Many studies have investigated the role of islet antigen reactive (IAR) CD4+ and CD8 + T cells in peripheral blood of individuals with type 1 diabetes (T1D). IAR T cells have been investigated for their role in disease mechanisms and as therapeutic targets and biomarkers for β cell destruction1,2,3,4,5,6,7. Levels of IAR T cells may be increased in the pancreas especially during the active phases of islet autoimmunity, which may last months to years before and after clinical diagnosis2,3. Since pancreatic biopsy is not tenable in living humans, most efforts have focused on peripheral blood. IAR CD4+ and CD8 + T cells are found in blood of at-risk and T1D donors, but also often in healthy controls (HC) donor8,9,10. Although distinctive phenotypic properties of IAR T cells in T1D donors suggest their association with disease, because IAR T cells are rare in the blood9, it remains difficult to ascribe a role for them in pathogenesis in the pancreas.

A defining feature of T cells is the expression of T cell receptors (TCRs) on their surface. T cells proliferate in response to TCR recognition of antigenic peptides, resulting in clonal expansion of a population of cells with identical TCR sequences at both the nucleic acid and protein sequence levels and the same antigen specificity11. We recently identified a population of expanded IAR T cells with restricted TRA junctions and germline-constrained antigen recognition properties12. In combination with our other studies13, these previous studies suggest the possibility that expanded IAR T cell TCR sequences in the blood of T1D patients represent the ontogeny of T cell autoimmune responses during disease. While this suggestion is attractive for both fundamental and translational investigations in T1D, evidence that IAR T cells with these characteristics are found in the pancreas is lacking. Studies have characterized islet infiltrating T cells in the pancreas from organ donors with T1D14,15,16, but their link to IAR T cells in the blood is unclear.

In this study, we examine IAR T cells in the pancreas by cross-sectional comparisons of TCR junction sequences between circulating IAR T cells and pancreatic infiltrating T cells (PIT) from different donors. We identify and characterize a sizable fraction of IAR TCRs from peripheral blood with TRA junctions that share perfect- or single-mismatch protein sequences with PIT TCRs.


Isolation of TCR sequences from IAR and PIT T cells

Our central hypothesis is that in vivo expansion of IAR T cells seen in peripheral blood13 reflects autoimmune responses in the pancreas during T1D. This hypothesis predicts the presence of IAR T cells in the pancreas during disease. To test this prediction, we utilized the extreme sequence diversity of TCRs to enable their use as “barcodes” for clonal populations of T cells recognizing specific antigens17. Significant overlap between IAR and PIT cell TCR sequences would therefore be a prerequisite for a biological role for IAR TCRs in both blood and the pancreas.

We used TCRs from IAR T cells isolated from the peripheral blood of two cohorts (Table 1, Supplementary Data 1). Cohort 1 was from our previous scRNA-seq comparisons of IAR T cells from HC donors (n = 12), new-onset T1D donors (newT1D) < 100 days from diagnosis (n = 24), and established T1D donors (T1D, n = 12)12,13. Cohort 2 comprised additional HC (with no islet-directed autoantibodies) (n = 6) and newT1D donors (n = 11); donors with single autoantibodies (moderate risk for developing T1D, n = 8); and donors with multiple autoantibodies (high risk, n = 6). Patient characteristics are summarized in Table 1 and Supplementary Data 1.

Table 1 Summary of islet T cell donors

We isolated IAR CD4 + T cells from these donors using a CD154 activation induced marker (AIM) assay, single-cell sorted them, and subjected them to paired TCR chain determination using scRNA-seq12,13. Paired in-frame αβ (TRA/TRB) IAR TCRs identified in single cells were subjected to several additional filtering steps, including removal of mucosa-associated invariant T cell (MAIT) and invariant natural killer T cell (iNKT) sequences before use in subsequent analyzes. TCRs analyzed are presented in Supplementary Data 2 and summarized in Table 2.

Table 2 Characteristics of the TCRs used in this study

For PIT cells, we utilized paired TCRs identified by reverse transcription PCR of islets or pancreatic tissues from organ donors provided by the Network for Pancreatic Organ Donors with Diabetes (nPOD)18; a protocol at Vanderbilt University Medical Center/University of Pittsburgh; the Integrated Islet Distribution program (IIDP); and the Alberta Diabetes Institute Islet Core (ADI). Pancreatic donor characteristics are summarized in Table 1 and Supplementary Data 1. Paired in-frame αβ (TRA/TRB) PIT TCRs are listed in Supplementary Data 2 and summarized in Table 2. PIT cell TCR sequences contained multiple perfectly matched amino acid junction sequences from preproinsulin reactive TCRs found in islets and peripheral blood19.

TCRs from IAR T cells in blood share TRA chain sequence identity with PIT cell TCRs

For our initial experiments, we focused on Cohort 1. Approximately 90% of these donors had high-risk DRB1*0401 HLA class II alleles, while ~10% had DRB1*0301 alleles12. Using molecular cloning, lentiviral re-expression, and TCR functional assays, we previously identified specific islet antigen epitopes that triggered 29/47 ( ~ 62%) of TCRs from expanded clones tested from this cohort12,13. Reason(s) why the remaining 18 TCRs tested did not demonstrate peptide specificity remain unknown but may involve suboptimal avidity, and/or presentation by MHC class II molecules not tested. These TCRs tested for specificity, therefore, represented IAR T cells from peripheral blood with a wide range of specificities.

To test for TCR overlap, we tested for perfectly matched recombined V-J or V-D-J junction sequence overlap between IAR TCRs from Cohort 1 with PIT cell TCRs. Although PIT TCRs represented a heterogeneous group of donors that varied among multiple parameters that may affect TCR repertoires, including cell type, patient group, and HLA class II allele (Supplementary Data 1, Supplementary Fig. 1), we elected to initially test all TCRs together. Unless otherwise noted, sequence comparisons were made at the amino acid level, as we reasoned that perfectly matched junctions were most likely to reflect conserved function. Multiple IAR junction sequences, mainly TRA junctions compared to TRB junctions (n = 55 and 7, respectively), perfectly matched pooled PIT junctions (Fig. 1A, Supplementary Fig. 1A). As expected, with PIT junctions, the same number of TRA than TRB junctions likewise perfectly matched IAR junctions (Fig. 1B). IAR junctions, which were isolated from CD4 + T cells, perfectly matched similar numbers of PIT junctions from both CD4+ and CD8 + T cells (Fig. 1C, Supplementary Fig. 1B). TCR chain sharing between CD4+ and CD8 + T cells has been noted previously, more frequently between individual TRA than TRB chains20. Overall, Cohort 1 had ~2.5% (55/2,174) of total unique IAR T cell TRA junctions that perfectly matched PIT junctions. Conversely, ~0.56% (55/9,757) of PIT cell TCRs matched IAR T cell TCRs. The distribution of IAR perfect matches between PIT TRA and TRB junctions (Fig. 1A) differed significantly from the distribution of TRA and TRB junctions in total PIT TCRs (Supplementary Fig. 1) (p-value = 1.1e-11, Fisher’s exact test). In contrast, distributions of IAR perfect matches by donor group (HC, T1D, and newT1D), cell type (CD4 + , CD8 + ), or HLA class II alleles did not differ significantly from the distributions in total PIT TCRs (Supplementary Fig. 1C and D) (all p-values > 0.05, Fisher’s exact test). These findings supported our decision to maximize power by utilizing combined PIT TCRs for most analyzes.

