Increased Genetic Risk for β-Cell Failure Is Associated With β-Cell Function Decline in People With Prediabetes

Abstract

Partitioned polygenic scores (pPS) have been developed to capture pathophysiologic processes underlying type 2 diabetes (T2D). We investigated the association of T2D pPS with diabetes-related traits and T2D incidence in the Diabetes Prevention Program. We generated five T2D pPS (β-cell, proinsulin, liver/lipid, obesity, lipodystrophy) in 2,647 participants randomized to intensive lifestyle, metformin, or placebo arms. Associations were tested with general linear models and Cox regression with adjustment for age, sex, and principal components. Sensitivity analyses included adjustment for BMI. Higher β-cell pPS was associated with lower insulinogenic index and corrected insulin response at 1-year follow-up with adjustment for baseline measures (effect per pPS SD −0.04, P = 9.6 × 10⁻⁷, and −8.45 μU/mg, P = 5.6 × 10⁻⁶, respectively) and with increased diabetes incidence with adjustment for BMI at nominal significance (hazard ratio 1.10 per SD, P = 0.035). The liver/lipid pPS was associated with reduced 1-year baseline-adjusted triglyceride levels (effect per SD −4.37, P = 0.001). There was no significant interaction between T2D pPS and randomized groups. The remaining pPS were associated with baseline measures only. We conclude that despite interventions for diabetes prevention, participants with a high genetic burden of the β-cell cluster pPS had worsening in measures of β-cell function.

Article Highlights

• Process-specific genetic scores (partitioned polygenic scores [pPS]) help explain clinical heterogeneity in type 2 diabetes, but whether they inform differential response to diabetes prevention strategies is unknown.

• In the Diabetes Prevention Program clinical trial, we examined pPS for associations with longitudinal glycemic measures and type 2 diabetes development.

• Higher β-cell pPS was associated with reduced baseline and even further reduced 1-year β-cell function despite diabetes prevention interventions.

• Diabetes prevention interventions may not optimally prevent β-cell function decline in people with genetic susceptibility for β-cell dysfunction, which may contribute to the heterogeneity seen in progression of diabetes development.

Introduction

An estimated 38% of adults in the U.S. general population have prediabetes (1). The lifetime risk of progressing from prediabetes to diabetes is as high as 74% (2). For this reason, interventions for diabetes prevention are essential. The Diabetes Prevention Program (DPP) investigated interventions of metformin, lifestyle, and placebo to prevent progression to type 2 diabetes (T2D) in a U.S. study population at high risk for developing T2D (3). While metformin and lifestyle led to 31% and 58% reduced incidence of diabetes in the DPP, respectively, many participants did not respond adequately to the trial interventions and developed T2D.

Given the substantial variability in response to interventions aimed at diabetes prevention, further investigations have pursued whether genetic variation impacts changes in glycemia over time or the response to interventions. Through targeted genotyping of diabetes-related genes, the DPP investigators have previously shown that common genetic variants are associated with a variable response to metformin and lifestyle intervention for diabetes prevention (4–6). The DPP has also examined the influence of the genetic burden of genome-wide significant T2D single nucleotide polymorphisms (SNPs) in a polygenic score (N = 34 SNPs), which was associated with increased risk of progression to diabetes and resistance to metformin intervention at the highest score quartile (7). The DPP investigators further examined a polygenic score that represented underlying T2D biology with an insulin resistance score (N = 17 SNPs) (8); they found that participants with an increased insulin resistance score had lower insulin sensitivity over time but lifestyle and metformin interventions improved insulin sensitivity regardless of the score burden.

More recently, we have shown that T2D genetic variants can be more clearly delineated into putative physiological processes using cluster analyses whereby the variants are grouped based on shared associations with T2D-related traits, such as BMI and serum lipid levels (9,10). In these studies, a shared set of five distinct clusters of T2D genetic variants was identified, including two related to pancreatic β-cell function (differing in associations with high vs. low fasting proinsulin) and three related to mechanisms of insulin resistance (obesity, fat distribution, and liver/lipid metabolism). Genetic clusters can be used to generate process-specific (partitioned) polygenic scores (pPS), which have been shown across multiple study populations to be associated with distinct clinical features and risks of other metabolic conditions (9,11).

