Insulin peptide recognition by early diabetogenic T cells
Type 1 diabetes is initiated by loss of T cell tolerance to pancreatic islet autoantigens including insulin. Gioia et al. developed peptide-MHC tetramer reagents capable of distinguishing separate populations of insulin-reactive CD4+ T cells from diabetes-prone NOD mice that differ in the register used for insulin peptide binding to class II MHC. Analysis of pancreatic islet tissue using these tetramers revealed that the earliest phase of the anti-insulin T cell response in islets is dominated by T cells with TCR sequence modifications that enable recognition of an Ins12–20 peptide–MHC class II complex. Identification of the primary mode of peptide recognition used by the early anti-insulin T cells opens the door to designing inhibitors capable of selectively blocking activation of these rogue T cells.
The class II region of the major histocompatibility complex (MHC) locus is the main contributor to the genetic susceptibility to type 1 diabetes (T1D). The loss of an aspartic acid at position 57 of diabetogenic HLA-DQβ chains supports this association; this single amino acid change influences how TCRs recognize peptides in the context of HLA-DQ8 and I-Ag7 using a mechanism termed the P9 switch. Here, we built register-specific insulin peptide MHC tetramers to examine CD4+ T cell responses to Ins12–20 and Ins13–21 peptides during the early prediabetic phase of disease in nonobese diabetic (NOD) mice. A single-cell analysis of anti-insulin CD4+ T cells performed in 6- and 12-week-old NOD mice revealed tissue-specific gene expression signatures. TCR signaling and clonal expansion were found only in the islets of Langerhans and produced either classical TH1 differentiation or an unusual Treg phenotype, independent of TCR usage. The early phase of the anti-insulin response was dominated by T cells specific for Ins12–20, the register that supports a P9 switch mode of recognition. The presence of the P9 switch was demonstrated by TCR sequencing, reexpression, mutagenesis, and functional testing of TCRαβ pairs in vitro. Genetic correction of the I-Aβ57 mutation in NOD mice resulted in the disappearance of D/E residues in the CDR3β of anti-Ins12–20 T cells. These results provide a mechanistic molecular explanation that links the characteristic MHC class II polymorphism of T1D with the recognition of islet autoantigens and disease onset.
The association between HLA genes and autoimmune diseases was uncovered more than 40 years ago (1). Among them were type 1 diabetes (T1D) (2) and a linkage to human leukocyte antigen–DR3 (HLA-DR3) and HLA-DR4 that explained the vast majority of the genetic component of this serious disease (3, 4) in which the exclusive destruction of the β cells of the islets of Langerhans of the endocrine pancreas leads to a lifelong dependency on insulin replacement therapy. The linkage to two HLA-DR haplotypes was later redefined as an association with the HLA-DQ haplotypes that segregate with these HLA-DR genes, HLA-DQ2 and HLA-DQ8 for HLA-DR3 and HLA-DR4, respectively. Relative risk is higher for HLA-DQ2 and HLA-DQ8 homozygotes than for heterozygotes and maximal for HLA-DQ2/HLA-DQ8 heterozygotes (5). In 1987, McDevitt’s group made the important observation that every HLA class II associated with T1D was carrying a distinct polymorphism at position 57 of the β chain that substituted the normal aspartic acid of all major histocompatibility complex (MHC) class II β chains at this position by a neutral residue (6). This notable observation has been confirmed since in one of the largest genetic studies of T1D (7). Structurally speaking, the consequences of this alteration are the loss of a salt bridge with the arginine 76 of the MHC class II α chain and the appearance of a surface-exposed positively charged patch that modifies both the P9 pocket and potential T cell receptor (TCR) contacts (8, 9). We, and others, have shown that the loss of the salt bridge had no consequence on the structural integrity of the molecule (8, 10) and that it remained stable and competent for peptide binding. As expected, the modifications of surface charges at the P9 pocket affect peptide binding profoundly, and the peptide repertoire of diabetogenic MHC class II molecules is heavily biased toward the selection of peptides with acidic residues at P9 (11, 12). However, similar to most I-A and HLA-DQ molecules, diabetogenic MHC class II proteins remain very promiscuous for peptide binding as they interact mainly with the peptide backbone instead of using anchor residues (13, 14). The consequence of this mode of binding is that all MHC class II molecules without Aspβ57 can also efficiently bind peptides that do not have a negatively charged residue at the P9 position (8, 14). In this case, a large positively charged patch remains surface-exposed and potentially accessible to T cell recognition. We have evaluated this situation by immunizing nonobese diabetic (NOD) mice and HLA-DQ8 transgenic NOD mice with peptides carrying or lacking a negatively charged residue at the P9 position. In both instances, we have shown that the absence of charge at that position in the peptide resulted in the selection of TCRs that encoded either an Asp or a Glu residue at position 2 or 3 of their complementarity-determining region 3β (CDR3β) (9, 15). For one of those peptides derived from hen egg lysozyme (HEL) that has a glycine at P9, we also demonstrated biophysically and structurally that the presence of a negatively charged residue at position 2 or 3 of the CDR3β increased the affinity of the TCR for its cognate peptide-MHC (pMHC) complex by more than 30-fold (9). We called this mode of TCR recognition the “P9 switch” and suggested that it might be important in the initiation of diseases such as celiac disease and T1D, which are tightly associated with non-Aspβ57 MHC class II molecules. However, in the absence of important reagents such as appropriate MHC tetramers and/or animal models (e.g., a celiac disease mouse model), we could not formerly test our hypothesis. The development of I-Ag7 tetramers capable of distinguishing the two main registers of the Ins9–23 peptide (16) was the first important step in testing the relevance of the P9 switch model in T1D because one of them, Ins12–20, has a glycine at P9, whereas the other one, Ins13–21, bears a glutamic acid at that position. Recognition of insulin requires its processing in intracellular compartments of antigen-presenting cells (APCs), resulting in the presentation of the Ins13–21 peptide segment. T cells that recognize the Ins13–21 peptide either from such processing of insulin or from exogenous peptides represent type A T cells reliant on conventional antigen presentation. In contrast, other peptide segments, such as Ins12–20, are eliminated by the intracellular processing of insulin and are therefore not normally presented by APCs. The Ins12–20 peptide segment can be presented when supplied as an exogenous peptide and is recognized by type B T cells. This alternate mode of presentation takes place when β cell crinosomes partially catabolize insulin molecules and release a number of insulin peptides that are then presented in islets and peripheral lymph nodes (17).
