Genetic Predictors of Response to Oral Insulin for Type 1 Diabetes Prevention

Abstract

OBJECTIVE

The TrialNet Oral Insulin Prevention Trial TN07 tested oral insulin to prevent stage 3 type 1 diabetes (T1D) in 560 relatives with stage 1 T1D of individuals with T1D. Of the three predefined risk strata, participants in secondary stratum 1 (SS1) (n = 55), characterized by low first-phase insulin release, responded significantly better to oral insulin. We aimed to identify genetic factors associated with treatment response.

RESEARCH DESIGN AND METHODS

The TEDDY-T1DExomeChip was used to genotype 552 participants with available DNA. Cox models examined associations between response to oral insulin and HLA haplotypes, 33 preselected T1D-associated single nucleotide polymorphisms, T1D genetic risk score 2 (GRS2), and type 2 diabetes (T2D)–associated polygenic scores. For primary analyses, P values were Benjamini-Hochberg (BH) corrected for multiple comparisons; results not passing correction were considered nominal.

RESULTS

GLIS3 rs7020673 was significantly associated with response to oral insulin in SS1 (BH-corrected P = 0.031 without and P = 0.022 with covariate adjustment). Additional nominal associations included better response with HLA-DRB1*04:01-DQA1*03:01-DQB1*03:02 (hazard ratio [HR] 0.22 vs. 1.09; unadjusted P = 0.031; adjusted P = 0.045) in SS1 and worse responses with TNFAIP3 and CTLA4 in at least one stratum. In exploratory analyses, participants with T1D-GRS2 >12.5 responded better to oral insulin (HR 0.68) than those with T1D-GRS2 ≤12.5 (HR 2.10; unadjusted P = 0.003; adjusted P = 0.006) in the overall cohort, and lower proinsulin- and obesity-partitioned T2D polygenic scores were associated with greater treatment benefit in SS1 and in another secondary stratum, respectively.

CONCLUSIONS

Genetic differences distinguish responders from nonresponders to oral insulin for T1D prevention. Genetics may enable precision medicine by identifying individuals likely to benefit from T1D-modifying therapies.

Graphical Abstract

Introduction

The TrialNet Oral Insulin Prevention Trial TN07 (TrialNet TN07) was a randomized placebo-controlled trial that tested the treatment effect of oral insulin in preventing stage 3 type 1 diabetes (T1D) in relatives with stage 1 T1D of individuals with T1D (1). The effect of oral insulin versus placebo was evaluated in several predefined analysis strata (definitions provided in the Supplementary Material). The oral insulin dose of 7.5 mg/day did not delay progression to stage 3 T1D in the primary stratum or all strata collectively (1). However, in secondary stratum 1 (SS1), characterized by low first-phase insulin release (FPIR), the oral insulin group showed a significantly decreased progression to stage 3 T1D compared with the placebo group. Although these results do not validate oral insulin treatment as an effective measure for T1D prevention in general, the study highlighted variations in the participants’ responses to oral insulin and the need for further investigation into factors that differentiate responders from nonresponders.

Genetics plays an essential role in the development of T1D, with >100 loci found to be associated with the disease (2). While there has been a growing understanding of the genetics of T1D, there is limited research exploring the interplay between genetics and response to therapies to modify the natural course of T1D. The rising interest in personalized medicine underscores the value of using genetics to predict individual responses to treatment. The primary aim of this study was to evaluate the genetic factors contributing to the differences in response to oral insulin in the TrialNet TN07 trial. Our findings will offer insights into how to use genetics to select individuals who will respond to oral insulin and potentially other treatments for T1D prevention.

