Genome-Wide Meta-analysis Identifies Genetic Variants Associated With Glycemic Response to Sulfonylureas

Abstract

OBJECTIVE Sulfonylureas, the first available drugs for the management of type 2 diabetes, remain widely prescribed today. However, there exists significant variability in glycemic response to treatment. We aimed to establish heritability of sulfonylurea response and identify genetic variants and interacting treatments associated with HbA_1c reduction.

RESEARCH DESIGN AND METHODS As an initiative of the Metformin Genetics Plus Consortium (MetGen Plus) and the DIabetes REsearCh on patient straTification (DIRECT) consortium, 5,485 White Europeans with type 2 diabetes treated with sulfonylureas were recruited from six referral centers in Europe and North America. We first estimated heritability using the generalized restricted maximum likelihood approach and then undertook genome-wide association studies of glycemic response to sulfonylureas measured as HbA_1c reduction after 12 months of therapy followed by meta-analysis. These results were supported by acute glipizide challenge in humans who were naïve to type 2 diabetes medications, cis expression quantitative trait loci (eQTL), and functional validation in cellular models. Finally, we examined for possible drug-drug-gene interactions.

RESULTS After establishing that sulfonylurea response is heritable (mean ± SEM 37 ± 11%), we identified two independent loci near the GXYLT1 and SLCO1B1 genes associated with HbA_1c reduction at a genome-wide scale (P < 5 × 10⁻⁸). The C allele at rs1234032, near GXYLT1, was associated with 0.14% (1.5 mmol/mol), P = 2.39 × 10⁻⁸), lower reduction in HbA_1c. Similarly, the C allele was associated with higher glucose trough levels (β = 1.61, P = 0.005) in healthy volunteers in the SUGAR-MGH given glipizide (N = 857). In 3,029 human whole blood samples, the C allele is a cis eQTL for increased expression of GXYLT1 (β = 0.21, P = 2.04 × 10⁻⁵⁸). The C allele of rs10770791, in an intronic region of SLCO1B1, was associated with 0.11% (1.2 mmol/mol) greater reduction in HbA_1c (P = 4.80 × 10⁻⁸). In 1,183 human liver samples, the C allele at rs10770791 is a cis eQTL for reduced SLCO1B1 expression (P = 1.61 × 10⁻⁷), which, together with functional studies in cells expressing SLCO1B1, supports a key role for hepatic SLCO1B1 (encoding OATP1B1) in regulation of sulfonylurea transport. Further, a significant interaction between statin use and SLCO1B1 genotype was observed (P = 0.001). In statin nonusers, C allele homozygotes at rs10770791 had a large absolute reduction in HbA_1c (0.48 ± 0.12% [5.2 ± 1.26 mmol/mol]), equivalent to that associated with initiation of a dipeptidyl peptidase 4 inhibitor.

CONCLUSIONS We have identified clinically important genetic effects at genome-wide levels of significance, and important drug-drug-gene interactions, which include commonly prescribed statins. With increasing availability of genetic data embedded in clinical records these findings will be important in prescribing glucose-lowering drugs.

Introduction

Sulfonylureas are potent glucose-lowering drugs that reduce HbA_1c by an average of 1.5% (18 mmol/mol) (1). Despite an increasing trend to use more modern, expensive treatments, sulfonylureas remain commonly prescribed in the U.K., making up 27% of new prescriptions, second only to metformin (2). Due to their very low cost, they are extensively used in low- and middle-income countries. However, considerable variation exists in response to sulfonylureas, with 10–20% of people with diabetes not responding at initiation of sulfonylurea therapy and 30–35% failing to respond to monotherapy after 5 years (3,4). It is likely that a combination of genetic and nongenetic modifying factors underlies the clinical variability of glycemic response to sulfonylureas. While many clinical risk factors such as baseline HbA_1c, sex, duration of diabetes, and dose are associated with glycemic response to sulfonylureas (5–7), modulatory genetic factors remain largely unexplored, with the exception of a few proof of concept studies with use of a candidate gene approach (8–12).

Glycemic response to metformin is heritable, with 34% of the variance in response explainable by common genetic variants (13–15). There have been no similar estimates for sulfonylurea response, and to date, no genome-wide association studies (GWAS) of glycemic response to sulfonylurea treatment have been reported, so the genetic contribution to how patients respond to sulfonylureas and clinical implication of this genetic variation have not been systematically studied. As an initiative of the Metformin Genetics Plus Consortium (MetGen Plus) and the DIabetes REsearCh on patient straTification (DIRECT) consortium, we report here the first genome-wide meta-analysis of glycemic response to sulfonylureas, measured as HbA_1c reduction after 12 months of therapy. Based on these findings we then explore the impact of interacting drugs and identify clinically important genotype-dependent statin-sulfonylurea interactions for this important class of diabetes therapies.

