Pharmacogenetic studies of T2D therapies lags significantly behind other complex diseases, despite the fact that pharmacologic interventions for T2D have been studied extensively at both the clinical and epidemiologic levels. For example, the sulphonylurea class of drugs has been in clinical use since the 1950s with the so-called “second generation” sulphonylureas approved for use in the US in 1984. Despite this long history, a simple PubMed search using the terms “pharmacogenetics” and “diabetes” yields only 88 original scientific papers spanning a 43 year period (1967-2010). If one replaces “diabetes” with “cardiovascular disease”, the same search yields 330 original scientific papers covering a 42 year period (1968-2010). Sadly, a search of review articles for diabetes pharmacogenetics yields 81 reviews, suggesting that while the basic idea of pharmacogenetics has percolated in scientific minds for decades, there has been little translation into the clinical research setting.
Rapid advances in genomic technologies have revolutionized studies of human genetics. As of this writing, 38 loci underlying susceptibility to T2D have been identified, mostly in populations of northern European ancestry [206
]. In contrast to the very small effect sizes of diabetes susceptibility loci, effect sizes for response to medications or adverse events may be substantially larger (2- to 50-fold), thus making it feasible to perform genome-wide association studies for such phenotypes without the need to obtain extremely large sample sizes. This point was illustrated by Nelson and colleagues who demonstrated that statistically significant, genome-wide associations can be detected with sample sizes in the low hundreds for certain adverse events [185
]. Thus, it should be possible to take existing drug trial data where DNA is available and perform genome-wide association analyses, much in the manner that was done for T2D. However, given the relatively smaller number of drug trials performed using identical patient ascertainment and treatment protocols, replication of primary findings from pharmacogenetics-based GWASs may be problematic. Replication of primary GWAS signals should pay careful attention to heterogeneity analysis or consider using a random effects meta-analysis rather than a fixed effects approach that has been traditionally implemented for disease-based GWAS. Regardless, DNA should be routinely collected in future drug trials to facilitate pharmacogenomic studies.
That said, the era of genome-wide association may be short-lived given the advent of “next generation” sequencing [218
]. Although this technology, and its variations like RNA-sequencing, is currently expensive, the cost-per-sample is rapidly dropping and soon may be the standard technology for assessing human genetic variation. Such technologies can provide rapid interrogation of >99% of the human genome covering both common and rare variants that may contribute to drug response or adverse events. Indeed, like complex diseases, it is likely that rare variants of high penetrance may contribute to drug response and sequencing is currently the only technology that allows efficient identification of such variation. Obviously, such technology would be overkill in the clinical setting, but is an obvious research tool to identify genetic variants that may be predictive of drug response. If predictive variants can be identified, it should be possible to develop a specific SNP chip that would be more appropriate for the clinical setting. In addition, one should consider the potential contributions of other genetic variation, such as copy number variants, insertion/deletions, and epigenetic modification, to drug response.
Regardless of what technologies are applied to pharmacogenetic studies, a critical feature of any pharmacogenetic study of T2D is the phenotype. As noted above, a variety of endpoints have been used to define response to T2D therapies. The most common has been changes in fasting glucose and/or HbA1C. This is a logical choice from a clinical perspective, given that the ultimate goal for a clinician is normalization of glycemia. However, this may not be the optimal choice to understand the role of genetic variation on drug response. First, as noted above, each class of T2D therapies operates through different mechanisms of action, including insulin secretion, hepatic glucose output, insulin sensitivity, etc.
Therefore, any change in fasting glucose or HbA1c is secondary to the drug effect at the molecular target and changes in these metrics may not accurately reflect the actions of the drug, given that multiple mechanisms are simultaneously working to regulate glucose levels. This effect is best illustrated in the examination of changes in fasting or 2-hour glucose levels as individuals progress towards T2D. Buchanan and colleagues were among the first to show that both fasting and 2-hour glucose rise minimally in individuals who did not develop T2D over a 5-year follow-up period; 1 mg/dl per year and 4 mg/dl per year, respectively [187
]. Although changes in these metrics were larger in individuals who eventually developed T2D over the same 5-year follow-up (19 mg/dl per year and 28 mg/dl per year, respectively), in both groups glucose levels were maintained in a relatively narrow range, mostly due to compensatory insulin secretion. However, as β-cell compensation approached and then fell below 10% of normal, there was a rapid rise in both fasting and 2-hour glucose in those individuals who developed T2D. The same pattern of changes in glucose was observed in Pima Indians [186
]. Therefore, accurate assessment of the specific physiologic parameter targeted by the given T2D therapy may be the optimal measure of drug response compared to fasting glucose or HbA1c.
Finally, genetic associations only provide information regarding specific genetic markers that may be predictive of drug efficacy. To date, association studies have not formally assessed specificity or sensitivity. While these metrics are not fully informative in genetic association studies, they do provide some information on the utility of a given marker as a predictive tool. Furthermore, there has not been a study to jointly examine all variants for a given therapy to assess whether the joint information accounts for a greater proportion of the variability in drug response compared to the individual markers alone. Clearly, prospective studies testing the power of genetic markers to predict drug response are requisite to fully endorse their introduction into the clinical care setting.