Research Design and Methods

Population

The participants in the TrialNet TN07 trial were relatives of individuals with T1D who met the primary inclusion criteria of having normal glucose tolerance and positivity for multiple islet autoantibodies (i.e., stage 1 T1D) with positive insulin autoantibodies (IAAs). The cohort was divided into four predefined analysis strata. The primary stratum and SS1 were characterized by confirmation of islet cell autoantibody (ICA), whereas SS2 and SS3 were characterized by lower FPIR. Detailed definitions of the analysis strata, inclusion criteria, and participant characteristics have been previously published in the primary outcome report (1), and a summary table is provided in section 1 of the Supplementary Material. SS1 was unique in that it was the only stratum demonstrating a positive response in the primary outcome report and had the highest overall risk, warranting greater emphasis in our analysis. Following the design of the parent study (1), we performed analyses in primary stratum, SS1, SS2, and all strata combined. SS3 was not analyzed separately as it only included two participants. All participating clinical centers received institutional review board approval, and informed consent was obtained from all participants.

Materials

Of 560 participants in TrialNet TN07, 552 had DNA samples that passed quality-control checks. All participants with available DNA were genotyped using the T1DExomeChip (a custom genotyping array with >90,000 custom single nucleotide polymorphisms [SNPs] added to the Infinium CoreExome-24 v.1.1 BeadChip [Illumina]) at the Department of Genome Sciences, University of Virginia. We selected 33 SNPs that were previously shown to be related to T1D progression or insulin production from previous publications (3–6) (a complete list is provided in Supplementary Table 1). We screened the selected SNPs for high linkage disequilibrium and retained those with the highest minor allele frequencies. A Consolidated Standards of Reporting Trials–style diagram illustrating the process of selecting the genetic factors evaluated in the primary aim is presented in Supplementary Fig. 1. Of these, 25 SNPs were directly genotyped, 6 were imputed using the Trans-Omics for Precision Medicine Imputation Server version R2 (rs111776337 with imputation quality score R² = 0.919, rs55900661 with R² = 0.913, rs113306148 with R² = 0.979, rs11705721 with R² = 0.991, rs73043122 with R² = 0.999, and rs12151883 with R² = 0.899) (7,8), and 2 SNPs could not be imputed. In addition, we calculated T1D genetic risk score 2 (GRS2) (9), which includes a total of 67 SNPs; of these, 30 were directly genotyped, 3 non-HLA SNPs were imputed with a median R² of 0.997 (minimum 0.858, maximum 0.999) using the Trans-Omics for Precision Medicine Imputation Server (7,8), and 5 in the HLA region (rs72848653 with R² = 0.999, rs9266268 with R² = 0.999, rs16899379 with R² = 0.998, rs2524277 with R² = 0.995, and rs9268500 with R² = 0.925) were imputed using the Michigan Imputation Server, with the high-resolution HLA reference panel spanning five global populations (n = 21,546) based on whole-genome sequencing data (10). The T1D-GRS2 is calculated as the sum of two components, namely, the risk score in the HLA region and the risk score in the non-HLA region. A type 2 diabetes (T2D) GRS (11) and several T2D partitioned polygenic scores (pPGSs) (12) were calculated using available genotype and imputation data. In this work, we named the pPGSs by the characteristics of clusters that were originally identified in DiCorpo et al. (12) (i.e., BMI, β-cell, lipodystrophy, liver, obesity, and proinsulin pPGSs). The HLA haplotypes of participants were typed at the HLA Core Facility, Barbara Davis Center for Diabetes, using Roche’s Molecular Systems HLA PCR and Linear Array, Abbott Laboratories’ DRB1 Sequencing Kit, or Bio-Rad’s Bio-Plex based on Luminex xMAP technology.

Statistical Analysis

The distribution of binary and categorical variables was summarized using frequencies and percentages. The distribution of continuous variables was summarized using means and SDs. The primary analyses involving binary genetic variants included only those with a frequency ≥20% in the analysis cohort.

