As discussed above, there are many advantages of single cell sequencing TCRs over bulk TCR repertoire sequencing. Before the advent of large scale scRNAseq, many people in the type 1 diabetes field appreciated the importance of sequencing immunoreceptors on a single cell basis and linking this to antigen specificity and affinity (40–42). As of 2017 there were 1655 clonotypes of known specificity for T1D autoantigens (41), a number which has increased substantially with the advent of higher throughput scRNAseq.

These have been identified through a numbers of methods. HLA class I or class II multimers may be used to select antigen specific T cells. This has the advantage of being able to select cells from the peripheral blood but is limited by HLA restriction and to known epitopes or mimotopes (43). In addition non-specific binding may yield false-positive TCRs. Alternatively peripheral blood T cells can be stimulated in vitro with islet peptide pools and selected on the basis of upregulation of activation markers, allowing wider specificities and HLA compatibilities, but with the risk of bystander activation again resulting in false negatives. A third approach is to sample T cells directly from the pLN or pancreas, where islet-specific T cells will be massively enriched. These cells can then either be stimulated in vitro with peptide pools the TCRs re-expressed ex vivo to determine specificity. Alternatively, TCR sequences can be compared to those in the literature known to be islet antigen specific.

Eugster and colleagues performed an heroic effort to sequence paired TCRs from 1650 T cells that either bound a GAD tetramer or responded to GAD in vitro, by sorting single cells from the peripheral blood and performing plate based scRNAseq (44). GAD specific TCRs were highly heterogenous both within and between donors, with no shared TCRs between donors, although individual TCRα or TCRβ chains were often shared. Moreover, there was limited overlap between the TCRs identified by tetramer binding and T cell activation methods, indicating that epitope recognition and MHC usage by GAD specific TCRs was likely to be broad (44).

Cerosaletti et al. performed scRNAseq of islet-reactive TCRs from the peripheral blood (identified by ex vivo response to stimulation with islet-peptide pools). They found that T cells from T1D had higher numbers of identical CDR3, which had arisen by clonal expansion (i.e. T cells with identical TCRα and TCRβ chains, that have arisen by division of a parent cell), rather than convergent recombination (24). By re-expressing the TCRs in cell lines it was found that many of these TCRs in people with T1D were IGRP specific (24). It was further shown that donors with T1D had large clonal expansions of IGRP-reactive T cells in the peripheral blood and frequently used a specific shared TCRα chain, which was paired with different beta chains in each donor (45). Preferential usage of TRAJ53 and TRAV29 and TCRα chains bearing the motif SGGSNYKLTF were identified in single cell TCR sequencing of people with T1D. When a bulk sequencing approach was taken, a particular TCRα chain bearing this motif was highly enriched in the memory CD8+ T cells of autoantibody positive people and those with T1D compared to controls. Clones bearing the motif were also shown to directly kill IGRP- peptide bearing cells (45). T cell clones bearing both IGRP (24, 45) and hybrid insulin peptide- responsive TCRs are persistent over time (46). However, others have examined TCR repertoires in children progressing to diabetes and shared TCRs were not seen either between children or within the same child over time, indicating high diversity in the peripheral blood at this age (27).

Early work examined T cells from the pancreatic lymph nodes (pLN) of people with T1D and found high clonal expansions (47). Additionally there were many T cells that shared a TCRβ but had divergent TCRα. Many clonally expanded CD4+ T cells recognised insulin A1-15 in the context of DR4 (47). Pathiraja et al. grew out CD4+T cells from the pancreatic islets of a donor with T1D using anti CD3 and cytokine stimulation. Over 25% of these clones had TCRs that responded to proinsulin peptides restricted by HLA-DQ8 or the HLA-DQ8 transdimer and 30% of clones used TRBV5–1*01 (48). Whilst it is difficult to make direct comparisons to frequencies of islet-reactive T cells in the peripheral blood (26), it is clear that in the peripheral blood frequencies are much lower [around 0.01-0.05% of T cells in people with T1D (3, 44, 49)]. Most T cells isolated from the pancreas had unique clonotypes, whilst the majority of in vivo clonally expanded T cells were specific for proinsulin (48). It has also been found that ZnT8- reactive T cells were present at similar frequencies in the blood of healthy controls and people with T1D, but were enriched in the pancreas of the latter (26). Single cell sequencing of TCRs found a public ZnT8 specific CDR3B in the peripheral blood, and enriched in the pancreas of people with T1D, although the full TCRβ had divergent sequences due to different gene usages. ZnT8 reactive T cells also showed a bias towards TRBV19 and TRAV12-2 usage (26).

