Research ArticleAUTOIMMUNITY

Islet-reactive CD8+ T cell frequencies in the pancreas, but not in blood, distinguish type 1 diabetic patients from healthy donors

See allHide authors and affiliations

Science Immunology  02 Feb 2018:
Vol. 3, Issue 20, eaao4013
DOI: 10.1126/sciimmunol.aao4013

At home in the pancreas

Type 1 diabetes (T1D) is associated with enrichment of autoreactive CD8+ T cells that target destruction of pancreatic islets. Culina et al. studied islet-reactive CD8+ T cells reactive to the zinc transporter 8186–194 (ZnT8186–194) and other islet epitopes in healthy individuals and T1D patients, which showed similar functionality and similar frequencies and naïve phenotypes in the peripheral circulation across both groups. In contrast, ZnT8186–194-reactive CD8+ T cells were enriched in the pancreas of T1D patients relative to healthy controls and showed cross-reactivity to an epitope from the commensal Bacteroides stercoris. These results indicate that incomplete central tolerance may allow the survival of these islet-reactive CD8+ T cells in the periphery, and that proinflammatory conditions in the islets can contribute to T1D progression.


The human leukocyte antigen–A2 (HLA-A2)–restricted zinc transporter 8186–194 (ZnT8186–194) and other islet epitopes elicit interferon-γ secretion by CD8+ T cells preferentially in type 1 diabetes (T1D) patients compared with controls. We show that clonal ZnT8186–194-reactive CD8+ T cells express private T cell receptors and display equivalent functional properties in T1D and healthy individuals. Ex vivo analyses further revealed that CD8+ T cells reactive to ZnT8186–194 and other islet epitopes circulate at similar frequencies and exhibit a predominantly naïve phenotype in age-matched T1D and healthy donors. Higher frequencies of ZnT8186–194-reactive CD8+ T cells with a more antigen-experienced phenotype were detected in children versus adults, irrespective of disease status. Moreover, some ZnT8186–194-reactive CD8+ T cell clonotypes were found to cross-recognize a Bacteroides stercoris mimotope. Whereas ZnT8 was poorly expressed in thymic medullary epithelial cells, variable thymic expression levels of islet antigens did not modulate the peripheral frequency of their cognate CD8+ T cells. In contrast, ZnT8186–194-reactive cells were enriched in the pancreata of T1D patients versus nondiabetic and type 2 diabetic individuals. Thus, islet-reactive CD8+ T cells circulate in most individuals but home to the pancreas preferentially in T1D patients. We conclude that the activation of this common islet-reactive T cell repertoire and progression to T1D likely require defective peripheral immunoregulation and/or a proinflammatory islet microenvironment.


In the setting of type 1 diabetes (T1D), insulitic lesions are enriched for CD8+ T cells, which are held as the final mediators of islet destruction. Concordantly, preproinsulin (PPI)–reactive CD8+ T cell clones can lyse β cells in vitro (1), and β cell–reactive CD8+ T cells infiltrate the islets of T1D patients (2). Autoimmune CD8+ T cells may therefore provide biomarkers for disease staging complementary to autoantibodies (aAbs). Although interferon-γ (IFN-γ)–secreting CD8+ T cells detected by enzyme-linked immunospot (ELISpot) distinguish T1D patients from healthy donors (3), the situation is more complex when nonfunctional human leukocyte antigen (HLA) class I multimer (MMr) assays are used. Although MMr+CD8+ T cells were often (4), but not invariably (5, 6), found at similar frequencies in both T1D and healthy individuals, they have been reported to exhibit more differentiated effector/memory phenotypes (4, 6) in T1D patients. A rather enigmatic state of “benign” autoimmunity therefore exists in healthy individuals.

To extend these observations, we aimed to determine the key features of islet-reactive CD8+ T cells that associate with T1D. We focused our efforts on immunodominant HLA-A*02:01 (HLA-A2)–restricted epitopes derived from PPI, glutamic acid decarboxylase (GAD), insulinoma-associated protein-2 (IA-2), and islet-specific glucose-6-phosphatase catalytic subunit-related protein (IGRP) (3), and on a highly immunoprevalent zinc transporter 8186–194 (ZnT8186–194) epitope that we recently described (7). The results indicate that incomplete central tolerance mechanisms allow the survival of an islet-reactive CD8+ T cell repertoire, which can be primed in the presence of defective peripheral immunoregulation and/or a proinflammatory islet microenvironment to progress toward T1D.


ZnT8186–194-reactive CD8+ T cell clones from T1D and healthy donors display equivalent functionality

Given that ZnT8186–194-reactive IFN-γ responses are highly prevalent in T1D patients (7), we started by generating ZnT8186–194-reactive CD8+ T cell clones from a new-onset T1D patient (D222D) (7). After in vitro expansion with the ZnT8186–194 peptide (8), HLA-A2 MMr+ cells were labeled with two different fluorochromes (9) and sorted (Fig. 1A). The three clones thus generated stained uniformly positive with ZnT8186–194 MMrs (Fig. 1B) and responded to ZnT8186–194 peptide stimulation by secreting tumor necrosis factor–α (TNF-α) (Fig. 1C), IFN-γ, interleukin-2 (IL-2), and, to a lesser extent, macrophage inflammatory protein-1β (MIP-1β) in a dose-dependent fashion (fig. S1, A to C). Cytotoxicity was then tested against an HLA-A2+ Epstein-Barr virus–transformed B-lymphoblastoid cell line (LCL) pulsed with the cognate ZnT8186–194 peptide. Increasing numbers of clonal CD8+ T cells led to the complete disappearance of ZnT8186–194-pulsed but not control-pulsed targets, with ≥90% lysis at a 1:2 effector/target (E/T) ratio (Fig. 1, D and E, and fig. S1, D to F). This lytic activity was mostly perforin-mediated (Fig. 1F) because it was inhibited significantly by concanamycin A, marginally by brefeldin A (suppressing cytokine secretion), and not at all by a blocking anti-FasL monoclonal antibody (mAb) (suppressing Fas-dependent cytotoxicity), and it was associated with CD107a up-regulation (Fig. 1G).

Fig. 1 ZnT8186–194-reactive CD8+ T cell clones from patient D222D.

(A) Frozen-thawed PBMCs were cultured with ZnT8186–194 or no peptide and stained with phycoerythrin (PE)/allophycocyanin (APC)–labeled ZnT8186–194 MMrs. (B) ZnT8186–194 and control MMr stains for one clone obtained from single-sorted ZnT8186–194/ZnT8186–194 double-MMr+ cells. (C) Percentage of intracellular TNF-α+ D222D clone 1 cells stimulated for 6 hours with K562-A2 cells pulsed with ZnT8186–194 or Flu MP58–66 peptide. (D) Percent lysis of Far Red–labeled LCL targets pulsed with ZnT8186–194 (top) or Flu MP58–66 peptide (bottom) and cultured for 24 hours with CFSE-labeled D222D clone 3 at increasing E/T ratios. (E) Percent lysis of LCL targets cultured with D222D clone 1, 2, or 3 (mean ± SEM; each clone is depicted in fig. S1, D to F). (F) Lysis of cognate peptide–pulsed LCLs cultured for 4 hours with D222D clone 2 or a MelanA26–35-reactive clone (E/T, 1:1) in the presence of concanamycin A (CMA), brefeldin A (BFA), CMA and BFA, anti-FasL (aFasL), or control IgG1 (immunoglobulin G1). *P = 0.015, **P = 0.009, and ***P < 0.001 by Student’s t test. Results are mean ± SEM of triplicate measurements from one of three experiments. (G) Percentage of surface CD107a+ D222D clone 1 cells stimulated as in (C). For (A), (C), and (G), the gate is on viable CD8+ cells.