Cox proportional hazards models were used to investigate the primary aim of identifying genetic factors that may contribute to the observed differences in oral insulin responses, where the time to stage 3 T1D was the survival outcome, and interactions between the treatment group and genetic variant were included as independent variables. For a binary genetic variant, we compared the hazard ratio (HR) of treatment versus placebo in participants with the genetic variant with those without it. An HR of 1 indicates no difference in time to stage 3 T1D in the two treatment groups, whereas an HR <1 indicates a slower progression to stage 3 T1D with the oral insulin treatment. Kaplan-Meier curves by treatment groups and genetic factors were used to visualize the results. For genetic factors with additive genotype coding based on imputed allele dosage, we compared the HRs of treatment versus placebo for every additional copy of the allele. For genetic factors with a continuous scale (e.g., T1D-GRS2), we examined whether differences in HRs can vary across the range of the continuous factors using Cox models with interaction terms between treatment assignment and the continuous factors (13). To facilitate interpretation and comparability, the results were visualized by the fitted survival curves from the Cox model at the lower and upper quartile values of the continuous genetic factor, accounting for differences in their scales (14). To evaluate potential thresholds of T1D-GRS2 that differentiate response to oral insulin treatment, we systematically examined a sequence of T1D-GRS2 cutoffs across the cohort distribution; the cutoff of 12.5 was selected as it lay at the midpoint of the empirically defined range (11.64–12.83) where associations with oral insulin response were most consistent. The difference in HRs was evaluated by the Wald test based on the interaction term, either unadjusted or adjusted for age, sex, and the first two principal components to account for population stratification (ancestral heterogeneity). The primary aim was to analyze the five high-risk HLA haplotypes identified by Erlich et al. (15), selected T1D-associated SNPs, and the T1D-GRS2 and its HLA and non-HLA subcomponents. A Consolidated Standards of Reporting Trials–style diagram is presented in Supplementary Fig. 1 to summarize the number of genetic factors investigated in the primary aim. P values for genetic analyses related to the primary aims were corrected for multiple comparisons using the Benjamini-Hochberg (BH) method. Results with BH-corrected P < 0.05 were considered statistically significant, whereas those with uncorrected P < 0.05 are described as nominal or hypothesis-generating findings. The association between treatment effects and T2D-GRS and pPGS (16) was examined in the exploratory analyses. The exploratory analyses also investigated thresholds of continuous factors that best differentiate treatment responses. Additionally, we conducted exploratory analyses to compare the genetic profiles of participants among these three strata. The means of the risk scores were compared using t tests, while Tukey method was used for multiple group comparisons. Frequencies of binary genetic variants were compared using Fisher exact test. An additional mediation analysis (17) was conducted by checking whether the gene-treatment interaction remained present after including IAA titer as a potential mediator. The method for the mediation analysis is detailed in section 2 of the Supplementary Material.

Data and Resource Availability

The code to generate the HLA interaction part of the T1D-GRS2 is freely available online (https://github.com/sethsh7/PRSedm) (9).

Results

The characteristics of the study population at treatment assignment are summarized in Table 1. Response to oral insulin treatment was assessed among participants with different HLA class II haplotypes, with results presented in Table 2. The presence of HLA-DRB1*04:01-DQA1*03:01-DQB1*03:02 was nominally associated with a better response to oral insulin treatment (HR 0.22 vs. 1.09; unadjusted P = 0.031; adjusted P = 0.045) (Supplementary Fig. 2) among participants in SS1. However, this association did not remain statistically significant after correction for multiple comparisons and is therefore considered hypothesis-generating. In the exploratory analysis comparing the frequency of HLA types in the analysis strata, SS1 had a higher frequency of the haplotype HLA-DRB1*04:01-DQA1*03:01-DQB1*03:02 compared with SS2 (P = 0.006) (Supplementary Fig. 3). Because IAA positivity was one of the inclusion criteria in the TrialNet TN07 trial, we examined the association between this haplotype and IAA titers. We found that participants with HLA-DRB1*04:01-DQA1*03:01-DQB1*03:02 had higher IAA titers (log scaled) in the primary stratum (P = 0.009) and when all strata were combined (P = 0.008). Mediation analysis showed that the HLA-treatment interaction remained evident after accounting for IAA titers, and IAA titers alone did not differentiate response to oral insulin treatment (P = 0.295), suggesting that the observed nominal association was not mediated by IAA levels.

Table 1.

