On the Verge of Precision Medicine in Diabetes

Abstract

The epidemic of type 2 diabetes (T2D) is a significant global public health challenge and a major cause of morbidity and mortality. Despite the recent proliferation of pharmacological agents for treatment of T2D, current therapies simply treat the symptom – hyperglycemia – and do not directly address the underlying disease process or modify the disease course. This article summarizes how genomic discovery has contributed to unraveling the heterogeneity in T2D, reviews relevant discoveries in the pharmacogenetics of five commonly prescribed glucose-lowering agents, presents evidence supporting how pharmacogenetics can be leveraged to advance precision medicine, and calls attention to important research gaps to its implementation to guide treatment choices.

1. Introduction

Diabetes is a major threat to public health, with over half a billion individuals living with the disease worldwide [1]. This number is projected to increase to 783.2 million globally in 2045 [2]. Type 2 diabetes (T2D), which accounts for approximately 90% of this total, is diagnosed based on the presence of hyperglycemia and without specific considerations to the underlying pathophysiology. Because elevated glycemia can be the sequelae of multiple pathogenic processes to varying degrees, including rising insulin resistance and defective insulin secretion (due to β-cell dysfunction or loss), T2D is a heterogeneous disease [3]. Disease severity and the need for escalation of therapy can differ greatly across individuals, as does the likelihood of progression to microvascular and macrovascular complications. Thus, the “palette model” of T2D has been proposed, in which each person has individual risks for dysregulation in one or more component pathways that contribute to the overall development of T2D [4]. This model can also be applied to the development of diabetic complications and organ dysfunction, as a result of disease progression.

Two main approaches have emerged to characterize the interindividual variability of T2D and identify subtypes that might be responsive to therapies targeting a specific pathophysiology. The first involves utilizing a limited set of clinical parameters to subclassify patients. Here, six variables (glutamate decarboxylase antibodies, age at diagnosis, body mass index [BMI], glycated hemoglobin [HbA1c], and homoeostatic model assessment 2 [HOMA2] estimates of β-cell function and insulin resistance) were used to cluster patients with new-onset diabetes into five reproducible subgroups that differed in their risk of diabetic complications and progression to insulin use [5, 6]. Other efforts have applied clustering methods to genome-wide association study (GWAS) results for T2D genetic variants and diabetes-related traits [7, 8]. Though this strategy minimizes the contribution of environmental risk factors, it produces subgroups reflecting genetically driven pathways that predispose individuals differentially to T2D-related metabolic disease outcomes [9].

Regardless of the approach, a deeper understanding of the multiple etiological processes contributing to T2D will permit pharmacological regimens to be tailored to the individual and appropriate recommendations to be developed for monitoring drug response and progression of disease. In the sections that follow, we review the contribution of genetic analyses and identification of individual polymorphisms and polygenic scores in understanding T2D risk and disease progression. We describe the current state of pharmacogenetics in T2D and summarize existing literature on responses to common glucose-lowering medications. Finally, we outline the challenges of and future directions for using genetic information to guide drug selection and advance precision medicine in T2D.

2. Genetic risk for T2D and implications for pharmacotherapy

2.1. GWAS

As a result of advances in genotyping technologies and the continued reduction in their cost, there has been an explosion of genomic discovery in recent years as it pertains to identifying genetic risk loci for T2D. To date, GWAS have discovered over 700 independent loci associated with T2D susceptibility, with an estimated genome-wide chip heritability of 19% [10, 11]. Given that the overall genetic contribution to T2D is estimated to lie in the 30-40% range, this result implies that as of the time of this writing we have captured over 50% of the genetic determinants for the disease. Other efforts have uncovered additional loci through examining diabetes-related quantitative traits, such as fasting glucose, fasting insulin, and 2-hour glucose following an oral glucose tolerance test (OGTT) [12]. Interestingly, many of the early T2D risk loci were found to impact β-cell function [13], demonstrating that variants that dually influence T2D risk and glucose homeostasis may provide insight into genetic defects related to insulin secretion or insulin action contributing to T2D development.

One major challenge is that many genetic signals reside in non-coding regions between genes, making it difficult to elucidate the mechanisms underlying these associations or prove causality. The causal variant and mechanism of action are better understood for some variants, such as rs7903146, an intron in the transcription factor 7-like gene (TCF7L2) and the most robustly and consistently replicated T2D risk locus with a per-allele odds ratio of ~1.4 [14]. Through follow-up molecular and physiological studies, this variant has been implicated in several mechanisms, including reduced β-cell mass, diminished insulin secretion, and alterations in the incretin response [15–17]. For many other associations, however, additional functional studies are needed to understand their biological relevance.

