Global and Partitioned Polygenic Risk Scores: Associations With Pathophysiology and Incidence of Type 2 Diabetes in People With Prediabetes

Abstract

Type 2 diabetes partitioned polygenic risk scores (pPRS) are related to their physiological impact on insulin secretion and sensitivity. We investigated associations of two pPRS sets obtained from multiancestry cohorts with the relationship between insulin secretion and sensitivity and diabetes incidence in the Diabetes Prevention Program (DPP). We generated 12 and 8 pPRS from soft- and hard-clustering approaches, respectively, in 2,052 DPP participants randomized to intensive lifestyle intervention, metformin, or placebo. Baseline insulin secretion demand and compensation were estimated through the relationship between secretion and sensitivity. Most expected associations of pPRS with insulin secretion and sensitivity were replicated. Some pPRS associated with lower secretion compensation. Global PRS associations with diabetes incidence were stronger than those of any pPRS in both sets. When insulin secretion compensation and demand modified pPRS associations with diabetes incidence, modification occurred only for demand: higher compensation associated with decreased diabetes incidence regardless of pPRS level. A β-cell dysfunction pPRS interacted with DPP treatments in a way that suggested that these treatments may be less effective in those genetically predisposed to diabetes due to insulin deficiency. Regardless of genetic risk as measured by pPRS, inadequate compensatory insulin secretion contributes to progression to diabetes in people with prediabetes.

Article Highlights

• The extent to which partitioned polygenic risk scores (pPRS) for type 2 diabetes influence the relationship between insulin secretion and sensitivity has not been examined in people with prediabetes.

• We investigated whether insulin secretion compensation and demand, estimated through the relationship between insulin secretion and sensitivity, mediated or modified the associations of two sets of pPRS with diabetes incidence in Diabetes Prevention Program participants.

• Several pPRS modified the effects of insulin demand on diabetes incidence, but none modified the effect of compensation.

• In people with prediabetes, inadequate compensatory insulin secretion contributes to progression to diabetes regardless of genetic risk as measured by pPRS.

Graphical Abstract

Introduction

Type 2 diabetes global polygenic risk score (PRS) and partitioned PRS (pPRS) studies help us understand diabetes development (1–4). pPRS are created by clustering diabetes genetic variants based on associations with diabetes-related traits (1,4,5), including decreased insulin sensitivity or secretion (4,6), the main physiological abnormalities leading to type 2 diabetes (1,4,7–9).

β-Cell function, estimated by a curve describing insulin secretion and sensitivity (10,11), declines as diabetes develops (9–14). This curve has been analyzed in several studies (13–15), including the Diabetes Prevention Program (DPP). An earlier study examined pPRS associations with diabetes incidence and changes in insulin secretion and sensitivity from baseline to year 1 during DPP’s active phase (5); however, those pPRS were derived from European cohorts only (5).

We now investigate how two sets of global PRS and pPRS (together referred to as PRSs) for type 2 diabetes, derived from multiancestry cohorts, associate with the baseline relationship between insulin secretion and sensitivity in people with prediabetes in the DPP. We examine the roles of this relationship and treatment in attenuating or modifying PRSs associations with diabetes incidence during DPP and DPP Outcomes Study (DPPOS).

Research Design and Methods

Data Collection

DPP was a clinical trial that randomized participants to treatment with an intensive lifestyle intervention (ILS), metformin, or placebo to determine whether ILS or metformin prevented or delayed type 2 diabetes in high-risk adults (16). DPPOS is its long-term follow-up (17). Throughout DPP/DPPOS, diabetes development was assessed by American Diabetes Association 1997 criteria using an annual glucose tolerance test or a semiannual fasting plasma glucose test. Both studies were approved by clinical centers’ institutional review boards. Participants gave written informed consent.

We used data from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) and dbGaP repositories from 2052 DPP/DPPOS participants of self-reported White, African American, or Hispanic race or ethnicity. Race or ethnicity was determined by self-report from the original DPP questionnaire. These participants had baseline data on fasting and 30-min serum insulin (I₀ and I₃₀ in μU/mL) and plasma glucose (G₀ and G₃₀ in mg/dL) during a glucose tolerance test, follow-up for diabetes, and genotypic data. Diabetes developed in 434 participants (21%) during DPP and in 1,236 participants (60%) by the closing date for this analysis (23 February 2020).

