Nature Reviews Cancer | AOP, published online 15 May 2014; doi:10.1038/nrc3729
OPINION
‘Toxgnostics’: an unmet need in cancer medicine David Church, Rachel Kerr, Enric Domingo, Dan Rosmarin, Claire Palles, Kevin Maskell, Ian Tomlinson and David Kerr
Abstract | If we were to summarize the rationale that underpins medical oncology in a Latin aphorism, it might be ‘veneno ergo sum’; that is, I poison, therefore I am. The burden of chemotherapy-associated toxicity is well recognized, but we have relatively few tools that increase the precision of anticancer drug prescribing. We propose a shift in emphasis from the focussed study of polymorphisms in drug metabolic pathways in small sets of patients to broader agnostic analyses to systematically correlate germline genetic variants with adverse events in large, well-defined cancer populations. Thus, we propose the new science of ‘toxgnostics’ (that is, the systematic, agnostic study of genetic predictors of toxicity from anticancer therapy). If it were not for the great variability among individuals, medicine might as well be a science and not an art. Sir William Osler,
Principles and Practice of Medicine, 1st Edition, Appleton and Company, New York, 1892
Although substantial advances have been made in the systemic therapy of solid tumours over the past two decades, most patients obtain only modest benefit from treatment, whereas toxicity is common and often associated with considerable morbid ity and mortality. For example, adjuvant chemotherapy following resection of stage II colorectal cancer benefits only ~4% of patients, whereas 30–40% of those treated will suffer common toxicity criteria (CTC) grade 3–4, and 0.5–1% will suffer fatal toxi cities1. Targeted therapies are also associated with significant adverse event profiles from both on- and off-target effects2. All drugs are defined by their therapeutic ratios (efficacy/toxicity). However, this is particularly true of antineoplastic agents, for which the dose–response curves are usually steep and the therapeutic window is correspondingly narrow. Although the past 10 years has seen much effort to improve the risk/benefit ratio of anticancer therapy, this has mostly focused on maximising the benefit side of the equation, with atten tion directed towards the identification of tumour biomarkers with prognostic impor tance that allow patient stratification according to the likelihood of recurrence or progression (see REFS 3–5 for examples), or companion diagnostics that predict
response to therapy 6–10 (reviewed in REF. 11). Despite thousands of reports that encompass hundreds of biomarkers in common solid tumours, the number that has entered rou tine clinical practice is depressingly small12–14. The reasons for this shortfall are manifold, but the high risk of false positives that result from multiple testing in small samples and the lack of validation cohorts are important contributors14,15. Recent biomarker studies using large (n >1,000 patients), adequately statistically powered clinical trials go some way to addressing this, but they remain lim ited by the need to focus on a few biomark ers defined a priori for reasons of cost, rather than the unbiased approach required for novel biomarker discovery. By comparison, efforts that aim to identify predictors of treatment-associated adverse events (the risk side of the thera peutic equation) have, to date, been limited. Although clinical predictors of toxicity, such as poor performance status or impaired renal and hepatic function, are well rec ognized, efforts that aim to decipher the potentially substantial contribution that genetic variation makes to the development of adverse events has mostly been restricted to candidate gene studies. Although they are often associated with large size effects, these variants are typically rare and are responsible for only a small fraction of therapy-associated morbidity, which makes screening for these impractical in clini cal practice16–19. Thus, at present, with the exception of adjustment for organ function, dose modifications are made in response to
NATURE REVIEWS | CANCER
PERSPECTIVES adverse events during antineoplastic ther apy. This retroactive approach is associated with increased morbidity for patients and greater costs for healthcare providers from hospitalization and supportive care20,21. The systematic unbiased identification and clinical application of germline variants that are predictive of anticancer therapyassociated toxicity — an approach that we refer to as ‘toxgnostics’ — should improve the risk/benefit ratio of treatment by identifying patients who require elective dose modifica tion, or for whom modest gains of therapy are likely to be outweighed by its costs. Although recent advances in polymorphism genotyping and next-generation sequencing mean that the technology platforms for such profiling are already accessible, the development of toxgnostics poses substantial challenges for scientists and clinicians. In this Opinion arti cle, we summarize the field to date, argue that an agnostic approach is likely to yield more clinically tractable results than most current strategies, and propose a roadmap for future toxgnostic development. Predicting toxicity of anticancer therapy Attempts to correlate toxicity and clinical outcomes with pharmacological parameters have mainly focussed on pharmacokinetic and pharmacodynamic endpoints. The premise that underpins this approach is that plasma drug concentrations are determined by the genetic and physiological factors that determine drug absorption, distribution, metabolism and excretion (ADME)22, and that there will be an equilibrium between drug levels in plasma and their sites of action in tumour and normal tissues. This field has been partially successful in that relation ships between plasma drug concentration profiles (‘area under the curve’ (AUC)) and the severity of neutropenia and other toxicities have been shown for several cyto toxics, including 5‑fluorouracil (5‑FU)23–26, methotrexate27,28 and carboplatin29. However, wide intra- and inter-individual variation in cytotoxic pharmacokinetics leads to similarly wide confidence intervals in these measurements, which makes the practi cal application of pharmacokinetic‑driven dosing challenging. Just over 55 years ago, glucose‑6‑phosphate dehydrogenase (G6PD) and pseudocholinesterase deficiency were shown to be manifestations of specific gene muta tions30,31, presaging the introduction of the term ‘pharmacogenetics’ (REF. 32). It was only towards the end of the past century — as a result of the Human Genome Project — that a broader term, ‘pharmacogenomics’, ADVANCE ONLINE PUBLICATION | 1
© 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES was introduced, reflecting the fact that drug responses and adverse effects are deter mined by multiple genetic variants. Over the past two decades, an expanding number of polymorphisms that affect anticancer drug metabolism have been identified, as discussed below. Candidate gene studies To date, the vast majority of attempts to identify germline variants that are predictive of anticancer drug toxicity have used can didate gene approaches, focussing on genes that are involved in various pathways con trolling ADME. As with tumour biomarker studies, despite hundreds of published toxic ity polymorphisms, few have been validated and entered clinical practice. Determination of thiopurine methyltransferase (TPMT) activity is recommended prior to treatment of patients with 6‑mercaptopurine and 6‑thoiguanine in the United Kingdom33–35. Moreover, an association of TPMT vari ants with cisplatin-induced ototoxicity in children36,37 prompted the US Food and Drug Administration (FDA) to list TPMT as a biomarker for adverse drug reactions to cisplatin. However, a subsequent study failed to replicate this finding 38, and it has been argued that the FDA decision was pre mature39. Despite high levels of cytochrome P450 polymorphism40–44, the only P450 assay that is currently recommended for routine clinical use is for the optimization of the prescription of warfarin45–47, despite the fact that drugs such as tamoxifen48 and potentially paclitaxel49,50 are metabolized by P450 enzymes that are known to be poly morphic. Initial data linking the occurrence of severe neutropenia to homozygosity for the UDP glucuronosyl transferase 1A1 poly morphism (UGT1A1*28) were sufficiently compelling that the FDA changed the pack age insert for irinotecan to include this as a risk factor 51,52. However, a subsequent metaanalysis has indicated that this phenomenon is dose-dependent, as the correlation with genotype could not be found when lower doses (100–125 mg per m2) of irinotecan were used53. These and other candidate gene studies are often statistically powered to detect a modest sized effect54–56. However, the limited number of markers that they analyse repre sents a major limitation for toxicity marker discovery, as even single nucleotide poly morphism (SNP) panels that are focused on drug metabolism and transport (for example, 1,936 variants in 231 genes)57 still only give information on a tiny fraction of the genome. Furthermore, the a priori nature of these
studies limits their use to the detection of polymorphisms in genes within known drug metabolic pathways. An additional problem with the vast majority of candidate gene stud ies is their small, single centre, retrospective nature, which contributes to a high risk of false positives15 and explains the failure of most markers identified in such studies to be subsequently validated.
toxgnostics … the systematic, agnostic study of genetic predictors of toxicity from anticancer therapy GWAS approach to toxicity variants Genome-wide association studies (GWAS) compare cases with the phenotype of inter est to controls at 500,000 to several million polymorphic loci across the genome. In order to obtain sufficient statistical power, GWAS generally require several hundred to thou sands of participants58 and a clearly defined phenotype. To date, GWAS have been used in cancer predominantly to identify SNPs that are associated with tumour susceptibility rather than therapeutic toxicity 59,60, although this approach has been successfully used to identify polymorphisms that are associated with drug toxicity in non-malignant condi tions61–63. It is noteworthy that in most of these studies, the variant would not have been predicted a priori, and would thus remain undetected using candidate gene approaches. Although the application of similar agnostic approaches to discovery of toxicity variants in cancer is in its infancy, several recent reports highlight both the strengths of this approach and its associated challenges. Arguably the greatest success is the demonstration of the association between a SNP in SLCO1B1 (encoding an organic anion transporter) and clearance of methotrexate in children with acute lymphoblastic leukaemia (ALL)64,65. It has been suggested that this SNP could be used to identify patients with ALL who have reduced drug clearance and therefore an increased risk of toxicity. Other studies have used GWAS to investigate predictors of toxicity associated with gemcitabine66 and epirubicin67, although in both cases the gener alizability of the results may be limited by the heterogeneous groups used for variant discov ery — highlighting the importance of a clearly defined population. Although GWAS do not, in general, iden tify causal variants, but rather polymorphisms linked to these, the polymorphisms identified
2 | ADVANCE ONLINE PUBLICATION
will often be adequate for the purposes of tox icity prediction. However, caution is required before generalizing results across populations, as linkage disequilibrium between the typed and causal variants may vary across ethnic backgrounds, and this should be reflected in recommendations and drug labelling when translating results into practice. Over the past 5 years, next-generation sequencing (NGS) of tumour exomes and, less commonly, genomes, has revolution ized our understanding of the mutational landscape of tumours68,69. Although financial, bioinformatic and statistical considerations currently limit the attractiveness of NGS for the detection of toxgnostic variants, this is likely to change as NGS becomes increasingly incorporated into routine practice. The dom inant challenge to this methodology will not be the accumulation of data but rather its rel evance, analysis and potential for over-fitting, which produces false-positive results70. Thus, large high-quality data sets are essential if we are not to succumb to the phenomenon of ‘junk in, junk out’. Toxgnostics Although the technologies discussed above have the potential to revolutionize our ability to predict the toxicity of anticancer therapy and improve care for patients, careful application is required if we are to realize this. In a related field, international guidelines were developed to standardize reporting of tumour biomarker studies and improve their quality 15,71,72. In the absence of similar guidelines for the study of toxicity variants, we hope that the term toxgnos tics — defined as the systematic, agnostic study of genetic predictors of toxicity from anticancer therapy — will serve to stimulate high-quality studies that will influence prac tice. The key elements of toxgnostics are highlighted in BOX 1.