In any instance, if the P9 switch model were correct, one would expect three simultaneous findings: (i) CD4+ T cells specific for pMHC with neutral P9 residues should be preferentially expanded; (ii) these T cells should be activated earlier than cells recognizing pMHC with Asp/Glu (D/E) at P9; (iii) these T cells should rely on coulombic forces for interaction. We have already shown that cells specific for a BDC2.5 mimotope with a neutral P9 position were preferentially expanded in the NOD mouse and that those cells had higher affinity for their pMHC than cells specific for BDC2.5 peptides with Asp at the P9 position (9). We now demonstrate that the nature of the amino acid at position β57 controls the selection and expansion of CD4+ T cells specific for insulin and the onset of T1D in the mouse. Using single CD4+ T cell analysis, we show that the tissue of origin of anti-insulin T cells could easily be identified on the basis of characteristic gene expression signatures. Hallmarks of TCR signaling and T cell expansion appeared to be localized exclusively to the islet rather than the pancreatic lymph nodes (PLNs). Although islet-infiltrating T cells were either classic T helper 1 (TH1) or regulatory T cells (Tregs), this dichotomy was not associated with particular TCRαβ pairs. The early T cell response seen in the islets was dominated by anti-Ins12–20 cells. A majority of these cells carried a D/E residue in their CDR3β and used a P9 switch for recognition. The restitution of an aspartic acid at position β57 of I-Ag7 in vivo eliminated this population of cells. These results provide a mechanistic molecular explanation of why and how diabetogenic MHC class II molecules increase the risk of autoimmunity.
MHC class II tetramers specific for the two registers of Ins9–23
The expression of functional MHC class II tetramers remains difficult especially when multiple registers can be displayed within short peptides. The length and nature of the linker when expressing tethered peptides, the on and off rates of the peptides themselves, and the potential role of flanking residues are all factors to consider when expressing MHC class II/peptide complexes. In addition, in many instances, expression does not mean proper display as many recombinant MHC class II molecules are unable to activate and/or bind T cells that are specific for the cognate peptide (18). We have successfully expressed the various registers of ovalbumin in I-Ad and I-Ab using a short seven-residue linker (19) but failed to express functional Ins12–20 and Ins13–21 peptides in I-Ag7 using the same approach. The reengineering of both peptide and MHC to introduce artificial disulfide bridges, as proposed by some (20, 21), led to expression of nonfunctional molecules (i.e., no stimulation of T cell hybridomas and no T cell staining ex vivo). We then applied various modifications at anchor residues or N- and C-terminal flanking residues (P1 and P10) with the idea to limit register sliding or inadequate display; none was successful (table S1). Last, we explored the ability of various sequences such as polyglycine stretches to disrupt type II polyproline helices (22), the normal required secondary structure of peptides for binding to MHC class II molecules. The addition of three C-terminal glycines allowed the expression of Ins13–21, whereas the addition of four glycines allowed the expression of both Ins12–20 and Ins13–21 (i.e., VEALYLVCG-GGGG and EALYLVCGE-GGGG, respectively). The specificity and functionality of the recombinant molecules were tested by staining and stimulating T cell clones specific for each register (Fig. 1). Accordingly, type A cells of which IIT3 is a representative (23) were stained and stimulated specifically by recombinant I-Ag7 Ins13–21 molecules, whereas type B cells such as 2D10 were stained and responsive to I-Ag7 Ins12–20 molecules.
Ex vivo detection of insulin-reactive CD4+ T cells in naïve NOD mice
After this characterization, the MHC tetramers were tested ex vivo to quantify the CD4+ T cells from the thymus, spleen, and PLNs, and islet infiltration from NOD mice of different ages. The specificity of the reagents and the existence of two distinct populations of cells were confirmed by double staining (Fig. 1, A and B). In all cases, insulin-reactive cells were detected in the thymus and peripheral lymph nodes; consistently, Ins12–20 cells were twice as numerous as Ins13–21 cells in prediabetic mice, whereas both populations, at difference with BDC2.5 cells (24), showed no notable expansion between the thymus and PLNs (Fig. 1C). However, in the islets at 6 weeks of age, the Ins12–20 cells showed a 200- to 400-fold expansion, which contributed to most of the infiltrating CD4+ population of a majority of mice (Fig. 2, A to C). The Ins13–21 cells, although also more numerous in the islets than in PLNs, never represented more than 20% of the infiltrating CD4+ population, and a pairwise analysis of each mouse showed that Ins12–20 cells dominated the early phase of autoimmunity (Fig. 2C). Across all 6-week-old mice (n = 11), the mean representation of Ins12–20-specific cells in the islets was 52.05% of all CD4+ T cells (SEM = 8.0%), whereas Ins13–21-specific cells had a mean representation of 11.26% (SEM = 1.96%), establishing the highly significant difference of frequencies between the two populations (P < 0.0001). The frequencies dropped to 3.31% (SEM = 1.42%) and 1.65% (SEM = 0.82%) for Ins12–20– and Ins13–21-specific cells, respectively, at 12 weeks of age, with no statistical difference of representation of either population (P = 0.352). These results indicated the likely importance of the P9 switch for early pathology but not for progression. This conclusion was reinforced by examining the anti-insulin response in overtly diabetic mice and seeing the same disappearance of the differential between the two populations with no statistical differences in both the spleen (P = 0.68) and the PLNs (P = 0.15) (Fig. 2D).
