Disentangling Sex Differences in Sulfonylurea Drug Response With Genome‐Wide Association Studies in Individuals With Type 2 Diabetes

Abstract

Sulfonylureas are a cornerstone of type 2 diabetes therapy despite interindividual variability in response. Despite well‐documented sex‐based differences, pharmacogenomic and genome‐wide association studies (GWAS) have largely overlooked sex as a biological variable. We conducted the first sex‐stratified GWAS of hemoglobin A1c (HbA1c) response to sulfonylureas in Action to Control Cardiovascular Risk in Diabetes (ACCORD) clinical trial participants (N = 871). Variants meeting genome‐wide (P < 5.0 × 10^–8) and suggestive (P < 5.0 × 10^–6) significance were assessed for replication in the Pharmacogenomics of Metformin (PMET1) cohort. Replicated variants were further analyzed in the Study to Understand the Genetics of the Acute Response to Metformin and Glipizide in Humans (SUGAR‐MGH) cohort to assess acute insulin and glucose responses to a single glipizide dose. Genome‐wide significant loci with sex‐specific effects were identified: KAZN, KIF2B, SLC39A10, and SPINK5 (combined‐sex); CRACR2A, KCNK2, and TENM2 (male‐only); and NACPH2 (female‐only). Two suggestive variants in the TMEM64/NECAB1 locus, associated with reduced HbA1c response to sulfonylureas in the male‐only ACCORD analysis, were directly replicated in the PMET1 male‐only cohort. In SUGAR‐MGH, one replicated variant (rs6471250‐C) was significantly associated with reduced peak insulin in males (P = 0.035) but not females (P = 0.40), demonstrating sex‐specific functional effects. This study identified statistically supported and biologically plausible loci with prior evidence linking nearby genes to pathways relevant to sulfonylurea action, including insulin secretion, insulin regulation/sensitivity, calcium signaling, potassium‐channel biology, and glucose transport. The findings highlight sex‐specific differences in sulfonylurea response, providing mechanistic insights and underscoring the importance of sex‐specific precision medicine. Identification of genetic variants influencing sex‐specific response could inform dosing to optimize sulfonylureas.

A sex‐stratified genome‐wide association study of sulfonylurea response identified sex‐specific genetic loci associated with HbA1c reduction in individuals with type 2 diabetes. Replication and functional validation supported a male‐specific association at the TMEM64/NECAB1 locus, highlighting the potential of sex‐informed pharmacogenomics to advance precision prescribing of sulfonylureas.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Sulfonylurea response varies substantially among individuals, but the contribution of sex‐specific genetic variation remains poorly understood.

• WHAT QUESTION DID THIS STUDY ADDRESS?

Do genetic variants associated with glycemic response to sulfonylureas differ between males and females, and can sex‐specific associations be replicated and supported by functional evidence?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

The first sex‐stratified GWAS of sulfonylurea response identified multiple sex‐specific loci, including replicated male‐specific associations at the TMEM64/NECAB1 locus. Functional analysis demostrated that rs6471250 was associated with reduced glipizide‐stimulated insulin secretion in males, providing biological support for sex‐specific genetic effects on sulfonylurea response.

• HOW MIGHT THIS CHANGE CLINICAL PHARMA‐COLOGY OR TRANSLATIONAL SCIENCE?

These findings support incorporating sex as a biological variable in pharmacogenomic studies and provide candidate genetic markers for future precision prescribing of sulfonlyureas. Although additional validation is needed before clinical implementation, sex‐specific genetic information may ultimately help optimize sulfonylurea therapy for individuals with type 2 diabetes.

Diabetes represents a major global health burden; the International Diabetes Federation estimates that approximately 589 million adults aged 20–79 years were living with diabetes worldwide in 2024 (https://diabetesatlas.org). Type 2 diabetes accounts for the majority of adult diabetes and remains the principal clinical context for sulfonylurea therapy. Despite the development of newer therapeutic agents, metformin and sulfonylureas remain among the most widely prescribed medications for type 2 diabetes. ¹ Pharmacogenomic and genome‐wide association studies (GWAS) have successfully identified actionable biomarkers and elucidated mechanisms of action for metformin. ² , ³ , ⁴ In contrast, genetic investigations of glycemic response to sulfonylureas remain limited, restricted to candidate‐gene approaches and a single genome‐wide study, with only a recent GWAS examining acute glucose and insulin responses to a single glipizide dose. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Although sex‐based differences in sulfonylurea efficacy have been documented in the literature, no sex‐stratified GWAS of sulfonylurea response has been conducted. ¹⁰

The mechanism of sulfonylurea action is well characterized at the molecular level, involving enhanced insulin secretion from remaining functional pancreatic β‐cells. These agents bind to the sulfonylurea receptor (SUR), inhibit potassium influx into pancreatic β‐cells, depolarize the cell membrane, and trigger calcium influx. Calcium influx promotes insulin exocytosis via actomyosin contraction, and the released insulin subsequently facilitates glucose uptake in muscle tissue. ¹¹ However, while this mechanism is well understood, the substantial heterogeneity in individual glycemic responses to sulfonylureas remains unexplained. Genetic studies offer a promising approach to identify the genomic contributions underlying this variable response.

