Key findings
- 0.023 → 0.487For HLA-A*02:01 the predicted probability of strong binding at the mutant SLC16A11 peptide rose from 0.023 for the wildtype peptide to 0.487 for the early-human glycine mutation.
- 0.014 → 0.207HLA-A*02:06 showed a smaller binding polymorphism at the same site, from 0.014 for the aspartic-acid peptide to 0.207 for the glycine mutation.
- R² = 0.975In the joint HLA-A*02:01, A*02:06 and SLC16A11 polygenic model, the weighted linear regression of the odds ratios across risk-allele categories was statistically significant at the <BMI 31 stratum (p = 0.0017).
- ~90% / 36%About 90% of this indigenous population carries at least one risk allele, while 36% have type 2 diabetes.
Abstract
The purpose of this work is to determine the relationship between the HLA-A alleles and SLC16A11 and estimate their risk/protection for type 2 diabetes mellitus (T2D) in a large indigenous American sample. The method uses binding prediction algorithms to identify risk 9mer peptides for the protein and develops polygenic statistical models that reflect the joint action of the two loci. The strength of peptide binding was estimated for the mutant form of the SLC16A11, T-C-G-T-T, with genotypes *11, *12, *22, risk allele = 2. Among the four polymorphic HLA-A alleles, binding polymorphisms were discovered for A*02:01 and A*02:06 at SLC16A11 missense site D127G. A binding polymorphism is defined as one in which the wildtype-consensus peptide has minimum binding while the mutation exhibits strong binding. For HLA-A*02:01 the binding double was [0.023, 0.487], respectively, while for A*02:06 it was [0.014, 0.207]. When A*02:01 genotypes were put in a logistic sensitivity analysis in <BMI kg/m2 strata 25-45, no significant associations were observed. When SLC16A11 genotypes were added, with interaction, the odds ratios for risk allele 2 genotypes were significantly increased. A new, polygenic logistic sensitivity analysis was created that generalized the risk alleles for the two loci (A*02:01, A*02:06, and T-C-G-T-T) with five risk allele (RA) categories, 4, 3, 2, 1, 0. The most significant stratum was <BMI 31 for risk categories 4 (OR=3.33, statistical power=0.819), 3 (2.64, 0.933), and 2 (1.91, 0.722). A statistically significant weighted linear regression across RA categories was OR = 0.583RA + 0.811 with R2 = 0.975. The binding polymorphisms, polygenic analysis, linear regressions, and power estimates lead to the conclusion that HLA peptide presenting molecules need a tripartite statistical model to test associations: disease, HLA allele, and target protein peptide.
Introduction
Type 2 Diabetes Mellitus (T2D) is a complex disease with both environmental and genetic components that is prevalent in indigenous Americans throughout the New World and has its largest manifestation in the desert southwest of the United States.1,2 While the clinical risk factors of increasing age, sedentary lifestyle, and obesity are well recognized, teasing out the genetic components of T2D has been more difficult. Many genome wide association studies (GWAS) have been performed searching for a clear genetic mechanism and locus. A recent review of a multi-ethnic, meta-analysis revealed a large array of single nucleotide polymorphisms (SNPs) and their associated loci.3 The current model is that this complex disease has a complex set of genetic loci each of which contributes a small amount to the overall genetic risk.
Here is taken a different approach, from logic and the history of research in this population. If one is looking for the genetic contribution to a highly prevalent disease, the obverse to the current model, a little from many, is to search for loci that have highly prevalent alleles. In 1981 an association of serological antigen HLA-A*02 with T2D in full indigenous heritage people was reported.4 This is the most common allele at the HLA-A locus in indigenous Americans. Furthermore, a 2024 report described an epistasis between HLA-DRB1*16:02 and the ancient human haplotype for SLC16A11 in the heavier body mass index (BMI, kg/m2) strata of southwestern American Indians that reduces their risk of T2D by 50%.5 This ancient haplotype is also one of the most common alleles in the population. The logical correlate of this is to examine the relationship between the HLA-A alleles and SLC16A11 and estimate their risk/protection for the disease. After more than forty years of refinement, HLA-A typing has become much more fine-scale revealing many underlying molecular alleles within the category of A*02.6 Here is performed a functional analysis that studies the role A*02:01 and A*02:06 and their relationship to donor 9mer peptides of the monocarboxylate transporter to determine their combined contributions to the genetic-immunological risk for T2D. The method uses binding prediction algorithms to identify risk peptides and develops polygenic statistical models that reflect the joint action of the HLA alleles and SLC16A11. Further, because BMI confounds the logistic analysis, sensitivity studies over the range 25-45 kg/m2 are performed to assess the statistical power of the results with a bootstrap within each BMI stratum. If such a relationship could be defined, it would be a large contribution to genetic medicine because a person's combined genotypes and BMI status would define their risk for the disease that in turn could be incorporated in counselling to modify their environmental susceptibility to T2D.