Statistical and methodological consider ations. Many of the considerations for toxgnostic investigation are similar to those for tumour biomarker studies, in terms of prospective sample and data collection, with the gold standard represented by embedd ing the toxgnostic study within a welldesigned, prospective, randomized, clinical trial. This has several advantages, including the mandated collection of toxicity data using CTC grades for trial participants in study case report forms (CRFs), a large sam ple size to minimize the risk of false posi tives and negatives, and the availability of high-quality outcome data — an important consideration, as it is plausible that variants www.nature.com/reviews/cancer
© 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES that influence toxicity might also affect treatment outcome73–75. For drugs that are seldom used in clinical trials, dedicated con sortia of investigators are likely to be essen tial for toxgnostic development to ensure the standardization of treatment and to capture data across different patient studies. We strongly recommend that studies incor porating toxgnostic analyses mandate col lection of a sample of peripheral blood from patients for purposes of DNA extraction, as this avoids the potential for confounding analysis by loss of heterozygosity if tumourderived DNA is used76,77. As with any associa tion study, the power to detect a variant that is predictive of an adverse event is a function of the study size, the frequency of toxicity and the frequency and effect size of the risk allele. For example, an average-size Phase III trial of 1,500 patients, in which 20% of patients incur grade 3 or grade 4 adverse events, will be adequately powered (>80%) at genomewide significance (P A and *2A (combined allelic odds ratio (OR) = 5.51 and P = 0.0013), as well as with the common TYMS polymorphisms 5′VNTR 2R/3R and 3′UTR 6bp ins-del (combined OR = 1.33 and P = 0.00018)80. There was weaker evi dence for these polymorphisms predicting toxicity from bolus and infusional 5‑FU monotherapy. No good evidence of associa tion with toxicity was found for the remain ing polymorphisms, including several that are currently included in predictive kits. No polymorphisms were associated with toxicity in combination regimens, perhaps reflect ing the lower dose of drugs when used in combination80. Arguably the greatest strength of toxgnostic screens is their ability to gen erate risk scores that predict toxicity more accurately than single-gene markers. In this study, the optimum panel for prediction of capecitabine monotherapy toxicity comprises four variants: TYMS 5′VNTR 2R/3R, 3′UTR 6bp ins-del, DPYD*2A and DPYD2846T>A. More recently, GWAS of the QUASAR2 clinical data set has identified several clini cally relevant, novel polymorphisms that are associated with capecitabine toxicity and unrelated to those uncovered by conventional pharmacogenetic screens (I.T., unpublished observations). Further studies will build on these and other 87 studies in the field. A well-performed recent analysis used a toxgnostic approach to attempt to define var iants associated with paclitaxel neuropathy in patients with breast cancer 88. The embedd ing of the study within a large Phase III clinical trial (CALGB 40101) permitted the investigators to analyse both the sever ity and the timing of onset of dysfunction, using a standardized grading system (US National Cancer Institute CTC). Although no polymorphisms reached genome-wide significance, a variant in FYVE, RhoGEF and PH domain containing 4 (FGD4) of marginal significance was validated in an independent replication set. FGD4 is a gene with a proven ADVANCE ONLINE PUBLICATION | 3
© 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES role in hereditary Charcot–Marie–Tooth neuropathy, and the variant described was associated with a hazard ratio (HR) of 1.57 per allele. Using a time to event model, the authors were able to generate risk plots for the development of grade 2 neuropathy according to the cumulative paclitaxel dose for patients lacking, heterozygous for or homozygous for the risk allele88. This case also shows the importance of recording the timing of adverse events on study CRFs, as toxgnostic variants that are predictive of early toxicities may differ from those associated with later, dose-dependent phenomena. We predict that the more frequently agnostic analyses such as GWAS are per formed, the more likely we are to discover non-pathway-associated but clinically relevant variants to guide toxgnostic development. Financial considerations A fundamental consideration for toxgnostics is whether its use in clinical practice will result in cost savings. Despite the costs of upfront testing, there are several indications that it may do. Taking adjuvant treatment with fluoropyrimidines as an example, the current model of dosing based on body surface area with modifications for toxic ity results in CTC grade 3 or grade 4 in 30–40% of treated patients in the first cycle, with resultant delays in treatment, and hospitalization owing to serious adverse events in 10–25% of cases across the period of treatment, with a toxic death rate of 1% (REF. 89). The cost of admission for an impor tant toxic event is substantial and includes supportive medication, bed and medical costs. It has been estimated at US$4,000 per day in the United States and an average of $20,000 per stay (with a range of $10,000 to $50,000, depending on the length of stay)90. A recent study that analysed DPYD testing in a community oncology practice found that prospective testing for DPYD muta tions in all patients with colorectal cancer starting 5‑FU would result in a cost saving of ~€1,000 per patient, through avoiding prolonged hospitalization of patients with high-risk variants91. Although the costs of SNP genotyping or NGS analysis are rapidly decreasing, they remain substantial, and careful health economic analysis is essential to determine the effects of the application of toxgnostics on healthcare costs. Clearly, potential savings will depend on toxgnostic test performance — if hospitalization is only avoided in one patient per 100 tested, the test is unlikely to be cost-effective. However, we hope that the broader study of toxgnostics
will lead to improvements that are sufficient to swing the balance in favour of testing as a cost-efficient means of avoiding the substan tial burden of morbidity and mortality of anticancer therapy. Will toxgnostics improve safety? Much effort is rightly focused on enhancing the safety of chemotherapy, both in the hos pital and community settings. Governmental and professional societies have produced practical guidelines that suggest how anti cancer drugs might be optimally delivered. For the most part, these guidelines detail pro cesses that are involved in reconstitution, pre scription, dosing and administration. Modern trial protocols have extensive dose modifica tion sections, which specify dose reductions and delays in response to grade 3 or 4 toxi cities. Unfortunately, this dose modification occurs post-hoc, leaving clinicians respond ing once potentially life-threatening toxicity has occurred. Classically, most acute side effects occur with the first cycle of therapy 92,93, and this is when a toxgnostic test would allow rational intervention through a priori dose modulation or, in some cases, specific drug avoidance. However, care will have to be taken to avoid compromising treatment efficacy in cases for which toxgnostic variants also predict treatment efficacy. Furthermore, although it is likely that toxgnostic panels will vary according to the treatment regimen, we envisage a situation in which personalized
dosing for patients who are treated with mul tiple lines of therapy will be informed by the results of genotyping at diagnosis, without the requirement for additional testing. Clearly, test performance is crucial and will govern clinical acceptance of toxgnostics and their integration into guidelines and treatment algorithms. There is no exist ing toxgnostic that explains and therefore leads to the avoidance of all severe toxicity. For example, the SNP set identified by our meta-analysis of DPYD and TYMS variants predictive of fluoropyrimidine toxicity has a sensitivity and specificity of approximately 65% (meaning that it will predict only two-thirds of the severe toxicity that will occur and will over-predict severe toxicity in one-third of patients) and has a positive predictive value of about 50% (REF. 80). This outperforms existing assays and is within the range of clinical usefulness, but further planned improvements in performance would clearly increase the likelihood of widespread use and give clinicians a real belief that its use would affect their treatment choices in clinical practice. Conclusions On the basis of the above discussions, we propose that, whenever possible, randomized trials of novel antineoplastic agents should function as vehicles for a genome-wide exploration of germline determinants of tox icity, as outlined in FIG. 1. The use of a blood
Clinical trial
Side effect profile and outcome data
DNA extracted from blood sample GWAS (1–4 million SNPs) Bioinformatics
Risk scores to identify patients most at risk of serious side effects Validation of associations in a subpopulation of patients • Improve risk/benefit ratio • Clinical use validation during Phase IV post marketing surveillance • Improve compliance to treatment regimens through a priori dose adjustment • Help regulators to register drugs with substantial side effect profiles
Figure 1 | Proposed clinical trial paradigm for the discovery of ‘toxgnostics’ markers though genome-wide association studies (GWAS). Schematic illustrating key steps in toxgnostic marker Nature Reviews | Cancer discovery and clinical implementation. The availability of high-quality standardized toxicity data within clinical trials enables accurate toxgnostic discovery, and outcome data ensure that markers or risk scores can be tested for association with outcome — ensuring that dose modification does not risk compromising treatment efficacy. Risk scores generally provide greater sensitivity and specificity than single gene biomarkers. SNP, single nucleotide polymorphism.