Seay et al. also found sharing of CDR3s between donors in the pancreas, with a TCRβ with homology to a known GAD reactive TCR found in 7/18 T1D donors (50). Furthermore a shared CDR3β chain was found in all people with T1D in the conventional T cell compartment, whilst in healthy controls it was predominantly in the Treg compartment (50). Interestingly TCR sequencing of GAD-responsive CD4+IL-13+ T cells from patients who had received injected GAD Alum found that they often used a highly public TCRβ (TCRα sequencing was not available) (51).

Direct capture of pancreatic T cells by the Nakayama group enabled single cell TCR sequencing and confirmed infiltration of proinsulin specific cells into the pancreas in T1D. Of the hundreds of TCRs sequenced, most were present only on a single cell, indicating a diversity of response even years after diagnosis. Clonal expansions were more likely in CD8+ T cells and these clones were found in multiple islets from the same donor, indicating in vivo migration. Furthermore, across three donors it was noted that whilst there were no identical TCRs, there were identical TCRα sequences and TCR subunits (52). When the TCRs were re-expressed, the B9-23 reactive TCRs isolated from the pancreas induced much higher IL-2 secretion compared to control B9-23 TCRs isolated from peripheral blood (52) which may indicate the former have a higher affinity for B9-23. Moreover, only the pancreas-derived TCRs were capable of a response to whole proinsulin presented by APCs (52).

Recently the Nakayama group has reconstructed individual TCRs from the pancreas of people with T1D. TCRs were selected for re-expression on the basis of clonal expansion or V gene usage previously associated with proinsulin C19-A3 specificity, and were found to recognise epitopes across preproinsulin and presented by a variety of MHC class II (53). Many TCRs recognised peptides in the region of B9-23, but others, (many from clonally expanded cells) recognised peptide right across from the signal peptide to the A chain. Furthermore, these TCRs recognised peptides in the context of diverse MHCII, although a preference was shown for DQ (53). Even with these constraints of the selection criteria in this study, this single cell approach showed a diversity of peptide and MHCII specificity that would have been missed using tetramers.

Taken together, the evidence suggests that T cells with TCRs with higher affinity for diabetes autoantigens are more likely to be found in the pancreas than in the peripheral circulation. This represents a major challenge in T1D research as in other autoimmune diseases it is relatively straightforward to obtain samples from the site of autoimmune attack (54). For example in psoriatic arthritis, extraction of viable T cells directly from the affected joints enabled sequencing of paired TCR receptors and scRNAseq profiling of cells phenotypes (55). Even in the pancreas, clonal expansion is modest and whilst CDR3 sequences specific for many diabetes autoantigens are shared between donors, there is not yet evidence of truly public TCRs with identical TCRα and β chains. However, more widespread use of the VDJdb repository (56), IEDB (57) and the JDRF/nPOD CloneSearch might allow enriched motifs to become apparent across different experiments, although this would still be limited by HLA restriction. To further complicate the picture, scRNAseq has demonstrated that islet antigen –reactive T cells (24) and HIP reactive T cells in particular (58) sometimes express two TCRα chains, which are known to contribute to autoimmunity (59, 60).

Combining TCR sequencing (or selection based on autoantigen reactivity), with scRNAseq has the potential to give further insights into T cell function. This has not always been straightforward to demonstrate, for example analysis of IGRP- specific T cells from the peripheral blood did not show a distinctive gene expression (GEX) pattern in response to stimulation (25). Similarly scRNAseq of ZnT8 reactive cells from the peripheral blood of people with T1D showed similar GEX profiles to healthy controls, indicating that these peripheral T cells may not be playing a driving role in T1D, although T1D patients had higher expression of aryl hydrocarbon receptor (AHR) and aurora kinase A (AURKA) and lower expression of RORA (26).

The approach was more successful for Heninger et al, who demonstrated that proinflammatory responses to diabetes autoantigens were dominant in children who progressed to autoantibody positivity, whilst regulatory T cell responses were seen in those who didn’t (27). An algorithm based on gene expression in response to autoantigens enabled identification of which children would later progress to autoantibody positivity. As this group developed autoantibodies the GEX profiles of their CD4+ T cells changed towards increased expression of Th1 genes (27). These findings suggest that biomarkers of T1D susceptibility may allow identification of at risk children prior to seroconversion (61).

In addition, Cerosaletti et al. used islet peptide pools to stimulate T cells from the peripheral blood in vitro and characterised those that activated by scRNAseq. They did not observe a significant level of differentially expressed genes between healthy controls to those from people with T1D. However, when they focussed on cells from people with T1D that were highly clonally expanded (termed T1D-E cells), they found that these cells did have a unique transcriptional profile compared to islet reactive T cells from healthy controls or those from people with T1D that were not clonally expanded. T1D-E cells preferentially expressed genes associated with T cell activation and leukocyte differentiation (24). These experiments demonstrate how focussing on antigen specificity can enhance findings from scRNAseq.

