Research ArticleAUTOIMMUNITY

How C-terminal additions to insulin B-chain fragments create superagonists for T cells in mouse and human type 1 diabetes

See allHide authors and affiliations

Science Immunology  05 Apr 2019:
Vol. 4, Issue 34, eaav7517
DOI: 10.1126/sciimmunol.aav7517

Superagonist diabetogenic peptides

Type 1 diabetes is an autoimmune disease triggered, in part, by activation of CD4+ T cells specific for insulin-derived autoantigens. Wang et al. prepared ternary complexes of diabetogenic T cell receptors attached to ligands formed by binding of insulin B-chain peptides to mouse or human MHC class II proteins known to be associated with genetic susceptibility to diabetes. Structural and functional analyses revealed that fusion peptides featuring covalent linkage of the C terminus of an insulin B-chain peptide to specific fragments of insulin C-peptide were highly potent ligands, enabling avid binding of both mouse and human TCRs. These findings point to transpeptidation reactions in lysosomes as a potential source of superagonist chimeric insulin peptides with an important driver role in autoimmune diabetes.


In type 1 diabetes (T1D), proinsulin is a major autoantigen and the insulin B:9-23 peptide contains epitopes for CD4+ T cells in both mice and humans. This peptide requires carboxyl-terminal mutations for uniform binding in the proper position within the mouse IAg7 or human DQ8 major histocompatibility complex (MHC) class II (MHCII) peptide grooves and for strong CD4+ T cell stimulation. Here, we present crystal structures showing how these mutations control CD4+ T cell receptor (TCR) binding to these MHCII-peptide complexes. Our data reveal stricking similarities between mouse and human CD4+ TCRs in their interactions with these ligands. We also show how fusions between fragments of B:9-23 and of proinsulin C-peptide create chimeric peptides with activities as strong or stronger than the mutated insulin peptides. We propose transpeptidation in the lysosome as a mechanism that could accomplish these fusions in vivo, similar to the creation of fused peptide epitopes for MHCI presentation shown to occur by transpeptidation in the proteasome. Were this mechanism limited to the pancreas and absent in the thymus, it could provide an explanation for how diabetogenic T cells escape negative selection during development but find their modified target antigens in the pancreas to cause T1D.


Insulin is a major target in type 1 diabetes (T1D) in humans and rodents (1). During the past several decades, many CD4+ T cell clones have been isolated from nonobese diabetic (NOD) mice responsive to an epitope(s) in the B:9-23 insulin peptide. There has been disagreement about how these T cells target this peptide. In particular, controversial has been the position or “register” this peptide takes in the groove of the NOD IAg7 major histocompatibility complex (MHC) class II molecule (MHCII) for presentation to these T cells. Some have suggested registers 1 and 2 (R1 and R2) that put B:12-20 and B:13-21, respectively, in the p1 to p9 positions of the IAg7-binding groove because these registers provide IAg7-compatible anchor amino acids at p1, p4, p6, and p9 (25). However, our previous data argue strongly that nearly all diabetogenic NOD CD4+ T cells recognize B:9-23 in register 3 (R3), which places B:14-22 (ALYLVCGER) in the p1 to p9 position (69). This register had not been previously considered because the peptide’s basic B:22R at p9 clashes with the IAg7 p9 pocket, which strongly prefers an acidic amino acid.

Our data showed that a substitution of E for R at p9 markedly improved binding of the peptide in R3 converting the weakly stimulatory natural peptide into a very strong agonist for a subset of NOD B:9-23–specific T cells, generally referred to as type A (2, 3). We also determined that, for a second group of T cells, type B (2, 3), an additional substitution of E to G at p8 (B:21), was needed to create the strong agonist for these T cells. Because p8 is usually a surface-exposed amino acid in MHCII-bound peptides, we postulated that the side chain of the p8E somehow interfered with the correct docking of the type B T cell receptors (TCRs) on the R3 IAg7-peptide complex. We have also studied human T cells specific for the B:9-23 peptide presented by HLA (human leukocyte antigen)–DQ8 (911). Like IAg7, the DQ8 β chain has a polymorphism at β57 that creates a preference for an acidic amino acid at p9. The substitution of R to E at B:22 in the peptide also greatly improved the presentation of the insulin peptide by DQ8 to these human T cells, establishing R3 as the relevant register. Most recently, we have published high-resolution crystal structures of these modified peptides bound to IAg7 and DQ8 (9), confirming the R3 position and activity of the peptides.