2.2. Polygenic scores

Because individual genetic loci associated with T2D have modest effect sizes, aggregating variants into a polygenic score can provide increased power to improve the prediction of future disease risk [18] and has the advantage of being able to be utilized from birth to infer the risk of disease before other risk factors emerge. In this way, the earliest polygenic scores incorporated a stringent set of genome-wide significant markers that represent true reproducible associations with T2D. However, whereas one such score developed with 34 T2D-associated variants was associated with an increased risk of progression to diabetes in the Diabetes Prevention Program (DPP), a lifestyle intervention attenuated this risk [19].

In recent years, new computational methods have permitted the construction of global extended polygenic scores (gePS), testing millions of sub-significant variants across the genome, weighted by effect size estimates obtained from large-scale discovery GWAS [18]. gePS have been increasingly shown to provide predictive ability equivalent to conventional clinical risk factors, with greater utility at the extremes of the population [20]. For instance, the top 2.5% of a gePS built from 4.6 million common variants in the UK Biobank had a 3.4-fold increased risk of T2D compared to those from the middle of the distribution and a 9.4-fold increased risk compared to the bottom 2.5% [21]. Individuals with a significantly elevated risk of T2D at the extreme end of the distribution may represent a subgroup that would benefit from diabetes prevention strategies or early pharmacologic intervention.

As previously described, clustering methods have aided in the development of partitioned polygenic scores (pPS), which classify genetic variants based on biological pathways contributing to T2D risk (e.g., β-cell function, obesity). One then can envision the potential of these scores to elucidate patterns of disease pathogenesis and aid in selecting a pharmacologic intervention that directly targets the underlying pathophysiological defect. However, published genetic clusters utilize only a subset of variants with the strongest genome-wide significant associations with T2D, in part due to the limitation of available GWAS summary statistics across multiple diabetes-related traits. As GWAS for these traits increase in number and expand to non-European ancestral groups, more robust clusters can be generated that include a greater number of variants and may have greater clinical translatability.

3. Pharmacogenetics in T2D

Traditionally, consensus guidelines for the treatment of T2D have recommended metformin as the first-line therapy due to its good safety profile, low cost, and glycemic efficacy [22]. The decision to add a second-line agent is based on inadequate glycemic control, typically on metformin. Apart from atherosclerotic cardiovascular or chronic kidney disease, for which glucagon-like peptide-1 (GLP-1) receptor agonists and sodium-glucose cotransporter-2 (SGLT2) inhibitors are recommended, there is a paucity of evidence to support the use of one second-line agent over another. Moreover, there is a lack of long-term comparative effectiveness studies of drug classes [23]. As a result, clinical practice is based largely on trial and error of additional therapeutic regimens, taking only a handful of characteristics (e.g., risk of hypoglycemia, weight gain, and cost) into consideration. In order to direct the choice of more individualized interventions, it is imperative to study the genetic characteristics that underlie the effectiveness and tolerability of various glucose-lowering agents.

Pharmacogenetics, the study of interactions between genetic loci and pharmacologic therapy, offers an opportunity to stratify those who are more likely to respond to a given medication and minimize the potential adverse effects of other therapies. Successful application of pharmacogenetics in monogenic diabetes has already been achieved; for instance, patients with neonatal diabetes caused by activating mutations in KCNJ11 can safely discontinue insulin injections in favor of a sulfonylurea, an oral agent [24]. Furthermore, those with maturity-onset diabetes of the young (MODY) due to inactivating GCK mutations can often be treated by diet alone and have a low prevalence of the vascular complications that are common in other forms of diabetes [25].

It is conceivable that in the future, the paradigm in monogenic diabetes can be extended to the polygenic disease of T2D. In this section, we present the current state of pharmacogenetics in T2D, featuring examples of studies in which evidence has been generated on how genetics can influence the response to five commonly prescribed glucose-lowering agents. The earliest studies examined candidate genes implicated in drug pharmacokinetics and transport; subsequent investigations employed genome-wide approaches for discovery of genetic loci related to drug response. We also discuss examples of how pharmacogenetics has also been employed to illuminate drug mechanism and gain insight into T2D disease heterogeneity.

3.1. Metformin

Despite being the most commonly used medication for initial treatment of T2D, a significant number of individuals develop progression of hyperglycemia on metformin, requiring escalation of therapy [26, 27]. The heritability of glycemic response to metformin has been estimated at 20-34% [28], suggesting a considerable genetic component of interindividual variability in response. Prior to the availability of GWAS, metformin pharmacogenetics was restricted to the study of cellular transporters that regulate metformin disposition. As a hydrophilic molecule, metformin has limited diffusion through cell membranes and requires transporters for its oral absorption, hepatic uptake, and renal excretion [29].