We hypothesized that the differential response to diabetes prevention interventions in the DPP could be explained in part by underlying genetics processes contributing to T2D development, as captured by T2D cluster pPS (9). Additionally, because prior analyses of the T2D cluster pPS involved cross-sectional data sets, it is not yet known whether the T2D pPS are associated longitudinally with changes in intermediate phenotypes related to diabetes progression. Therefore, we aimed to leverage the prospective DPP study to investigate whether genetic T2D pPS 1) influence change in β-cell function and insulin resistance parameters over 1 year in all participants and 2) modify the impact of the diabetes prevention interventions on these parameters.

Research Design and Methods

Description of DPP Study Design and Participants

The DPP was a U.S.-based multicenter randomized controlled trial that enrolled participants at high risk of developing diabetes from 27 clinical sites to test the effects of intensive lifestyle and metformin interventions on the incidence of diabetes, as previously described (12,13). Briefly, 3,234 participants with overweight or obesity, ages ≥25 years, and with fasting plasma glucose level between 95 and 125 mg/dL (5.3 and 6.9 mmol/L) and 2-h plasma glucose level between 140 and 199 mg/dL (7.8 and 11.0 mmol/L) during a standard 75-g oral glucose tolerance test (OGTT) were randomized to intensive lifestyle intervention (n = 1,079), metformin (850 mg twice daily [n = 1,073]), or placebo (n = 1,082).

The primary end point in DPP was the development of diabetes as defined according to the American Diabetes Association guidelines of fasting glucose level ≥126 mg/dL from semiannual tests or 2-h glucose level ≥200 mg/dL during yearly 75-g OGTT, with confirmation of either on a second occasion within 6 weeks (3).

This analysis includes 2,647 participants from DPP with genotyping and baseline and 1-year measurements. Each clinical center and the coordinating center obtained institutional review board approval. All participants included in these analyses provided written informed consent for the main study and subsequent genetic investigations.

Quantitative Glycemic Physiologic Traits

The methods for measuring HbA_1c, glucose, insulin, proinsulin, and triglyceride levels have previously been described (13). The insulin sensitivity index (ISI) was calculated as the reciprocal of HOMA of insulin resistance with use of the equation [(fasting insulin (mU/L) × fasting glucose (mmol/L)) / 22.5] (14) based on the fasting glucose and insulin levels during the OGTT. Insulin secretory response was estimated with two methods (15,16): 1) the corrected insulin response (CIR) = (100 × 30-min insulin) / (30-min glucose × [30-min glucose − 70 mg/dL]) and 2) the insulinogenic index (IGR) = (30-min insulin − fasting insulin) / (30-min glucose − fasting glucose [units μU/mg]). We calculated the proinsulin-to-insulin ratio (PIR) by dividing fasting proinsulin level by fasting insulin level. Participants were asked not to take metformin or placebo on the morning of the OGTT. For longitudinal analyses, the 1-year end point was chosen because the sample size was largest at that time point (95% of participants completed the 1-year follow-up visit) and weight loss was the most pronounced at 1 year in the intervention arms.

Genotyping

DNA was extracted from peripheral blood leukocytes, and genotyping was done with the HumanCoreExome genotyping array from Illumina at the Genomics Platform at the Broad Institute. Genotypes were called with Birdsuite (https://www.broadinstitute.org/birdsuite/birdsuite-analysis). A two-stage procedure consisting of prephasing the genotypes into whole chromosome haplotypes followed by imputation was conducted. The prephasing was performed with SHAPEIT2 (17). We used 1000 Genomes Phase 3 haplotypes as reference panel (18), and the genotype imputation for 9 million SNPs was done with IMPUTE2 (19).