Methods
Population data
The published data and reports from which these analyses were performed are derived from a sample of Southwest American Indians who have a portion of American Indian heritage and who have participated at some time in a study of type 2 diabetes mellitus from 1965 to 2007. A list of papers underlying the present work is available from the author. Diabetes diagnosis and genotyping were performed as previously described.5,7–10 The new peptide binding analyses used a vector of published HLA-A alleles common in indigenous Americans and incorporate publicly available protein sequences and algorithms. No new unpublished genotypes or phenotypes were used in the present analysis.
HLA typing
HLA typing was performed from methods as previously described.11–14 HLA alleles were used at the 4-field level of resolution.
SLC16A11 locus
A mutation haplotype of SLC16A11 is defined by five SNPs—rs117767867(T), rs13342692(C), rs13342232(G), rs75418188(T), and rs75493593(T)—and segregates in this population in a diallelic model. The wildtype is defined as allele1 and the mutation allele 2 yielding three genotypes: SLC16A11*11, *12, and *22. The mutation has its origin in the early hominins, 500,000 to 750,000 years ago, an ancient human haplotype.15,16 Its association with T2D in Hispanic and Indigenous persons, the effect of weight as a confounder, and the search for the mechanism, has been reported in many studies.5,15,17–20
Statistics
Allele frequencies were calculated either by gene counting or the EM method.21 Haplotypes frequencies were computed by the estimator-maximum algorithm.22 Haplotypes were defined for each functional HLA-A heterodimer assigned to genotypes by a weighted probability function based on the estimated EM haplotype frequencies. Descriptive statistics and general linear models for the association of T2D with covariates were calculated with SAS version 9.4.23 Covariates for the linear models included age, sex, and PC1-PC5. The model was corrected for first-degree relationship using the REPEATED option in the SAS PROC GENMOD procedure that corrects for the familial correlation of the members of the sibship. When the REPEATED option in the SAS PROC GENMOD model is used to control relatedness among the subjects, a -2loglikelihood test to determine the significance of the differences of the models cannot be employed. Instead, SAS provides the Quasi-likelihood Information Criterion (QICu) to compare different GEE regressions. A lower value of the statistic indicates a better fit. The statistical power of the odds ratios within strata was calculated by a 1,000-iteration bootstrap of the fully parametrized logistic regression with the power defined by the proportion of iterations <0.05.5 Sensitivity analysis was calculated from logistic regressions in strata defined by BMI categories <BMI in BMI range 25-45.5 For instance, the <BMI 25 stratum was all persons who had BMI kg/m2 less than 25.
MHC binding prediction
MHC binding prediction for HLA was performed using the algorithm NetMHCIIpan 4.0 EL and NetMHCpan 4.1 EL from the Immune Epitope Database and Analysis Resource (IEDB).24–26 For input the program uses the first peptide from a FASTA file for a given protein and a vector of HLA alleles for each of which are to be calculated an IC50nM score and a percentile rank. The percentile rank compares the peptide's score against the scores of five million random peptides selected from the SWISSPROT database. A small percentile rank indicates high affinity of the peptide with the HLA allele. The program moves over one amino acid, reads the second peptide, calculates the two parameters for each MHC allele, and then, by this method, continues to the end of the amino acid sequence.
Results
Allele and haplotype frequencies
Table 1 presents the EM-estimated frequencies of the SLC16A11 and HLA-A alleles in the present study. Risk allele HLA-A*02:01 has one of the highest reported world frequencies in this indigenous group, 0.420. The second most frequent allele at the HLA-A locus is A*24:02, 0.335. Combined, the two alleles represent 75% of the total allele frequency variation at the locus and are found in American Indians and admixed groups throughout the Americas.27 A second variant of A*02, A*02:06, is also found in the sample with frequency 0.061. When considering individual copies of the risk alleles, approximately 65% have at least one copy of HLA-A*02:01 while about 11% have at least one copy of A*02:06.