Additional ZnT8186–194-reactive CD8+ T cell clones were generated from other T1D patients and healthy donors. The final panel comprised 16 clones (9 clones from five T1D patients and 7 clones from five healthy donors), most of which were isolated directly ex vivo (table S1). All but one clone (H328C 9B3) stained with HLA-A2 MMrs loaded with ZnT8186–194 (VAANIVLTV), and several of them also with a longer ZnT8185–194 variant (AVAANIVLTV) reported to exhibit similar immunoprevalence (fig. S2A) (10). Higher staining intensities were observed with ZnT8186–194 MMrs for all clones barring D010R 1D3. Concordantly, ZnT8186–194 bound to recombinant HLA-A2 molecules with higher affinity (KD, 15 nM versus 207 nM) and similar stability (t1/2, 1.8 hours versus 2.3 hours) relative to the longer ZnT8185–194 peptide (fig. S2, B and C). The ZnT8186–194 epitope was therefore retained for subsequent experiments, except for clone D010R 1D3.

Clones from T1D and healthy donors were first compared for the intensity of ZnT8186–194 MMr staining. The tyrosine kinase inhibitor dasatinib, which stabilizes T cell receptor (TCR) interactions with peptide-HLA complexes (11), enhanced MMr staining, particularly for MMrlow clones (Fig. 2A). Heterogeneous staining patterns were observed when comparing normalized MMr fluorescence intensities (Fig. 2B). Although dasatinib reduced variability, even in its absence, no differences were apparent between the T1D and healthy groups.

Fig. 2 Ag avidity, Ag sensitivity, and polyfunctionality of ZnT8186–194-reactive CD8+ T cell clones.

(A) ZnT8186–194 MMr staining in the absence (light gray) or presence (dark gray) of dasatinib. The dotted profile indicates the unstained control. (B) ZnT8186–194 MMr MFI for the indicated clones in the absence (left) or presence (right) of dasatinib. Bars indicate median values. Results are representative of two separate experiments. (C) The indicated clones were stimulated for 6 hours with ZnT8186–194-pulsed K562-A2 cells, and the percentage of cytokine+ cells out of viable CD8+ cells was calculated. Results are representative of three independent experiments. (D and E) EC50 (D) and maximal cytokine response (percent cytokine+ cells at optimal peptide concentrations) (E) for clones stimulated as above. Bars indicate median values. Results are representative of two to four separate experiments. *P = 0.014 by Mann-Whitney test. (F) Polyfunctionality distribution of T1D (left) and healthy clones (right). Percentage of T cells producing zero to four cytokines among TNF-α, IFN-γ, IL-2, and MIP-1β upon exposure to ZnT8186–194-pulsed K562-A2 cells (100 μM) are shown.

Next, we performed in vitro recall assays with increasing ZnT8186–194 peptide concentrations. Representative data are shown in Fig. 2C, and results are summarized in Fig. 2 (D and E). In line with the MMr staining profiles, the half-maximal effective peptide concentration (EC50) required to elicit cytokine responses (TNF-α, IFN-γ, and IL-2) and the maximal cytokine responses were again heterogeneous but not significantly different between T1D and healthy clones (Fig. 2, D and E), with one exception noted for the lower MIP-1β EC50 of T1D clones. Clones obtained after in vitro expansion displayed an equivalent range of antigen (Ag) sensitivities, either high (D222D and H017N) or low (H328C), arguing against a bias compared with clones isolated directly ex vivo. Moreover, the polyfunctionality index, which reflects the number of cells secreting multiple cytokines (12), was similar for T1D and healthy clones (Fig. 2F). As expected, EC50 values correlated negatively with MMr staining and polyfunctionality indices (fig. S3). Collectively, these results show that ZnT8-reactive CD8+ T cells isolated from T1D and healthy individuals exhibit similar Ag avidity, sensitivity, and polyfunctionality.

ZnT8186–194-reactive CD8+ T cell clones from T1D and healthy donors display equivalent β cell cytotoxicity

To determine whether ZnT8186–194-reactive CD8+ T cells can recognize naturally processed ZnT8 epitopes, we performed cytotoxicity assays using ZnT8-transduced K562-A2 targets (K562-A2/ZnT8). High-avidity ZnT8186–194-reactive clones lysed unpulsed K562-A2/ZnT8 targets almost as efficiently as targets pulsed with the ZnT8186–194 peptide, whereas unpulsed K562-A2 control targets remained largely intact (Fig. 3, A and B). In contrast, low-avidity ZnT8186–194-reactive clones were unable to lyse K562-A2/ZnT8 targets in the absence of exogenous ZnT8186–194 peptide (Fig. 3C).

Fig. 3 Target cell lysis by ZnT8186–194-reactive CD8+ T cell clones.

(A to C) Lysis of K562-A2 cells transfected (open triangles) or not (open circles) with a full-length ZnT8 plasmid and cultured for 24 hours with clones D222D 2 (A), H017N A1 (B), or H314C 6C4 (C). Filled symbols indicate ZnT8186–194-pulsed target cells (10 μM). Results are presented as mean ± SEM of triplicate wells from two separate experiments. (D to G) Real-time cytotoxicity for the indicated clones versus HLA-A2+ ECN90 (white triangles) or control HLA-A2 EndoC-βH2 β cell targets (white circles) (E/T, 2:1). Black and gray symbols indicate the corresponding targets pulsed with 10 μM ZnT8186–194 or GAD114–122 peptide, respectively [ZnT8186–194 or MelanA26–35 for the H004N clone MelanA in panel (G)]. Means ± SEM of triplicate measurements are shown at each time point. Results are representative of at least two separate experiments. (H) Percent maximal HLA-A2+ ECN90 and HLA-A2 EndoC-βH2 β cell lysis by the indicated clones (T1D, gray symbols; healthy, white symbols; control H004N clone MelanA, horizontal dotted line) in the absence or presence of the ZnT8186–194 peptide. Bars indicate median values. Lysis was calculated from the cytotoxicity profiles as in (D) to (G).

We then evaluated cytotoxicity against T1D-relevant targets by using HLA-A2+ ECN90 and control HLA-A2 EndoC-βH2 human β cell lines. Although the ECN90 but not the EndoC-βH2 line expressed HLA class I in the unstimulated state, expression levels were similarly up-regulated by pretreatment with different cocktails of inflammatory cytokines, without inducing significant β cell death (fig. S4, A and B). IFN-γ was chosen as the single cytokine that up-regulated HLA class I expression to comparable levels in both lines, and pretreatment was carried out for 18 hours before a real-time cytotoxicity assay. As observed with K562-A2/ZnT8 targets, high-avidity clones lysed unpulsed HLA-A2+ ECN90 cells presenting naturally processed ZnT8-derived epitopes (Fig. 3, D and E), whereas low-avidity clones displayed marginal lytic activity (Fig. 3F). Cytotoxicity increased in all cases with the addition of the ZnT8186–194 peptide, suggesting a more limited natural presentation compared with ZnT8-transduced targets. Lysis of HLA-A2 EndoC-βH2 cells was negligible, and a control clone reactive to the melanocyte self-epitope MelanA26–35 only lysed ECN90 cells in the presence of exogenous MelanA26–35 peptide (Fig. 3G). Microscopic inspection confirmed the lysis measured in real-time cytotoxicity assays (fig. S4C). β cell lysis was not different between T1D and healthy clones, either in the absence or in the presence of exogenous ZnT8186–194 peptide (Fig. 3H). Moreover, IFN-γ pretreatment of HLA-A2+ ECN90 cells neither up-regulated ZnT8 protein expression (fig. S4D) nor increased the functional activation of co-incubated CD8+ T cell clones (fig. S4, E and F). Collectively, these results demonstrate that ZnT8186–194-reactive CD8+ T cells display similar cytotoxicity in T1D and healthy individuals and that ZnT8 expression is not modulated by inflammation.