Single-cell expression profiling of insulin-reactive CD4+ T cells reveals organ residence signature and key steps of pathogenesis
Having appropriate reagents, anti-insulin register-specific T cells were sorted by cytometry as single cells in 96-well plates using a dump negative (CD8−B220−CD11b)−CD3+CD4+tetramer+ gate. Each cell was examined for the expression of a set of 96 genes surveying surface receptors, cytokines, signaling molecules, chemokines and chemokine receptors, transcription factors, and cytokine response genes (25). Profiles were established from individual 6- and 12-week-old mice and based on the analysis of 1580 individual cells, of which 501 were from the islets, 729 were from the PLNs, and 350 from the spleen; 967 cells were from 6-week-old mice (n = 11) and 613 cells were from 12-week-old mice (n = 4). With or without glyceraldehyde-3-phosphate dehydrogenase (GAPDH) normalization, unsupervised clustering could define six subsets. However, as illustrated in Fig. 3A, 1060 cells, besides CD3, CD4, and housekeeping genes, expressed less than 10 of the 96 chosen genes at detectable levels and needed to be removed from further statistical analysis to evaluate the differential expression between cells expressing larger numbers of genes. Accordingly, 87.2% of cells from the spleen, 66.4% from the PLNs, and 54% from the islet were removed from the unsupervised cluster analysis (Fig. 3B). In the spleen and PLNs, these numbers correlated with the recently described phenotype of functional anergy defined as FR4+CD73+ in the CD44+ population (fig. S1) (26). After this curation, the unsupervised clustering revealed four different groups (Fig. 3, C and D, and fig. S2). This cluster definition was not influenced by age (6 versus 12 weeks) or the reactivity of the cell (Ins12–20 versus Ins13–21). Cells from cluster 1 were represented in all three tissues and, besides CD3 and CD4, expressed genes associated with cell adhesion and sustained activation of T cells, CD86, ICAM-1, CD80, CD40 (27, 28), and two transcription factors (Zeb2 and PPARγ) involved in terminal differentiation and effector function of T cells (29). This cluster of cells was the only one that increased significantly between 6 and 12 weeks. Cluster 2 was overrepresented in the PLNs and notable by absent features: no signs of TCR- or cytokine-mediated signaling and no markers associated with the major CD4+ subsets. CCR7, a chemokine receptor strongly linked to lymph node homing, was the highest expressed gene in this cluster (Table 1). Clusters 3 and 4 identified the cells resident in islets and a very small percentage of PLN cells (Fig. 3). Cluster 3 cells were prototypic TH1 cells with high expression of TH1 cytokines, such as interferon-γ (IFN-γ), tumor necrosis factor–α (TNF-α), and interleukin-21 (IL-21); TCR-triggered signaling molecules, such as CD3, Zap-70, and Fyn; and coactivator genes, including CD28 and inducible T-cell costimulator (ICOS). The expression of IFN-γ and TNF-α was confirmed by flow cytometry for the whole intra-islet CD4+ T cell population, and very few cells were found to be double expressors (fig. S3). Cluster 4 was more intriguing, with high expression of typical Treg-associated genes, forkhead box P3 (Foxp3), IL2Rα, and IL-10, as well as persistent expression of T-bet and, in some cells, IFN-γ and TNF-α (Fig. 3E). The presence of Foxp3+CD25+ cells was detected by flow cytometry at increased frequency in infiltrated islets compared with peripheral lymphoid tissues for both Ins12–20– and Ins13–21-specific cells (fig. S4). The overlap of highly expressed genes between clusters 3 and 4 included programmed cell death protein 1 (PD-1), cytotoxic T lymphocyte–associated protein 4 (CTLA-4), ICOS, and the TCR signaling molecules Zap70, Fyn, and Nur77 (Table 1 and Fig. 3E); we interpreted this overlap as a sign of the possible transition between these two subsets, indicative of the plasticity of T cells. In any instance, the near-equal representation of the two clusters, 45.6% of cells in cluster 3 and 34.5% in cluster 4 (Fig. 3D), in the entire anti-insulin T cell population regardless of register usage indicated that antigen alone was not the determining factor for effector function. We conclude from the single-cell gene analysis that the intra-islet events determine much of the final fate of the T cells, regardless of their specificity.