Table 1. Allele Frequencies
| Allele | Frequency | 95% C. I. |
|---|---|---|
| HLA-A, N = 6634 | ||
| 02:01 | 0.420 | 0.412, 0.429 |
| 02:06 | 0.061 | 0.057, 0.065 |
| 24:02 | 0.335 | 0.327, 0.343 |
| 31:01 | 0.089 | 0.084, 0.093 |
| Other | 0.095 | 0.090, 0.100 |
| SLC16A11, N = 6512 | ||
| C-T-A-C-G | 0.603 | 0.595, 0.611 |
| T-C-G-T-T | 0.397 | 0.389, 0.405 |
The frequency of the wildtype SLC16A11 allele (C-T-A-C-G) is 0.603 while the early human risk allele (T-C-G-T-T) frequency is the simple additive inverse, 0.397. Goodness-of-fit tests cannot reject the null hypothesis of Hardy-Weinberg equilibrium. Approximately 64-68% of the population have at least one copy of the early human haplotype. (Table 2) The EM-estimated probability for the SLC16A11 5-SNP genotypes is 1.000 for all persons. Comparative data for the locus in other American Indian populations with this level of genomic typing and haplotype estimation does not exist. However, studies in North and Central American Hispanic populations with Native American admixture suggest similar frequencies might be expected in other full indigenous heritage samples when typed.15
Table 2. Description of the Southwestern American Indian Sample
| Stratum | N | T2D | HLA-A | SLC16A11 | Combined Risk Alleles | |
|---|---|---|---|---|---|---|
| % | %*02:01 | %*02:06 | %T-C-G-T-T | %*02+T-C-G-T-T | ||
| Total | 5666 | 36.4 | 65.6 | 11.9 | 63.7 | 89.5 |
| BMI<25 | 588 | 24.3 | 65.0 | 10.9 | 67.6 | 91.3 |
| BMI<35 | 3167 | 33.9 | 65.3 | 11.4 | 64.8 | 89.8 |
| BMI<45 | 5005 | 35.5 | 65.3 | 11.8 | 64.0 | 89.3 |
A binding polymorphism
The binding pattern of 9mers was computed for four polymorphic alleles, HLA-A*02:01, *02:06, *24:02, *31:01, with the predictive peptide binding tool NetMHCpan 4.1 EL. Of particular interest was the pattern across the four missense mutations of SLC16A11. We found a high probability of strong binding at missense D127G for A*02:01. (Table 3) For the wildtype peptide DLLHLYLGL, while the rank% is 3.3, the probability of strong binding score is only 0.023. However, with the glycine mutation and peptide GLLHLYLGL, the rank% represents 0.27% of the highest binding scores in the SWISSPROT database, while the probability of strong binding at 0.487 is among the highest values for the protein.
Table 3. Predicted Binding of HLA-A*02:01 and A*0206 with Wildtype and Early Human Mutation from SLC16A11 at D127G Polymorphism
| Wildtype | Early Human Mutation | |||||||
|---|---|---|---|---|---|---|---|---|
| Core Peptide | Rank% | Probability | Strength | Core Peptide | Rank% | Probability | Strength | |
| HLA-A*02:01 9mers | ||||||||
| 1 | FVFSAFASD | 23.0 | 0.000307 | NL | FVFSAFASG | 15.00 | 0.00096 | NL |
| 2 | VFSAFASDL | 19.0 | 0.000490 | NL | VFSAFASGL | 12.00 | 0.00156 | NL |
| 3 | FSAFASDLL | 7.0 | 0.004963 | WB | FSAFASGLL | 11.00 | 0.00188 | NL |
| 4 | SAFASDLLH | 22.0 | 0.000318 | NL | SAFASGLLH | 19.00 | 0.00047 | NL |
| 5 | AFASDLLHL | 2.9 | 0.028636 | WB | AFASGLLHL | 2.70 | 0.03316 | WB |
| 6 | FASDLLHLY | 3.1 | 0.026311 | WB | FASGLLHLY | 5.40 | 0.00840 | WB |
| 7 | ASDLLHLYL | 2.5 | 0.037258 | WB | ASGLLHLYL | 8.50 | 0.00307 | WB |
| 8 | SDLLHLYLG | 63.0 | 0.000012 | NL | SGLLHLYLG | 54.00 | 0.00002 | NL |
| 9 | DLLHLYLGL | 3.3 | 0.023070 | WB | GLLHLYLGL | 0.27 | 0.48690 | SB |
| HLA-A*02:06 9mers | ||||||||
| 1 | FVFSAFASD | 12.0 | 0.00320 | NL | FVFSAFASG | 7.00 | 0.00954 | WB |
| 2 | VFSAFASDL | 25.0 | 0.00055 | NL | VFSAFASGL | 17.00 | 0.00149 | NL |
| 3 | FSAFASDLL | 3.5 | 0.03309 | WB | FSAFASGLL | 5.90 | 0.01331 | WB |
| 4 | SAFASDLLH | 19.0 | 0.00111 | NL | SAFASGLLH | 15.00 | 0.00217 | NL |
| 5 | AFASDLLHL | 3.1 | 0.03989 | WB | AFASGLLHL | 3.00 | 0.04272 | WB |
| 6 | FASDLLHLY | 1.2 | 0.14326 | SB | FASGLLHLY | 2.30 | 0.06318 | WB |
| 7 | ASDLLHLYL | 1.2 | 0.14433 | SB | ASGLLHLYL | 7.00 | 0.00972 | WB |
| 8 | SDLLHLYLG | 50.0 | 0.00007 | NL | SGLLHLYLG | 44.00 | 0.00010 | NL |
| 9 | DLLHLYLGL | 5.7 | 0.01413 | WB | GLLHLYLGL | 0.85 | 0.20694 | SB |
When the four polymorphic HLA-A alleles were entered in the program and tested against all 9mers for SLC16A11 ancient human haplotype, there were 1852 observations within which the strong binding probability of the glycine peptide was among the top 1.5% of values. (Table 4) There is also a binding polymorphism for A*02:06, but smaller, with strong binding score of 0.014 for the aspartic acid allele and 0.207 for the glycine mutation. (Table 3) No binding polymorphism occurs for A*24:02; however, there is strong binding for both the wildtype and early human mutation at peptides AFAS(D/G)LLHL. The glycine mutation slightly enhanced strong predicted binding for the allele, 0.322.