4 | ADVANCE ONLINE PUBLICATION
www.nature.com/reviews/cancer © 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES sample should also enable all patients to be analysed, rather than a subset. A systematic and agnostic approach to discovering toxicityassociated SNPs, which goes beyond the relatively rigid constructs of genes encoding determinants of ADME, embedded in well designed and adequately powered trials might lead to the discovery of high-performing molecular toxgnostics with sufficient clinical usefulness to change practice and improve chemotherapy safety, thereby saving money and substantially reducing morbidity and mortality. It is now time to integrate prog nostic biomarkers, companion diagnostics and toxgnostics into adjuvant treatment pro tocols for colorectal and other cancers, as we develop algorithms that embrace the totality of precision medicine and allow optimal patient stratification, drug and dose selection to improve the risk/benefit ratio to favour the patient as much as possible. David Church and Rachel Kerr are at the Oxford Cancer Centre, Department of Oncology, University of Oxford, Churchill Hospital, Old Road, Headington, Oxford, OX3 7LE, UK. David Church is also at the Molecular and Population Genetics Laboratory, The Wellcome Trust Centre for Human Genetics, University of Oxford, Roosevelt Drive, Oxford, OX3 7BN, UK. Enric Domingo, Dan Rosmarin, Claire Palles and Ian Tomlinson are at the Molecular and Population Genetics Laboratory, The Wellcome Trust Centre for Human Genetics, University of Oxford, Roosevelt Drive, Oxford, OX3 7BN, UK. Ian Tomlinson is also at the Genomic Medicine Theme, Oxford Comprehensive Biomedical Research Centre, The Wellcome Trust Centre for Human Genetics, University of Oxford, Roosevelt Drive, Oxford, OX3 7BN, UK. Kevin Maskell is at Oxford Cancer Biomarkers, The Magdalen Centre, Oxford Science Park, Robert Robinson Avenue, Oxford, OX4 4GA, UK. David Kerr is at the Nuffield Department of Clinical Laboratory Sciences, University of Oxford, John Radcliffe Hospital, Headley Way, Oxford, OX3 9DU, UK. Correspondence to D.K. e-mail: [email protected] doi:10.1038/nrc3729 Published online 15 May 2014 1. Gray, R. et al. Adjuvant chemotherapy versus observation in patients with colorectal cancer: a randomised study. Lancet 370, 2020–2029 (2007). 2. Widakowich, C. et al. Review: side effects of approved molecular targeted therapies in solid cancers. Oncologist 12, 1443–1455 (2007). 3. Bertagnolli, M. M. et al. Microsatellite instability and loss of heterozygosity at chromosomal location 18q: prospective evaluation of biomarkers for stages II and III colon cancer—a study of CALGB 9581 and 89803. J. Clin. Oncol. 29, 3153–3162 (2011). 4. Roth, A. D. et al. Prognostic role of KRAS and BRAF in stage II and III resected colon cancer: results of the translational study on the PETACC‑3, EORTC 40993, SAKK 60–00 trial. J. Clin. Oncol. 28, 466–474 (2010). 5. Hutchins, G. et al. Value of mismatch repair, KRAS, and BRAF mutations in predicting recurrence and benefits from chemotherapy in colorectal cancer. J. Clin. Oncol. 29, 1261–1270 (2011).
6. Romond, E. H. et al. Trastuzumab plus adjuvant chemotherapy for operable HER2‑positive breast cancer. N. Engl. J. Med. 353, 1673–1684 (2005). 7. Piccart-Gebhart, M. J. et al. Trastuzumab after adjuvant chemotherapy in HER2‑positive breast cancer. N. Engl. J. Med. 353, 1659–1672 (2005). 8. Sosman, J. A. et al. Survival in BRAF V600‑mutant advanced melanoma treated with vemurafenib. N. Engl. J. Med. 366, 707–714 (2012). 9. Amado, R. G. et al. Wild-type KRAS is required for panitumumab efficacy in patients with metastatic colorectal cancer. J. Clin. Oncol. 26, 1626–1634 (2008). 10. Van Cutsem, E. et al. Cetuximab and chemotherapy as initial treatment for metastatic colorectal cancer. N. Engl. J. Med. 360, 1408–1417 (2009). 11. Parkinson, D. R., Johnson, B. E. & Sledge, G. W. Making personalized cancer medicine a reality: challenges and opportunities in the development of biomarkers and companion diagnostics. Clin. Cancer Res. 18, 619–624 (2012). 12. Walther, A. et al. Genetic prognostic and predictive markers in colorectal cancer. Nature Rev. Cancer 9, 489–499 (2009). 13. Church, D., Midgley, R. & Kerr, D. Biomarkers in early-stage colorectal cancer: ready for prime time? Digestive Diseases 30 (Suppl. 2), 27–33 (2012). 14. Buyse, M., Sargent, D. J., Grothey, A., Matheson, A. & de Gramont, A. Biomarkers and surrogate end points—the challenge of statistical validation. Nature Rev. Clin. Oncol. 7, 309–317 (2010). 15. Simon, R. M., Paik, S. & Hayes, D. F. Use of Archived specimens in evaluation of prognostic and predictive biomarkers. J. Natl Cancer Institute 101, 1446–1452 (2009). 16. Ciccolini, J., Gross, E., Dahan, L., Lacarelle, B. & Mercier, C. Routine dihydropyrimidine dehydrogenase testing for anticipating 5‑fluorouracil-related severe toxicities: hype or hope? Clin. Colorectal Cancer 9, 224–228 (2010). 17. Locker, G. Y. et al. ASCO 2006 update of recommendations for the use of tumor markers in gastrointestinal cancer. J. Clin. Oncol. 24, 5313–5327 (2006). 18. Gold, H. T., Hall, M. J., Blinder, V. & Schackman, B. R. Cost effectiveness of pharmacogenetic testing for uridine diphosphate glucuronosyltransferase 1A1 before irinotecan administration for metastatic colorectal cancer. Cancer 115, 3858–3867 (2009). 19. Schmoll, H. J. et al. ESMO Consensus Guidelines for management of patients with colon and rectal cancer. a personalized approach to clinical decision making. Ann. Oncol. 23, 2479–2516 (2012). 20. Delea, T. E. et al. The incidence and cost of hospitalization for 5‑FU toxicity among Medicare beneficiaries with metastatic colorectal cancer. Value Health 5, 35–43 (2002). 21. Calhoun, E. A. et al. Evaluating the total costs of chemotherapy-induced toxicity: results from a pilot study with ovarian cancer patients. Oncologist 6, 441–445 (2001). 22. Wishart, D. S. Improving early drug discovery through ADME modelling: an overview. Drugs R. D. 8, 349–362 (2007). 23. Gamelin, E. et al. Long-term weekly treatment of colorectal metastatic cancer with fluorouracil and leucovorin: results of a multicentric prospective trial of fluorouracil dosage optimization by pharmacokinetic monitoring in 152 patients. J. Clin. Oncol. 16, 1470–1478 (1998). 24. Gamelin, E. C. et al. Relationship between 5‑fluorouracil (5‑FU) dose intensity and therapeutic response in patients with advanced colorectal cancer receiving infusional therapy containing 5‑FU. Cancer 77, 441–451 (1996). 25. Hillcoat, B. L., McCulloch, P. B., Figueredo, A. T., Ehsan, M. H. & Rosenfeld, J. M. Clinical response and plasma levels of 5‑fluorouracil in patients with colonic cancer treated by drug infusion. Br. J. Cancer 38, 719–724 (1978). 26. Di Paolo, A. et al. Relationship between 5‑fluorouracil disposition, toxicity and dihydropyrimidine dehydrogenase activity in cancer patients. Ann. Oncol. 12, 1301–1306 (2001). 27. Newell, D. R. Clinical pharmacokinetics of antitumor antifolates. Semin. Oncol. 26, 74–81 (1999).
NATURE REVIEWS | CANCER
28. Jolivet, J., Cowan, K. H., Curt, G. A., Clendeninn, N. J. & Chabner, B. A. The pharmacology and clinical use of methotrexate. N. Engl. J. Med. 309, 1094–1104 (1983). 29. Calvert, A. H. et al. Carboplatin dosage: prospective evaluation of a simple formula based on renal function. J. Clin. Oncol. 7, 1748–1756 (1989). 30. Lehmann, H. & Ryan, E. The familial incidence of low pseudocholinesterase level. Lancet 271, 124 (1956). 31. Alving, A. S., Carson, P. E., Flanagan, C. L. & Ickes, C. E. Enzymatic deficiency in primaquinesensitive erythrocytes. Science 124, 484–485 (1956). 32. Vogel, F. Moderne Probleme der Humangenetik. Ergebn. Inn. Med. Kinderheilkd. 12, 52–125 (1959). 33. Lennard, L., Lilleyman, J. S., Van Loon, J. & Weinshilboum, R. M. Genetic variation in response to 6‑mercaptopurine for childhood acute lymphoblastic leukaemia. Lancet 336, 225–229 (1990). 34. Relling, M. V. et al. Clinical Pharmacogenetics Implementation Consortium guidelines for thiopurine methyltransferase genotype and thiopurine dosing. Clin. Pharmacol. Ther. 89, 387–391 (2011). 35. Ford, L. T. & Berg, J. D. Thiopurine S‑methyltransferase (TPMT) assessment prior to starting thiopurine drug treatment; a pharmacogenomic test whose time has come. J. Clin. Pathol. 63, 288–295 (2010). 36. Ross, C. J. et al. Genetic variants in TPMT and COMT are associated with hearing loss in children receiving cisplatin chemotherapy. Nature Genet. 41, 1345–1349 (2009). 37. Pussegoda, K. et al. Replication of TPMT and ABCC3 genetic variants highly associated with cisplatininduced hearing loss in children. Clin. Pharmacol. Ther. 94, 243–251 (2013). 38. Yang, J. J. et al. The role of inherited TPMT and COMT genetic variation in cisplatin-induced ototoxicity in children with cancer. Clin. Pharmacol. Ther. 94, 252–259 (2013). 39. Ratain, M. J., Cox, N. J. & Henderson, T. O. Challenges in interpreting the evidence for genetic predictors of ototoxicity. Clin. Pharmacol. Ther. 94, 631–635 (2013). 40. Zanger, U. M. & Schwab, M. Cytochrome P450 enzymes in drug metabolism: regulation of gene expression, enzyme activities, and impact of genetic variation. Pharmacol. Ther. 138, 103–141 (2013). 41. Mahgoub, A., Idle, J. R., Dring, L. G., Lancaster, R. & Smith, R. L. Polymorphic hydroxylation of Debrisoquine in man. Lancet 2, 584–586 (1977). 42. Heim, M. & Meyer, U. A. Genotyping of poor metabolisers of debrisoquine by allele-specific PCR amplification. Lancet 336, 529–532 (1990). 43. Kupfer, A. & Preisig, R. Pharmacogenetics of mephenytoin: a new drug hydroxylation polymorphism in man. Eur. J. Clin. Pharmacol. 26, 753–759 (1984). 44. Sullivan-Klose, T. H. et al. The role of the CYP2C9‑Leu359 allelic variant in the tolbutamide polymorphism. Pharmacogenetics 6, 341–349 (1996). 45. Jonas, D. E. & McLeod, H. L. Genetic and clinical factors relating to warfarin dosing. Trends Pharmacol. Sci. 30, 375–386 (2009). 46. Schwab, M. & Schaeffeler, E. Warfarin pharmacogenetics meets clinical use. Blood 118, 2938–2939 (2011). 47. Klein, T. E. et al. Estimation of the warfarin dose with clinical and pharmacogenetic data. N. Engl. J. Med. 360, 753–764 (2009). 48. Schroth, W. et al. Breast cancer treatment outcome with adjuvant tamoxifen relative to patient CYP2D6 and CYP2C19 genotypes. J. Clin. Oncol. 25, 5187–5193 (2007). 49. Hertz, D. L. et al. CYP2C8*3 predicts benefit/risk profile in breast cancer patients receiving neoadjuvant paclitaxel. Breast Cancer Res. Treat. 134, 401–410 (2012). 50. Hertz, D. L. et al. CYP2C8*3 increases risk of neuropathy in breast cancer patients treated with paclitaxel. Ann. Oncol. 24, 1472–1478 (2013). 51. Iyer, L. et al. Genetic predisposition to the metabolism of irinotecan (CPT‑11). Role of uridine diphosphate glucuronosyltransferase isoform 1A1 in the glucuronidation of its active metabolite (SN‑38) in human liver microsomes. J. Clin. Invest. 101, 847–854 (1998).