We estimated insulin secretion using the corrected insulin response (CIR₃₀) (CIR₃₀ = [100*I₃₀]/[G₃₀*(G₃₀ − 70)]) and insulin sensitivity using the reciprocal of HOMA of insulin resistance (ISI₀ = 22.5/[I₀*(G₀/18.01)]).

Genotyping and Imputation

DNA was obtained from peripheral blood leukocytes. Genotyping was completed with the HumanCoreExome genome-wide array (Illumina, San Diego, CA). Genotype phasing and imputation were conducted separately for each race or ethnicity group in Beagle 5.2 using the 1000 Genomes Project Phase 3 data set as the reference panel (18,19) (see the Supplementary Material).

Generation of PRSs

We calculated PRSs following two previous analyses. We generated pPRS for eight single nucleotide polymorphism (SNP) clusters (Table 1) identified by Suzuki et al. (2) using hard-cluster analysis. We also generated pPRS for 12 clusters (Table 1) identified by Smith et al. (3) using soft-cluster analysis. We calculated a type 2 diabetes global PRS for each meta-analysis across all 1,289 and 650 SNPs identified by Suzuki et al. and Smith et al. (2,3), respectively (see the Supplementary Material). Supplementary Tables 1 and 2 show numbers of directly typed or successfully imputed SNPs in each cluster by race or ethnicity.

Table 1.