Fig. 1: IAR T-PIT TCR sequence matching.
figure 1

Panels AG utilized TCRs from Cohort1; panels H-K, TCRs from combined Cohorts 1 + 2. A) Perfectly matched IAR TRA and TRB junctions from Cohort 1 (Table 2) in PIT TCRs (n = 9,757 unique TRA and TRB junctions, Table 2). B Perfectly matched PIT TRA and TRB junctions in IAR TCRs (n = 4,310 unique TRA and TRB junctions). C Perfectly matched PIT T cell TCR junctions from CD4+ and CD8+ cells in IAR TCRs. D Perfectly PIT-matched versus non-matched TRA junctions from Cohort 1 IAR T cell TCRs (n = 55 matched and 2119 non-matched) versus TRA junctions from HC20 (185 matched and 24,968 non-matched) and COVID-19 patients20 (1341 matched and 197,412 non-matched). E As in D but using TRB junctions (n = 7 matched and 2129 unmatched). F PIT junction matching in IAR TRA junctions versus junctions from HC donors20. Dot sizes, Levenshtein index values. Diagonal line, equivalency line. Blue line, best fit line from linear modeling. The pAdj for a slope different from 1 was calculated using linear modeling and ANOVA. Gray shading, 95% confidence intervals. G As in F but using TRB junctions. H PIT matching by different TRA chain segments in combined Cohorts 1 and 2 (n = 3264 unique TRA junctions). Lv0 and Lv1, Levenshtein index values of 0 and 1 (n = 74 and 2025 unique TRA junctions, respectively). I PIT matching by TRA junction segments in combined Cohorts 1 and 2 according to HLA-DRB1 alleles in PIT donors. There were: n = 484, 153, 199, and 95 junctions from donors having 03:01, 04:01, 07:01, and other HLA-DRB1 alleles, respectively. J Junction sharing by TCRs with PIT-matched TRA junctions. One TCR in each group recognized13 the indicated autoantigen epitope. K Total TRA chains (Table 2) in expanded (n = 1000) and non-expanded (n = 2907) cells from combined Cohorts 1 and 2 were tested for differences in PIT-matched and non-PIT-matched junctions from HC, Aab + , newT1D, and T1D donors using two-sided Fisher’s exact tests.

PIT-matched junctions are enriched in IAR CD4 + T cells relative to unselected TCR repertoires

We hypothesized that junction sequence matching between IAR and PIT TCRs was greater than would occur by chance in unselected repertoires. An unselected repertoire determined by the same technology from healthy individuals matched by donor characteristics and HLA type to IAR T cell donors was not available. We selected instead two PBMC repertoires from Su et al. (16 uninfected (HC) and 129 donors with COVID-19, comprising 2513 and 198,753 unique junctions for HC and COVID-19 patients, respectively)21. It should be noted that these TCRs were determined by different technologies (a 5’RACE method21 versus the RT-PCR method optimized for rare cells that was used in our data set12,13), and were not matched by donor characteristics and HLA type to IAR T cell donors.

We used Fisher’s exact test to compare numbers of perfectly PIT-matched to non-matched TRA junctions from IAR T cells versus unselected HC and COVID-19 T cells. We found that a higher fraction of PIT junction matches with IAR TRA junctions than with junctions from unselected repertoires: there were 55 perfect matches with IAR TRA junctions (2.5%) versus 154 (1.4%) with HC junctions (n = 11,274) and 1342 (0.7%) for COVID patient junctions (n = 198,753). These comparisons yielded log2 odds ratios >1 (or >2-fold in linear units), outside the 95% confidence intervals (Fig. 1D) (p-values 1.9e-4 and 8.5e-16, respectively, by Fisher’s exact test). In contrast, the log2 odds ratio was <0 (or 1 in linear units), for PIT-matched to non-matched junctions from COVID-19 patients versus HCs donor(Fig. 1D) (p-value > 0.05). Thus, there was a higher fraction of PIT TCR perfect matches with IAR TRA junctions than unselected human TCR repertoires from HC and COVID-19 patients. In contrast, PIT TCRs did not show greater numbers of perfect matches with IAR TRB junctions (Fig. 1E).

TCRs from IAR T cells with mismatched junctions may also share functional properties. While we expect that shared function would be more likely with mismatches that conserve amino acid electrochemical properties, we have ignored these properties to simplify subsequent analyzes. We tested for enrichment of IAR T cell TRA chains in PIT TCRs over a range of mismatch values by calculating pairwise Levenshtein index values. A plot of the numbers of PIT TRA junction matches with IAR versus HC T cells at different PIT mismatch thresholds (Fig. 1F) showed significant off-diagonal skewing in the direction of IAR TRA junctions (slope = 0.872, p-value (that the slope was not equal to 1) = 3e-3, by linear modeling). Thus, enrichment of PIT-matched to non-matched TRA junctions versus unselected HC TCRs was not limited to perfect sequence matches. Ninety-five % confidence intervals for PIT junction sequence overlaps with 0 and 1 mismatches showed greatest divergence between IAR T cell and HC TCRs. Taken together, we identified 582 unique TRA chains from a total of 2174 unique TRA junctions (~27%) IAR CD4 + T cells in Cohort 1 with perfect or single mismatches with PIT TCR junctions. We did not observe significant off-diagonal divergence of TRB junctions that were paired with PIT-matched TRA junctions, indicating more highly mismatched sequences (Fig. 1G). Although there were no perfect PIT-matches to IAR TCRs of known specificity, selected IAR TCRs of unknown specificity with perfect PIT matches are shown in Table 3. There were, however, also numerous IAR TCRs with known specificity having single mismatches to PIT TCRs (Table 4). With both perfect and single PIT mismatches, the paired TRB junctions were markedly more divergent than the matched TRA junctions.

Table 3 Examples of PIT TCRs sharing perfectly matched TRA chains with IAR T cells. A
Table 4 PIT TCRs sharing a single mismatched TRA chain with IAR T cells of known specificity. A

PIT junction matching in a second cohort of IAR T cells

Since the results in Fig. 1A–I were obtained from only a single donor cohort (Cohort 1), we did not know how broadly applicable our findings were. To examine this question, we repeated the analyzes with an independent cohort of IAR T cells (Cohort 2) (Table 1 and Supplementary Data 1). These analyzes yielded similar results (Supplementary Fig. 2). There were again more TRA junctions compared to TRB junctions (n = 30 and 4 TRA and TRB, respectively) showing perfect matches with total PIT junctions (Fig. S2A, B), spread between CD4+ and CD8+ cells (Fig. S2C). There also were more TRA than TRB junctions showing 0 or 1 mismatches with PIT junctions (Supplementary Fig. 2D–G). In total, there were 360 unique TRA chains from Cohort 2 with perfect matching or single mismatches with PIT TCR junctions in a total of 1,090 unique TRA junctions (~33%). These findings demonstrated that our observations with Cohort 1 were not restricted to a single data set and therefore had a potentially broader range of islet specificities. To increase power in subsequent analyzes, we combined TCRs from Cohorts 1 and 2, yielding 3264 and 3187 unique TRA and TRB junctions (Table 2). In both Cohorts 1 and 2, there were more PIT-matches with IAR TRA junctions than TRB junctions (Fig. 1, Supplementary Fig. 1, and Supplementary Fig. 2). Since the combined data were more optimally powered to analyze TRA than TRB junctions, we chose to focus on them in later experiments. In combination, there were 942 unique TRA junctions with perfect- or single mismatches to PIT junctions, from a total of 3264 unique IAR TRA junctions (~29%). We hereafter refer to these junctions as “PIT-matched”.