ADVANCE ONLINE PUBLICATION | 5 © 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES 52. Ando, Y. et al. UGT1A1 genotypes and glucuronidation of SN‑38, the active metabolite of irinotecan. Ann. Oncol. 9, 845–847 (1998). 53. Hoskins, J. M., Goldberg, R. M., Qu, P., Ibrahim, J. G. & McLeod, H. L. UGT1A1*28 genotype and irinotecan-induced neutropenia: dose matters. J. Natl Cancer Inst. 99, 1290–1295 (2007). 54. Schwab, M. et al. Role of genetic and nongenetic factors for fluorouracil treatment-related severe toxicity: a prospective clinical trial by the German 5‑FU Toxicity Study Group. J. Clin. Oncol. 26, 2131–2138 (2008). 55. Loganayagam, A. et al. Pharmacogenetic variants in the DPYD, TYMS, CDA and MTHFR genes are clinically significant predictors of fluoropyrimidine toxicity. Br. J. Cancer 108, 2505–2515 (2013). 56. McLeod, H. L. et al. Pharmacogenetic predictors of adverse events and response to chemotherapy in metastatic colorectal cancer: results from North American Gastrointestinal Intergroup Trial N9741. J. Clin. Oncol. 28, 3227–3233 (2010). 57. Burmeister, H., Aebi, S., Studer, C., Fey, M. F. & Gautschi, O. Adherence to ESMO clinical recommendations for prophylaxis of chemotherapyinduced nausea and vomiting. Supportive Care Cancer 20, 141–147 (2012). 58. Watanabe, R. M. Statistical issues in gene association studies. Methods Mol. Biol. 700, 17–36 (2011). 59. Dunlop, M. G. et al. Common variation near CDKN1A, POLD3 and SHROOM2 influences colorectal cancer risk. Nature Genet. 44, 770–776 (2012). 60. Houlston, R. S. et al. Meta-analysis of three genomewide association studies identifies susceptibility loci for colorectal cancer at 1q41, 3q26.2, 12q13.13 and 20q13.33. Nature Genet. 42, 973–977 (2010). 61. Daly, A. K. et al. HLA‑B*5701 genotype is a major determinant of drug-induced liver injury due to flucloxacillin. Nature Genet. 41, 816–819 (2009). 62. McCormack, M. et al. HLA‑A*3101 and carbamazepine-induced hypersensitivity reactions in Europeans. N. Engl. J. Med. 364, 1134–1143 (2011). 63. Link, E. et al. SLCO1B1 variants and statin-induced myopathy—a genomewide study. N. Engl. J. Med. 359, 789–799 (2008). 64. Trevino, L. R. et al. Germline genetic variation in an organic anion transporter polypeptide associated with methotrexate pharmacokinetics and clinical effects. J. Clin. Oncol. 27, 5972–5978 (2009). 65. Ramsey, L. B. et al. Rare versus common variants in pharmacogenetics: SLCO1B1 variation and methotrexate disposition. Genome Res. 22, 1–8 (2012). 66. Kiyotani, K. et al. A genome-wide association study identifies four genetic markers for hematological toxicities in cancer patients receiving gemcitabine therapy. Pharmacogenet. Genomics 22, 229–235 (2012). 67. Srinivasan, Y. et al. Genome-wide association study of epirubicin-induced leukopenia in Japanese patients. Pharmacogenet. Genomics 21, 552–558 (2011).
68. Meyerson, M. et al. Advances in understanding cancer genomes through second-generation sequencing. Nat. Rev. Genet. 11, 685–696 (2010). 69. McDermott, U., Downing, J. R. & Stratton, M. R. Genomics and the continuum of cancer care. N. Engl. J. Med. 364, 340–350 (2011). 70. Liu, J. et al. Predictive Power Estimation Algorithm (PPEA)—a new algorithm to reduce overfitting for genomic biomarker discovery. PLoS ONE 6, e24233 (2011). 71. McShane, L. M. et al. Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK). J. Natl Cancer Institute 97, 1180–1184 (2005). 72. Riley, R. D. et al. Prognosis Research Strategy (PROGRESS) 2: prognostic factor research. PLoS Med. 10, e1001380 (2013). 73. Martinez-Balibrea, E. et al. UGT1A and TYMS genetic variants predict toxicity and response of colorectal cancer patients treated with first-line irinotecan and fluorouracil combination therapy. Br. J. Cancer 103, 581–589 (2010). 74. Glimelius, B. et al. Prediction of irinotecan and 5‑fluorouracil toxicity and response in patients with advanced colorectal cancer. Pharmacogenomics J. 11, 61–71 (2011). 75. Cecchin, E. et al. Predictive role of the UGT1A1, UGT1A7, and UGT1A9 genetic variants and their haplotypes on the outcome of metastatic colorectal cancer patients treated with fluorouracil, leucovorin, and irinotecan. J. Clin. Oncol. 27, 2457–2465 (2009). 76. Pharoah, P. D., Abraham, J. & Caldas, C. Re: CYP2D6 genotype and tamoxifen response in postmenopausal women with endocrine-responsive breast cancer: the Breast International Group 1–98 trial and Re: CYP2D6 and UGT2B7 genotype and risk of recurrence in tamoxifen-treated breast cancer patients. J. Natl. Cancer Inst. 104, 1263–1264; author reply 1266 –1268 (2012). 77. Regan, M. M. et al. CYP2D6 genotype and tamoxifen response in postmenopausal women with endocrineresponsive breast cancer: the breast international group 1–98 trial. J. Natl Cancer Inst. 104, 441–451 (2012). 78. Salazar, R. et al. Gene expression signature to improve prognosis prediction of stage II and III colorectal cancer. J. Clin. Oncol. 29, 17–24 (2011). 79. Gray, R. G. et al. Validation study of a quantitative multigene reverse transcriptase-polymerase chain reaction assay for assessment of recurrence risk in patients with stage II colon cancer. J. Clin. Oncol. 29, 4611–4619 (2011). 80. Rosmarin, D. et al. Genetic markers of toxicity from capecitabine and other fluorouracil-based regimens: investigation in the QUASAR2 study, systematic review, and meta-Analysis. J. Clin. Oncol. 32, 1031–1039 (2014). 81. Longley, D. B., Harkin, D. P. & Johnston, P. G. 5‑fluorouracil: mechanisms of action and clinical strategies. Nature Rev. Cancer 3, 330–338 (2003).
6 | ADVANCE ONLINE PUBLICATION
82. Diasio, R. B., Beavers, T. L. & Carpenter, J. T. Familial deficiency of dihydropyrimidine dehydrogenase. Biochemical basis for familial pyrimidinemia and severe 5‑fluorouracil-induced toxicity. J. Clin. Invest. 81, 47–51 (1988). 83. Harris, B. E., Carpenter, J. T. & Diasio, R. B. Severe 5‑fluorouracil toxicity secondary to dihydropyrimidine dehydrogenase deficiency. A potentially more common pharmacogenetic syndrome. Cancer 68, 499–501 (1991). 84. Etienne, M. C. et al. Population study of dihydropyrimidine dehydrogenase in cancer patients. J. Clin. Oncol. 12, 2248–2253 (1994). 85. van Staveren, M. C., Jan Guchelaar, H., van Kuilenburg, A. B., Gelderblom, H. & Maring, J. G. Evaluation of predictive tests for screening for dihydropyrimidine dehydrogenase deficiency. Pharmacogenomics J. 13, 389–395 (2013). 86. Pullarkat, S. T. et al. Thymidylate synthase gene polymorphism determines response and toxicity of 5‑FU chemotherapy. Pharmacogenomics J. 1, 65–70 (2001). 87. Fernandez-Rozadilla, C. et al. Pharmacogenomics in colorectal cancer: a genome-wide association study to predict toxicity after 5‑fluorouracil or FOLFOX administration. Pharmacogenomics J. 13, 209–217 (2013). 88. Baldwin, R. M. et al. A genome-wide association study identifies novel loci for paclitaxel-induced sensory peripheral neuropathy in CALGB 40101. Clin. Cancer Res. 18, 5099–5109 (2012). 89. Twelves, C. et al. Capecitabine as adjuvant treatment for stage III colon cancer. New Engl. J. Med. 352, 2696–2704 (2005). 90. Kuderer, N. M., Dale, D. C., Crawford, J., Cosler, L. E. & Lyman, G. H. Mortality, morbidity, and cost associated with febrile neutropenia in adult cancer patients. Cancer 106, 2258–2266 (2006). 91. Ahmed, G. et al. Cost implications of reactive versus prospective testing for dihydropyrimidine dehydrogenase (DPD) deficiency in patients with colorectal cancer. J. Clin. Oncol. 31, (Suppl.; Abstr 3627) (2013). 92. Blay, J. Y. et al. Early lymphopenia after cytotoxic chemotherapy as a risk factor for febrile neutropenia. J. Clin. Oncol. 14, 636–643 (1996). 93. Crawford, J., Dale, D. C. & Lyman, G. H. Chemotherapy-induced neutropenia: risks, consequences, and new directions for its management. Cancer 100, 228–237 (2004).
Acknowledgements
Core funding to the Wellcome Trust Centre for Human Genetics was provided by the Wellcome Trust (090532/Z/09/Z).