The most studied transporter is OCT1, which is highly expressed in the liver and encoded by the highly polymorphic gene SLC22A1 [30]. Though reduced function variants were associated with an impaired response to an OGTT following a short course of metformin in healthy volunteers [31], subsequent findings on their impact on HbA1c reduction in individuals with T2D have been inconsistent [32, 33], potentially owing to differences in study design or population characteristics. However, there appeared to be a signal for metformin intolerance, which manifests primarily as gastrointestinal side effects in 20-30% of patients. In a GoDARTS study of side effects, individuals with two or more reduced-function alleles in SLC22A1 were over twice as likely to develop intolerance to metformin [34].

More recent studies have focused on agnostic searches across the genome for variants associated with glycemic efficacy. The first GWAS identified rs11212617 in the ATM locus, which was associated with a greater response to metformin, defined as achieving a HbA1c <7% within 18 months [35]; this finding was later replicated in independent datasets [36]. Subsequently, a second GWAS meta-analysis of over 10,000 participants revealed a genome-wide significant association between rs8192675, an intronic variant in the GLUT2 glucose transporter gene SLC2A2, and greater HbA1c reduction [37]. The absolute HbA1c reduction conferred by the C allele was clinically relevant, equivalent to the dose impact of an additional 550 mg of metformin. Since the variant was also associated with lower SLC2A2 expression in the human liver, one plausible biological mechanism is that C-allele carriers may have decreased glucose clearance, which is improved and overcome by metformin therapy.

However, neither of these GWAS findings was replicated in the DPP [37, 38], which raised the question of whether pharmacogenetic interactions may differ depending on the stage of the disease process. At the time of writing, a GWAS for metformin response has been completed in the DPP, a diverse cohort of individuals at a higher risk of developing T2D in the US. In examining the one-year change in metformin-related quantitative traits, the study revealed novel ancestry-specific allelic associations that remain to be replicated, thus illustrating the importance of generating pharmacogenomic resources in diverse populations [39].

Finally, an incomplete understanding of metformin’s mechanism of action continues to hinder attempts to elucidate the genetics of response [40]. While future molecular studies will enable a greater comprehension of metformin’s benefits in relation to glucose metabolism, an alternative strategy has been to perform pharmacological experiments that perturb the glucose homeostatic system. In this way, the Study to Understand the Genetics of the Response to Metformin and Glipizide in Humans (SUGAR-MGH) utilizes two acute drug challenges to identify the functional relevance of pharmacogenetic loci [41]. By characterizing the physiological response to a drug and how it differs by genotype, one can gain insight into the potential mechanism of action of the variant in question. With regard to metformin, a gePS for fasting glucose is associated with reduced metformin response, indicating that people genetically predisposed to fasting hyperglycemia may not be the best candidates for initial metformin therapy [42]. We discuss relevant findings related to sulfonylurea use in this study in the next section.

3.2. Sulfonylureas

Sulfonylureas are commonly utilized second-line agents for the treatment of T2D and stimulate glucose-independent insulin secretion by binding to the ATP-sensitive K+ (K_ATP) channel on the cell membrane of pancreatic β cells [43]. Despite their widespread use, sulfonylureas can cause weight gain and carry the deleterious side effect of hypoglycemia. As with metformin pharmacogenetics, the earliest studies of sulfonylurea response examined relevant candidate genes. Carriers of two reduced-function alleles in CYP2C9, the gene encoding the liver enzyme responsible for the rate-limiting step of sulfonylurea metabolism, were 3.4 times more likely to attain a HbA1c <7%, equivalent to a greater HbA1c reduction of 0.5% [44]. With respect to side effects, the evidence has been mixed, with some studies showing a higher risk of hypoglycemia in two-copy carriers of CYP2C9 [45, 46] and others not observing an association [47].

Other lines of investigation have focused on genetic variation in the sulfonylurea receptor (SUR1, ABCC8) and the potassium inward rectifier channel (Kir6.2, KCNJ11), which together form the K_ATP channel. In two independent cohorts of patients with T2D treated with gliclazide for 8 weeks, the non-synonymous variant rs757110 in ABCC8 was associated with greater glycemic response [48]. Similarly, carriers of the K-allele at rs5219 in KCNJ11 experienced a greater effect of sulfonylurea treatment [49]. However, these findings were not confirmed when examining one-year failure to treatment with sulfonylureas in the UK Prospective Diabetes Study (UKPDS) [50].

