Perspective Genomic biomarkers in predictive

Genomic biomarkers in predictive medicine. An interim analysis

Keywords: biomarkers; early detection; ; ; translational research

Current genomics and biotechnology markers and the limited use of such denotes the prevalence of the high-risk promise the development of biomarkers markers outside research. Failure to allele and RR denotes the relative risk of to predict individual risk, enable focus also results in misleading claims disease for a subject with high-risk versus early detection of disease, and improve for early detection biomarkers based on standard risk allele. Most have diagnostic classification to better inform studies with inappropriate controls. Here, sufficiently low prevalences that the OR individualized treatment. I discuss these I shall discuss separately several broad and RR are essentially equal. From this objectives, commenting on progress to categories of intended use as illustrated in formula one can show that the PPV is no date and obstacles to future success. The Fig 1 (summarized in Box 1). greater than RRp. So if the RR is 1.22 and discussion mainly uses examples from the disease prevalence is 5% (p ¼ 0.05), where the nature of the disease then the absolute risk of developing the has expedited genomic approaches for Biomarkers of Disease Risk disease for a subject with a high-risk developing biomarkers. Many of the allele is no greater than 6.1%. If the lessons being learned in oncology, how- The development of genetic biomarkers RR were 5, then the absolute risk of ever, should be applicable to other for predicting risk of disease in indivi- developing disease for a subject with a chronic diseases. duals has had limited success to date. high-risk allele could be as large as 10% Biomarkers are biological measure- Numerous large whole-genome associa- for a disease with 5% prevalence in the ments that can be used to predict risk tion studies (Ioannidis et al, 2010) population. of disease, to enable early detection of involving thousands of patients have There are several possible explana- disease, to improve treatment selection been conducted for many chronic dis- tions for the ‘low penetrance’ (low PPV) and to monitor the outcome of therapeu- eases. These studies have genotyped of the identified genetic loci. For oncol- tic interventions. One major motivation cases and controls in order to identify ogy, a major reason is the genetic of the was the germ-line polymorphisms that put indi- heterogeneity of most cancers. For exam- identification and development of such viduals at higher risk for developing a ple, estrogen receptor negative and biomarkers for ‘personalized, preventive specific disease. Many genetic loci have estrogen receptor positive and predictive medicine’. Although the been identified as statistically significant are different in terms of the somatic sequencing of the human genome has and, in some cases, are providing valu- that characterize them, as well had profound impacts on biomedical able leads for understanding the biologi- as with regard to natural history and research in many other fields, and while cal basis of the diseases. However, the responsiveness to treatment. From most it is still too early to fully assess its impact strength of the associations is often far perspectives they appear to be different on biomarker development (Lander, too weak to provide much value for diseases and lumping them together in 2011), I will provide an interim analysis counselling individuals (Bloss et al, searching for polymorphisms of disease and identify some of the roadblocks to 2011). Ioannidis et al reviewed 56 GWAS susceptibility is problematic (Kraft and progress. reporting 92 statistically significant asso- Halman, 2010) Indeed, the success of One of the greatest problems in the ciations between cancer phenotypes and early studies that identi- development and validation of biomar- genetic variants and found a median per- fied the highly penetrant BRCA1 locus kers is the ambiguity of the term and the allele odds ratio (OR) of 1.22 with an owed, in large part, to restricting the failure to recognize that biomarker valid- interquartile range of 1.15–1.36. The studies to cases with early onset breast ity means fitness for intended use. An absolute risk of disease for a subject cancer. Many other chronic diseases are enormous amount of resources is simply with a high-risk allele can be considered phenotypically and molecularly hetero- wasted because researchers do not focus the ‘positive predictive value’ of the geneous. They are also probably geneti- clearly on an intended use. This is seen, genetic test and can be expressed cally heterogeneous and thus very diffi- for example, in the gap between the PPV ¼ RRp/(1 þ g(RR 1)), where p cult to study with broad genome-wide enormous literature on prognostic bio- denotes the prevalence of the disease, g association studies.

