|Home | About | Journals | Submit | Contact Us | Français|
Recently we demonstrated that zinc transporter 8 (ZnT8) is a major target of autoantibodies in human type 1 diabetes (T1D). Since the molecules recognized by T1D autoantibodies are typically also targets of autoreactive T cells, we reasoned that this would likely be the case for ZnT8. To test this hypothesis IFN-γ producing T cells specific for ZnT8 in the peripheral blood of 35 patients with T1D (< 6mo post-onset at blood draw) and 41 age-matched controls were assayed by ELISPOT using a library of 23 overlapping di-peptide pools covering the entire 369aa primary sequence. Consistent with our hypothesis, patients showed significantly higher T cell reactivity than the matched controls, manifest both in terms of the breadth of the overall response, and the magnitude of responses to individual pools. Thus, the median number of pools giving positive responses (stimulation index ≥ 3) in the control group was 1.0 (range 0 – 7) compared to 6.0 (range 1 – 20; p < 0.0001) for the patients. Similarly, the median SI of positive responses in controls was 3.1 versus 5.0 in the patients (p < 0.0001). Individually 7/23 pools showed significant disease-association (p < 0.001), with several of the component peptides binding the disease associated HLA-DR3 (0301) and -DR4 (0401) molecules in vitro. We conclude that ZnT8 is also a major target of disease-associated autoreactive T cells in human T1D, and suggest that reagents that target ZnT8-specific T cells may have therapeutic potential in preventing or arresting the progression of this disease.
Although autoantibodies are currently the best biomarkers of type 1 diabetes (T1D) in humans, it is generally accepted that the destruction of pancreatic β-cells is principally the result of the activation and expansion of autoreactive T cells specific for β-cell antigens (reviewed by (1)). Consequently, considerable efforts have been made over the past 25 years to identify the molecular targets of potentially diabetogenic T cells, and the immunodominant epitopes within them (2). Such information has contributed significantly to our understanding of the pathophysiology of T1D, for example by revealing possible molecular mimicry between viral and islet antigens (3). It also underpins the development of peptide-based antigen-specific therapeutic strategies, and is responsible for the identification of novel biomarkers of pre-clinical disease (4).
Zinc transporter 8 (ZnT8) is primarily restricted to the islets of Langerhans, with the highest expression being in pancreatic beta cells (5). Recently, we demonstrated that, depending upon the age of onset, autoantibodies to ZnT8 (ZnT8A) are present in 60 – 80% of newly diabetic subjects attending the Barbara Davis Center clinic ((6) and J. M. Wenzlau, H. W. Davidson and J. C. Hutton unpublished findings). This result has since been replicated by ourselves and others in different populations (7, 8); it is now evident that ZnT8A occur with a similar prevalence to those against the “gold standard” T1D autoantigens proinsulin, the 65kDa form of glutamic acid decarboxylase (GAD65), and insulinoma antigen 2 (IA-2), and that ZnT8A overlap with, but are independent of, these other biomarkers. Consequently, protocols that measure ZnT8A in addition to antibodies against the other 3 major targets have a significantly enhanced detection rate of diabetes-related autoimmunity (6, 7, 9, 10).
Although humoral and cellular responses to diabetes autoantigens are believed to be independently activated (11), numerous previous studies have shown that T cells recognizing epitopes from proinsulin, GAD65, and IA-2 can be detected in the peripheral blood of patients at the onset of clinical disease (reviewed by (2)). Given this precedent we reasoned that ZnT8 is also likely to be a target of autoreactive T cells in human T1D. To investigate this hypothesis we analyzed the frequency of proinflammatory IFN-γ secreting cells in PBMCs from 35 newly diabetic subjects (<6 mo post-diagnosis) and 41 age-matched controls, using a library of overlapping peptides spanning the entire 369 amino acid primary sequence of ZnT8. The results of this initial study suggest that ZnT8 is indeed a significant T cell target in T1D, and identify 6 regions of the molecule that likely contain key disease associated epitopes.
