HLA Class II (DR, DQ, DP) Genes Were Separately Associated With the Progression From Seroconversion to Onset of Type 1 Diabetes Among Participants in Two Diabetes Prevention Trials (DPT-1 and TN07)

Type 1 diabetes (T1D) is an autoimmune disease (1), and its development typically runs from seroconversion (stage 1), to hyperglycemia (stage 2), to onset of symptomatic T1D (stage 3) (2). The progression phase, the focus of this investigation, includes the three stages from the seroconversion to T1D onset, during which the β-cell function quantitatively deteriorates and dysglycemia develops, over highly variable periods extending from months to years (2). It has been hypothesized that genetic mechanisms underlying initiation and progression differ (3,4). Given their retrospective nature, most T1D genetic studies using stage 3 T1D patients are unable to differentiate between initial or progression genetic mechanisms, but have clearly demonstrated that DQA1/DQB1 genes are the main genetic contributors (5–8), together with DRB1, DRB3, DRB4, DRB5 (9–12), and DPA1/DPB1 (13–15). With the recent completion of a large birth cohort study, The Environmental Determinants of Diabetes in the Young (TEDDY), which entailed frequent follow-up collections of biospecimens and clinical data on over 8,500 participants for nearly 15 years, it is now possible to begin to distinguish genetic mechanisms associated with initiation and progression stages separately (16,17). Recently, TEDDY investigators reported that HLA class II genes (HLAII) play important roles in the occurrence of seroconversions (17–19); that is, DQ, DR, and DP contribute to disease susceptibility and to disease onset (seroconversion). An obvious follow-up question, which has not been studied extensively to date, is whether HLAII genes play any role in disease progression.

In that regard, one genetic association study, using archived clinical and genetic data from the Diabetes Prevention Trial–Type 1 (DPT-1) clinical trial (20), found roles of DQB1 alleles; individuals with DQB1*03:02 and DQB1*03:01 tended to be faster and slower progressors, respectively (20). Following up on their finding, we genotyped participants in DPT-1 to obtain high-resolution HLA allele types. Besides confirming their findings, our work drilled into DQB1 molecular structure and uncovered two amino acids within DQB1 (−18β, β57), in which −18β could take the amino acid A or V and β57 could have A or D amino acid. Two amino acids jointly form motifs (e.g., DQB1*VA, DQB1*AD) that explain dual faster or slower progression, respectively (21). In another related study of DPT-1 participants, the 5-year risk of progression according to DQ types confirmed earlier reports of strong protection afforded by the DQB1*06:02 allele among autoantibody-positive relatives (22).

The HLA system is critically important to host immunity, and further investigation could provide insights into genetic mechanisms useful for developing secondary prevention strategies. DQ, DR, and DP contribute to host immunity, but their roles in T1D progression are largely unexplored. Hence, determining their contributions to disease progression, in addition to well-documented DQ associations, would enhance our knowledge of the genetic factors that contribute to T1D risk and disease progression. To optimize the power of this investigation, we accessed DNA samples from participants in the Randomized Diabetes Prevention Trial with Oral Insulin (TN07) (23) and investigated roles of class II genes in disease progression with pooled TN07 and DPT-1 samples, forming an integrated cohort (iCohort). Additionally, the larger sample size enables us to explore heterodimeric associations of class II genes and their possible interactions with insulin, age, and risk level.

The DPT-1 study was the first clinical trial to assess efficacy of oral insulin for preventing T1D or slowing disease progression among participants at moderate risk and that of parenteral insulin among high-risk participants (20,24). While overall results were inconclusive, it was noted that participants on oral insulin tended to exhibit delayed progression, even though the improvement was not statistically significant (25). Encouraged by this initial result, TrialNet sponsored the follow-up randomized trial on oral insulin (TN07) (26). The follow-up trial again revealed an insignificant improvement regarding disease progression (26). Participants in both DPT-1 and TN07 trials had a family history of T1D and elevated autoantibody levels, in which nearly half of the participants in DPT-1 were projected to have more than a 50% risk of developing T1D. From both trials, investigators have collected and archived extensive clinical trial data as well as extensive collections of biospecimen samples, and are making them available for additional studies through the National Institute of Diabetes and Digestive and Kidney Diseases Central Repository portal (https://repository.niddk.nih.gov/home/).