Sex‐based differences in sulfonylurea response have been documented, with men typically achieving greater HbA1c reduction while women experience higher rates of hypoglycemia and weight gain, potentially influenced by hormonal milieu and other sex‐related biological or clinical factors. ¹⁰ , ¹² , ¹³ Despite this evidence, current clinical guidelines do not recommend sex‐specific dosing adjustments. More broadly, although genetic effects on drug response are frequently sex‐specific, pharmacogenomic studies have rarely incorporated sex‐stratified analyses. ¹⁴ , ¹⁵ , ¹⁶ This gap is exemplified by the only previous GWAS of HbA1c response to sulfonylureas, which did not examine sex‐specific associations, and similarly by SUGAR‐MGH (Study to Understand the Genetics of the Acute Response to Metformin and Glipizide in Humans). ⁸ , ⁹ Candidate‐gene studies have examined genetic variation in genes involved in sulfonylurea metabolism or therapeutic mechanism, including CYP2C9, the primary metabolizing enzyme, ABCC8, KCNJ11, and TCF7L2, and their effects on drug response. ⁵ , ⁶ , ⁷ , ¹⁷ These studies, however, adjusted for biological sex as a demographic variable rather than investigating sex‐specific genetic contributions.

We present the first sex‐stratified GWAS of HbA1c response to sulfonylureas and validate key findings in external datasets. We observed concordant sex‐specific effects at the TMEM64/NECAB1 locus across independent studies and present functional evidence supporting a role for this locus in insulin regulation. Our findings provide new mechanistic insights into genetic contributions to sulfonylurea response and underscore the importance of sex‐specific approaches in precision medicine for type 2 diabetes.

RESEARCH DESIGN AND METHODS

The action to control cardiovascular risk in diabetes (ACCORD) trial

Study participants

The ACCORD randomized controlled trial (RCT) (clinicaltrials.gov‐NCT00000620) is described in detail elsewhere. ¹⁸ Briefly, ACCORD employed a double 2 × 2 factorial design comparing intensive vs. standard glycemic control and investigating strategies for blood pressure and lipid management. ACCORD enrolled 10,251 individuals with type 2 diabetes who had established cardiovascular disease (CVD) or at least two CVD risk factors. ¹⁸ Informed consent was obtained using procedures reviewed and approved by each site's institutional review board, and the trial received central approval and monitoring from the National Heart, Lung, and Blood Institute (IRB: FWA00003429). Participants self‐reported their racial and ethnic identity across five categories: non‐Hispanic White (NHW), non‐Hispanic Black (NHB), Hispanic, Asian, and Other. Sex was taken from ACCORD trial enrollment records.

Phenotype definition

We evaluated glycemic response to four sulfonylureas (gliclazide, glimepiride, glipizide, and glyburide) among ACCORD participants who initiated sulfonylurea treatment during trial participation. Participants reporting sulfonylurea use before enrollment were excluded to isolate medication effects. To isolate the effect of sulfonylurea initiation and minimize confounding from polytherapy, we restricted analysis to participants whose only medication change during the analysis window was sulfonylurea initiation. To account for concomitant medication use, medication scores derived from aggregated medication types, dosages, and treatment durations were included as potential covariates. ¹⁹

Hemoglobin A1c (HbA1c) was measured every 4 months, and sulfonylurea response was calculated as the change in HbA1c (ΔHbA1c = on‐treatment minus pre‐treatment). The four‐month measurement schedule was defined by the original ACCORD protocol. We used the first eligible post‐initiation HbA1c value rather than an average across the window to capture early treatment response and reduce confounding from subsequent dose titration, adherence changes, treatment intensification, or additional medication changes. Pre‐treatment HbA1c was defined as the measurement recorded within 30 days before sulfonylurea initiation (≤ 30 days). On‐treatment HbA1c was defined as the first measurement recorded between 90 and 270 days after sulfonylurea initiation. This timeframe allows sufficient time for HbA1c stabilization following treatment initiation while minimizing confounding from subsequent treatment modifications.

Medication compliance was assessed from participant self‐reports during the 90‐ to 270‐day treatment window. Participants were required to meet both criteria ¹ : adherence of 80–100% at ≥ 80% of study visits, and ² no visits with reported 0% adherence. For missing compliance records, the subsequent record was carried forward.

Genotyping

Genotyping data processing, quality control, and imputation are described elsewhere. ² , ²⁰ Briefly, genotypes were imputed to the 1,000 Genomes reference panel for 8,174 ACCORD participants who provided genetic consent.

Covariate selection

Relevant covariates were identified using backward variable selection and Bayesian information criterion (BIC). ¹⁹ Trial arm and pre‐treatment HbA1c were retained as a priori covariates based on clinical relevance. All considered covariates and the respective univariate associations with sulfonylurea are listed in Table S1–S11 . Only samples with complete phenotype and covariate data were included in final analyses. To address multicollinearity, one variable was excluded from each correlated pair (Pearson's |r| > 0.5). Genetic ancestry was controlled using principal component analysis (PCA) implemented in EIGENSTRAT (v4.2). The first three principal components (PCs) were included in all models; PCs 4–10 were eligible for inclusion through variable selection. Final covariate models used in association analyses are presented in Tables S2–S4 .

Common variant analysis

Associations between sulfonylurea response (defined as on‐treatment minus pre‐treatment HbA1c), covariates, and common autosomal variants (minor allele frequency [MAF] > 1%) were tested using linear regression under an additive genetic model in PLINK v1.9. X‐chromosome variants were analyzed separately using PLINK 2 with the default X‐chromosome model. ²¹ Y‐chromosome variants were not included in this study. Because Y‐chromosome analyses are necessarily male‐only, they would not provide a comparable female stratum and would require a separate analytic framework outside the prespecified combined‐sex, male‐only, and female‐only design. Genome‐wide association studies were conducted across three strata: multi‐ancestry combined‐sex (N = 870), male‐only (N = 520), and female‐only (N = 350). Given the limited sample size in pharmacogenomic studies, we employed a tiered approach for variant selection using genome‐wide significance (P < 5.0 × 10⁻⁸) and suggestive significance (P < 5.0 × 10⁻⁶) thresholds. Suggestive thresholds are frequently used in genetics studies because variants that do not initially achieve genome‐wide significance may still represent biologically relevant associations that replicate in independent cohorts or achieve significance in larger meta‐analyses. ²² , ²³ , ²⁴ Accordingly, suggestive loci were not interpreted as definitive associations, but rather as candidates for replication and functional follow‐up. Variants meeting either threshold were carried forward for external replication. To minimize spurious associations, we required that significant and suggestive loci (MAF 1–3%) demonstrated multiple independent associations within a 1 Mb window (i.e., ≥ 2 significant and/or suggestive variants per locus). SNP‐based heritability was estimated using genome‐wide complex trait analysis (GCTA v1.22) separately for each stratum, excluding related individuals (genetic relatedness > 0.05) and adjusting for selected covariates. ²⁵