Associations of PRS with diabetes incidence

Set	Polygenic score	Model 1				Model 2
		DPP		DPPOS		DPP		DPPOS
		HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
Suzuki et al.	Global PRS	1.181 (1.074, 1.299)	0.0006	1.157 (1.094, 1.223)	<0.0001	1.122 (1.021, 1.234)	0.017	1.143 (1.081, 1.208)	<0.0001
	β-Cell + PI	1.080 (0.983, 1.187)	0.110	1.067 (1.009, 1.128)	0.022	1.096 (0.996, 1.207)	0.061	1.081 (1.021, 1.144)	0.007
	β-Cell − PI	1.053 (0.956, 1.159)	0.294	1.086 (1.027, 1.148)	0.004	1.065 (0.965, 1.176)	0.212	1.100 (1.039, 1.164)	0.001
	Residual glycemic	1.027 (0.936, 1.127)	0.577	1.057 (0.999, 1.117)	0.053	0.985 (0.897, 1.082)	0.752	1.038 (0.982, 1.098)	0.187
	Body fat	1.099 (0.999, 1.208)	0.514	1.049 (0.992, 1.110)	0.094	1.091 (0.992, 1.199)	0.072	1.048 (0.990, 1.109)	0.104
	Metabolic syndrome	1.084 (0.985, 1.194)	0.098	1.059 (1.001, 1.121)	0.046	1.044 (0.948, 1.150)	0.379	1.036 (0.979, 1.096)	0.225
	Obesity	1.045 (0.952, 1.147)	0.356	1.015 (0.960, 1.073)	0.600	1.003 (0.915, 1.099)	0.949	1.012 (0.958, 1.069)	0.664
	Lipodystrophy	1.102 (1.000, 1.213)	0.050	1.119 (1.057, 1.184)	0.0001	1.059 (0.964, 1.164)	0.234	1.103 (1.043, 1.167)	0.0006
	Liver/lipid metabolism	0.990 (0.903, 1.086)	0.833	0.993 (0.940, 1.049)	0.795	1.003 (0.913, 1.101)	0.953	1.009 (0.955, 1.066)	0.753
Smith et al.	Global PRS	1.139 (1.035, 1.253)	0.008	1.157 (1.094, 1.223)	<0.0001	1.074 (0.977, 1.179)	0.139	1.135 (1.074, 1.199)	<0.0001
	β-Cell 1	1.089 (0.990, 1.197)	0.078	1.058 (1.000, 1.119)	0.051	1.050 (0.956, 1.153)	0.310	1.040 (0.984, 1.101)	0.166
	β-Cell 2	0.993 (0.903, 1.092)	0.885	1.041 (0.984, 1.102)	0.165	1.023 (0.927, 1.128)	0.656	1.069 (1.009, 1.133)	0.023
	Proinsulin	0.966 (0.878, 1.063)	0.477	1.033 (0.976, 1.094)	0.267	0.977 (0.887, 1.077)	0.639	1.045 (0.986, 1.107)	0.140
	Obesity	1.056 (0.961, 1.159)	0.257	1.002 (0.947, 1.061)	0.936	0.984 (0.902, 1.074)	0.720	0.994 (0.942, 1.050)	0.840
	Hyper insulin	1.072 (0.975, 1.179)	0.151	1.032 (0.976, 1.092)	0.270	1.030 (0.936, 1.133)	0.550	1.020 (0.964, 1.080)	0.486
	Cholesterol	1.031 (0.937, 1.134)	0.529	0.986 (0.932, 1.042)	0.611	1.019 (0.925, 1.122)	0.708	0.988 (0.935, 1.045)	0.674
	Lipodystrophy 1	1.019 (0.925, 1.121)	0.706	1.077 (1.018, 1.139)	0.0093	0.997 (0.905, 1.099)	0.954	1.066 (1.009, 1.128)	0.024
	Lipodystrophy 2	0.999 (0.908, 1.098)	0.978	1.050 (0.954, 1.068)	0.745	1.000 (0.907, 1.101)	0.994	1.050 (0.992, 1.111)	0.093
	Liver/lipid	0.976 (0.887, 1.073)	0.610	1.009 (0.954, 1.068)	0.745	0.955 (0.868, 1.050)	0.341	0.997 (0.943, 1.055)	0.922
	Bilirubin	1.021 (0.929, 1.122)	0.660	1.016 (0.961, 1.074)	0.577	0.981 (0.892, 1.079)	0.688	0.994 (0.941, 1.051)	0.845
	SHBG-LpA	1.005 (0.914, 1.105)	0.920	0.995 (0.940, 1.053)	0.859	0.975 (0.887, 1.071)	0.597	0.973 (0.920, 1.029)	0.343
	ALP negative	1.090 (0.993, 1.197)	0.068	1.036 (0.980, 1.095)	0.208	1.137 (1.036, 1.248)	0.0068	1.063 (1.005, 1.123)	0.032

All models adjusted for sex, race or ethnicity, treatment, and baseline age. Model 2 additionally adjusted for baseline insulin secretion compensation and baseline insulin secretion demand. All polygenic scores were centered at 0 within race or ethnicity groups. HRs for polygenic risk scores are given per 1 SD in the total sample. Statistically significant associations have been highlighted using boldface. SHBG, sex hormone–binding globulin.

Statistical Analysis

The baseline relationship between insulin secretion and sensitivity was studied by estimating ln(CIR₃₀) as a function of ln(ISI₀) using standardized major axis (SMA) regression (13) without assuming a hyperbolic relationship between the variables. Estimated insulin secretion compensation (distance away from SMA line) quantifies the ability of secreted insulin to compensate for changes in insulin sensitivity representing a measure of homeostatic response; higher compensation associates with lower diabetes incidence (13). Estimated insulin secretion demand (distance along SMA line) quantifies demand on insulin secretion imposed by low insulin sensitivity representing a measure of allostatic load/physiological stress on β-cells; higher demand associates with higher diabetes incidence (13). We modeled PRSs associations with baseline compensation and demand using multivariate linear mixed models; for each PRS, models had compensation and demand as outcomes, participant identifier as a random intercept, and the PRS, sex, baseline age, self-reported race or ethnicity, and treatment as covariates. To determine whether each PRS associated with this relationship, we divided participants into quartiles of that PRS and tested whether the SMA lines for the different quartiles shared common slopes or intercepts. Significantly different slopes and/or intercepts indicate that the relationship differs by PRS quartiles.