Here, we show the structures of a mouse type A, a mouse type B, and a human type A–like TCR, bound to their optimal versions of the MHCII-R3 insulin peptide ligands. The structures confirmed the R3 recognition of the T cells and showed that the specificity differences among mouse type A and type B T cells lie in how they deal with the amino acid at p8 (B:21) of the insulin peptide. Despite differences among the mouse and human TCRs in the sequences of their Vα and Vβ domains and their orientations on their ligands, there were some notable common features among the complexes pointing out the similarities in human and mouse both in how these ligands are formed and in how TCRs engage them. The structures also show how the peptide modifications were essential to the formation of the complexes, suggesting a role for modification of the peptide in vivo to initiate the CD4+ T cell response in T1D. On the basis of the recent work showing that fused peptides can form neoepitopes for MHCI and MHCII (1220), we predicted that the process of transpeptidation, a version of reverse proteolysis, could cause internal proteolytic deletions of the appropriate portions of proinsulin between the B chain and C-peptide, generating chimeric peptides similar to our modified ones. We show here that synthetic versions of the predicted fused peptides act as superagonists for a variety of mouse and human insulin-reactive T cells.


Mouse and human CD4+ T cells recognize B:9-23 bound similarly to IAg7 and DQ8 but with different docking modes

Mouse and human CD4+ T cells reactive to epitopes involving the B:9-23 insulin peptide have been reported and characterized in numerous publications (2, 3, 6, 7, 9, 11, 2123). A subset of these T cells are listed in Table 1, along with the sequences of their TCR Vα and Vβ complementary-determining regions (CDR) 1, 2, and 3 (CDR1, CDR2, and CDR3) [this table is an updated version of that presented in a previous publication (9)]. For the structural studies presented here, we used three of these T cells: an NOD mouse type A T cell, I.29 (6, 7); a NOD mouse type B T cell, 8F10 (3, 9); and a human type A–like T cell, T1D3, isolated from a patient with T1D (9, 11). As described in Materials and Methods, we used T cell hybridomas or TCR-transduced T cell avatars for functional studies. For structural studies, we prepared soluble versions of the TCRs, fusing the V regions of the TCRs to the extracellular domains of human Cα and Cβ, which were expressed separately in Escherichia coli inclusion bodies, solubilized, mixed, and refolded to prepare the soluble functional TCRs (24, 25).

Table 1 Properties of the TCRs of the mouse and human B:9-23–specific CD4T cells.
View this table:

Also, as previously described (6, 7, 9), we prepared soluble versions of IAg7 and DQ8 bound to an epitope of insulin B:9-23, which was modified optimally for each of the T cells, as shown schematically in Fig. 1A. Briefly, for all three ligands, the peptide was covalently attached to the N terminus of the MHCII β chain via a flexible linker (26) and carried substitutions at p1 and p9 to provide optimal anchor amino acids for either IAg7 (p1A > R and p9R > E) or DQ8 (p1A > E and p9R > E) in R3. The p1 mutations were used in constructs for soluble MHC-peptide complexes, assuring uniformity and stability of the covalent complexes, but they were not used for soluble peptides used in the in vitro stimulation assays below.

Fig. 1 Mouse and human TCR engagement of IAg7 and DQ8 with an R3-bound insulin peptide.