To formally test for sex‐specific genetic effects, we examined the 13 lead variants (i.e., the variant with the lowest P value at each locus) identified in the male‐only and female‐only ACCORD analyses. For each variant, a linear regression model was fit in R using combined‐sex data, including the following terms: additive genotype effect, sex, genotype:sex interaction, and covariates (trial arm [glycemia, blood pressure, fibrate], pre‐treatment HbA1c, PCs 1–3, diabetes duration, baseline CVD status, dyslipidemia duration, and pre‐treatment eGFR). Statistical significance was set at P < 0.00128 (Bonferroni‐corrected for 39 tests: 3 terms [genotype, sex, genotype:sex] × 13 variants).

Common variant gene‐based analysis and colocalization

To improve statistical power and identify genes with aggregated effects across multiple variants, we conducted gene‐level analysis using FUMA (MAGMA) for 20,161 genes. ²⁶ , ²⁷ Variants were included in gene sets if located within ±50 kb of untranslated regions of target genes. FUMA used the 1,000 Genomes Project reference panel for linkage disequilibrium (LD) estimation within gene sets. ²⁸ Gene‐level discovery was defined as Bonferroni‐corrected significance (P < 2.48 × 10⁻⁶; 0.05/20,161 genes) with suggestive significance at P < 2.48 × 10⁻⁴. Gene Ontology enrichment analysis was performed using FUMA to identify overrepresented biological processes (GOBP) and molecular functions (GOMF) among genes mapped from the GWAS results.

To identify potential causal variants, we assessed colocalization between GWAS signals and expression quantitative trait loci (eQTL). Colocalization was assessed using LocusFocus (simple‐sum method) incorporating expression data from GTEx across tissue‐gene combinations. ²⁹ Colocalization analysis examined characterized genes within ±250 kb of each lead variant across four tissues: pancreas, brain, blood, and muscle (16 total tissue‐gene pair combinations). For colocalization analysis, the Bonferroni‐corrected significance threshold was P < 0.0056 (0.05/9 gene‐tissue pairs tested).

Pharmacogenomics of metformin study

The Pharmacogenomics of Metformin (PMET1) cohort was derived from the Genetic Epidemiology Research on Adult Health and Aging (GERA) study, a subsample of the Kaiser Permanente Research Program on Genes, Environment, and Health (RPGEH). Genotyping, quality control, and imputation methods are described in detail elsewhere. ⁴ Phenotype data were extracted from electronic health records (EHRs).

Phenotype definition

Glycemic response to four sulfonylureas (gliclazide, glimepiride, glipizide, and glyburide) was defined as the change in HbA1c (on‐treatment minus pre‐treatment). Pre‐treatment HbA1c was defined as the eligible measurement closest to 90 days before sulfonylurea prescription, with measurements recorded between 180 days before and 21 days after prescription considered eligible. On‐treatment HbA1c was defined as the eligible measurement closest to 365 days after sulfonylurea prescription, with measurements recorded between 90 and 450 days after prescription considered eligible. Using the closest eligible post‐prescription HbA1c measurement, rather than averaging all available EHR values, minimized heterogeneity due to unequal numbers and timing of clinical measurements across participants.

Concomitant HbA1c‐altering medications (other than sulfonylureas) were identified if prescribed within the pre‐treatment or on‐treatment measurement windows and categorized into five levels: (1) none; (2) stable before baseline; (3) initiated at baseline; (4) initiated with sulfonylurea; (5) initiated during the on‐treatment period. This categorical variable was included as a covariate in all PMET1 analyses.

Common variant analysis

Change in HbA1c (ΔHbA1c = on‐treatment minus pre‐treatment) was regressed against: age at diabetes diagnosis, sex, pre‐treatment HbA1c, days from pre‐treatment measurement to sulfonylurea prescription, days from sulfonylurea prescription to on‐treatment measurement, average sulfonylurea dose, and interaction terms for metformin, insulin, and thiazolidinedione use. The residual ΔHbA1c (adjusted for all covariates) was used as the outcome in GWAS analyses with SNPTEST v2.5.4, stratified by genetic ancestry and sex. Summary statistics from ancestry‐specific analyses were meta‐analyzed using inverse‐variance weighting to generate three multi‐ancestry strata: combined‐sex (N _max = 3,514), male‐only (N _max = 1,908), and female‐only (N _max = 1,606).

Genome‐wide association study replication

ACCORD variants (MAF ≥ 1%, P < 5.0 × 10⁻⁶) were tested for replication in the PMET1 study, which served as an independent replication cohort. ACCORD variants were assigned to loci using a ± 500 kb interval around each lead variant. Given the limited sample size, two levels of replication support were defined (1): direct replication, defined as variants showing consistent direction of effect in the same PMET1 stratum with Bonferroni‐corrected significance (combined‐sex: P < 6.70 × 10⁻⁴, n = 75 loci; male‐only: P < 5.30 × 10⁻⁴, n = 95 loci; female‐only: P < 1.90 × 10⁻³, n = 27 loci); and (2) meta‐analysis support, defined as meta‐analysis P value smaller than the original ACCORD P value, indicating consistency across studies despite not meeting study‐specific significance thresholds. Meta‐analyses were performed using inverse‐variance weighted fixed‐effects models in METAL, combining like strata (combined‐sex, male‐only, female‐only) across ACCORD and PMET1.