Characteristics of the study population at enrollment by strata

Characteristic	Primary stratum (n = 389)	SS1 (n = 55)	SS2 (n = 112)†
Islet autoantibodies	Confirmed ICA+ or both GADA+ and IA-2A+	Confirmed ICA+ or both GADA+ and IA-2A+	ICA+ and confirmed GADA+ or IA-2A+
First-phase insulin release	Above threshold	Below threshold	Above threshold
Treatment assignment
Oral insulin	203 (52)	28 (51)	50 (45)
Placebo	186 (48)	27 (49)	62 (55)
Sex
Female	143 (37)	17 (31)	56 (50)
Male	246 (63)	38 (69)	56 (50)
Race
Other	11 (2.9)	1 (2.0)	6 (5.8)
White	349 (93)	50 (98)	96 (93)
White and another race	14 (3.7)	0 (0)	1 (1.0)
Unknown	15	4	9
Ethnicity
Hispanic or Latino	36 (9.3)	3 (5.5)	12 (11)
Not Hispanic or Latino	344 (88)	49 (89)	96 (86)
Unknown	9 (2.3)	3 (5.5)	4 (3.6)
Age (years)	8 (5, 12)	8 (6, 13)	7 (5, 10)
T1D-GRS2	14.0 (12.7, 15.1)	14.3 (13.1, 15.3)	13.5 (12.0, 14.7)
Unknown	5	0	1
HLA component of T1D-GRS2	9.9 (8.7, 11.0)	9.9 (8.8, 11.3)	9.2 (8.2, 10.5)
Unknown	5	0	1
Non-HLA component of T1D-GRS2	4.1 (3.5, 4.6)	4.2 (3.8, 4.8)	4.2 (3.5, 4.6)
Unknown	5	0	1
DRB1*04:01-DQA1*03:01-DQB1*03:02 present	178 (46)	32 (58)	37 (33)
DRB1*03:01-DQA1*05:01-DQB1*02:01 present	162 (42)	18 (33)	39 (35)

Data are n (%) or median (interquartile range).

†SS3 was defined as ICA⁺ and confirmed GADA⁺ or IA-2A⁺, FPIR below threshold. It only included two individuals and, thus, is not presented.

Table 2.

Effects of HLA class II haplotypes on the effect of oral insulin treatment for stage 3 T1D prevention

		Unadjusted			Adjusted
DRB1*-DQA1*-DQB1* and stratum	n †	HR (95% CI) in participants without vs. with the haplotype	P	BH-corrected P	HR (95% CI) in participants without vs. with the haplotype	P	BH-corrected P
03:01-05:01-02:01
All	221 of 558	1.03 (0.69–1.55) vs. 0.65 (0.43–0.98)	0.113	0.527	1.02 (0.68–1.55) vs. 0.66 (0.43–1.00)	0.139	0.703
Primary	162 of 389	1.23 (0.75–2.02) vs. 0.62 (0.38–1.01)	0.055	0.603	1.17 (0.70–1.95) vs. 0.65 (0.40–1.07)	0.109	0.601
SS1	18 of 55	0.32 (0.13–0.81) vs. 0.71 (0.21–2.43)	0.301	0.897	0.28 (0.11–0.73) vs. 0.72 (0.19–2.72)	0.252	0.793
SS2	39 of 112	1.22 (0.31–4.89) vs. 0.65 (0.20–2.14)	0.498	0.987	1.18 (0.28–4.95) vs. 0.93 (0.25–3.41)	0.817	0.976
04:01-03:01-03:02
All	247 of 558	0.88 (0.59–1.31) vs. 0.85 (0.56–1.30)	0.920	0.965	0.85 (0.57–1.27) vs. 0.88 (0.57–1.35)	0.910	0.983
Primary	178 of 389	0.85 (0.53–1.38) vs. 0.96 (0.58–1.58)	0.741	0.874	0.80 (0.49–1.31) vs. 0.99 (0.59–1.65)	0.561	0.729
SS1	32 of 55	1.09 (0.38–3.17) vs. 0.22 (0.08–0.61)	0.031	0.336	1.04 (0.35–3.16) vs. 0.23 (0.08–0.64)	0.045	0.500
SS2	37 of 112	0.91 (0.31–2.71) vs. 1.67 (0.30–9.12)	0.557	0.987	0.98 (0.32–3.02) vs. 2.12 (0.35–12.94)	0.479	0.976

Statistical tests were based on Cox proportional hazards models with interaction terms between HLA genotypes (binary variables indicating presence of the haplotype) and treatment assignment. The size of the interaction effects was measured by the difference in HR (treatment vs. placebo) in participants with and without the HLA haplotype. P values were based on the Wald test. Results are presented with and without correction for multiple comparisons using the BH method.