PIT TRA junction matches extend to include the J gene but not the V gene

To test the extent of sequence identity of PIT-matched IAR TRA chains, we extended the requirement for sequence matches into the V and J gene segments flanking matched junctions (Fig. 1H). Since the V and J gene segments in TCRs are much longer than junctions22, they are less amenable to string match comparisons, prompting us to consider V and J genes as identical if they simply had the same name. For all junctions with 0 or 1 PIT mismatches, sequence identity extended towards the C terminus to include use of identical J genes (Fig. 1H). In contrast, sequence identity towards the N terminus was less marked, with only ~20-25% of matched junctions having identical V genes.

Since the TCR V gene contains the CDR1 and CDR2 regions that contact the MHC class I or class II molecules23, it was possible that the low frequency of V gene identical matches reflected peptide presentation to PIT TCRs by different HLA molecules. IAR T cell TCRs were associated primarily with high-risk HLA genotypes12, whereas PIT TCRs were from individuals having a wider variety of HLA genotypes. We, therefore, broke down the PIT-matching data from Fig. 1H by HLA class II DRB1 genotype. Individuals having 03:01, 04:01, 07:01 and pooled other DRB1 genotypes all showed similar frequencies of V gene identical matches (Fig. 1I). Thus, IAR TCRs (from ~90% HLA DRB1-04 donors)12 showed neither preferential matches with PIT junctions from donors with different class II alleles (Supplementary Fig. 1D), nor did they show class II allele-dependent differences in the extent of matches in to V gene segments (Fig. 1I). Together, these findings suggest that peptide presentation by different PIT HLA molecules did not have a major effect on matching with IAR TRA junctions. Other studies have noted HLA-independent associations with TRB chains24.

PIT-matched TRA junctions are paired with diverse TRB chains

Although we observed a few TRB perfect matches (Fig. 1A), these were not paired with perfectly matched TRA chains. Likewise, we observed that perfect or single mismatch PIT-matched TRA chains were paired with quite different TRB chain sequences. This was true with TCRs of both unknown (Table 3) and known12 (Table 4, and Fig. 1J) specificity. The divergence of TRB sequences with perfect or single mismatch PIT-matched TRA junctions makes it uncertain that these TCRs share epitope specificity. We conclude that PIT-matched TRA chains were paired with diverse TRB chains.

PIT TRA matches with IAR T cell junctions were expanded early in disease

We speculated that PIT TRA matches were related to T cell expansion and T1D progression. As a test, we quantified levels of PIT TRA matches in IAR TCR junctions along the continuum of stages during the T1D disease process25, using combined Cohorts 1 and 2 (Fig. 1K). Since we were able to sample relatively few 1AAb and 2AAb individuals in Cohort 2, we combined these donors into a single category (termed AAb + ). We then compared levels of PIT-matched TRA junctions from expanded (>1 cell) and non-expanded (1 cell only) T cell clones. In cells with expanded TCRs, we saw an increase in the fraction of PIT-matched junctions from Donors in both AAb+ and newT1D donors, then a decrease in T1D donors to a similar level as in HCs donor(Fig. 1K). These comparisons were underpowered with respect to both donor and cell numbers and group differences did not reach significance after multiple correction. In contrast, there were strongly significant decreases in PIT-matched TRA junctions in the more numerous non- expanded TCRs that were the opposite of the expanded TRA junctions, i.e. higher numbers of non-expanded PIT-matched TRA junctions in HC and T1D compared to AAb+ and newT1D (Fig. 1K).

To control for p-value inflation by increased numbers of junctions from relatively few donors, we repeated this analysis at the donor level and observed the same trends as with junction frequency (Supplementary Fig. 3). The effects at the donor level were weaker, suggesting that these studies were underpowered. Together, these analyzes suggest that an elevation of expanded PIT-matched TRA chains in blood occurred prior to the onset of clinical disease.

Publicity, generational probability, and TCR convergence of PIT-matched TCRs

Since different cohorts were used to isolate IAR and PIT TCRs, PIT-matched TRA junctions were shared between different donors (public). We hypothesized that PIT-matched TCRs were related to the public TCRs with diverse specificities and germline-constrained antigen recognition properties that we previously described in IAR T cells12. To test this hypothesis, we asked whether PIT-matched TRA and associated TRB junctions were enriched with public and private junctions. Public IAR and PIT-matched TRA junctions showed significant overlap, whereas private IAR TRA junctions and PIT-matched TRA junctions did not (Supplementary Fig. 4A, B). Supplementary Fig. 4 Public IAR TRB junctions showed weak overlap with PIT-matched TRB junctions, but not private IAR TRB junctions did not (Supplementary Fig. 4C, D). Thus, there was a strong overlap between TRA junctions from public TCRs and PIT-matching TRA junctions.

Public TCR sharing may result from the expansion of clones with a high probability of recombination26. The probability of generating individual recombination events may be estimated by the generation probability (Pgen) of junction sequences during TCR recombination26. To estimate the generational probabilities of PIT-matched TCRs, we calculated their Pgen values (Supplementary Fig. 5). Pgen values were higher overall (i.e., a higher propensity for generation by chance) for TRA than TRB junctions, presumably reflecting the greater numbers of random non-templated nucleotides in TRB junctions27. -log10(Pgen) values for both PIT-matched TRA (Fig. S5A), and their associated TRB junctions (Fig. S5B), were significantly higher than for non-PIT-matched junctions, especially for TRA junctions (median -log10(Pgen) values of 7.9 and 9.2, respectively). The differences between PIT-matched and non-PIT-matched TCRs were smaller than the overall spread of Pgen scores (>10 orders of magnitude), suggesting heterogeneity in generational probabilities of different PIT-matching sequences.

Public TCR AA sequences TCRs may arise from use of identically rearranged nucleotide (nt) sequences across donors or from post-recombination antigen selection of clones with identical protein sequences encoded by different nt sequences (TCR convergence)28,29,30. To distinguish these possibilities, we examined TCR convergence in PIT-matched junctions (Supplementary Fig. 6). A total of 93/6451 IAR junctions (69 TRA and 24 TRB, 1.1%) were convergent (Supplementary Fig. 6A). This number is likely an underestimate because of the relatively few cells sampled and the resulting shallowness of the repertoire estimation. Significantly more TRA than TRB junctions were convergent (p-value = 7.4e-7, Fisher’s exact test) (Supplementary Fig. 6A). Nearly all public (Supplementary Fig. 6B) but only a few private TRA junctions (Supplementary Fig. 6C) were convergent, despite private TRA junctions being found in more cells (more expanded) than public junctions (Supplementary Fig. 6D). This argues against sequencing errors as an alternative explanation for different nt sequences. Finally, convergent TRA junctions were more enriched with PIT-matched sequences than convergent TRB junctions (Supplementary Fig. 6E and Supplementary Fig. 6F). Thus IAR TCRs, particularly public TRA junctions, showed evidence of TCR convergence. Taken together, results from this section demonstrate that PIT-matched TCRs were enriched with public sequences, have high generational probability, and show TCR convergence.

