mineral density in community-dwelling older adults. J. Clin. Periodontol. 31, 680–684 (2004).

92. Redlich, K. et al. Repair of local bone erosions and reversal of systemic bone loss upon therapy with anti-tumor necrosis factor in combination with osteoprotegerin or parathyroid hormone in tumor necrosis factor-mediated arthritis. Am. J. Pathol. 164, 543–555 (2004).

93. Lin, C. L., Moniz, C., Chambers, T. J. & Chow, J. W. Colitis causes bone loss in rats through suppression of bone formation. Gastroenterology 111, 1263–1271 (1996).

94. Mattila, K. J., Valle, M. S., Nieminen, M. S., Valtonen, V. V. & Hietaniemi, K. L. Dental infections and coronary atherosclerosis. Atherosclerosis 103, 205–211 (1993).

95. Reddy, M. S. Oral osteoporosis: is there an association between periodontitis and osteoporosis? Compend. Contin. Educ. Dent. 23, 21–28 (2002).

96. Kawai, T. et al. B and T lymphocytes are the primary sources of RANKL in the bone resorptive lesion of periodontal disease. Am. J. Pathol. 169, 987–998 (2006).

97. Teng, Y. T. et al. Functional human T-cell immunity and osteoprotegerin ligand control alveolar bone destruction in periodontal infection. J. Clin. Invest. 106, R59–R67 (2000).

98. Korn, T., Bettelli, E., Oukka, M. & Kuchroo, V. K. IL-17 and Th17 cells. Annu. Rev. Immunol. 27, 485–517 (2009).

99. Hundorfean, G., Neurath, M. F. & Mudter, J. Functional relevance of T helper 17 (Th17) cells and the IL-17 cytokine family in inflammatory bowel disease. Inflamm. Bowel Dis. 18, 180–186 (2012).

100. Wong, P. K. et al. Interleukin-6 modulates production of T lymphocyte-derived cytokines in antigen-induced arthritis and drives inflammation-induced osteoclastogenesis. Arthritis Rheum. 54, 158–168 (2006).This study provides evidence of the role of IL-6 and associated IL-17 production on osteoclast generation.

101. Dinarello, C. A. Interleukin-1 and interleukin-1 antagonism. Blood 77, 1627–1652 (1991).

102. Tracey K. J. & Cerami, A. Tumor necrosis factor: a pleiotropic cytokine and therapeutic target. Annu. Rev. Med. 45, 491–503 (1994).

103. Dinarello, C. A. et al. Tumor necrosis factor (cachectin) is an endogeneous pyrogen and induces production of interleukin 1. J. Exp. Med. 163, 1433–1450 (1986).

104. Naka, T., Nishimoto, N. & Kishimoto, T. The paradigm of IL-6: from basic science to medicine. Arthritis Res. 4 (Suppl. 3), 233–242 (2002).

105. Kishimoto, T. IL-6: from its discovery to clinical applications. Int. Immunol. 22, 347–352 (2010).

106. Wallach, D. et al. Tumor necrosis factor receptor and Fas signaling mechanisms. Annu. Rev. Immunol. 17, 331–367 (1999).

107. Goeddel, D. V. Signal transduction by tumor necrosis factor: the Parker B. Francis Lectureship. Chest 116 (Suppl. 1), 69–73 (1999).

108. Sethi, G., Sung, B. & Aggarwal, B. B. TNF: a master switch for inflammation to cancer. Front. Biosci. 13, 5094–5107 (2008).

109. Ikejima, T., Okusawa, S., Ghezzi, P., van der Meer, J. W. & Dinarello, C. A. Interleukin-1 induces tumor necrosis factor (TNF) in human peripheral blood mononuclear cells in vitro and a circulating TNF-like activity in rabbits. J. Infect. Dis. 162, 215–223 (1990).

110. Legendre, F., Bogdanowicz, P., Boumediene, K. & Pujol, J. P. Role of interleukin 6 (IL-6)/IL-6R-induced signal tranducers and activators of transcription and mitogen-activated protein kinase/extracellular signal-related kinase in upregulation of matrix metalloproteinase and ADAMTS gene expression in articular chondrocytes. J. Rheumatol. 32, 1307–1316 (2005).

111. Rowan, A. D. et al. Synergistic effects of glyco -protein 130 binding cytokines in combination with interleukin-1 on cartilage collagen breakdown. Arthritis Rheum. 44, 1620–1632 (2001).

112. Shingu, M. et al. The effects of cytokines on metalloproteinase inhibitors (TIMP) and collagenase production by human chondrocytes and TIMP production by synovial cells and endothelial cells. Clin. Exp. Immunol. 94, 145–149 (1993).

113. Solis-Herruzo, J. A. et al. Interleukin-6 increases rat metalloproteinase-13 gene expression through stimulation of activator protein 1 transcription factor in cultured fibroblasts. J. Biol. Chem. 274, 30919–30926 (1999).