Participants of either gender aged 6 – 30y (Table I) were recruited from patients, relatives, and volunteers attending the Barbara Davis Center for Childhood Diabetes in accordance with protocols approved by the Colorado Multiple Institute Review Board. Subjects with clinical disease of greater than 6 mo duration at the time of blood draw were excluded from the study, as were control subjects who tested positive for 1 or more diabetes autoantibodies (IAA, GADA, ICA512, or ZnT8A). PBMCs were prepared from heparinized blood using Ficoll-paque Plus™ (GE Healthcare, Piscataway, NJ) and aliquots used immediately for either ELISPOT analysis or DNA preparation.
Control peptides P1 (Proinsulin C19-A3; GSLQPLALEGSLQKRGIV), Insulin B9-23 (SHLVEALYLVCGERG), GAD3 (GAD65 335–352; TAGTTVYGAFDPLLAVAD), GAD4 (GAD65 554–575; VNFFRMVISNPAATHQDIDFLI), R2 (IA-2 853–872; SFYLKNVQTQETRTLTQFHF), and R5 (IA-2 709–736; LAKEWQALCAYQAEPNTCATAQGEGNIK) (12, 13) were synthesized at >95% purity (University of Colorado Cancer Center Proteomics Core) and added from stocks of 10mM in DMSO. A Pepscreen™ library of overlapping 20mers spanning the entire 369 residue primary sequence of ZnT8 was obtained from Sigma Genosys (St Louis, MO). The library was designed to contain 51 20mers, with consecutive peptides overlapping by 13 residues. However, 5 failed synthesis, providing an actual library of 46 peptides. Individual peptides were resuspended in DMSO to a final concentration of 50mM. (Fig. 1A)
Indirect ELISPOT analyses were conducted essentially as described elsewhere (13) using the human interferon-gamma ELISPOT kit (U-CyTech Biosciences, Utrecht, The Netherlands). To optimize the use of the available blood draw (typically 30 – 70ml) while minimizing potential competition between peptides, the ZnT8 library was divided into 23 pools each containing 2 sequential peptides (each at 10μM) (Supplementary Table II). Freshly isolated PBMCs (1 × 106) were cultured in 250μl RPMI 1640 containing 10% heat-inactivated human AB serum (PAA Laboratories Inc., Dartmouth, MA) and 10μM control peptide or di-peptide pool. An additional 250μl of medium was added after 24h, and the cells harvested 24h later. After washing, the cells were resuspended in 300μl medium and transferred as three 100μl aliquots to 96-well clear polystyrene culture plates previously coated with the anti-IFN–γ capture monoclonal and subsequently treated with 1 × blocking solution (U-CyTech). Seventeen hours later, the cells were removed by decanting, and the wells extensively washed (2 × PBS, and 5 × PBS containing 0.05% Tween-20). Spots were then formed by sequential incubations with the biotinylated 2nd site anti-IFN–©, gold-labeled goat anti-biotin, and a precipitating silver substrate, and enumerated with a Bioreader 4000 Pro × (BIOSYS GmbH, Karben, Germany). Results are expressed either as the total number of specific spots or as stimulation indices (SIs). The former is calculated by adding the total number of spots formed in the 3 wells derived from incubations in the presence of peptide(s) and subtracting the total number of spots detected in the 3 wells containing cells incubated in the presence of vehicle (DMSO) only to define the specific signal for each peptide or peptide pool. Negative values are set to zero. Total (spots - background) is the sum of the specific signals for each individual. SIs are calculated by adding the total number of spots formed in the 3 wells derived from incubations in the presence of peptide(s) and dividing by either the total number of spots detected in the 3 wells containing cells incubated in the presence of DMSO alone, or 1 if no spots were detected in these wells. Based on the behavior of ELISPOT assays for other diabetes autoantigens (14) an SI ≥ 3 was selected as the cut-off for positivity. Positive control samples comprising incubations with Pentacel™ (Sanofi Pasteur Inc, Swiftwater, PA: a mixture of diptheria and tetanus toxoids, acellular pertussis, adsorbed and inactivated poliovirus, and H. influenzae type b capsular polysaccharide conjugated to tetanus toxoid) were also included in each assay.