SNP Selection and Construction of Genetic Risk Score

Five pPS corresponding to each of the T2D genetic clusters (β-cell, proinsulin, obesity, lipodystrophy, and liver/lipid) identified by Udler et al. (9) were computed, as previously described. Briefly, in our prior work we performed soft cluster analysis of T2D variant-trait associations, which generated cluster-specific weights for the T2D SNPs. We used the top-weighted SNPs (as defined in 9) and their cluster weights to generate the T2D pPS. Each pPS accounted for the number of risk alleles present for each SNP in the cluster, multiplied by each SNP’s cluster weight, and then summed over all the SNPs in the cluster. Since six SNPs had A/T or C/G alleles, which present a potential source of error in determining the T2D-risk allele, for these SNPs we chose a proxy in high linkage disequilibrium (r² > 0.8) to avoid error in pPS generation (Supplementary Table 1). Furthermore, we also calculated a T2D global extended polygenic score for participants (methods described in 20) and examined it for associations with glycemic traits (Supplementary Table 3).

Statistical Analyses

Continuous baseline characteristics were reported as mean ± SD if normally distributed or as median with 25th and 75th percentiles if not. Categorical variables were presented as frequency (percent). Baseline measures were examined to validate the previous findings of Udler et al. (9) with these traits and pPS.

The primary outcomes were changes at 1 year in measures that relate to metabolic function: ISI, CIR, IGR, fasting insulin, PIR, HbA_1c, BMI, waist circumference, and triglyceride levels. To assess the change in these outcomes at year 1, we modeled the outcome value at year 1 adjusted for the respective trait measurement at baseline.

Generalized linear models were used to estimate the association of each pPS with the above measures at baseline and year 1 measures adjusted for baseline. Models included adjustment for age at randomization, sex, and the top 10 principal components for genetic ancestry. The natural log or the original scale was used in baseline and year 1 values (as noted in Table 2 footnotes). Model fit was assessed with residual analysis, including testing for normally distributed residuals and outliers. Residual plots were used to confirm that the assumption of homogeneity of variance was met.

The association between pPS and diabetes incidence over the course of the main trial (mean follow-up time of 3.2 years) was tested with Cox proportional hazards models. We confirmed proportional hazards and used interval censoring with the Breslow method to account for ties.

Analyses of the year 1 measures adjusted for baseline measures as well as diabetes incidence were stratified by treatment group and included a test for interaction between pPS and treatment group. When the interaction between pPS and treatment group was nonsignificant, an analysis including all participants was performed with adjustment for treatment arm.

Because the pPS were initially generated from a primarily European ancestry population, in a sensitivity analysis we focused on the participants of self-reported White race to see whether these analyses were consistent with those from the primary analysis. We also performed sensitivity analyses including adjustment for BMI.

Effect estimates per sample SD, 95% CIs, and P values are reported for each pPS. All tests performed were two sided, with a Bonferroni α-level of 0.001 (0.05 / (10 × 5)) used to determine statistical significance across the 10 novel outcomes assessed (changes over 1 year for nine continuous measures and incidence of T2D). SAS, version 9.4, was used for all analyses (SAS Institute, Cary, NC).

Data and Resource Availability

In accordance with the National Institutes of Health (NIH) Public Access Policy, we continue to provide all manuscripts to PubMed Central including this article. For DPP/Diabetes Prevention Program Outcomes Study (DPPOS) the protocols and lifestyle and medication intervention manuals have been provided to the public through the public website www.dppos.org. The DPPOS abides by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) data sharing policy and implementation guidance as required by the NIDDK, NIH (https://repository.niddk.nih.gov/studies/dppos/).

Results

T2D pPS Are Associated With Relevant Clinical Characteristics at Baseline

Baseline DPP participant characteristics were similar across the treatment groups for demographics and anthropometric measurements, including sex, age, self-reported race, BMI, and waist circumference (Table 1). We generated the five T2D pPS in participants (Supplementary Table 1) and noted <5% missingness in genotyping related to all scores. Additionally, median values and variances for the pPS and glycemic parameters were similar across the treatment groups at baseline (Table 1).

Table 1.