We estimated PRSs associations with diabetes incidence during DPP (median of 3.2 years) and DPPOS (up to 23 years) using Cox models adjusted for the same covariates. To make hazard ratios (HRs) comparable across the whole sample, for each race or ethnicity group, we centered all PRSs around their corresponding mean, then standardized the variance of each PRS to 1 across the whole sample. We defined follow-up for diabetes as time to diabetes diagnosis or time to final visit during period of interest if diabetes was not diagnosed.

We calculated percentage of PRS associations with diabetes incidence attenuated by insulin demand and compensation by conducting exploratory mediation analyses using the causal steps approach (20) (see the Supplementary Material). We tested interactions of PRSs with insulin secretion compensation and demand and with treatment to investigate whether the PRSs modified associations of these variables with diabetes incidence. Subgroup analyses of PRS associations with diabetes incidence were conducted within each treatment and race or ethnicity group. Main analyses were conducted with self-reported race or ethnicity. We also conducted analyses adjusting for the first 10 principal components to assess potential influence of residual population stratification; principal components largely separated participants by self-reported race or ethnicity (Supplementary Fig. 1).

Data Resource and Availability

Data sets are available in the NIDDK (for DPP: https://repository.niddk.nih.gov/study/38; for DPPOS: https://repository.niddk.nih.gov/study/40) and dbGaP (for DPP genome-wide association study: https://dbgap.ncbi.nlm.nih.gov/beta/study/phs000681.v2.p1/) repositories.

Results

Supplementary Table 3 shows descriptive statistics by treatment group. Distributions of most PRSs differed by race or ethnicity but not by treatment; some differed by sex (Supplementary Figs. 2–4 and Supplementary Tables 4–9). Overall, greater age associated with lower PRSs (Supplementary Tables 10 and 11).

Correlations between β-cell function–related pPRS and global PRS within both sets were >0.30 (Supplementary Figs. 5 and 6). The correlation between the global PRSs of Suzuki et al. (2) and Smith et al. (3) was r = 0.70 (Supplementary Fig. 7). Correlations between diabetes-related traits with most PRSs were generally weak (Supplementary Tables 12 and 13).

Relationship Between Insulin Secretion and Sensitivity

Figure 1 shows relationships between insulin secretion and sensitivity by quartiles of selected associated PRSs; Supplementary Tables 14 and 15 show common slopes/intercepts tests results. Supplementary Figs. 8 and 9 contain curves for all PRSs.

Figure 1.

Relationship between insulin secretion and sensitivity by quartiles of Suzuki et al. (2) and Smith et al. (3) PRSs. A–D: The 25th to 75th centiles of ISI₀ are shown in the x-axis and the 25th and 75th centiles of CIR₃₀ are shown in the y-axis. The points on each line represent the geometric means of ISI₀ and CIR₃₀. T2D, type 2 diabetes.

Suzuki et al. PRSs

Intercepts of the SMA line describing the relationship between ln(CIR₃₀) and ln(ISI₀) differed across the Suzuki et al. (2) global PRS (P = 0.049) quartiles; participants in different global PRS quartiles had different insulin secretion levels at the same sensitivity. Slopes describing this relationship across β-cell − proinsulin (PI) (P = 0.044) pPRS quartiles differed significantly; that is, secretion relative to sensitivity differed by pPRS.

The Suzuki et al. (2) global, β-cell + PI, and β-cell − PI PRSs associated with lower ln(CIR₃₀) and higher ln(ISI₀) (Fig. 2A and Supplementary Table 16). The global PRS associated with lower compensation and demand, (Fig. 2B and Supplementary Table 17). The β-cell + PI and β-cell − PI pPRS associated with lower demand; these associations were stronger in older participants (Supplementary Table 18).

Figure 2.