IAA, GADA, and ICA512 were determined by the UCD DERC clinical core using established assays (15–17). ZnT8A were either determined using the “standard” radioimmunoassay (6), or a modified procedure using a trimeric probe containing sequentially the R, Q, and W variants of the ZnT8 C-terminal domain. The detailed design of the trimeric probe will be reported elsewhere (JMW, HWD, and JCH in preparation).
HLA genotyping was performed by the UCD DERC clinical core. Individual DRB1 and DQB1 alleles were identified by reverse hybridization of PCR amplicons (18) to either sequence specific oligonucleotide bead arrays (DRB1) (Luminex xMAP; One Lambda, Inc., Canoga Park, CA), or linear arrays (DQB1) (Roche Molecular Systems, Alameda, CA), respectively.
Binding of ZnT8 peptides to recombinant HLA-DR3 (0301) and -DR4 (0401) was performed by Proimmune Ltd., (Oxford, UK) using their Class II REVEAL™ binding assay.
Statistical analyses were conducted using Prism 5 software (GraphPad Software Inc., La Jolla, CA). Group comparisons used the Mann Whitney U test, and categorical variables used Fischer’s exact test. In each case p < 0.05 was considered significant.
ZnT8 is a recently described humoral autoantigen in T1D (6). However, at present little is known regarding its relevance as a target of potentially diabetogenic T cells. The goal of this study therefore was to determine if indeed ZnT8 is a significant target of pro-inflammatory T cells in patients with recently diagnosed T1D. In contrast to proinsulin, GAD65, and IA-2, ZnT8 is a polytopic integral membrane protein with 6 predicted transmembrane helices (5), and we are currently unable to express the intact recombinant molecule in a form suitable for conducting T cell assays. Thus, as a surrogate for the intact protein, we used a library of overlapping peptides encompassing the entire 369 amino acid primary sequence (Fig. 1A). Given the relatively low frequency of islet antigen specific autoreactive T cells in the peripheral blood of most patients with T1D, and the complexity of the human MHC, which exhibits co-dominant expression of alleles thereby creating the possibility for both cis and trans-pairing of the polymorphic HLA-DP and -DQ alpha and beta chains (for example (19)) and potential expression of up to 12 distinct class II molecules by a single individual, each with its own unique binding motif, it is unsurprising that previous studies have suggested that T1D association is most evident when multiple epitopes are considered together, rather than on the basis of responses to a single antigenic peptide (12, 14, 20, 21). Accordingly, we first examined the total number of ZnT8-specific T cells detected in the samples from each donor (Fig. 1B). A highly significant disease association was observed. Although there was no statistically significant difference between the background signals in the 2 groups (median 0 spots (range 0 – 16) for controls and 1 spot (range 0 – 24) for patients; p = 0.11), the median number of spots above background in the summed incubations from the control group was 13 (range 0 – 300) compared to 81 (range 18 – 689; p < 0.0001) in PBMCs from the recently diabetic subjects. Using a cut-off defined as the upper 99% confidence limit of the control subjects (50.2 spots) 24/35 (68.6%) of the patients but only 3/41 (7.3%) of controls, showed significant ZnT8-specific responses.
To estimate the breadth of the autoresponse in each individual we next calculated the total number of peptide pools giving a positive response (SI ≥ 3; (14)) in the 2 groups (Fig. 1C). Again a highly significant expansion in the patient group was observed. In the control group the median number of positive peptide pools was 1.0 (range 0 – 7), compared to 6.0 (range 1 – 20; p < 0.0001) in PBMCs from the patients. By this criterion all of the subjects with T1D, but only 29/41 (70.7%) of the control subjects, responded to at least one ZnT8 peptide pool, although all control subjects responded to the positive control that contained a mixture of pediatric recall antigens (data not shown). The significant association between this measure of ZnT8 autoreactivity and T1D was unchanged if either less stringent (SI ≥ 2.1) or more stringent (SI ≥ 5) cut-offs were applied (1.0 v 9.0 pools; p < 0.0001) and (0 v 3.0 pools; p < 0.0001) respectively (Fig. 1D, E). However, using the more stringent cut-off only 29/35 (82.9%) of the subjects with T1D, and 10/41 (24.4%) of the control subjects, responded to any ZnT8 peptide pool.