114. Richards, C., Gauldie, J. & Baumann, H. Cytokine control of acute phase protein expression. Eur. Cytokine Netw. 2, 89–98 (1991).

115. Dinarello, C. A. Interleukin-1 and the pathogenesis of the acute-phase response. N. Engl. J. Med. 311, 1413–1418 (1984).

116. Kushner, I. Regulation of the acute phase response by cytokines. Perspect. Biol. Med. 36, 611–622 (1993).

117. Andus, T., Geiger, T., Hirano, T., Kishimoto, T. & Heinrich, P. C. Action of recombinant human interleukin 6, interleukin 1β and tumor necrosis factor α on the mRNA induction of acute-phase proteins. Eur. J. Immunol. 18, 739–746 (1988).

118. Zhang, Y. H., Lin, J. X. & Vilcek, J. Interleukin-6 induction by tumor necrosis factor and interleukin-1 in human fibroblasts involves activation of a nuclear factor binding to a kappa B-like sequence. Mol. Cell Biol. 10, 3818–3823 (1990).

119. Everaerdt, B., Brouckaert, P. & Fiers, W. Recombinant IL-1 receptor antagonist protects against TNF-induced lethality in mice. J. Immunol. 152, 5041–5049 (1994).

120. Devlin, R. D., Reddy, S. V., Savino, R., Ciliberto, G. & Roodman G. D. IL-6 mediates the effects of IL-1 or TNF, but not PTHrP or 1,25(OH)2D3, on osteoclast-like cell formation in normal human bone marrow cultures. J. Bone Miner. Res. 13, 393–399 (1998).

121. Ma, T. et al. Human interleukin-1-induced murine osteoclastogenesis is dependent on RANKL, but independent of TNF-α. Cytokine 26, 138–144 (2004).

122. Lee, Z. H. et al. IL-1α stimulation of osteoclast survival through the PI 3-kinase/Akt and ERK pathways. J. Biochem. 131, 161–166 (2002).

123. Kobayashi, K. et al. Tumor necrosis factor α stimulates osteoclast differentiation by a mechanism independent of the ODF/RANKL–RANK interaction. J. Exp. Med. 191, 275–286 (2000).

124. Kotake, S. et al. Interleukin-6 and soluble interleukin-6 receptors in the synovial fluids from rheumatoid arthritis patients are responsible for osteoclast-like cell formation. J. Bone Miner. Res. 11, 88–95 1996.

125. De Benedetti, F. et al. Impaired skeletal development in interleukin-6-transgenic mice: a model for the impact of chronic inflammation on the growing skeletal system. Arthritis Rheum. 54, 3551–3563 (2006).

126. Lam, J. et al. TNF-α induces osteoclastogenesis by direct stimulation of macrophages exposed to permissive levels of RANK ligand. J. Clin. Invest. 106, 1481–1488 (2000).This report reveals the role of the pro-inflammatory cytokine TNF in osteoclastogenesis.

127. Abu-Amer, Y., Ross, F. P., Edwards, J. & Teitelbaum S. L. Lipopolysaccharide-stimulated osteoclastogenesis is mediated by tumor necrosis factor via its p55 receptor. J. Clin. Invest. 100, 1557–1565 (1997).

128. Bluml, S. et al. Antiinflammatory effects of tumor necrosis factor on hematopoietic cells in a murine model of erosive arthritis. Arthritis Rheum. 62, 1608–1619 (2010).This study dissects the osteoclastogenic effects of TNF signals, showing that these effects are mediated via activation of TNF receptor 1 rather than TNF receptor 2, and that the latter may have protective effects.

129. Zhang, Y. H., Heulsmann, A., Tondravi, M. M., Mukherjee, A. & Abu-Amer, Y. Tumor necrosis factor-α (TNF) stimulates RANKL-induced osteoclastogenesis via coupling of TNF type 1 receptor and RANK signaling pathways. J. Biol. Chem. 276, 563–568 (2001).

130. Ravetch, J. V. & Bolland, S. IgG Fc receptors. Annu. Rev. Immunol. 19, 275–290 (2001).

131. Zou, W. et al. Syk, c-Src, the αvβ3 integrin, and ITAM immunoreceptors, in concert, regulate osteoclastic bone resorption. J. Cell Biol. 176, 877–888 (2007).

132. Vignery, A. Macrophage fusion: the making of osteoclasts and giant cells. J. Exp. Med. 202, 337–340 (2005).

133. Zvaifler, N. J. Rheumatoid synovitis. An extravascular immune complex disease. Arthritis Rheum. 17, 297–305 (1974).

134. Weissmann, G. Rheumatoid arthritis and systemic lupus erythematosus as immune complex diseases. Bull. NYU Hosp. Jt Dis. 67, 251–253 (2009).

135. Brown, E. E., Edberg, J. C. & Kimberly, R. P. Fc receptor genes and the systemic lupus erythematosus diathesis. Autoimmunity 40, 567–581 (2007).