Baseline characteristics

	Total	Placebo	Metformin	Lifestyle
n	2,647	887	875	885
Male, n (%)	864 (32.6)	281 (31.7)	294 (33.6)	289 (32.7)
Female, n (%)	1,783 (67.4)	606 (68.3)	581 (66.4)	596 (67.3)
Self-reported race, n (%)
White	1,468 (55.5)	490 (55.2)	503 (57.5)	475 (53.7)
Black	537 (20.3)	186 (21.0)	178 (20.3)	173 (19.5)
Hispanic	450 (17.0)	147 (16.6)	143 (16.3)	160 (18.1)
Asian/Pacific Islander	117 (4.4)	37 (4.2)	31 (3.5)	49 (5.5)
American Indian	75 (2.8)	27 (3.0)	20 (2.3)	28 (3.2)
Age, years	50.7 ± 10.7	50.6 ± 10.4	50.9 ± 10.3	50.6 ± 11.4
BMI, kg/m2	34.1 ± 6.6	34.3 ± 6.7	34.0 ± 6.6	34.0 ± 6.6
Waist, cm	105.3 ± 14.5	105.5 ± 14.3	105.0 ± 14.4	105.3 ± 14.8
ISI	0.16 (0.11, 0.24)	0.16 (0.11, 0.24)	0.16 (0.11, 0.24)	0.16 (0.11, 0.24)
IGR (μU/mg)	104.6 (68.2, 157.6)	105.2 (66.7, 160.8)	102.3 (68.2, 152.2)	106.7 (70.2, 158.2)
Triglycerides, mg/dL	143 (99, 204)	148 (104, 210)	139 (98, 199)	139 (96, 200)
HbA1c, %	5.9 ± 0.5	5.9 ± 0.5	5.9 ± 0.5	5.9 ± 0.5
HbA1c, mmol/mol	41 ± 1.5	41 ± 1.5	41 ± 1.5	41 ± 1.5
Fasting insulin, mIU/L	24 (16, 34)	24 (16, 33)	24 (16, 34)	24 (16, 34)
CIR	0.54 (0.35, 0.79)	0.54 (0.35, 0.81)	0.53 (0.35, 0.79)	0.54 (0.37, 0.77)
PIR	0.17 (0.13, 0.23)	0.17 (0.13, 0.24)	0.17 (0.13, 0.23)	0.18 (0.13, 0.24)
T2D pPS (50% median)
β-Cell	47.2 ± 6.0	47.2 ± 5.9	47.3 ± 6.0	47.2 ± 6.1
Proinsulin	12.0 ± 2.4	12.0 ± 2.3	11.9 ± 2.4	12.0 ± 2.5
Obesity	8.8 ± 4.2	8.8 ± 4.2	8.7 ± 4.2	8.9 ± 4.2
Lipodystrophy	34.7 ± 4.4	34.5 ± 4.4	34.7 ± 4.3	34.8 ± 4.5
Liver/lipid	9.7 ± 3.8	9.8 ± 3.8	9.8 ± 3.8	9.5 ± 3.9

Data are n (%) for categorical variables and mean ± SD for continuous and median (25th, 75th percentile) for skewed distributions.

We initially assessed whether the five T2D pPS were associated with baseline participant characteristics that would be expected from previously reported findings (9). Indeed, we replicated prior key cluster pPS associations with baseline clinical phenotypes (at P < 0.05) (Table 2 and Supplementary Table 2): a higher β-cell cluster pPS was associated with baseline clinical phenotypes (at P < 0.05) (Supplementary Table 2); a higher β-cell cluster pPS was associated with lower measures of insulin secretory response (CIR and IGR) and fasting insulin and higher PIR; a higher proinsulin pPS was associated with lower fasting insulin and PIR (reaching significance in the sensitivity analysis examining only individuals of White race); a higher lipodystrophy pPS was associated with lower ISI, BMI, and HDL and higher triglycerides; a higher obesity pPS was associated with higher BMI; and a higher liver/lipid pPS was associated with lower triglycerides. We did not replicate associations between proinsulin pPS and waist circumference, obesity pPS and fasting insulin, and liver/lipid pPS and fasting insulin.

Table 2.