Our use of a library of overlapping peptides was essential to ensure that as many specificities as possible were detected, but may have the effect of inflating the apparent breadth of the response measured by the sum of positive pools. We therefore repeated our analysis disregarding the second of sequential positive responses where any duplication could occur (i.e. if positive responses were observed in pools 1–4 only the non-overlapping pools 1 and 3 were counted) (Fig. 1F). As expected, this correction selectively reduced the value for the median number of positive pools (SI ≥ 3) in the patients (controls 1.0 v 1.0; patients 5.0 v 6.0), but did not alter the level of statistically significant difference between them.
There was also a significant difference in the magnitude of the positive responses between the two groups. Overall the median SI in control individuals who responded (SI ≥ 3) to at least 1 peptide pool was 3.1 (29 individuals, 84 positive responses, range 3.0 – 30) compared to 5.0 in the patient group (35 individuals, 258 positive responses, range 3.0 – 222; p < 0.0001). Similarly, on an individual basis the median SIs of responders were also significantly different in the two groups (controls: 29 individuals, median SI 3.2, range 3.0 – 9.0 versus patients: 35 individuals, median SI 4.5, range 3.2 – 18.0; p < 0.0001) (Fig. 1G). However there was no correlation between the number of peptide pools that an individual responded to and the median SI of the positive responses in that individual (controls p = 0.73, patients p = 0.63: Spearman’s test).
We also analyzed responses in our subjects to a series of 6 previously validated control peptides from proinsulin, IA-2, and GAD65 (12, 13). Of these only GAD65335–352 showed statistically significant disease association when considered alone (p = 0.0047) (Supplemental Fig. 1A). In contrast, as has been reported previously (14), when the combined responses to the control peptides were examined a highly significant association was revealed (Supplemental Fig. 1B).
T1D shows a strong association with the expression of particular MHC class II alleles, with the HLA-DRB1, -DQA1, and -DQB1 loci on chromosome 6p21 responsible for approximately 50% of familial aggregation of the disease (22). In particular there is a strong positive association with the DRB1*0301-DQA1*0501-DQB1*0201, DRB1*0401-DQA1*0301-DQB1*0302, and DRB1*0405-DQA1*0301-DQB1*0302 (DR3/DQ2 and DR4/DQ8) haplotypes. Consistent with this association, 30/35 of the patient group, but only 28/41 of the control group, expressed at least 1 HLA-DQB1*0201 or -DQB1*0302 allele (Supplementary Table I). Preliminary bioinformatics analyses predicted that both low and high risk molecules would likely bind similar numbers of peptides from ZnT8. Consequently there was no a priori reason to expect that the expression of high risk alleles alone would result in increased autoreactivity to ZnT8. However, to exclude the possibility that the higher proportion of individuals in our patient group who expressed high risk haplotypes was significantly influencing our analysis, we examined the degree of disease association after stratification on the basis of the disease associated DQB1 alleles. When only HLA-DQ2 and/or -DQ8 positive individuals were considered the median number of spots above background in the patient samples was 94.0 (range 20 – 689) compared to 13.0 (range 0 – 140; p < 0.0001) in the samples from the control subjects (Fig. 2A). Similarly, the increased breadth of the response was maintained (controls median = 1.0 pools, patients median = 6.0 pools; p < 0.0001) (Table II). Moreover, disease associated expanded responses were not restricted to either the HLA-DR3/DQ2 or HLA-DR4/DQ8 haplotypes, being evident in both DQ2+ individuals who do not express DQ8 and DQ8+ individuals who do not express DQ2 (Table II). However, the greatest breadth of reactivity in the patient group was seen in individuals expressing HLA-DR1. Thus the median response in DR1+ patients (n = 6) was 11.0 pools (range 4 – 20) versus 5.0 pools (range 1 – 20) in DR1− subjects (n = 29; p = 0.035). Consistent with this observation, DRB1*0101 containing molecules are predicted to bind peptides from 22 of the 23 ZnT8 peptide pools (data not shown), although no significant difference in reactivity associated with DR1 expression was observed in control subjects (DR1+ (n = 10) median 1.5 pools, DR1− (n = 31) median 1.0 pools; p = 0.852).