136. Mocsai, A. et al. The immunomodulatory adapter proteins DAP12 and Fc receptor γ-chain (FcRγ) regulate development of functional osteoclasts through the Syk tyrosine kinase. Proc. Natl Acad. Sci. USA 101, 6158–6163 (2004).

137. Scott, D. L., Symmons, D. P., Coulton, B. L. & Popert, A. J. Long-term outcome of treating rheumatoid arthritis: results after 20 years. Lancet 1, 1108–1111 (1987).

138. Aringer, M. & Smolen, J. S. Therapeutic blockade of TNF in patients with SLE — promising or crazy? Autoimmunity Rev. 18 May 2011 (doi:10.1016/j.autrev.2011.05.001. 2011).

139. Sokolove, J., Zhao, X., Chandra, P. E. & Robinson, W. H. Immune complexes containing citrullinated fibrinogen costimulate macrophages via Toll-like receptor 4 and Fcγ receptor. Arthritis Rheum. 63, 53–62 (2011).

140. Guerne, P. A., Carson, D. A. & Lotz, M. IL-6 production by human articular chondrocytes. Modulation of its synthesis by cytokines, growth factors, and hormones in vitro. J. Immunol. 144, 499–505 (1990).

141. van Gool, J., van Vugt, H., Helle, M. & Aarden, L. A. The relation among stress, adrenalin, interleukin 6 and acute phase proteins in the rat. Clin. Immunol. Immunopathol. 57, 200–210 (1990).

142. Chrousos, G. P. The hypothalamic–pituitary–adrenal axis and immune-mediated inflammation. N. Engl. J. Med. 332, 1351–1362 (1995).

R E V I E W S

248 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved

143. Perlstein, R. S., Whitnall, M. H., Abrams, J. S., Mougey, E. H. & Neta, R. Synergistic roles of interleukin-6, interleukin-1, and tumor necrosis factor in the adrenocorticotropin response to bacterial lipopolysaccharide in vivo. Endocrinology 132, 946–952 (1993).

144. Rachon, D., Mysliwska, J., Suchecka-Rachon, K., Wieckiewicz, J. & Mysliwski, A. Effects of oestrogen deprivation on interleukin-6 production by peripheral blood mononuclear cells of postmenopausal women. J. Endocrinol. 172, 387–395 (2002).

145. Poli, V. et al. Interleukin-6 deficient mice are protected from bone loss caused by estrogen depletion. EMBO J. 13, 1189–1196 (1994).

146. Girasole, G., Passeri, G., Jilka, R. L. & Manolagas, S. C. Interleukin-11: a new cytokine critical for osteoclast development. J. Clin. Invest. 93, 1516–1524 (1994).

147. Okamoto, H. et al. The synovial expression and serum levels of interleukin-6, interleukin-11, leukemia inhibitory factor, and oncostatin M in rheumatoid arthritis. Arthritis Rheum. 40, 1096–1105 (1997).

148. Sasai, M. et al. Delayed onset and reduced severity of collagen-induced arthritis in interleukin-6-deficient mice. Arthritis Rheum. 42, 1635–1643 (1999).

149. Gilbert, L. et al. Expression of the osteoblast differentiation factor RUNX2 (Cbfa1/AML3/Pebp2αA) is inhibited by tumor necrosis factor-α. J. Biol. Chem. 277, 2695–2701 (2002).

150. Gilbert, L. C., Rubin, J. & Nanes, M. S. The p55 TNF receptor mediates TNF inhibition of osteoblast differentiation independently of apoptosis. Am. J. Physiol. Endocrinol. Metab. 288, E1011–E1018 (2005).

151. Kaneki, H. et al. Tumor necrosis factor promotes Runx2 degradation through up-regulation of Smurf1 and Smurf2 in osteoblasts. J. Biol. Chem. 281, 4326–4333 (2006).

152. Abbas, S., Zhang, Y. H., Clohisy, J. C. & Abu-Amer, Y. Tumor necrosis factor-α inhibits pre-osteoblast differentiation through its type-1 receptor. Cytokine 22, 33–41 (2003).

153. Mukai, T. et al. TNF-α inhibits BMP-induced osteoblast differentiation through activating SAPK/JNK signaling. Biochem. Biophys. Res. Commun. 356, 1004–1010 (2007).

154. Ding, J. et al. TNF-α and IL-1β inhibit RUNX2 and collagen expression but increase alkaline phosphatase activity and mineralization in human mesenchymal stem cells. Life Sci. 84, 499–504 (2009).

155. Hughes, F. J. & Howells, G. L. Interleukin-6 inhibits bone formation in vitro. Bone Miner. 21, 21–28 (1993).

156. Hughes, F. J. & Howells, G. L. Interleukin-11 inhibits bone formation in vitro. Calcif. Tissue Int. 53, 362–364 (1993).

157. Li, X. et al. Sclerostin binds to LRP5/6 and antagonizes canonical Wnt signaling. J. Biol. Chem. 280, 19883–19887 (2005).