PIT-matched TRA junctions were more germline-like, and more hydrophobic than non-PIT-matched junctions

To test for molecular differences between PIT-matched and non-matched junctions, we compared sequence features that have been correlated with TCR autoreactivity12,31,32(Fig. 2). We utilized density plots for these comparisons to emphasize the range of TCR junction lengths. We found that the distribution of TRA junctions was significantly left shifted (shorter) in PIT-matched junctions than in non-matched junctions (p-value < 1e-4, Kolmogorov–Smirnov test) (Fig. 2A). PIT-matched TRA junctions had a median length of ~39 nt, versus a median length of ~42 nt for PIT-non-matched junctions (i.e., 3 nt or 1 AA residue difference) (Fig. 2A). We did not see a significant difference in length with paired TRB junctions; TRB junctions paired with both PIT-matched and non-matched TRA junctions had identical median length values of ~42 nt (Fig. 2B). Furthermore, TRA junctions (Fig. 2C), but not paired TRB junctions (Fig. 2D), were more hydrophobic in PIT-matched versus non-matched TRA junctions (p-value < 1e-4). PIT-matched and non- matched TRA junctions had median hydrophobicity values of 0.23 and 0.18, respectively, on the Eisenberg hydrophobicity scale33, falling between values for proline and tyrosine (0.12 and 0.26, respectively). Thus, PIT-matched TRA junctions, but not their paired TRB junctions, were shorter and more hydrophobic than PIT-non-matched TRA chains.

Fig. 2: PIT-matched TRA junctions are shorter and more hydrophobic than non- PIT-matched junctions.
figure 2

Distributions of sequence lengths (in nt) for features of unique PIT-matched (n = 942 and 832) and non-matched (n = 2322 and 2355) TRA and TRB junctions, respectively, as delineated by IMGT/HighV-QUEST61 from combined Cohorts 1 + 2. A, B junction lengths (Junc); C, D hydrophobicity. The significance of differences between PIT-matched (teal) and PIT-non-matched (peach) junctions was assessed using two-sided Kolmogorov-Smirnov tests. Solid vertical line, median value from PIT-matched junctions; dashed vertical line, median value from PIT-non-matched junctions.

Length differences between PIT-matched and non-matched TRA junctions map to peptide-contact regions

We hypothesized that sequence features of PIT-matching TRA junctions were keys to the function(s) of IAR T cells having these TCRs. We first determined the positions of single AA mismatches between PIT and IAR T cell TRA junctions (n = 927). We found that mismatched residues were distributed with a mode at amino acid position 3 (Supplementary Fig. 7A). By comparison, the C terminal AA from V genes (3’ end) and the N terminal AA of J genes (5’ end), which mark the boundaries of germline-encoded residues in the respective gene segments, were centered at AA residues 3 and 5, respectively (Supplementary Fig. 7A). Considering individual IAR TRA junctions with single PIT mismatches, we observed that ~74% (687/927) AA mismatches were located at the C-terminus of the V gene or between the C-terminus of the V gene and the N-terminus of the J gene (Supplementary Fig. 7B). Thus, PIT mismatches occurred in a mixture of germline and non-germline-encoded residues in the V-J recombination region.

We next focused our comparisons on regions important for TCR binding and function. TCR complementary determining regions (CDRs) convey the specificity for antigen and major histocompatibility complex molecules34. A comparison of CDR1 region lengths showed that the median CDR1 length was identical in PIT-matched and non-matched junctions, but the distribution was left-shifted significantly for PIT-matched junctions (Supplementary Fig. 8A). CDR2 regions showed no little difference between PIT-matched and non-matched TRA chains (Supplementary Fig. 8B), whereas CDR3 regions, as expected from the junction sequences comparisons in Fig. 2, were shorter in in PIT-matched TRA chains (Supplementary Fig. 8C), by ~3 nt (1 AA).

We wished to determine the source(s) of junction length differences between PIT-matched and non-matched TRA junctions. We compared N region lengths between the two groups of TRA chains and found that PIT-matched junctions had median N region lengths of ~3 nt (1 AA) whereas PIT-non-matched junctions had median N region lengths of ~5 nt (1-2 AA) (Supplementary Fig. 8D). We also quantified contributions of the V gene and J gene segments adjacent to the N region. The 3’ end of the V region was slightly shorter in PIT-matched than non-matched TRA junctions (10 versus 11 nt) (Supplementary Fig. 8E), whereas the 5’ end of the J region was slightly longer (27 versus 25 nt) (Supplementary Fig. 8F). This is consistent with our observed sharing of J genes but not V genes. Thus, the difference in CDR3 lengths was a complex product of genome- and non-genome-encoded regions in PIT-matched junctions. Framework (FR) regions FR1, FR2, and FR3 showed more modest differences in length between PIT-matched and non-matched TRA chains (Supplementary Fig. 8G–I), demonstrating some selectivity of the differences in CDR1 and CDR3 lengths. Together, these analyzes showed that PIT-matched and non-matched TRA junctions differed regions important for peptide binding, including templated V gene region CDR1 and both templated and non-templated regions of CDR3..

We also wished to determine whether there were TRA V gene differences between PIT-matched and non-matched TRA junctions. We thus compared enrichment of V gene segments in PIT-matched versus non-matched TCRs. This comparison identified TRAV41*01V genes as significantly over-represented (pAdj ~1e-4) in PIT-matched IAR T cell TRA chains (Supplementary Fig. 9A). In contrast, we did not identify any V genes as significantly over- or under-represented (pAdj >0.05) in PIT-matched IAR T cell TRB chains (Supplementary Fig. 9B). TRAV41*01V genes were previously identified, together with MAIT and iNKT TCRs, in innate T cells having invariant TRA (iTRA) chains35. The over-represented V gene, TRAV41*01, had a CDR1 region length of 15 nt (5 AA), compared to the median of 18 nt (6 AA), thus contributing to the overall shorter CDR1 regions in PIT-matched TRA junctions.

Determining predicted peptide binding contacts for PIT-matching and non-matching TRA chains

CDR1 and CDR3 regions are involved in binding to peptides presented by MHC molecules23, leading us to hypothesize that the observed length variation of these regions suggested altered peptide binding properties for PIT-matched IAR TRA chains. As a test, we used TCRmodel236 to predict peptide binding residues in tri-molecular structures of PIT-matched and PIT-non-matched TCRs complexed with peptide-class II MHC molecules. For model input, it was desirable to use TCRs with cognate peptides. We therefore selected paired TRA and TRB sequences from a set of 30 IAR CD4+ TCRs, largely from expanded clones, with known specificity (16 PIT-matched and 14 PIT-non-matched TCRs), together with their cognate peptides12; and sequences of HLA DRA*01:01 and DRB1*0401 MHC class II subunits37.

From best fit models for each TCR, we identified likely peptide contact residues (TCR residues <5 Å in distance from the bound peptide chain) and mapped these to TRA and TRB sequence features. We first visually compared TCR-peptide contact residues in representations of PIT-matched and PIT-non-matched TCRs with extremes in TRA junction length (Fig. 3A, B). This comparison showed that IAR Clone 640 (10 AA, PIT-matched), specific for GAD65 113-132, had more predicted peptide contacts that mapped to CDR1, and fewer that mapped to CDR3, than IAR Clone 2353 (17 AA, PIT-non-matched), specific for IGRP 305–324. To extend these observations to the larger data set, we compared the numbers of TCR contacts of each chain from the set of 30 IAR TCRs (Supplementary Data 3). We displayed the numbers of contacts in different CDR sequences in both chains from each TCR as a function of TRA and TRB junction lengths (Supplementary Fig. 10). As expected, the TRA and TRB CDR3 regions contributed the most peptide contacts. There was a significant positive relationship between the number of peptide contacts mapping to the TRA CDR3 region and TRA junction length, and a weaker negative relationship with the TRB CDR3 region (Supplementary Fig. 10). While the number of contacts was low, there was also a negative relationship between the TRA CDR1 region and TRA junction length.