The following clinical and demographic factors—race, sex, age, treatment type, risk level, and source of data—may confound the association analysis of HLA genes and progression, and, if so, should be adjusted as confounding variables. The current study includes 670 and 546 participants from DPT-1 and TN07, respectively, totaling 1,216 participants (Supplementary Table 1). Nearly half of the participants were treated with insulin during the studies, and 25% of them were at high risk. Demographically, most participants were Caucasians, and there were 200 more male participants than females. While most participants were younger than 20 years old, the rest were distributed across ages 20 to 46 years, and age distributions across six subgroups were largely comparable (Supplementary Fig. 1).

All participants were followed until the clinical diagnosis of T1D or termination of the trial, and progression from islet autoimmunity to T1D was quantified through a censored time-to-T1D. Incidences in the six groups combinatory with two risk levels, treatment/placebo status, and two clinical trials were computed (Supplementary Fig. 2). Associations of all risk factors with time-to-T1D were evaluated through both univariate (left panel of Supplementary Table 1) and multivariate (right panel of Supplementary Table 1) Cox regression models, and results suggested that risk level and age were significantly associated with accelerated and decelerated progression, respectively, as expected (Supplementary Table 1).

HLA typing was carried using the ScisGo HLA v6 typing kit (Scisco Genetics Inc., Seattle, WA) following the kit protocol. Briefly, the method uses an amplicon-based two-stage PCR, followed by sample pooling and sequencing using a MiSeq v2 PE500 (Illumina, San Diego, CA). The protocol yielded three-field coverage of HLAII genes (DRB1, DRB3, DRB4, DRB5, DQA1, DQB1, DPA1, and DPB1). Phase within each gene was determined, in part, by bridging amplicons and, when not available, by database lookup HLAII (27).

DQA1 and DQB1 are in strong linkage disequilibrium (LD), which is used to infer their haplotypes, referred to as DQA1-DQB1. Similarly, DPA1 and DPB1 are haplotyped as DPA1-DPB1. Because of evolutionary reasons, DRB1 alleles shared one of either DRB3, DRB4, or DRB5, and such empirical “LD” also allows for haplotyping, denoted generically as DR. Statistically, haplotyping DQA1-DQB1, DPA1-DPB1, and DR is highly accurate for nearly all samples. Among eight cases with ambiguously inferred haplotypes (six cases with posterial probability of 0.96 and two cases with less than 0.95 posterial probabilities), haplotypes with the highest higher confidence (posterial probabilities) were chosen for further analysis.

All analyses in this project relied on statistical functions and packages in R. For haplotyping genes, we used the haplo.em function from the R package haplo.stats. Details on statistical analyses and significance assessments are included in the Supplementary Material.

Earlier, with DPT-1, we carried out the association analysis with high-resolution DQA1-DQB1 genotypes and confirmed the previous association results with DQB1*03:02 and DQB1*03:01 (20). We repeated the same association analysis with resequenced DQA1-DQB1 genotypes in the iCohort, effectively doubling the sample size. As expected, DQA1-DQB1 haplotype DQA1*03:01-DQB1*03:02, with 822 copies, was found to significantly increase the progression risk (hazard ratio [HR] = 1.25, P = 0.0035), while DQA1*03:03-DQB1*03:01 delayed progression (HR = 0.54, P = 0.00116) (Supplementary Table 2). Interestingly, perhaps due to the larger sample size, the haplotype DQA1*05:01-DQB1*02:01 (commonly known as DQ2.5) was also found to have elevated progression risk (HR = 1.19, P = 0.04). Further exploring their genotypic associations, homozygous and heterozygous genotypes of haplotype DQA1*03:01-DQB1*03:02 have comparable hazard ratios (HR = 1.44 and 1.37, P = 0.04 and 0.0037, respectively), suggesting possible recessive associations. On the other hand, the association with DQ2.5 appeared to be in line with the additive effect for heterozygote and homozygote (HR = 1.21 and 1.34, P = 0.06 and 0.23, respectively). Association results of all DQ haplotypes are listed in Supplementary Table 2. To assist the interpretation of these results, the data in Fig. 1 show cumulative incidence curves for homozygous and heterozygous individuals with corresponding haplotypes, along with participants without specific haplotypes (Fig. 1A–C, respectively). Clearly, heterozygous and homozygous individuals for DQA1*03:01-DQB1*03:02 have higher incidence curves (green and red, respectively, in Fig. 1A) than the curve of noncarriers (black curve in Fig. 1A). Similarly, heterozygous and homozygous individuals for DQA1*05:01-DQB1*02:01 have their incidence curves (green and red curves, respectively, in Fig. 1B) slightly higher than the incidence curve of noncarriers (black curve in Fig. 1B), and homozygous carriers appear to have higher incidence rates than heterozygous carriers. Lastly, heterozygous and homozygous individuals with DQA1*03:03-DQB1*03:01 have lower incidence rates than noncarriers, and homozygous carriers appeared to have even lower incidence rates, although the sample size of eight participants with one observed event is too small to draw a meaningful conclusion. Note that counts of observed haplotypes are computed (Supplementary Table 2) and can be used to compute haplotypic frequencies by dividing counts over their total. However, haplotype frequencies cannot be taken as estimates in the general population, since participants in clinical trials were highly selected by their family history and other risk factors.