Competing interests statement
The authors declare competing interests: see Web version for details.
www.nature.com/reviews/cancer © 2014 Macmillan Publishers Limited. All rights reserved
OPINION
‘Toxgnostics’: an unmet need in cancer medicine David Church, Rachel Kerr, Enric Domingo, Dan Rosmarin, Claire Palles, Kevin Maskell, Ian Tomlinson and David Kerr
Abstract | If we were to summarize the rationale that underpins medical oncology in a Latin aphorism, it might be ‘veneno ergo sum’; that is, I poison, therefore I am. The burden of chemotherapy-associated toxicity is well recognized, but we have relatively few tools that increase the precision of anticancer drug prescribing. We propose a shift in emphasis from the focussed study of polymorphisms in drug metabolic pathways in small sets of patients to broader agnostic analyses to systematically correlate germline genetic variants with adverse events in large, well-defined cancer populations. Thus, we propose the new science of ‘toxgnostics’ (that is, the systematic, agnostic study of genetic predictors of toxicity from anticancer therapy). If it were not for the great variability among individuals, medicine might as well be a science and not an art. Sir William Osler,
Principles and Practice of Medicine, 1st Edition, Appleton and Company, New York, 1892
Although substantial advances have been made in the systemic therapy of solid tumours over the past two decades, most patients obtain only modest benefit from treatment, whereas toxicity is common and often associated with considerable morbid ity and mortality. For example, adjuvant chemotherapy following resection of stage II colorectal cancer benefits only ~4% of patients, whereas 30–40% of those treated will suffer common toxicity criteria (CTC) grade 3–4, and 0.5–1% will suffer fatal toxi cities1. Targeted therapies are also associated with significant adverse event profiles from both on- and off-target effects2. All drugs are defined by their therapeutic ratios (efficacy/toxicity). However, this is particularly true of antineoplastic agents, for which the dose–response curves are usually steep and the therapeutic window is correspondingly narrow. Although the past 10 years has seen much effort to improve the risk/benefit ratio of anticancer therapy, this has mostly focused on maximising the benefit side of the equation, with atten tion directed towards the identification of tumour biomarkers with prognostic impor tance that allow patient stratification according to the likelihood of recurrence or progression (see REFS 3–5 for examples), or companion diagnostics that predict
response to therapy 6–10 (reviewed in REF. 11). Despite thousands of reports that encompass hundreds of biomarkers in common solid tumours, the number that has entered rou tine clinical practice is depressingly small12–14. The reasons for this shortfall are manifold, but the high risk of false positives that result from multiple testing in small samples and the lack of validation cohorts are important contributors14,15. Recent biomarker studies using large (n >1,000 patients), adequately statistically powered clinical trials go some way to addressing this, but they remain lim ited by the need to focus on a few biomark ers defined a priori for reasons of cost, rather than the unbiased approach required for novel biomarker discovery. By comparison, efforts that aim to identify predictors of treatment-associated adverse events (the risk side of the thera peutic equation) have, to date, been limited. Although clinical predictors of toxicity, such as poor performance status or impaired renal and hepatic function, are well rec ognized, efforts that aim to decipher the potentially substantial contribution that genetic variation makes to the development of adverse events has mostly been restricted to candidate gene studies. Although they are often associated with large size effects, these variants are typically rare and are responsible for only a small fraction of therapy-associated morbidity, which makes screening for these impractical in clini cal practice16–19. Thus, at present, with the exception of adjustment for organ function, dose modifications are made in response to
NATURE REVIEWS | CANCER
PERSPECTIVES adverse events during antineoplastic ther apy. This retroactive approach is associated with increased morbidity for patients and greater costs for healthcare providers from hospitalization and supportive care20,21. The systematic unbiased identification and clinical application of germline variants that are predictive of anticancer therapyassociated toxicity — an approach that we refer to as ‘toxgnostics’ — should improve the risk/benefit ratio of treatment by identifying patients who require elective dose modifica tion, or for whom modest gains of therapy are likely to be outweighed by its costs. Although recent advances in polymorphism genotyping and next-generation sequencing mean that the technology platforms for such profiling are already accessible, the development of toxgnostics poses substantial challenges for scientists and clinicians. In this Opinion arti cle, we summarize the field to date, argue that an agnostic approach is likely to yield more clinically tractable results than most current strategies, and propose a roadmap for future toxgnostic development. Predicting toxicity of anticancer therapy Attempts to correlate toxicity and clinical outcomes with pharmacological parameters have mainly focussed on pharmacokinetic and pharmacodynamic endpoints. The premise that underpins this approach is that plasma drug concentrations are determined by the genetic and physiological factors that determine drug absorption, distribution, metabolism and excretion (ADME)22, and that there will be an equilibrium between drug levels in plasma and their sites of action in tumour and normal tissues. This field has been partially successful in that relation ships between plasma drug concentration profiles (‘area under the curve’ (AUC)) and the severity of neutropenia and other toxicities have been shown for several cyto toxics, including 5‑fluorouracil (5‑FU)23–26, methotrexate27,28 and carboplatin29. However, wide intra- and inter-individual variation in cytotoxic pharmacokinetics leads to similarly wide confidence intervals in these measurements, which makes the practi cal application of pharmacokinetic‑driven dosing challenging. Just over 55 years ago, glucose‑6‑phosphate dehydrogenase (G6PD) and pseudocholinesterase deficiency were shown to be manifestations of specific gene muta tions30,31, presaging the introduction of the term ‘pharmacogenetics’ (REF. 32). It was only towards the end of the past century — as a result of the Human Genome Project — that a broader term, ‘pharmacogenomics’, ADVANCE ONLINE PUBLICATION | 1
© 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES was introduced, reflecting the fact that drug responses and adverse effects are deter mined by multiple genetic variants. Over the past two decades, an expanding number of polymorphisms that affect anticancer drug metabolism have been identified, as discussed below. Candidate gene studies To date, the vast majority of attempts to identify germline variants that are predictive of anticancer drug toxicity have used can didate gene approaches, focussing on genes that are involved in various pathways con trolling ADME. As with tumour biomarker studies, despite hundreds of published toxic ity polymorphisms, few have been validated and entered clinical practice. Determination of thiopurine methyltransferase (TPMT) activity is recommended prior to treatment of patients with 6‑mercaptopurine and 6‑thoiguanine in the United Kingdom33–35. Moreover, an association of TPMT vari ants with cisplatin-induced ototoxicity in children36,37 prompted the US Food and Drug Administration (FDA) to list TPMT as a biomarker for adverse drug reactions to cisplatin. However, a subsequent study failed to replicate this finding 38, and it has been argued that the FDA decision was pre mature39. Despite high levels of cytochrome P450 polymorphism40–44, the only P450 assay that is currently recommended for routine clinical use is for the optimization of the prescription of warfarin45–47, despite the fact that drugs such as tamoxifen48 and potentially paclitaxel49,50 are metabolized by P450 enzymes that are known to be poly morphic. Initial data linking the occurrence of severe neutropenia to homozygosity for the UDP glucuronosyl transferase 1A1 poly morphism (UGT1A1*28) were sufficiently compelling that the FDA changed the pack age insert for irinotecan to include this as a risk factor 51,52. However, a subsequent metaanalysis has indicated that this phenomenon is dose-dependent, as the correlation with genotype could not be found when lower doses (100–125 mg per m2) of irinotecan were used53. These and other candidate gene studies are often statistically powered to detect a modest sized effect54–56. However, the limited number of markers that they analyse repre sents a major limitation for toxicity marker discovery, as even single nucleotide poly morphism (SNP) panels that are focused on drug metabolism and transport (for example, 1,936 variants in 231 genes)57 still only give information on a tiny fraction of the genome. Furthermore, the a priori nature of these
studies limits their use to the detection of polymorphisms in genes within known drug metabolic pathways. An additional problem with the vast majority of candidate gene stud ies is their small, single centre, retrospective nature, which contributes to a high risk of false positives15 and explains the failure of most markers identified in such studies to be subsequently validated.
toxgnostics … the systematic, agnostic study of genetic predictors of toxicity from anticancer therapy GWAS approach to toxicity variants Genome-wide association studies (GWAS) compare cases with the phenotype of inter est to controls at 500,000 to several million polymorphic loci across the genome. In order to obtain sufficient statistical power, GWAS generally require several hundred to thou sands of participants58 and a clearly defined phenotype. To date, GWAS have been used in cancer predominantly to identify SNPs that are associated with tumour susceptibility rather than therapeutic toxicity 59,60, although this approach has been successfully used to identify polymorphisms that are associated with drug toxicity in non-malignant condi tions61–63. It is noteworthy that in most of these studies, the variant would not have been predicted a priori, and would thus remain undetected using candidate gene approaches. Although the application of similar agnostic approaches to discovery of toxicity variants in cancer is in its infancy, several recent reports highlight both the strengths of this approach and its associated challenges. Arguably the greatest success is the demonstration of the association between a SNP in SLCO1B1 (encoding an organic anion transporter) and clearance of methotrexate in children with acute lymphoblastic leukaemia (ALL)64,65. It has been suggested that this SNP could be used to identify patients with ALL who have reduced drug clearance and therefore an increased risk of toxicity. Other studies have used GWAS to investigate predictors of toxicity associated with gemcitabine66 and epirubicin67, although in both cases the gener alizability of the results may be limited by the heterogeneous groups used for variant discov ery — highlighting the importance of a clearly defined population. Although GWAS do not, in general, iden tify causal variants, but rather polymorphisms linked to these, the polymorphisms identified
2 | ADVANCE ONLINE PUBLICATION
will often be adequate for the purposes of tox icity prediction. However, caution is required before generalizing results across populations, as linkage disequilibrium between the typed and causal variants may vary across ethnic backgrounds, and this should be reflected in recommendations and drug labelling when translating results into practice. Over the past 5 years, next-generation sequencing (NGS) of tumour exomes and, less commonly, genomes, has revolution ized our understanding of the mutational landscape of tumours68,69. Although financial, bioinformatic and statistical considerations currently limit the attractiveness of NGS for the detection of toxgnostic variants, this is likely to change as NGS becomes increasingly incorporated into routine practice. The dom inant challenge to this methodology will not be the accumulation of data but rather its rel evance, analysis and potential for over-fitting, which produces false-positive results70. Thus, large high-quality data sets are essential if we are not to succumb to the phenomenon of ‘junk in, junk out’. Toxgnostics Although the technologies discussed above have the potential to revolutionize our ability to predict the toxicity of anticancer therapy and improve care for patients, careful application is required if we are to realize this. In a related field, international guidelines were developed to standardize reporting of tumour biomarker studies and improve their quality 15,71,72. In the absence of similar guidelines for the study of toxicity variants, we hope that the term toxgnos tics — defined as the systematic, agnostic study of genetic predictors of toxicity from anticancer therapy — will serve to stimulate high-quality studies that will influence prac tice. The key elements of toxgnostics are highlighted in BOX 1.