www.embomolmed.org EMBO Mol Med 3, 1–7 ß 2011 EMBO Molecular Medicine 1 Perspective Genomic biomarkers in predictive medicine

BOX 1: Progress in genomic biomarker development Biomarkers are biological measure- mentioned. In this paper, I provide a short life-threatening is very challenging ments that can be used for a variety and personal assessment of the progress and the validation of such biomarkers of purposes, including identifying achieved in these areas of genomic requires very large randomized individuals who are at high risk of biomarker development. screening trials. The development of developing a disease, detecting dis- It is indicated that (i) progress has biomarkers for personalizing treat- ease early at a stage when it is been slow in personalized risk prediction ment selection, particularly in oncol- treatable and diagnostic classification and in early detection; (ii) genome-wide ogy, has seen greater progress. Key for personalized treatment based on a association studies are more likely to bottlenecks that limit progress in the biological characterization of the provide leads for understanding the translation of discoveries in genomics disease of each individual patient. pathogenesis of diseases than useful to biomarkers and treatments The sequencing of the human genome information on personalized risk assess- that reduce mortality and morbidity has provided an important body of ment; and (iii) development of bio- from chronic diseases are also information for the development of markers sufficiently sensitive and specific discussed. biomarkers for all of the purposes for early detection of diseases that will be

Other possible explanations for the listic to expect that individual poly- Early Detection Biomarkers failure of finding highly penetrant sus- morphisms will have substantial expla- ceptibility is the fact that chronic natory or predictive power in elucidating Many diseases can be more effectively diseases are caused by the combined these relationships. For other chronic treated at an early stage. Most solid effects of multiple genetic polymorph- diseases, however, GWAS may be more tumours have a long sub-clinical course isms and/or that chronic diseases are essential for generating leads concerning prior to diagnosis and hence, there caused by a combination of genetic and the biology of the disease. In most should be substantial opportunity for environmental causes. The greatest cases, however, these leads must be early detection. There has been little potential value of genome-wide associa- followed by fine mapping of the regions success to date, however, in developing tion studies is to shed light on the of the detected polymorphisms and and validating early detection biomar- biological basis of the disease. For then years of biological investigations kers with medical utility. Early detection oncology, this may be less essential since to understand the relevance of the research has been severely hindered by cancers are in large part diseases of DNA disease alleles. It is probably too early the use of poor study methodology. Many modification and the tumour genomes to evaluate the impact of GWAS on cancer biomarkers are ‘discovered’ by can be directly evaluated. Of course medical utility. It is clear, however, that comparing levels of candidate proteins in interpreting tumour genomes to find the initial expectations of easy and direct tumour tissue, collected at diagnosis, to the mutations, which are key to oncogen- translation of GWAS findings to patient normal tissue. Numerous ‘discoveries’ esis, is difficult (Ledford, 2010) and benefit were unrealistic. GWAS studies have been published and publicized GWAS could provide additional useful in heterogeneous diseases such as cancer based on such evidence. Finding such a information. Most cancers, however, could be improved by evaluating difference, however, is very weak evi- result from complex sequences of associations with biologically meaning- dence that the marker will be useful for somatic mutations, which interact with ful subsets of patients with central early detection. A recent publication each other to influence tumour evolution review of cases to ensure accuracy of evaluated 28 candidate biomarkers using (Ashworth et al, 2011). It seems unrea- classification. serum samples obtained from subjects in

Figure 1. Broad categories of intended use of biomarkers.

2 ß 2011 EMBO Molecular Medicine EMBO Mol Med 3, 1–7 www.embomolmed.org Perspective Richard Simon