Associations of Suzuki et al. (2) and Smith et al. (3) PRSs with baseline ln(ISI₀), ln(CIR₃₀), insulin secretion compensation and insulin secretion demand. A and C: The b units represent differences in ln(ISI₀) per 1 SD higher PRS and differences in ln(CIR₃₀) per 1 SD higher PRS. B and D: The b units are SD of insulin secretion compensation per 1 SD higher PRS and SD of insulin secretion demand per 1 SD higher PRS. A multivariate linear mixed-effects (MLME) models were used to model associations of PRSs with ln(ISI₀) and ln(CIR₃₀): the MLME fit for each PRS is ln(ISI₀) = ${\mathit{\beta }}_{10}+{\mathit{\beta }}_{11}\times \mathrm{PRS}+{\beta }_{12}\times \mathrm{male}+{\beta }_{13}\times \mathrm{age}+{\beta }_{14}\times \mathrm{African\hspace{0.5em}American}+{\beta }_{15}\times \mathrm{Hispanic}+{\beta }_{16}\times \mathrm{metformin}+{\beta }_{17}\times \mathrm{ILS}+b_{i1}$ and ln(CIR₃₀) = ${\beta }_{20}+{\beta }_{21}\times \mathrm{PRS}+{\beta }_{22}\times \mathrm{male}+{\beta }_{23}\times \mathrm{age}+{\beta }_{24}\times \mathrm{African\hspace{0.5em}American}+{\beta }_{25}\times \mathrm{Hispanic}+{\beta }_{26}\times \mathrm{metformin}+{\beta }_{27}\times \mathrm{ILS}+b_{i2}$ where $b_{i1}$ and $b_{i2}$ are random intercepts with normal distribution centered around 0 with variance covariance matrix capturing correlation between the ln(ISI₀) and ln(CIR₃₀) in the off-diagonal element. A separate MLME model with similar structure was used to study associations of PRSs with compensation and demand. The flipping of the triangles (sensitivity and demand) from above to below zero in A–D are due to A and C showing associations of PRSs with insulin sensitivity (the opposite of insulin resistance) and B and D showing insulin secretion demand (determined by resistance).

Smith et al. PRSs

Intercepts of SMA lines describing the relationship between ln(CIR₃₀) and ln(ISI₀) differed across the Smith et al. (3) global (P = 0.042) and β-cell 1 (P = 0.044) PRSs quartiles. Curves for higher quartiles of these PRSs were left- and downward-shifted compared with curves for lower quartiles, suggesting worse β-cell function. The Smith et al. global, β-cell 1, and β-cell 2 PRSs associated with lower ln(CIR₃₀) and higher ln(ISI₀), (Fig. 2C and Supplementary Table 19). The Smith et al. global and β-cell 1 PRSs associated with lower compensation and demand (Fig. 2D and Supplementary Table 20). In participants ≥60 years old, a 1 SD higher global, β-cell 1, and β-cell 2 PRSs resulted in lower adjusted insulin demand than in participants <60 years old (Supplementary Table 21).

Diabetes Incidence

Baseline secretion compensation associated with decreased diabetes incidence during DPP (HR = 0.60 per SD, 95% CI 0.56–0.65; P < 0.0001) and DPPOS (HR = 0.66, 95% CI 0.62–0.69; P < 0.0001). Baseline demand associated with higher incidence during DPPOS (HR = 1.08 per SD, 95% CI 1.02–1.14; P = 0.009) but not DPP (HR = 1.05, 95% CI 0.96–1.16; P = 0.278).

Suzuki et al. PRSs and Incident Diabetes

The Suzuki et al. (2) global PRS associated with diabetes incidence during DPP (HR = 1.18 per SD, 95% CI 1.07–1.30) and DPPOS (HR = 1.16, 95% CI 1.09–1.22). Higher β-cell + PI and β-cell − PI pPRS increased incidence during DPPOS (Table 1). Baseline insulin secretion compensation and demand attenuated <40% of the Suzuki et al. global PRS association with diabetes incidence during both periods (Supplementary Table 22).

When modeling diabetes incidence during DPP, none of the Suzuki et al. (2) PRSs interacted with baseline compensation and demand, while treatment interacted with the obesity pPRS (P = 0.008). With higher obesity pPRS values, the effect of ILS and metformin on diabetes incidence, compared with placebo, diminished (Fig. 3D). During DPPOS, none of the Suzuki et al. PRSs interacted with treatment.

Figure 3.