Exploring functional roles of variants associated with sulfonylurea response

The Study to Understand the Genetics of the Acute Response to Metformin and Glipizide in Humans (SUGAR‐MGH) has been described in detail elsewhere. ⁹ Briefly, SUGAR‐MGH was a pharmacogenomic study in which 1,000 individuals at risk for or with untreated type 2 diabetes (controlled by lifestyle modification alone) received a short course of metformin and a single 5 mg dose of glipizide, with glucose and insulin measurements at 30, 60, 90, 120, 180, and 240 minutes. Functional effects of replicated variants (rs80334907, rs6471250) were assessed using two analytical approaches: (1) multi‐ancestry combined‐sex linear regression including a genotype:sex interaction term, and (2) sex‐stratified linear regression. Outcomes were rank‐inverse normalized peak log insulin (adjusted for baseline insulin) and glucose trough (adjusted for baseline glucose), with all models adjusted for age, BMI, and the first 10 principal components of genetic ancestry.

RESULTS

Sex‐based differences in HbA1c response to sulfonylureas were statistically significant (P < 0.001; Table 1 ). Males achieved greater mean HbA1c reduction (−1.58%; 95% CI: −1.68 to −1.47) compared to females (−1.30%; 95% CI: −1.43 to −1.17), consistent with previous reports of sex‐specific differences in sulfonylurea efficacy. Race/ethnicity‐based differences in sulfonylurea response were also significant. Non‐Hispanic White (NHW) individuals showed significantly better response than non‐Hispanic Black (NHB) individuals (P = 0.0013), while no significant difference was detected between NHW and Hispanic individuals (P = 0.14) (Figure 1 ). The estimated heritability of HbA1c response to sulfonylurea in the multi‐ancestry male‐only stratum was 0.87 [SE: 0.22, P = 4.9 × 10⁻⁵]. The female‐only stratum showed no significant heritability (h ² = 0.01, SE = 0.44, P = 0.50). The combined‐sex stratum showed significant intermediate heritability (h ² = 0.69, SE = 0.19, P = 0.0002).

Table 1.

Cohort demographics and concomitant medications

	Multi‐ancestry all (N = 871)	Multi‐ancestry male‐only (N = 520)	Multi‐ancestry female‐only (N = 351)
Female (%)	40.30	—	—
Age (mean [SD])	61.60 [6.18]	61.51 [6.25]	61.72 [6.09]
BMI (mean [SD])	32.56 [5.59]	31.85 [5.43]	33.60 [5.66]
Years with type 2 diabetes (mean [SD])	8.59 [7.08]	8.32 [7.18]	9.00 [6.90]
History of cardiovascular disease (%)	31.92	36.92	24.50
Non‐Hispanic White (%)	64.4	68.5	58.4
Non‐Hispanic Black (%)	14.8	10.8	20.8
Hispanic (%)	5.9	4.8	7.4
Other (%)	14.9	15.9	13.4
Pre‐treatment HbA1c (%) (median [IQR])	8.2 [7.6, 9.2]	8.2 [7.6, 9.2]	8.2 [7.6, 9.2]
On‐treatment HbA1c (%) (median [IQR])	6.7 [6.2, 7.4]	6.6 [6.1, 7.3]	6.9 [6.4, 7.5]
∆ HbA1c (%) (median [IQR])	−1.4 [−2.2, −0.6]	−1.5 [−2.3, −0.7]	−1.2 [−2.1, −0.5]
Intensive glycemia arm (%)	56.14	56.92	54.99
Standard glycemia arm (%)	43.86	43.08	45.01
Intensive blood pressure arm (%)	22.39	20.77	24.79
Standard blood pressure arm (%)	23.53	19.23	29.91
Fibrate lipid treatment arm (%)	27.10	30.19	21.71
Placebo lipid treatment arm (%)	26.98	29.81	22.79
Angiotensin type 2 antagonists (%)a	15.73	13.85	18.52
Ace inhibitors (%)a	45.24	47.31	42.17
Alpha‐glucosidase inhibitors (%)a	1.03	0.58	1.71
Cholesterol absorption inhibitors (%)a	1.84	1.15	2.28
Statin (%)a	52.12	57.69	43.87
Lisinopril (%)a	15.04	15.00	15.10
Loop diuretics (%)a	8.04	5.77	11.40
Meglitinides (%)a	5.63	5.38	5.98
Metformin (%)a	81.06	83.27	77.78
Nitrates (%)a	5.28	6.73	3.13
Thiazolidinediones (%)a	37.31	40.58	32.48
Insulin (%)a	24.57	19.04	32.76

^a Tabulated percentages represent the percentage of participants with at least one record of taking the medication, or a medication in the drug class, during the sulfonylurea treatment window. This is not an exhaustive list of medications recorded in ACCORD. Rather, it is a representative list of the medications commonly used by participants in the trial.

Figure 1.

Median, interquartile range, and distributions of HbA1c response to sulfonylurea treatment. Violin and box plots of pre‐treatment, on‐treatment, and HbA1c difference across (a–c) Multi‐ancestry male and multi‐ancestry female‐only strata, and (d–f) Non‐Hispanic White, non‐Hispanic Black, and Hispanic strata.