a randomized screening trial. None of the 28 showed early detection performance BOX 2: Statistics of early detection either individually or in combination as Sens ¼ probability marker is positive at screening T years before diagnosis of good as or better than the traditional lethal tumour. CA-125 (Mai et al, 2011). This prompted Spec ¼ probability marker is negative at screening in absence of lethal tumour. a commentary in Nature titled ‘Missing p ¼ prevalence of lethal tumours in population (0.01 means 1/100 patients the Mark’. (Bucher, 2011) One way to screened). improve early detection research would DCure ¼ improvement in cure proportion of lethal tumours detected T years be to perform genome-wide or proteome- before diagnosis. wide discovery using serum samples PPV ¼ probability of lethal tumour given marker is positive at screening archived from retrospective longitudinal psens PPV ¼ cohorts rather than samples from patients psens þð1pÞð1specÞ at diagnosis. Such ‘phase 3’ studies have NSaved ¼ increase in lives saved per 10,000 individuals screened. been generally reserved for validation of NSaved ¼ 10,000 p sens DCure. candidate markers discovered in the quick NFalseþ¼number of false positive tests per 10,000 individuals screened. and easy phases 1 and 2 studies based on NFalseþ¼10,000 (1 p)(1 spec). diagnostic samples. If one of the candidates in the ovarian p Sens Spec T DCure PPV NSaved NFlaseþ cancer study had been found elevated in 0.01 0.95 0.95 1 0.10 0.16 9.5 495 serum samples prior to diagnosis based on 0.01 0.90 0.90 5 0.20 0.087 18 990 the retrospective analysis of the screening trials, this would have given an indication of the lead-time achievable with that marker, but because the study was retro- order to conduct effective population improvement in probability of cure by spective it would not provide information screening, we need a test with very high detecting disease T years before clinical about whether such detection would specificity and to restrict screening to diagnosis. The table uses hypothetical reduce mortality or morbidity from the high-risk populations. improvements in cure probabilities of disease. One would not know whether the There is also the issue of what do we 0.10 and 0.20 corresponding to lead times cases detected had localized disease at the mean by ‘disease’. We not only need very of 1 or 5 years. It also assumes that a time of detection. In general, although high specificity, we need very high longer lead time results in slightly lower perhaps not the case for ovarian cancer, specificity for detecting the form of the sensitivity and specificity of the test. without a prospective randomized screen- disease which is life threatening. But, the Using the assumptions shown in the ing trial one would not know what earlier the point at which the disease is table, a screening test that provides a 1- proportion of the detected cases might detected, the more difficult it may be to year lead-time for disease detection not represent a tumour that would be life distinguish a life-threatening cancer from results in 9.5 lives saved and 495 false threatening within the lifetime of the an nodule which may be indolent during positives per 10,000 screened. With a test patient (Etzioni et al, 2003). the patient’s lifetime, given that the early providing a 5-year lead-time, 18 lives are The identification of early detection steps of oncogenesis are variable and saved and there are 990 false positive markers that are sufficiently specific for stochastic. For example, a large propor- tests per 10,000 patients screened. These use in population screening is challen- tion of individuals have a BCR-ABL numbers depend on the assumed 10 and ging. Some important aspects of the fusion protein detectable in their blood, 20% increases in cure rate with 1- and 5- statistics of early detection screening is yet only a small percentage of those year lead-times, respectively, and on the shown in Box 2. If the prevalence of the develop chronic myelogenous leukaemia assumed 1% disease prevalence. disease is denoted by p and the sensitivity (CML). Not all early lesions may progress Progress in the identification of early and specificity are denoted by sens to invasive cancer and yet early detection detection biomarkers has been very lim- and spec, respectively, then the prob- may encourage treatment with serious ited. Most of the focus has been on funding ability that a test positive case has the adverse effects. the application of new technologies to disease (positive predictive value) equals In addition to showing the PPV result- identification of tumour markers. Each p (sens)/(p (sens) þ (1 p)(1 spec)). ing from values of prevalence, sensitivity new round of technology development has When sensitivity and specificity are both and specificity, Box 2 also shows the generated overly optimistic claims based 0.95 and the prevalence of disease in the expected number of lives saved by on poor research design and methodology. population is 1%, the positive predictive screening and the expected number of In order to expedite progress and to make value is only about 0.16 – only 16% of the false positives for every 10,000 indivi- more efficient use of limited resources, test positive individuals will actually duals screened. The expected number of new strategies for biomarker discovery are have the disease. The remaining 84% lives saved is the product of the pre- required that make greater use of archived will be possibly subjected to unnecessary valence of lethal tumours in the popula- samples from longitudinal studies (Zhu and invasive follow-up procedures. In tion, the sensitivity of the test and the et al, 2011). Improved policies for funding