Table 4. Distribution of SLC16A11 Peptide Binding Scores in Decile Intervals in Southwest American Indians
| HLA-A Locus Binding Probability 9mers, N = 1852 | |||
|---|---|---|---|
| Number | % | Cum. % | |
| 0.0-.10 | 1752 | 94.60 | 94.60 |
| >.10-.2 | 33 | 1.78 | 96.38 |
| >.20-.3 | 17 | 0.92 | 97.30 |
| >.30-.4 | 12 | 0.65 | 97.95 |
| >.40-.5 | 10 | 0.54 | 98.49 |
| >.50-.6 | 7 | 0.38 | 98.87 |
| >.60-.7 | 6 | 0.32 | 99.19 |
| >.70-.8 | 7 | 0.38 | 99.57 |
| >.80-.9 | 4 | 0.22 | 99.78 |
To obtain perspective on the strength of the binding polymorphism for HLA-A*02:01, the same set of polymorphic alleles were taken at HLA-A with binding estimated across all the 359,516 9mer peptides of a 113-protein set with NetMHCpan 4.1 EL (Figure 1). The mean probability of binding score was 0.018 with a 90% quantile of 0.014 in a highly skewed distribution for which 96% of binding scores were <0.1. The binding score of 0.487 for A*02:01 at D127G of the ancient human SLC16A11 haplotype places it in the top 1.4% of all probabilities of strong binding.
Observing the binding polymorphisms for HLA-A as a function of the SLC16A11 protein suggested a model to test the association of the genotypes for risk of T2D with logistic regression controlled for age, sex, first degree relationship, and PC1-PC5.5 These were performed in a sensitivity analysis for <BMI strata 25-45 kg/m2; in the model SLC16A11 without HLA-A demonstrates the well-known strong associations as reported earlier.5,15 For the SLC16A11*22 genotype the maximum odds ratios (OR) are found at strata <BMI 31 (2.50, p<0.0001) and 32 (2.48, p<0.0001). Genotype SLC16A11*12 also has maximum values at these same <BMI strata: 31 (1.51, p=0.0041); 32 (1.58, p=0.0006). When a similar analysis is performed for HLA-A*02:01 without the SLC16A11 genotypes, with T2D as the dependent variable and the standard set of covariates, none of the models had a type 3 GEE significance <0.05.
Interaction model
A logistic regression and sensitivity analysis were next performed with T2D as the dependent variable and both HLA-A*02:01 and SLC16A11 as explanatory variables with interaction. While, again, there was no association with A*02:01 and none of the interaction terms were significant, the presence of the HLA-A allele significantly increased the vector of odds ratios for SLC16A11. (Figure 2) Without a significant interaction term and lack of independence between the BMI sensitivity categories, because they are cumulative, a resampling approach is required to test the significance of the effect. A bootstrap of b=1000 for the twenty-one differences between the odds ratios for SLC16A11*22 without A*02:01 in the model, and with the allele and interaction, was performed. The mean OR difference was 0.67 with standard error 0.26 and empirical confidence interval (0.23, 1.24) that excludes 0.0. The vector of QICu statistics across the range of BMI sensitivities for models with SLC16A11 alone was subtracted from one with both SLC16A11 and HLA-A*02:01, N=21. The mean difference was -7.07, se=2.28, with empirical 95% confidence interval (-12.06, -2.74). A similar analysis substituting the QICu values for models with the interaction term had mean=-6.19, se=2.97, and empirical 95% C.I. (-12.20, -0.80).