Common variants associated with response to sulfonylureas

In the combined‐sex stratum, 47 variants reached genome‐wide significance (P < 5.0 × 10⁻⁸), mapping to five distinct loci: KAZN, KIF2B, SLC39A10, SPINK5, and VSX2 (Figure 2 a , Table 2 a ). Sex‐stratified analyses identified genome‐wide significant variants in the male‐only (21 variants, 10 loci) and female‐only (1 variant, 1 locus) strata, collectively representing 13 unique loci: BCAS3, BMP2, C4orf32, CRACR2A, FBXO33, KCNK2, TENM2, VSX2, WASL, and XNPEP1 (male‐only; Figure 2 b , Table 2 a ) and NACPH2 (female‐only; Figure 2 c , Table 2 a ). Of the 13 loci identified in sex‐stratified analyses, five showed significant genotype:sex interactions: BMP2, CRACR2A, NACPH2, TENM2, and WASL (Table 2 b ). Genomic inflation factors (λ _GC) ranged from 1.022 to 1.082 across analyses (combined‐sex: 1.081; male‐only: 1.082; female‐only: 1.022), indicating minimal systematic bias.

Figure 2.

Genetic associations with HbA1c response to sulfonylurea treatment in ACCORD. Panels (a–c) show Manhattan plots of HbA1c response in the multi‐ancestry combined‐sex (N = 871), male‐only (N = 520), and female‐only (N = 351) strata. Horizontal dashed red lines indicate genome‐wide significance (P = 5.0 × 10⁻⁸).

Table 2.

Genome‐wide significant loci identified in combined and sex‐stratified ACCORD analyses

(A) Discovery ACCORD GWAS												(B) Sex interaction model
Strata	Gene	Chr	rsID	EA	OA	EAF	AFRa	EURa	AMRa	β (95% CI)	P	Model term	β	SE	P
Multi‐ancestry	KAZN	1	rs74059606	T	C	0.076	0.222	0.041	0.073	0.43 (0.28, 0.59)a	3.3 x 10−8	—	—	—	—
	SLC39A10	3	rs199540697	A	AAAGT	0.015	0.001	0.011	0.004	1.06 (0.70, 1.42)	8.5 x 10−9
	SPINK5	5	rs3777130	G	A	0.46	0.581	0.39	0.348	−0.22 (−0.30, −0.15)	1.7 x 10−8
	VSX2	14	rs80258392	T	C	0.012	0.046	—	—	1.03 (0.69, 1.38)	5.8 x 10−9
	KIF2B	17	rs9907326	C	T	0.011	0.056	—	0.007	1.10 (0.74, 1.47)	6.3 x 10−9
Multi‐ancestry male‐only	KCNK2	1	rs75060629	C	T	0.038	0.24	0.003	0.046	−0.88 (−1.19, −0.57)	3.5 x 10−8	Genotype	−0.55	0.14	1.1 x 10 −4
												Male	−0.18	0.06	0.0016
												Interaction	−0.40	0.18	0.03
	C4orf32	4	rs200233793	A	ACT	0.012	—	—	—	−1.40 (−1.89, −0.92)	2.1 x 10−8	Genotype	−1.12	0.24	3.3 x 10 −6
												Male	−0.19	0.06	5.7 x 10 −4
												Interaction	−1.12	0.33	5.9 x 10 −4
	TENM2	5	rs116498056	A	G	0.013	0.104	0.003	0.008	1.55 (1.07, 2.03)	5.1 x 10−10	Genotype	1.57	0.24	1.2 x 10 −10
												Male	−0.27	0.06	9.9 x  −7
												Interaction	1.33	0.29	3.9 x 10 −6
	WASL	7	rs73718476	C	T	0.012	0.084	—	0.009	1.24 (0.80, 1.68)	5.0 x 10−8	Genotype	1.26	0.22	9.9 x 10 −9
												Male	−0.27	0.06	1.2 x 10 −6
												Interaction	1.12	0.28	6.7 x 10 −5
	XNPEP1	10	rs74154924	C	T	0.01	0.108	—	0.009	1.66 (1.09, 2.22)	1.4 x 10−8	Genotype	1.64	0.28	3.5 x 10 −9
												Male	−0.28	0.06	4.8 x 10 −7
												Interaction	1.68	0.36	3.0 x 10 −6
	—	10	rs74137131	T	C	0.021	0.028	—	—	−0.98 (−1.31, −0.66)	7.7 x 10−9	Genotype	−0.77	0.16	1.9 x 10 −6
												Male	−0.19	0.06	5.8 x 10 −4
												Interaction	−0.58	0.21	0.0054
	LINC00999	10	rs114747652	G	T	0.025	0.002	—	—	1.05 (0.71, 1.38)	1.2 x 10−9	Genotype	1.00	0.16	9.6 x 10 −10
												Male	−0.30	0.06	8.9 x 10 −8
												Interaction	1.03	0.24	2.0 x 10 −5
	CRACR2A	12	rs7306983	A	G	0.01	0.11	0.026	0.015	−1.45 (−1.95, −0.94)	3.3 x 10−8	Genotype	−1.22	0.26	2.6 x 10 −6
												Male	−2.50	0.68	2.6 x 10 −4
												Interaction	−1.15	0.34	8.8 x 10 −4
	FBXO33	14	rs17110805	T	C	0.018	0.08	–	0.004	1.08 (0.71, 1.46)	2.2 x 10−8	Genotype	1.05	0.18	1.8 x 10 −8
												Male	−0.28	0.06	3.6 x 10 −7
												Interaction	1.08	0.29	1.7 x 10 −4
	VSX2	14	rs74061765	C	T	0.017	0.112	0.002	0.018	1.16 (0.76, 1.57)	3.7 x 10−8	Genotype	1.19	0.2	5.8 x 10 −9
												Male	−0.27	0.06	1.2 x 10 −6
												Interaction	0.71	0.27	0.0099
	BCAS3	17	rs144032083	A	G	0.06	0.507	0.003	0.078	−0.75 (−1.02, −0.49)	4.0 x 10−8	Genotype	−0.41	0.12	5.1 x 10 −4
												Male	−0.20	0.06	8.8 x 10 −4
												Interaction	−0.24	0.11	0.036
	BMP2	20	rs75313787	A	G	0.013	0.098	—	0.024	1.25 (0.81, 1.69)	4.5 x 10−8	Genotype	1.20	0.22	3.7 x 10 −8
												Male	−0.27	0.06	1.0 x 10 −6
												Interaction	1.09	0.27	4.5 x 10 −5
Multi‐ancestry female‐only	NACPH2	20	rs73172241	T	C	0.13	0.081	0.165	0.100	0.54 (0.37, 0.72)	4.8 x 10−9	Genotype	−0.03	0.08	0.67
												Female	0.08	0.06	0.19
												Interaction	0.56	0.12	1.6 x 10 −6