TCR sequencing of insulin-reactive CD4+ T cells: Confirmation of the P9 switch model and islet pauciclonal expansion
Single-cell TCR sequencing of tetramer-sorted T cells was carried out using the same reaction used for RNA expression profiling (25). Briefly, after a preamplification step, a second round of polymerase chain reaction (PCR) was used to amplify and bar code each cell with a common tag on the Cα and Cβ primers. The final library was produced using a nested PCR approach with internal Cα and Cβ primers and analyzed on a MiSeq next-generation sequencer (25). The efficiency of sequencing of TCR chains was highly dependent on the state of activation of the cell as measured on our 96-gene set; chain pairing was successful in ~70% of cases, whereas in the remaining 30%, only one of the two TCR chains could be sequenced. In the absence of up-regulation of at least 50% of the 96 genes that we evaluated in our panel, TCR sequencing efficiency was always low (<20%). Consequently, very few sequences were recovered from the spleen, and fewer from the PLNs than from the islets. The first information that came from this analysis is that both Ins12–20 and Ins13–21 cells in the PLNs and islets expressed diverse T cell receptor alpha variable region (TRAV) and T cell receptor beta variable region (TRBV) segments with slightly more diversity in the islets (Fig. 4). This diversity came with some bias and an overrepresentation of TRAV6.5, TRAV6.6, TRAV5D4, TRBV2, TRBV5, and TRBV13.1 for Ins12–20 cells in both locations, whereas TRBV1, TRBV5, and TRBV13.2 were overrepresented for Ins13–21 cells in the islets. The profiles of Ins12–20 V segment usage, although very similar between the islets and PLNs, did not correspond to matching TCR pairs in the two locations; similarly, no identical TCRs could be found between mice. A pairwise analysis (Fig. 4B) revealed some preferential Vα/Vβ pairing for Ins12–20 cells, but all of these pairs corresponded to clones. As we defined the occurrence of clonal expansion by the sequencing of the same TCR pair in three or more cells in the same location, clones were only detected in the islets and never in the PLNs, and most were found at 6 weeks of age and were specific for Ins12–20 (Figs. 4B and 5A). When cluster identity was examined for each clone, although most of them belonged to either cluster 3 or 4, some were split between clusters (Fig. 5A). These observations strongly supported the hypothesis that antigen recognition and expansion took place in the islets and not in the lymph node and demonstrated that the functional fate, TH1 versus Treg, was not exclusively dependent on TCR usage. To support the first conclusion, we stimulated Ins12–20 and Ins13–21 T cell hybridomas with fresh islet cells and PLN cells in the absence of exogenous peptides. As shown in Fig. 5B, only islet cells activated Ins-specific clones, whereas PLN cells, even in much larger numbers, did not (Fig. 5B).
Examination of the CDR3 regions of TCRα and TCRβ isolated from islets and PLNs revealed the same length (~14.0 ± 1.0 amino acids) and an increase in frequency of negatively charged residues at position +3 of CDR3β of Ins12–20 T cells, suggesting the existence of a P9 switch (Fig. 6A). In most cases, this D/E residue was encoded by a non-germline sequence, indicating the role of the antigen recognition in the selection and expansion of these populations (table S2). To confirm the functionality of this switch, three TCRαβ pairs isolated from islet cells and retaining a negative charge in CDR3β were cloned and re-expressed in BW5147 TCR-negative cells using the lentiviral vector pMIG2 (30), and used for functional testing on plastic-bound pMHC molecules. Beyond confirming specificity for Ins12–20, the titration of ligand demonstrated that the functional avidity of the three TCR pairs varied substantially with stimulatory index 50 (SI50) corresponding to between 60 and 200 ng of ligand per well (Fig. 6B). Exposure of these clones to an I-Ag7 Ins12–20 complex mutated at position 9 of the peptide for a glutamic acid (Ins12–20 P9E) resulted in the complete absence of stimulation (Fig. 6B). The same result was obtained for 2D10 and 8F10 T cells, two T cells isolated from immunized mice, and whose SI50 values are 5 and 50 ng of ligand per well, respectively (Fig. 6C). The reciprocal experiment was performed by mutating the glutamic acid residue 114 in the CDR3β of Cl2 TCR and testing both wild-type (WT) and mutant TCRs on the I-Ag7 Ins12–20 complex (Fig. 6D); the mutant Cl2 T cells were unresponsive. Last, the same experiment was performed using recombinant 2D10 TCR and its CDR3β116D➔G mutant using plasmon resonance for affinity measurements (Fig. 6E). Both WT and mutant TCRs were captured at the surface of a lipid bilayer by their C-terminal histidine tails using a 1,2-dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl)imidodiacetic acid)succinyl] (DOGS) lipid and tested for binding to I-Ag7 Ins12–20. Whereas the WT 2D10 TCR bound with measurable affinity (Kass = 1.9 × 104 M s−1, Kdiss = 0.01 s−1, and Kd = 4.44 × 10−7), the point mutation resulted in an almost complete loss of measurable binding (association rate was not quantifiable), confirming the functionality of the switch in this TCR.