Early interest in autoantibodies in T1D, before the advent of scRNAseq, focussed on isolating GAD-specific B cells from people with T1D (62) and sequencing the BCRs from clones, which provided evidence that GAD autoantibodies have frequently undergone somatic maturation and are therefore from antigen-experienced B cells (63, 64). Similarly IA-2 specific antibodies sequenced from B cells from people with T1D also show evidence of somatic mutation (65–67). Anti-insulin antibodies have been sequenced from people with T1D, but may have arisen in response to injected insulin rather than endogenous insulin (68, 69) [reviewed (70)]. BCR sequencing combined with phenotyping of B cells has given great insight into B cell response in other autoimmune diseases (71) and in response to vaccinations (72) and in B cell lymphoma (73). Yet without the equivalent of a tetramer approach to identifying autoreactive B cells, phenotyping and characterisation of the BCR has lagged behind T cell research in T1D.

Smith et al. developed an approach to isolate insulin reactive B cells from the peripheral blood of people with T1D, by flow cytometric sorting B cells that bound insulin conjugated to fluorescent tags. The authors were then able to sequence BCRs from individual cells. They demonstrated that insulin binding BCRs preferentially used JH6 gene segments which have previously been associated with autoreactivity and were biased towards use of positively charged amino acids in the CDR3 region (74). When re-expressed as antibodies, BCRs from anergic naive IgD⁺, IgM⁻ B cells demonstrated binding at levels thought to induce anergic B cell responses, whilst those from naïve B cells bound weakly and would likely be ignorant of insulin under physiological conditions (74).

scRNAseq has been used to characterise a novel lymphocyte population that express both TCRs and BCRs (28). It is suggested that these “dual expressors” (DE) are increased in frequency in type 1 diabetes and that in people with type 1 diabetes there is a public BCR which can stimulate insulin-reactive CD4+ T cells. However, this work remains controversial as others have been unable to replicate the enrichment of DE in T1D nor the specific public BCR sequence (75). This highlights the importance of good quality control at every step of scRNAseq experiments.

Isolation of CD19+IgG+ B cells from pancreatic lymph nodes from autoantibody positive donors and single cell sequencing of their BCRs demonstrated that no clonally expanded B cells were identified in the pLN. Antibodies were reconstructed from BCR sequencing, although very few of these were found to be specific for IA2 (none were specific for GAD and insulin was not tested) (76). Seay et al. also sorted and single cell sequenced the BCRs from pancreatic LNs. They found an enrichment of insulin binding motifs in pLN from people with T1D compared to controls (50). They also observed sequence overlap with autoreactive BCRs cloned from precursor (early immature) B cells from healthy donors previously published by Wardemann et al. (77). Wardemann et al. observed that not only are many BCRs from healthy donor precursor B cells insulin reactive, they are often also polyreactive to other autoantigens for example dsDNA, ssDNA or nuclear proteins (77). This polyreactivity has also been noted for both IgM and IgG insulin antibodies (68). Similarly Smith et al. demonstrated that all of their high affinity insulin binding BCRs were also reactive to LPS and chromatin (74). Polyreactive antibodies have been postulated to play a key role in the healthy immune system but are also implicated in a variety of autoimmune diseases (78, 79). It therefore appears that autoreactive B cells in T1D may span a wide range of phenotypes and the antibodies produced may often be polyreactive, however the limited number of studies make it difficult to draw firm conclusions.

scRNAseq is beginning to give fascinating insights into type 1 diabetes and new approaches may yield further discoveries. The first of these is spatial transcriptomics (80). In this technique, indexed oligos capture RNA from either fresh-frozen or formalin-fixed, paraffin-embedded tissue sections. This allows determination of gene expression on a level that is fast approaching single cell resolution. It has already been used to give insights into cell interactions in other diseases such as rheumatoid arthritis, where infiltrating leukocytes interact with target cells (81, 82). Spatial transcriptomics therefore has great potential to unravel lymphocyte interactions with beta cells in the pancreas and to give insight into different patterns of immune cell infiltration (2). In both type 1 (23) and type 2 diabetes (83) scATACseq has recently been used to link GWAS to epigenetic regulation of gene expression. New methodologies enabling combination of ATACseq, and CITEseq with scRNAseq in the same experiment will also contribute to the field (84). New analysis tools such as CellPhoneDB give the ability to map interactions between subsets of cells, based on DEG in scRNAseq datasets, which would allow identification of novel interactions between immune cells and beta cells in the pancreas (85, 86). This may become increasingly important as we begin to understand the role of beta cell stress and signalling in type 1 diabetes (6) as well as the involvement of other pancreatic cells in diabetes development (23). CellPhoneDB has been used to identify crosstalk between T cells and epithelial cells in ulcerative colitis (87) whilst in rheumatoid arthritis scRNAseq has revealed interaction pathways between B cells, fibroblasts and monocytes (88). Additionally, recent work from the Satija lab has brought together previously published scRNAseq datasets of pancreatic cells, including immune cells from healthy pancreatic samples (12), which will facilitate this type of analysis. This would be further enhanced were there a unified repository for T1D scRNAseq datasets, similar to those for COVID-19 (89).