Associations of T2D pPS with baseline and 1-year baseline-adjusted values

Outcome (N = 2,647)	β-Cell pPS		Proinsulin pPS		Obesity pPS		Lipodystrophy pPS		Liver/lipid pPS
	Estimate	P	Estimate	P	Estimate	P	Estimate	P	Estimate	P
BMI, kg/m2
Baseline	−0.6496	1.7 × 10−7	−0.0301	0.814	0.2678	0.032	−0.5483	1.3 × 10−5	0.0061	0.961
1-year adj baseline	0.0519	0.21	−0.0574	0.180	−0.0303	0.470	0.0500	0.230	−0.0204	0.620
Waist, cm
Baseline	−1.2290	1.1 × 10−5	−0.0299	0.918	0.6096	0.030	−1.4584	2.5 × 10−7	0.1025	0.715
1-year adj baseline	0.1025	0.43	0.0076	0.960	−0.0568	0.660	0.2432	0.070	−0.2308	0.080
Triglycerides, mg/dL
Baseline	−2.8280	0.128	0.3748	0.845	0.8656	0.641	4.0744	0.030	−9.3715	3.9 × 10−7
1-year adj baseline	1.5997	0.23	0.1291	0.930	1.6676	0.210	1.5290	0.260	−4.3715	1.0 × 10−3
HbA1c
Baseline, %	−0.0006	0.943	0.0160	0.104	−0.0067	0.491	0.0075	0.430	0.0058	0.530
Baseline, mmol/mol	−23.504		−23.3220		−23.571		−23.4150		−23.434
1-year adj baseline, %	0.0066	0.33	0.0000	0.990	0.0055	0.410	0.0026	0.680	−0.0108	0.100
1-year adj baseline, mmol/mol	−23.425		−23.4970		−23.437		−23.4690		−23.615
Fasting insulin, μU/mL
Baseline	−1.2027	4.6 × 10−5	−0.6455	0.034	−0.4197	0.157	0.4934	0.100	0.2062	0.486
1-year adj baseline	−0.3308	0.23	0.1283	0.670	0.7801	0.010	0.0399	0.890	0.1962	0.490
PIR
Baseline	0.0101	8.9 × 10−5	−0.0045	0.078	−0.0051	0.048	0.0000	0.972	−0.0015	0.587
1-year adj baseline	0.0042	0.14	0.0017	0.560	−0.0029	0.326	0.0013	0.690	−0.0038	0.190
ISI
Baseline	0.0054	0.04	0.0050	0.061	0.0025	0.295	−0.0070	0.007	−0.0019	0.498
1-year adj baseline	0.0048	0.10	−0.0002	0.930	−0.0004	0.850	0.0009	0.800	−0.0035	0.190
IGR (μU/mg)
Baseline	−9.1105	6.4 × 10−7	−2.2354	0.244	−3.1251	0.090	2.9536	0.113	1.5114	0.411
1-year adj baseline	−8.4978	5.6 × 10−6	−2.7241	0.160	0.2252	0.910	−0.0259	0.100	0.1498	0.940
CIR
Baseline	−0.0465	1.5 × 10−8	−0.0155	0.073	−0.0114	0.178	0.0149	0.072	−0.0008	0.913
1-year adj baseline	−0.0417	9.6 × 10−7	−0.0108	0.230	0.0042	0.630	−0.0035	0.700	0.0027	0.760

Estimates are reported per SD. The outcome value at year 1 includes adjustment for the respective trait measurement at baseline (1-year adj baseline).

P values in boldface type indicate significant replication (P < 0.05) for associations with baseline measures and study-wide significant (P < 0.001) for 1 year adjusted for baseline. For baseline, results are reported in natural log transformation scale. For all 1-year baseline-adjusted results, the original scale was used for point estimates (no log transformation).