158. Winkler, D. G. et al. Osteocyte control of bone formation via sclerostin, a novel BMP antagonist. EMBO J. 22, 6267–6276 (2003).

159. Mason, J. J. & Williams, B. O. SOST and DKK: antagonists of LRP family signaling as targets for treating bone disease. J. Osteoporos. 2010, 460120 (2010).

160. Viswanathan, A. & Sylvester, F. A. Chronic pediatric inflammatory diseases: effects on bone. Rev. Endocr. Metab. Disord. 9, 107–122 (2008).

161. Capriles, V. D., Martini, L. A. & Areas, J. A. Metabolic osteopathy in celiac disease: importance of a gluten-free diet. Nutr. Rev. 67, 599–606 (2009).

162. Payne, J. B. & Golub, L. M. Using tetracyclines to treat osteoporotic/osteopenic bone loss: from the basic science laboratory to the clinic. Pharmacol. Res. 63, 121–129 (2011).

163. Elborn, J. S. How can we prevent multisystem complications of cystic fibrosis? Semin. Respir. Crit. Care Med. 28, 303–311 (2007).

164. Haugeberg, G., Conaghan, P. G., Quinn, M. & Emery, P. Bone loss in patients with active early rheumatoid arthritis: infliximab and methotrexate compared with methotrexate treatment alone. Explorative analysis from a 12-month randomised, double-blind, placebo-controlled study. Ann. Rheum. Dis. 68, 1898–1901 (2009).

165. Strand, V. & Simon, L. S. Low dose glucocorticoids in early rheumatoid arthritis. Clin. Exp. Rheumatol. 21 (Suppl. 31), 186–190 (2003).

166. Kim, H. J. et al. Glucocorticoids suppress bone formation via the osteoclast. J. Clin. Invest. 116, 2152–2160 (2006).

167. Hoes, J. N. et al. EULAR evidence-based recommendations on the management of systemic glucocorticoid therapy in rheumatic diseases. Ann. Rheum. Dis. 66, 1560–1567 (2007).

168. Teitelbaum, S. L., Seton, M. P. & Saag, K. G. Should bisphosphonates be used for long-term treatment of glucocorticoid-induced osteoporosis? Arthritis Rheum. 63, 325–328 (2011).

169. Lange, U., Teichmann, J., Muller-Ladner, U. & Strunk, J. Increase in bone mineral density of patients with rheumatoid arthritis treated with anti-TNF-α antibody: a prospective open-label pilot study. Rheumatology (Oxford) 44, 1546–1548 (2005).

170. Bernstein, M., Irwin, S. & Greenberg, G. R. Maintenance infliximab treatment is associated with improved bone mineral density in Crohn’s disease. Am. J. Gastroenterol. 100, 2031–2035 (2005).

171. Veerappan, S. G., O’Morain, C. A., Daly, J. S. & Ryan, B. M. Review article: the effects of antitumour necrosis factor-α on bone metabolism in inflammatory bowel disease. Aliment. Pharmacol. Ther. 33, 1261–1272 (2011).

172. Kremer, J. M. et al. Tocilizumab inhibits structural joint damage in rheumatoid arthritis patients with inadequate responses to methotrexate: results from the double-blind treatment phase of a randomized placebo-controlled trial of tocilizumab safety and prevention of structural joint damage at one year. Arthritis Rheum. 63, 609–621 (2011).

173. Ito, H. et al. A pilot randomized trial of a human anti-interleukin-6 receptor monoclonal antibody in active Crohn’s disease. Gastroenterology 126, 989–996 (2004).

174. Hsu, B., Zhou, B., Smolen, J. S. & Weinblatt, M. Proof-of-concept for CNTO 136, a human anti-interleukin-6 monoclonal antibody, in a multicenter, randomized, double-blind, placebo-controlled, Phase 2 study in patients with active rheumatoid arthritis despite methotrexate therapy. Ann. Rheum. Dis. 70 (Suppl. 3), 459 (2011).

175. Mease, P. et al. Inhibition of IL-6 with ALD518 improves disease activity in rheumatoid arthritis in a randomized, double-blind, placebo-controlled, dose ranging Phase 2 clinical trial. Ann. Rheum. Dis. 69 (Suppl. 3), 98 (2011).

176. Hickling, M., Golor, G., Juillon, A., Shaw, S. & Kretsos, K. Safety and pharmacokinetics of CDP6038, an anti-IL-6 monoclonal antibody, administered by subcutaneous injection and intravenous infusion to healthy male volunteers: a Phase 1 study. Ann. Rheum. Dis. 70 (Suppl. 3), 471 (2011).

177. Nam, J. L. et al. Current evidence for the management of rheumatoid arthritis with biological disease-modifying antirheumatic drugs: a systematic literature review informing the EULAR recommendations for the management of RA. Ann. Rheum. Dis. 69, 976–986 (2010).