Polygenic logistic model
A polygenic logistic model with BMI sensitivity analysis was designed that defined the combined risk alleles (RA) as HLA-A*02:01 and the SLC16A11*T-C-G-T-T early human mutation. The reference is persons with no risk alleles. It defines four risk categories, 4, 3, 2, and 1. For instance, the double homozygote [A*02:01/A*02:01; *T-C-G-T-T / *T-C-G-T-T] would be scored 4, whereas the double heterozygote [A*02:01/A*24:02; *C-T-A-C-G / *T-C-G-T-T] would be scored as 2. The type 3 GEE model significance was greatest in strata <BMI 31, p = 0.0007, and 32, p = 0.0009. The vector of five odds ratios for the two <BMI strata were entered in weighted linear regressions: for 31 the fitted line was OR = 0.521RA + 0.956, p = 0.0027, R2 = 0.966; while for 32 OR = 0.426RA + 0.928, p = 0.0035, R2 = 0.959. A similar joint model was developed for a combined HLA-A risk category of A*02:01 and A*02:06 with the early human mutation because both HLA-A alleles share a binding polymorphism at D127G (Table 3). The increase in the odds ratios over that of SLC16A11*22 alone demonstrates a significant linear trend at type 3 GEE maximum model significance at <BMI 31 (OR = 0.583RA + 0.811, p = 0.0017, R2 = 0.975) and 32 (OR = 0.463RA + 0.776, p = 0.0046, R2 = 0.952). (Figure 3)
To test the statistical power of the association of the risk allele categories within these strata, a bootstrap of b=1000 iterations was performed within each stratum. For stratum 31 the risk allele categories of 4, 3, and 2 had power values 0.819, 0.950, and 0.722, whereas the respective power numbers for stratum 32 were 0.762, 0.933, and 0.633, the proportions of bootstrap values p<0.05.
Figure 4 illustrates the enhancement of HLA-A*02:01 and A*02:06 in the polygenic models when the risk category "4" odds ratio is compared to the SLC16A11*22, the homozygous risk genotype, when it is in a regression without the HLA-A alleles (blue bars). The black bars represent all three risk alleles in the polygenic model while the red bars remove A*02:06 from the risk categories. The analyses were performed across <BMI categories 25-35 and demonstrate a decreased effect of the enhancement as the BMI categories increase, which confirms the confounding effect of BMI on the SLC16A11 association with T2D.
A control allele segregates at the HLA-A locus, A*24:02, that is nearly as common as A*02:01. (Table 1) When we repeated the same analyses with it there was no amplification of the SLC16A11 odds ratios when A*24:02 was included as a second explanatory variable and, when a new polygenic model was created with the early human mutation as risk alleles, there also was no similar effect. (Figure 5) There is no binary polymorphism for the allele. Therefore, both the binding polymorphisms and statistical relationship with SLC16A11 were confined to A*02:01 and A*02:06.
Discussion
The mechanics of the human immune system are complicated because of the subtyping of the functions of B-, T-, and dendritic cells along with their suites of cytokines, chemokines, receptors, ligands, gene enhancers and repressors, and microRNAs.28,29 However, a fundamental reaction among the cascades is the presentation of peptide by the HLA heterodimers. In the past, many studies of the association of HLA alleles alone have shown a statistical relation, risk or protection, with diseases. This model was expanded to identify and include the peptide and its origin in the analysis and to create an appropriate statistical model. This expansion implies that past binary models of HLA association, protein, and disease, might be incomplete, that, as an antigen presenting molecule, the peptide or peptides presented are needed to complete the analysis, an HLA tripartite model: HLA heterodimer, disease, peptide.
A surprising finding was the binding polymorphism found for class I HLA-A*02:01 at the glycine substitution for aspartic acid at SLC16A11 D127G and peptide GLLHLYLGL, which early human haplotype is associated with T2D (Table 3). Allele A*02:01 is ubiquitous in all human populations but has one of its highest representations in indigenous Americans. When non-zero frequencies in the Allele Frequency Net Database (AFND) are tabulated for HLA-A*02:01 and HLA-A*02:01:01, there are 357 records with a mean frequency of 0.159 and a 99% quantile of 0.474.27 Among the top ten frequencies in the world, three of them were reported from this indigenous population11–14, and nine out of ten came from indigenous or admixed populations in North, Central, and South America.27 When we define the frequent SLC16A11 early human haplotype and A*02:01 as the risk alleles in the joint model, about 90% of all persons in this indigenous population have at least one risk allele (Table 2), while 36% of the total population has T2D with approximately 50% of affected persons over the age of thirty-five.1,2 Combined with the observation that there is a significant linear trend that associates the numbers of risk alleles with the disease, this finding suggests that the two loci together compose a large proportion of the genetic component for susceptibility. Further, because the two risk alleles, SLC16A11*T-C-G-T-T and HLA-A*02:01, are likely to be found in all indigenous and admixed groups throughout the United States, Canada, Central and South America, these observations could have implications for their risk of T2D.