Unlike HLA-DRB1*0301, *0401, and *0405, alleles such as HLA-DRB1*0403 and HLA-DQB1*0602 show a negative association with T1D, and are therefore considered “protective” (22). At present the mechanism responsible for this genetic protection from T1D remains a matter of debate, but has variously been proposed to be due to a failure of the protective molecules to productively present key diabetogenic peptides (23) or an enhanced ability to bind key tolerogenic peptides (24). Consequently, we were also concerned by the fact that 13/41 controls but 0/35 patients expressed 1 or more of the molecules encoded by these “protective” alleles. Accordingly we repeated our analysis stratifying the control group on the basis of the expression of either HLA-DRB1*0403 or HLA-DQB1*0602 (“protected”), or the absence of these protective alleles (“un-protected”). No difference between the 2 control groups was observed. Thus the median number of spots above background in the samples from the “protected” subjects was 13.0 (n = 13, range 3 – 124) compared to 13.5 (n = 28, range 0 – 300; p = 0.5277) in the samples from the “un-protected” subjects (Fig. 2B). Similarly, the breadth of the responses in the 2 groups was equivalent (“un-protected” median = 1.0 pool, range 0–7, “genetically protected” median = 2.0 pools, range 0–7; p = 0.336). Unsurprisingly, comparison of each group with the patient cohort again revealed a highly statistically significant disease association in each case (p < 0.0001) (Fig. 2C, D).
Together this data indicate that individuals with clinical T1D show a significantly greater number of circulating proinflammatory T cells reactive with ZnT8 than age and HLA-matched controls, and that this auto-response is present in individuals with both the DR3/DQ2 and DR4/DQ8 haplotypes. They also suggest that HLA molecules present in individuals with both genetically protective and susceptible haplotypes are equally able to present ZnT8 peptides to proinflammatory T cells, and that the increased response in the patient group is directly related to disease, rather than simply a consequence of the particular MHC class II molecules that they express.
Unlike many autoimmune conditions T1D does not show any significant gender association (25). Consequently gender was considered irrelevant in the recruitment of our study subjects. However, despite the fact that ~50% of individuals attending the Barbara Davis Center clinic are female (PAG, unpublished observation) there was a distinct male bias in the T1D group, with females comprising only 12/35 (34.3%) of the total (Table I). In contrast, only 15/41 (36.6%) of the control group was male (Table I). Thus, although we considered it highly unlikely, we examined whether this gender discrepancy contributed to the significant differences in ZnT8 T cell responses that we observed. As expected, no significant correlation with gender was observed in either the patient or control groups (Supplementary Fig. 1C, D).
The generation of high affinity class switched antibodies, including diabetes autoantibodies, is a T-dependent process. Nevertheless, previous studies that investigated potential relationships between humoral and cellular autoimmunity to proinsulin, IA-2, and GAD65 did not reveal any significant association (11, 26), likely because of the dependence of B cell help and tissue inflammation on separate effector T cell populations (27). Consistent with the results obtained for the other diabetes autoantigens, no correlation between the ZnT8A index and the magnitude of the T cell response was observed (Supplementary Fig. 1E). Likewise, there was also no association between T cell reactivity to ZnT8 and humoral autoimmunity to either insulin, GAD65, or IA-2 (data not shown).