Research Design and Methods

List of abbreviations used throughout this article and their corresponding explanations are shown in Supplementary Table 1.

Study Design and Participants

We established an international consortium allowing recruitment of 5,485 unrelated individuals of European ancestry from six referral centers in Europe and North America as part of MetGen Plus and the DIRECT consortium (Supplementary Table 2). Included participants had a clinical diagnosis of type 2 diabetes and were treated with sulfonylureas as monotherapy or as an add-on to metformin. This study was approved by respective research ethics review boards, and participants provided written informed consent.

Sample Ascertainment

Clinical, prescription, and biochemical data were retrieved from the electronic medical record systems. Participants with type 2 diabetes aged >35 years at diagnosis who used sulfonylureas with no history of insulin use were identified. They were stably treated with sulfonylureas for at least 6 months with no other glucose-lowering drug started or stopped within the study period. The baseline HbA_1c was between 7% (53.0 mmol/mol) and 14% (129.5 mmol/mol) at sulfonylurea initiation.

Measurement of Glycemic Response and Definition of Variables

Participants’ glycemic response to sulfonylurea was modeled as the quantitative phenotype of HbA_1c reduction between baseline HbA_1c and treatment HbA_1c while the patients were maintained on stable treatment. Baseline HbA_1c was defined as the HbA_1c measure closest to sulfonylurea initiation and within 6 months before and 7 days after this date. The treatment HbA_1c was the HbA_1c measure closest to 12 months after initiation of sulfonylureas (between 6 and 15 months).

In all the studies, covariates were selected based on previous reports and univariate association between the outcome variable (HbA_1c reduction) and explanatory variables. The best fit linear regression model was determined using stepwise backward elimination. Accordingly, baseline HbA_1c, sex, age at diagnosis, baseline BMI, average daily dose, time between baseline HbA_1c and treatment HbA_1c, and drug group (sulfonylurea monotherapy or sulfonylurea added to metformin) were considered in the final model as available in each cohort (Supplementary Table 3). Average daily dose was calculated as the mean daily dose of prescriptions filled during the study period (mean of percentage of each sulfonylurea divided by maximum prescribable according to the British National Formulary). Baseline weight was the measure nearest to the sulfonylurea start date (index date) and within 180 days on either side of the index date. Each study was adjusted for the top n principal components (PCs) to account for 80–90% of the variation in population structure.

The final response model was as follows: HbA_1c reduction ∼ baseline HbA_1c + PCs + study-specific covariates.

Genome-Wide Array Genotyping, Quality Control, and Imputation

For each respective cohort genome-wide genotyping was performed on a variety of arrays as illustrated in Supplementary Table 3. Genotyping and quality control procedures for the Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS), Hoorn Diabetes Care System (DCS), and Pharmacogenomics of Metformin (PMET) cohorts have previously been described (13,15,16). Genotyping data for each platform were individually cleaned by each study center. Standard postgenotyping quality control procedures were applied to each data set (Supplementary Fig. 1). Monomorphic single nucleotide polymorphisms (SNPs) with minor allele frequency (MAF) <1%, call rate <98%, or Hardy-Weinberg equilibrium <10⁻⁶ were removed. Samples with genotyping calls <98% or heterozygosity >3 SDs from the mean or correlated with another sample (identity by descent >0.125) were filtered out. All genetic variants were mapped to and reported with Genome Reference Consortium Human genome build 37 (GRCh37). Each data set was then imputed to the 1000 Genomes CEU reference panel (phase 1, version 3) with IMPUTE software (17), except PMET2 and Geisinger where imputation was performed with the HRC.r1-1 EUR reference genome (GRCh37 build) using the Michigan server. Postimputation, SNPs with poor imputation quality (INFO < 0.6), monomorphic variants, or MAF <5% were excluded (Supplementary Fig. 1).