(A) Sentinel loci of discovery GWAS, (B) Sex‐interaction model of sex‐specific sentinel loci. Bolded values reached Bonferroni‐corrected signficance. AFR, African Genetic Ancestry; AMR, Admixed‐American Genetic Ancestry; EA, effect allele; EAF, effect allele frequency; EUR, European Genetic Ancestry; OA, other allele.

^a Population level ancestry‐specific allele frequency reported by dbSNP.

At the suggestive significance threshold (P < 5.0 × 10⁻⁶), 395 variants mapped to 50 loci in the combined‐sex stratum, 816 variants mapped to 68 loci in the male‐only stratum, and 89 variants mapped to 19 loci in the female‐only stratum (Table S5 ). Notable suggestive loci included: ADAM19, CALCB, COL4A2, DLGAP1, KAZN, KCNK1, PIK3R1, and SPINK5 in the combined‐sex stratum; BCAS3, BMP2, CRACR2A, KCNK2, TCEA3, TENM2, TMEM64/NECAB1, VSX2, and WASL in the male‐only stratum; and AGAP1, CACNA1S, KCNN3, MAGI2, and NACPH2 in the female‐only stratum.

Gene‐based analysis

No genes met the Bonferroni‐corrected significance threshold. However, several genes showed suggestive significance, including CNGB3, SLC7A14, SLC23A2, SLC50A1, and SPINK5 (Tables S6–S8 ). In the combined‐sex stratum, variants in the SPINK5 locus demonstrated significant colocalization with expression quantitative trait loci (eQTL) in skeletal muscle tissue (P < −log₁₀(2.25); Figure S1 A,B). Gene Ontology analysis identified multiple biological processes and molecular function pathways containing SPINK5, including those involved in regulation of serine proteases, endopeptidases, and hydrolase activity (Figure S1 C,D , Table S9 ). ACCORD gene‐based analysis results were not significantly replicated in PMET1.

Direct variant replication and meta‐analysis support with PMET1

ACCORD findings were tested for replication in PMET1 (N = 3,514) using variants meeting suggestive significance across any ACCORD stratum (MAF ≥ 1%, P < 5.0 × 10⁻⁶; Table S5 ). Two suggestive variants in TMEM64/NECAB1 achieved direct replication in the male‐only stratum. Specifically, rs80334907‐A (β = 1.00, 95% CI: 0.63–1.35, P = 1.3 × 10⁻⁷) and rs6471250‐C (β = 0.72, 95% CI: 0.42–1.02, P = 4.4 × 10⁻⁶) were associated with reduced sulfonylurea effectiveness. Both variants showed direct replication in the PMET1 male‐only stratum with consistent effect direction: rs80334907‐A (β = 1.43, 95% CI: 0.73–2.14, P = 6.4 × 10⁻⁵) and rs6471250‐C (β = 1.25, 95% CI 0.62–1.88, P = 1.1 × 10⁻⁴) (Table 3 ).

Table 3.

Direct replication of ACCORD genetic associations in PMET1

Gene	rsID	CHR	POS	EA	OA	ACCORD				PMET1
						EAF	β (95% CI)	P	N	EAF	β (95% CI)	P	N
TMEM64/NECAB1	rs80334907	8	91653113	A	C	0.014	1.00 (0.63, 1.36)	1.3 x 10 −7	520	0.016	1.43 (0.73, 2.14)	6.4 x 10 −5	248
	rs6471250	8	91730728	C	T	0.033	0.72 (0.42, 1.02)	4.4 x 10 −6		0.028	1.25 (0.62, 1.88)	1.1 x 10 −4

Bolded values reached Bonferroni‐corrected significance.

CHR, chromosome; EA, effect allele; EAF, effect allele frequency; OA, other allele; POS, base pair position.

Meta‐analysis identified directional consistency for 10 variants in the combined‐sex stratum, 136 in the male‐only stratum, and 11 in the female‐only stratum (Table S10 ). These variants mapped to genes involved in biologically plausible pathways related to sulfonylurea mechanism of action, including insulin signaling (ADGRB3, COL5A3, RAB28, SCARB1) and calcium transport (CRACR2A, TMEM64/NECAB1). ³⁰ , ³¹ , ³² , ³³

Functional role of the replicated sex‐specific locus TMEM64 / NECAB1

The functional effects of directly replicated variants were assessed in SUGAR‐MGH participants receiving a single glipizide dose. Of the two TMEM64/NECAB1 variants that replicated in PMET1, only rs6471250‐C showed functional evidence in SUGAR‐MGH. In the combined‐sex SUGAR‐MGH analysis, rs6471250‐C carriers demonstrated significantly lower peak insulin (β = −0.97, 95% CI: −1.58 to −0.36, P = 0.0019); however, the genotype:sex interaction term was associated with increased peak insulin (β = 0.58, 95% CI: 0.18–0.98, P = 0.0041), indicating sex‐specific effects (Table 4 ). Sex‐stratified analyses revealed that rs6471250‐C was significantly associated with lower peak insulin in males (β = −0.33, 95% CI: −0.64 to −0.02, P = 0.035) but not females (β = 0.15, 95% CI: −0.20 to 0.50, P = 0.40), demonstrating male‐specific functional effects (Figure 3 , Table S11 ). While not statistically significant, rs6471250‐C showed a trend toward increased glucose trough (indicating delayed glucose recovery) in the combined‐sex (β = 0.65, 95% CI: −0.07 to 1.37, P = 0.078) and male‐only (β = 0.32, 95% CI: −0.05 to 0.69, P = 0.091) analyses, consistent with reduced sulfonylurea responsiveness in males. In contrast, no significant associations between rs6471250‐C and either peak insulin or glucose trough were detected in the female‐only stratum, confirming the male‐specific nature of this variant's functional effect.