Interactions of pPRS with insulin secretion compensation and demand and with treatment. A–E: Likelihood ratio tests were performed to evaluate whether interaction terms of compensation and demand with PRSs were statistically significant when adding these interaction terms to model 2 in Table 1. Likelihood ratio tests were performed to evaluate whether interaction terms of treatment with PRSs were statistically significant when adding these interaction terms to model 1 in Table 1. The panels are arranged to first show significant interactions with insulin secretion compensation and demand and then significant interactions with treatments during the DPP or DPPOS periods for pPRS that had significant interactions (P < 0.05).

There were no clear patterns of associations of Suzuki et al. (2) PRSs with diabetes incidence by treatment groups during DPP or DPPOS (Supplementary Tables 23 and 24). During both periods, PRSs associations with diabetes incidence were generally stronger in White participants (Supplementary Tables 25 and 26).

Smith et al. PRSs and Incident Diabetes

Higher Smith et al. (3) global PRS increased diabetes incidence during DPP (HR = 1.14 per SD, 95% CI 1.04–1.25) and DPPOS (HR = 1.16, 95% CI 1.09–1.22) (Table 1). Baseline compensation and demand attenuated <46% of these associations (Supplementary Table 27).

Compensation and demand interacted with Smith et al. (3) global PRS during DPP (P = 0.014) and DPPOS (P = 0.001) and with Smith et al. obesity pPRS (P = 0.040) during DPPOS. At higher values of global and obesity PRSs, the damaging effect of higher demand diminished, while the protective effect of higher compensation was only modestly attenuated (Fig. 3A–C). During DPPOS, treatment interacted with the β-cell 1 (P = 0.005) pPRS. Metformin and ILS were effective compared with placebo only in people with β-cell 1 pPRS at or below the mean (Fig. 3E).

Supplementary Tables 28 and 29 show Smith et al. PRSs associations with diabetes incidence within each treatment group for DPP and DPPOS. Within each race or ethnicity group, few of the Smith et al. PRSs predicted diabetes incidence (Supplementary Tables 30 and 31).

Discussion

Associations of type 2 diabetes global PRS and cluster pPRS with the relationship between insulin secretion and sensitivity has not, to our knowledge, been examined in people with prediabetes. We investigate how these scores relate to insulin secretion compensation and demand, two variables that summarize the relationship between insulin secretion and sensitivity, and determine whether this relationship attenuates or modifies PRSs associations with diabetes incidence in DPP participants. We tested many hypotheses and present nominal P values to limit false negatives associated with correction for multiple nonindependent hypotheses; thus, some of the significant associations observed may be spurious.

Most PRSs weakly correlated with diabetes-related traits. We replicated most expected associations of pPRS from both sets with insulin secretion and sensitivity. For pPRS associated with the baseline relationship between insulin secretion and sensitivity, higher pPRS generally resulted in worse β-cell function. β-Cell dysfunction pPRS from both sets associated with lower insulin secretion demand and greater sensitivity. This suggests that DPP participants with high genetic burden for β-cell dysfunction, with no diabetes at trial entry, were more insulin-sensitive, resulting in low secretion demand, allowing them to compensate. This is consistent with a previous DPP study where carriers of the risk allele for two SNPs in the TCF7L2 gene, thought to decrease insulin secretion, had enhanced insulin sensitivity at baseline (21).

Global PRS from both sets associated with diabetes incidence during both follow-up periods; these associations were stronger than those of any pPRS, which were modest or nonexistent. Baseline insulin secretion compensation and demand attenuated some associations and interacted with several pPRS, such that diabetes incidence decreased with higher insulin secretion compensation at any pPRS level (Supplementary Fig. 10). In contrast, higher insulin demand did not always increase incidence significantly, especially at higher pPRS values. This indicates that, in people with higher genetic diabetes risk, baseline compensation was more important for diabetes incidence than demand. This agrees with suggestions that, in people at higher risk, the most important factor for delaying or preventing diabetes is insulin compensation (13–15). Efforts should focus on improving or finding treatments that improve or maintain adequate compensation when insulin sensitivity diminishes.