(A) Schematic representations of optimal versions of the insulin B:10-23 peptide for the I.29 (left, 8E9E6ss with IAg7), 8F10 (middle, 8G9E with IAg7), and T1D3 (right, 8E9E11ss with DQ8). The core positions (p1 to p9) of the peptide in the MHCII-binding groove are numbered. Engineered disulfides are shown. Mutations from the natural sequences are in red. (B) Ribbons representing the three TCRs docked on their optimal MHCII-peptide ligands are shown for I.29 (left), 8F10 (middle), and T1D3 (right), viewed from the C-terminal end of the peptide. Colors are blue (TCR Vα), red (TCR Vβ), cyan (MHC α1), magenta (MHC β1), and yellow (peptide). (C) The footprints of the three TCRs on the solvent accessible surfaces of their MHCII-peptide ligands are shown from above: I.29, IAg7-8E9E6ss (left); 8F10, IAg7-8G9E (middle); and T1D3, DQ8-8E9E11ss (right). The surfaces are colored cyan (MHCIIα), magenta (MHCIIβ), and yellow (peptide). The surface of atoms involved in TCR contact (≤4.5 Å from a TCR atom) are colored in darker shades of the same colors. Also shown are tube representations of the tips of the TCR CDR loops: αCDR1 (green), αCDR2 (cyan), αCDR3 (blue), βCDR1 (yellow), βCDR2 (orange), and βCDR3 (red). For each complex, the arrow represents the angle of engagement of the TCR CDRs, placing αCDR2, βCDR3, and βCDR1 on one side and αCDR1, αCDR3, and βCDR1 on the other.

In addition, for I.29, α62N in IAg7 was changed to C, introducing a disulfide bond to the natural p6C in the peptide when the peptide bound to IAg7 in R3 (6). For 8F10, we changed B:21E to G at p8 to remove the p8E side chain from the surface, which is inhibitory for this and other type B T cells (7, 9). For the human T1D3 T cell, we introduced C at the first amino acid of the linker (p11) and at DQ8 α72 while changing the natural B:19C at p6 to A to form an unambiguous disulfide bond between the C terminus of the peptide and DQ8 α-chain helix. These mutations were previously determined to create optimal stimulating ligands for the three CD4+ T cells (6, 7, 9, 11). We refer to these three peptides below as 8E9E6ss, 8G9E, and 8E9E11ss (Fig. 1A).

Our first attempts to cocrystallize the NOD mouse TCRs bound to their IAg7-peptide ligands produced crystals that contained only the free TCR. The crystal structures of these two free TCRs were solved (Materials and Methods and table S1). They had crystallized with packing arrangements that, in each crystal, blocked the site for IAg7-peptide binding to the TCR. Examination of structures revealed several amino acids that appeared to be important for this packing (fig. S1). Because these amino acids were not on the TCR face predicted to interact with the IAg7-peptide complex, we introduced mutations (Vβ 56 K to A for I.29 and Cβ 202R to A for 8F10) at these positions to disrupt this crystal packing and reexpressed the proteins. In subsequent crystallization trials using these two modified mouse TCRs and the human T1D3 TCR, we obtained cocrystals with their optimal ligands and were able to solve the structures of the ternary complexes (table S1).

As seen in the scores of structures of other TCRs bound to MHC-peptide ligands (27), in our crystals, the TCRs bound to their MHCII-peptide complexes in the familiar diagonal orientation but with differences in the pitches and angles of docking (Fig. 1, B and C). Views of the complexes from the C-terminal ends of the peptides (Fig. 1B) show that the I.29 TCR was pitched toward the IAg7 β-chain helix of the IAg7-8E9E6ss complex, whereas the 8F10 and T1D3 TCRs sat more flatly on their IAg7-8G9E and DQ8-8E9E11ss ligands. Views from above (Fig. 1C) show that the CDRs of the I.29 and T1D3 TCRs were docked at a shallow angle of only about 30° to the peptide backbone, a feature that is at one extreme of the range seen with other TCR-MHCII complexes, whereas the 8F10 TCR sits on the IAg7-8G9E complex at an angle of about 60°, which is more typical of TCR-MHCII complexes (27).