Fig. 3: Increased ratios of germline-encoded to recombined peptide contacts in PIT-matched TCRs.
figure 3

A Gaussian Surface representation of a model of Clone 640, a PIT-matching TCR with a 10 AA TRA junction (7 CDR contacts) paired with a 12 AA TRB junction (8 CDR contacts), that binds the GAD65 113-132 peptide presented by the HLA DRB1*0401 class II molecule12. The viewing plane is from the interface with the MHC molecules, which together with the peptide, have been removed from the representation for clarity. Labels indicate the CDR loops, with letters denoting the chain (α or β); and numbers, the CDR loop. Numbers in parentheses are the number of predicted peptide contacts with the CDR loop. B A model of Clone 2353, a PIT-non-matching TCR with a 17AA TRA junction (7 CDR contacts) paired with a 14 AA TRB junction (8 CDR contacts), that binds the IGRP305-324 peptide presented by the HLA DRB1*0401 class II molecule. C The number of total peptide contacts for either the TRA or TRB chains did not differ significantly between PIT-matched and PIT-non-matched TCRs (p-value > 0.05, two-sided Wilcoxon signed rank test). Each dot represents a value for an individual donor. The width of the violins represent frequency; horizontal lines within the violins represent median values. D Ratios of TRA CDR1 to CDR3 peptide contacts decrease with increasing TRA junction length. The P value is for a null hypothesis of a slope of 0, calculated using linear modeling. Gray shading, 95% confidence intervals.

Increased dependence on germline-encoded residues for binding of shorter and PIT-matching TCRs

The variation in contact residues by junction length (and PIT match) may indicate increased overall numbers of peptide contacts by longer TCRs, whereby increased numbers of CDR3 contacts would augment contacts CDR1 and CDR2. Alternatively, decreased numbers of CDR3 contacts in shorter TCRs could come with increased contacts in other regions. To distinguish these possibilities, we compared the overall number of contacts by TCR chain in IAR PIT-matched versus PIT-non-matched TCRs. This revealed no significant difference in overall peptide contacts in either TRA or TRB chains between the two groups (Fig. 3C). The ratio of TRA CDR1 to CDR3 contacts varied inversely and significantly with TRA junction length (p-value < 1e-4 for line slope, by linear modeling), while maintaining the separation in length between PIT-matched and PIT-non-matched junctions (Fig. 3D). This indicates that peptide recognition by shorter PIT-matched TCRs shows greater reliance on the germline-encoded residues in the TRA CDR1 region. Longer PIT-non-matched TCRs, in turn, show greater reliance on CDR3 residues generated by V(D)J recombination.

IAR CD4 + T cell TCRs with PIT-matching TRA junctions show evidence of multi-specificity

We hypothesized that different peptide binding modes of PIT-matched and non-matched TRA junctions would lead to altered peptide recognition properties, such as strength and/or specificity of binding. As a measure of binding strength, we compared the functional avidities of TCRs having PIT-matched and non-matched TRA junctions specific for different islet peptide epitopes12. We plotted cell proliferation12, versus peptide concentration (Fig. 4A) for the subset of PIT-matched or non-matched TRA junctions with known specificity. To increase power, we aggregated results with different peptides from each of three different islet antigens (GAD65, IGRP and ZNT8). These results showed wide variability in range of dose responses, but no consistent differences between TCRs with PIT-matched and non-matched TRA junctions. EC50 values12 for aggregated GAD65- and IGRP-specific TCRs with PIT-matched and non-matched TRA junctions did not differ significantly by unpaired Wilcox tests; since there were only two ZNT8 TCRs, a p-value could not be calculated. We obtained similar overall results when individual peptides were considered. These data suggest that there are not large differences in functional avidity between TCRs with PIT-matched and non-matched TRA junctions.

Fig. 4: Functional properties of selected PIT-matching and non-matching TRA chains of known specificity.
figure 4

A Dose response curves for TCRs recognizing GAD65 (n = 11), IGRP (n = 8) and ZNT8 (n = 4) epitopes12. Curves represent single values for each point. Curves for multiple epitopes per target antigen were pooled to increase power; examination of curves for individual epitopes yielded conclusions consistent with the pooled data. B Cross-reactivity of TCRs with PIT-matching TRA junctions (Table 5) for autoimmune (GAD65) and viral (influenza M protein) epitopes in a CFSE proliferation assay.

Based on these results, we next hypothesized that TCRs with PIT-matched and non-matched TRA junctions differed in peptide binding specificity. This was supported by our previous finding of multi-specificity of some public IAR TCRs12. Though limited in number (n = 3), the frequency of these multi-specific clones was higher in public compared with private clones12. In the present studies, we noted that TRA chains from all three of these multi-specific TCRs were PIT-matching. This suggested that multi-specific TRA junctions were modestly more frequent in PIT-matching than non-matching TRA junctions (one-sided p-value = 0.030, Fisher’s exact test).

Two of these PIT-matching multi-specific TCRs (Clones 81 and 566) recognized non-overlapping GAD65 epitopes12. In parallel and independent experiments, we unexpectedly found that the TRA chain of a TCR (P196-1) from Influenza A/MP54- reactive CD4 + T cells (Methods) perfectly matched the TRA chain from Clones 81 and 566 (Table 5). Furthermore, P196-1 had TRB chains that differed from Clones 81 and 566 at only a single AA position (Table 5). This sequence similarity led us to reason that closely related Clone 81, Clone 566 and P196-1 were all multi-specific. As a test, we compared the ability of activating peptides (MP54 97-116 and GAD65 377-396), a non-activating peptide (GAD65 113–132) to trigger proliferation of TCR-transduced primary CD4 + T cells (Fig. 4B). We found that both MP54 97-116 and GAD65 377–396 peptides identically activated Clone 81, Clone 566, and P196-1 TCRs. Thus, these TCRs with PIT-matched TRA junctions were all multi-specific, supporting our hypothesis that distinctive TCR sequence features, including shorter and more hydrophobic TRA junctions, are linked to an inherent tendency for multi-specificity.

Table 5 Mismatched TRB chains from multi-specific TCRs with perfectly matched TRA chains. A

To elucidate possible mechanisms of cross-reactivity of the Clone 81 TCR for the MP54 97-116 and GAD65 377-396 peptides, we constructed molecular models36 of the Clone 81 TCR12 and complexes of these peptides together with HLA-DRA1*0101/DRB1*0401 molecules37. The cognate GAD65 377-396 (HKWKLSGVERANSVTWNPHK, where bold font denotes core sequence complexed with the TCR) and MP74 97-116 (VKLYRKLKREITFHGAKEIS) peptides did not show compelling sequence similarity by multiple sequence alignment. They also showed no evidence of structure38 or aromatic side chain39 conservation characterizing “hotspots” of molecular mimicry used by some cross-reactive TCRs. Despite their low degree of peptide sequence similarity, molecular models of Clone 81 TCR with both peptides yielded identical scores of 0.88. In addition, TRA and TRB chains in both models showed similar predicted topography in their interactions with peptide-MHC class II complexes (Supplementary Fig. 11A, B). Overlapping but non-identical residues in the modeled TCR TRA and TRB chain CDR1, CDR2 and CDR3 regions contacted the different peptides (Supplementary Fig. 11C). Alignment of predicted structures of the TRA (Supplementary Fig. 11D) and TRB (Supplementary Fig. 11E) chains from models made with the different peptides showed nearly perfect superposition except in the CDR3 regions. This suggests that the PIT-matched Clone 81 TCR accommodates quite different peptide sequences through interactions involving conformationally conserved CDR1 (and CDR2 chains for TRB) regions, together with more variable CDR3 regions.