ZnT8186–194-reactive CD8+ T cells display private TCR gene usage but public CDR3β amino acid sequences

Molecular analysis of expressed TCRα (TRA) and TCRβ (TRB) gene transcripts revealed that no sequences were shared (public) among ZnT8186–194-reactive clones isolated from T1D or healthy individuals (fig. S5). The three clones from patient D222D harbored an identical TCR. However, this observation did not affect our previous functional comparisons because the measured parameters were even more similar between T1D and healthy donors when only one D222D clone was considered.

We then interrogated an in silico database of TRB sequences compiled by high-throughput sequencing of naïve/terminal effector (CD45RA+CD45RO) or central memory (CD45RO+CD45RACD62Lhi) CD4+ and CD8+ T cells obtained from HLA-A2+ recent-onset T1D patients, at-risk aAb+ individuals, and healthy individuals (table S2). The D010R 1E2, H328C 8E8, and H034O 141B9 complementarity-determining region 3β (CDR3β) amino acid sequences were found among CD8+ and CD4+ T cells isolated from several individuals (Fig. 4, A to C), mostly within the CD45RA+CD45RO pool for CD8+ T cells. Several different TRBV genes were used in conjunction with these identical CDR3β loops to generate “mosaic” sequences. The same CDR3β amino acid sequences were detected in silico among the polyclonal TCR repertoires compiled from conventional and regulatory CD4+ and CD8+ T cells isolated from pancreatic lymph node (PLN) and spleen samples by the Network for Pancreatic Organ Donors with Diabetes (nPOD) (n = 67 identical CDR3β; n = 6 identical TCRβ; Fig. 4D). Most of the CD8+ T cell hits in PLNs (11 of 15, 73%) were from HLA-A2+ patients but with no obvious association with T1D.

Fig. 4 In silico search for CDR3β amino acid sequences from ZnT8186–194-reactive CD8+ T cell clones.

(A to C) Prevalence of the CDR3β amino acid sequences from clones D010R 1E2 (A), H328C 8E8 (B), and H034O 141B9 (C) among HLA-A2+ T1D (n = 5), aAb+ (n = 5), and healthy individuals (n = 10), as assessed by in silico analysis of TCRβ repertoires obtained from the indicated CD8+ and CD4+ T cell subsets. (D) In silico search for the same CDR3β amino acid sequences in the repertoire of CD8+, conventional CD4+ (Tconv; CD127+), and regulatory CD4+ (Treg; CD25+CD127) T cells obtained from nPOD PLN, spleen, and inguinal lymph node (ILN) samples via the online database For each cell type and tissue, the first, second, and third columns refer to clones D010R 1E2, H034O 141B9, and H328C 8E8, respectively. Dark and light gray cells indicate negative and positive samples, respectively. Frequencies per 106 TCRs are annotated, and underlining indicates samples with a nucleotide sequence match. White cells indicate unavailable samples. Pancreatic NET, neuroendocrine tumor.

To extend these findings, we developed D010R 1E2, H328C 8E8, and D222D TRA and TRB clonotype–specific TaqMan real-time quantitative PCR (qPCR) assays (fig. S6, A and B). Applied to CD4+ and CD8+ T cell complementary DNA (cDNA) preparations from two independent cohorts of T1D patients (n = 97 and n = 53) and healthy individuals (n = 97 and n = 38), these assays detected the D010R and H328C TRB among CD8+ T cells from a single HLA-A2+ T1D patient in each case (fig. S6, C and D). Collectively, these results show that ZnT8186–194 recognition is mediated primarily by private clonotypes, some of which share CDR3β amino acid sequences among individuals to form mosaic TCRs.

Circulating islet-reactive CD8+ T cells display similar ex vivo frequencies and a predominantly naïve phenotype in T1D and healthy individuals

In further experiments, we used ex vivo combinatorial HLA-A2 MMr assays (9) to analyze ZnT8186–194-reactive CD8+ T cells in 39 HLA-A2+ recent-onset T1D patients (16 children and 23 adults) and 39 age- and sex-matched healthy donors (17 children and 22 adults) (table S3 and fig. S7). Control specificities included MelanA26–35, which is recognized by a large naïve pool in humans (13), and Flu MP58–66, to which most individuals harbor Ag-experienced CD8+ T cells reflecting previous viral exposure. Importantly, donors yielding <3 × 105 total CD8+ T cells or <5 MMr+ cells were excluded from the analysis to avoid undersampling. Parallel ELISpot assays confirmed that ZnT8186–194-reactive IFN-γ responses were more frequent in T1D versus healthy donors (fig. S8) (7). In contrast, similar frequencies of ZnT8186–194 and MelanA26–35 MMr+CD8+ T cells were detected in age-stratified T1D and healthy donors (Fig. 5A). Healthy children and adults displayed higher frequencies of Flu MP58–66 MMr+CD8+ T cells compared with their age- and sex-matched T1D counterparts, whereas children displayed about fourfold higher frequencies of ZnT8186–194 and MelanA26–35 MMr+CD8+ T cells compared with adults, irrespective of T1D status.

Fig. 5 Ex vivo frequencies and Ag-experienced phenotypes of circulating islet-reactive CD8+ T cells.

(A) ZnT8186–194, MelanA26–35, and Flu MP58–66 MMr+CD8+ cells were stained ex vivo and counted (see fig. S7). Frequencies out of total CD8+ T cells are depicted for T1D adults (red circles), T1D children (crossed red circles), age- and sex-matched healthy adults (blue circles), and children (crossed blue circles). *P ≤ 0.05, **P = 0.002, and ***P ≤ 0.0003. (B) Percentage of Ag-experienced cells out of total MMr+ cells. *P ≤ 0.03, **P = 0.004, and ***P = 0.0007. (C) Absolute frequencies of the corresponding Ag-experienced fractions. *P ≤ 0.03, **P ≤ 0.01, and ***P ≤ 0.0001. (D) MMr+CD8+ cells reactive to the indicated islet epitopes were stained ex vivo and counted (see fig. S10A). Frequencies out of total CD8+ T cells are depicted as in (A). *P = 0.02. (E) Percentage of Ag-experienced cells out of total MMr+ cells. (F) Absolute frequencies of the corresponding Ag-experienced fractions. Bars display median values. The median number of MMr+ events and total CD8+ T cells analyzed are indicated for each distribution. Significance was determined using the Mann-Whitney test. For (A) and (D), data points with <300,000 CD8+ T cells and <5 MMr+ cells were excluded. For (B), (C), (E), and (F), data points with <5 MMr+ cells were excluded.

Among ZnT8186–194 MMr+CD8+ cells, the Ag-experienced fraction (CD45RA+CCR7 or CD45RACCR7+/−) was similarly limited (generally ≤25% of all ZnT8186–194 MMr+ events) in T1D and healthy adults (Fig. 5B). In contrast, higher proportions of Ag-experienced ZnT8186–194 MMr+ cells were present in T1D children versus adults, although these values were not different in healthy children. T1D children also harbored significantly higher proportions of Ag-experienced ZnT8186–194 versus MelanA26–35 MMr+ cells. MelanA26–35-reactive CD8+ T cells were mostly naïve, whereas Flu MP58–66-reactive CD8+ T cells were mostly Ag-experienced in all groups. Comparable results were obtained using the absolute frequencies of Ag-experienced MMr+CD8+ T cells (Fig. 5C).