As noted earlier, DRB3, DRB4, and DRB5 are distinct genes, and each locus is linked to a specific DRB1 allele (see Table 1 for a list of DR haplotypes). For example, DRB1*03:01 is in LD with either DRB3*01:01 or DRB3*02:02 but never both. Also, some DRB1 alleles are not found in linkage with any of these three homologous genes, noted as Null in the DR haplotypes. Applying the Cox regression with an adjustment for risk level and age, we assessed DR associations with disease progression, and found that individuals with haplotype DRB1*03:01-DRB3*02:02 tend to be faster progressors (HR = 1.59, P = 0.001) than those with DRB1*13:02-DRB3*03:01, who tended to be slower progressors (HR = 0.63, P = 0.023) (“Without adjusting DQ” in Table 1). To account for LD between DR and DQ, we adjusted DQ polymorphisms, and repeated the same DR association analysis (“With adjusting DQ” in Table 1). The adjusted DR association analysis still led to risk association with DRB1*03:01-DRB3*02:02 (HR = 1.44, P = 0.019), and, interestingly, the haplotype DRB1*03:01-DRB3*01:01 was now associated with the protective association (HR = 0.73, p-0.04). In addition, the adjusted association analysis suggested that DRB1*12:01-DRB3*02:02 and DRB1*01:03-Null appeared to have marginal risk associations (HR = 2.60 and 2.32, P = 0.018 and 0.039, respectively).

While genetic roles of DP in T1D have not been established as well as those for DQ or DR, it is of interest to investigate DP association with progression. Assessing immunogenetic associations of DPA1-DPB1 with T1D progression with and without accounting for DQA1-DQB1 polymorphisms are listed in Table 2, as “Without adjusting DQ” and “With adjusting DQ,” respectively. Without the DQ adjustment, individuals with DPA1*01:03-DPB1*03:01 in 282 observed DP haplotypes tended to be slower progressors (HR = 0.71, P = 0.007 in “Without adjusting DQ” in Table 2). Evidently, this protective association was independent of DQ, since the adjusted association retained a comparable association (HR = 0.70, P = 0.006). Interestingly, the DQ-adjusted analysis identified one additional DPA1*02:01-DPB1*11:01 haplotype that was marginally significant with respect to its risk associations with disease progression (HR = 1.98, P = 0.021), although it had a relatively low haplotype frequency, with 29 observed haplotypes in the iCohort.

aired class II genes, such as DQA1 and DQB1, are transcribed and translated to form a heterodimer with α- and β-chains, respectively. Since every participant has two haplotypes, these two haplotypes could form up to four distinct heterodimers if both DQA1 and DQB1 genotypes are heterozygous (provided there are no structural restrictions to αβ pairing). For example, DQA1*03:01-DQB1*03:02 (DQ8.1) and DQA1*05:01-DQB1*02:01 (DQ2.5) could form four different heterodimers: DQA1*03:01-DQB1*02:01 and DQA1*05:01-DQB1*03:02 in addition to empirically observed DQ8.1 and DQ2.5. For clarity, to contrast with naturally observed DQA1-DQB1 haplotypes, somatically recombined haplotypes DQA1*-DQB1* are referred to as somatic haplotypes (Table 3). As expected, some somatic haplotypes are the same as empirical haplotypes. Since somatic haplotypes of DQ and DP are identified by somatic recombination, some of them are structurally prohibited and are thus not observed. For our current evaluation, we retained them as they are, and systematically evaluated their associations with disease progression.