Genetic demonstration of the role of the switch in insulin recognition
The importance of mutations at positions β56 and β57 of I-Ag7 in the onset of insulitis and diabetes was tested 25 years ago by transgenesis in the NOD mouse using Aβk and Aβd transgenes (31–34). The more focused targeting of position I-Ag7β57S➔D was also tested by transgenesis and resulted in a marked reduction in diabetes but a persistence of peri-insulitis (35, 36). These transgenic mouse models were very imperfect, mainly because they all still carried a normal I-Ag7 molecule. We showed that the simple dilution of a WT IAg7 onto a NOD-Ab−/− background (NOD-Ab−/− × NOD F1 mouse) was sufficient to prevent diabetes, establishing that the quantity of MHC and/or antigen displayed by MHC on APCs controlled disease onset (fig. S5). Because of these limitations, none of the transgenic experiments could address the mechanisms by which position β57 altered T cell responses and progression toward autoimmunity. To access this information directly, the serine β57 of I-Ag7 was mutated to an aspartic acid using zinc-finger nuclease (ZFN) technology directly in NOD mice (37). Before examination, the new mouse line was backcrossed twice into the WT NOD background to eliminate any potential off-target effect of ZFN cleavage and repair. Neither males nor females of this mutant mouse line developed diabetes, and no periinsulitis could be observed in mice, even after 20 weeks. Mutant I-Ag7S57➔D tetramers were recombinantly expressed for both insulin peptides and used to quantify anti-insulin CD4+ T cells in the mutant mouse. As shown in Fig. 7, the numbers of Ins12–20 and Ins13–21 T cells were comparable in the thymus, spleen, and PLNs of 8-week-old female mice; cell numbers could not be obtained from islets because so few CD4+ T cells could be isolated. Although tetramer+ cells were in almost identical numbers in the spleen of the mutant mouse as in WT mice, numbers in PLNs were three times lower in the β57 mutant animal (Figs. 1 and 7). The TCRs of the rare insulin-specific cells isolated from the PLNs were sequenced and compared with WT PLN TCR sequences (Fig. 7B). The disappearance of D/E residues in CDR3β of Ins12–20-specific T cells from mutant animals as compared with WT mice was highly significant (13.3% versus 28.2%; P < 0.001), confirming the direct role of serine β57 in the selection of this particular TCR repertoire. The influence of β57 was also shown functionally by comparing the presentation of HEL and insulin by APCs from WT or mutant mice. As shown in Fig. 7C, the presentation of HEL was diminished when the antigen was presented to the anti-HEL TCR 21.30 by mutant as compared with presenting cells from regular NOD mice, confirming our previous studies (9), although the recognition of the Ins9–23 peptide was completely abrogated for all Ins12–20-specific T cells that we tested and of which clone 9 is a representative. The presentation of Ins9–23 peptide to the prototypic anti-Ins13–21 11T3 T cell hybridoma was nearly identical between NOD and β57 mutant mice (fig. S6).
The development of reagents that could selectively detect the two main adjacent registers of the Ins9–23 peptide allowed us to dissect some fundamental aspects of the biology of I-Ag7 molecules and their impact in the pathogeny of T1D. Beyond showing the importance of peptide registers in the biology of I-A and HLA-DQ molecules (15, 19, 38), these new tetramers confirmed the existence of two adjacent registers in the B9–23 segment of insulin (16, 39).
The interrogation of the biology of autoreactive T cells against a nominal antigen using single-cell analysis and a set of selected T cell–related genes revealed some important aspects of autoimmunity. Foremost, the fact that a majority of anti-insulin T cells were expressing very low levels of any of the 96 genes that we selected as indicators of T cell functions, in the spleen, lymph nodes, and even islets, seems to correlate well with the strong state of peripheral anergy toward insulin that is observed in most mouse genetic backgrounds, including the NOD mouse. This “anergy” was also correlated phenotypically with the expression of CD73 and FR4 in the CD44+ population (26). Functionally, this overall concept of “high degree of tolerance toward insulin” is confirmed by the outcome of immunization of mice with B9–23 peptide or insulin; we and others have shown that insulin administration rarely produces measurable CD4+ T cell expansion, regardless of the adjuvant used, although it induces cytokine secretion measurable by enzyme-linked immunospot (23, 40). Recent studies indicate that there is a continuous and low level of presentation of insulin peptides in all peripheral lymph nodes including the pancreatic nodes. The peptides are derived from crinosomes released into the circulation after a glucose challenge (17). This low-level and persistent presentation in peripheral nodes induced in the 8F10 T cells that recognize Ins12–20 represents weak activation but not anergy. Ablation of lymph nodes results in complete absence of diabetes and T cell priming (41). This response at the periphery will likely vary among T cells depending on their TCR specificities and affinities. Regardless, the islet presentation will profoundly influence the response based on its display of a number of regulatory molecules such as PD-1. Therefore, it seems logical that tolerance to insulin will have a higher chance of being broken in the environment of the islet where all forms of insulin and its by-products and abortive products are presented. Studies in TCR transgenic mice have supported this hypothesis (39), and our gene expression and T cell repertoire analyses of insulin-specific cells strongly supported the same conclusion. What allows the breakage of tolerance in the islets? Is it linked to antigen processing and presentation? The effects of the dilution of I-Ag7 in MHC knockout NOD mice demonstrate that this is likely the case. However, besides the existence of a threshold for the quantities of particular pMHC complexes needed to activate autoreactive cells, the classical model of T cell activation tells us that effective costimulation as well as a milieu providing cytokines and growth factors (42) will be necessary. In an inflammatory context, it appears that the islet-resident macrophage is the most likely cell responsible for self-antigen presentation. Not only do they capture immunogenic peptides from the crinophagic granules upon insulin stimulation (17), they also express a very efficient antigen processing and presentation machinery (43), and their elimination results in the prevention of disease (44). Of course, it cannot be completely ruled out that in the context of the islet, β cells themselves contribute to T cell stimulation because they have been shown to express MHC class I and class II upon TNF-α/IFN-γ stimulation (45), two cytokines that are found in the lesions. A scenario involving antigen presentation by two different cell types might explain one of the most unexpected findings of our study: the ability of T cell clones to differentiate either as TH1 or as Tregs (Fig. 5A). For instance, the engagement of PD-1 by the PD-1 ligand expressed by β cells might favor Treg over TH1 differentiation (46). In addition, the ability for a single T cell to receive signals from two presenting cells might also explain the persistence of TH1-associated genes such as T-box transcription factor X21 (Tbx21) and IFN-γ in some Tregs (47, 48). It is also important to contrast findings in the islets and the PLNs, a location where we never found clonal expansion and in which Tregs did not express Tbx21 in addition to Foxp3, yielding only classical Tregs. These differences suggest that Tregs received additional differentiation signals in the islets.