Newly diabetic individuals who express at least one copy of the risk-conferring HLA-DR4/DQ8 and/or -DR3/DQ2 haplotypes exhibit a significantly higher frequency of proinflammatory ZnT8-specific T cells in their peripheral blood than do age and HLA-matched control subjects (Fig. 2A). Preliminary bioinformatics analyses using several public domain MHC class II prediction servers suggested that multiple peptides from ZnT8 would be expected to bind to the HLA molecules expressed by these subjects (data not shown). However, there was considerable divergence between the various algorithms, precluding a definitive conclusion regarding the behavior of individual peptides from being made. To begin to address this issue empirically, and hence corroborate our ex vivo data, we utilized a commercial in vitro binding assay (Reveal and Prove™; ProImmune) to directly examine binding of a library of 52 15mer peptides spanning the entire primary sequence of ZnT8 to recombinant HLA-DR3 (0301) and -DR4 (0401). As shown in Fig. 3B, 21/52 (40.4%) peptides showed positive binding to HLA-DR4 in vitro, of which 8 had assay values more than twice the threshold for positivity. Comparison between the ex vivo and in vitro libraries indicated that 16 of the 23 pools used for the T cell assays contained at least one peptide capable of binding to HLA-DR4 (0401), including 5/6 pools that showed statistically significant disease association in subjects expressing HLA-DR4 but not -DR3 (Fig. 3B). In contrast, only 4/52 (7.7%) peptides showed significant binding to recombinant HLA-DR3 (0301), with only 1 giving a value more than 200% of the threshold (Fig. 3A). Nevertheless, HLA-DR3 binding peptides were present in 2/4 pools that showed statistically significant disease association in subjects expressing HLA-DR3 but not -DR4.
The magnitude and diversity of the overall response to ZnT8 (Fig. 1) suggests that T cells recognizing multiple epitopes within this autoantigen are expanded in the peripheral blood of subjects with recent onset T1D. Although the data shown in Fig. 3 indicate that multiple peptides from ZnT8 can bind both HLA-DR3 and -DR4, they do not determine which are naturally processed and presented. For the other major autoantigens CD4+ T cell responses restricted to a particular MHC molecule are typically focused on a limited subset of immunodominant peptides (for example (28)). We therefore analyzed the frequency of positive responses to individual ZnT8 peptide pools in the diabetic and control groups (Table III). Perhaps surprisingly, all of the peptide pools elicited a response in at least 4/35 (11%) of the newly diabetic subjects, with T cells reactive to 4 pools (7, 9, 17, and 21) being present in greater than 40% of the members of this group. In contrast, only 13/23 pools elicited a positive response in more than 2/41 (4.9%) of the control subjects, with reactivity to 2 pools (3 and 18) being undetectable in these individuals. Pair-wise analysis (Fischer’s exact test) revealed that 13/23 pools showed a statistically significant disease association at the 95% confidence level, with 7 (pools 1, 5, 9, 17, 18, 19, and 21) also being significant at the 99.9% confidence interval (Table III). The same group of pools also showed the highest statistical significance when the data was stratified to include only those individuals who expressed at least one HLA-DRB1*03 or *04 allele (Supplementary Table III).
Pairwise analysis of the responses to individual peptide pools in PBMCs from subjects expressing HLA-DR4 but not HLA-DR3 revealed that 6/23 showed T1D association, with pools 1 and 17 having the greatest statistical significance (Supplementary Table III). Comparison with the results of the in vitro binding assay indicated that the positive responses to pools 1 and 17, which were each detectable in 9/19 (47%) of subjects with T1D, might be explained by DR4-binding peptides (Fig. 3B). Consistent with the lower level of in vitro binding to HLA-DR3 (Fig. 3A), analysis of the responses to individual peptide pools in PBMCs from subjects expressing HLA-DR3 but not HLA-DR4 revealed only 4/23 that showed significant disease association (Supplementary Table III). The most significant of these (pool 21) gave positive responses in 4/6 (66.7%) subjects with T1D, but failed to elicit a positive response in 12/12 control subjects (p = 0.0049). Consistent with the involvement of this molecule, pool 21 contains a peptide that showed modest in vitro binding to HLA-DR3 (Fig. 3A).
In addition to the disease association in individuals who express HLA-DR4 but not HLA-DR3, pool 17 also showed significant disease association in subjects expressing HLA-DR3 but not HLA-DR4 (Supplementary Table III). Overall this pool could detect 57.1% of T1D subjects with 95.1% specificity (Table III). However, the most effective assay was obtained when analysis was restricted to the 6 non-overlapping pools that showed the most significant disease association (pools 1, 5, 9, 17, 19, and 21), (Fig. 4). In the control group the median number of positive pools was 0 (range 0 – 2), compared to 2.0 (range 0 – 6; p < 0.0001) in PBMCs from the patients. Using a randomly selected cut-off pre-defined as the mean + 3SD of the control group (1.9 pools) the assay showed 74.3% sensitivity at 97.5% specificity (Fig. 4A). A similar result was obtained when only those individuals expressing one or more HLA-DQ2 or -DQ8 molecule were included (Fig. 4B), although in this case the calculated cut-off was marginally higher (2.03), giving an assay with 46.7% sensitivity at 100% specificity.