Rs7903146 in TCF7L2 has also been a pharmacogenetic variant of interest, given its strong association with T2D. In GoDARTS, a cohort of individuals with established T2D, carriers of the T risk allele were less likely to respond to sulfonylureas, with an odds ratio for failure of 1.73 [51]. On the other hand, the acute response to glipizide was enhanced in risk allele carriers in SUGAR-MGH, in which participants are either healthy or at increased risk for T2D [17], suggesting that the variant has a differential effect depending on whether β cell deterioration has taken place. Furthermore, the T allele was associated with higher incretin levels across the study, potentially implicating altered incretin signaling as a mechanism for its influence on T2D risk and demonstrating the utility of this pharmacogenetic resource in illuminating functional mechanism.

Expanding beyond single variant associations, the first GWAS meta-analysis of glycemic response to sulfonylureas was recently published and reported a heritability estimate of approximately 37%, comparable to that of metformin [52]. Using an outcome of HbA1c reduction following 12 months of therapy, the study identified associations with two genome-wide significant variants, rs1234032 near GXYLT1 and rs10770791 located in a hepatic transporter gene SLCO1B1, which were supported by functional studies and replication efforts. Interestingly, a significant interaction was observed between rs10770791 genotype and concomitant statin use, whereby the genotype had no influence on sulfonylurea response in the presence of statins. In the absence of statin exposure, C allele homozygotes had a 0.48% reduction in HbA1c compared to T allele homozygotes, which has implications for the clinical utility of this variant when prescribing sulfonylureas.

3.3. GLP-1 receptor agonists

GLP-1 receptor agonists have risen to the forefront of the treatment algorithm for T2D due to their glycemic benefit, metabolic impact, low hypoglycemia risk, and cardiorenal protection [53]. Despite their efficacy, there is considerable heterogeneity in their glycemic and metabolic effects; some clinical trial participants fail to lose any weight on these agents whereas others can lose up to 25 kg over 6 months [54, 55]. Moreover, clinical markers of poor β-cell function, such as low C-peptide and longer duration of T2D appear to associate with worse glycemic response at 6 months [56], suggesting that certain subtypes or physiologically-defined clusters of T2D may not benefit from these agents.

Due to the interindividual variability in response to GLP-1 receptor agonists, understanding the genetic predictors of response will help target treatment to those most likely to benefit. Most pharmacogenetic studies to date have been limited to missense variants in the drug target gene GLP1R, which affect the function of the GLP-1 receptor and alter the insulin secretion response to GLP-1 [57]. For instance, in 90 individuals receiving liraglutide for 14 weeks, carriers of the A allele in rs6923761 had a 2.9 kg larger weight reduction, but no effect was seen for HbA1c [58]. The T variant in rs10305420, which interestingly is T2D-protective [11], was associated with smaller reductions in HbA1c (0.4% per allele) and weight (1.27 kg per allele) after 6 months of exenatide treatment in obese individuals with T2D [59].

Recently, the first GWAS for glycemic response to GLP-1 receptor agonists was completed across four observational studies and two randomized control trials, and included a candidate gene analysis of rs6923761 and rs10305420 [60]. The A allele of rs6923761, which was found by others to alter the insulin secretory response to infused GLP-1 [57], was associated with a lower HbA1c reduction. The authors did not observe a significant effect of rs10305420 on HbA1c reduction, and the GWAS on common variants did not discover any genome-wide significant associations. However, gene-based burden testing of low frequency variants revealed a novel association between risk alleles in Arrestin beta 1 (ARRB1) and greater HbA1c reduction; the effect was mainly driven by the A allele of rs140226575, which interestingly is more commonly found in Hispanic (minor allele frequency [MAF] 6%) and American Indian/Alaskan Native (MAF 11%) populations. Finally, a genetic risk score analysis identified that ~5% of the population with low frequency variants in ARRB1 responded 30% better to GLP-1 receptor agonists than 43% of the population with wild-type ARRB1 variants and at least one variant allele at GLP1R, therefore signifying a potential subgroup of individuals who may derive a greater benefit from earlier use of GLP-1 receptor agonists.

As one of the most studied variants associated with T2D, rs7903146 in TCF7L2 has also been investigated. In a study of 162 individuals with T2D, carriers of the T risk allele experienced a greater reduction in insulin levels (but had similar levels of plasma glucose) during a mixed-meal test following 8 weeks of treatment with exenatide, which was construed by the authors to represent a more efficient insulin response to a GLP-1 receptor agonist in allele carriers [61]. However, the evidence is limited, and the interplay between the mechanisms underlying T2D risk and drug response need to be further elucidated.