The second most important conclusion of our study is the demonstration of the critical role that the P9 switch mode of recognition plays in the pathogeny of T1D, as illustrated with the Ins12–20 peptide. Building on the prior finding that the P5 position of the Ins12–20 peptide has been shown to be essential for diabetes development (49), we show here that the P9 switch is essential for the function of Ins12–20-specific T cells and that at the population level, a D/E residue in CDR3β is selected and necessary for T cell recognition. Examination of infiltrating CD4+ T cells at 6 weeks of age when insulitis is minimal also demonstrated the preferential expansion of Ins12–20-specific T cells at the early preclinical phase of disease, indicating the proliferative advantage that the P9 switch provides to T cells. It is quite remarkable to see that more than 50% of the CD4+ islet infiltrate is made of these cells at 6 weeks. It will be of upmost interest to determine whether the remaining cells also use a switch for recognition and expansion and whether peptides are provided by other fragments of insulin. Particular attention will be given to the C-peptide, a major source of antigen in humans (50) and mice (51, 52); almost all humans C-peptide–derived epitopes (50), as well as the immunodominant murine C-peptide determinant (51), have no negative charge at P9. In humans, the recognition of these peptides by TCRs fulfilling the canons of a P9 switch mode of T cell recognition with D/E amino acids in their CDR3β has already been reported (53). In mice, similar observations were made for noninsulin epitopes devoid of D/E P9 residues (9).
Last, the kinetics of the P9 switch–driven anti-insulin response in mice with its early burst and rapid disappearance have important translational implications. If they also exist in humans, it is only in at-risk patients that these cells would be detectable; their capture from peripheral blood could constitute the earliest possible marker of islet autoimmunity and change the timing of therapeutic intervention.
Our observations regarding the potential role of the P9 switch mode of T cell recognition in the onset of diabetes in the NOD mouse were challenged by the examination of the mutant β57S➔D mouse. Our prediction was that the disappearance of the positively charged surface at P9 would eliminate T cells capable of using the P9 switch. As a result, the Ins12–20 peptide should select a normal T cell repertoire without D/E residues in the N-terminal segment of the CDR3β. Although this expectation was proven correct by direct sequencing of the anti-Ins12–20 TCRs, the enumeration of anti-insulin cells showed their marked reduction in PLNs as compared with WT NOD mice and their near-complete elimination in the islet. This latest point established the direct role of P9 switch cells in the initiation of infiltration and the inability of cells without switch to substitute for them. Therefore, the binding advantage that we have documented for P9 switch cells appears necessary and sufficient to break tolerance in the context of islet antigen presentation. This simple relationship between position β57 of diabetogenic MHC molecules and the selection of a unique T cell repertoire appears to be the molecular mechanism that links MHC and T1D. Further studies will attempt to confirm this conclusion in humans. This task might prove challenging because although the switch plays an essential role in initiation, it appears not necessary for continuation of disease. However, the examination of at-risk and just-diagnosed patients should allow us to overcome this difficulty. The initial analysis of anti–C-peptide CD4+ T cells that showed the presence of D/E residues in CDR3β makes us optimistic (53). It is likely that antibodies or small molecules capable of interfering with the position β57 of HLA-DQ8, if applied early enough in the development of the autoimmune process, would be capable of preventing progression and onset of disease.
MATERIALS AND METHODS
The study was intended to examine the very early phase of T1D in mice and dissect the kinetics and characteristics of the anti-insulin response. We took advantage of new tetramer reagents we developed to follow anti-insulin CD4+ T cells in mice and performed single-cell analysis. The deep examination of the T cells specific for the Ins12–20 fragment was aimed at testing the relevance of the P9 switch model of T cell recognition in disease onset and progression.
NOD/LtJ mice were purchased from the Jackson laboratory and housed under pathogen-free conditions. NOD/LtJ mice in which MHC class II genes were deleted were also from the Jackson laboratory [NODCaj.129S2(B6)-H2-Ab1tm1Doi/LwnJ] but were bred in-house for the past decade and genotyped in the laboratory. The mutant NOD/LtJ mouse strain in which position β57 of I-Ag7 was mutated from S➔D by ZFN technology (NOD/ShiLtJ-H2-Ab1em2Ygch) was produced by C.M. and Y.-G.C. (37) and was bred and genotyped in our facility. Care and handling of mice followed Institutional Animal Care and Use Committee rules. Glycemia of all animals was measured weekly from a drop of blood collected from the tip of the tail.
Cells and transfection
BW5147 (TCRα−β−) cells and T cell hybridomas were maintained in Dulbecco’s modified Eagle’s medium–10% fetal bovine serum (FBS) medium. For reexpression of TCR in BW5147 after sequencing, Vα and Vβ segments were synthesized and cloned into pMIG II (Addgene) and spinfected with polybrene for 2 hours at room temperature at 1000g (30). After recovery, cells expressing TCR were sorted by staining with the anti–TCRβ chain antibody H57-597 (BD Biosciences).
Construction, expression, and purification of I-Ag7/peptide tetramers and recombinant TCR molecules
The generation of I-Ag7/peptide complexes has previously been reported in detail (54). The peptide sequences of I-Ag7 HEL11–27, I-Ag7 Ins12–20, and I-Ag7 Ins13–21 molecules are AMKRHGLDNYRGYSLGN, VEALYLVCG, and EALYLVCGE, respectively. The biotinylation sequence #85 from Schatz (55) follows the acidic zipper on the α chain. Biotinylation was performed in vivo by cotransfecting an inducible BirA plasmid and adding 50 μM biotin at the time of induction. After purification by ion exchange and size exclusion, biotinylated molecules were tetramerized at 4°C overnight using phycoerythrin (PE)–labeled streptavidin (BioSource International) or PE–cyanine 7– or APC-labeled streptavidin (eBioscience) using a 5:1 molar ratio of biotinylated molecules to labeled streptavidin. For surface plasmon resonance (SPR) measurements on lipid bilayers, the same pMHC molecules were produced with a C-terminal strep-tag II sequence and purified on Strep-Tactin columns (IBA Lifesciences).