Single-cell gene expression analysis of sorted ZnT8186–194 MMr+ cells (fig. S9A) revealed few differences, with T1D patients displaying higher expression of the T helper 17–related aryl hydrocarbon receptor (AHR) and the mitotic aurora kinase A (AURKA) and lower expression of the transcriptional activator RAR-related orphan receptor A (RORA). In line with our in vitro data, clonotypic analyses of individual ZnT8186–194 MMr+CD8+ T cells sorted ex vivo yielded unique CDR3α and CDR3β sequences (fig. S9B). Of the 21 CDR3β amino acid sequences obtained, 7 (33%) were found in our in silico TRB database across all groups (fig. S9C). Among CD8+ T cells, these seven sequences were again most frequent within the CD45RA+CD45RO compartment and predominantly found in T1D versus healthy individuals. Preferential usage of certain TRBV (mostly TRBV19, 25%) and TRAV (mostly TRAV12-2, 38%) genes was observed (fig. S9, D and E), with TRBV19 also shared among the ZnT8186–194-reactive clones. However, only 8 of 63 matching sequences (13%) expressed the TRBV gene of the corresponding ZnT8186–194-reactive T cell. Thus, despite biased TRAV and TRBV gene usage, the ex vivo data confirmed that ZnT8186–194 recognition is mediated primarily by private clonotypes expressing mosaic TCRs.

Next, we used an extended combinatorial MMr panel to compare ZnT8186–194-reactive CD8+ T cells with those recognizing other immunodominant β cell epitopes in adult donors (fig. S10A). Assay reproducibility across panels was confirmed using separate blood draws from the same individuals (fig. S10B). As observed for ZnT8186–194, other β cell–reactive CD8+ T cell populations displayed similar frequencies (typically 1 to 50 MMr+ cells per 106 CD8+ T cells) in T1D and healthy adults (Fig. 5D). An exception was noted for PPI15–24-reactive CD8+ T cells, whose frequencies were lower than those for other β cell–reactive fractions and higher in T1D versus healthy donors. In all instances, the Ag-experienced fraction among MMr+CD8+ T cells was limited in both subject groups (Fig. 5, E and F). The few children included in this extended analysis harbored β cell–reactive CD8+ T cells with a more Ag-experienced phenotype. CD45RA+CCR7+ islet-reactive cells were bona fide naïve because they were largely CD27+CD28+ and CD95 (fig. S11).

Collectively, these results show that circulating CD8+ T cells reactive to ZnT8186–194 and other β cell epitopes occur at similar frequencies and exhibit a predominantly naïve phenotype in T1D and healthy adults, whereas higher frequencies of total and Ag-experienced ZnT8186–194-reactive CD8+ T cells are present in children, irrespective of T1D status.

Poor promiscuous ZnT8 gene expression in human thymic medullary epithelial cells

The combined observations that CD8+ T cells reactive to ZnT8186–194 and other β cell epitopes are predominantly naïve and circulate at similar frequencies in T1D and healthy donors are compatible with frequent escape from thymic deletion due to poor islet Ag expression in thymic medullary epithelial cells (mTECs) (14). We tested this hypothesis by examining ZnT8 (SLC30A8) gene expression in total or immature (AIREHLA class IIlo) versus mature (AIRE+HLA class IIhi) human mTECs from five children undergoing heart surgery. As misinitiated mRNA transcription is described in mTECs (13, 15), forward primers were designed to identify potential alternative start sites via hybridization with exons 5 to 8 of the SLC30A8 gene (Fig. 6A). The ZnT8186–194 region is encoded by exons 7 and 8, with the exon 8 primer located just downstream. Combinations of these forward primers with reverse primers located either in exon 11 (Fig. 6B) or in the 3′ untranslated region (3′UTR) (Fig. 6C) yielded the expected bands using human islet mRNA. However, mTECs did not express SLC30A8. One sample (#64) displayed a faint band amplified with the exon 8 forward primer, which matched SLC30A8 by sequencing, suggesting low-level misinitiated transcription starting at exon 8 (i.e., downstream of the ZnT8186–194 region). Collectively, these results demonstrate that ZnT8 is poorly expressed in mTECs and that SLC30A8 transcription is limited to a misinitiated mRNA isoform skipping the ZnT8186–194 sequence.

Fig. 6 SLC30A8 and INS gene expression in mTECs and circulating islet-reactive CD8+ T cell frequencies in HLA-A2+ and HLA-A2 healthy donors.

(A) SLC30A8 RT-PCR strategy. Forward primers spanned exons 5 to 8, and reverse primers spanned either exon 11 or the 3′UTR. The position of the ZnT8186–194-coding region is shown. (B) SLC30A8 expression in mTECs, using the indicated forward primers and the exon 11 reverse primer. (C) SLC30A8 expression in mTECs from donor #64 (previously testing positive with the exon 8 forward primer) and #211 (previously testing negative), using the 3′UTR reverse primer. bp, base pair. (D) INS RT-PCR strategy. Forward primers spanned both or neither of the PPI6–14 and PPI15–24 regions, and reverse primers spanned either the 3′UTR or exon 2 (PCR products 1 and 2, respectively). (E) INS expression in thymuses pooled from five to eight donors. (F) Ex vivo MMr+CD8+ cell frequencies in age- and sex-matched, EboV- and HCV-seronegative HLA-A2+ and HLA-A2 healthy donors. (G) Percent Ag-experienced MMr+ cells. Bars indicate median values. The median number of MMr+ events is indicated, with a median of 1 × 106 total CD8+ T cells analyzed. *P ≤ 0.03, **P = 0.008, and ***P ≤ 0.0004 by Mann-Whitney test. For (G), data points with <5 MMr+ cells were excluded. NA, not available.

Islet-reactive CD8+ T cells circulate at similar frequencies irrespective of thymic expression of their cognate epitopes

In contrast to ZnT8, other β cell Ags are expressed by mTECs, including PPI (16, 17), IA-2 (17, 18), and IGRP (19), or by thymic Ag-presenting cells, including GAD65 (20). Moreover, circulating PPI6–14-reactive CD8+ T cells displayed higher frequencies than PPI15–24-reactive CD8+ T cells, despite the fact that the complete INS transcript incorporating both epitopes was detected in the thymus (Fig. 6, D and E). In conjunction with the finding that circulating CD8+ T cells reactive to ZnT8186–194 and other islet epitopes also occur at similar frequencies (Fig. 5D), these observations suggest a limited role of thymic Ag gene expression in setting such frequencies.

Potential confounders in this scenario include the efficiency of epitope presentation and the odds of encountering cognate peptide-HLA complexes in the thymic environment. Accordingly, we compared the frequencies of islet-reactive CD8+ T cells in age- and sex-matched HLA-A2+ and HLA-A2 healthy adults, reasoning that thymic deletion cannot occur in the absence of the appropriate HLA-A2 restriction (21). To exclude promiscuous presentation by non–HLA-A2 molecules, we selected donors based on HLA-A and HLA-B molecules incapable of binding the selected islet epitopes (table S4). Moreover, all patients were seronegative for Ebola (EboV) and hepatitis C virus (HCV). Representative flow cytometry plots are shown in fig. S12. Except for PPI6–14, HLA-A2–restricted islet-reactive CD8+ T cells occurred at largely equivalent frequencies in HLA-A2+ and HLA-A2 donors (Fig. 6F). Although higher overall relative to most islet specificities, the frequencies of MMr+CD8+ cells recognizing the HLA-A2–restricted foreign epitopes EboV NP202–210 and HCV PP1406–1415 were also similar between HLA-A2+ and HLA-A2 groups. No significant phenotypic differences were observed between groups for any of these MMr+CD8+ T cell populations (Fig. 6G). As expected, control Flu MP58–66 MMr+CD8+ cells were more prevalent and more Ag-experienced in HLA-A2+ donors. Collectively, these findings suggest that thymic presentation of these islet epitopes does not trigger significant clonal deletion in HLA-A2+ donors.