While a full list of DQ empirical and somatic haplotypes and their association results are listed in Supplementary Table 3, we extracted three haplotypes (DQ8.1, DQ7.3, DQ2.5) and their six somatic recombinants (top nine rows in Table 3) for closer analysis. While DQ8.1 and DQ7.3 retain their respective risk and protective associations in the heterodimeric association analysis, DQ2.5 became marginally insignificant, as somatic recombination increased 576 observed haplotypes to 665 heterodimers (HR = 1.10, P = 0.079). Interestingly, two somatic haplotypes, DQA1*03:01-DQB1*02:01 and DQA1*05:01-DQB1*03:02, resulting from somatic recombination from DQ8.1 and DQ2.5 were significantly associated with progression (HR = 1.32 and 1.34, P = 0.013 and 0.0055, respectively). On the other hand, somatic haplotypes DQA1*03:01-DQB1*03:01 and DQA1*03:03-DQB1*03:02, resulting from somatic recombination of DQ8.1 and DQ7.3, became insignificant (HR = 0.94 and 1.08, P = 0.754 and 0.502, respectively), as if risk and protective associations with two haplotypes cancel each other out. Lastly, the somatic haplotype DQA1*05:01-DQB1*03:01, resulting from somatic recombination of DQ2.5 and DQ7.3, has significant protective association (HR = 0.42, P = 0.0362), and the haplotype DQ6.4, with additional heterodimers from somatic recombination, appeared to be significantly protective (HR = 0.68, P = 0.049).

Paired DPA1 and DPB1 also form a heterodimer with α- and β-chains. While DPA1-DPB1 haplotypes are empirically observed, their somatic recombination leads to additional DP haplotype DPA1*-DPB1* (see Supplementary Table 4 for a complete list of both empirical and somatic haplotypes). Here we examined one empirical haplotype, DPA1*01:03-DPB1*03:01, that has marginal significance with protective association (HR = 0.85, P = 0.017). Interestingly, two somatic recombinants, DPA1*01:03-DPB1*124:01 and DPA1*01:04-DPB1*04:01, were significantly associated with faster progression (HR = 1.78 and 3.07, P = 0.005 and 0.004, respectively).

DRB1 allele groups are in linkage with one of the DRB3/B4/B5 genes or with none, forming empirical DR haplotypes (28). For heterozygous DR haplotypes, they could be somatically recombined to form two somatic recombinants. Following the same analytic process, we assessed associations of both empirical and somatic DR haplotypes with disease progression (see Supplementary Table 5 for a complete list). Examining only those DR haplotypes with P values less than 0.05, we found that one somatic haplotype, DRB1*03:01-DRB4*01:03, was associated with faster progression (HR = 1.26, P = 0.027), and another somatic haplotype, DRB1*10:01-Null, i.e., by DRB1*10:01, was protective, as none of these progressed to T1D. However, DRB1*10:01 was relatively uncommon.

The relatively large sample size of the iCohort allows us to explore immunogenetic associations with T1D progression in subgroups, namely, participants in placebo versus in treatment groups (Supplementary Fig. 3), young (<10 years old) versus older participants (Supplementary Fig. 4), and low- versus high-risk participants (Supplementary Fig. 5). In subgroup analysis, we evaluated haplotype-specific associations within each subgroup with estimated coefficients (log HR) and their 95% CIs, which are then displayed as paired red and green lines. Separately, we evaluated whether the differences of coefficients between two groups are statistically significant, and those less than 0.05 are flagged. For example, in the subgroup analysis of placebo versus insulin treatment (Supplementary Fig. 3), most haplotype-specific coefficients between placebo and insulin coefficients overlap, except that those individuals with DQA1*05:05-DQB1*03:01 and DRB1*11:01-DRB3*02:02 tended to progress faster when taking insulin (P = 0.0219 and 0.015, respectively). However, such associations could be attributed to an inflated type I error rate with multiple comparisons. Similar results are observed for subgroup analysis with age and risk level.

Through assessment of immunogenetic associations with class II genes, we identified three protective haplotypes (DQA1*03:03-DQB1*03:01, DRB1*13:02-DRB3*03:01 and DPA1*01:03-DPB1*03:01). As expected, individuals with such protective haplotypes tend to be slower progressors (green, red, and blue lines in Fig. 1D), relative to the general incidence curve (gray line), but are similar to individuals with any protective haplotypes (brown line). These results suggest that, when designing future preventive clinical trials, exclusion of slower progressors could be considered as a way to improve trial efficiency. When excluding known slow progressors, the incidence curve for the remaining participants (black line in Fig. 1D) would be higher than the overall incidence curve (gray line in Fig. 1D). To achieve the anticipated incidence by year 4 of the trial (vertical dotted line in Fig. 1D) with ∼35% incident cases, a trial, excluding known slow progressors, could achieve 35% incident cases earlier (the left dotted vertical line in Fig. 1D).