www.embomolmed.org EMBO Mol Med 3, 1–7 ß 2011 EMBO Molecular Medicine 3 Perspective Genomic biomarkers in predictive medicine

early detection research are required the study on node negative, hormone drugs in broad patient populations with that place greater emphasis on proper receptor positive patients who received molecularly uncharacterized tumours is research design as well as use of state-of- hormonal treatment but no chemotherapy. no longer based on solid science and is the-art technology. Even with a candidate Most prognostic studies use a convenience unlikely to be successful. Development of marker that has been demonstrated to sample of heterogeneous patients and cancer therapeutics with companion diag- enable earlier disease detection, however, develop signatures that have no therapeu- nostics is the dominant theme today. The demonstrating that it has medical utility tic relevance (Subramanian & Simon, predictive biomarkers that are used to for reducing mortality from the disease 2010). The objectives of genomic prog- guide treatment selection for molecularly requires very large and expensive rando- nostic studies are generally not clearly targeted drugs are mostly based on the mized screening trials. considered. The purpose of prognostic or amplification of a single . signatures like the Oncotype Dx recurrence The genes are usually either the target of score and the MammaPrint signature is to the drug or a non-target gene with Treatment Selection Biomarkers help patients and in making relevance to the pathways involving the informed therapeutic decisions. Useful target gene. Protein over-expression has In some areas of therapeutics, such as tools like Oncotype Dx and MammaPrint tended to be a less reliable basis for oncology, predictive and prognostic bio- should not be criticized because they do credentialing gene targets or for develop- markers have been effectively developed not provide biological insight into the ment of predictive biomarkers. Some of to help guide treatment decisions. For disease. the drugs being developed, such as kinase example, the OncotypeDx recurrence ‘Predictive biomarkers’ indicate which inhibitors, have multiple targets and score and MammaPrint signature are patients are most likely or unlikely to development of a predictive biomarker used to determine whether a woman benefit from a specific treatment. For is made difficult by uncertainty in the with node-negative hormone-receptor- example, estrogen receptor expression molecular basis for anti-tumour effect. positive breast cancer has a sufficiently levels have been used for many years to The extreme clonal heterogeneity of most good prognosis with local treatment and select patients for anti-estrogen hormo- tumours is also a challenge for proper adjuvant hormonal treatment that she nal treatment and HER2 over-expression evaluation of predictive biomarkers in does not require cytotoxic chemotherapy or amplification is widely used to select oncology. In the future, patients at major (Paik et al, 2004; Van de Vijver et al, patients for treatment with anti-HER2 cancer centres will likely undergo whole 2002). In oncology, there is an enormous drugs. EGFR mutation is used to select exome sequencing of multiple samples literature of claims for improved prog- patients for small-molecule EGFR inhibi- from each accessible tumour site in order nostic factors that have never found tors in non-small-cell lung cancer and to develop an optimal therapeutic strategy. clinical application. This gap between KRAS mutation to de-select patients from Gene expression signatures have less research and application has accelerated with anti-EGFR antibodies in frequently been used as predictive bio- with the development of gene expression advanced colorectal cancer. Discovery markers for new drugs. Use of a genomic profiling. Identifying a prognostic impact of the BCR-ABL fusion protein in CML led mutation or amplification of a gene of a gene mutation may suggest an to the development of imatinib and related to the mechanism of action of important role of that gene product as a the analysis of mutations of that gene the drug as a predictive biomarker is more molecular target as was the case for HER2 determines second-line treatments scientifically satisfying. Gene expression in breast cancer. But the numerous gene (Drucker et al, 2001). The identification profiling has been frequently used for expression-based signatures that have of the EML4-ALK fusion gene in a small developing prognostic signatures but been developed based on prognosis or subset of patients with non-small-cell much less frequently for developing pre- on clustering expression profiles (Perou lung cancer led to using a kinase inhibitor dictive markers of benefit from specific et al, 2000) have had limited utility for that targets that gene with extremely treatments. This may be in part because either elucidating underlying biology or promising results (Kwak et al, 2010). frozen tissue samples have rarely been informing treatment decision-making. Similarly, the discovery of a single point archived from patients in clinical trials Developing a prognostic gene expres- mutation in the BRAF gene in 60% of suitable for the development of predictive sion signature is not likely to be useful patients with metastatic led to signatures. When archived tissues are unless the signature is developed with an the development of an inhibitor with available from randomized trials, the intended use clearly in mind from the start. increased specificity for the mutated tissues are generally formalin fixed and That intended use should drive the selec- protein with extremely promising results paraffin preserved (FFPP). Because of tion of cases and the interpretation of (Flaherty et al, 2010). RNA degradation in FFPP tissues, such results. In order to identify a gene In fact, most oncology drug develop- samples, until recently, were not suitable expression signature like the Oncotype ment today is driven by molecular targets for microarray gene expression analysis. DX recurrence score for use in determining because cancers of most primary sites are Development of prognostic signatures for which breast cancer patients with node- heterogeneous with regard to oncogenesis a heterogeneous set of samples, which negative estrogen receptor positive disease and sensitivity to treatment. Conse- may be derived from different types of have such good prognosis that they do not quently, the blockbuster strategy is unli- progenitor cells, also is easier than devel- require chemotherapy, one needs to focus kely to work for most cancers; testing new oping a predictive signature for a set of