β-Cell pPS Is Associated With Longitudinal Decline in β-Cell Function

We next assessed whether the pPS were associated with changes in metabolic traits over 1 year. Higher β-cell cluster pPS was associated with worsening β-cell function over time, with reduced CIR and IGR 1-year values adjusted for baseline measurement and treatment group (CIR effect per SD −0.042 [95% CI −0.058, −0.025], P = 9.6 × 10⁻⁷; IGR −8.498 μU/mg [−12.166, −4.830], P = 5.6 × 10⁻⁶) (Fig. 1, Table 2, and Supplementary Tables 3, 4, and 5). The 1-year baseline-adjusted values of both CIR and IGR showed consistent directional associations in all treatment arms, and there was no evidence of an interaction with treatment group. Higher liver/lipid pPS was significantly associated with lower 1-year baseline-adjusted serum triglyceride values (effect per SD −4.37 [−6.98, −1.76], P = 1.0 × 10⁻³). Again directional associations with treatment arm were consistent, and there was no evidence of interaction with treatment arm (Table 2 and Supplementary Tables 3, 4, and 5). No other cluster pPS was significantly associated with changes in traits over 1 year.

Figure 1.

Greater decline in insulin secretory response over 1 year in the highest quartile of β-cell cluster pPS. In participants in the highest quartile of β-cell cluster pPS compared with those in the lowest quartile, the adjusted means of measures of β-cell secretory response estimated with IGR (A) and CIR (B) both significantly decline from baseline to 1 year (P < 0.001). IGR (μU/mg) = (insulin_{30 min} − insulin_{0 min}) / (glucose_{30 min} − glucose_{0 min}). CIR = (100 × insulin_{30 min}) / [glucose_30-min × (glucose_30-min − 70 mg/dL)].

In considering diabetes incidence, none of the cluster pPS reached study-wide significance within treatment arms for risk of disease onset and no interactions with treatment group were identified (Supplementary Table 6). The β-cell cluster pPS was nominally significantly associated with diabetes incidence in the BMI-adjusted model (hazard ratio 1.10 per weighted allele [95% CI 1.00, 1.20], P = 0.035).

Discussion

Given the rapidly rising rates of T2D worldwide, it is of great interest to determine whether stratification can be done for people at risk for T2D to help optimize strategies for diabetes prevention. These investigations are important and timely in the setting of the recently published second international consensus report of precision diabetes medicine, where the authors highlight and address the challenges posed by heterogeneity in pathogenesis and clinical course in diabetes (21). Using data from a randomized controlled trial of prevention strategies for diabetes, we have examined five different genetic pPS intended to capture T2D risk related to pancreatic β-cell function (β-cell, proinsulin) and insulin resistance (liver/lipid, obesity, lipodystrophy). We have assessed associations of the pPS with changes in diabetes-related traits over time as well as diabetes incidence and whether these differ based on diabetes prevention interventions. The findings of our study support prior findings in cross-sectional studies of association of the pPS and glycemic and metabolic traits. In addition to replicating these associations (Table 2 and Supplementary Table 2), for the first time, we identify two significant 1-year longitudinal associations with the T2D pPS: 1) increased β-cell cluster pPS with reduced CIR and IGR and 2) increased liver/lipid pPS with reduced serum triglyceride levels.

Our key findings show that high genetic burden of the β-cell cluster pPS was associated with lower baseline insulin secretory response and decline in insulin secretory response over 1 year, with no detected difference in these findings across the DPP treatment groups. We have thus demonstrated longitudinally that genetic clusters defining underlying pathophysiology for diabetes development can identify a subset of people at risk for diabetes who also are at increased risk of decline in β-cell function, which is unfortunately not mitigated by metformin or lifestyle interventions enough to be detectable in our study. We also observed a nominally significant association with the β-cell cluster pPS and increased incidence of diabetes. This finding was nominally significant at P < 0.05 but did not meet our study-wide significance threshold. We have high suspicion that it is a true association because all the individual alleles in the pPS are associated with T2D risk (by design) and the β-cell cluster pPS has been shown in prior work to have the strongest association with prevalent T2D of any cluster (11).