An earlier association with the antigen HLA-A*02 as the sole explanatory variable, with covariates, could not be reproduced. When A*02:01 is alone in the logistic regression, the reference stratum has many people who do not have a SLC16A11 risk allele. The binding polymorphism implies that the T-C-G-T-T haplotype must be present with the HLA-A risk allele to amplify the effect of the SLC16A11 genotypes. Including the two loci in the same regression, with an interaction term, indirectly associates the heterodimer and peptide that results in the increased magnitudes of the ORs. The new polygenic logistic model explicitly associates A*02:01 and T-C-G-T-T. In the four-allele risk category there is total strong binding that yields the largest odds ratios (Figures 3,4). Likewise, in the three-allele risk category there is guaranteed strong binding because the HLA-A allele and the resulting glycine peptide will occur in some combination. Further, this interpretation of the genotypes and statistics is supported by the statistically significant linear trend of risk as a function of number of alleles in which the R2 values were all >95%, a measure of the amount of variation explained by the linear model. The strong statistical power for the association of the genotypes with T2D for the joint analysis that adds A*02:06 to it is further evidence for the correctness of the model and its interpretation (Figure 4).
This binding polymorphism is the third component for an HLA tripartite association model for antigen presenting cells and implies a way forward for further studies. In contrast to mutations in genes such as the sickle cell beta hemoglobin protein or the cystic fibrosis transmembrane conductance regulator (CFTR) gene, in which the primary genetic lesion leads to a change in the action of the molecule itself, in the peptide presenting HLA molecules the missense mutation is in a second, polymorphic protein which degradation product has two forms, wildtype and mutant. Therefore, there are three elements that need to be considered, HLA, disease, and target protein and its variants. A further complication in the analysis is that classical HLA plays a crucial role in the maintenance of self-tolerance as well as a potential source of autoimmunity. Teasing out the tolerance versus the disease roles of HLA is the challenge.
A recent paper emphasized the role of MCT11 (SLC16A11) in the treatment of cancer and its interference in clinical therapies.30 HLA-A molecules, such as A*02:01 and A*02:06, present their peptide to CD8+ T cells that help initiate the cell-mediated immunological response. It reported that SLC16A11 promotes the influx of lactic acid into tumor-infiltrating CD8+ Tex cells, allowing the maturation of Tex cell exhaustion that reduces their effector functions. Targeting the MCT11 molecule with antibody decreased lactic acid uptake in Tex cells in tumor-bearing mice and reduced tumor growth. While T2D is not cancer, the observation is another example of histocompatibility class I molecules, here mouse, working with monocarboxylate transporter 11 in modifying the course of disease, and provides an additional immunological pathway to explore for its action in human diabetes susceptibility and protection.
Modern computer storage technology, search engines, artificial intelligence, and machine learning algorithms would aid this discrimination by cataloguing strong binding peptides — and, more particularly, binding polymorphisms, as seen in the present paper — into a universal library of HLA molecules and their potential targets in the human proteome. Genome wide association studies have revealed millions of genetic polymorphisms. Modern computer algorithms that predict the binding of class I and class II peptides would be applied to the proteome, starting with the major classical loci, HLA-A, -B, -C, -DR, -DQ, and -DP. Each HLA molecular allelic variant would be tested against each peptide in the proteome from which a topography of binding across the human genome would be created. Special emphasis would be placed on missense mutations in exons to compare wildtype-consensus peptides with amino acid substitutions. When a protein is associated with a disease, and there is reason to believe it involves an immune component, such as T2D, the library would be searched for patterns of binding. This search would include strong binding at non-polymorphic sites that might also be informative. For instance, in the present analysis, HLA-A*02:01 has twelve peptides in monomorphic regions of SLC16A11 for which the probabilities for strong binding have the range 0.522-0.930 with rank% 0.25-0.03 in the SWISSPROT database (data not shown). Further, the library could be extended to model three-dimensional HLA / T-Cell interaction as a function of identified strong binding peptides and/or binding polymorphisms.
An example of a potential application of such a library is presented by a recent paper that used two sophisticated statistical analyses and implicated forty-seven plasma proteins with the onset and complications of T2D.31 Interestingly, the HLA-DRA molecule, which allelic variants form the second protein in the heterodimer antigen presenting molecule with HLA-DRB1, has one of the highest odds ratios and narrowest standard errors. It was recently reported, using the HLA tripartite model, that haplotype HLA-DRA*01:01:01-DRB1*16:02:01 and SLC16A11*T-T-C-T-G were protective in the heavier BMI stratum of the sample with the likely core peptide at D127G being FSAFASGLL.5 This report, in turn, had been motivated by an earlier one that demonstrated a DRB1*02 (16:02) tagged SNP associated with protection from T2D that was first found by a significant relation with HLA-DRA.32 The question naturally arises, given an immune component to the disease, is there a common pattern of binding within the set of peptides of these forty-seven proteins and their variants when tested against the vector of common class I and II loci in the studied populations? With the proposed HLA-proteome-peptide library it would be a matter of defining the parameters for the search and machine-learning to reveal potential commonalities for risk. It takes forty-seven binary, unrelated observations in the study and creates a method for a fine scale molecular search to reveal a potential network of common immunological risk among associated molecules.