Genome-Wide Association Analysis

Following imputation, GWAS was conducted for each respective cohort under an additive genetic model for assessment of the role of common variants (MAF ≥5%) in glycemic response to sulfonylureas. Each SNP was tested for association with quantitative measure of sulfonylurea-related HbA_1c reduction with SNPTEST v2.536 (18) using multiple linear regression correcting for baseline HbA_1c, genotypic PCs, and other study-specific variables (Supplementary Table 3). Genome-wide association analyses were carried out separately by respective study centers. Prior to meta-analysis, we performed post-GWAS harmonization and quality control of GWAS results from each cohort to track possible errors in the study-specific analyses. We used the standard protocol accompanied by the EasyQC R package (19). Specifically, we removed SNPs with MAF <5%, low imputation quality (<0.6), large absolute values of β-coefficients and SEs (≥10), low call rate (<0.98), and deviations from Hardy-Weinberg equilibrium (P < 10⁻⁶). Meta-analysis was then performed with use of an inverse variance–weighted fixed-effects model, implemented in GWAMA v2.1.34 (20). Post–meta-analysis, SNPs with MAF <5%, available in fewer than six studies, with large absolute values of β-coefficients and SEs (≥10) were excluded (Supplementary Fig. 1). Heterogeneity was assessed with the I² metric from the complete study-level meta-analysis. Between-study heterogeneity was tested with the Cochran Q statistic and considered significant at P < 0.1. We used the commonly accepted threshold of 5.0 × 10⁻⁸ for joint P values to determine statistical significance. Nominal significance was considered to be P < 0.05. The CMplot package (21) in R was used to generate Manhattan and quantile-quantile plots. Regional plots around genome-wide or suggestive genes were visualized using LocusZoom (22). The final meta-analysis included 5,385,635 common autosomal SNPs from 5,485 independent individuals of European ancestors treated with sulfonylureas (λ = 1.008) (Supplementary Fig. 2).

Common Variant Heritability

We used the generalized restricted maximum likelihood approach under the LDAK assumptions using SumHer v5.1 (23) to estimate how much of the variance in HbA_1c reduction after sulfonylurea treatment could be attributed to common genetic variants (SNP-based heritability [h² SNP]). This method is a valid approach for estimating heritability in studies in which acquisition of data of family members with the same diagnosis who have received the same medication and were assessed with use of the same treatment outcome is not feasible. In addition, SumHer uses GWAS summary without requiring individual-level data (23). Therefore, we estimated the SNP heritability using summary statistics from the meta-GWAS. To avoid the impact of extreme linkage disequilibrium (LD) regions and disproportionately large effect size SNPs on heritability estimates, we exclude SNPs within the MHC (chromosome 6: 25–34 Mb) and SNPs that individually explain >1% of phenotypic variation and SNPs in LD with these (within 1 cM).

Conditional Analysis

Given rs10770791 is in partial LD with previously established nonsynonymous variants, rs4149056 (*5; V174A, D′ = 1; r² = 0.17) and rs2306283 (*1B; N130D, D′ = 0.98; r² = 0.63), we performed conditional analysis by including these SNPs in the model together. This analysis was carried out with individual-level data from the GoDARTS and PMET cohorts (65% of the total population) and baseline HbA_1c, PCs, and other study-specific covariates.

Biochemical Response to Glipizide

To test weather meta-GWAS–identified genetic variants are associated with trough glucose levels, we performed a lookup using data from the Study to Understand the Genetics of the Acute Response to Metformin and Glipizide in Humans (SUGAR-MGH). SUGAR-MGH enrolled 1,000 participants at risk for antidiabetes therapy in the future or individuals with lifestyle-controlled type 2 diabetes naïve to treatment. Participants received a single dose of 5 mg glipizide followed by measurement of glucose and insulin levels at 30, 60, 90, 120, 180, and 240 min. This was used to construct phenotypes of acute glipizide response. The association of rs1234032 and rs10770791 with glipizide response was assessed with linear regression with baseline glucose, age, sex, and the first 10 PCs as a covariate (see Supplementary Notes).

Drug-Drug-Gene Interaction Analysis

Given we have identified a genetic variant in the SLCO1B1 (a gene encoding hepatic transporter of statins) associated with glycemic response to sulfonylureas, we checked for interaction between SLCO1B1 rs10770791 and statin use in a drug-drug-gene interaction model using linear regression, with HbA_1c reduction as the dependent variable. This analysis was performed with use of individual-level data from the GoDARTS and PMET cohorts where we have access to prescription data.

Statin-treated case subjects were recipients of sulfonylureas who were also prescribed statins for at least the 3 months prior to the measurement of treatment HbA_1c. Statin untreated control subjects were those recipients of sulfonylureas who did not receive a statin prescription for at least 1 year prior to measurement of the treatment HbA_1c.

Expression Quantitative Trait Locus Lookups

Expression quantitative trait loci (eQTL) analysis seeks to identify genetic variants that affect the expression of one or more genes: a gene-SNP pair for which the expression of the gene is associated with the allelic configuration of the SNP is referred to as an eQTL. eQTL lookups were performed in human liver and whole blood samples for rs10770791 and rs1234032, respectively. Additional lookups were performed using publicly available data from the Genotype-Tissue Expression (GTEx) consortium.

The human liver eQTL lookups were carried out using data from a previous study performed by the group of F.I. (24). In brief, this eQTL study was performed with 1,183 liver samples, combined from four data sets (24). We looked up the top associated SNP, rs10770791, from this study, as it is in the SLCO1B1 and SLCO1B3 region, which are genes that are abundantly expressed in the liver.

