RESEARCH DESIGN AND METHODS—ZnT8A radioimmunoprecipitation assays were performed in 421 new-onset type 1 diabetic Caucasians using COOH-terminal constructs incorporating the known human aa₃₂₅ variants (Trp, Arg, and Gln). Genotypes were determined by PCR-based SNP analysis.

RESULTS—Sera from 224 subjects (53%) were reactive to Arg₃₂₅ probes, from 185 (44%) to Trp₃₂₅probes, and from 142 (34%) to Gln₃₂₅probes. Sixty subjects reacted only with Arg₃₂₅ constructs, 31 with Trp₃₂₅ only, and 1 with Gln₃₂₅ only. The restriction to either Arg₃₂₅ or Trp₃₂₅ corresponded with inheritance of the respective C- or T-alleles. A strong gene dosage effect was also evident because both Arg- and Trp-restricted ZnT8As were less prevalent in heterozygous than homozygous individuals. The SLC30A8 SNP allele frequency (75% C and 25% T) varied little with age of type 1 diabetes onset or the presence of other autoantibodies.

ZnT8 is a newly discovered target of type 1 diabetes autoimmunity (3) localized to the insulin granule of the pancreatic β-cell. It is encoded by SLC30A8, one of nine human genes for multispanning transmembrane proteins facilitating Zn²⁺ efflux from the cell and sequestration into intracellular compartments (4,5). Recent genome-wide association studies demonstrate association of ZnT8 gene polymorphisms with human type 2 diabetes (6–9), notably a nonsynonymous SNP encoding either Arg or Trp at aa₃₂₅. The major, Arg₃₂₅-encoding C-allele confers a minor risk (odds ratio 1.07–1.18) of disease. In nondiabetic subjects with a family history of type 2 diabetes, the C-allele was associated with increased insulin sensitivity (10), increased circulating proinsulin-to-insulin ratio (11), and decreased insulin responses in intravenous glucose tolerance tests (12), indicating a dominant effect on insulin secretion, β-cell mass, or both.

We report here that the type 1 diabetes autoimmune response to ZnT8 is focused on a few key epitopes, two of which are defined by the polymorphic aa₃₂₅ residue. To our knowledge, this is the first reported instance where a polymorphic variant determines the specificity of the autoimmune response. It indicates that the autoreactive B-lymphocyte repertoire is restricted to a few ZnT8 epitopes and is truly self-reactive as opposed to arising as a bystander response to a foreign antigen.

Genomic DNA was extracted from heparinized blood from 352 of the above type 1 diabetes patients using standard procedures. Polymorphic variations in the SLC30A8 gene were determined by qPCR using Taqman probes and an ABI7000 (ABI, Waltham, MA) targeting the nonsynonymous SNPs rs13266634, rs2466295 in the 3′ untranslated region, and rs6469675 in intron 2. Ascertainment rates were >99%.

A comparison of the levels of autoantibody reactivity to the Arg and Gln probes (Fig. 1B) showed that the majority of individuals either reacted equivalently to the probes (falling within the bounds of the diagonal of the x-y plot ± 3 SD) or responded to the Arg probe alone. Trp and Gln reactivities (Fig. 1C) were similarly separated. Of the 34 patients who reacted equivalently to Arg and Trp probes (within the bounds of the diagonal ± 3 SD of Fig. 1D), 29 (85.3%) had an equivalent response as determined by the Gln probe. This indicated that for these individuals, the aa at position 325 was not a determinant of autoantibody reactivity. A series of preabsorption experiments was therefore performed using peptides and recombinant proteins as competing ligands (Fig. 2). Selected type 1 diabetes sera that reacted with the Arg probe alone were blocked by recombinant NUS-C-term Arg₃₂₅ protein but not by NUS-C-term Trp₃₂₅ or NUS-C-term Gln₃₂₅. Similarly, Trp-only responses were blocked by NUS-C-term Trp₃₂₅ but not NUS-C-term Arg₃₂₅ or NUS-C-term Gln₃₂₅. Sera that reacted equivalently to Arg, Trp, and Gln probes were blocked by any of the NUS-C-term ZnT8 recombinants. Overlapping 20-mer peptides spanning the ZnT8 COOH-terminal from 268–369 did not compete for reactivity, suggesting that the epitopes were conformational rather than linear. Overall, these results suggest that ZnT8A reactivity could be accounted for by three classes of conformational epitopes: one for which Arg₃₂₅ was an essential determinant, a second Trp₃₂₅ restricted, and a third not affected by aa₃₂₅.