Conclusions
Four observations support the idea that these three loci partially underly the genetic and immunological risk and protection for T2D. First, risk alleles HLA-A*02:01 and the SLC16A11 ancient human haplotype have high frequencies that are coincident with the highest frequency of T2D. Second, when the HLA-A*0201 risk allele is entered into the polygenic logistic regression with SLC16A11, an amplification of risk is observed; the resulting hypothesis is that this amplification is strong evidence of a functional, synergistic relationship reflected in the binding polymorphisms in Table 3. Third, the significant weighted linear regressions for the risk genotypes with R2 > 95% suggest an additive model with each additional joint allele increasing risk. Fourth, the pattern of the probabilities of strong binding and rank% at mutation SLC16A11 D127G are consistent with the action of both arms of the immune system, humoral and protective and HLA-DRB1 restricted, as previously described, and cell-mediated-risk, and HLA-A restricted as presented here.
Finally, these data and results illustrate the importance of the genomes of indigenous people, but particularly in populations whose geographical isolation has shielded them from large proportions of gene flow from non-indigenous populations. Communities with large proportions of full indigenous heritage members carry a history that is many hundreds of thousands of years old, not just in individual alleles or loci, but in a pattern of variation over the entire genome that is a valuable resource for the elucidation of human genetic history. In the present case, this pattern of variation potentially reveals underlying genetic and immunological components for protection from, and risk for, type 2 diabetes mellitus that possibly has implications across all human populations independent of heritage and gene flow.
Declarations
- Conflict of Interests
- The author has no conflict of interest to declare.
- Funding
- None.
- Acknowledgements
- None.
References
- Knowler WC, Pettitt DJ, Saad MF, Bennett PH. Diabetes mellitus in the Pima Indians: incidence, risk factors and pathogenesis. Diabetes Metab Rev 1990;6:1–27
- Knowler WC, Bennett PH, Hamman RF, Miller M. Diabetes incidence and prevalence in Pima Indians: a 19-fold greater incidence than in Rochester, Minnesota. Am J Epidemiol 1978;108:497–505
- Mahajan A, Spracklen CN, Zhang W, etc. Multi-ancestry genetic study of type 2 diabetes highlights the power of diverse populations for discovery and translation. Nat Genet 2022;54:560-572. doi: 10.1038/s41588-022-01058-3
- Williams RC, Knowler WC, Butler WJ, et al. HLA-A2 and type 2 (insulin independent) diabetes mellitus in Pima Indians: An association of allele frequency with age. Diabetologia 1981;21:460-463
- Williams RC, Hanson RL, Peters B, et al. Epistasis between HLA-DRB1*16:02:01 and SLC16A11 T-C-G-T-T reduces odds for type 2 diabetes in southwest American Indians. Diabetes 2024;73:1002-1011
- Marsh SGE. Nomenclature for factors of the HLA system. https://hla.alleles.org/nomenclature/index.html
- Traurig M, Hanson RL, Marinelarena A, et al. Analysis of SLC16A11 variants in 12,811 American Indians: genotype-obesity interaction for Type 2 diabetes and an association with RNASEK expression. Diabetes 2016;65:510–519
- Shen GQ, Abdullah KG, Wang QK. The TaqMan method for SNP genotyping. Methods Mol Biol 2009;578:293–306
- Schleinitz D, Distefano JK, Kovacs P. Targeted SNP genotyping using the TaqMan assay. Methods Mol Biol 2011;700:77–87
- Hui L, DelMonte T, Ranade K. Genotyping using the TaqMan assay. Curr Protoc Hum Genet 2008; Chapter 2:10
- Williams RC and McAuley JE. HLA class I variation controlled for genetic admixture in the Gila River Indian Community of Arizona: A model for the Paleo-Indians. Hum Immunol 1992;33:39-46
- Williams R, Chen YF, Endres R, et al. Molecular variation at the HLA-A, B, C, DRB1, DQA1, and DQB1 loci in full heritage American Indians in Arizona: private haplotypes and their evolution. Tissue Antigens 2009;74:520–533
- Williams RC, Knowler WC, Shuldiner AR, et al.; Regeneron Genetics Center. Next generation sequencing and the classical HLA loci in full heritage Pima Indians of Arizona: defining the core HLA variation for North American Paleo-Indians. Hum Immunol 2019;80:955–965