ZnT8186–194-reactive CD8+ T cells cross-recognize a Bacteroides stercoris mimotope

Although islet-reactive CD8+ T cells were predominantly naïve in T1D and healthy adults, substantial Ag-experienced fractions were noted in some individuals, irrespective of disease status (Fig. 5, B and E). Moreover, CD8+ T cell frequencies correlated with the size of the Ag-experienced fraction for some individual (ZnT8186–194 and IGRP265–273) and pooled islet epitopes (fig. S13, A to C), consistent with the patterns observed for Flu MP58–66-reactive CD8+ T cells (fig. S13D). These observations raise the possibility of cross-priming by unrelated homologous epitopes (mimotopes). Sequence homology was observed between the ZnT8186–194 epitope (VAANIVLTV) and a peptide (KAANIVLTV) derived from protein WP_060386636.1 of the intestinal commensal B. stercoris. To assess potential cross-reactivity, we performed ex vivo MMr assays on duplicate peripheral blood mononuclear cell (PBMC) samples. One sample was stained with pairs of ZnT8186–194 MMrs, whereas the other was stained with one ZnT8186–194 MMr and one B. stercoris MMr. Similar frequencies were detected for ZnT8186–194-ZnT8186–194 and ZnT8186–194B. stercoris double-MMr+CD8+ T cells in three of four donors (Fig. 7A). As a negative control, ZnT8186–194–EboV NP202–210 double-MMr+CD8+ T cells were undetectable (Fig. 7B), whereas EboV NP202–210–EboV NP202–210 double-MMr+CD8+ T cells were present at similar frequencies in duplicate samples (Fig. 7C). Cross-reactivity with the B. stercoris mimotope was confirmed for one of four ZnT8186–194-reactive CD8+ T cell clones via MMr costaining (Fig. 7D) and in vitro recall (Fig. 7E), with the B. stercoris mimotope displaying stronger agonist potency than the native ZnT8186–194 peptide. Collectively, these results show that ZnT8186–194-reactive CD8+ T cells can cross-recognize a bacterial mimotope.

Fig. 7 ZnT8186–194-reactive CD8+ T cells cross-recognize a B. stercoris mimotope.

(A to C) Four donors with sizable ZnT8186–194 MMr+CD8+ T cell fractions were selected. A first PBMC aliquot received PE/BV786-labeled ZnT8186–194 MMrs and PE/BV711-labeled EboV NP202–210 MMrs. For the second aliquot, PE-labeled B. stercoris MMrs replaced the PE-labeled ZnT8186–194 MMrs. (A) Overlay of ZnT8186–194/ZnT8186–194 MMr+ (blue) and ZnT8186–194/B. stercoris MMr+ cells (red). (B) Negative control staining of ZnT8186–194/EboV NP202–210 MMr+ cells. (C) Positive control staining of EboV NP202–210/EboV NP202–210 MMr+ cells from the first and second aliquot. The frequencies of MMr+ out of total CD8+ T cells are indicated. (D) Four ZnT8186–194-reactive CD8+ T cell clones (D222D 2, D349D 178B9, H017N A1, and H328C 9C8) were stained with BV786/PE-labeled ZnT8186–194, PE-labeled B. stercoris, and BV650-labeled MelanA26–35 MMrs. The ZnT8186–194/B. stercoris cross-reactive clone H017N is shown, from left to right: ZnT8186–194/B. stercoris MMr+, ZnT8186–194/ZnT8186–194 MMr+, negative control ZnT8186–194/MelanA26–35, and B. stercoris/MelanA26–35 MMr+ cells. (E) The H017N clone was stimulated with peptide-pulsed LCLs (0.1 μM; 6 hours). Percentages of cytokine+ cells are shown as mean ± SEM of two experiments. P = 0.008 for Wilcoxon signed-rank comparison of pooled cytokine responses between B. stercoris and ZnT8186–194, MelanA26–35, or no peptide and between ZnT8186–194 and MelanA26–35 or no peptide.

ZnT8186–194-reactive cells are enriched in the pancreas of T1D patients

To reconcile the finding that equivalent frequencies of predominantly naïve islet-reactive CD8+ T cells circulate in most individuals, we hypothesized that the T1D-relevant fraction may be sequestered in the pancreas. We therefore performed in situ ZnT8186–194 MMr staining on frozen pancreatic sections from HLA-A2+ T1D (n = 9), aAb+ (n = 9), nondiabetic (n = 11), and type 2 diabetes (T2D) cases (n = 3) from nPOD (table S5). Representative images are shown in Fig. 8 (A to E), with scattered ZnT8186–194 MMr+ cells either within islets or within the exocrine tissue. Consecutive sections from ZnT8186–194 MMr+ pancreata were probed with MelanA26–35 MMrs, which, in conjunction with positive control vitiligo skin sections, confirmed staining specificity (Fig. 8F). Donor-matched PLN sections were stained in parallel (Fig. 8, G and H). Whereas ZnT8186–194 MMr+ cells were significantly more abundant than MelanA26–35 MMr+ cells in T1D, aAb+, and nondiabetic cases, ZnT8186–194 MMr+ cells were enriched in the pancreata of T1D versus nondiabetic and T2D cases (Fig. 8I). In contrast, ZnT8186–194 MMr+ cells were present at similar densities across all groups in PLN sections (Fig. 8J). Several of these nPOD cases were previously analyzed in silico for the presence of ZnT8186–194 CDR3β sequences in PLNs (Fig. 4D). These sequences were present in four of five cases with ZnT8 MMr+ pancreata and three of five cases with ZnT8 MMr+ PLNs (table S5). A donor with chronic pancreatitis (#6288) and very high ZnT8186–194-reactive CD3Rβ counts among spleen and PLN CD8+ T cells also displayed high densities of ZnT8 MMr+ cells in the pancreas. Collectively, these results show that ZnT8-reactive cells are preferentially enriched in the pancreas of T1D patients.

Fig. 8 In situ ZnT8186–194 MMr staining of nPOD pancreas and PLN sections.

(A to E) Representative pancreas images from cases T1D #6161 (A), aAb+ #6347 (B), and nondiabetic (ND) #6289 (C) (magnification, ×20; scale bars, 100 μm). Red arrows indicate MMr+ cells, and the dotted areas of (A) and (B) are magnified in (D) (scale bar, 80 μm) and (E) (scale bar, 50 μm), respectively. (F) Consecutive sections from ZnT8186–194 MMr+ pancreata were probed with negative control MelanA26–35 MMrs. A representative image from T1D case #6211 is shown on the left, and a positive control staining on skin sections from a vitiligo patient is shown on the right (magnification, ×20; scale bars, 100 μm). (G) Representative PLN image (magnification, ×20; scale bar, 100 μm) from T1D case #6161. (H) Magnification of the dotted area of (G) (scale bar, 40 μm). (I and J) Number of ZnT8186–194 and MelanA26–35 MMr+ cells/mm2 section area of pancreas (I) and PLNs (J). Each point represents an individual case (detailed in table S5). Bars indicate median values. *P ≤ 0.05 and **P ≤ 0.009 by Mann-Whitney test. NA, not assessed.