†Number of participants with the HLA haplotype of the number of participants with known HLA haplotype in the stratum.

We then analyzed the association between the selected SNPs and response to oral insulin. SNPs with significant associations are summarized in Table 3, and the complete results are provided in Supplementary Table 1. After applying the predefined correction for multiple comparisons, rs7020673 within the GLIS3 locus remained significantly associated with response to oral insulin treatment in SS1 (BH-corrected P = 0.031 with no covariate adjustment; P = 0.022 with covariate adjustment). Without correction, the rs7020673 C allele was associated with worse response to oral insulin treatment in SS1 (unadjusted P = 0.001; adjusted P < 0.001) and SS2 (unadjusted P = 0.040; adjusted P = 0.090). In addition, two other SNPs showed nominal associations with treatment response in at least one analysis stratum. Within the TNFAIP3 locus, the rs2327832 G allele was associated with a worse response in all strata combined (unadjusted P = 0.011; adjusted P = 0.025) and in primary stratum (unadjusted P = 0.017; adjusted P = 0.038). Within the CTLA4 locus, the rs3087243 A allele was associated with a worse response to treatment in SS2 (unadjusted P = 0.040; adjusted P = 0.086). The fitted survival curves of treatment response rates for the reported SNPs are also presented in Supplementary Fig. 4A–C. In exploratory analysis comparing allele frequencies in the analysis strata, SS1 had more participants with the IL27 locus rs4788084 C allele than the primary stratum (P = 0.012), and the primary stratum had more participants with the INS locus rs1004446 A allele than SS1 (P = 0.021) (Supplementary Fig. 5).

Table 3.

Selected SNPs with significant effects on response to oral insulin treatment

						Unadjusted			Covariate adjusted
SNP	Major allele	Minor allele	Chromosome	Gene	Stratum	HR (95% CI) in participants without minor alleles vs. with one additional copy of the minor allele	P	BH-corrected P	HR (95% CI) in participants without minor alleles vs. with one additional copy of the minor allele	P	BH-corrected P
rs3087243	G	A	2	CTLA4	SS2	0.33 (0.08–1.45) vs. 1.53 (0.51–4.53)	0.040	0.509	0.42 (0.09–2.01) vs. 1.55 (0.52–4.59)	0.086	0.492
rs2327832	A	G	6	TNFAIP3	All	0.66 (0.46–0.95) vs. 1.25 (0.83–1.87)	0.011	0.233	0.67 (0.47–0.97) vs. 1.19 (0.79–1.79)	0.025	0.557
rs2327832	A	G	6	TNFAIP3	Primary	0.66 (0.43–1.03) vs. 1.40 (0.85–2.32)	0.017	0.376	0.67 (0.43–1.04) vs. 1.29 (0.78–2.13)	0.038	0.601
rs7020673	G	C	9	GLIS3	SS1	0.06 (0.01–0.27) vs. 0.44 (0.21–0.95)	0.001	0.031	0.03 (0.01–0.19) vs. 0.33 (0.14–0.78)	<0.001	0.018
rs7020673	G	C	9	GLIS3	SS2	0.39 (0.10–1.59) vs. 1.74 (0.57–5.34)	0.046	0.509	0.46 (0.10–2.04) vs. 1.80 (0.56–5.82)	0.090	0.492

The statistical tests were based on Cox proportional hazards models with interaction terms between SNPs (binary variables indicating presence of the allele) and treatment assignment. The size of the interaction effects was measured by the difference in HR (treatment vs. placebo) in participants with and without the allele. P values were based on the Wald test. Results are presented with and without correction for multiple comparisons using the BH method. The risk alleles of the identified SNPs, as reported in the literature, are presented in section 4 of the Supplementary Material along with the corresponding references.