The results of this initial study of cellular autoimmunity to ZnT8 clearly suggest that, like the other “gold-standard” humoral targets, ZnT8 is a significant target of autoimmune T cells in human T1D. Based on overall activity, over 68% of patients but less than 8% of controls, showed significantly expanded numbers of ZnT8-specific pro-inflammatory T cells in their peripheral blood. The presence of positive responses in some controls may appear surprising, but has often been reported in studies of human T cell responses to other diabetes autoantigens (reviewed by (1)), and may be indicative of the inherent immunogenicity of the antigen. In addition, it should be noted that some of the control subjects are autoantibody negative first-degree relatives of diabetic subjects, who might be more prone to islet autoimmunity than those with more protective genotypes. Using an established cut-off, all of the newly diabetic subjects tested responded to at least one peptide from ZnT8, with the majority exhibiting elevated numbers of pro-inflammatory T cells specific to multiple non-contiguous regions of the protein, likely indicative of “epitope-spreading.” Moreover, the magnitude of the signal elicited by individual pools among responders was also considerably greater in the patients than controls. The time course of the ELISPOT assay is too short to allow significant maturation of naive cells to occur (29). Consequently the responses we observed presumably reflect the presence of elevated pools of ZnT8-specific effector and/or memory T cells in the peripheral blood of the diabetic subjects, with the most likely explanation for this being that it is indicative of a role of ZnT8-specific T cells in diabetogenesis. However, as is also true for all other human T cell responses to diabetes autoantigens, the relationship between peripheral T cells and those in the target organ remains a subject of debate (for example (30, 31)). Nevertheless, consistent with an antigen-driven process, post-onset the peripheral frequency of ZnT8-specific IFN-γ producing T cells appeared greatest within 6 months of clinical diagnosis, with those individuals re-tested at later ages typically showing a much reduced response (data not shown). Given that cross-presentation of free 20mer peptides is typically inefficient (32), we assume that the responses we observed were predominantly, if not exclusively, derived from CD4+ T cells. However, as the library used may contain some truncated peptides we cannot entirely exclude the possibility that CD8+ T cells may also have contributed, although our recent studies using highly purified peptides suggest that this is unlikely to have significantly impacted the results we obtained.
When considered individually, 13 of the 23 di-peptide pools showed statistically significant disease association at the 95% confidence level. Although at first glance this may seem unexpectedly high, the human MHC is highly complex, with any individual expressing up to 12 different class II molecules, most if not all of which have the potential to bind multiple peptides from ZnT8. The existence of ZnT8-specific T cells restricted to many different class II molecules, both within an individual, and in the control and patient populations as a whole, likely explains both the breadth of the auto-response we observed, and the fact that all of the pools gave positive responses in at least 11% of the patients tested. Indeed the degree of complexity of the response only increased when the components of the most reactive pools were tested individually in a separate cohort of patients (data not shown). To date we have only conducted direct binding studies with two of the HLA-DR molecules expressed by the subjects in our study. However, the level of T cell reactivity we observed appears consistent with in silico predictions that suggest that most of the HLA-DR molecules expressed by our subjects, including some of those deemed “protective”, have the capacity to bind peptides from multiple pools of the library. The bioinformatics based predictions suggest that ZnT8 is inherently immunogenic, and the magnitude of the response we observe implies that its potential as a target of cellular immunity is realized in many patients, perhaps because ZnT8-specific T cells are subject to minimal negative selection. For tissue-specific antigens such as proinsulin central tolerance is believed to depend upon thymic and extra-thymic expression driven by transcriptional activators such as AIRE (33, 34) and Deaf1 (35). Interestingly, ZnT8 (slc30a8) was not identified among the genes selectively expressed by mouse thymic medullary epithelial cells (36) and we are unaware of any evidence of thymic expression in humans. Thus it is tempting to speculate that tolerance to ZnT8 is largely the result of immunological ignorance, and that once this is overcome a robust response may ensue.
