Amino acid polymorphisms in human histocompatibility leukocyte antigen class II and proinsulin epitope have impacts on type 1 diabetes mellitus induced by immune-checkpoint inhibitors

Introduction Immune-checkpoint inhibitors are effective in various advanced cancers. Type 1 diabetes mellitus induced by them (ICI-T1DM) is a serious complication requiring prompt insulin treatment, but the immunological mechanism behind it is unclear. Methods We examined amino acid polymorphisms in human histocompatibility leukocyte antigen (HLA) molecules and investigated proinsulin epitope binding affinities to HLA molecules. Results and Discussion Twelve patients with ICI-T1DM and 35 patients in a control group without ICI-T1DM were enrolled in the study. Allele and haplotype frequencies of HLA-DRB1*04:05, DQB1*04:01, and most importantly DPB1*05:01 were significantly increased in patients with ICI-T1DM. In addition, novel amino acid polymorphisms in HLA-DR (4 polymorphisms), in DQ (12 polymorphisms), and in DP molecules (9 polymorphisms) were identified. These amino acid polymorphisms might be associated with the development of ICI-T1DM. Moreover, novel human proinsulin epitope clusters in insulin A and B chains were discovered in silico and in vitro peptide binding assays to HLA-DP5. In conclusion, significant amino acid polymorphisms in HLA-class II molecules, and conformational alterations in the peptide-binding groove of the HLA-DP molecules were considered likely to influence the immunogenicity of proinsulin epitopes in ICI-T1DM. These amino acid polymorphisms and HLA-DP5 may be predictive genetic factors for ICI-T1DM.

We have previously identified thyrotropin receptor epitopes to HLA-DR molecules in Graves' disease in silico, in vitro, and human studies (8,9). Further, we have also previously in part examined of histocompatibility leukocyte antigen (HLA) alleles and haplotypes in ICI-T1DM. However, immunological mechanisms in ICI-T1DM are largely unknown due to the rarity of the disease (0.8% prevalence among ICI-treated patients (5).
In the current study, novel amino acid polymorphisms in HLA class II molecules in patients with ICI-T1DM, and in vitro proinsulin epitope binding affinities to HLA-DP molecules were revealed. These genetic factors may be utilized for prediction of ICI-T1DM, and also contribute to elucidate the mechanism of cancer immunotherapy and ICI-T1DM. Therefore, current study offers novel management and monitoring options for cancer immunotherapy.

Patients and methods Patients
Patients were recruited from the Japanese Red Cross Society Wakayama Medical Center (JRCW), Wakayama Medical University Hospital (WMU), and Nagoya University (NU).

Assessment of irAEs and ICI-T1DM
Assessment of irAEs was made based on the descriptions and grading scales of the National Cancer Institute (NCI) Common Terminology Criteria for Adverse Events version 3.0. Diagnostic criteria for T1DM were based on the hyperglycemic symptoms, and continuous requirement of insulin therapy irrespective of autoimmune diabetes-related autoantibodies (11). ICI-controls were defined as those who were treated with ICI but did not develop any irAEs including T1DM (5,10). Healthy Japanese individuals were used as general controls (12)(13)(14), and participants had no clinical or demographic differences and they had the same ethnic background.
In vitro human proinsulin peptides binding assay to HLA-DP5 and HLA-DP15 Fifteen-mer peptides derived from human proinsulin were synthesized based on the predicted affinities to HLA-DP5: 1) strong binders, 2) intermediate binders, and 3) others to cover the entire sequence (PEPscreen ® peptide library, ProImmune, Oxford, UK). Peptides that were known to have high affinities with HLA-DP5 and HLA-DP15 were also synthesized. The synthesized peptides were subjected to an in vitro peptide binding assay (ProImmune REVEAL ® MHC class II-peptide binding assay: (ProImmune, Oxford, UK) using recombinant HLA-DP5 and HLA-DP15 proteins.Detection of binding peptides is based on the presence or absence of the native conformation of the MHCpeptide complex in an immunoassay. Each test peptide was given a 'REVEAL SCORE' relative to positive control peptides, which were known to bind HLA-DP5 or HLA-DP15 with high affinity (signal of known positive control peptide which was known to bind each HLA-DP with high affinity, divided by each test peptide x 100%). The high-throughput assay quantifies the ability of the test peptides to bind to HLA-DP5 and HLA-DP15. The results of the three separate assays were in close agreement and are presented as an average of the results.
Three-dimensional modeling of the HLAclass II molecules Three-dimensional modeling of HLA-class II molecules (HLA-DR, HLA-DQ, and HLA-DP) was downloaded from the Protein Data Bank database (15), and visualized with PyMOL (16).