Despite the common gastrointestinal side effects of these agents, predominantly nausea (25-60%) and vomiting (5-15%) [53], which can hamper their use and result in their discontinuation, none have studied the genetic predictors of adverse effects. While the genetic effects on non-glycemic outcomes such as weight have been explored, additional studies are needed on their pleiotropic effects and benefits on the cardiovascular system. In one genomic approach, the cardioprotective effects of GLP-1 receptor agonists were validated by analysis of the GLP1R variant rs10305492, which is associated with both glucose-lowering and diminished coronary heart disease risk [62]. Thus, genetics can be used as a tool to provide insight into the safety of therapeutic glucose-lowering agents. However, there remains a major knowledge gap in understanding the degree to which such variation contributes to clinical outcomes, including diabetic complications and mortality, which will necessitate access to long-term outcomes data, accompanied by available genetic data.

3.4. Dipeptidyl peptidase-4 (DPP-4) inhibitors

DPP-4 inhibitors are a class of oral agents that act by inhibiting the enzyme DPP-4 and increasing the levels of GLP-1 and glucose-dependent insulinotropic polypeptide (GIP) by preventing their degradation [63]. They have favorable safety profiles and tolerability, contributing to their use as an oral second-line medication. Because DPP-4 inhibitors, with the exception of linagliptin, are not primarily metabolized in the liver [64], pharmacogenetic studies have focused on variants in incretin receptors or associated with T2D risk rather than those in drug transporters. Given modest glycemic efficacy and absence of weight loss benefit of DPP-4 inhibitors compared to their GLP-1 receptor agonist counterparts, there are relatively few pharmacogenetic breakthroughs for this drug class.

Genetic variation in DPP4, encoding the target molecule for DPP-4 inhibitors, has been hypothesized to be a predictor of drug response, such that mutations that result in reduced degradation of incretins would confer improvements in glycemic control. However, the evidence is limited, with only one study to date showing that carriers of the T allele of rs2909451 and C allele of rs759717 have increased activity of DPP-4 enzymatic activity during sitagliptin therapy, with no further evaluation of glycemic outcomes [65]. As a downstream indirect target of DPP-4 inhibitors, the GLP-1 receptor has also been studied. Indeed, the missense rs6923761 variant in GLP1R was associated with a smaller HbA1c reduction following 6 months of treatment with either sitagliptin or vildagliptin in 140 individuals with T2D, possibly due to reduced GLP-1-stimulated insulin secretion [66]. This finding was subsequently confirmed in a later study, where in a recessive model, AA homozygotes had a 0.26% smaller reduction in HbA1c compared to G allele carriers, similar to the effect size of pharmacogenetic studies for metformin and sulfonylureas [67]. With respect to the TCF7L2 variant, individuals homozygous for the T risk allele had a 0.26% smaller HbA1c reduction in response to 24 weeks of linagliptin compared to wild-type homozygous, though findings should be interpreted with caution given the small sample size [68].

In a study that leveraged complementary genetic, pharmacogenetic, and physiologic data, the G allele of rs7202877 near CTRB1/2, a protective factor for the development of T2D [69], was associated with increased GLP-1 stimulated insulin secretion [70]. Carriers of the G allele were subsequently observed in a pharmacogenetic cohort to have decreased responsiveness to DPP-4 inhibitor treatment, though there was notably no effect of genotype on response to GLP-1 receptor agonist treatment. In whole pancreas and islets, the G allele was also found to raise mRNA expression of CTRB1/2, which encode the digestive enzyme chymotrypsinogen; consistent with the expression data, chymotrypsin activity from stool samples was augmented in G-allele carriers. The study results suggest an intriguing model by which the increase in chymotrypsin activity improves the sensitivity of pancreatic β-cells to GLP-1, resulting in increased insulin secretion and decreased risk of T2D. Moreover, increased chymotrypsin activity may cause G-allele carriers to be relatively insensitive to DPP-4 inhibition, which could explain the reduced effect of DPP-4 inhibitors in these individuals.

3.5. SGLT2 inhibitors

The most recent drug class to be added to the therapeutic armamentarium for T2D, SGLT2 inhibitors act by inhibiting glucose reabsorption in the proximal tubules of the kidney [53]. Because of their glucoretic effects, SGLT2 inhibitors have beneficial effects on body weight and blood pressure. Moreover, multiple clinical trials have demonstrated the protective advantages of these agents in those with established cardiovascular or kidney disease [71]. While the risk of hypoglycemia is low since their mechanism of action is independent of insulin, other side effects associated with this medication include genitourinary infections, dehydration, and euglycemic diabetic ketoacidosis.