The Suzuki et al. (2) obesity pPRS interacted with treatment for diabetes incidence during DPP, with ILS and metformin conferring protective at low, but not high, obesity-associated genetic risk for diabetes. This contrasts with findings that metformin worked better at higher BMIs (16). This might be due to the weak correlation between BMI and this pPRS or might suggest that metformin prevents diabetes in obesity caused by environmental, but not genetic, factors. This highlights the importance of considering both population-specific scores and environmental contributions to diabetes (22).

The Smith et al. (3) β-cell 1 pPRS, describing genetic diabetes risk due to β-cell dysfunction related to glucose homeostasis, also interacted with treatment, suggesting that both treatments were protective only at low or mean values of this pPRS. This is consistent with participants with a high β-cell cluster pPRS having worsening insulin secretion (CIR₃₀) from baseline to year 1, despite interventions (5). It is also consistent with both interventions succeeding primarily by improving insulin sensitivity, making them less effective in people whose glucose intolerance is primarily due to insulin deficiency (21,23,24).

Each clustering approach has advantages and disadvantages (2,3). Each SNP in the Suzuki et al. (2) analysis was assigned to only one cluster, not accounting for pleiotropy of SNPs associated with multiple diabetes-related traits. Although the Smith et al. (3) approach accounts for such pleiotropy, about half of these SNPs did not contribute to any pPRS, because of the cutoff point for the cluster weights, resulting in information loss. Most of the significant associations with diabetes incidence and its pathophysiology measures that we identified were from the Suzuki et al. (2) set.

In a Finnish study, glycemic responses to lifestyle treatment in prediabetes were similar in those with high or low global PRS, but the study did not examine pPRS (25). Our findings build on a previous investigation of pPRS in DPP participants (5) that concluded, as we do, that therapies in addition to or instead of DPP interventions may be necessary to prevent progression from prediabetes to diabetes in people with high genetic susceptibility for β-cell dysfunction (5). We assessed β-cell function through the relationship between insulin secretion and sensitivity, while they used only measures of insulin secretion without accounting for sensitivity. They investigated pPRS derived from primarily European cohorts, while we studied pPRS derived from multiancestry cohorts, which is important because DPP participants reflect the diversity of the U.S. population (16). They only studied diabetes incidence during the DPP follow-up and did not find any modifiers of the pPRS associations. In contrast, we also investigated diabetes incidence during the DPPOS long-term follow-up, during which baseline insulin secretion compensation and demand as well as treatment modified the associations of some pPRS with diabetes incidence. We obtained similar results by adjusting for principal components for genetic ancestry instead of self-reported race or ethnicity (Supplementary Tables 32–34).

In conclusion, in DPP/DPPOS, global PRS had stronger associations with diabetes incidence than any individual pPRS, and pPRS associations with diabetes-related traits were weak. Thus, further work to improve pPRS is necessary before they can be incorporated into prevention efforts for people with prediabetes. The Smith et al. (3) β-cell 1 pPRS interacted with ILS and metformin, suggesting these treatments may be less effective in those with high genetic predisposition to diabetes because of β-cell dysfunction. Regardless of genetic risk as measured by cluster pPRS for type 2 diabetes, inadequate compensatory insulin secretion contributes to progression to diabetes in people with prediabetes despite preventive treatment measures.

This article contains supplementary material online at https://doi.org/10.2337/figshare.32085156.

Funding Statement

This research was supported by the Intramural Research Program of NIDDK within the National Institutes of Health (NIH). DPP was conducted by the DPP Research Group and supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (RO1-DK0702041), the General Clinical Research Center Program, the National Institute of Child Health and Human Development, the National Institute on Aging, the Office of Research on Women's Health, the Office of Research on Minority Health, the Centers for Disease Control and Prevention (CDC), and the American Diabetes Association. The resources from the DPP study were supplied by NIDDK Central Repository (NIDDK-CR). The DPPOS study was conducted by the study investigators and supported by NIDDK. The resources from the DPPOS (https://doi.org/10.58020/66×5-8y21) study reported here were supplied by NIDDK-CR and are available for request at https://repository.niddk.nih.gov.