4 ß 2011 EMBO Molecular Medicine EMBO Mol Med 3, 1–7 www.embomolmed.org Perspective Richard Simon

patients homogeneous enough to have (Hawk et al, 2008; Nathan and Varmus, of multiple samples from individual been included in a single clinical trial. 2000; Sung et al, 2003). The key scientific tumours will enable us to characterize Unfortunately, progress in develop- and structural roadblocks have received the clonal heterogeneity of each tumour ment of new therapeutics in oncology less attention, however, and they influ- (Jones et al, 2008; Navin et al, 2011). has primarily been restricted to mono- ence our ability to use genomic technol- With sufficient sequencing power, we clonal antibodies and small-molecule ogies for developing useful biomarkers as can phylogenetically reconstruct the kinase inhibitors and improvements in well as for developing effective thera- evolution of individual tumours and therapeutics have only infrequently peutics. Some of these roadblocks are identify the founder mutations (Campbell resulted in cures. Many of the most discussed below and some suggested et al, 2008). These founder mutations important molecular targets, such as the approaches for improving translational represent the initial rate-limiting genomic tumour suppressor genes p53 and Rb, are research are depicted in Fig 2. changes that enabled the developing not effectively amenable to drug interven- First, basic research does not go far tumour to grow to a size in which tion. Using synthetic lethality approaches enough in identifying the key steps in the numerous subsequent mutations could to target the effects of key mutations in development and pathogenesis of most develop in a non rate-limiting manner such genes holds great promise and will chronic diseases in order to enable (Simon, 2010). Because these founder likely be the dominant theme for future translational research to proceed effec- mutations are present in all sub-clones drug development in oncology (Ashworth tively. Even in oncology, our very limited and because subsequent mutations et al, 2011; Haber et al, 2011). understanding of the oncogenesis of developed in the context of these muta- Personalization of therapy is only cancer is a major hurdle to effective tions, they may represent the key mole- effective if the therapeutic strategies for translational research (Simon, 2010). cular targets for that individual tumour. the identified subsets of patients are Once basic research identifies a key step Even with identification of tumour spe- effective. For instance, the point muta- of oncogenesis and a druggable molecular cific founder mutations, it may be tion that causes sickle-cell anaemia was target, the pharmaceutical and biotech- necessary to treat early and with combi- identified more than 60 years ago, but nology industries are often adept at nations of drugs selected based on that discovery has not yet led to effective developing potent inhibitors of that target. knowledge of the networks in which treatment for that disease (Pauling et al, We still do not fully understand the the founder genes participate. Imatinib is 1949). The development of predictive development and progression of any type highly effective in treating CML if treat- biomarkers for guiding treatment for of cancer even if, in rare cases such as ment begins prior to transition towards other diseases has in many cases lagged CML, our knowledge of oncogenesis has blast crisis. The blast crisis of CML may behind developments in oncology, which been sufficient to develop effective treat- represent a mutational meltdown that has the advantage that it is a disease of ments. Development of more effective also occurs in solid tumours. In CML we DNA and hence much information about treatments likely requires the character- have the benefit of detecting the disease tumour sensitivity to treatment options ization of key founder mutations that before that meltdown occurs. The onco- can be gained by using the plethora of drive the pathogenesis of the individual gene addiction to founder mutations that new whole-genome technologies. Yet, tumour, understanding the networks in tumours sometimes exhibit can be dis- there is substantial research in using which these genes are involved and sipated by later mutations (Jonkers & high-throughput genomic and proteomic treating early enough with combinations Berns, 2004; Weinstein, 2002). Ashworth technologies to identify biomarkers for of drugs to overcome resistant sub- et al (Ashworth et al, 2011) provide a many other diseases. clones. Deep single molecule sequencing penetrating discussion of the possible