Multi-specific TCRs shared sequence features with PIT-matched TCRs

We wished to test in greater depth our hypothesis that PIT-matched and non-matched TRA junctions differed in peptide binding specificity. Recognizing that the number of known multi-specific TCRs from IAR T cells was too small at present to enable firm population-based conclusions about multi-specificity, we took an alternative approach to test our hypothesis. In other experiments, we observed that VDJdb, a curated public database of TCRs with known specificities40, contains TCR sequences that recognize both single and multiple peptide epitopes. These designations were likely biased, with some TCRs being designated as single specificity because they were understudied, and others being designated as multi-specific based on reactivity with closely related peptides. We reasoned, however that these errors would tend to offset each other, and that the size and scope of this database would provide a more comprehensive source of specific and multi-specific TCRs, which could be used to test their junction sequence features. There were n = 17,826 unique TRA chains in VDJdb that recognized single epitopes and n = 1664 TRA chains that recognized multiple epitopes (Fig. 5A). We found that TRA CDR3 regions (Fig. 5B) from TCRs that recognized multiple epitopes were significantly shorter (by ~3 nt in median length, or ~1 AA) than junctions from TCRs that recognized single epitopes. Likewise, TRA chains from multi-specific TCRs contained fewer non-templated (N region) nucleotides (~1 nt in median length) than TCRs with single specificity (Fig. 5C). Finally, TRA junction amino acid sequences from multi-specific TCRs were more hydrophobic in TCRs that recognized multiple epitopes (Fig. 5D). None of these differences were seen with TRB chains (Supplementary Fig. 12A–D). Thus, multi-specific TCRs from VDJdb shared multiple sequence features with PIT-matched TCRs, including shorter TRA but not TRB chains, had fewer N region nt and encoded more hydrophobic AA sequences.

Fig. 5: TRA junctions from cross-reactive TCRs in the VDJdb database shared junction sequence features with PIT-matched TRA junctions.
figure 5

A Density of TRA chains from TCRs with one versus multiple specificities in VDJdb. There were n = 17,826 unique TRA chains from TCRs with single specificity and n = 1664 with multiple specificities. B Distribution of TRA CDR3 nt lengths from TCRs with one versus multiple specificities; C Distribution of TRA N region nt lengths. D Distribution of TRA junction AA hydrophobicity. Solid vertical lines, median values from TRA chains from multi-specific TCRs; dashed vertical lines, median values from TCRs with single reported specificities. Significance of differences in distributions were assessed using two-sided Kolmogorov-Smirnov tests.


It has long been unclear whether and how rare autoreactive cells in peripheral blood represent autoimmunity in target organs. In other words, are these cells drivers or passengers in autoimmune processes? To help resolve this question, we show here that IAR CD4 + T cell TCRs from peripheral blood share matching TRA chains, and lesser numbers of TRB chains, with PIT TCRs. We also show suggestive, although non-conclusive evidence that frequencies of PIT-matching TRA chains in blood increase prior to the time of diagnosis. This may indicate a temporal linkage of levels of PIT-matched TRA chains in blood with disease progression. Shared TRB chains between peripheral blood and insulin-reactive TCRs from the pancreas were reported recently19. These combined studies place potentially autoreactive public TCR chains at the scene of disease (the pancreas) at the right time (early in the disease process) to influence disease development.

There are several potential functional roles for these shared, public TRA junctions. Since PIT-matched TCRs, especially TRA junctions, show TCR convergence, it is possible that some PIT-matched TCRs share islet autoantigen binding. Convergence in biology is often associated with selection, and convergence of TCR sequences has been associated with antigen specificity in other systems28,29,30. Direct evidence to support shared specificity in the present case is lacking, largely because it is unknown how many PIT-matched IAR TRA chains show functional autoreactivity. It is not yet possible to accurately predict that TCRs share specificity without complete identity of both chains, necessitating laborious experimental work to determine TCR specificity.

It is also possible that the sharing of TRA and divergence of TRB junctions reflects different roles of these chains in TCR binding. For example, others have shown that the TRA chain in a “TRA centric” MHC class I-restricted TCR determines antigen specificity, with paired TRB chains regulating avidity41. Finally, it is possible that there has been selection for another function of PIT-matched TCRs other than directly recognizing islet antigens. Perhaps sharing and divergence arise from selection in the thymus. Yet another possibility, suggested by enrichment of TRAV41*01 V genes in PIT-matched TCRs, is that sharing and convergence reflect evolutionary remnants of a system of innate-like T cells in mammals35. Although out of the scope of the present study, it is also worth considering whether PIT-matched TRA junctions are related to dual TRA chain TCRs that have been proposed to modify various aspects of T cell function, including thymic selection42.

T cell specificity is key to cellular immunity43. Paradoxically, TCR cross-reactivity is necessary because the diversity of TCRs, while huge, is dwarfed by the vast array of potential foreign peptide-MHC complexes44,45. Cross-reactivity expands the potential range of a given repertoire towards foreign antigens, but it comes at the expense of greater potential for increased self-reactivity or autoimmunity. These considerations emphasize the importance of understanding of sequence and structural features determining TCR cross-reactivity. Unfortunately, at present such an understanding remains incomplete43,46. Our data show that cross-reactive (multi-specific) TCRs share sequence features with PIT-matched TCRs, namely shorter and more hydrophobic TRA junctions and diverse TRB chains, reminiscent of a previous report in mouse47. Supporting this conclusion, the few islet-specific TCRs for which we have demonstrated cross-reactivity are PIT-matched TCRs. Earlier work demonstrated shortened32 and more hydrophobic31 TRB chains in bulk T cell subsets from T1D donors, and suggested that these features are important in development of self-reactive T cells.

Molecular modeling showed that germline-encoded CDR1, rather than CDR3, residues from TRAV12-2 contributed critically to the binding of an immunodominant TCR to theYellow fever virus48. This feature may contribute to the high precursor frequency and immunodominance of this TCR48. PIT-matching IAR T cell TRA chains also employed less non-templated and more germline-like (innate) mode(s) of epitope engagement. TRA CDR3 regions of PIT-matching IAR varied from the genome by relatively few N region nucleotide sequences. In addition, PIT-matching IAR T cell TRA chains showed shorter CDR3 regions and longer TRAJ segments than non-matching IAR T cell TRA chains. TCR CDR1 regions contact both peptide epitope and MHC residues, while CDR2 regions generally contact only MHC residues23. Thus, TCR regions important for peptide binding were shortened in PIT-matching TRA chains and had altered peptide binding properties. Our results also suggest that these shorter and more hydrophobic, germline-like, TRA chains are a feature of many cross-reactive TCRs. While some TCRs that predominantly utilize germline-like TRA CDR3 regions are cross-reactive49, this has not been previously demonstrated on the global level we show here.