Because it encodes the major target of SGLT2 inhibitors, SLC5A2 has been investigated as a relevant pharmacogene for drug response. While variants in SLC5A2 have been previously identified with development of familial renal glucosuria, no association was observed between five common variants and response to 24 weeks of empagliflozin in individuals with T2D, as measured by HbA1c, fasting glucose, or body weight [72]. Interestingly, in a later meta-analysis, the minor allele of rs9934336 in SLC5A2 was significantly associated with a reduced risk of T2D, suggesting that long-term prevention of hyperglycemia conferred by this variant can protect against development of T2D [73]. However, the meta-analysis was not poised to assess for glycemic response to SGLT2 inhibitors, and the functional relevance of this particular variant remains to be understood.

Other pharmacogenetic studies have evaluated variation in genes coding for uridine diphosphate glucuronosyltransferases (UGTs), which are involved in the disposition of SGLT2 inhibitors. In a small pharmacokinetic study of 134 individuals, carriers of reduced-function variant UGT1A9*3 had increased plasma concentration of canagliflozin [74]. Further limiting the clinical relevance of this finding is the lack of a difference in the incidence of drug adverse effects between genotype groups and the low frequency of the UGT1A9*3 allele in the population. Despite the paucity of current evidence on the pharmacogenetics of SGLT2 inhibitors, they represent a promising therapeutic option and the proper trials with the appropriate cohort size and defined phenotypic outcomes must be designed to guide stratified use of these agents.

4. Pharmaco-omics

Precision medicine in diabetes has largely focused on genomic variation, but the field has expanded to investigate other forms of human variation, considering differences in epigenetic marks, gene expression, as well as changes in circulating metabolites and proteins within a system. Though GWAS have generated insight into pharmacogenetic associations, the underlying molecular mechanisms often remain to be determined; the integration of GWAS findings with other -omics data can help shed functional insights. Present challenges of these additional techniques include difficulties in accessing the relevant tissues (e.g. beta cells) and the potential for reverse causation, in that non-genetic changes on epigenetic marks or transcript, metabolite, or protein levels may be caused by the disease or its treatment rather than contribute to its etiology. While it would be too lengthy to provide an exhaustive list of examples in this review, we will briefly illustrate the use of these technologies with respect to metformin pharmacotherapy.

Blood-based epigenetic markers can aid in clinical decision-making for newly diagnosed individuals with T2D and guide selection of the optimal first-line therapy. DNA methylation has been demonstrated to contribute to T2D disease susceptibility [75, 76] and conceivably could associate with drug response. To test this hypothesis, one study examined genome-wide DNA methylation in drug-naive patients with T2D and performed association with glycemic response and intolerance to metformin [77]. They found that those with higher degrees of methylation were 2.5 times more likely to not respond to metformin and 3 times more likely to tolerate the drug due to side effects, illustrating its use as a potential biomarker.

Additionally, metabolomics, the large-scale study of small molecules, when studied in conjunction with genetics, can facilitate the discovery of novel biological pathways and reveal unique signatures associated with disease risk that may benefit differentially by treatment. [78]. By assessing the metabolome before and after a drug, one can characterize signatures that correspond to a particular metabolic state or desirable drug response (i.e., glycemia). In the DPP, lower baseline levels of the metabolite betaine was associated with reduced diabetes incidence, and betaine levels were significantly increased by the intensive lifestyle intervention and numerically increased by metformin [79]. This observation suggests a potential clinical utility of betaine measurements in monitoring the efficacy of preventive therapies for T2D.

Finally, the gut microbiome has also been recognized as playing an important role in drug metabolism [80]. In a large metagenome-wide study that controlled for metformin treatment, a reduction in butyrate-producing bacteria was observed in T2D [81]. Moreover, in a double-blind, placebo-controlled study of individuals with newly diagnosed T2D randomized to metformin or placebo for 4 months, metformin was found to alter the composition and function of the gut microbiota, which may mediate some of the glucose-lowering effects of metformin [82]. Thus, studying how the host microbiome changes with respect to disease and its interaction with drug treatment will continue to be an area of research attention.

5. Challenges and future directions

5.1. Need for diverse populations

There is a lack of well-powered genetic studies in diverse populations; in 2019, nearly 80% of GWAS participants were of European descent despite comprising only 16% of the global population [83]. Because the frequency and effect of risk alleles differ across populations and linkage disequilibrium (the correlation structure of the genome) varies across ancestries, polygenic risk scores for T2D derived from European populations transfer poorly to other ancestral groups. Given that the prevalence of diabetes and diabetic complications is higher in people of African or Native American descent, the omission of these underrepresented populations in genetic studies and in the development of disease prediction models may result in greater health disparities.