178. Baltzer, A. W. et al. Gene therapy for osteoporosis: evaluation in a murine ovariectomy model. Gene Ther. 8, 1770–1776 (2001).

179. Genovese, M. et al. Secukinumab (ain457) showed a rapid decrease of disease activity in patients with active rheumatoid arthritis including those with high inflammatory burden. Ann. Rheum. Dis. 70 (Suppl. 3), 472 (2011).

180. Bluml, S. et al. Essential role of microRNA-155 in the pathogenesis of autoimmune arthritis in mice. Arthritis Rheum. 63, 1281–1288 (2011).

181. Yang, M. & Mattes, J. Discovery, biology and therapeutic potential of RNA interference, microRNA and antagomirs. Pharmacol. Ther. 117, 94–104 (2008).

182. Smolen J. S. & Steiner, G. Therapeutic strategies for rheumatoid arthritis. Nature Rev. Drug Discov. 2, 473–488 (2003).

183. Westhovens, R. et al. Clinical efficacy and safety of abatacept in methotrexate-naive patients with early rheumatoid arthritis and poor prognostic factors. Ann. Rheum. Dis. 68, 1870–1877 (2009).

184. Tak, P. P. et al. Inhibition of joint damage and improved clinical outcomes with rituximab plus methotrexate in early active rheumatoid arthritis: the IMAGE trial. Ann. Rheum. Dis. 70, 39–46 (2011).

185. Hein, G. et al. Influence of rituximab on markers of bone remodeling in patients with rheumatoid arthritis: a prospective open-label pilot study. Rheumatol Int. 31, 269–272 (2011).

186. van der Heijde, D. et al. Tofacitinib (CP-690,550), an oral Janus kinase inhibitor, in combination with methotrexate reduced the progression of structural damage in patients with rheumatoid arthritis: a 24-month Phase 3 study. [meeting abstract] Arthritis Rheum. 63 (Suppl. 10), 2592 (2011).

187. Weinblatt, M. E. et al. An oral spleen tyrosine kinase (Syk) inhibitor for rheumatoid arthritis. N. Engl. J. Med. 363, 1303–1312 (2010).

188. Massey, D. C., Bredin, F. & Parkes, M. Use of sirolimus (rapamycin) to treat refractory Crohn’s disease. Gut 57, 1294–1296 (2008).

189. Cejka, D. et al. Mammalian target of rapamycin signaling is crucial for joint destruction in experimental arthritis and is activated in osteoclasts from patients with rheumatoid arthritis. Arthritis Rheum. 62, 2294–2302 (2010).

190. Westenfeld, R. et al. Impact of sirolimus, tacrolimus and mycophenolate mofetil on osteoclastogenesis —implications for post-transplantation bone disease. Nephrol. Dial. Transplant. 12, 4115–4123 (2011).

191. Dobashi, Y., Watanabe, Y., Miwa, C., Suzuki, S. & Koyama, S. Mammalian target of rapamycin: a central node of complex signaling cascades. Int. J. Clin. Exp. Pathol. 4, 476–495 (2011).

192. Markman, B., Dienstmann, R. & Tabernero, J. Targeting the PI3K/Akt/mTOR pathway — beyond rapalogs. Oncotarget 1, 530–543 (2010).

193. Kim, J. E. & Chen, J. Regulation of peroxisome proliferator-activated receptor-γ activity by mammalian target of rapamycin and amino acids in adipogenesis. Diabetes 53, 2748–2756 (2004).

194. Wan, Y. PPARγ in bone homeostasis. Trends Endocrinol. Metab. 21, 722–728 (2010).

195. Glantschnig, H., Fisher, J. E., Wesolowski, G., Rodan, G. A. & Reszka, A. A. M-CSF, TNFα and RANK ligand promote osteoclast survival by signaling through mTOR/S6 kinase. Cell Death. Differ. 10, 1165–1177 (2003).

196. Garrett, I. R. et al. Selective inhibitors of the osteoblast proteasome stimulate bone formation in vivo and in vitro. J. Clin. Invest. 111, 1771–1782 (2003).

197. Chen, D., Frezza, M., Schmitt, S., Kanwar, J. & Dou, P. Bortezomib as the first proteasome inhibitor anti-cancer drug: current status and future perspectives. Curr. Cancer Drug Targets 11, 239–253 (2011).

198. von Metzler, I. et al. Bortezomib inhibits human osteoclastogenesis. Leukemia 21, 2025–2034 (2007).

199. Zavrski, I. et al. Proteasome inhibitors abrogate osteoclast differentiation and osteoclast function. Biochem. Biophys. Res. Commun. 333, 200–205 (2005).

200. Hongming, H. & Jian, H. Bortezomib inhibits maturation and function of osteoclasts from PBMCs of patients with multiple myeloma by downregulating TRAF6. Leuk. Res. 33, 115–122 (2009).

201. Schmidt, N. et al. Targeting the proteasome: partial inhibition of the proteasome by bortezomib or deletion of the immunosubunit LMP7 attenuates experimental colitis. Gut 59, 896–906 (2010).