Bridging the Gap between Basic Genomic Research and Patient Benefit

The gap between basic research and clinical benefit has been termed the ‘valley of ’ in the popular press (Begley, 2008; Butler, 2008). Much attention has been devoted to the numer- ous infrastructure and financial complex- ities of translational research including regulatory issues, human subject approvals, intellectual property issues, lack of funding, lack of patients, lack of training for -investigators and a fragmented research infrastructure Figure 2. Suggested approaches to improve translational research.

www.embomolmed.org EMBO Mol Med 3, 1–7 ß 2011 EMBO Molecular Medicine 5 Perspective Genomic biomarkers in predictive medicine

basis of oncogene addiction and other these problems are scientifically difficult, » The scientific challenges kinds of gene interactions, which may be it is that existing mechanisms for support- of understanding the therapeutically exploitable but elucida- ing research and most existing research pathogenesis of chronic tion of such interactions is at an early organizations do not provide an effective stage of development. framework for a concerted effort to tackle diseases to the extent that For many diseases the challenges in these problems. Consequently, the road- we can effectively prevent, understanding disease pathogenesis are blocks remain, in some cases for decades detect, diagnose and treat even greater than for cancer. Germline as in the point mutation causing sickle-cell them are substantial. polymorphisms may provide leads, but anaemia (Pauling et al, 1949). Nonetheless, there is an the process of elucidating the biology of Bridging the broad gaps between basic the disease to find key molecular targets research and clinical benefit is likely to enormous amount of talent for treatment is often painstakingly slow. require major changes in the interactions available to meet these A second area of scientific bottleneck between industry and academia, and challenges. « is related to the fact that there is little more public funding of ‘mini-Manhattan’ focus outside of industry on identifying project teams to overcome key road- key breakthroughs in basic research and blocks. The investigator-initiated frame- funding prioritized programs to translate work is highly effective for basic research overcoming the key bottlenecks to pro- those breakthroughs into products that and has yielded major biomedical dis- gress. We need to treat the biomedical benefit patients. Much of the public coveries. It has not, however, elucidated research enterprise as a system that needs funding for translational research is the basic steps in the development and to be optimized to achieve our objectives. devoted to making biological measure- pathogenesis of many major chronic This may require creating new kinds of ments on patient tissues in an attempt to diseases nor has it provided adequate organizations to better foster innovation, understand the nature of the disease or identification of key targets to enable encourage transdisciplinary fertilization providing infrastructure to help investi- effective translational research. It is not and ensure that resources are optimally gators bring their personal research to the necessarily the most appropriate frame- allocated for overcoming the key obsta- clinic. After the V600E point mutation in work for bridging the ‘valley of death’. cles to progress. the BRAF gene was discovered to be The US National Science Foundation and present in about 60% of patients with Defense Advance Research Projects malignant melanoma, there was little Agency have utilized strategies involving Acknowledgements drug discovery activity among NIH investigator-initiated approaches to pub- The author is appreciative of the valu- grantees to exploit this finding. Fortu- licly prioritized objectives in some of able comments of the referees. nately, BRAF was druggable with stan- their programs. dard chemistry and two companies To tackle the scientific bottlenecks to The author declares that he has no developed extremely promising specific key translational opportunities, new conflict of interest. inhibitors of the mutated form of the horizontally integrated organizations of gene. When the translational challenges experts in basic, clinical and quantitative are more difficult, or when the financial research are needed. Such organizations incentives are either too limited or in can play an important role in training the conflict with industry concerns about next generation of biomedical research- market segmentation, however, such ers to work in settings without silos or complete and uncoordinated dependence hierarchies of disciplines where creative on industry for therapeutic development basic scientists, clinicians and quantita- may not server the public well. tive scientists can be part of a single team For example, mutations of the p53 and working together daily to focus on Rb tumour suppressor genes are preva- bridging the gap separating basic research lent and important in many types of from products that benefit patients. Richard Simon cancer but their gene products are not The scientific challenges of under- Richard Simon is Chief of the Biometric Research easily druggable; neither industry nor standing the pathogenesis of chronic Branch in the US National Cancer Institute and academic research have developed suc- diseases to the extent that we can head of the section on Molecular Statistics cessful approaches for exploiting these effectively prevent, detect, diagnose and and Bioinformatics. This article represents his mutations. Developing feasible pharma- treat them are substantial. Nonetheless, personal opinion on the current status and key cologic ways of interfering with mutated there is an enormous amount of talent roadblocks to progress in developing genomic p53 or Rb in tumours are difficult, long- available to meet these challenges. To biomarkers to reduce mortality and morbidity from cancer and other chronic diseases. term, high-risk endeavours that are not take advantage of the opportunities