Statistical and methodological consider ations. Many of the considerations for toxgnostic investigation are similar to those for tumour biomarker studies, in terms of prospective sample and data collection, with the gold standard represented by embedd ing the toxgnostic study within a welldesigned, prospective, randomized, clinical trial. This has several advantages, including the mandated collection of toxicity data using CTC grades for trial participants in study case report forms (CRFs), a large sam ple size to minimize the risk of false posi tives and negatives, and the availability of high-quality outcome data — an important consideration, as it is plausible that variants www.nature.com/reviews/cancer
© 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES that influence toxicity might also affect treatment outcome73–75. For drugs that are seldom used in clinical trials, dedicated con sortia of investigators are likely to be essen tial for toxgnostic development to ensure the standardization of treatment and to capture data across different patient studies. We strongly recommend that studies incor porating toxgnostic analyses mandate col lection of a sample of peripheral blood from patients for purposes of DNA extraction, as this avoids the potential for confounding analysis by loss of heterozygosity if tumourderived DNA is used76,77. As with any associa tion study, the power to detect a variant that is predictive of an adverse event is a function of the study size, the frequency of toxicity and the frequency and effect size of the risk allele. For example, an average-size Phase III trial of 1,500 patients, in which 20% of patients incur grade 3 or grade 4 adverse events, will be adequately powered (>80%) at genomewide significance (P A and *2A (combined allelic odds ratio (OR) = 5.51 and P = 0.0013), as well as with the common TYMS polymorphisms 5′VNTR 2R/3R and 3′UTR 6bp ins-del (combined OR = 1.33 and P = 0.00018)80. There was weaker evi dence for these polymorphisms predicting toxicity from bolus and infusional 5‑FU monotherapy. No good evidence of associa tion with toxicity was found for the remain ing polymorphisms, including several that are currently included in predictive kits. No polymorphisms were associated with toxicity in combination regimens, perhaps reflect ing the lower dose of drugs when used in combination80. Arguably the greatest strength of toxgnostic screens is their ability to gen erate risk scores that predict toxicity more accurately than single-gene markers. In this study, the optimum panel for prediction of capecitabine monotherapy toxicity comprises four variants: TYMS 5′VNTR 2R/3R, 3′UTR 6bp ins-del, DPYD*2A and DPYD2846T>A. More recently, GWAS of the QUASAR2 clinical data set has identified several clini cally relevant, novel polymorphisms that are associated with capecitabine toxicity and unrelated to those uncovered by conventional pharmacogenetic screens (I.T., unpublished observations). Further studies will build on these and other 87 studies in the field. A well-performed recent analysis used a toxgnostic approach to attempt to define var iants associated with paclitaxel neuropathy in patients with breast cancer 88. The embedd ing of the study within a large Phase III clinical trial (CALGB 40101) permitted the investigators to analyse both the sever ity and the timing of onset of dysfunction, using a standardized grading system (US National Cancer Institute CTC). Although no polymorphisms reached genome-wide significance, a variant in FYVE, RhoGEF and PH domain containing 4 (FGD4) of marginal significance was validated in an independent replication set. FGD4 is a gene with a proven ADVANCE ONLINE PUBLICATION | 3
© 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES role in hereditary Charcot–Marie–Tooth neuropathy, and the variant described was associated with a hazard ratio (HR) of 1.57 per allele. Using a time to event model, the authors were able to generate risk plots for the development of grade 2 neuropathy according to the cumulative paclitaxel dose for patients lacking, heterozygous for or homozygous for the risk allele88. This case also shows the importance of recording the timing of adverse events on study CRFs, as toxgnostic variants that are predictive of early toxicities may differ from those associated with later, dose-dependent phenomena. We predict that the more frequently agnostic analyses such as GWAS are per formed, the more likely we are to discover non-pathway-associated but clinically relevant variants to guide toxgnostic development. Financial considerations A fundamental consideration for toxgnostics is whether its use in clinical practice will result in cost savings. Despite the costs of upfront testing, there are several indications that it may do. Taking adjuvant treatment with fluoropyrimidines as an example, the current model of dosing based on body surface area with modifications for toxic ity results in CTC grade 3 or grade 4 in 30–40% of treated patients in the first cycle, with resultant delays in treatment, and hospitalization owing to serious adverse events in 10–25% of cases across the period of treatment, with a toxic death rate of 1% (REF. 89). The cost of admission for an impor tant toxic event is substantial and includes supportive medication, bed and medical costs. It has been estimated at US$4,000 per day in the United States and an average of $20,000 per stay (with a range of $10,000 to $50,000, depending on the length of stay)90. A recent study that analysed DPYD testing in a community oncology practice found that prospective testing for DPYD muta tions in all patients with colorectal cancer starting 5‑FU would result in a cost saving of ~€1,000 per patient, through avoiding prolonged hospitalization of patients with high-risk variants91. Although the costs of SNP genotyping or NGS analysis are rapidly decreasing, they remain substantial, and careful health economic analysis is essential to determine the effects of the application of toxgnostics on healthcare costs. Clearly, potential savings will depend on toxgnostic test performance — if hospitalization is only avoided in one patient per 100 tested, the test is unlikely to be cost-effective. However, we hope that the broader study of toxgnostics
will lead to improvements that are sufficient to swing the balance in favour of testing as a cost-efficient means of avoiding the substan tial burden of morbidity and mortality of anticancer therapy. Will toxgnostics improve safety? Much effort is rightly focused on enhancing the safety of chemotherapy, both in the hos pital and community settings. Governmental and professional societies have produced practical guidelines that suggest how anti cancer drugs might be optimally delivered. For the most part, these guidelines detail pro cesses that are involved in reconstitution, pre scription, dosing and administration. Modern trial protocols have extensive dose modifica tion sections, which specify dose reductions and delays in response to grade 3 or 4 toxi cities. Unfortunately, this dose modification occurs post-hoc, leaving clinicians respond ing once potentially life-threatening toxicity has occurred. Classically, most acute side effects occur with the first cycle of therapy 92,93, and this is when a toxgnostic test would allow rational intervention through a priori dose modulation or, in some cases, specific drug avoidance. However, care will have to be taken to avoid compromising treatment efficacy in cases for which toxgnostic variants also predict treatment efficacy. Furthermore, although it is likely that toxgnostic panels will vary according to the treatment regimen, we envisage a situation in which personalized
dosing for patients who are treated with mul tiple lines of therapy will be informed by the results of genotyping at diagnosis, without the requirement for additional testing. Clearly, test performance is crucial and will govern clinical acceptance of toxgnostics and their integration into guidelines and treatment algorithms. There is no exist ing toxgnostic that explains and therefore leads to the avoidance of all severe toxicity. For example, the SNP set identified by our meta-analysis of DPYD and TYMS variants predictive of fluoropyrimidine toxicity has a sensitivity and specificity of approximately 65% (meaning that it will predict only two-thirds of the severe toxicity that will occur and will over-predict severe toxicity in one-third of patients) and has a positive predictive value of about 50% (REF. 80). This outperforms existing assays and is within the range of clinical usefulness, but further planned improvements in performance would clearly increase the likelihood of widespread use and give clinicians a real belief that its use would affect their treatment choices in clinical practice. Conclusions On the basis of the above discussions, we propose that, whenever possible, randomized trials of novel antineoplastic agents should function as vehicles for a genome-wide exploration of germline determinants of tox icity, as outlined in FIG. 1. The use of a blood
Clinical trial
Side effect profile and outcome data
DNA extracted from blood sample GWAS (1–4 million SNPs) Bioinformatics
Risk scores to identify patients most at risk of serious side effects Validation of associations in a subpopulation of patients • Improve risk/benefit ratio • Clinical use validation during Phase IV post marketing surveillance • Improve compliance to treatment regimens through a priori dose adjustment • Help regulators to register drugs with substantial side effect profiles
Figure 1 | Proposed clinical trial paradigm for the discovery of ‘toxgnostics’ markers though genome-wide association studies (GWAS). Schematic illustrating key steps in toxgnostic marker Nature Reviews | Cancer discovery and clinical implementation. The availability of high-quality standardized toxicity data within clinical trials enables accurate toxgnostic discovery, and outcome data ensure that markers or risk scores can be tested for association with outcome — ensuring that dose modification does not risk compromising treatment efficacy. Risk scores generally provide greater sensitivity and specificity than single gene biomarkers. SNP, single nucleotide polymorphism.