That said, it is possible that the actual diversity of the auto-response might be slightly exaggerated by our methodology, with the overlap of the constituent peptides in consecutive pools meaning that some of the disease-associated responses we observed might be non-redundant. For example, part or all of the reactivity to pool 18 could be due to epitopes also in pools 17 and/or 19. Nonetheless, our data strongly suggests the presence of a minimum of 9 disease associated T cell epitopes in human ZnT8. Coincidentally, this value is similar to those obtained in studies using overlapping peptide libraries encompassing GAD65 (37), and the cytoplasmic domain of IA-2 (38). The fact that the HLA haplotypes exhibited by the 2 groups were not identical means that we cannot discount the possibility that variations in detection limits within the 2 groups might have influenced the magnitude of the differences in overall reactivity that we observed. However, its contribution is clearly insufficient to account for the results we obtained, and a highly statistically significant disease association was also evident when our analysis was limited to non-overlapping peptide pools (Fig. 1F, ,4),4), when this concern no longer applies.
Our decision to utilize a library of overlapping peptides rather than focusing solely on those predicted to bind to high risk alleles was intended to ensure an unbiased analysis of the overall proinflammatory CD4+ response in patients and controls. However, the fine mapping of epitopes is highly problematic using this experimental approach, and at present we can neither be certain of the total number of immunodominant ZnT8 epitopes, nor of their restriction to any particular class II molecule. Nevertheless, our results clearly indicate that like ZnT8A, ZnT8-directed T cell autoimmunity is associated with, but not restricted to, both the DR4/DQ8 and DR3/DQ2 high-risk haplotypes. In particular the HLA-DR4 (0401) molecule, which was expressed by 19/35 (54.3%) of the diabetic group, appears to be a major element in ZnT8 autoimmunity, with several peptides that bind this molecule including ZnT88–22 and ZnT815–29 (pool 1), ZnT8120–134 and ZnT8134–148 (pool 9), ZnT8260–274 (pool 17), ZnT8267–281 (pools 17 and 18), and ZnT8295–309 (pool 19) possibly containing disease-related epitopes. Similarly, ZnT8155–169 (pool 10) and ZnT8323–337 (pool 21) may contain HLA-DR3 restricted epitopes.
In a previous study of autoreactivity to IA-2 and proinsulin, Arif and colleagues reported a reciprocal polarization towards IL-10 or IFN-γ production for peripheral T cells from patients and controls specific for the same epitope (14). In the present study we were unable to measure both IFN-γ and IL-10 responses to the entire peptide library in the majority of our study subjects due to the limited amount of blood available, although preliminary results from the sub-set tested for both cytokines revealed essentially equivalent ZnT8-specific IL-10 responses in both patients and controls (data not shown). Our identification of 7 key disease associated ZnT8 peptide pools will now enable us to address this and other important questions relating to the role of ZnT8 in T1D pathogenesis, such as the temporal appearance/disappearance of ZnT8-specific T cells during disease progression, and whether ZnT8 is also a significant target of “natural” regulatory T cells. It also provides the basis for the development of novel reagents that either alone or in combination could be used for antigen-specific therapeutic intervention to arrest the progression of the disease.
This work was supported by a Juvenile Diabetes Research Foundation autoimmunity prevention consortium center grant (4-2007-1056), NIH R01 DK052068 (to JCH), NIH R56 DK052068 (to JCH and HWD), NIH P30 DK57516 “University of Colorado Health Sciences Center Diabetes and Endocrinology Research Center”, and the Children’s Diabetes Foundation of Denver.
We gratefully acknowledge the patients, relatives, and volunteers attending the Barbara Davis Center for Childhood Diabetes Clinic who donated blood for this study, and Whitney Kastelic, Jesse Temple-Trujillo, and Rachael Jenison for co-ordinating sample collection. We also thank Danny Zipris and George Eisenbarth for many helpful discussions, Sunanda Babu, Taylor Armstrong, Lisa Fitzgerald-Miller, and Lisa Frisch for technical assistance, and Kim McFann for statistical advice.