Statistical analysis
Differences between the two groups were analyzed by a Mann-Whitney U test. The association of allele frequencies was analyzed using Fisher's exact test. Frequencies of HLA alleles and amino acid polymorphisms were analyzed by univariate and multivariate logistic regression analysis with stepwise selection of covariates. Bonferroni test was applied if the variables were significant.
Statistical analyses were performed using JMP, version 15 (SAS Institute Inc., Cary, N.C., USA). P values < 0.05 were considered to be statistically significant.

Clinical characteristics of patients with ICI-T1DM
A total of twelve patients with ICI-T1DM (six patients from JRC, two patients from WMU, and four patients from NU) were identified and enrolled in the study (Tables 1A, 1B; Supplementary  Table 1). Clinical profiles of seven patients with ICI-T1DM were partly described previously (5). Thirty-five independent ICIcontrols (13 from WMU and 22 from NU) were prospectively identified as subjects without irAE and were analyzed. Anti-GAD65 antibody was negative for patients measured. Patient #6 only developed insulin autoantibodies (IAA).

HLA alleles and haplotypes analysis
The plasma glucose levels of patients with ICI-T1DM and HLA typing results are summarized in Table 1B. We have used HLAclass I and II four-digit allelic typing results in the allele or haplotype analysis (Tables 2A-E; Supplementary Table 2A-C). Allele frequencies of HLA-DRB1*04:05 (Table 2A) and HLA-DQB1*04:01 (Table 2B), both alleles in complete linkage disequilibrium, were significantly higher in patients with ICI-T1DM than in general controls and also in ICI-controls. HLA-DPB1*05:01 allele frequency was more significantly associated with an increased risk of ICI-T1DM when compared with general controls and also in ICI-controls (P=0.005 and 0.004, respectively) (Table 2C).
To investigate the importance of amino acid polymorphisms mentioned above in detail (Figures 1-3), a stepwise selection of covariate amino acid residues was applied in multivariate logistic analysis across the HLA-DR, DQ, and DP (Table 3). Subsequently, b205 at HLA-DPb1 was found to be most significant among them.
Differences in amino acids at HLA-DPB1*05:01 and DPB1*15:01 were shown in Supplementary Table 3C and Supplementary Table 6C). Of those, amino acid at positions 85 and 86 compose pocket 1 (P1) of peptide-binding groove on the HLA molecule, and amino acid at position 9 composes P9 ( Figure 4).
Amino acid residues located in the nine peptide-binding grooves (referred to as pockets) were previously described (17,18). Amino acids at positions 9 and 57 of HLA-DRb1 compose P9, and the amino acid at position 86 composes P1 as well ( Figures 1A, B Epitope predictions and in vitro binding of human proinsulin peptides to HLA-DP5 and HLA-DP15 Regarding binding of HLA-DP5 and human proinsulin, binding affinity of the signal peptide (AA 1-24) was predicted to be low   (possessing high %RANK) ( Table 4) ( Figure 6A). In the remaining region (AA 25-110), two epitope candidate regions (AA 43-60 and AA 53-67) were predicted. Five peptides in the regions were strong binders. Other 13 peptides were predicted as intermediate binders.
In addition to the 18 peptides, 16 overlapping peptides were synthesized to cover the whole portion. Subsequently, a total of 34 overlapping 15-mer human proinsulin-derived peptides were synthesized and subjected to in vitro peptide binding assay    (Table 4). Besides, proinsulin epitope binding predictions to HLA-DP15 were similarly shown in Table 4. The proinsulin binding predictions to HLA-DR5 or to HLA-DP15 were different, but all portions in proinsulin including C-peptide region (AA57-87) showed binding predictions to both alleles. Then in vitro peptide binding assay was conducted, and remarkably, in the insulin B chain (AA25-54) and insulin A chain (AA90-110), two clusters were identified (peptides 6-10: AA 29-57 as cluster 1 and peptides 31-34: AA 84-110 as cluster 2) for HLA-DP5 and also for HLA-DP15 (Table 4) ( Figure 6B). Moreover, peptides in the C-peptide region (AA57-87) bound to neither HLA-DP5 nor HLA-DP15.