4 | ADVANCE ONLINE PUBLICATION
www.nature.com/reviews/cancer © 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES sample should also enable all patients to be analysed, rather than a subset. A systematic and agnostic approach to discovering toxicityassociated SNPs, which goes beyond the relatively rigid constructs of genes encoding determinants of ADME, embedded in well designed and adequately powered trials might lead to the discovery of high-performing molecular toxgnostics with sufficient clinical usefulness to change practice and improve chemotherapy safety, thereby saving money and substantially reducing morbidity and mortality. It is now time to integrate prog nostic biomarkers, companion diagnostics and toxgnostics into adjuvant treatment pro tocols for colorectal and other cancers, as we develop algorithms that embrace the totality of precision medicine and allow optimal patient stratification, drug and dose selection to improve the risk/benefit ratio to favour the patient as much as possible. David Church and Rachel Kerr are at the Oxford Cancer Centre, Department of Oncology, University of Oxford, Churchill Hospital, Old Road, Headington, Oxford, OX3 7LE, UK. David Church is also at the Molecular and Population Genetics Laboratory, The Wellcome Trust Centre for Human Genetics, University of Oxford, Roosevelt Drive, Oxford, OX3 7BN, UK. Enric Domingo, Dan Rosmarin, Claire Palles and Ian Tomlinson are at the Molecular and Population Genetics Laboratory, The Wellcome Trust Centre for Human Genetics, University of Oxford, Roosevelt Drive, Oxford, OX3 7BN, UK. Ian Tomlinson is also at the Genomic Medicine Theme, Oxford Comprehensive Biomedical Research Centre, The Wellcome Trust Centre for Human Genetics, University of Oxford, Roosevelt Drive, Oxford, OX3 7BN, UK. Kevin Maskell is at Oxford Cancer Biomarkers, The Magdalen Centre, Oxford Science Park, Robert Robinson Avenue, Oxford, OX4 4GA, UK. David Kerr is at the Nuffield Department of Clinical Laboratory Sciences, University of Oxford, John Radcliffe Hospital, Headley Way, Oxford, OX3 9DU, UK. Correspondence to D.K. e-mail: [email protected] doi:10.1038/nrc3729 Published online 15 May 2014 1. Gray, R. et al. Adjuvant chemotherapy versus observation in patients with colorectal cancer: a randomised study. Lancet 370, 2020–2029 (2007). 2. Widakowich, C. et al. Review: side effects of approved molecular targeted therapies in solid cancers. Oncologist 12, 1443–1455 (2007). 3. Bertagnolli, M. M. et al. Microsatellite instability and loss of heterozygosity at chromosomal location 18q: prospective evaluation of biomarkers for stages II and III colon cancer—a study of CALGB 9581 and 89803. J. Clin. Oncol. 29, 3153–3162 (2011). 4. Roth, A. D. et al. Prognostic role of KRAS and BRAF in stage II and III resected colon cancer: results of the translational study on the PETACC‑3, EORTC 40993, SAKK 60–00 trial. J. Clin. Oncol. 28, 466–474 (2010). 5. Hutchins, G. et al. Value of mismatch repair, KRAS, and BRAF mutations in predicting recurrence and benefits from chemotherapy in colorectal cancer. J. Clin. Oncol. 29, 1261–1270 (2011).
6. Romond, E. H. et al. Trastuzumab plus adjuvant chemotherapy for operable HER2‑positive breast cancer. N. Engl. J. Med. 353, 1673–1684 (2005). 7. Piccart-Gebhart, M. J. et al. Trastuzumab after adjuvant chemotherapy in HER2‑positive breast cancer. N. Engl. J. Med. 353, 1659–1672 (2005). 8. Sosman, J. A. et al. Survival in BRAF V600‑mutant advanced melanoma treated with vemurafenib. N. Engl. J. Med. 366, 707–714 (2012). 9. Amado, R. G. et al. Wild-type KRAS is required for panitumumab efficacy in patients with metastatic colorectal cancer. J. Clin. Oncol. 26, 1626–1634 (2008). 10. Van Cutsem, E. et al. Cetuximab and chemotherapy as initial treatment for metastatic colorectal cancer. N. Engl. J. Med. 360, 1408–1417 (2009). 11. Parkinson, D. R., Johnson, B. E. & Sledge, G. W. Making personalized cancer medicine a reality: challenges and opportunities in the development of biomarkers and companion diagnostics. Clin. Cancer Res. 18, 619–624 (2012). 12. Walther, A. et al. Genetic prognostic and predictive markers in colorectal cancer. Nature Rev. Cancer 9, 489–499 (2009). 13. Church, D., Midgley, R. & Kerr, D. Biomarkers in early-stage colorectal cancer: ready for prime time? Digestive Diseases 30 (Suppl. 2), 27–33 (2012). 14. Buyse, M., Sargent, D. J., Grothey, A., Matheson, A. & de Gramont, A. Biomarkers and surrogate end points—the challenge of statistical validation. Nature Rev. Clin. Oncol. 7, 309–317 (2010). 15. Simon, R. M., Paik, S. & Hayes, D. F. Use of Archived specimens in evaluation of prognostic and predictive biomarkers. J. Natl Cancer Institute 101, 1446–1452 (2009). 16. Ciccolini, J., Gross, E., Dahan, L., Lacarelle, B. & Mercier, C. Routine dihydropyrimidine dehydrogenase testing for anticipating 5‑fluorouracil-related severe toxicities: hype or hope? Clin. Colorectal Cancer 9, 224–228 (2010). 17. Locker, G. Y. et al. ASCO 2006 update of recommendations for the use of tumor markers in gastrointestinal cancer. J. Clin. Oncol. 24, 5313–5327 (2006). 18. Gold, H. T., Hall, M. J., Blinder, V. & Schackman, B. R. Cost effectiveness of pharmacogenetic testing for uridine diphosphate glucuronosyltransferase 1A1 before irinotecan administration for metastatic colorectal cancer. Cancer 115, 3858–3867 (2009). 19. Schmoll, H. J. et al. ESMO Consensus Guidelines for management of patients with colon and rectal cancer. a personalized approach to clinical decision making. Ann. Oncol. 23, 2479–2516 (2012). 20. Delea, T. E. et al. The incidence and cost of hospitalization for 5‑FU toxicity among Medicare beneficiaries with metastatic colorectal cancer. Value Health 5, 35–43 (2002). 21. Calhoun, E. A. et al. Evaluating the total costs of chemotherapy-induced toxicity: results from a pilot study with ovarian cancer patients. Oncologist 6, 441–445 (2001). 22. Wishart, D. S. Improving early drug discovery through ADME modelling: an overview. Drugs R. D. 8, 349–362 (2007). 23. Gamelin, E. et al. Long-term weekly treatment of colorectal metastatic cancer with fluorouracil and leucovorin: results of a multicentric prospective trial of fluorouracil dosage optimization by pharmacokinetic monitoring in 152 patients. J. Clin. Oncol. 16, 1470–1478 (1998). 24. Gamelin, E. C. et al. Relationship between 5‑fluorouracil (5‑FU) dose intensity and therapeutic response in patients with advanced colorectal cancer receiving infusional therapy containing 5‑FU. Cancer 77, 441–451 (1996). 25. Hillcoat, B. L., McCulloch, P. B., Figueredo, A. T., Ehsan, M. H. & Rosenfeld, J. M. Clinical response and plasma levels of 5‑fluorouracil in patients with colonic cancer treated by drug infusion. Br. J. Cancer 38, 719–724 (1978). 26. Di Paolo, A. et al. Relationship between 5‑fluorouracil disposition, toxicity and dihydropyrimidine dehydrogenase activity in cancer patients. Ann. Oncol. 12, 1301–1306 (2001). 27. Newell, D. R. Clinical pharmacokinetics of antitumor antifolates. Semin. Oncol. 26, 74–81 (1999).
NATURE REVIEWS | CANCER
28. Jolivet, J., Cowan, K. H., Curt, G. A., Clendeninn, N. J. & Chabner, B. A. The pharmacology and clinical use of methotrexate. N. Engl. J. Med. 309, 1094–1104 (1983). 29. Calvert, A. H. et al. Carboplatin dosage: prospective evaluation of a simple formula based on renal function. J. Clin. Oncol. 7, 1748–1756 (1989). 30. Lehmann, H. & Ryan, E. The familial incidence of low pseudocholinesterase level. Lancet 271, 124 (1956). 31. Alving, A. S., Carson, P. E., Flanagan, C. L. & Ickes, C. E. Enzymatic deficiency in primaquinesensitive erythrocytes. Science 124, 484–485 (1956). 32. Vogel, F. Moderne Probleme der Humangenetik. Ergebn. Inn. Med. Kinderheilkd. 12, 52–125 (1959). 33. Lennard, L., Lilleyman, J. S., Van Loon, J. & Weinshilboum, R. M. Genetic variation in response to 6‑mercaptopurine for childhood acute lymphoblastic leukaemia. Lancet 336, 225–229 (1990). 34. Relling, M. V. et al. Clinical Pharmacogenetics Implementation Consortium guidelines for thiopurine methyltransferase genotype and thiopurine dosing. Clin. Pharmacol. Ther. 89, 387–391 (2011). 35. Ford, L. T. & Berg, J. D. Thiopurine S‑methyltransferase (TPMT) assessment prior to starting thiopurine drug treatment; a pharmacogenomic test whose time has come. J. Clin. Pathol. 63, 288–295 (2010). 36. Ross, C. J. et al. Genetic variants in TPMT and COMT are associated with hearing loss in children receiving cisplatin chemotherapy. Nature Genet. 41, 1345–1349 (2009). 37. Pussegoda, K. et al. Replication of TPMT and ABCC3 genetic variants highly associated with cisplatininduced hearing loss in children. Clin. Pharmacol. Ther. 94, 243–251 (2013). 38. Yang, J. J. et al. The role of inherited TPMT and COMT genetic variation in cisplatin-induced ototoxicity in children with cancer. Clin. Pharmacol. Ther. 94, 252–259 (2013). 39. Ratain, M. J., Cox, N. J. & Henderson, T. O. Challenges in interpreting the evidence for genetic predictors of ototoxicity. Clin. Pharmacol. Ther. 94, 631–635 (2013). 40. Zanger, U. M. & Schwab, M. Cytochrome P450 enzymes in drug metabolism: regulation of gene expression, enzyme activities, and impact of genetic variation. Pharmacol. Ther. 138, 103–141 (2013). 41. Mahgoub, A., Idle, J. R., Dring, L. G., Lancaster, R. & Smith, R. L. Polymorphic hydroxylation of Debrisoquine in man. Lancet 2, 584–586 (1977). 42. Heim, M. & Meyer, U. A. Genotyping of poor metabolisers of debrisoquine by allele-specific PCR amplification. Lancet 336, 529–532 (1990). 43. Kupfer, A. & Preisig, R. Pharmacogenetics of mephenytoin: a new drug hydroxylation polymorphism in man. Eur. J. Clin. Pharmacol. 26, 753–759 (1984). 44. Sullivan-Klose, T. H. et al. The role of the CYP2C9‑Leu359 allelic variant in the tolbutamide polymorphism. Pharmacogenetics 6, 341–349 (1996). 45. Jonas, D. E. & McLeod, H. L. Genetic and clinical factors relating to warfarin dosing. Trends Pharmacol. Sci. 30, 375–386 (2009). 46. Schwab, M. & Schaeffeler, E. Warfarin pharmacogenetics meets clinical use. Blood 118, 2938–2939 (2011). 47. Klein, T. E. et al. Estimation of the warfarin dose with clinical and pharmacogenetic data. N. Engl. J. Med. 360, 753–764 (2009). 48. Schroth, W. et al. Breast cancer treatment outcome with adjuvant tamoxifen relative to patient CYP2D6 and CYP2C19 genotypes. J. Clin. Oncol. 25, 5187–5193 (2007). 49. Hertz, D. L. et al. CYP2C8*3 predicts benefit/risk profile in breast cancer patients receiving neoadjuvant paclitaxel. Breast Cancer Res. Treat. 134, 401–410 (2012). 50. Hertz, D. L. et al. CYP2C8*3 increases risk of neuropathy in breast cancer patients treated with paclitaxel. Ann. Oncol. 24, 1472–1478 (2013). 51. Iyer, L. et al. Genetic predisposition to the metabolism of irinotecan (CPT‑11). Role of uridine diphosphate glucuronosyltransferase isoform 1A1 in the glucuronidation of its active metabolite (SN‑38) in human liver microsomes. J. Clin. Invest. 101, 847–854 (1998).