The human whole blood eQTL lookup was performed with use of data from the DIRECT consortium in a total of 3, 029 subjects at high risk of developing type 2 diabetes or with recently diagnosed type 2 diabetes (25). A detailed explanation of the eQTL analysis has previously been published (26), and summary statistics are available (DOI: 10.5281/zenodo.4475681).

Cell Culture and In Vitro Transport and Inhibition Studies

Human embryonic kidney (HEK)-293 Flp-In cells stably expressing empty vector (EV), OATP1B1, and OATP1B3, were used for performance of in vitro transport and inhibition studies to establish the potency of inhibitors as IC₅₀ (i.e., concentration of inhibitor required to inhibit 50% uptake of a particular OATP1B1 and OATP1B3 substrate). Stably transfected HEK-293 Flp-In cells were maintained in DMEM H-21 supplemented with 10% FBS, 100 units/mL penicillin, 100 units/mL streptomycin, and 500 µg/mL geneticin. For transport studies, 150,000 cells/well were seeded the day before the experiment on a poly-d-lysine–coated 48-well plate. After 16–24 h, media were removed and cells were incubated at 37°C for 5–10 min in 0.5 mL Hanks’ balanced salt solution (HBSS) (Thermo Fisher Scientific). Uptake studies were initiated after removal of 0.3 mL of the HBSS above and addition of 0.15 mL HBSS containing a trace amount of ³H-glyburide (NET1024250UC; PerkinElmer), ³H-glipizide (MT1855; Moravek), or ³H-esterone sulfate (as positive control, NET203250UC; PerkinElmer). After 5 min, radioactive substrates were removed and washed twice with 1 mL ice-cold HBSS. For inhibition studies, the same methods above were used, where ³H-glyburide was used as substrate and various concentrations of atorvastatin (Cayman Chemical) or simvastatin (Cayman Chemical) were added together with ³H-glyburide. For comparison of the uptake of ³H-glyburide and ³H-glipizide in OATP1B1 reference and OATP1B1-174A (*5)–expressing cells, studies were performed using stable and transiently transfected cells. The stable and transient experiments were carried out with HEK-293 Flp-In cell lines expressing EV, OATP1B1 reference, and OATP1B1-174A (*5), previously established by our group (27). These cell lines were used to determine the uptake of ³H-glyburide, ³H-glipizide, and 3H-esterone sulfate (as positive control). In brief, each well was transfected with 200 ng DNA vector with 0.4 µL Lipofectamine LTX transfection reagent (Thermo Fisher Scientific) in a 48-well poly-d-lysine–coated plate. Uptake studies were then performed after 48 h with the methods described above and in triplicate wells.

Data and Resource Availability

Summary-level data that underlie the results reported in this article are available upon request to the corresponding author.

Results

Glycemic Response to Sulfonylureas Is Heritable

The SNP heritability estimate (h²) for a model-adjusted absolute reduction in HbA_1c was mean ± SEM 37 ± 11%, comparable with our previous estimate for metformin (h² = 34%) (14). This suggests that approximately one-third of the total variance of glycemic response to sulfonylureas is due to the additive effects of common variants.

GWAS Identifies Two Variants Associated With Altered Glycemic Response to Sulfonylureas

Meta-GWAS identified two genome-wide significant variants, rs1234032 and rs10770791, both on chromosome 12 (Fig. 1, Supplementary Fig. 2, and Table 1). The most significant association was obtained for rs1234032, with a mean ± SEM −0.14 ± 0.03% (−1.5 ± 0.3 mmol/mol) difference in HbA_1c reduction per C allele; P = 2.39 × 10⁻⁸. No statistical evidence for difference in effect size between studies was observed (P for heterogeneity [P_het] = 0.55) (Fig. 3). We then examined data from a healthy volunteer population (SUGAR-MGH, N = 857) given a single dose of glipizide (28) and found that the C allele of rs1234032 was associated with higher postdose glucose trough levels (β = 1.61, P = 0.005), and thus worse response, consistent with our GWAS findings. rs1234032 is an intergenic SNP, near GXYLT1 (Fig. 2 and Fig. 3), a gene that encodes a xylose transferase. rs1234032 is a cis eQTL to GXYLT1 in the whole blood with use of 3,029 samples from the DIRECT consortium, with the C allele being associated with increased expression (β = 0.21, P = 2.04 × 10⁻⁵⁸). rs1234032 also showed a significant association with GXYLT1 expression in multiple tissues including adipose subcutaneous (P = 8.1 × 10⁻⁵), artery tibial (P = 2.8 × 10⁻⁹), artery aorta (P = 3.4 × 10⁻⁶), nerve tibial (P = 3.6 × 10⁻⁶) and whole blood (P = 0.01) from the GTEx consortium (29), with the C allele associated with increased expression. These significant eQTL analyses could be due to strong linkage of rs1234032 (D′ = 1 and R² = 0.95) to rs7958582, which is within the cis-regulatory elements (https://screen.wenglab.org/). The C allele of rs1234032 is also in LD with the A allele of rs7964383 (D′ = 0.98, r² = 0.41), which is highly associated with increased whole blood gene expression (P = 1.7 × 10⁻⁴) (29) and circulating protein levels of GXYLT1 (30). Both rs7958582 (β per G allele = −0.10, P = 1.84 × 10⁻⁰⁶) and rs7964383 (β per A allele = −0.06, P = 0.003) were also nominally associated with glycemic response to sulfonylureas.