This study explores the DR, DQ, and DP linked to progression to T1D in 1,216 participants, positive for islet autoimmunity and participating in two disease prevention studies, DPT-1 (20) and TN07 (23). It reproduces the findings of previous studies regarding alleles predisposing to susceptibility and prevention of T1D, while it also reveals some new associations. Progression and prevention here must be viewed in the context of already established islet autoimmunity as well as receiving a treatment for expected disease prevention, or a placebo.

There are no surprises in the DRB1 linkages to T1D progression or resistance thereof (Table 1). However, the extended DRB1-DRB3/B4/B5 haplotype holds a few surprises, both when adjusted for age and risk level and when adjusted to DQ, age, and risk level (Table 1). For one, an earlier report (29), showed that the extended haplotype DRB1*03:01-DRB3*02:02 is linked to susceptibility to T1D, without differentiating between initiation and progression. The selected participants in the iCohort allowed us to conclude that DRB3*02:02 conferred risk when in linkage to DRB1*03:01 or to DRB1*12:01. The conclusion is that DRB3*02:02 contributes to the disease progression. Furthermore, the extended haplotype DRB1*03:01-DRB3*01:01 is linked to protection from T1D (Table 1). This is indeed revealing, as the susceptibility of DRB1*03:01 (DR3 in the abbreviated nomenclature) to T1D seems dependent on the linkage to particular DRB3 alleles, with DRB3*01:01 conferring protection, a finding that was apparent also after correction for DQ, age, and risk level.

Despite the established correlation with susceptibility of DRB3*02:02, there have been no CD4⁺ T cell reactivities to any peptides of the known autoantigenic proteins linked to T1D (PPI, GAD65, IA-2, and ZnT8) restricted to this particular DRB3 molecule (29). The amino acid sequences of HLA-DR proteins and their key differences, especially in the antigen-binding β1 domain, are shown in Supplementary Fig. 6. By contrast, DRB4*01:01, in linkage with DR4 alleles, several of which are susceptible to T1D, shows considerable reactivity to T1D peptides arising from the signal sequence of PPI (30).

The DQ molecules predisposing to progression to T1D (listed in Table 3) are essentially the ones found to predispose to the condition in the general population (DQA1*03:01-DQB1*03:02, DQA1*05:01-DQB1*03:02, DQA1*03:01-DQB1*02:01, and DQA1*05:01-DQB1*02:01, albeit with the latter not achieving statistical significance). Of these molecules, the second and the third are encoded in trans; that is, each of the α- and β-chains are derived from different homologous chromosomes, and significantly elevate progression risks (HR = 1.34 and 1.32, P = 0.006 and 0.01, respectively). By contrast, the DQA1*03:03-DQB1*03:02 and DQA1*03:03-DQB1*02:01 haplotypes, which are somatic recombinations from DQ.81 and DQ2.5 and share the same α-chain, have an HR around 1.00 (P = 0.50 and 0.64, respectively). On the other hand, considerable protection from T1D was seen in various DQ7 serotype haplotypes such as DQA1*03:03-DQB1*03:01, and DQA1*05:01-DQB1*03:01, but not DQA1*03:01-DQB1*03:01. In addition, the DQ6 serotype molecule DQA1*01:02-DQB1*06:04 shows near-comparable protection from T1D. It binds to proinsulin peptides with similar or slightly lower affinity than DQA1*01:02-DQB1*06:02 (31).