In this study, we found that ZnT8186–194-reactive CD8+ T cell clones exhibited heterogeneous functional profiles but no consistent differences between T1D and healthy individuals. Most ZnT8186–194-reactive clones originated from naïve precursors and expressed private TCRs. Ex vivo analyses on larger donor cohorts revealed that the frequency of circulating ZnT8186–194-reactive CD8+ T cells was similar in age-matched T1D versus healthy donors, but higher in children versus adults, irrespective of T1D status. A similar pattern was noted for CD8+ T cells recognizing the extrapancreatic self-epitope MelanA26–35, but the corresponding ZnT8186–194-reactive populations were more Ag-experienced in T1D children. Thus, although most children harbor a larger autoimmune repertoire not restricted to islet Ags, the activation of the islet-reactive fraction occurs preferentially in T1D children, which may reflect a more aggressive islet autoimmunity leading to earlier disease onset. On the other hand, some Ag-experienced β cell–reactive CD8+ T cells were invariably detected in healthy donors, supporting the possibility that foreign epitopes may prime autoreactive clonotypes expressing promiscuous TCRs (2224). Some ZnT8186–194-reactive CD8+ T cell clonotypes were able to cross-recognize a B. stercoris mimotope. B. stercoris belongs to the Bacteroidetes phylum, which is enriched in the gut microbiome of T1D (25) and at-risk aAb+ individuals (26). Circulating CD8+ T cells reactive to other HLA-A2–restricted β cell epitopes were also detected at equivalent frequencies (1 to 50 MMr+ cells per 106 CD8+ T cells) and with a predominantly naïve phenotype in T1D and healthy adults. This coherent pattern across islet specificities suggests that Ag-driven recruitment involves a limited fraction of naïve precursors, which fits with the paucity of public clonotypes found for ZnT8186–194-reactive CD8+ T cells, as reported for PPI15–24 (4). The lower frequencies of Flu-reactive CD8+ T cells observed in T1D versus healthy donors may reflect impaired antiviral responses (27).

One strength of our ex vivo studies is the highly specific and reproducible quantification of MMr+ cells. The observed frequencies of β cell–reactive CD8+ T cells fall below previous estimates obtained without enrichment (4, 5) but mirror those described using stringent MMr-based magnetic enrichment (22). Although higher frequencies of β cell–reactive CD8+ T cells have been observed in T1D versus healthy donors (5), our data align with another study reporting no difference, both with and without dasatinib enhancement of MMr staining (4). Although the T1D children in our cohort had a longer disease duration than T1D adults, comparable age-related differences were observed in healthy donors, and similar trends remained when restricting the analysis to more recently diagnosed T1D children. Other studies reporting higher Ag-experienced β cell–reactive CD8+ T cell fractions in T1D versus healthy donors are potentially limited by undersampling and single MMr labeling (4). Moreover, such differences were not observed for all islet epitopes (4), and significant naïve fractions (~40%) were also present in T1D patients (4, 6).

T cell precursors can escape thymic negative selection due to “blind spots” that result from poor or incomplete expression of certain tissue Ags (13, 17, 2830), due to alternative splicing and promoter usage and misinitiated transcription (1315, 30). These features favor the generation of truncated transcripts lacking certain T cell epitopes, as observed for ZnT8. However, ZnT8186–194-reactive CD8+ T cells circulated at similar frequencies relative to other islet-reactive populations. Together with the finding that HLA-A2–restricted islet-reactive CD8+ T cells circulated at similar frequencies in HLA-A2+ and HLA-A2 donors, these results show that the thymus does not eliminate all autoreactive CD8+ T cells. Nonetheless, thymic self-Ag expression may “prune” autoreactive clonotypes bearing high-affinity TCRs (22, 31).

Perhaps the most outstanding question raised by our findings relates to the benign state of islet autoimmunity “licensed” by incomplete central tolerance. Evidence for such thymic defects in human T1D has been limited to INS polymorphisms, which rank second among T1D susceptibility loci after DQB1 and modulate thymic INS expression (16). Given the ongoing debate surrounding the role of thymic negative selection for autoreactive T cells (22, 32, 33), it is not altogether unexpected that even this insulin paradigm does not fully explain T1D development. Homozygous INS susceptibility alleles are present in ~55% of Caucasians (34), yet very few develop T1D. We propose that incomplete clonal deletion of autoreactive T cells involves several other β cell Ags besides insulin and is not restricted to T1D patients.

So what distinguishes benign from pathogenic autoimmune T cells? The evidence suggests that neither their blood frequency nor their phenotype is at play. However, naïve T cells circulate perpetually, in contrast to memory T cells. The body load of Ag-experienced autoreactive T cells may therefore be underestimated in the blood of diseased individuals due to tissue sequestration (35). In line with this, the number of ZnT8186–194-reactive cells was higher than that of MelanA26–35-reactive ones in the pancreas, similar between aAb+ and nondiabetic donors, and enriched in the pancreas of T1D patients. Moreover, ZnT8186–194-reactive CD8+ T cell clones displayed potent lytic activity, much higher than other islet specificities (1). Although these findings suggest a pathogenic role for ZnT8-reactive CD8+ T cells, their precise localization and Ag-experienced status within the pancreas and whether their numbers are inversely correlated with age, as observed in the blood, remain to be verified.

The second possibility is that autoimmune T cells may be differentially regulated in T1D and healthy donors (33, 36), either intrinsically (e.g., anergy or exhaustion) or extrinsically by regulatory T cells. In support of this notion, T1D-specific islet-reactive CD8+ T cells have been repeatedly detected using functional IFN-γ ELISpot readouts (3, 7, 10), which use unfractionated PBMCs, thus preserving regulatory networks. This observation also argues that peripheral blood can be informative under appropriate assay conditions. ZnT8186–194-reactive CD8+ T cells from T1D patients also expressed higher levels of aurora A kinase, which might hint at increased mitotic activity in the T1D setting. Mirroring this observation, an anergic phenotype has been reported for self-reactive CD8+ T cells in nonautoimmune patients (22, 36), which may be at least partially imprinted in the thymus (33, 37).

The third possibility is that the central diabetogenic ingredient may be enhanced β cell vulnerability caused by islet inflammation in the face of similar autoimmune T cell repertoires across individuals. In this scenario, tolerance to β cell Ags may depend primarily on T cell ignorance (32, 37). Three observations are noteworthy in this respect. First, the higher T1D risk at younger age contrasts with the lower risk of other autoimmune diseases, which may indicate greater β cell vulnerability due to childhood stressors, such as islet-tropic enteroviruses and the metabolic demands imposed by growth. Second, high densities of ZnT8186–194 MMr+ cells were detected in the pancreas of an organ donor with chronic pancreatitis, suggesting that islet-reactive CD8+ T cells can expand under inflammatory conditions that are not autoimmune ab initio. Third, PPI15–24-reactive CD8+ T cells were more frequent in T1D versus healthy donors, with similar results reported for PPI2–10 (22). These epitopes map to the PPI leader sequence, which may undergo enhanced processing and presentation under metabolic stress (1). Collectively, the present data pose new challenges toward development of circulating T cell biomarkers for T1D staging and suggest novel therapeutic strategies based on mimicking benign autoimmunity or complementing incomplete central tolerance (38).