Figure 1

Manhattan plot of genome-wide results from single marker association with glycemic response to sulfonylureas with use of an additive genetic model in a meta-analysis consisting of 5,485 individuals with type 2 diabetes on sulfonylureas.

Figure 2

Regional association plots around genome-wide significant SNPs, rs1234032 (left) and rs10770791 (right) locus at chromosome 12, for the meta-GWAS. The purple diamonds in both plots indicate the top SNPs in the locus.

Figure 3

Forest plot of the meta-analysis of the association of HbA_1c reduction with rs1234032 (left) and rs10770791 (right) variants after sulfonylurea treatment. Information on the various cohorts can be found in Supplementary Data. The numbers in parentheses indicate the number of individuals in each of the cohorts. The last column shows the effect size [95% CI].

View this table:

Table 1

Results for index variants in the top 15 independent loci (P < 1.0 × 10⁻⁵) associated with glycemic response

The second variant, rs10770791, is located in an intron of SLCO1B1 (Fig. 2), and each copy of the C allele (frequency 49.8%) was associated with a mean ± SEM 0.11 ± 0.02% (1.2 ± 0.2 mmol/mol) greater HbA_1c reduction; P = 4.80 × 10⁻⁸. Stratified analyses showed a consistent direction of association across cohorts with similar effect sizes with no significant heterogeneity (P_het = 0.94) (Fig. 3). rs10770791 genotype was not significantly associated with sulfonylurea dose modification (P = 0.16) or drug group (the likelihood of being on mono- or dual therapy) (P = 0.29). No significant association between rs10770791 and postglipizide trough glucose concentration was observed in healthy volunteers given glipizide in SUGAR-MGH (β = −0.37, P = 0.46).

rs10770791 Is an eQTL for SLCO1B1 That Encodes OATP1B1, a Transporter of Sulfonylureas

Focusing on the SLCO1B1 locus, we performed locus-wide meta-analysis to identify the candidate causal gene (Fig. 2). We also examined two established common nonsynonymous variants in SLCO1B1, rs4149056 (*5; V174A) and rs2306283 (*1B; N130D) (30). rs4149056 (D′ = 1; r² = 0.17) and rs2306283 (D′ = 0.98; r² = 0.63) were in partial LD with rs10770791, with both rs4149056 (mean ± SEM β = 0.10 ± 0.03% [1.1 ± 0.3 mmol/mol], P = 2.72 × 10 ⁻⁴) and rs2306283 (β = 0.08 ± 0.02% [0.9 ± 0.2 mmol/mol], P = 4.32 × 10⁻⁵) nominally associated with sulfonylurea response. However, in a conditional analysis where we have individual-level data from the GoDARTS and PMET cohorts, n = 3,557 (65% of the total population), only rs10770791 remained strongly associated with sulfonylurea response (β = 0.15 ± 0.05% [2 ± 0.4 mmol/mol], P = 1.4 × 10⁻³), with rs4149056 (β = 0.03 ± 0.05% [0.3 ± 0.4 mmol/mol], P = 0.58) and rs2306283 (β = 0.06 ± 0.05% [0.7 ± 0.4 mmol/mol], P = 0.19) not significant.

We then undertook eQTL lookups of SLCO1B1 expression in 1,183 liver samples of European ancestry (24) and demonstrated that the C-allele of rs10770791 was associated with decreased SLCO1B1 expression (β = −5.24, P = 1.61 × 10⁻⁷) and, marginally, with decreased SLCO1B3 expression (β = −2.46, P = 0.01). We found directionally consistent but nonsignificant associations in the 208 liver samples examined in the GTEx project (β = −0.06, P = 0.13 for SLCO1B1).