The correlations of progression to T1D with different DPA1-DPB1 alleles (Table 2) are similar to a previous publication (32). It should be noted that a recent publication of peptides extracted from the groove of DP heterodimers has revealed bound peptides both in the classical forward orientation (amino terminus, N, near the α52-β90 interphase, and carboxy terminus, C, near the α76-β57 interphase; see designations in the graphical abstract) and in the reverse orientation, that is, C terminus where the N terminus was situated and vice versa (33). The DPA1*02:01-DPB1*11:01 heterodimer, associated with progression to T1D, was found to be loaded with peptides in the forward orientation in 87% of the extracted peptides, with the remaining 13% being in the reverse orientation (33). By contrast, the DPA1*01:03-DPB1*03:01 heterodimer that is associated with resistance to progression to T1D was found to be loaded with peptides in the forward orientation in 98% of the extracted peptides (33). Unfortunately, there is a lack of T1D-related epitopes linked to any DP molecules (32). As there are peptides in the reverse orientation in the groove of the T1D-susceptible DPA1*02:01-DPB1*11:01 molecule, future research needs to take into account that T1D-associated peptides to be discovered may be presented in the reverse orientation.

The association between distinct DR, DQ, or DP heterodimers and progression to diabetes suggests that HLAII may contribute to the pathogenesis by virtue of the ability of the respective protein products to present antigen. For example, it may be hypothesized that the pathogenesis toward clinical onset is driven by a sustained presentation of insulin peptides on DQA1*03:01-DQB1*03:02 or of GAD65 peptides on DQA1*05:01-DQB1*02:01. However, an alternative explanation needs to be sought, as data in the TEDDY study suggest that it is only the first appearing islet autoantibody, but not the subsequent second or additional autoantibodies, that is associated with HLA (3,34). The data rather suggest that the HLAII association with the first appearing autoantibody was indeed also related to the subsequent progression to clinical onset. The TEDDY study has shown that insulin autoantibody as the first appearing autoantibody was related to a more rapid progression compared with GAD autoantibody-first (34). These findings were recently supported in the Finnish Diabetes Registry, indicating that the heterogeneity of T1D is associated with HLAII and the first appearing islet autoantibody (35).

The current investigation takes an analytic strategy of pooling DPT-1 and TN07, rather than the conventional genome-wide analytic strategy (discovery, replication, and pooling). The primary motivation is that our investigation centers on HLAII genes that are known to be immunogenically functional, and highly polymorphic with many alleles; each allele combination potentially corresponds to a functional molecule, and the analytic powers of our approach are directly influenced by sample sizes. Further, an immunogenetic association with an HLAII molecule observed in one study population tends to be similar to that in another study population, unless we assume that differential gene-environmental interactions between studies lead to heterogeneous associations. To verify this assumption, we performed stratified association analyses separately within DPT-1 and TN07, and tested whether differences in their log hazard ratios equal zero, the results of which indicate that allele-specific differences are not statistically significant from zero, i.e., that there is no evidence for heterogeneous DQ, DR, and DP associations with the progression (not shown).

While identifying progression haplotypes, this investigation has identified three resistant haplotypes (DQA1*03:03-DQB1*03:01, DRB1*13:02-DRB3*03:01, and DPA1*01:03-DPB1*03:01), and their carriers seem to be associated with slower progression rates. While these associations may prompt mechanistic investigations in the future, they could be readily used as an additional exclusion criterion. Given the estimated incidence curves in DPT-1 and TN07 trials, one could achieve the same statistical observations in a shorter clinical trial. Of course, adopting this exclusion criterion would require financial resources and time to genotype and recruit. The net benefit requires a balanced consideration of both gains and losses from analyzing genetic frequencies in the general population. Similarly, one could consider high-risk haplotypes as an inclusion criterion, which may improve clinical trial efficiency, especially for investigating immunotherapies targeting at HLAII genes.

Acknowledgments. The authors would like to express appreciation to the steering committee of TN07 in TrialNet for approving our request to access DNA samples and to thank NIDDK Central Repository for providing clinical data from both DPT-1 and TN07. Equally, they would like to thank reviewers whose comments have substantially improved the presentation of this work.

Funding. The study was supported by a grant (R01 DK132406) from National Institutes of Health/National Institute of Diabetes and Digestive and Kidney Diseases.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. L.P.Z., G.K.P., and Å.L. researched and analyzed the data and wrote the manuscript. J.S.S. and A.P. were investigators who conducted the DPT-1 trial. H.M.P. engaged in reviewing data and analysis. W.W.K. reviewed the manuscript and expanded on implications of the findings in the context of DRB3 binding peptides. T.P.L. reviewed and commented on the manuscript. D.E.G. led the team of R.W., C.-W.P., and W.C.N. in the next-generation sequencing, and researched data. G.P.B. and A.K.M. carried out graphical representations of select DR molecules and contributed to the Discussion section of the manuscript. L.P.Z. and Å.L. are guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Jennifer E. Posey.