One caveat to our study is that we have addressed only sharing of public TCRs. This was by necessity, since TCR sequences from IAR and PIT T cells from the same individuals were not available. While this feature enhances broadness of potential translational applications, it also limits mechanistic conclusions. Future studies may be able to address this weakness by testing of additional nPOD tissues18, including spleen and pancreas draining lymph nodes, from the same donors used for PIT TCR identification. Another limitation is that our study is underpowered with respect to numbers of donors and for TCRs of known specificity needed to directly test our hypothesis that PIT-matched TCRs tend to be multi-specific. It also is important to note that the TCR repertoires we compared were all determined using different technologies, which can lead to systematic biases50. A more comprehensive collection of autoreactive TCRs with known specificity from more donors, and matched comparator repertoires determined using the same technology, remain aspirational goals for future studies.

Our studies have potential translational implications. Typically, studies on TCRs as biomarkers have taken a reductionist approach by focusing on one or a few TCRs to monitor disease progression and/or response to therapy19,51,52. Our broader-based studies place TCRs with distinctive TRA sequence features and potential cross reactivity in both the pancreas and blood in early stages of T1D progression. These distinctive TRA-centric TCRs may provide an alternative source of potential biomarkers and targets for future translational studies.


Donors and samples

Donor group characteristics are summarized in Supplementary Data 1. The donor groups were defined as follows: HC, age-matched healthy donors; 1 Aab, donors at risk for T1D having one islet-directed autoantibody; 2Aab, at risk donors having two or more islet-directed autoantibodies; newT1D, T1D donors, within three months of original diagnosis; and T1D, donors with established T1D for a mean of 3.79 years (range, 1.6 −7.1 years). IAR CD4 + T cells were isolated from peripheral blood or cryopreserved PBMC samples for these studies (Supplementary Data 1).

Cohort 1 samples from HC, newT1D, and T1D participants were described previously12. Briefly, T1D donors were from the Benaroya Research Institute Disease Registry and Repository (BRI DRR). newT1D samples were from the placebo arms of the AbATE (NCT00129259) and START (NCT00515099) clinical trials sponsored by the Immune Tolerance Network53,54. HC samples were collected following informed written consent from healthy volunteers, matched for age and sex to T1D individuals with T1D and had no personal or family history of T1D. HC, T1D, and most newT1D donors had high-risk DRB1*0401 HLA class II alleles. Participant characteristics were summarized previously12. Samples from the BRI DRR were collected after informed written consent from donors, with approval from the BRI Institutional Review Board (IRB7109-332).

Cohort 2 samples were from the Type 1 Diabetes TrialNet Pathways to Prevention study (TN01) for the at-risk donors. Samples from TN01 participants were collected under the auspices of TrialNet ancillary study #201 and analyzed under ancillary study #200. Samples from TN01 donors were collected after informed written consent by the BRI Immune-mediated Disease Registry and Repository for sample storage and distribution (IRB07109).

Pancreas tissue or purified islet cell samples were obtained from organ donors identified by the Network for Pancreatic Organ Donors with Diabetes program (nPOD; RRID:SCR_01464118, a protocol at Vanderbilt University Medical Center/University of Pittsburgh16; the Integrated Islet Distribution program (IIDP; RRID:SCR_014387); and the Alberta Diabetes Institute Islet Core (ADI). Organ Procurement Organizations (OPO) partnering with nPOD to provide research resources are listed at nPOD obtains pancreas and other tissues from deidentified, deceased organ donors from partnering organ procurement organizations (OPOs) in the US under IRB Protocol #201600029 from the University of Florida. OPOs obtain consent for research from the donors’ families. The use of sequencing data from cadaver spleen and pancreas nPOD samples was approved by the BRI IRB as presenting no more than minimal risk to human donors and, therefore, qualifying for a Certificate of Exemption under 45 CFR 46.104(d)(4)(ii). Information, which may include information about biospecimens, was recorded by the investigator in such a manner that the identity of the human subjects cannot readily be ascertained directly or through identifiers linked to the subjects.

IAR CD4 + T cell TCRs

We isolated IAR T cells from peripheral blood using an overnight activation induced marker (AIM) assay based on upregulation of CD154 and CD6912,13. Samples from Cohort 1 were stimulated with a peptide pool of immunodominant DRB1*0401, DRB1*0301, and DQ8 restricted peptides from the islet proteins GAD65, IGRP, ZnT8, and preproinsulin (20 aa in length)12,13. In short, PBMC were stimulated with a pool of 35 immunodominant peptides from GAD65, IGRP, ZnT8, and preproinsulin restricted to the high-risk HLA class II DRB1*0401, *0301, or DQ8 molecules12,13. PBMC from donors in Cohort 2 were stimulated in an HLA agnostic approach with overlapping peptide libraries (20 amino acids (AA) in length, 12 AA overlap) from the above islet proteins (Supplementary Data 4). For both groups, CD154+ islet peptide activated cells were magnetically enriched, isolated as CD154 + CD69+ by single cell flow sorting and subjected to scRNA-seq to identify paired TRA and TRB chains in IAR T cells. The flow cytometry gating strategy used for Cohort 2 samples is shown in Supplementary Fig. 13 and antibodies are listed in Supplementary Data 5. Prior to use in this study, IAR CD4 + T cell TCRs with in-frame protein sequences were selected and filtered by removing iNKT cells with the CVVSDRGSTLGRLYF junction (n = 21 cells, 47 junctions); and MAIT cells with the TRAV1-2 V gene and TRAJ33, TRAJ20 or TRAJ12 J genes (n = 18 cells, 45 junctions). Complied and filtered TCR sequences are presented in Supplementary Data 2 and at Some sequences were described previously12 and are also available at

The P196-1 TCR was identified in experiments similar to those previously described for IAR CD4 + T cells12, but using the Influenza A MP54 97–116 peptide instead of islet peptides. The near identity of the P196-1 TCR to IAR CD4+ TCRs was discovered by sequence matching against an internal database of experimentally determined antigen specific TCRs identified using scRNA-seq. A single codon-optimized DNA fragment flanked by Not1-Spe1 restrictions and encoding the TRA and TRB junctions of the P196-1 TCR, integrated into the ‘TCR flex’ pMP71 sequence upstream of the murine Trac and Trbc genes55 (Supplementary Data 6), was prepared synthetically (GenScript, Piscataway, NJ). The synthetic fragment was then incorporated as a NotI and SpeI restriction fragment in place of the green fluorescent protein (GFP) sequence in the lentiviral vector, pRLL-MND-GFP56.

PIT TCR sequences

TCR alpha and beta chain sequences expressed by T cells in the islets or pancreas tissues were amplified by PCR57,58. Briefly, single cells stained with anti-CD3 and anti-CD4 or anti-CD8 antibodies were sorted into each well of 96-well plates, followed by reverse transcription synthesis of cDNA. TCR alpha and beta chain genes were amplified by multiplex PCR using primers binding to V gene and the constant region, Illumina sequencing linkers were added by additional PCR, and fragments were sequenced on Illumina sequencers. V gene, J gene, and junction sequences were identified using IMGT/HighV-QUEST59. TCRs having with in-frame protein sequences were selected for further analysis. IAR TRA junctions having 0 or 1 mismatches with PIT TCRs were designated as “PIT-matched”. TRB junctions were considered “PIT-matched” if they were paired with a PIT-matched TRA junction.