ADVANCE ONLINE PUBLICATION | 5 © 2014 Macmillan Publishers Limited. All rights reserved
PERSPECTIVES 52. Ando, Y. et al. UGT1A1 genotypes and glucuronidation of SN‑38, the active metabolite of irinotecan. Ann. Oncol. 9, 845–847 (1998). 53. Hoskins, J. M., Goldberg, R. M., Qu, P., Ibrahim, J. G. & McLeod, H. L. UGT1A1*28 genotype and irinotecan-induced neutropenia: dose matters. J. Natl Cancer Inst. 99, 1290–1295 (2007). 54. Schwab, M. et al. Role of genetic and nongenetic factors for fluorouracil treatment-related severe toxicity: a prospective clinical trial by the German 5‑FU Toxicity Study Group. J. Clin. Oncol. 26, 2131–2138 (2008). 55. Loganayagam, A. et al. Pharmacogenetic variants in the DPYD, TYMS, CDA and MTHFR genes are clinically significant predictors of fluoropyrimidine toxicity. Br. J. Cancer 108, 2505–2515 (2013). 56. McLeod, H. L. et al. Pharmacogenetic predictors of adverse events and response to chemotherapy in metastatic colorectal cancer: results from North American Gastrointestinal Intergroup Trial N9741. J. Clin. Oncol. 28, 3227–3233 (2010). 57. Burmeister, H., Aebi, S., Studer, C., Fey, M. F. & Gautschi, O. Adherence to ESMO clinical recommendations for prophylaxis of chemotherapyinduced nausea and vomiting. Supportive Care Cancer 20, 141–147 (2012). 58. Watanabe, R. M. Statistical issues in gene association studies. Methods Mol. Biol. 700, 17–36 (2011). 59. Dunlop, M. G. et al. Common variation near CDKN1A, POLD3 and SHROOM2 influences colorectal cancer risk. Nature Genet. 44, 770–776 (2012). 60. Houlston, R. S. et al. Meta-analysis of three genomewide association studies identifies susceptibility loci for colorectal cancer at 1q41, 3q26.2, 12q13.13 and 20q13.33. Nature Genet. 42, 973–977 (2010). 61. Daly, A. K. et al. HLA‑B*5701 genotype is a major determinant of drug-induced liver injury due to flucloxacillin. Nature Genet. 41, 816–819 (2009). 62. McCormack, M. et al. HLA‑A*3101 and carbamazepine-induced hypersensitivity reactions in Europeans. N. Engl. J. Med. 364, 1134–1143 (2011). 63. Link, E. et al. SLCO1B1 variants and statin-induced myopathy—a genomewide study. N. Engl. J. Med. 359, 789–799 (2008). 64. Trevino, L. R. et al. Germline genetic variation in an organic anion transporter polypeptide associated with methotrexate pharmacokinetics and clinical effects. J. Clin. Oncol. 27, 5972–5978 (2009). 65. Ramsey, L. B. et al. Rare versus common variants in pharmacogenetics: SLCO1B1 variation and methotrexate disposition. Genome Res. 22, 1–8 (2012). 66. Kiyotani, K. et al. A genome-wide association study identifies four genetic markers for hematological toxicities in cancer patients receiving gemcitabine therapy. Pharmacogenet. Genomics 22, 229–235 (2012). 67. Srinivasan, Y. et al. Genome-wide association study of epirubicin-induced leukopenia in Japanese patients. Pharmacogenet. Genomics 21, 552–558 (2011).
68. Meyerson, M. et al. Advances in understanding cancer genomes through second-generation sequencing. Nat. Rev. Genet. 11, 685–696 (2010). 69. McDermott, U., Downing, J. R. & Stratton, M. R. Genomics and the continuum of cancer care. N. Engl. J. Med. 364, 340–350 (2011). 70. Liu, J. et al. Predictive Power Estimation Algorithm (PPEA)—a new algorithm to reduce overfitting for genomic biomarker discovery. PLoS ONE 6, e24233 (2011). 71. McShane, L. M. et al. Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK). J. Natl Cancer Institute 97, 1180–1184 (2005). 72. Riley, R. D. et al. Prognosis Research Strategy (PROGRESS) 2: prognostic factor research. PLoS Med. 10, e1001380 (2013). 73. Martinez-Balibrea, E. et al. UGT1A and TYMS genetic variants predict toxicity and response of colorectal cancer patients treated with first-line irinotecan and fluorouracil combination therapy. Br. J. Cancer 103, 581–589 (2010). 74. Glimelius, B. et al. Prediction of irinotecan and 5‑fluorouracil toxicity and response in patients with advanced colorectal cancer. Pharmacogenomics J. 11, 61–71 (2011). 75. Cecchin, E. et al. Predictive role of the UGT1A1, UGT1A7, and UGT1A9 genetic variants and their haplotypes on the outcome of metastatic colorectal cancer patients treated with fluorouracil, leucovorin, and irinotecan. J. Clin. Oncol. 27, 2457–2465 (2009). 76. Pharoah, P. D., Abraham, J. & Caldas, C. Re: CYP2D6 genotype and tamoxifen response in postmenopausal women with endocrine-responsive breast cancer: the Breast International Group 1–98 trial and Re: CYP2D6 and UGT2B7 genotype and risk of recurrence in tamoxifen-treated breast cancer patients. J. Natl. Cancer Inst. 104, 1263–1264; author reply 1266 –1268 (2012). 77. Regan, M. M. et al. CYP2D6 genotype and tamoxifen response in postmenopausal women with endocrineresponsive breast cancer: the breast international group 1–98 trial. J. Natl Cancer Inst. 104, 441–451 (2012). 78. Salazar, R. et al. Gene expression signature to improve prognosis prediction of stage II and III colorectal cancer. J. Clin. Oncol. 29, 17–24 (2011). 79. Gray, R. G. et al. Validation study of a quantitative multigene reverse transcriptase-polymerase chain reaction assay for assessment of recurrence risk in patients with stage II colon cancer. J. Clin. Oncol. 29, 4611–4619 (2011). 80. Rosmarin, D. et al. Genetic markers of toxicity from capecitabine and other fluorouracil-based regimens: investigation in the QUASAR2 study, systematic review, and meta-Analysis. J. Clin. Oncol. 32, 1031–1039 (2014). 81. Longley, D. B., Harkin, D. P. & Johnston, P. G. 5‑fluorouracil: mechanisms of action and clinical strategies. Nature Rev. Cancer 3, 330–338 (2003).
6 | ADVANCE ONLINE PUBLICATION
82. Diasio, R. B., Beavers, T. L. & Carpenter, J. T. Familial deficiency of dihydropyrimidine dehydrogenase. Biochemical basis for familial pyrimidinemia and severe 5‑fluorouracil-induced toxicity. J. Clin. Invest. 81, 47–51 (1988). 83. Harris, B. E., Carpenter, J. T. & Diasio, R. B. Severe 5‑fluorouracil toxicity secondary to dihydropyrimidine dehydrogenase deficiency. A potentially more common pharmacogenetic syndrome. Cancer 68, 499–501 (1991). 84. Etienne, M. C. et al. Population study of dihydropyrimidine dehydrogenase in cancer patients. J. Clin. Oncol. 12, 2248–2253 (1994). 85. van Staveren, M. C., Jan Guchelaar, H., van Kuilenburg, A. B., Gelderblom, H. & Maring, J. G. Evaluation of predictive tests for screening for dihydropyrimidine dehydrogenase deficiency. Pharmacogenomics J. 13, 389–395 (2013). 86. Pullarkat, S. T. et al. Thymidylate synthase gene polymorphism determines response and toxicity of 5‑FU chemotherapy. Pharmacogenomics J. 1, 65–70 (2001). 87. Fernandez-Rozadilla, C. et al. Pharmacogenomics in colorectal cancer: a genome-wide association study to predict toxicity after 5‑fluorouracil or FOLFOX administration. Pharmacogenomics J. 13, 209–217 (2013). 88. Baldwin, R. M. et al. A genome-wide association study identifies novel loci for paclitaxel-induced sensory peripheral neuropathy in CALGB 40101. Clin. Cancer Res. 18, 5099–5109 (2012). 89. Twelves, C. et al. Capecitabine as adjuvant treatment for stage III colon cancer. New Engl. J. Med. 352, 2696–2704 (2005). 90. Kuderer, N. M., Dale, D. C., Crawford, J., Cosler, L. E. & Lyman, G. H. Mortality, morbidity, and cost associated with febrile neutropenia in adult cancer patients. Cancer 106, 2258–2266 (2006). 91. Ahmed, G. et al. Cost implications of reactive versus prospective testing for dihydropyrimidine dehydrogenase (DPD) deficiency in patients with colorectal cancer. J. Clin. Oncol. 31, (Suppl.; Abstr 3627) (2013). 92. Blay, J. Y. et al. Early lymphopenia after cytotoxic chemotherapy as a risk factor for febrile neutropenia. J. Clin. Oncol. 14, 636–643 (1996). 93. Crawford, J., Dale, D. C. & Lyman, G. H. Chemotherapy-induced neutropenia: risks, consequences, and new directions for its management. Cancer 100, 228–237 (2004).
Acknowledgements
Core funding to the Wellcome Trust Centre for Human Genetics was provided by the Wellcome Trust (090532/Z/09/Z).
Competing interests statement
The authors declare competing interests: see Web version for details.
www.nature.com/reviews/cancer © 2014 Macmillan Publishers Limited. All rights reserved