- Williams RC, Koroglu C, Knowler WC, et al. Next generation sequencing for HLA loci in full heritage Pima Indians of Arizona, part II: HLA-A, -B, and -C with selected non-classical loci at 4-field resolution from whole genome sequences. Hum Immunol 2021;82:385–403
- SIGMA Type 2 Diabetes Consortium; Williams AL, Jacobs SBR, Moreno-Macías H, et al. Sequence variants in SLC16A11 are a common risk factor for type 2 diabetes in Mexico. Nature 2014;506:97–101
- Prüfer K, Racimo F, Patterson N, et al. The complete genome sequence of a Neanderthal from the Altai Mountains. Nature 2014;505:43–49
- Lamadrid-Figueroa H, Fritz J. Genetic ancestry of the Mexican population: diversity and implications for health research. Am J Hum Biol 2026;38:e70191
- Sevilla-González M, Garibay-Gutiérrez MF, Vargas-Vázquez A, et al. Metabolomic profile alterations associated with the SLC16A11 risk haplotype following a lifestyle intervention in people with prediabetes. Curr Dev Nutr 2024; https://doi.org/10.1016/j.cdnut.2024.104444
- Aguilar-Salinas CA, Luna MTT. The role of SLC16A11 variations in diabetes mellitus. Curr Opin Nephrol Hypertens 2023;32:455-450
- Maynard AG, Bhardwaj R, Jones TR, et al. Bridging the variant-to-function gap in type 2 diabetes: advances and challenges. Diabetologia 2025;69:274-294
- Weir B. Genetic Data Analysis. Sunderland, MA, Sinaurer Associates, 1990
- Long JC, Williams RC, Urbanek M. An E-M algorithm and testing strategy for multiple-locus haplotypes. Am J Hum Genet 1995;56:799–810
- SAS/STAT 14.2 User's Guide. Cary, NC, SAS Institute Inc., 2016
- Reynisson B, Alvarez B, Paul S, Peters B, Nielsen M. NetMHCpan-4.1 and NetMHCIIpan-4.0: improved predictions of MHC antigen presentation by concurrent motif deconvolution and integration of MS MHC eluted ligand data. Nucleic Acids Res 2020;48:W449–W454
- Jensen KK, Andreatta M, Marcatili P, et al. Improved methods for predicting peptide binding affinity to MHC class II molecules. Immunology 2018;154:394–406
- Andreatta M, Karosiene E, Rasmussen M, Stryhn A, Buus S, Nielsen M. Accurate pan-specific prediction of peptide-MHC class II binding affinity with improved binding core identification. Immunogenetics 2015;67:641–650
- Gonzalez-Galarza FF, McCabe A, Melo Dos Santos EJ, et al. Allele frequency net database (AFND) 2020 update: gold-standard data classification, open access genotype data and new query tool. Nucleic Acids Res 2020;48:783-788
- Vinuesa CG, Linterman MA, Yu D, MacLennan ICM. Follicular helper T cells. Annu Rev Immunol 2016;34:335–368
- Deng Q, Luo Y, Chang C, Wu H, Ding Y, Xiao R. The emerging epigenetic role of CD8+T cells in Autoimmune Diseases: a systematic review. Front Immunol 2019;10:856
- Peralta RM, Xie B, Lontos K, et al. Dysfunction of exhausted T cells is enforced by MCT11-mediated lactate metabolism. Nature Immunology 2024;25:2297-2307
- Yuan S, Xu F, Li X, et al. Plasma proteins and onset of type 2 diabetes and diabetic complications: proteome-wide Mendelian randomization and colocalization analyses. Cell Reports Medicine 2023;4:101174 https://doi.org/10.1016/j.xcrm.2023.101174
- Williams RC, Muller YL, Hanson RL, et al. HLA-DRB1 reduces the risk of type 2 diabetes mellitus by increased insulin secretion. Diabetologia 2011;54:1684–1692
Article information
Authors
- Robert C. WilliamsProfessor Emeritus, SHESC, Arizona State University, 900 Cady Mall, Tempe, Arizona 85281Corresponding author · [email protected]
How to cite
Williams, R. C., 2026. Exploring the Binding Polymorphism for HLA-A Alleles at the Ancient Human SLC16A11 Haplotype in Indigenous Americans with High Prevalence Type 2 Diabetes. Medical Research Archives, [online] 14(10). https://doi.org/10.18103/mra.2026.0657
Record
- Journal
- Medical Research Archives
- Issue
- Vol 14 No 10 (2026)
- Section
- Research Articles
- Published
- 31 October 2026
- ISSN
- 2375-1924
- Downloads
- 73
Copyright and licence
Published by the European Society of Medicine under a Creative Commons Attribution 4.0 International licence (CC BY 4.0), which permits unrestricted use, distribution and reproduction in any medium, provided the original authors and source are credited.