We then examined the association between T1D-GRS2 and the response to oral insulin treatment, both collectively across all strata and separately for each stratum. The results in Table 4 indicate that T1D-GRS2, as a continuous variable, did not significantly relate to the response to oral insulin treatment in the study population (BH-corrected unadjusted P = 0.477; adjusted P = 0.703). We further explored thresholds for T1D-GRS2 and found that when all strata were combined, participants with T1D-GRS >12.5 had better responses to oral insulin compared with those with T1D-GRS2 ≤12.5 (unadjusted P = 0.003; adjusted P = 0.006) (Supplementary Fig. 6). This threshold was derived from systematic evaluation across a range of cutoffs, with 12.5 falling at the midpoint of the empirically defined interval (11.64–12.83) where associations with treatment response were most consistent (additional information provided in section 4 of the Supplementary Material). No significant associations were found between response to treatment and the HLA and non-HLA components of T1D-GRS2 when examining individual strata (Table 4 and Supplementary Fig. 6). In exploratory analyses, we found that T1D-GRS2 was significantly lower in the SS2 than the primary stratum (P = 0.042) and SS1 (P = 0.008); the scores of HLA components in the T1D-GRS2 were lower in SS2 than the primary stratum (P = 0.014) and SS1 (P = 0.011) (Supplementary Fig. 7), while we found no significant differences for the non-HLA component. We note that since the cohort largely comprised White participants (94%), differences in T1D-GRS2 across analysis strata are unlikely due to ancestry stratification.

Table 4.

Effects of T1D-GRS2 and its HLA and non-HLA components on oral insulin treatment for stage 3 T1D prevention

		Unadjusted			Adjusted
Variable and stratum	n	HR (95% CI) at lowest vs. highest quartiles	P	BH-corrected P	HR (95% CI) at lowest vs. highest quartiles	P	BH-corrected P
T1D-GRS2
All	537	1.05 (0.72–1.53) vs. 0.71 (0.51–1.00)	0.064	0.477	0.99 (0.68–1.45) vs. 0.75 (0.53–1.05)	0.167	0.703
Primary	372	1.08 (0.69–1.67) vs. 0.77 (0.51–1.17)	0.177	0.605	0.95 (0.61–1.48) vs. 0.83 (0.55–1.26)	0.586	0.729
SS1	52	0.48 (0.19–1.22) vs. 0.37 (0.14–0.97)	0.653	0.916	0.51 (0.20–1.34) vs. 0.28 (0.10–0.81)	0.353	0.863
SS2	111	0.89 (0.24–3.35) vs. 1.00 (0.36–2.77)	0.870	0.987	0.83 (0.21–3.36) vs. 1.30 (0.45–3.71)	0.543	0.976
HLA component
All	537	1.06 (0.72–1.57) vs. 0.70 (0.49–0.99)	0.065	0.477	1.01 (0.68–1.50) vs. 0.72 (0.51–1.02)	0.137	0.703
Primary	372	1.10 (0.70–1.74) vs. 0.76 (0.50–1.15)	0.147	0.605	0.99 (0.62–1.56) vs. 0.81 (0.53–1.24)	0.465	0.729
SS1	52	0.40 (0.15–1.12) vs. 0.45 (0.18–1.10)	0.874	0.916	0.39 (0.14–1.10) vs. 0.35 (0.13–0.94)	0.889	0.987
SS2	111	0.89 (0.24–3.34) vs. 0.88 (0.32–2.39)	0.987	0.987	0.92 (0.23–3.74) vs. 1.07 (0.38–2.97)	0.833	0.976
Non-HLA component
All	537	0.88 (0.63–1.24) vs. 0.85 (0.60–1.22)	0.867	0.965	0.86 (0.61–1.21) vs. 0.87 (0.61–1.24)	0.983	0.983
Primary	372	0.89 (0.59–1.33) vs. 0.94 (0.62–1.42)	0.803	0.883	0.84 (0.56–1.28) vs. 0.93 (0.61–1.42)	0.662	0.767
SS1	52	0.38 (0.16–0.91) vs. 0.24 (0.08–0.76)	0.390	0.897	0.37 (0.16–0.89) vs. 0.17 (0.05–0.59)	0.178	0.793
SS2	111	0.84 (0.27–2.58) vs. 1.33 (0.41–4.29)	0.520	0.987	0.79 (0.24–2.55) vs. 1.73 (0.52–5.76)	0.281	0.883