202. Lee, S. W., Kim, J. H., Park, Y. B. & Lee, S. K. Bortezomib attenuates murine collagen-induced arthritis. Ann. Rheum. Dis. 68, 1761–1767 (2009).

203. Yannaki, E. et al. The proteasome inhibitor bortezomib drastically affects inflammation and bone disease in adjuvant-induced arthritis in rats. Arthritis Rheum. 62, 3277–3288 (2010).

204. Russell, R. G. Bisphosphonates: the first 40 years. Bone 49, 2–19 (2011).

205. Lin, C. L., Moniz, C. & Chow, J. W. Treatment with fluoride or bisphosphonates prevents bone loss associated with colitis in the rat. Calcif. Tissue Int. 67, 373–377 (2000).

206. Eggelmeijer, F. et al. Increased bone mass with pamidronate treatment in rheumatoid arthritis. Results of a three-year randomized, double-blind trial. Arthritis Rheum. 39, 396–402 (1996).

207. Kwak, H. B. et al. Risedronate directly inhibits osteoclast differentiation and inflammatory bone loss. Biol. Pharm. Bull. 32, 1193–1198 (2009).

208. Watts, N. B. & Diab, D. L. Long-term use of bisphosphonates in osteoporosis. J. Clin. Endocrinol. Metab. 95, 1555–1565 (2010).

209. McClung, M. R. et al. Denosumab in postmenopausal women with low bone mineral density. N. Engl. J. Med. 354, 821–831 (2006).

210. Cummings, S. R. et al. Denosumab for prevention of fractures in postmenopausal women with osteoporosis. N. Engl. J. Med. 361, 756–765 (2009).

R E V I E W S

NATURE REVIEWS | DRUG DISCOVERY VOLUME 11 | MARCH 2012 | 249

© 2012 Macmillan Publishers Limited. All rights reserved

211. Cohen, S. B. et al. Denosumab treatment effects on structural damage, bone mineral density, and bone turnover in rheumatoid arthritis: A twelve-month, multicenter, randomized, double-blind, placebo-controlled, Phase II clinical trial. Arthritis Rheum. 58, 1299–1309 (2008).The results from this trial provide proof of concept that selectively targeting osteoclasts prevents local inflammatory bone loss but does not affect inflammation or cartilage damage.

212. Dore, R. K. et al. Effects of denosumab on bone mineral density and bone turnover in patients with rheumatoid arthritis receiving concurrent glucocorticoids or bisphosphonates. Ann. Rheum. Dis. 69, 872–875 (2010).

213. Adler, R. A. & Gill, R. S. Clinical utility of denosumab for treatment of bone loss in men and women. Clin. Interv. Aging 6, 119–124 (2011).

214. Saad, F. et al. Incidence, risk factors, and outcomes of osteonecrosis of the jaw: integrated analysis from three blinded active-controlled Phase III trials in cancer patients with bone metastases. Ann. Oncol. 10 Oct 2011 (doi:10.1093/annonc/mdr435).

215. Takahata, M., Awad, H. A., O’Keefe, R. J., Bukata, S. V. & Schwarz, E. M. Endogenous tissue engineering: PTH therapy for skeletal repair. Cell Tissue Res. 31 May 2011 (doi:10.1007/s00441-011-1188-4).

216. Vincent, C. et al. Pro-inflammatory cytokines TNF-related weak inducer of apoptosis (TWEAK) and TNFα induce the mitogen-activated protein kinase (MAPK)-dependent expression of sclerostin in human osteoblasts. J. Bone Miner. Res. 24, 1434–1449 (2009).

217. Heiland, G. R. et al. Neutralisation of Dkk-1 protects from systemic bone loss during inflammation and reduces sclerostin expression. Ann. Rheum. Dis. 69, 2152–2159 (2010).This study reveals the efficacy of blocking DKK1 on inflammatory bone loss in an experimental model.

218. Betts, A. M. et al. The application of target information and preclinical pharmacokinetic/pharmacodynamic modeling in predicting clinical doses of a Dickkopf-1 antibody for osteoporosis. J. Pharmacol. Exp. Ther. 333, 2–13 (2010).

219. Ominsky, M. S. et al. Inhibition of sclerostin by monoclonal antibody enhances bone healing and improves bone density and strength of nonfractured bones. J. Bone Miner. Res. 26, 1012–1021 (2011).

220. Tian, X. et al. Treatment with a sclerostin antibody increases cancellous bone formation and bone mass regardless of marrow composition in adult female rats. Bone 47, 529–533 (2010).

221. Li, X. et al. Sclerostin antibody treatment increases bone formation, bone mass, and bone strength in a rat model of postmenopausal osteoporosis. J. Bone Miner. Res. 24, 578–588 (2009).

222. Rickels, M. R., Zhang, X., Mumm, S. & Whyte, M. P. Oropharyngeal skeletal disease accompanying high bone mass and novel LRP5 mutation. J. Bone Miner. Res. 20, 878–885 (2005).