Study design

The objective of this study was to identify the features of islet-reactive CD8+ T cells that associate with T1D. Hypotheses were formulated on a prospective basis guided by the data. On the basis of a detailed analysis of ZnT8186–194-reactive CD8+ T cell clones (listed in table S1), we initially hypothesized that peripheral autoreactivity occurs independently of disease status. This hypothesis was substantiated using HLA-A2 MMrs to quantify and characterize islet-reactive CD8+ T cells directly ex vivo (donors listed in table S3). Next, we hypothesized that this widespread autoimmune repertoire stems from a universal leakiness of central tolerance, which was verified by thymic gene expression studies and by comparing HLA-A2–restricted Ag-reactive CD8+ T cell population frequencies in HLA-A2+ versus HLAA2 donors (listed in table S4). Last, we hypothesized that the lack of distinguishing features in the periphery reflects sequestration of the disease-relevant subset in the pancreas. This hypothesis was confirmed by in situ MMr staining of pancreatic sections (donors listed in table S5). After power analysis, age- and sex-matched, unblinded case-control sets were selected from donors recruited at affiliated Diabetology Units. Samples were processed in batch, and no outliers were excluded. All in vitro experiments were performed on at least two separate occasions. For ex vivo MMr analyses, undersampled data points were excluded, as detailed in Figs. 5 and 6.

Study approval

All individuals provided written informed consent. Ethical approval was granted by the Comité de Protection des Personnes Ile de France 1-2 (AOR10049, K091101, and A01094-53) and by the Institutional Review Boards of the Cambridge Royal Free Hospital (08/H0720/25), the Benaroya Research Institute (7109-147), the University of Heidelberg (367/2002), and the University of Florida Health Center (nPOD Project). The ImMaDiab study is registered at (NCT01747967).

Peptides, MMrs, and HLA-A2 binding measurements

Peptides ZnT8186–194 (VAANIVLTV) and its B. stercoris WP_060386636.1 mimotope (KAANIVLTV), ZnT8185–194 (AVAANIVLTV), MelanA26–35 (A27L variant; ELAGIGILTV), Flu MP58–66 (GILGFVFTL), PPI6–14 (RLLPLLALL), PPI15–24 (ALWGPDPAAA), GAD114–122 (VMNILLQYV), IA-2805–813 (VIVMLTPLV), IGRP265–273 (VLFGLGFAI), EboV NP202–210 (RLMRTNFLI), and HCV PP1406–1415 (KLVALGINAV) were synthesized at >85% purity (ChinaPeptides). Peptide–HLA-A2 affinity and stability assays were performed as detailed in fig. S2. HLA-A2 MMrs were produced as described (39), and staining was performed in the presence of 50 nM dasatinib (11). For Fig. 2B, MMr median fluorescence intensities (MFIs) were normalized to that of D222D clone 2 in the presence of dasatinib.

Cloning of ZnT8186–194-reactive CD8+ T cells

Frozen-thawed PBMCs (2 × 106 to 10 × 106) were stained with dual fluorochrome-labeled ZnT8186–194 MMrs, either directly ex vivo or after 10 days of acDC culture (8) in the absence or presence of 1 μM peptide (ZnT8186–194 or ZnT8185–194). Double-positive events were then sorted as single cells into individual wells of a U-bottom 96-well plate. Each sort well contained 200,000 irradiated PBMCs, 5% Cellkines (ZeptoMetrix), proleukin (200 IU/ml), IL-15 (25 ng/ml), phytohemagglutinin-L (PHA-L) (1 μg/ml; Sigma-Aldrich), penicillin/streptomycin, and amphotericin B. Medium was replenished every 3 days without PHA-L. Expanding clones were selected by visual inspection, transferred into 48-well plates for specificity testing, and restimulated as above every 2 to 3 weeks.

Ag recall assays

Peptide-pulsed HLA-A2+ LCLs or K562 cells transduced with HLA-A2, CD80, and 4-1BBL (a gift from J. Riley, University of Pennsylvania, Philadelphia, PA) were labeled with CellTrace Violet (Life Technologies) and incubated with T cells at an E/T ratio of 2:1 for 6 hours in the presence of brefeldin A (10 μg/ml). Intracellular cytokine staining was performed using BD Cytofix/Cytoperm reagents and analyzed using a BD LSR Fortessa cytometer. CD107a staining was performed with a fluorescein isothiocyanate–labeled mAb (clone H4A3, BD). Polyfunctionality indices were calculated as described (12).

Cytotoxicity assays

LCL, K562-A2, or K562-A2/ZnT8 target cells were labeled with CellTrace FarRed (Life Technologies), dispensed into 96-well flat-bottom plates at 105 cells per well, and cocultured with different numbers of carboxyfluorescein diacetate succinimidyl ester (CFSE)–labeled T cells for 6 to 24 hours. After staining with Live/Dead Aqua (Life Technologies) and fixation, a set number of CompBeads (BD) was added to each well. Flow cytometric analysis was performed by counting the numbers of CFSEFarRed+Live/Dead targets for each condition, normalized to equal numbers of CompBeads. Percent lysis was calculated as 100 × (live targets cultured alone) − (live targets in the presence of T cells)/(live targets cultured alone). Blocking experiments were conducted with concanamycin A (100 nM; Sigma-Aldrich), brefeldin A (5 μg/ml; Sigma-Aldrich), and the anti-FasL antibody NOK-1 (5 μg/ml; BD).

The EndoC-βH2 cell line (HLA-A*01/03, HLA-B*07/08, and HLA-C*07/07) was described previously (40), and the ECN90 cell line (HLA-A*02:01/03, HLA-B*40/49, and HLA-C*03/07) was derived from a human neonatal pancreas using similar protocols. Real-time cytotoxicity assays on β cell lines were performed using the xCELLigence system (ACEA Biosciences). Briefly, β cells were dispensed into 96-well E-plates and pretreated as indicated. After resting for 20 hours and pulsing with 10 μM peptide or dimethyl sulfoxide for 2 hours, T cells were added at an E/T ratio of 2:1, and impedance was recorded every 5 to 15 min for 4 hours. Cell indices were normalized to values at the time of T cell addition (t = 0) and transformed to percent lysis.

TCR sequencing, in silico analyses, and clonotype-specific TaqMan assays

TRA and TRB gene expression was analyzed using a template-switch anchored reverse transcription PCR (RT-PCR) (41) for T cell clones and a multiplex nested PCR (42) for single-sorted cells. Gene usage was determined according to the ImMunoGeneTics nomenclature.

The TRB database (Adaptive Biotechnologies) used for in silico analyses was derived from the donors listed in table S2. TaqMan assays (Life Technologies; fig. S6) were applied to cDNA samples from naïve (CD45RA+CCR7+) and Ag-experienced (CD45RA+CCR7 or CD45RACCR7+/−) CD4+ and CD8+ T cells bulk-sorted from age- and sex-matched cohorts incorporating 83 T1D patients [age, 34 years (17 to 59); 51% females; T1D duration, 8 years (0.1 to 55); 51% HLA-A2+] and 93 healthy donors [age, 34 years (17 to 60); 47% females; 41% HLA-A2+]. cDNA samples were amplified using clonotype-specific TaqMan primers for 18 cycles, followed by real-time qPCR using clonotype-specific TaqMan assays on Fluidigm 96.96 microfluidic chips with a BioMark HD qPCR system. Amplification curves for individual assays were examined and compared with curves from a TRB constant region assay as a control for TRB templates in each reaction.

Ex vivo analysis of ZnT8186–194-reactive CD8+ T cells

Cryopreserved PBMCs from T1D and healthy donors (table S3) were magnetically depleted of CD8 cells (STEMCELL Technologies), stained with the combinatorial MMr panels (9) detailed in figs. S7 and S10, and acquired using a BD FACSAria III cytometer. IFN-γ ELISpot assays were performed as described (7). Single-cell gene expression analysis is detailed in fig. S9A and table S6.