The statistical tests were based on Cox proportional hazards models with interaction terms between the variable and treatment assignment, and the variables were continuous in the model. The size of the interaction effects was measured by the difference in HR comparing treatment vs. placebo at the lower quartiles and the upper quartiles of the variable distribution. P values were based on the Wald test. Results are presented with and without correction for multiple comparisons using the BH method.

We also explored the associations between T2D-GRS and pPGSs and the response to oral insulin treatment (Supplementary Table 2). Lower proinsulin pPGSs were associated with better response to treatment in SS1 (unadjusted P = 0.019; adjusted P = 0.027). Lower obesity pPGSs were associated with better response to treatment in SS2 (unadjusted P = 0.037; adjusted P = 0.016).

Conclusions

We analyzed whether genetic factors were associated with response to oral insulin in 560 participants in the TrialNet TN07 study. Our primary analyses focused on the susceptible HLA haplotypes of T1D, SNPs that were previously associated with T1D progression, and T1D-GRS2. After correction for multiple comparisons, GLIS3 rs7020673 remained significantly associated with response to oral insulin in SS1, which is characterized by low FPIR and higher autoantibody positivity. Hypothesis-generating findings with significant uncorrected P values include HLA-DRB1*04:01-DQA1*03:01-DQB1*03:02 in SS1, TNFAIP3 in the primary stratum or all strata combined, GLIS3 in SS1 and SS2, and CTLA4 in SS2. In the exploratory analyses of other genetic factors, we observed better responses to oral insulin among participants with T1D-GRS >12.5.

In a secondary analysis of the TrialNet TN07 study combining all strata, Zhao et al. (18) reported that oral insulin delayed progression to stage 3 T1D among participants carrying HLA-DR4-DQ8 (HR 0.59; P = 0.027). However, this difference was driven by participants with a high IA-2 antigen (IA-2A) level (HR 0.50; P = 0.028), while the effect of HLA-DR4-DQ8 was not significant among those with low IA-2A levels, and HLA-DR4-DQ8 was not a significant factor of response among participants in the primary stratum. In our study, we observed a nominal interaction between oral insulin and HLA-DRB1*04:01-DQA1*03:01-DQB1*03:02 (HR 0.22) in SS1, characterized by low FPIR. Similarly, in the Diabetes Prevention Trial–Type 1 (DPT-1), oral insulin delayed the progression to T1D in individuals with higher IAA (19), which is also associated with HLA-DR4 (20). Moreover, better response to teplizumab treatment for the prevention of T1D has been reported in participants with HLA-DR4 (21). Taken together, these findings suggest that DRB1*04:01-DQA1*03:01-DQB1*03:02 and related markers (e.g., higher IAA titers, multiple autoantibody positivity) may help identify individuals more likely to respond favorably to oral insulin treatment for T1D prevention (22). Prospective studies testing this hypothesis are warranted.

We analyzed 25 genes previously associated with T1D and identified 3 genes with SNPs showing nominal associations with response to treatment. TNFAIP3 has been identified as a risk factor for islet autoimmunity (4), T1D (3,23), and other autoimmune diseases (23). GLIS3 is also associated with risk of T1D (3) and islet autoantibodies (24). The CTLA4 gene encodes cytotoxic T-cell–associated protein 4, an immune checkpoint inhibitor primarily expressed in T cells. (25) This gene has also been found to be related to T1D (26,27), as well as other autoimmune diseases (28).