Discussion
A total of 47 patients with malignancies who had been treated with ICI were subjected to HLA typing by next generation sequencing. T1DM developed in twelve and the remaining thirtyfive served as controls. In allele and haplotype analyses, the patients had an increase of HLA-DRB1*04:05, DQB1*04:01, and in particular of DPB1*05:01. Indeed, 17/24 (71%) alleles were DPB1*05:01 among the patients compared with 24/70 (34%) in the controls. Moreover, significant amino acid polymorphisms at HLA-DR, DQ, and DP allele were identified that might contribute to the development of ICI-T1DM, probably with conformational    alterations in the peptide-binding groove in each HLA-class II molecule. HLA-DP5 was found to be more strongly related to amino acid polymorphisms. As the peptide-binding groove dictate peptide binding, we scanned proinsulin in silico as a hypothetical autoantigen and novel human proinsulin epitope clusters in insulin B and A chains were discovered for HLA-DP in vitro. Our findings are promising suggestion of a possible association between HLA and ICI-T1DM through proinsulin peptide binding studies.
In comparison with our previous study (5), the current study was conducted on a larger scale with confirmation of HLA-DPB1*05:01 predominance in ICI-T1DM than those of general controls and ICIcontrols with the same ethnic background. Current study may also support other reports describing that HLA-DR4 alleles are increased in patients with ICI-T1DM in the United States (21, 22), although these studies are not at the detailed allelic levels. Based on the conditional multiple regression analysis, in addition to HLA-DPB1*05:01 allele, alleles of HLA-DRB1*04:05 and DQB1*04:01, and HLA-DRB1*04:05-DQB1*04:01 haplotype were found to be susceptible to ICI-T1DM (Tables 2A-E) Figures 5A-C). These amino acids contribute to the formation of the peptide-binding grooves on HLA-class II molecules, therefore may functionally contribute to the epitope presentations. Notably, patients with ICI-T1DM in Japan have been reported to have HLA-DRB1*11:01/*13:02 (23), DRB1*04:05 (24-26), DRB1*04:06 (26), and DRB1*09:01 (25), and all of patients above had G86 at HLA-DRb1. Particularly, both HLA-DRB1*04:05 and DRB1*09:01 have G86 at HLA-DRb1, and were reported to be associated with T1DM (27). Conversely, HLA-DRB1*15:01, a protective allele for T1DM has V86 at HLA-DRb1 (27), and HLA-DRB1*15:01 was not seen in patients with ICI-T1DM in the current study. Therefore, G86 and V86 seemed to be diseasepromotion and protection alleles, respectively, at HLA-DRb1, in association with (P1). Todd, et al. reported strongly conserved Asp (D) at position 57 (P1, P9) of HLA-DQb1 with disease susceptibility in patients with T1DM (28), and D at position 57 (P1, P9) of HLA-DRb1 seemed to play a strong protective role in ICI-T1DM in the current study ( Figure 5A). Then, stepwise selection and multivariate analysis revealed that b205 at HLA-DPb1 was most important   ( Table 3). Therefore, we speculated the amino acid polymorphism b205 at HLA-DPb, which locates outside of the peptide-binding groove in HLA-DPb1, may be related to interactions such as HLA-DM, or to alterations in signal transduction within the HLA molecule. Taken together, significances of HLA-DP5 in ICI-T1DM were observed. Next, we conducted human proinsulin peptide binding prediction to HLA-DP molecules to ensure effective epitope presentation in ICI-T1DM (Table 4) ( Figure 6A). Predicted binding affinities and in vitro binding results were quite different for HLA-DP5, especially in the C-peptides region, for reasons unknown (Figures 6A,  B). Remarkably, insulin B chain epitope (cluster 1): AA 29-57, and the insulin A chain epitope (cluster 2): AA 84-110 were established in vitro binding assay ( Figure 6B). Among them, peptide 9 (AA 42-56), VCGERGFFYTPKTRR (core sequence underlined), was predicted as intermediate binders and also exhibited the strongest in vitro peptide binding, thus is mostly expected to be immunogenic T-cell epitope. Insulin peptide B9-23 (AA 33-47) has been reported to be a major autoantigen to induce immunity in the nonobese diabetic mouse, and is also included in the cluster 1 (19). Mannering et al. reported that cells transfected with HLA-DRB1*04:05 presented insulin A1-13 (AA 88-102) which is the identical to the peptide 32 (20). HLA-DP15 has not been reported as a risk allele in ICI-T1DM, but showed similar in vitro proinsulin epitopes to HLA-DP5 (cluster 1 and cluster 2) ( Figure 6B). Despite the dissimilarities in amino acid sequence between HLA-DPB1*05:01 and HLA-DPB1*15:01, the importance of epitope cluster 1 and cluster 2 was reinforced across the alleles in silico and in vitro (Supplementary Table 6C).