We have seen how combining GEX with V(D)J sequencing has increased insights into T1D. The recent development of DNA barcoded multimers will allow now the determination of T cell antigen specificity in scRNAseq experiments (90, 91), whilst conjugation of whole proteins or large folded protein fragments to DNA barcodes will facilitate identification of antigen specific B cells (92).

Computational approaches to determine the likely interaction of an immunoreceptor with target antigen also have the potential to revolutionise the search for antigen specific TCRs and BCRs. Approaches such as tcrdist (93), GLIPH (94) and immune receptor network generation for BCRs (95) enable BCR and TCR sequences to be mapped and visualised, and those that differ by only one or two amino acids are assumed to target the same antigens. NetTCR (96) and TCRex (97) use neural networks and machine learning algorithms to cluster TCRs predicited to bind the same epitope. Recent advances such as ICON and TCRAI leverage scRNAseq technology along with oligo labelled dextramers. They utilise the paired TCRα and TCRβ transcripts to build libraries of antigen specific receptors, with a neural network to predict antigen specificity of TCRs. However, many of these approaches have been validated using viral or tumour antigens with well-defined epitopes. As we have seen in the sections above, whilst there definitely are peptide sequences from diabetes autoantigens that are widely recognised, the immune response also targets diverse sequences in different individuals. Furthermore, auto-antigenic TCRs tend to bind pMHC with lower affinity than TCRs targeting pathogens (98, 99) as high affinity self-reactive TCRs are generally deleted in the thymus. It is not clear how this lower affinity and lack of public TCRs may impact upon the usefulness of computational approaches for T1D.

In 2019, it was demonstrated that teplizumab could delay progression to T1D in high risk individuals (100). Further work confirmed a correlation between fold change in C-peptide and change in frequency of CD8+KLRG1+TIGIT+ T cells (101). scRNAseq of T cells from the clinical trials of teplizumab and other immunotherapies in T1D could offer an amazing opportunity to identify all biomarkers predictive of successful treatment. For example, scRNAseq studies have shown a variety of phenotypic markers induced in vitro with anti-CD3 antibodies in human PBMC, including a variety of interleukin receptors and markers of regulation and exhaustion including FOXP3, CTLA4, TNFRSF18, LAG3 and PDCD1 (102). In contrast, anti-CD3/CD28 stimulation of PBMC analysed with scRNAseq and CITEseq, showed phenotypes strongly associated with activation (although memory subsets also upregulated senescence) (103).

In the future, a deeper understanding of TCRs and BCRs has the potential to better quantify the risk of progression in autoantibody positive people. Monitoring the abundance and phenotypes of lymphocytes bearing specific CDR3 sequences or using specific V genes may also prove useful in monitoring immunotherapies, particularly antigen specific immunotherapies, where phenotypic changes in whole lymphocyte populations may not be so obvious (1, 104–106). In addition, BCRs also have the potential to be used in CAR-Treg cell immunotherapy as has been demonstrated in the NOD mouse (107).

scRNAseq has demonstrated its great potential to identify novel biomarkers both in T1D and other autoimmune diseases. However, it is both technically challenging and expensive. Therefore it is crucial that researchers should be able to translate findings from scRNAseq into more accessible diagnostic and monitoring tests, for example using standardised flow cytometry or qPCR panels as is starting to happen in cancer research (108, 109). Similarly in IBD, a machine learning approach allowed identification of a CD8+ T cell signature that could predict prognosis. These biomarkers were then developed into a commercially available whole blood qPCR test to facilitate personalised therapy (110).

In T1D, recent advances in computational analysis are beginning to allow discrimination of changes in cell subsets from bulk RNAseq. Mehdi et al. identified a peripheral blood transcriptomic signature that predicted autoantibody development (111). Of the DEG identified, many were associated with the ubiquitin-proteasome pathway, DC and T cell function and were potentially targets of drugs approved for other conditions (111). Xhonneux et al. (112), demonstrated from transcriptomics of whole blood that they could undertake “digital cytometry”, by mapping groups of genes back to cell types. Children who developed autoantibodies against insulin first, had a signature of increased NK cells and CD4+ memory T cells. In contrast, those who first developed autoantibodies to GAD had a reduced percentage of CD4+ memory T cells and NK cells, but increased activated NK cells. Harmonizome (113) was used to identify a G protein–coupled receptor, GPR171, predicted to control the immune signature found in IAA+ children (112). Adding gene expression information to predictive models, increased their accuracy in predicting later T1D development in children under 18 months (112).