The associations of high genetic burden of β-cell cluster pPS with reduced baseline insulin secretory response, 1-year decline in insulin secretory response, and a nominally greater risk of incident diabetes exemplify the role of β-cell dysfunction in the progression from prediabetes to diabetes. Indeed, higher diabetes incidence in all treatment groups was found among individuals with lower measures of insulin secretory response at baseline in the DPP (22). It has been hypothesized that environmental factors causing obesity and insulin resistance may also contribute to decreased insulin secretion due to β-cell dysfunction in the pathophysiology of T2D development (23). Our study demonstrates that genetic factors alone can predict decline in β-cell insulin secretory response with no detectable abrogation by the DPP diabetes interventions. Our β-cell pPS included a subset of T2D-associated SNPs predicted using an unbiased clustering approach to share a disease mechanism. In contrast, we did not find associations for the insulin resistance clusters (obesity, lipodystrophy, and liver/lipid) with estimates of baseline or longitudinal β-cell function.

The second of the significant longitudinal findings was that increased genetic burden in the liver/lipid cluster was associated with decline in triglyceride levels over 1 year. The liver/lipid cluster includes loci GCKR, PNPLA3, and TM6SF2, which have been shown through functional experiments to relate to liver lipid metabolism (24–31). The alleles at these loci that increase T2D risk are interestingly associated with reduced serum triglyceride levels, therefore representing a process distinct from traditional triglyceride elevation seen with metabolic syndrome; this underlying mechanism may be due to deregulated triglyceride processing in the liver leading to a liver-mediated form of insulin resistance (24,30,31). We previously showed using epigenomic data that this genetic cluster is enriched for enhancers/promoters active in liver tissue (9). The present findings suggest that regardless of treatment arm, individuals with high genetic burden for liver/lipid cluster had a decline in triglyceride levels over time, suggesting worsening of dysfunctional lipid metabolism. While we were unable to observe a significant association with change in insulin sensitivity over 1 year, we did see trends toward reduced insulin sensitivity (increased insulin resistance) based on higher 1-year fasting insulin levels and lower ISI adjusted for baseline (Table 2), supporting the idea that the increased dysregulation of triglyceride levels may correlate with worsening insulin resistance. Failure to reach significant associations with fasting insulin and ISI may have been related to the study ascertainment of individuals with elevated BMI leading to greater difficulty in discriminating BMI-independent effects on insulin levels. Further research is needed to clarify the underlying mechanisms connecting the liver/lipid pPS and lower triglyceride levels, which may also be independent of insulin action.

Our findings build on previous results in the DPP from examining the influence of genetic burden as measured according to polygenic risk scores, glycemic parameters, and diabetes incidence. Hivert et al. (7) developed a 34-SNP T2D polygenic score. Similar to our β-cell cluster pPS findings, the authors determined that DPP participants in the highest T2D polygenic score quartile had reduced β-cell secretory response (IGR) at baseline compared with those in the lowest quartile; however, the 1-year longitudinal change in IGR was not significantly different between the two groups. In contrast, we found a significant association between β-cell–specific pPS and insulin secretory responses (IGR and CIR) over the first year. We also assessed an updated T2D polygenic score (including >1 million variants) that was associated with reduced IGR and CIR at 1 year adjusted for baseline values. Interestingly, both associations were less significant than those for β-cell–specific pPS, which contained fewer SNPs (Supplementary Table 3). While the full polygenic scores for T2D included variants influencing many T2D pathophysiological pathways, the β-cell pPS included a specific subset of 30 T2D-associated SNPs specifically related to β-cell function (including HNF1A, HNF4A, TCF7L2, and SLC30A8) (9). Our findings support the use of process-specific polygenic risk scores to more precisely define the underlying pathophysiology that contributes to progression from prediabetes to diabetes.

Metformin treatment, thiazolidinediones, and interventions that promote weight loss have been found to reduce the progression from prediabetes to diabetes. In the current study, we showed that despite metformin and lifestyle interventions, the DPP participants with higher β-cell cluster pPS had a greater decline in insulin secretory response during the first year, indicating that these interventions could not substantially prevent β-cell deterioration. Therefore, alternative diabetes prevention strategies to preserve β-cell function and maintain or improve insulin secretory response may be key for preventing diabetes in people at higher risk of β-cell dysfunction in this high β-cell pPS cluster.