Table 4.

Validation of Replicated Genetic Associations in SUGAR‐MGH

Gene	rsID	EA	OA	Term	Log peak insulin				Glucose trough
					EAF	β (95% CI)	P	N	EAF	β (95% CI)	P	N
TMEM64/NECAB1	rs6471250	C	T	Genotype	0.04	−0.97 (−1.58, −0.36)	0.0019	531	0.05	0.65 (−0.07, 1.37)	0.078	550
				Sex		−0.11 (−0.23, 0.02)	0.10			−0.03 (−0.19, 0.12)	0.66
				Genotype:Sex		0.58 (0.18, 0.98)	0.0041			−0.36 (−0.82, 0.09)	0.12

Bolded values reached Bonferroni‐corrected significance. EA, effect allele; EAF, effect allele frequency; OA, other allele.

Figure 3.

Validation of replicated TMEM64/NECAB1 locus in SUGAR‐MGH. Panels (a) present a regional zoom plot of the TMEM64/NECAB1 locus stratified by sex (males in orange, females in blue). Panels (b) and (c) display mean log insulin (μIU/mL) following glipizide dose by rs6471250 genotype (CT/CC Effect Allele) in males and females, respectively.

CONCLUSIONS

Although approximately 35% of individuals with type 2 diabetes are prescribed sulfonylureas, few studies have systematically assessed the genetic contributions to sulfonylurea response. ⁸ , ⁹ This first sex‐stratified GWAS of sulfonylurea response presents genetic heritability estimates in a multi‐ancestry RCT cohort and identifies novel sex‐specific loci associated with sulfonylurea response.

Sex‐specific genetic effects were identified in the male‐only ACCORD analysis. Variants in the TMEM64/NECAB1 locus showed significantly reduced sulfonylurea effectiveness in men when evidence was combined across ACCORD and PMET1. The rs6471250‐C variant, which is predominantly found in individuals of African genetic ancestry (population allele frequency (rs6471250‐C: 25.3%)), was significantly associated with lower peak insulin following a single glipizide dose in SUGAR‐MGH. The HbA1c response to sulfonylureas in men is estimated to decrease by 0.72% (95% CI: 0.42–1.02) per allele copy. While not statistically significant, rs6471250‐C showed a trend toward increased glucose trough (β = 0.32, 95% CI: −0.05 to 0.69, P = 0.091), supporting reduced insulin secretion. Given that the HbA1c‐lowering efficacy of sulfonylureas ranges from 1.0 to 1.5%, the 0.72% effect represents a substantial proportion of the expected therapeutic response for individuals who may not respond optimally.

Based on the literature, we propose that the TMEM64/NECAB1 locus influences sulfonylurea response through regulation of intracellular calcium homeostasis. In mice, Tmem64 interacts with sarcoplasmic/endoplasmic reticulum calcium ATPase 2 (Serca2/Atp2a2) and modulates calcium activity. ³³ Additionally, Necab1, which encodes a calcium‐binding protein, functions as a negative regulator of insulin secretion in pancreatic β‐cells; changes in its expression modulate calcium signaling in these cells. ³² These findings suggest that sex‐specific genetic variation contributes to the documented sex differences in HbA1c response to sulfonylureas. Furthermore, sulfonylureas inhibit ATP‐dependent potassium channels (via SUR1 and SUR2 receptors on pancreatic β‐cells), resulting in calcium influx. We identified significant and suggestive loci involved in calcium and potassium regulation in the male‐only stratum, including CRACR2A, which encodes a specialized plasma membrane calcium channel. When cellular calcium is depleted, CRACR2A is activated to restore calcium in the endoplasmic reticulum. ³⁴

Genetic variation affecting insulin regulation, secretion, or sensitivity may additionally influence sulfonylurea effectiveness. Variants in ADGRB3 and RAB28 were identified in the combined‐sex stratum and supported by external replication. ADGRB3 encodes an angiogenesis inhibitor receptor expressed on pancreatic β‐cells that binds complement 1q‐like‐3 (C1QL3) in a calcium‐dependent manner. Prior studies demonstrated that siRNA‐mediated knockdown of ADGRB3 increased insulin secretion through cAMP signaling. ³⁰ This suggests that elevated ADGRB3 expression suppresses insulin secretion, which may compromise sulfonylurea effectiveness. From the male‐only stratum, the RAB28 locus showed support for increased sulfonylurea response in the PMET1 replication cohort. RAB28 encodes a small GTPase expressed in adipose and skeletal muscle cells, and its GTP‐binding state is acutely regulated by insulin. In mice, siRNA‐mediated knockdown of Rab28 decreased basal glucose uptake, while overexpression of Rab28 increased cell‐surface levels of the glucose transporter GLUT4. ³¹

Genetic variation associated with sulfonylurea response was identified in the combined‐sex stratum. Specifically, SPINK5 was identified, which encodes a serine protease inhibitor. Genetic variation in SPINK5 was associated with improved glycemic response to sulfonylureas and showed colocalization with eQTL in skeletal muscle tissue. Sulfonylureas may enhance skeletal muscle glucose uptake by potentiating insulin effects, suggesting a mechanistic link between sulfonylurea action and skeletal muscle insulin signaling. ³⁵

Genetic variation associated with sulfonylurea response was found to have pleiotropic effects on retinal phenotypes. The TENM2, SCARB1, and VSX2 loci may affect vision through microvascular abnormalities, contributing to both acute vision impairment from hyperglycemia and chronic vision loss from diabetes. ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ SCARB1 and VSX2 showed replication support in PMET1.

