Personalizing Supportive Care: Pharmacogenomics and Risk Prediction

Personalizing Supportive Care: Pharmacogenomics and Risk Prediction

By Stephen T. Sonis, DMD, DMSc

Article Highlights

  • Optimal treatment of patients with cancer will depend on accurate regimen-related risk assessment prior to the initiation of treatment.
  • Of the elements that potentially affect the likelihood of toxicity, genomics has a major, consistent, and reproducible influence as a risk prognosticator. Genes that govern metabolism and pathogenesis are key drivers of risk.
  • Accurate discovery, validation, and clinical implementation of genomic predictors present broad challenges, such as the logistics of creating large-scale studies, clinician confidence in and the understanding of pharmacogenomic-predictive tools, and third-party payer reimbursement for genomic testing.

The importance of treatment-related side effects in impeding effective cancer therapy was recognized in the recently articulated “cancer moonshot” initiative. Regimen-related toxicities impact the overall success of treatment outcomes—in some sense, a tail capable of wagging the dog. To truly impact the tolerability of treatment, effective, personalized, biologically-based prevention of side effects is more desirable than reactive palliative management. Not only does such an approach mitigate the miseries associated with cytotoxic or targeted regimens in a cost-effective way, but it opens a window for even more aggressive and definitive cancer treatment.

Obstacles to Toxicity Prophylaxis

One impediment to toxicity prophylaxis is knowing who to treat. We know that in the vast majority of patients with cancer, the distribution and incidence of treatment complications are uneven. Although some patients might suffer from a hurricane of toxicities that affect their ability to tolerate treatment, disrupt their lifestyle, and decrease their quality of life, others sail through cancer therapy with barely a ripple in their lives. We have now come to understand that the determinants of toxicity risk are not random, thus creating an opportunity for us to define a patient’s likelihood of potential treatment complications before administering a drug or therapy.

But knowing who is at risk of specific toxicities is of little value unless there are treatment alternatives or effective interventions to prevent the toxicity from occurring. And, even if there are treatment alternatives or toxicity interventions, their efficacy will not be consistent across all individuals, as is the case with all drugs. Thus, it would be especially valuable to both identify the at-risk population and then individualize the most effective treatment for each patient. One cannot escape the supposition that treating everyone in the same way simply makes no sense.

This conclusion is no epiphany; the desirability to personalize or individualize treatment is longstanding. And with the emphasis on increasing efficiencies in health care outcomes, this goal is on the U.S. national political radar, as evidenced by the mention of a Precision Medicine Initiative during the 2015 State of the Union Address. Interestingly, of the $215 million designated in the President’s budget for this effort, $70 million is allocated to the National Cancer Institute (NCI) to “scale up efforts to identify genomic drivers of cancer and apply that knowledge to the development of more effective approaches to cancer treatment.”1 Although it is unknown how much of this $70 million, if any, will end up supporting efforts to define regimen-related toxicity risk or individual responses to interventions, now is an opportune time to incorporate elements of supportive care into a comprehensive oncology precision medicine paradigm.

Genomics as a Risk Prognosticator

Advances in the science, epidemiology (including molecular epidemiology), and pharmacokinetics of regimen-related toxicities have now converged at a time when technology and computational analytics have developed to a level that permits integration of these disparate elements into a useable format that can effectively predict toxicity risk. Furthermore, as interest and research in regimen-related toxicities have increased, the number of factors that are associated with risk has increased exponentially. A PubMed query with the keywords “toxicity risk and chemotherapy” will generate more than 16,000 hits, and the same exercise with “radiotherapy” results in 4,590. Given the vast number of elements that seem to have the potential to influence toxicity risk, a major practical challenge is the identification of a limited number of core predictors that can be consistently and reliability assessed prior to treatment. This exercise is called “feature selection,” and it should be a critical component of any risk-estimate algorithm.

Of the elements that potentially affect toxicity risk, genomics has a major, consistent, and reproducible influence as a risk prognosticator in at least two ways: first, it affects drug metabolism, and second, genomics’ control of pathways associated with the pathogenesis of regimen-related injury is profound. From the standpoint of risk determination, genomic impact on drug metabolism (pharmacokinetics) is the most linear and straightforward.