Strengths of this study include the randomized design, which allowed us to characterize how these genetic variants might influence response to diabetes preventive interventions. To the best of our knowledge, this is the first study examining T2D pPS in the longitudinal setting of a randomized clinical trial and looking at treatment response. Limitations of the study include limited sample size from an existing clinical study impacting power. Because the DPP population had a much narrower weight range based on recruitment inclusion, our ability to replicate some baseline associations noted in the results may be further reduced by power. Another limitation is the development of the genetic clusters in studies of predominantly European genetic ancestry; additional efforts are underway to develop genetic clusters using studies from more diverse ancestral groups. Additionally, while significant associations were identified, the effect sizes were small, suggesting that stratification with cluster pPS would not be clinically useful at this point in time but potentially could be in the future with development of more robust pPS.

In summary, by studying individuals at risk for diabetes in a randomized control trial, we show that high genetic burden in β-cell cluster pPS was associated with a 1-year decline in insulin secretory response, regardless of intervention allocation. Commonly used diabetes prevention interventions, such as lifestyle intervention and metformin therapy, may not optimally preserve β-cell function in people with genetic susceptibility for β-cell dysfunction, and alternative therapies may be necessary to prevent progression to diabetes. Therefore, further studies are needed to find more precise and effective therapies for diabetes prevention given the heterogeneous pathophysiology contributing to diabetes development.

This article contains supplementary material online at https://doi.org/10.2337/figshare.25833832.

Funding Statement

L.K.B. is supported by the NorthShore Auxiliary Scholars Award from NorthShore University HealthSystem. M.S.U. was supported by NIDDK, NIH, grant K23DK114551. J.M. was partially supported by the American Diabetes Association (7-21-JDFM-005), the Nutrition Obesity Research Center at Harvard (P30 DK040561), and NIH (UG1 HD107691). S.R. is supported by a Webb-Waring Biomedical Research Award from the Boettcher Foundation and by U.S. Department of Veterans Affairs award IK2-CX001907. J.M.M. is supported by American Diabetes Association grant 11-22-ICTSPM-16 and National Human Genome Research Institute grant U01HG011723. A.H.-C. is supported by American Diabetes Association grant 11-23-PDF-35. J.C.F. is supported by National Heart, Lung, and Blood Institute, NIH, grant K24 HL157960. Genotyping in the DPP was supported by NIDDK, NIH, grant R01 DK072041 to J.C.F. Research reported in this publication was supported by the NIDDK, NIH, under award nos. U01 DK048489, U01 DK048339, U01 DK048377, U01 DK048349, U01 DK048381, U01 DK048468, U01 DK048434, U01 DK048485, U01 DK048375, U01 DK048514, U01 DK048437, U01 DK048413, U01 DK048411, U01 DK048406, U01 DK048380, U01 DK048397, U01 DK048412, U01 DK048404, U01 DK048387, U01 DK048407, U01 DK048443, and U01 DK048400, through provision of funding during DPP and DPPOS to the clinical centers and the Coordinating Center for the design and conduct of the study and collection, management, analysis, and interpretation of data. Funding was also provided by the Eunice Kennedy Shriver National Institute of Child Health and Human Development; the National Institute on Aging; the National Eye Institute; the National Heart, Lung, and Blood Institute; the National Cancer Institute; the Office of Research on Women’s Health; the National Institute on Minority Health and Health Disparities; the Centers for Disease Control and Prevention; and the American Diabetes Association. Genetic analyses in the DPP were funded in part through a grant to P.W.F. from the Novo Nordisk Foundation. The Southwestern American Indian Centers were supported directly by the NIDDK, including its Intramural Research Program, and the Indian Health Service. The General Clinical Research Center program, National Center for Research Resources, and Department of Veterans Affairs supported data collection at many of the clinical centers. Merck KGaA provided medication for DPPOS. DPP and DPPOS have also received donated materials, equipment, or medicines for concomitant conditions from Bristol-Myers Squibb, Parke-Davis, LifeScan, Health o meter, Hoechst Marion Roussel, Merck-Medco Managed Care, Merck and Co., Nike Sports Marketing, Slim Fast Foods, and The Quaker Oats Company. The Henry M. Jackson Foundation provided support services under subcontract with the Coordinating Center. The sponsor of this study was represented on the Steering Committee and played a part in study design, how the study was done, and publication.