A Strategy for PGx Discovery – Millions of Variants: Where to Start?
To appreciate the need for advancement in pharmacogenomics (PGx), let’s start by taking an inventory of the success in this space since the approval of Herceptin (the first drug with PGx / biomarker information in its label). A review of the FDA’s Pharmacogenomic Biomarkers in Drug Labeling Table revealed only 12% of drugs since Herceptin had PGx / biomarker information in their label and only 14 of these labels direct clinicians to utilize testing prior to prescription. Clearly, there is need for improvement in PGx-driven patient selection for therapeutic intervention; however, the current paradigm and PGx analytics are failing to produce biomarkers with clinical and commercial utility.
Part of the problem is that most compounds enter Phase II/III clinical development with a weak or hypothesis-generating PGx program due to a lack of solid a priori information on the drug-gene relationship with clinical response. While there are other avenues to conduct PGx discovery, the majority of this type of translational research is occurring in the clinical trial setting with the corresponding unique stressors: (1) relatively small, finite sample sizes; (2) ethnically, and genetically, diverse clinical cohorts; (3) heterogeneous patient populations from trial to trial (potentially impacting ability to replicate findings); (4) potential polygenic nature of drug response; and (5) business implications of cost associated with both false positives and false negatives.
To realize the potential of translational research, advancements need to be made in how we approach PGx discovery. The roadblock to progress in PGx studies is no longer in obtaining high dimensional genetic data on patient populations, but rather how to effectively harness the wealth of information available. Consider a PGx study with DNA samples assayed using one of many SNP genotyping assays in which the objective is to identify genes / variants associated with treatment-specific response. Traditionally, PGx discovery has directly searched for the proverbial needle in a haystack, i.e., a single SNP; however, this approach has failed to produce any DNA-based classifier for drug efficacy and generally has a low rate of replication even for association in a confirmation trial population. Alternatively, a gene is a biologically relevant unit of genetic variation with boundaries that are independent of ancestry and generalizable across studies.
Recent research has produced various methods to jointly test the set of SNPs in a gene, many of which are based in a variance components framework where the genotypes are modeled by a kernel function.1-3 These genomic region-based testing approaches have been shown to be more powerful (by leveraging the underlying genetic architecture and reducing the multiplicity burden), better suited for the complexities of the clinical trial setting, flexible at handling both SNPs and rare variation, and capable of producing actionable results for use in therapeutic development. To demonstrate value, a comparison of single SNP testing and genomic region-based testing was conducted for three ‘proof-of-concept’ scenarios. Results were recently presented at the Biomarker Summit in San Diego. Whether evaluating the impact of genomic variation on treatment response or informing development of similar compounds, genomic region-based testing has the potential to increase the value of your portfolio and ultimately impact patient care.
- Wu, M. C., Kraft, P., Epstein, M. P., Taylor, D. M., Chanock, S. J., Hunter, D. J., & Lin, X. (2010). Powerful SNP-set analysis for case-control genome-wide association studies. The American Journal of Human Genetics, 86(6), 929-942.
- Ionita-Laza, I., Lee, S., Makarov, V., Buxbaum, J. D., & Lin, X. (2013). Sequence kernel association tests for the combined effect of rare and common variants. The American Journal of Human Genetics, 92(6), 841-853.
- Qu, L., Guennel, T., & Marshall, S. L. (2013). Linear Score Tests for Variance Components in Linear Mixed Models and Applications to Genetic Association Studies. Biometrics, 69(4), 883-892.