In the development of ICI-T1DM, we hypothesized that the inhibition of immune-checkpoint molecules by ICI induced immunity to pancreatic b-cells, as observed in thyroid follicular epithelial cells during thyroid irAE (5,29). Antibody-dependent cellular cytotoxicity by ICIs, by cytotoxic T-cells (30), or by both of them, would contribute to the development of ICI-T1DM. We speculate that ICI-T1DM-predisposing HLA may also be involved with malignant diseases. Proinsulin epitope, as well as tumorassociated antigen/neoantigen, could be bound to peptide-binding cleft of HLA, and cross-presented on the surface of antigenpresenting cells due to molecular mimicry (3,4,9). Another important topic is the correlation of ICI treatment effectivity and ICI-T1DM. Considering that 8/12 (75%) were ICI-responders in ICI-T1DM group and that generally only 20-30% are responders, common mechanisms between ICI treatment and ICI-T1DM were suggested. Therefore, HLA seemed to be associated with both ICI treatment outcome and risk for an endocrine adverse event.
Considering that proinsulin is processing into insulin A chain and B chain, and that epitope clusters were identified in the two chains, evaluation of IAA in the time course may be of interest. As well, anti-GAD65 antibody titers in the course seem to be important.
This study has limitations. Firstly, we have tested proinsulin binding to HLA-DP. Binding studies for HLA-DR and HLA-DQ molecules are also desirable. Secondly, more evidence of immunogenicity of HLA-DP5 from in vitro and in vivo experiments in comparison with HLA-DR/DQ are preferable to confirm the peptide binding data in this study. Thirdly, better investigation of the frequency of HLA-DPB1*05:01, irrespective of ICI treatment is still desirable. Finally, the study consists of a small A B C  Peptide-binding grooves in an HLA class-II molecule (amino acid position 1-9) are shown. Amino acids in positions 1, 4, 6, 7, and 9 bind to HLA and those in positions 2, 3, 5, and 8 are assumed to be outward facing in order to stimulate the T-cell receptor (TCR).   sample size, so more data with increased numbers may help to establish our results.
In conclusion, HLA-DP5 as a predisposition molecule, and significant amino acid polymorphisms at HLA-class II molecules in patients with ICI-T1DM were established in this study. Based on the silico and in vitro proinsulin peptide-binding study, conformational changes in the peptide-binding groove of the HLA-DP molecules may influence the immunogenicity of proinsulin epitopes in ICI-T1DM. These genetic factors may be utilized for prediction of ICI-T1DM, and also could contribute to elucidation of the mechanism of cancer immunotherapy and ICI-T1DM. Therefore, current study offers safer and more effective management and monitoring options for cancer immunotherapy. Further investigations are warranted to elucidate the relation of ICI-treatment effectiveness and development of ICI-T1DM.

Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.

Ethics statement
The studies involving human participants were reviewed and approved by Red Cross Society Wakayama Medical Center (JRCW), Wakayama Medical University Hospital (WMU), and Nagoya University (NU). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

Author contributions
HI: Conceptualization, designed the study. original draft preparation. SM: Visualization, writing-reviewing and editing. All authors performed the clinical studies statistically analyzed the data and wrote the manuscript. All authors approved the final version of the manuscript for publication. All authors contributed to the article and approved the submitted version.

Funding
This work was partially supported by the Takeda Science Foundation, the Japanese Foundation for Multidisciplinary Treatment of Cancer, Japan Diabetes Foundation, and the Grantin-Aid for Scientific Research from the Ministry of Education, Science, Sports, and Culture, Japan (20K17541 and 21K08544). Pharmaceutical Co., Ltd., MSD K.K., and Chugai Pharmaceutical Co., Ltd., and personal fees from Ono Pharmaceutical Co., Ltd., Bristol-Myers Squibb and MSD K.K. outside of this study.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.