223. Zeckey, C., Hildebrand, F., Frink, M. & Krettek, C. Heterotopic ossifications following implant surgery — epidemiology, therapeutical approaches and current concepts. Semin. Immunopathol. 33, 273–286 (2011).

224. Boraiah, S., Paul, O., Hawkes, D., Wickham, M. & Lorich, D. G. Complications of recombinant human BMP-2 for treating complex tibial plateau fractures: a preliminary report. Clin. Orthop. Relat. Res. 467, 3257–3262 (2009).

225. Balemans, W., Cleiren, E., Siebers, U., Horst, J. & Van Hul, W. A generalized skeletal hyperostosis in two siblings caused by a novel mutation in the SOST gene. Bone 36, 943–947 (2005).

226. Hoff, M. et al. Adalimumab reduces hand bone loss in rheumatoid arthritis independent of clinical response: subanalysis of the PREMIER study. BMC Musculoskelet. Disord. 12, 54 (2011).

227. Eklund, S. A. & Burt, B. A. Risk factors for total tooth loss in the United States; longitudinal analysis of national data. J. Public Health Dent. 54, 5–14 (1994).

228. Gladman, D. D. et al. Consensus on a core set of domains for psoriatic arthritis. J. Rheumatol. 34, 1167–1170 (2007).

229. van der Heijde, D. et al. How to report radiographic data in randomized clinical trials in rheumatoid arthritis: guidelines from a roundtable discussion. Arthritis Rheum. 47, 215–218 (2002).

230. Smolen, J. S., Aletaha, D., Koeller, M., Weisman, M. & Emery, P. New therapies for the treatment of rheumatoid arthritis. Lancet 370, 1861–1874 (2007).

231. Smolen, J. S. et al. Evidence of radiographic benefit of infliximab plus methotrexate in rheumatoid arthritis patients who had no clinical improvement: a detailed subanalysis of the ATTRACT trial. Arthritis Rheum. 52, 1020–1030 (2005).

232. Smolen, J. S. et al. The need for prognosticators in rheumatoid arthritis. Biological and clinical markers — where are we now? Arthritis Res. Ther. 10, 208 (2008).

233. Smolen, J. S., Martinez-Avila, J. C. & Aletaha, D. Tocilizumab inhibits progression of joint damage in rheumatoid arthritis irrespective of its anti-inflammatory effects: disassociation of the link between inflammation and destruction. Ann. Rheum. Dis. 25 Nov 2011 (doi:10.1136/annrheumdis- 2011-200395).

234. Mattes, J., Yang, M. & Foster, P. S. Regulation of microRNA by antagomirs: a new class of pharmacological antagonists for the specific regulation of gene function? Am. J. Respir. Cell. Mol. Biol. 36, 8–12 (2007).

235. Sims, N. A. et al. Targeting osteoclasts with zoledronic acid prevents bone destruction in collagen-induced arthritis. Arthritis Rheum. 50, 2338–2346 (2004).

236. Herrak, P. et al. Zoledronic acid protects against local and systemic bone loss in tumor necrosis factor-mediated arthritis. Arthritis Rheum. 50, 2327–2337 (2004).

237. Jarrett, S. J. et al. Preliminary evidence for a structural benefit of the new bisphosphonate zoledronic acid in early rheumatoid arthritis. Arthritis Rheum. 54, 1410–1414 (2006).

238. Schett, G. et al. Osteoprotegerin protects against generalized bone loss in tumor necrosis factor-transgenic mice. Arthritis Rheum. 48, 2042–2051 (2003).

239. Schett, G. et al. Additive bone-protective effects of anabolic treatment when used in conjunction with RANKL and tumor necrosis factor inhibition in two rat arthritis models. Arthritis Rheum. 52, 1604–1611 (2005).

240. Aletaha, D., Funovits, J. & Smolen, J. S. Physical disability in rheumatoid arthritis is associated with cartilage damage rather than bone destruction. Ann. Rheum. Dis. 70, 733–739 (2011).

241. van der Heijde, D. et al. Expert agreement confirms that negative changes in hand and foot radiographs are a surrogate for repair in patients with rheumatoid arthritis. Arthritis Res. Ther. 9, R62 (2007).

242. Machold, K. P. et al. Very recent onset arthritis — clinical, laboratory and radiological findings during the first year of disease. J. Rheumatol. 29, 2278–2287 (2002).

243. Plant, M. J., Jones, P. W., Saklatvala, J., Ollier, W. E. & Dawes, P. T. Patterns of radiological progression in rheumatoid arthritis: results of an 8 year prospective study. J. Rheumatol. 25, 417–426 (1998).

244. Van der Heijde, D. M. Joint erosions and patients with early rheumatoid arthritis. Br. J. Rheumatol. 34, (Suppl. 2), 74–78 (1995).