adequately addressed either by industry provided by the genomic, biotechnology Biometric Research Branch, National Cancer or by the culture of the NIH investigator- and information revolutions, however, Institute, Bethesda, MD, USA initiated grant system. It is not just that we need to better focus this talent on E-mail: [email protected]

6 ß 2011 EMBO Molecular Medicine EMBO Mol Med 3, 1–7 www.embomolmed.org Perspective Richard Simon

References Hawk ET, Matrisian LM, Nelson WG, Dorfman GS, Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, Ashworth A, Lord CJ, Reis-Filho JS (2011) Genetic Stevens L, Kwok J, Viner J, Hautala J, Grad O Baehner FL, Walker MG, Watson D, Park T, et al interactions in cancer progression and (2008) The translational research working (2004) A multigene assay to predict recurrence treatment. Cell 145: 30-38 group developmental pathways: Introduction of tamoxifen-treated, node-negative breast Begley S (2008) We fought cancer ... and cancer and overview. Clin Cancer Res 14: 5664- cancer. N Engl J Med 351: 2817-2826 won. Newsweek, Sep 15, 43-66 5671 Pauling L, Itano HA, Singer SJ, Wells IC (1949) Bloss CS, Schork NJ, Topol EJ (2011) Effect of Ioannidis JPA, Castaldi P, Evangelou E (2010) A Sickle cell anemia, a molecular disease. Science direct-to-consumer genomewide profiling to compendium of genome-wide associations for 25: 543-548 assess disease risk. N Engl J Med 364: 524-534 cancer: critical synopsis and reappraisal. J Natl Perou CM, Sorlie T, Eisen MB, van De Rijn M, Jeffrey Bucher L (2011) Missing the mark: Why is it so Cancer Inst 102: 846-858 SS, Reese CA, Pollak JR, Ross DT, Jonsen H, Akslen hard to find a test to predict cancer? Nature Jones S, Chen WD, Parmigiani G, Diehl F, LA, et al (2000) Molecular portraits of human 471: 428-432 Beerenwinkel N, Antal T, Traulsen A, Nowak MA, breast tumors. Nature 406: 747-752 Bucher L (2011) Missing the mark: Why is it so Siegel C, Velculescu VE, et al (2008) Comparative Simon R (2010) Translational research in hard to find a test to predict cancer? Nature lesion sequencing provides insights into tumor oncology: key bottlenecks and new paradigms. 471: 428-432 evolution. Proc Natl Acad Sci 105: 4283-4288 Expert Rev Mol Med 12: e32 Butler D (2008) Crossing the valley of death. Jonkers J, Berns A (2004) Oncogene addiction. Subramanian J, Simon R (2010) Gene expression Nature 453: 840-842 Cancer Cell 6: 535-538 based prognostic signatures in lung cancer: Campbell PJ, Pleasance ED, Stephens PJ, Dicks E, Kraft P, Halman CA (2010) GWAS identifies a Ready for clinical use? J Natl Cancer Inst 102: Rance R, Goodhead I, Follows GA, Green