Pharmacokinetics

Genes that govern metabolism are associated with the pharmacokinetics of medications and play a major role in affecting the levels of available/active drugs. We’re all familiar with a typical pharmacokinetics curve in which blood levels peak and then disappear over time, with the latter being dependent on the drug’s half-life. That typical curve is based on the “normal” metabolism of the drug. But what if a patient is deficient in the drug’s metabolizing enzyme? Not only might the peak level of the drug be higher than expected, but the drug’s elimination will be slowed. And if the patient is scheduled to have sequential infusions of the drug, it accumulates to levels that increase the likelihood of toxicity.

A classic oncology example can occur among patients receiving 5-fluorouracil (5-FU) who have an aberrant allele associated with dihydropyrimidine dehydrogenase, the gene that initially catabolizes 5-FU. Because this mutation results in ineffective metabolism of the drug, the risk of 5-FU–related toxicity skyrockets in these individuals. The availability of a gene-based test to identify gene variations associated with 5-FU toxicity represents an early successful effort to translate genomic toxicity risk prediction to the clinic.

Other similar examples have been reported for a number of cancer drugs. Nonetheless, the percentage of the population that suffers gene-associated metabolic deficiencies is relatively small (< 10%)—certainly less than the reported incidence of a wide range of toxicities.2

Risk-Predicting Genes

The identification of genes controlling the fate of drugs has been relatively easy compared to discovering risk-predicting genes or single nucleotide polymorphisms (SNPs) related to pathogenesis. Whereas enzymes are often controlled by one or two critical genes, mechanistic pathways function as the result of a concert of interacting genes whose function is defined by a specific sequence of expression. Often, assessing the presence of a gene associated with a specific protein can be misleading relative to risk. Consequently, the variability of activity of a specific biologic pathway noted between one patient and another is not the result of a single gene, but rather it is a manifestation of the cumulative activity of a team of genes.

This complexity has presented challenges to discovering risk predictors. The results of candidate gene approaches, in which a group of experts hypothesize the genes of choice to evaluate, have been consistently disappointing. Alternatively, predictive SNPs or genes identified by genome-wide association studies have been notoriously difficult to reproduce. Fortunately, the application of innovative analytical approaches and software and the availability of the Cloud and big data have resulted in studies in which predictive genomics have been validated for a number of toxicities.

Validation of Innovative Analytical Approaches

Validation and enhancement of these results present broad challenges and opportunities for collaborations to confirm, expand, and clinically implement genome-based risk prediction. Typically, accurate discovery and validation of genomic predictors require large numbers of patients, so as to include individuals who develop the toxicity of interest and those who did not. The requirement for large numbers is magnified when we deal with risk in the broad and real-world context in which clusters of mechanistically similar side effects are the norm and not the exception. As past experience has taught us, the accuracy and reproducibility of genome-based risk prediction studies requires large numbers of patients—both to discover the “right” genes or SNPs and to validate their predictive accuracy.

An obvious challenge focuses on the logistics of accruing sufficient patient numbers for such studies. One approach is being successfully implemented by the Radiogenomics Consortium. With seed funding from the NCI’s Division of Cancer Control and Population Sciences, this international consortium seeks to “expand knowledge of the genetic basis for differences in radiosensitivity and to develop assays to help predict the susceptibility of [patients with] cancer for the development of adverse effects resulting from radiotherapy.”3

Alternatively, industry-sponsored research of large numbers of patients, coupled with new analytics that provide for more scientifically efficient outcomes, has been effectively applied to identify genomic risk factors for both chemo- and radiotherapy-induced toxicities.

Three important considerations arise as we transit from theoretical use of genomic risk predictors into a routine clinical asset. First, some critics of pharmacogenomic-predictive tools suggest that clinicians might rely totally on them to guide treatment to the exclusion of other factors. Clearly, that is not the intention. The information garnered by such tests should, conversely, provide clinicians and patients with more data around which informed, personalized treatment selections can be based in the framework of patient preferences and other factors. Second, although oncologists are among the specialists most comfortable with understanding, using, and discussing genomic data, there appears to be a need and opportunity to increase provider confidence and knowledge with this information. A challenge for any commercialized genomic test for regimen-related toxicity risk will be to provide information clearly and in an easily interpretable and actionable format. Third, we are in the midst of vast economic transitions in health care. If genomic testing is to become clinically viable, third-party payer reimbursement will be critical. The value proposition for genomic testing must be unambiguously defined such that the incremental cost of such tests results is overshadowed by the savings achieved by individualized use of toxicity prophylaxis.

Individualizing cancer care has become a banner objective for clinicians, patients, and payers. Integrating this approach so that it can be effectively and proactively applied to mitigate regimen-related toxicities makes enormous sense in the comprehensive treatment of patients with cancer.