245. Redlich, K. et al. Tumor necrosis factor α-mediated joint destruction is inhibited by targeting osteoclasts with osteoprotegerin. Arthritis Rheum. 46, 785–792 (2002).

246. Muller-Ladner, U. et al. Synovial fibroblasts of patients with rheumatoid arthritis attach to and invade normal human cartilage when engrafted into SCID mice. Am. J. Pathol. 149, 1607–1615 (1996).

247. Colucci, S. et al. Lymphocytes and synovial fluid fibroblasts support osteoclastogenesis through RANKL, TNFα, and IL-7 in an in vitro model derived from human psoriatic arthritis. J. Pathol. 212, 47–55 (2007).

248. Redlich, K. et al. Osteoclasts are essential for TNF-α-mediated joint destruction. J. Clin. Invest. 110, 1419–1427 (2002).

249. Partsch, G. et al. Highly increased levels of tumor necrosis factor-α and other proinflammatory cytokines in psoriatic arthritis synovial fluid. J. Rheumatol. 24, 518–523 (1997).

250. Gortz, B. et al. Arthritis induces lymphocytic bone marrow inflammation and endosteal bone formation. J. Bone Miner. Res. 19, 990–998 (2004).

251. Jimenez-Boj, E. et al. Bone erosions and bone marrow edema as defined by magnetic resonance imaging reflect true bone marrow inflammation in rheumatoid arthritis. Arthritis Rheum. 56, 1118–1124 (2007).

252. Arnett, F. C. et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum. 31, 315–324 (1988).

253. Shimizu, S., Shiozawa, S., Shiozawa, K., Imura, S. & Fujita, T. Quantitative histologic studies on the pathogenesis of periarticular osteoporosis in rheumatoid arthritis. Arthritis Rheum. 28, 25–31 (1985).

254. Mansson, B., Gulfe, A., Geborek, P., Heinegard, D. & Saxne, T. Release of cartilage and bone macromolecules into synovial fluid: differences between psoriatic arthritis and rheumatoid arthritis. Ann. Rheum. Dis. 60, 27–31 (2001).

255. Furie, R. et al. A Phase III, randomized, placebo-controlled study of belimumab, a monoclonal antibody that inhibits B lymphocyte stimulator, in patients with systemic lupus erythematosus. Arthritis Rheum. 63, 3918–3930 (2011).

256. Hsu, B., Sheng, S., Smolen, J. & Weinblatt, M. Results from a 2-part, proof-of-concept, dose-ranging, randomized, double-blind, placebo-controlled, Phase 2 study of sirukumab, a human anti- interleukin-6 monoclonal antibody, in active rheumatoid arthritis patients despite methotrexate therapy. [meeting abstract] Arthritis Rheum. 63 (Suppl. 10), 2631 (2011).

257. Griffiths, C. E. et al. Comparison of ustekinumab and etanercept for moderate-to-severe psoriasis. N. Engl. J. Med. 362, 118–128 (2010).

258. Taylor, P. C. et al. Ofatumumab, a fully human anti-CD20 monoclonal antibody, in biological-naive, rheumatoid arthritis patients with an inadequate response to methotrexate: a randomised, double-blind, placebo-controlled clinical trial. Ann. Rheum. Dis. 70, 2119–2125 (2011).

259. van Vollenhoven, R. F., Kinnman, N., Vincent, E., Wax, S. & Bathon, J. Atacicept in patients with rheumatoid arthritis and an inadequate response to methotrexate: results of a Phase II, randomized, placebo-controlled trial. Arthritis Rheum. 63, 1782–1792 (2011).

260. Hartung, H. P. & Kieseier, B. C. Atacicept: targeting B cells in multiple sclerosis. Ther. Adv. Neurol. Disord. 3, 205–216 (2010).

261. Fulciniti, M. et al. Anti-DKK1 mAb (BHQ880) as a potential therapeutic agent for multiple myeloma. Blood 114, 371–379 (2009).

262. Cusick, T. et al. Odanacatib treatment increases hip bone mass and cortical thickness by preserving endocortical bone formation and stimulating periosteal bone formation in the ovariectomized adult rhesus monkey. J Bone Miner. Res. 23 Nov 2011 (doi: 10.1002/jbmr.1477).

263. Eisman, J. A. et al. Odanacatib in the treatment of postmenopausal women with low bone mineral density: three-year continued therapy and resolution of effect. J. Bone Miner. Res. 26, 242–251 (2011).

AcknowledgementsThe authors would like to thank B. Niederreiter for expert support.

Competing interests statementThe authors declare competing financial interests: see Web version for details.

FURTHER INFORMATIONAmgen website — 21 April 2011 press release: http://www.amgen.com/media/media_pr_detail.jsp?releaseID=1553039

ALL LINKS ARE ACTIVE IN THE ONLINE PDF

R E V I E W S

250 | MARCH 2012 | VOLUME 11 www.nature.com/reviews/drugdisc

© 2012 Macmillan Publishers Limited. All rights reserved