Higher T1D-GRS2 and T1D-GRS (a previous version, with 30 SNPs) increase the risk of progression to stage 3 T1D in individuals at risk (29,30). However, to our knowledge, whether GRSs can be used to select individuals who will respond to disease-modifying therapies has not been previously reported in the literature. We observed that although the T1D-GRS2 as a continuous variable was not associated with response to oral insulin, exploratory analyses suggested T1D-GRS2 >12.5 as a threshold that predicts better response. Importantly, the 12.5 cutoff was selected because it reflects the midpoint of the empirically defined range (11.64–12.83) where T1D-GRS2 consistently predicted response in TrialNet TN07. The worse response to oral insulin, an antigen-specific therapy, in participants with a decreased T1D genetic burden may suggest that a subset of individuals may develop clinical T1D through a different pathogenic mechanism.

The pPGSs for T2D have been proposed to separate the genetic risk of T2D into biologically meaningful subtypes (31). Given prior evidence on associations between T1D development and obesity (32–35) and T2D-linked genetic factors (36), we explored associations between T2D pPGSs and response to oral insulin for T1D prevention. Among TrialNet TN07 participants in SS1 (i.e., with low FPIR), response to oral insulin improved with higher proinsulin pPGSs, which has been associated with defects in proinsulin-to-insulin processing in T2D. Of note, among individuals at risk for T1D, there is a positive correlation between proinsulin–to–C-peptide ratio and IAA titers (37), which in turn predict response to oral insulin for T1D prevention (19). In SS2, a lower obesity pPGS predicted better response to oral insulin, an antigen-specific therapy, for T1D prevention. Overall, our findings support that metabolic characteristics relevant to T2D may also influence the timing of T1D progression. Furthermore, these metabolic differences could be identified by genetic markers and potentially used to tailor T1D preventive approaches.

In addition to studying genetic associations with response to treatment, we conducted exploratory analyses on genetic differences among study strata. Participants in the SS2 had a lower T1D-GRS2, lower HLA component, and lower frequency of DRB1*04:01-DQA1*03:01-DQB1*03:02 than those in the primary stratum and SS2. Overall, this lower burden of genes typically associated with T1D is consistent with more preserved insulin secretion capacity, as suggested by FPIR above threshold. We also observed differences in the proportions of participants carrying alleles in INS, IL27, and UBASH3A. While the known involvement of INS in the production of insulin (38) may underlie the differences and explain its association not only with T1D (6) but also with T2D (38), additional research is needed to fully understand genetic influences on phenotypic heterogeneity of T1D.

The findings of this study were constrained by the limited sample size in the stratum where differences in response to oral insulin were identified (i.e., SS1). The analysis took on an exploratory approach to examining potential genetic factors that may interact with treatment, while underscoring the necessity for careful interpretation. The identified genetic factors tended to associate with higher disease risk potentially because response differences were more detectable in the higher-risk strata with a greater number of events. Nevertheless, our analyses showed that genetic variation may distinguish individuals who are more likely to respond to treatment for T1D prevention from those who are less likely to respond. Incorporating genetic information holds promise for optimizing T1D-modifying therapies through precision medicine. Importantly, genetic predictors of oral insulin response were detectable specifically within SS1, the subgroup in which the parent TrialNet TN07 trial demonstrated clinical benefit. This pattern supports the view that the strata represent biologically distinct subgroups rather than arbitrary statistical partitions and that genetic effects on treatment response may only emerge when analyzed within an appropriate biological context.

This article contains supplementary material online at https://doi.org/10.2337/figshare.30660524.

Funding Statement

This research was funded by the National Institutes of Health (NIH) through National Institute of Diabetes and Digestive and Kidney Diseases grant R01 DK121843. The Type 1 Diabetes TrialNet Study Group is a clinical trials network currently funded by NIH through the National Institute of Diabetes and Digestive and Kidney Diseases, the National Institute of Allergy and Infectious Diseases, and The Eunice Kennedy Shriver National Institute of Child Health and Human Development, through the cooperative agreements U01 DK061010, U01 DK061034, U01 DK061042, U01 DK061058, U01 DK085465, U01 DK085453, U01 DK085461, U01 DK085466, U01 DK085499, U01 DK085504, U01 DK085509, U01 DK103180, U01 DK103153, U01 DK085476, U01 DK103266, U01 DK103282, U01 DK106984, U01 DK106994, U01 DK107013, U01 DK107014, U01 DK106993, UC4 DK117009-01, and Breakthrough T1D (formerly JDRF).