TCR complementary DNA for the α and β chains of insulin-reactive T cell hybridomas and primary cells were subcloned into the fly expression vector pRMHa3 or pMT/Bip/V5-His (Invitrogen) and sequenced. The final constructs code for the VαCα and the VβCβ chains, respectively, followed by a linker sequence (SSADL), a thrombin site (LVPRGS), a leucine zipper (acidic for the α chain and basic for the β chain), and a hexahistidine tag. Vectors were cotransfected into SC2 cells along with a vector encoding a puromycin-resistance gene, and stable cell lines were established. Soluble TCRs were purified from culture supernatants, as previously described (56).
Synthetic peptides HEL11–27 (AMKRHGLDNYRGYSLGN), Ins9–23 (SHLVEALYLVCGERG), and forms of Ins12–20 (TEGVEALYLVCGGGS) and Ins13–21 (TEGEALYLVCGEGGS) in which the N and C termini were extended by TEG- and GGS- sequences, respectively (39), were synthesized by AnaSpec Inc. and were ≥95% pure.
PCR site-directed mutagenesis was carried out using the QuikChange II PfuUltra High-Fidelity DNA Polymerase (Agilent) and verified by DNA sequencing.
Islet isolation and cell preparation
Islets were isolated using a modified protocol derived from Li et al. (57). Briefly, pancreata were isolated and incubated for 60 min at 37°C after injection of collagenase P through the common bile duct. After several washes, islets were isolated on a metal strainer and hand-picked under a dissection microscope. An average of 200 to 300 islets were obtained per mouse. For making single-cell suspension, islets were digested with trypsin or dispase at 37°C and washed extensively before being counted.
Cell preparation, cell staining, and flow cytometry analysis
Non-islet single-cell suspensions were prepared by mechanical disruption of the corresponding organs in Hanks’ balanced salt solution buffer, and erythrocytes were removed by osmotic lysis. Cells were washed with fluorescence-activated cell sorting (FACS) buffer [phosphate-buffered saline (PBS) containing 3% FBS and 0.05% NaN3] and incubated with avidin (0.5 mg/ml; Sigma-Aldrich) and Fc block (BD Biosciences) in FACS buffer for 1 hour at room temperature. Cells were then washed once and stained with PE-labeled or PE-Cyanine 7–labeled MHC/peptide tetramers at a final concentration of 10 μg/ml in FACS buffer for 1 hour at room temperature. For costaining of surface markers, fluorescein isothiocyanate–anti-CD4 (clone RM4-4, BD Biosciences), allophycocyanin–anti-B220 (clone RA3-6B2, eBioscience), and allophycocyanin–anti-CD8 (clone 53-6.7, eBioscience) were used. Exclusion of dead cells was performed by propidium iodide labeling (Invitrogen). Flow cytometry was performed using a FACS LSR-II instrument (Becton Dickinson) or a MACSQuant analyzer (Miltenyi Biotec), and data were analyzed using FlowJo software (TreeStar Inc.). Staining of T cell hybridomas with MHC tetramers was carried out in the same way.
CD4+ T cell responses
For plate-bound activation assays, pMHCs were coated overnight in PBS at 4°C at various concentrations in flat-bottom 96-well plates and washed twice in PBS before plating 4 × 104 T cell hybridomas per well. After 24 hours of incubation at 37°C, supernatants were harvested and assayed for IL-2 content using a radioactive IL-2–dependent natural killer cell line bioassay. For assays with live APCs, T cell hybridomas were incubated with 1 × 105 APCs per well and increasing concentration of peptides. All assays were performed using biological triplicates and performed in at least three independent experiments.
For single-cell analysis, T cells were sorted with pMHC tetramers on a FACSAria III machine (BD Biosciences) directly into reverse transcriptase (RT) buffer (10 μl per well). Protocols for preamplification and amplification for gene expression analysis and TCR sequencing have been detailed in recent publications (25). Gene expression analysis was performed on a Biomark unit with 96 × 96 Dynamic Array IFCs, whereas sequencing was performed on an Access Array instrument using 48 × 48 Library Prep IFCs (Fluidigm). Gene expression data quality control was performed using the Fluidigm software package and manual inspection of each melting curve to remove cells in which primer dimers were interfering with signal. After curation, data were analyzed with a homemade R package (see below). DNA sequencing was performed on a MiSeq instrument (Illumina). After stitching, TCR sequences were confronted to the IMGT database to determine TRAV and TRBV assignment (58). The following list of TRAV (x axis) and TRBV (y axis) segments appears in Fig. 4: TRAV2, TRAV4-2, TRAV5-1, TRAV5D-4, TRAV6-1, TRAV6-2, TRAV6-3, TRAV6-5, TRAV6-6, TRAV6D-4, TRAV6D-7, TRAV7-2, TRAV7D-2, TRAV7D-4, TRAV7N-4, TRAV8-1, TRAV8D-2, TRAV9-1, TRAV9-3, TRAV9D-3, TRAV9D-4, TRAV10, TRAV10D, TRAV11, TRAV12-1, TRAV12-2, TRAV14D-2, TRAV14D-3/DV8, TRAV16, TRAV16D/DV11, TRAV21/DV12, and TRBV1 TRBV2 TRBV5 TRBV12-2 TRBV13-1 TRBV13-2 TRBV13-3 TRBV14 TRBV15 TRBV16 TRBV17 TRBV19 TRBV20 TRBV24 TRBV29 TRBV31.3.