Therefore, efforts have emerged to conduct trans-ancestry meta-analyses of T2D GWAS, enabling improved risk prediction [10]. Leveraging the population diversity in the DIAMANTE study, the authors compared the performance of trans-ancestry and ancestry-specific T2D polygenic risk scores, and found that the trans-ancestry score, which was weighted with population-specific allelic effect estimates, had the greatest predictive power in all populations. The NIH-funded PRIMED Consortium has a similar goal of utilizing multi-ancestry summary statistics to improve the applicability of polygenic scores in diverse populations [84]. Additionally, large-scale GWAS in understudied populations are still needed to reveal novel associations for variants that are common in one ancestral group but rare or absent in others, such as the association of SLC16A11 with T2D in Latino populations, which confers a ~20% increased risk of T2D per haplotype copy [85]. As illustrated by this discovery, increased inclusion can facilitate enhanced disease prediction and uncover new aspects of disease pathophysiology. To improve upon representation of ancestry groups in genetic studies, genotyping technologies must also have broad variant coverage across multiple ancestries, available imputation panels must be cosmopolitan, and statistical methods must be developed to integrate such genetic information. Finally, researchers and funding institutions need to prioritize the creation and enrollment of diverse cohorts for future genomic investigation.

Research in pharmacogenetics must follow this example to enhance the generalizability of findings for clinical translation. Similar to GWAS for complex disease, discovery GWAS for drug response suffer from lack of representation, with Europeans accounting for 90% of participants, compared to 3% each for Africans, Asians, and Hispanics [86]. Failing to consider ancestral diversity can result in associations that do not replicate in other populations. Additionally, the influence of an actionable variant can differ in magnitude due to its minor allele frequency in a given population, and the failure to include ancestry-specific variants is problematic. One example seen in cardiovascular pharmacogenetics is warfarin dosing, which is characterized by high interindividual variability with a genetic basis, but dosing algorithms based on pharmacogenetic findings in Europeans has performed poorly in African populations [87]. Thus, studies in minorities are necessary to find and validate novel ancestry-specific variants to mitigate the gaps in performance of dosing algorithms and facilitate the development of ancestry-specific recommendations, with the goal of improving patient care across all populations.

5.2. Overcoming small sample size

Another major shortcoming of current pharmacogenetic GWAS is the limited sample size of available cohorts. It is a challenge to assemble large cohorts containing phenotypes of drug response, as each individual needs to have available information on multiple component measures, including medication dose, degree of drug adherence, presence of both baseline and follow-up measurements of the outcome variable (e.g., HbA1c, fasting glucose), and concomitant medications that could influence glycemia [88]. Furthermore, future studies need to expand beyond glycemia and evaluate susceptibility to side effects as well as long-term outcomes such as development of diabetic complications. Clinical trials may represent an ideal cohort for pharmacogenetic analyses as they contain standardized drug exposures and pre-defined phenotypic outcomes, but they can be difficult to access if conducted by the pharmaceutical industry, often lack DNA collection for genotyping, and may not have extended follow-up to be able to detect long-term outcomes.

Consortium science has emerged to combat this problem and expand the sample size of primary and replication cohorts. The Metformin Genetics (MetGen) consortium is one such international collaborative effort whose goal is to gather datasets for metformin and other glucose-lowering medications that can be combined for meta-GWAS, which have enhanced power to detect gene-drug associations [37]. Another approach is to employ a retrospective observational design in electronic health record (EHR)-linked biobanks, which offer rich longitudinal clinical data paired with genomic information. Whereas biobanks are widely accessible, contain a wealth of clinical outcomes, and provide a major increase in sample size, the data can be rather heterogeneous and require careful implementation of phenotyping algorithms to capture the desired outcome of interest.

5.3. Leveraging polygenic scores

As discussed earlier, polygenic scores provide an opportunity to boost power by combining millions of variants across the genome and can explain a greater proportion of the variance observed for a given outcome. While they have primarily been employed to estimate disease risk, polygenic scores that inform on T2D pathophysiology can help predict response to therapy. One approach for understanding gene-drug interactions is to evaluate the influence of polygenic scores constructed from known risk loci for T2D on drug response. In an analysis in SUGAR-MGH, individuals with a higher burden of T2D risk variants had a more robust response to acute glipizide exposure [39]. This finding is remarkable because it appears that treatment with a sulfonylurea that stimulates insulin secretion can overcome genetic defects in β-cell function early on in disease pathogenesis, recapitulating what is seen in monogenic neonatal diabetes.