Glyburide is a substrate of both OATP1B1 and OATP1B3 (31–35), whereas there are conflicting reports about glipizide, which has been shown to be a substrate of OATP1B3 but not OATP1B1 (31). We therefore undertook functional studies on sulfonylurea transport and observed that both glyburide and glipizide were substrates of OATP1B1 and OATP1B3 in HEK-293 cells recombinantly expressing the transporters (Fig. 4A). Further, we observed that OATP1B1 Ala174 (c.521C) had a significantly lower uptake of glyburide (P < 0.002) and a trend toward a lower uptake of glipizide (P = 0.06) compared with OATP1B1 Val174 (c.521T) (Fig. 4B).

Figure 4

Glyburide and glipizide uptake in HEK-293 Flp-ln cells recombinantly expressing SLCO1B1 or SLCO1B3. A: Uptake of [3H]-glyburide and [3H]-glipizide in HEK-293 Flp-ln stable cells expressing EV, SLCO1B1, or SLCO1B3. Rifampicin (50 μmol/L) is used as a canonical inhibitor of SLCO1B1 and SLCO1B3 P values, representing significance from EV, were determined by one-way ANOVA followed by Dunnett two-tailed test. ****P < 0.0001; ***P < 0.001; **P < 0.01; *P < 0.05. Bars represent the mean ± SEM uptake from three wells. Values shown are from a representative experiment of at least three independent studies. B: Uptake of [3H]-estrone sulfate, [3H]- glyburide, and [3H]-glipizide in HEK-293 Flp-ln stable cells expressing EV, SLCO1B1, and SLCO1B1 V174A. Estrone sulfate is a canonical substrate of SLCO1B1 and is used as a positive control in this assay. P values, for significance from EV, were determined by one-way ANOVA followed by Dunnett two-tailed test. ****P < 0.0001; ***P < 0.001; **P < 0.01; *P < 0.05. Bars represent the mean ± SEM uptake from four wells from a representative experiment. The uptake values for [3H]-glyburide and [3H]-glipizide shown are from at least four independent studies with three or four replicates per study. C: Inhibition of [3H]-glyburide uptake by atorvastatin and simvastatin acid in HEK-293 Flp-ln stable cells expressing SLCO1B1 and SLCO1B3. Each point represents the mean ± SEM uptake from four wells. Values shown are from a representative experiment of two independent studies.

Statins Inhibit Sulfonylurea Transport via OATP1B1; Genetically Reduced OATP1B1 Transport Has a Large Effect in Nonstatin Users

Given the high frequency with which hypercholesterolemia and diabetes co-occur, statins are often taken concomitantly with sulfonylureas. OATP1B1, expressed on the basolateral membrane of human hepatocytes (36), contributes to the hepatic uptake of sulfonylureas and statins from portal blood (37). We therefore sought to examine whether the initiation of statins in patients receiving sulfonylurea is associated with glycemic response in a drug-drug-gene interaction model with a sample of 3,566 adults, where we have access to individual-level data. On the basis of retrospective data from the GoDARTS and PMET cohorts, 2,096 (59%) sulfonylurea users were coprescribed statins and 1,470 (41%) were not. In a multiple linear regression model adjusted for baseline HbA_1c, statin cotreatment was associated with greater HbA_1c reduction on initiation of sulfonylurea, but only with adjustment for rs10770791 (mean ± SEM 0.22 ± 0.09% [2 ± 1.0 mmol/mol], P = 0.02). These results highlight a significant interaction between statin use and SLCO1B1 genotype (rs10770791) (P = 0.001) (Supplementary Table 4). In support of these results, we show that atorvastatin acid and simvastatin acid inhibited OATP1B1- and OATP1B3-mediated uptake of glyburide, with IC₅₀ values ranging between 0.2 and 2.9 μmol/L (Supplementary Table 5), consistent with previous studies showing that these two statins inhibit OATP1B1-mediated uptake of estradiol-17β-glucuronide (38).

We then performed stratified analysis to see whether statin use modifies the association between rs10770791 and sulfonylurea-related HbA_1c reduction using a similar model. We observed that the effect of rs10770791 was abolished in sulfonylurea users prescribed statins (mean ± SEM β = 0.053 ± 0.03% [0.6 ± 0.3 mmol/mol)], P = 0.11). However, among users of sulfonylureas without statins, we found a pronounced HbA_1c reduction associated with the C allele of rs10770791 (β = 0.23 ± 0.049% [2.4 ± 0.6 mmol/mol], P = 3.1 × 10⁻⁶) (Supplementary Table 6). C allele homozygotes at rs10770797 had a 0.48 ± 0.12% (5.2 ± 1.26 mmol/mol) greater absolute HbA_1c reduction than T allele homozygotes.