AR, common breast cancer risk allele among BRCA1 464-474 Futreal PA, Stratton MR (2008) Subclonal carriers. Nat Genet 42: 819-820 Sung NS, Crowley WF, Genel M, Salber P, Sandy L, phylogenetic structures in cancer revealed by Kwak EL, Bang YJ, Camidge R, Shaw AT, Solomon B, Herwood LM, Johnson SB, Catanese V, Tilson H, ultra-deep sequencing. Proc Natl Acad Sci 105: Maki RG, Ou SHI, Dezube BJ, Janne PA, Costa DB, Getz K (2003) Central challenges facing the 13081-13086 et al (2010) Anaplastic lymphoma kinase national clinical research enterprise. JAmMed Drucker B, Talpaz M, Resta DJ, Peng B, Buchdunger inhibition in non-small-cell lung cancer. N Engl J Assoc 289: 1278-1287 E, Ford JM, Lydon NB, Kantarjian H, Capdeville R, Med 363: 693-703 Van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart Ohno-Jones S, et al (2001) Efficacy and safety of Lander ES (2011) Initial impact of the sequencing AAM, Voskuil DW, Schreiber GJ, Peterse JL, a specific inhibitor of the BCRABL tyrosine of the human genome. Nature 470: 187-213 Roberts C, Marton MJ, et al (2002) A gene- kinase in chronic myeloid leukemia. N Engl J Ledford H (2010) The cancer genome challenge. expression signature as a predictor of survival in Med 344: 1031-1037 Nature 464: 972-974 breast cancer. N Engl J Med 347: 1999-2009 Etzioni R, Urban N, Ramsey S, Mcintosh M, Mai PL, Wentzensen N, Greene MH (2011) Weinstein IB (2002) Addiction to oncogenes—the Schwartz S, Reid B, Radich J, Anderson G, Challenges related to developing serum-based Achilles heel of cancer. Science 297: 63-64 Hartwell L (2003) The case for early detection. biomarkers for early ovarian cancer detection. Zhu CS, Pinsky PF, Cramer DW, Ransohoff DF, Nat Rev Cancer 3: 1-10 Cancer Prev Res 4: 303-306 Hartge P, Pfeiffer RF, Urban N, Mor G, Bast RC, Flaherty KT, Puzanov I, Kim KB, Ribas A, Mcarthur Nathan DG, Varmus HE (2000) The National Moor LE, et al (2011) A framework for evaluating GA, Sosman JA, O’dwyer PJ, Lee RJ, Grippo JF, Institutes of Health and clinical research: a biomarkers for early detction: validation of Nolop K, et al (2010) Inhibition of mutated, progress report. Nat Med 6: 1201-1204 biomarker panels for ovarian cancer. Cancer activated BRAF in metastatic melanoma. N Engl Navin N, Kendall J, Troge J, Andrews P, Rodgers L, Prev Res 4: 375-383 J Med 363: 809-819 mcindoo J, Cook K, Stepansky A, Levy D, Esposito Haber DA, Gray NS, Baselga J (2011) The evolving D, et al (2011) Tumor evolution inferred by war on cancer. Cell 145: 19-24 single-cell sequencing. Nature 472: 90-94 DOI 10.1002/emmm.201100153

www.embomolmed.org EMBO Mol Med 3, 1–7 ß 2011 EMBO Molecular Medicine 7