The relationship between ZnT8 autoantibody reactivity and genetic variation at the SLC30A8 locus was examined using the SNP (rs13266634) encoding the Arg/Trp₃₂₅ variant and two adjacent noncoding SNPs identified in a type 2 diabetes genome-wide association study (6), rs2466295, located 259 bp distally in the 3′ UTR, and rs6469675, located 19635 bp proximally in intron 2. The minor allele frequency (MAF) for rs13266634 in our type 1 diabetic population of 0.266 (n = 351) approximated the reference frequency of 0.256 (n = 168) for Europeans in the NLM SNP database (http://www.ncbi.nlm.nih.gov/SNP/snpref.cgi?rs=13266634). The distribution of genotypes (55.3% CC, 36.2% CT, and 8.5% TT) was consistent with Hardy-Weinberg distribution (53.9, 39.0, and 7.1, respectively). Similar correlations were observed for the MAF for rs6469675 (0.285 vs. 0.220 in our study vs. the NLM SNP database, respectively) and rs2466295 (0.361 vs. 0.407).

The specificity of the ZnT8A response reflected the rs13266634 genotype (Table 1), with little or no association observed with the adjacent rs2466295 and rs6469675 SNPs. The ZnT8A response assessed by the Gln probe showed no significant variation with the rs13266634 genotype, whereas responses to the Arg probe were highest in CC homozygotes, lowest in TT homozygotes, and intermediate in the heterozygote group. Conversely, responses to the Trp probe were highest in TT homozygotes, lowest in CC, homozygotes, and intermediate in heterozygotes. An even stronger relationship with genotype was seen in the groups having only Arg₃₂₅- and Trp₃₂₅-restricted responses. Arg₃₂₅-only responses were observed only in individuals bearing the rs13266634 C-allele, with a 4.2-fold higher frequency in homozygotes than heterozygotes. With one exception, all Trp₃₂₅-restricted responses were associated with the rs13266634 T-allele, with a 10.2-fold higher frequency in homozygotes than heterozygotes. The single Gln₃₂₅-restricted ZnT8A patient (Fig. 1A) was genotyped CC, confirmed by sequencing. The small number of sera positive for both Trp₃₂₅ and Arg₃₂₅ but not Gln₃₂₅ probes were associated with heterozygote genotypes (11 of 13 cases) as expected (data not shown). The prevalence of insulin autoantibody, GAD antibody, or IA-2 antigen response did not change as a function of rs13266634 genotype.

The median age of onset of disease in the genotyped individuals was 11.2 years (range 0.6–58), with more than half (57.3%) diagnosed between ages 8 and 16 years and 88.9% before age 18 years. A frequency distribution analysis based on binning at 4-year intervals showed no statistically significant difference in the representation of the CC, CT, and TT genotypes at any age of onset, though a trend was observed for a higher CC and lower CT genotype frequency in the youngest onset group (0.6–4 years) compared with that of a 5- to 15-year-old reference group (Fishers exact test, P = 0.24; n = 219) (Fig. 3A).

The autoantibody responses to the ZnT8 Arg- and Trp-restricted isoepitopes segregated with the alleles encoding the respective variant amino acids, indicating that humoral type 1 diabetes autoimmunity to ZnT8 is directed against self and not nonself epitope determinants. This argues against the molecular mimicry hypothesis that suggests that autoimmunity is triggered by an initial immune response to a infectious agent that in turn triggers reactivity to self because of sequence homology between the pathological agent and a self protein (16–20). Our results favor the idea that ZnT8 autoreactivity arises because of a defect in induction of self-tolerance, since the mimicry model would more likely favor one isoepitope over another, which in turn would be manifest as genetic dissociation of the encoding allele with the disease. The MAF of the rs13266634 SNP in the type 1 diabetic population under study was, however, similar to reference populations, and no association of the SNP was seen with age of diabetes onset or the prevalence of antibodies to ZnT8 or other diabetes autoantigens. We cannot, however, preclude a role for molecular mimicry in T-cell recognition of antigenic peptides or in antigen presentation to CD4⁺ T-cells because antigen/antibody binding can directly influence the peptides presented from the antigen by virtue of altering intracellular proteolytic processing (21,22).