Gene expression in human mTECs

Human thymus samples were obtained from children undergoing corrective cardiac surgery at the University of Heidelberg, Germany. mTECs were purified as described (13). Sorted total, immature, and mature mTECs (CD45EpCAM+CDR2) were independently validated for gene expression of the tissue-restricted Ags β-casein and MelanA (13). Amplified bands were sequenced to confirm identity with the expected SLC30A8 and INS regions. SLC30A8 exons were annotated with reference to Ensembl ID ENST00000427715. The INS PCR covered all INS transcripts except ENST00000512523 (product 1) and ENST00000421783 (product 2).

In situ ZnT8186–194 MMr staining

In situ staining was performed as described (2). Briefly, unfixed, frozen sections were dried for 2 hours, loaded with 1 μg of MMrs overnight at 4°C, washed gently with phosphate-buffered saline, and fixed in 2% paraformaldehyde for 10 min. After a further wash, endogenous peroxidase activity was blocked with 0.3% H2O2. Sections were then incubated serially with rabbit anti-phycoerythrin, horseradish peroxidase–conjugated swine anti-rabbit, and 3,3′-diaminobenzidine tetrahydrochloride substrate (Thermo Fisher Scientific). After a final wash, sections were counterstained with hematoxylin, dehydrated via sequential passages in 95 to 100% ethanol and xylene, mounted, and analyzed using a Nikon Eclipse Ni microscope with NIS-Elements Analysis D software v4.40.

Statistical analysis

Data are shown as median (range) or mean ± SEM. Significance was assessed using two-tailed tests with a cutoff value of α = 0.05, as detailed for each figure.


Fig. S1. Cytokine secretion and cytotoxicity of ZnT8186–194-reactive CD8+ T cells from T1D patient D222D.

Fig. S2. CD8+ T cell recognition and HLA-A2 binding of ZnT8186–194 and ZnT8185–194 epitope variants.

Fig. S3. Ag sensitivity correlates with Ag avidity and polyfunctionality in ZnT8186–194-reactive CD8+ T cell clones.

Fig. S4. Modulation of HLA class I and ZnT8 expression in human β cell lines.

Fig. S5. TCR sequences of ZnT8186–194-reactive CD8+ T cell clones.

Fig. S6. ZnT8186–194-reactive clonotype-specific TaqMan assays.

Fig. S7. Gating strategy for the analysis of ZnT8186–194, MelanA26–35, and Flu MP58–66 MMr+CD8+ T cells.

Fig. S8. IFN-γ secretion by ZnT8186–194-reactive CD8+ T cells.

Fig. S9. Gene expression in ex vivo single-sorted ZnT8186–194 MMr+CD8+ T cells.

Fig. S10. Extended combinatorial MMr panel for the analysis of multiple islet-reactive CD8+ T cell populations, and reproducibility of ex vivo MMr assays.

Fig. S11. CD27, CD28, and CD95 expression on ZnT8186–194-reactive CD8+ T cells.

Fig. S12. Representative MMr and CD45RA/CCR7 dot plots for HLA-A2+ and HLA-A2 healthy donors depicted in Fig. 6 (F and G).

Fig. S13. Correlation between the frequency of MMr+CD8+ T cells and the Ag-experienced fraction within the same MMr+CD8+ population.

Table S1. Summary of ZnT8186–194-reactive CD8+ T cell clones.

Table S2. Characteristics of study patients for in silico TRB analyses.

Table S3. Characteristics of HLA-A2+ study patients for ex vivo MMr studies.

Table S4. Characteristics of HLA-A2+ and HLA-A2 healthy donors for ex vivo MMr studies.

Table S5. Characteristics of nPOD cases for in situ ZnT8186–194 MMr staining.

Table S6. Primers used for gene expression analysis of the individual ZnT8186–194 MMr+CD8+ T cells depicted in fig. S9A.

Members of the ImMaDiab Study Group

Data file S1. Raw data from figure graphs (Excel).

Reference (43)


Acknowledgments: We thank C. Maillard, M. Scotto, and S. Rozlan for technical assistance; Univercell Biosolutions for providing the ECN90 β cell line; K. Kedzierska and her laboratory (University of Melbourne) for help with single-cell TCR sequencing; T. Brusko (University of Florida, Gainesville) for help with the nPOD TCR database; the CyBio platform of the Cochin Institute and DKFZ Flow Cytometry Core Facility for assistance with flow cytometry; T. Loukanov (University of Heidelberg) for providing human thymic tissue; K. Boniface and J. Seneschal (INSERM U1035, Bordeaux) for providing vitiligo skin sections; and S. You for reviewing the manuscript. Funding: This work was performed within the Département Hospitalo-Universitaire AutHorS, supported by the Programme Hospitalier de Recherche Clinique ImMaDiab, Lilly France, the INSERM-Transfert Proof-of-Concept program acDC, the Ile-de-France CORDDIM, and grants from the JDRF (1-PNF-2014-155-A-V and 2-SRA-2016-164-Q-R), the Aviesan/Astra Zeneca “Diabetes and the Vessel Wall Injury” program, the Société Francophone du Diabète, the Agence Nationale de la Recherche (ANR-2015-CE17-0018-01), the Fondation pour la Recherche Médicale (Equipe FRM DEQ20140329520), and the Helmsley Charitable Trust George S. Eisenbarth nPOD Award for Team Science (2015PG-T1D052) (to R.M.); a JDRF Biomarkers grant (17-2012-598) (to K.C.); an NIH R01 grant (DK052068) (to H.W.D.); Lilly and Fondation Bettencourt-Schueller funds (to R.S.); a European Research Council grant (ERC-2012-AdG) (to B.K.); and funds from JDRF, the Wellcome Trust, and the National Institute for Health Research Cambridge Biomedical Research Centre [to the JDRF/Wellcome Trust Diabetes and Inflammation Laboratory (Cambridge Institute for Medical Research, University of Cambridge), which provided PBMC samples from the JDRF D-GAP study]. Samples from at-risk patients were obtained through a TrialNet ancillary study to the TN-01 Pathway to Prevention study funded by NIH grants U01 DK061010, U01 DK061034, U01 DK061042, U01 DK061058, U01 DK085465, U01 DK085453, U01 DK085461, U01 DK085463, U01 DK085466, U01 DK085499, U01 DK085504, U01 DK085505, U01 DK085509, U01 DK103180, U01-DK103153, U01-DK085476, and U01-DK103266 and by JDRF. D.A.P. is a Wellcome Trust Senior Investigator (100326Z/12/Z). This research was performed with the support of the nPOD, a collaborative T1D research project funded by JDRF. Organ procurement organizations partnering with nPOD to provide research resources are listed at This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking (INNODIA, no. 115797). This Joint Undertaking receives support from the Union’s Horizon 2020 research and innovation program and the European Federation of Pharmaceutical Industries and Associations, JDRF, and the Leona M. and Harry B. Helmsley Charitable Trust. Author contributions: S.C., A.I.L., G.A., K.C., S.P., G.S., K.K., L.N., A.E., T.Ø., J.E.M., and M.K. performed experiments and analyzed data. A.M., K.L., E.L., J.-P.B., A.L., V.A., H.W.D., S.B., D.A.P., E.B., M.B., S.C.-Z., F.D., and R.S. provided critical reagents, experimental assistance, and intellectual input. M.K. and R.M. performed statistical analyses. S.C., A.I.L., G.A., K.C., S.P., D.A.P., B.K., and R.M. designed and interpreted experiments and wrote the manuscript. The ImMaDiab Study Group recruited patients and performed clinical phenotyping. Competing interests: The authors declare that they have no competing interests. Some TCR sequences described herein are covered by Inserm-Transfert patent WO/2017/046335. Data and materials availability: Available materials will be provided with the appropriate material transfer agreement.
View Abstract

Navigate This Article