Conclusions

We report the first meta-GWAS on glycemic response to sulfonylureas and establish that this trait is heritable with a 37% heritability estimate. We have identified two novel loci at chromosome 12 and confirmed a potential involvement of the GXYLT1 and SLCO1B1 genes in glycemic response to sulfonylureas. We report large clinical effects of variants in SLCO1B1, which encodes a transporter for sulfonylureas in the liver where it is metabolized, and report interaction with coprescription of statins.

The SNP rs1234032 is an eQTL for GXYLT1 in multiple tissues including whole blood. GXYLT1 adds the first xylose to O-glucose–modified residues in NOTCH1 (31), which is a major determinant of pancreatic islet cell mass and insulin secretion and is a risk factor for diabetes (32). The C allele at rs1234032 was associated with increased expression of GXYLT1. Transgenic overexpression of human GXYLT1 was previously shown to impair Notch signaling (39). Notch signaling pathway is known to play an important role in regulating development of pancreas and also shown to be expressed in adult pancreas (40). In a recent study, Eom et al. (40) compared glucose levels, insulin secretion, and islet and β-cell masses in Notch1 antisense transgenic (NAS) and control mice after intraperitoneal glucose tolerance test. Higher glucose levels, lower insulin secretion, and decreased total islet and β-cell masses were shown in NAS in comparison with control mice. In line with this, we have shown increased trough glucose concentration with the C allele in healthy volunteers who were naïve to type 2 diabetes medications who received a glipizide challenge and, hence, worse response.

The C allele at rs10770791 was significantly associated with reduced expression of SLCO1B1 mRNA in the liver and worse glycemic response to sulfonylureas. SLCO1B1 encodes the organic anion-transporting polypeptide, OATP1B1, which facilitates the hepatic uptake of clinically relevant drugs such as statins. Gliclazide, glipizide, glyburide (glibenclamide), glimepiride, tolazamide, and tolbutamide were prescribed for the subjects in this study. Approximately 90% of the prescriptions in GoDARTS were for gliclazide, and glipizide was the main sulfonylurea in the PMET cohorts. While gliclazide and glimepiride are substrates of OATP1B1 (31,34), glyburide has been shown to be a substrate of both OATP1B1 and OATP1B3 (31,34–36,41,42). However, there are conflicting reports about glipizide, which has been shown to be a substrate of OATP1B3 but not OATP1B1 (36). Here we show that both glyburide and glipizide were substrates of OATP1B1 and OATP1B3. Further, we observed a significantly lower uptake of glyburide (P < 0.002) and a trend toward a lower uptake of glipizide (P = 0.06) for OATP1B1 Ala174 (c.521C) compared with OATP1B1 Val174 (c.521T). Examination of other known missense variants (rs60140950 [p.Gly256Ala], rs11045681 [p.Tyr311Ser], and rs11045819 [p.Pro155Thr]) in the SLCO1B3 and SLCO1B3–SLCO1B7 regions that are in partial LD with rs10770791 showed no significant association. Taken together these results suggest that the pharmacogenetic mechanism for the effect of rs10770791 on sulfonylurea response is primarily a result of altered hepatic expression of SLCO1B1 and, to a lesser extent, SLCO1B3. Partial LD of rs10770791 with various missense variants may contribute to its effect on sulfonylurea response; however, conditional analysis demonstrated association of rs10770791 with glycemic response independent of the missense variants. The reduced SLCO1B1 expression likely results in less OATP1B1-mediated transport of sulfonylurea into the liver and potentially higher plasma concentrations available at the site of action (pancreas).

There is a high prevalence of multimorbidity and subsequent polypharmacy in type 2 diabetes, highlighting a need to consider drug-drug as well as drug-drug-gene interactions in prediction models of glycemic response to sulfonylureas. Given that statins are often taken concomitantly with sulfonylureas, with both being substrates of OATP1B1, we examined for a possible drug-drug-gene interaction and showed a significant interaction between statin use and SLCO1B1 genotype (rs10770791) (P = 0.001). Stratified analysis by statin use showed differential effects of rs10770791 in statin users and nonusers. While the association between rs10770791 and glycemic response to sulfonylureas was abolished in statin users, it was more pronounced in statin nonusers. In those not treated with statins nearly one-quarter of the population who carry two C alleles at rs10770791 had a 0.48% (5.2 mmol/mol) greater HbA_1c reduction compared with T allele homozygotes. These large effects are the equivalent of those in starting a dipeptidyl peptidase 4 inhibitor (43) and equated to a dose difference of 28 mg gliclazide. Our findings suggest that the previous reported observational association between statins and hypoglycemia in sulfonylurea users (44) may be explained by interactions at SLCO1B1, depending on the underlying genotype. The findings are consistent with previous studies in healthy volunteers and rodents demonstrating that atorvastatin administration is associated with increased levels of glimepiride (45) and glyburide (46), respectively. Given that there is a strong recommendation to use statins by recent guidelines, statin use is increasing among people with diabetes (47). Therefore, integrating comedications with genetic data could improve optimization of polypharmacy regimens.