Surface plasmon resonance
A Biacore T200 instrument was used to determine interactions between purified MHC/peptide complexes and TCR molecules. TCR was immobilized on lipid bilayers coated on L1 chips by capture of their histidine tag. Lipid bilayers were produced by fusion of liposomes on the hydrophobic surface. Phosphatidylcholine (PC), cholesterol (C), dioleoyl phosphatidylcholine (DOPC), l-α-PE, l-α-phosphatidylglycerol (PG), and DOGS-NTA were all from Avanti Polar Lipids, and liposome composition was 60:20:5:5:10 (PC:C:DOPC:PE:PG:DOGS-NTA) weight ratio. Liposomes were produced by extrusion at 1 mg/ml in PBS on 400-nm polycarbonate disks (59). Strep-tag–modified pMHC molecules were injected over the surface after stabilization of baseline in filtered and degassed PBS at a flow rate of 20 to 30 μl/min. To avoid regeneration of the surface between injections, kinetics studies were performed using the single cycle mode using an irrelevant TCR in the subtraction flow cell. Kd values, as well as on and off rates, were obtained by nonlinear curve fitting of subtracted curves using the 1:1 Langmuir binding model using the BIAevaluation program (version 3.0.2).
Fisher’s exact test and Student’s t test (two-tailed) were used for statistical analyses.
R analysis methods
Single-cell expression data, in the form of quantitative PCR (qPCR) Ct values from the Fluidigm Biomark HD system, were analyzed in the R programming environment according to recommendations from the Fluidigm SINGuLAR Analysis Toolset documentation. Log2 expression values were calculated using the default limit of detection Ct value of 24, and negative log2 expression values were set to 0. Log2 expression values were then normalized per individual cell by dividing each cell’s gene expression values by its Gapdh expression value. These normalized values were median-centered for further analysis. The expression of 96 genes was quantified in 1746 individual cells. Cells were removed from the analysis if no expression of any gene was detected in the isolate (111 cells removed) or if Gapdh expression was markedly low (55 cells removed). Cells with relatively low total expression were filtered by clustering the cells and removing clusters with low cumulative normalized log2 expression values (1131 cells removed). Differential gene expression was calculated using the nonparametric Kruskal-Wallis test by ranks, due to the multimodal distributions of most genes. P values were corrected by the Benjamini-Hochberg procedure.
The “Rtsne” R package was used for t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction. Violin, table, and t-SNE plots were created using the “ggplot2” R package (60). Heatmaps are based on the heatmap.2 function from the “gplots” R package. The columns of the heatmaps are ordered from lowest to highest differential expression P values, and the rows are ordered by hierarchical clustering based on Euclidean distance.
Fig. S1. Examination of the “anergic state” of total and Ins-tetramer+CD4+ T cells in the spleen and PLNs based on FR4+/CD73+ surface labeling of the CD44− and CD44++ populations.
Fig. S2. Violin plot representation of the top most up-regulated and differentially expressed genes in clusters 1 to 4.
Fig. S3. Direct IFN-γ and TNF-α staining in islet-infiltrating CD4+ T cells in three 9-week-old NOD mice.
Fig. S4. Direct detection of CD25+Foxp3+ cells in the spleen, PLNs, and islets of 10-week-old female NOD mice.
Fig. S5. Diabetes incidence in NOD and heterozygote NOD × MHC class II−/− female mice.
Fig. S6. Response of the Ins13–21-specific T cell hybridoma 11T3 to the Ins9–23 peptide presented by NOD and NODβ57D splenocytes.
Table S1. Summary of the various peptide constructs that were used to produce recombinant I-Ag7/peptide complexes in fly cells.
Table S2. Comparison of all CDR3β analyzed for the present study to their genomic segments.
Table S3. Excel spreadsheet with raw data.
Acknowledgments: We acknowledge the outstanding support we received throughout the performance of the current study from personnel associated with the flow cytometry and DNA sequencing cores at Scripps Research and the Scripps Center for Computational Biology’s Bioinformatics Core. We also thank the staff of the NCBI Gene Expression Omnibus center who helped us make our data publicly available. Funding: Funding was provided by the NIH Clinical and Translational Science Award issued to the Scripps Translational Science Institute UL1TR002550 and TL1TR002551 to S.S. and L.G., KL2TR001112 to B.A., and NIH grants RO1DK058177 to E.U., R01DK099317 to M.N., RO1DK097605 and DK107541 to Y.-G.C., P01 AI042288 to C.M., and 1R01DK117138 to L.T. B.A. was a resident of the Scripps Clinic and Scripps Green Hospital Internal Medicine Residency ABIM Research pathway during the course of the study. Author contributions: L.G., M.H., A.C., and S.S. designed and performed all experiments included in the manuscript and worked as a team. B.A. performed some of the early experiments that initiated the study. L.K. was in charge of the animal colony, genotyping, and production of tetramers. M.N. provided critical DNA and cellular reagents. X.W. validated essential reagents independently of us. A.S. supervised the bioinformatics component of the single-cell analysis. C.M. and Y.-G.C. produced the mutant mouse used in the last set of experiments. E.U. was critical in data evaluation and manuscript editing. L.T. oversaw and coordinated the entire study and wrote the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The single-cell qPCR gene expression data and the single-cell TCRα and TCRβ sequencing data are available from the Gene Expression Omnibus under accession number GSE134885. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. The new tetramer constructs for I-Ag7 Ins12–20 and I-Ag7 Ins13–21 described in this paper are available upon request through the NIH Tetramer Facility.
- Copyright © 2019 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works