Other TCR sequences

TCR sequences from COVID-19 patients and HC21 were obtained from VDJdb40 TCR sequences (versions 2022-11-24 and 23-09-16) were downloaded from VDJdb V/J/CDR3 protein sequences from VDJdb (updated 09/16/23) were converted into complete coding sequences representing fully spliced TCR cDNA sequences using Stitchr60. TCR sequence features were derived from Stitchr-produced cDNA sequences by IMGT/HighV-QUEST59. In some cases, comparisons were made between groups of TCRs of different sizes. To confirm that this disparity between different set sizes did not bias the results, we normalized group sizes, down sampling by randomly selecting with replacement subsets of the larger group(s) matching in size with the smaller group. Comparisons were made between results obtained before and after down sampling to ensure that the comparison made with the full data set was accurate. We repeated these comparisons multiple times to ensure that the down sampling comparison was representative. Reported results utilized the full data sets.

Transduction of human CD4 + T cells

Human CD4 + T cells were transduced with lentiviral TCR expression vectors12. Briefly, CD4 + T cells from HC donors were stimulated for with human anti-CD3/CD28 monoclonal antibodies (mAbs), then incubated with recombinant TCRs cloned into a modified ‘TCR flex’ pMP71 vector12 upstream of the murine Trac and Trbc genes55. Transduction efficiency was determined by staining with a mAb targeting the murine Trbc chain (H57-597, allophycocyanin (APC)-labeled, BD Biosciences, San Jose, CA). Transduced T cells were labeled with 5(6)-Carboxyfluorescein diacetate N-succinimidyl ester (CFSE) and mixed with irradiated antigen-presenting cells (APC) loaded with antigenic peptides. APC were either autologous PBMC; PMBC from MHC class II-matched individuals; or Priess lymphoblastoid cells, (ATCC, Rockville, MD). After incubation at 37 °C, cells were stained with anti- human CD4+ (Clone RPA-T4, phycoerythrin-labeled, BioLegend, San Diego CA), and anti- murine Trbc chain to identify CFSE dye dilution in transduced cells by flow cytometry12.

Flow cytometry

Antibodies used for flow cytometry are described in Supplementary Data 5. Peptide simulated PBMC cultures were surface stained with anti-CD154-PE followed by anti-PE-magnetic microbeads (Miltenyi) according to the manufacturer’s instructions. Magnetically enriched CD154+ cells were surfaced stained with PerCP-Cy5.5-coupled monoclonal antibodies specific for CD8, CD19, CD14, CD56, iNKT, and CD161 along with Viaprobe live/dead stain (BD, Franklin Lakes, NJ) as a dump channel. Monoclonal antibodies specific for the following surface markers were included to identify and phenotype IAR CD4 + T cells: CD3, CD4, CD45RO, CD45RA, CCR7, CD69, CCR4, CCR6, CXCR3, CD95, CD25, CD2, CD38, and SLAMF6. IAR CD4 T cells were identified as CD3 + CD4 + CD154 + CD69 + T cells according to the gating scheme shown in Supplementary Fig. 13 and single cell sorted into a 96 well plate using a BD FACSAria flow cytometer. Proliferation of CFSE-labeled TCR transduced CD4 T cells was determined by surface staining cultured cells with anti-CD4-PE and anti-murine Trbc-APC monoclonal antibodies (Supplementary Data 5) followed by flow cytometry using a BD LSRFortessa cytometer. CFSE intensity was assessed in CD4+mTrbc+ cells (Fig. 4B).

TCR expansion and sharing

TCR expansion was defined as the number of individual cells expression the same junction AA sequence. A junction was designated as “expanded” if the number of cells with the same junction AA sequence was >1. Conversely, a junction was designated as “non-expanded” if it was found in only one cell. TCR publicity for each junction in a TCR was defined as the number of donors expressing the same junction AA sequence. A “public” junction was defined as one found in multiple donors; a “private” junction was expanded and found only in a single donor. We used IGoR software (v. 1.4.0) to estimate the generation probability (Pgen) of junction sequences during TCR recombination26. Junction sequences with higher (or less negative) Pgen values have a higher probability of generation by V(D)J recombination. TCR convergence was defined as the number of nucleotide (nt) sequences encoding each junction AA sequence. Junctions having >1 nt sequence associated with a single AA sequence were designated as “Converged” and junctions having only a single nt sequence per AA sequence, “non-Converged”.

Sequence comparisons

Pairwise Levenshtein distances between peptide sequences were calculated using the stringdist (v. 0.9.10) software package in R. The Levenshtein index is the number of residue changes needed to transform one sequence into another. Motif mapping using the MEME software was accomplished at Sequence features of TCR chains were determined by IMGT/HighV-QUEST61 analysis of nt sequences.

Sequence features from VDJdb TCRs

Since nt sequences used to identify junction sequence features are not readily available for VDJdb TCRs, we utilized the software package, Stitchr (v., to produce complete TCR cDNAs from V/J/CDR3 AA sequences60. We then used IMGT/HighV-QUEST (v. 1.9.4)61 to identify TCR sequence features from the predicted cDNA sequences.

TCR modeling

For modeling, we used TCRmodel2, which is based on the AI system, AlphaFold v2.362, a newly described method that predicts protein structure with high accuracy at unmatched scale63. TCRmodel2 uses focused databases of TCR and MHC sequences to expedite multiple sequence alignment feature building; optimization of the TCR template selection; and utilization of peptide–MHC complex structures as templates to improve modeling accuracy. We used the TCRmodel236 web server ( to predict TCR–pMHC complex structures. We saved best fit models for further analysis; these had Model Confidence scores36 from 0.74–0.91 (0.85 ± 0.04, mean + SD), where a score ≥ 0.85 is considered a “likely good” model. Models were viewed and analyzed using the Mol* Viewer64 ( TCR-peptide contact residues, (i.e., TCR residues <5 Å in distance from the bound peptide chain) were identified and mapped to TCR sequence features identified using IMGT HighV-QUEST59. Structural alignment was performed using the RCSB PDB65 web-based viewer (


Statistical tests were performed using the R programming language and software environment (v. 4.2.3). We utilized t-tests for group comparisons of continuous, normally distributed variables; Wilcoxon signed rank tests for non-normally distributed variables; Kolmogorov–Smirnov tests for distributions of numeric values; the hypergeometric test to determine the significance of set overlaps; and Fisher’s exact test for categorical variables. Unless otherwise noted, we performed non-paired, two-sided tests and assumed equivalent variation in the groups compared. Where appropriate, multiple testing corrections were made66. The term “significant” is reserved for p-values (single tests) or false-discovery rate (FDR)-adjusted p-values (pAdj) (multiple tests) of <0.05. Only significant P- and pAdj values are indicated on the Figures. Specific tests used to derive each listed P- or pAdj value are given in the text or in the Figure legends.

Inclusion and ethics

Benaroya Research Institute is committed to creating an environment where all people are, and perceive themselves to be, welcomed, valued, respected, and heard in order to develop, contribute to and achieve aligned organizational and scientific goals.

Study approval

Protocols for these studies were approved by the Institutional Review Board of Benaroya Research Institute (IRB7109-332 and IRB7109-460.02, for Cohorts 1 and 2, respectively). Protocols for collection of nPOD samples were approved under an IRB from the University of Florida (#201600029). Protocols for the clinical studies were approved under the auspices of NCT00129259 for the AbATE trial53 and NCT00515099 for the START study54

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.