These results were consistent with the separate observation that those in the severely insulin-deficient diabetes cluster [5], which is similarly enriched for variants influencing β-cell function, had a strong initial response to sulfonylureas in the ADOPT study but then rapidly failed the agent over time, presumably due to deteriorating β-cell function [89]. Moreover, the clusters had differing risks of progression to diabetic complications; for instance, clinicians may choose to prioritize insulin sensitizing or renoprotective therapies in the severe insulin-resistant diabetes cluster with the highest risk for diabetic kidney disease. As polygenic scores for T2D-related traits and physiological clusters become more sophisticated, they can begin to drive treatment decisions that address the underlying pathophysiology for a given individual.

Additionally, as larger pharmacogenetic GWAS are conducted, polygenic scores can be created for drug response outcomes themselves. Per a systematic review, over half of the 51 identified applications of polygenic scores in pharmacogenomics were in psychiatric disorders and only four were in endocrine, nutritional, and metabolic disease [90], illustrating a promising opportunity for advancement in this area. However, several limitations were highlighted among these published analyses, including Eurocentric biases of cohorts, substantial variability in methodology, inconsistency in reporting of results, and lack of validation in independent cohorts, all of which must be addressed before these scores can be useful for clinical practice.

5.4. Genetics-guided approach to drug discovery

Importantly for therapeutic development, agnostic genetic studies can pave the way for identification of new drug targets [91]. This concept is illustrated by gene-drug pairs in which the T2D medication existed before the GWAS discovery was made, such as PPARG and thiazolidinediones [92], KCNJ11 and sulfonylureas [93], and GLP1R and GLP-1 receptor agonists [94]. Therefore, continued advancements in GWAS have the potential to uncover intriguing genes that may serve as the basis for future drug development.

Furthermore, genetics can be leveraged to enhance drug discovery efforts (Table 1). In the case of SGLT2 inhibitors, the process began with the observation of families who developed familial renal glucosuria (FRG), a relatively benign condition [95]. Subsequently, inactivating mutations in SLC5A2 were found to be the cause of this phenotype [96], and compounds that inhibit the target protein SGLT2 were discovered and synthesized. The benign clinical phenotype suggested that permanent inhibition of glucose reabsorption in the nephron did not seem to have long-term deleterious consequences. The result was the development of a pharmacologic class that mimics the effect of dysfunctional SGLT2 in individuals with FRG and harnesses the mechanism of renal glucose excretion to reduce hyperglycemia in patients with T2D. Thus, taking SGLT2 inhibitors as an example, a genetics-guided approach can be implemented in identifying untapped mechanisms to treat a disease, as well as evaluating potential off-target effects.

Table 1.

Model for a genetics-guided approach to drug discovery

6. Conclusions

With the diabetes epidemic accelerating at an unprecedented pace, we need to embrace precision diabetology, which involves not only a deeper understanding of the complex biological mechanisms contributing to T2D development and progression but also the generation of evidence to support a stratified approach to drug prescription based on an individual’s genetic information. Whereas large-scale genomic discovery efforts have enriched our knowledge of the genetic architecture of T2D, much remains to be answered about the functional relevance of T2D-associated loci, which have the potential to expedite the development of novel drugs.

Moreover, pharmacogenetics remains in its early stages, especially with respect to the influence of genetic variation on newer glucose-lowering medications. The majority of studies have examined candidate genes that are selected based on their relevance to drug pharmacokinetics, association with the drug target gene, or with the etiological process underlying the development of T2D. As illustrated through GWAS of metformin and, more recently, sulfonylureas and GLP-1 receptor agonists, unbiased genome-wide searches have greater potential to uncover novel variation and shed insight on new molecular pathways.

To obtain reproducible results and avoid false positive findings, appropriately sized replication cohorts in diverse populations are needed to confirm associations and to generate recommendations for all ancestral groups. Future studies should examine adverse drug reactions with the goal of minimizing unnecessary exposure to individuals who are more susceptible. As newer therapeutic agents with demonstrated benefits beyond glycemia permeate clinical care, long-term outcomes should also be examined. Overcoming these research gaps will enable the transformation of the treatment paradigm for T2D into one that incorporates genetics-based prediction alongside clinical characteristics.

KEY POINTS.

• T2D is a heterogeneous disease for which the treatment guidelines are algorithmic; current recommendations do not consider the pathophysiological defects of the underlying disease or incorporate individualized genetic information to optimize treatment response.

• Pharmacogenetic studies of metformin and sulfonylurea response have generated evidence for a precision medicine approach to drug selection that can be applied to newer glucose-lowering agents.

• Next steps for the field of pharmacogenetics include recruiting more diverse populations into genetic studies, establishing large cohorts with well-defined phenotypic outcomes apart from glycemia, and maximizing the utility of polygenic scores.

• As illustrated through the emergence of SGLT2 inhibitors, genetics holds promise for the identification of new drug targets that can inspire development of novel therapeutic agents.