The substantial difference in heritability between our results (h ² = 0.69 ± 0.19) and previous GWAS studies (h ² = 0.37 ± 0.11) may reflect several methodological and biological factors. First, comprehensive covariate adjustment, including baseline HbA1c, cardiovascular disease history, and concomitant medications, reduced environmental variance and improved detection of genetic effects. Second, the RCT design ensured standardized dosing and rigorously documented adherence, minimizing measurement error that typically attenuates heritability estimates in observational studies. Third, pharmacogenomic traits often demonstrate larger individual genetic effect sizes than complex disease traits, likely reflecting the more direct relationship between genotype and drug response and the more constrained biological mechanisms involved. Our sex‐stratified analyses revealed striking heterogeneity, with males demonstrating substantial heritability (h ² = 0.87 ± 0.22, P = 4.9 × 10⁻⁵) and females showing negligible heritability (h ² = 0.01 ± 0.44, P = 0.50). This supports true biological heterogeneity in sulfonylurea response; if our estimates were inflated by population structure or other confounding factors, similar inflation would be expected across sex strata. These patterns are consistent with sex‐specific heterogeneity in the genetic contribution to sulfonylurea response, but the female‐only estimate is imprecise and should not be interpreted as evidence that genetic contribution is absent in females.

The inclusion of Black and Hispanic participants influenced heritability estimates and discovery power, as demonstrated by observed ancestry‐specific allele frequency patterns. Many of the identified significant variants showed substantially higher allele frequencies in individuals of African genetic ancestry compared to other populations. For example, rs116498056 in TENM2 has a frequency of 10.4% in individuals of African genetic ancestry compared to 0.3% in European genetic ancestry populations. The replicated variant rs6471250 in TMEM64/NECAB1 similarly shows the highest frequency in individuals of African genetic ancestry. This pattern suggests that the inclusion of Black participants (14.8% of the cohort) was essential for detecting these associations. Ancestry‐specific heritability estimation was attempted; however, models did not converge due to limited sample sizes in ancestry‐specific subgroups.

The primary limitation of this study is the relatively modest sample size, which precluded ancestry‐specific analyses and likely accounts for the smaller number of significant loci identified in females (N = 351) compared to males (N = 520). To address this limitation, external replication was conducted to increase confidence in the identified loci. Loci identified in the RCT cohort successfully replicated in the EHR cohort, demonstrating the complementary value of combining both approaches: RCT data for discovery and EHR data for independent validation. Additionally, the absence of hormone data limits mechanistic interpretation of the sex‐specific analyses but does not undermine the sex‐stratified genetic analysis. Additionally, since pharmacokinetic measurements were not collected in ACCORD or PMET1, the HbA1c‐response analyses cannot distinguish absorption, distribution, metabolism, or elimination effects from drug‐target response. SUGAR‐MGH provided acute pharmacodynamic readouts after glipizide, but not a full pharmacokinetic profile. Finally, as in most GWAS studies, parent‐of‐origin effects could not be assessed because parental genotypes, family‐based transmission data, and parent‐of‐origin phasing were unavailable from the genotyping platforms used.

The prevalence of CVD in ACCORD (31.92%) is comparable to or lower than in general type 2 diabetes populations, where 32–44% have established CVD. ⁴² This suggests the findings are generalizable to individuals with type 2 diabetes receiving sulfonylureas. Successful replication of findings in independent EHR cohorts with broader population representation and no CVD risk selection criteria further supports their generalizability.

In summary, this first sex‐stratified GWAS of sulfonylurea response identified sex‐specific genetic associations supported by external replication and, for rs6471250‐C, functional follow‐up in SUGAR‐MGH. These findings provide biologically plausible leads for understanding sex‐specific sulfonylurea response and support further validation before clinical implementation. Specifically, genetic variation influencing sex‐specific sulfonylurea response could inform individualized dosing strategies to optimize the use of this affordable and widely used drug class in type 2 diabetes treatment. ⁴³

FUNDING

This research was supported by a National Heart, Lung, and Blood Institute grant to the University of North Carolina at Chapel Hill (R01 HL110380‐04), the University of Virginia (R01 HL110400), and the Joslin Diabetes Center. S.W.Y. and K.M.G. were supported by GM61390 and GM117163. J.H.L. is supported by NIDDK K23 DK131345. This research was funded in part by the Intramural Research Program of the National Institute of Environmental Health Sciences.

CONFLICTS OF INTEREST

D.M.R. has received research funding from Novo Nordisk. All other authors declared no competing interests for this work.

AUTHOR CONTRIBUTIONS

J.H.B., J.S.H., D.M.R., and A.A.M.‐R. wrote the manuscript; J.H.B., J.S.H., J.H.L., S.W.Y., D.M.R., and A.A.M.‐R. designed the research; J.H.B., J.S.H., M.K., S.N., and D.M.R. performed the research; and J.H.B., J.S.H., F.S.A., M.K., S.N., H.S.S., M.M.H., K.M.G., J.C.M., A.D., M.J.W., J.H.L., S.W.Y., J.B.B., R.P.W., D.M.R., and A.A.M.‐R. analyzed the data.

Supporting information

Tables S1‐S11

Figure S1