This study has some limitations. First, the modest sample size does not have sufficient power to detect the contribution of rare and low-frequency variants in heritability estimation and/or glycemic response to sulfonylureas. However, this is the first GWAS and largest pharmacogenomic study on sulfonylureas response so far. Second, this study was conducted in Whites of European descent, and therefore the results may not generalize to other populations. Third, even though we have performed several validation studies, direct replication of the findings in an independent study is warranted. Finally, further studies need to be done to elucidate the biological mechanism of the identified associations, especially for GXYLT1.

In conclusion, we have established that common genetic variants contribute to the variation in glycemic response to sulfonylureas, with an estimated heritability of 37%. This result shows that a moderate proportion of the variance in glycemic response is genetic, with an important role for common genetic variation in glycemic response to sulfonylureas. We report that a variant that modulates gene expression and circulating GXYLT1 reduces response to sulfonylureas. We have also revealed a robust association between rs10770791, a cis eQTL for SLCO1B1 expression in the liver, and glycemic response to sulfonylureas, with reduced SLCO1B1 expression associating with increased response to sulfonylureas. Our results suggest the potential of rs10770791 to be a biomarker for stratified medicine in diabetes. In addition, we have highlighted significant drug-drug-gene interactions for sulfonylurea, statin use, and rs10770791, with clinically actionable genetic effects with pronounced differences in HbA_1c reduction in a subgroup of patients treated with sulfonylureas without statins. Over the next 5 years we will see an ever-increasing availability of genotype or sequence data embedded in the medical records; given replication, the SLCO1B1-statin interaction could be clinically actionable and will need to be taken into account at the point of prescribing sulfonylureas.

Article Information

Acknowledgments. The authors acknowledge the following individuals for their contributions in providing information required for the studies: Dana Fraser from Parexel, who completed imputation for the Metformin Response (METRO) cohort; Dilrini Ranatunga from Kaiser Permanente Northern California Division of Research, for providing the phenotype and genotype data for the PMET1 cohort and phenotype data for PMET2 cohort; Sara R. Rashkin Center for Applied Bioinformatics, St. Jude Children’s Research Hospital, Memphis, TN, for providing advice related to computational coding and analyses required for the PMET cohorts; and Xujia Zhou, University of California, San Francisco, CA, for experimental assistance required for this study. The authors also acknowledge Jennifer L. Aponte, Genomic Medicine, Parexel; Margaret G. Ehm, Target Sciences, GlaxoSmithKline; and Dawn M. Waterworth, Target Sciences, GlaxoSmithKline for their contribution to data analysis and revising the manuscript. The authors also acknowledge clinicalstudydatarequest.com for access to the HARMONY data. Finally, the authors thank all study participants.

Funding and Duality of Interest. The work leading to this publication has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no. 115317 (DIRECT), resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and European Federation of Pharmaceutical Industries (EFPIA) companies’ in kind contribution. Funding was in part from the National Institutes of Health (NIH), R01-GM117163, to J.C.F., M.M.H., and K.M.G. E.R.P. holds a Wellcome New Investigator Award (102820/Z/13/Z). Funding for SUGAR-MGH was provided by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), NIH, R01-DK088214. J.H.L. is supported by NIDDK, NIH, T32-DK007028. J.C.F. is supported by NIDDK, NIH, K24-DK110550. Geisinger MyCode type 2 diabetes project was supported by the Geisinger Health Plan Quality Pilot Fund (Principal Investigator: M.T.M.L.). E.R.P. has received honoraria for speaking from Lilly and Sanofi. J.C.F. has received honoraria for speaking at scientific conferences from Novo and for consulting from Goldfinch Bio. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.Y.D, S.W.Y, L.M.‘t.H., K.M.G., and E.R.P. contributed to conception and design of the study. A.Y.D., S.W.Y., N.V.L., Y.Z., M.K.S., A.E., F.I, F.X., J.H.L., R.C.S., and Y.-C.C. contributed to data analysis. S.W.Y., A.E., F.I., J.M.M, V.K., J.S.W., M.T.M.L., Y.K., Y.M., M.K., C.N.A.P., J.C.F., M.M.H., and L.M.‘t.H. contributed to data collection and genotyping. A.Y.D., S.W.Y., K.M.G., and E.R.P. contributed to manuscript writing, with contributions from all authors on the final version. A.Y.D. S.W.Y, K.M.G., and E.R.P are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.