The combination of these differences in pitch and angle led to different footprints among the TCRs on their ligands as measured by the number of atom-to-atom contacts and the buried surface area (BSA) at the interface (Fig. 1C, Table 2, and table S1). The tilt and twist of the I.29 TCR created a footprint on IAg7 dominated by TCR interactions with the MHCII β1 helix at the expense of the α1 helix. The contacts usually seen between the Vβ CDR1 and CDR2 loops with the α1 helix of MHCII were absent, with the CDR3 of Vα now providing nearly all of the contact with this helix. Meanwhile, there was extensive I.29 TCR contact along the entire length of the IAg7 β1 helix, mediated by the Vα CDR1 and CDR2, as well as the CDR3s of Vα and Vβ. In contrast, the 8F10 and T1D3 TCR footprints were much more evenly spread in a conventional way over the MHC helices, with their Vβ CDR1 and CDR2 loops making multiple contacts with the MHC α1 helix and their Vα CDR1 and CDR2 loops doing the same with the IAg7 β1 helix. Thus, overall, these structures show that the TCRs specific for MHCII-insulin bind their ligands in a variety of ways and confirm that the ligands of these three TCRs are indeed the insulin peptides bound to MHCII in R3.

Table 2 I.29, 8F10, and T1D3 TCR contacts with their ligand.

AA, amino acid. BSA, buried surface area, Å2 (% of total BSA).

View this table:

The p8 amino acid determines type A recognition versus type B recognition of the MHCII insulin complexes

In the solved structures, the various interactions of the TCRs with the C-terminal ends of their optimal R3-bound peptides gave a structural explanation for the phenomena associated with the specificities of NOD mouse type A and type B CD4+ T cells. Type A T cells require the natural p8E in the peptide for strong reactivity (7, 9). For the I.29 TCR, interaction with this amino acid was the only TCR contact at this end of the peptide. The unusual angle and tilt of engagement of the I.29 TCR with IAg7-8E9E6ss brought its Vβ2 CDR2 loop into a position such that β51R formed a salt bridge with the carboxylate of the p8E side chain (Fig. 2A, Table 2, and table S2). Our previous work had shown the importance of this contact in that the removal of the p8E side chain by its mutation to G reduced the peptide stimulatory potency by about 30-fold and markedly reduced the ability of the I.29 TCR to bind to the IAg7-8E9E6ss complex (9). Reciprocally, as shown in Fig. 2B, mutation of the TCR β51R to A eliminated the ability of the IAg7-8E9E6ss tetramer to bind to the I.29 T cell. Mutations of many of the other I.29 TCR amino acids to alanine, making the most contact with the ligand in Table 2, also eliminated the binding of the fluorescent IAg7-p8E9Ess tetramer to the I.29 T cell (fig. S2). Many other B:9-23–reactive type A T cells also use Vβ2 (Table 1), but as of now, no type B T cells have been reported to use this Vβ element. This suggests that the unusual rotation of the TCR and the selection of Vβ elements that contain an R in this position of CDR2 may be a common feature of type A T cells.

Fig. 2 TCR interactions with the peptide p8 amino acid determine type A recognition versus type B recognition.

(A) The salt bridge between β51R of the I.29 TCR βCDR2 loop and the p8E of the 8E9E6ss peptide is shown. (B) Binding of a soluble fluorescent tetramer version of the IAg7-8E9E6ss complex to three versions of the 5KC TCR-negative T cell is shown: the untransduced 5KC control (red), 5KC transduced with the unmutated I.29 TCR (blue), and 5KC transduced with the I.29 TCR (green) bearing an R > A mutation at the CDR2 β51 amino acid. The transductants were prescreened for similar surface TCR expression. Additional data from this experiment are shown in fig. S2. (C) Extensive interaction of the 8F10 βCDR3 loop with the backbone of the 8G9E peptide in the region of p7G and p8G, approaching to within 3.1 Å at the p8G backbone nitrogen. (D) Disruption of the 8F10 T cell response to the 8G9E peptide by mutation of the p8G to L or V. Representative of three separate experiments with similar results. More data are shown in Fig. 5, C and D. (E) Extensive interaction is shown between β30R of the T1D3 TCR βCDR1 loop and the 8E9E11ss peptide from p8E to p10G, which includes a salt bridge to p8E, as well as an H-bond to Y60 of the DQ8 β-chain helix. (F) Inhibitory effect of mutating the β30R to A within T1D3 TCR βCDR1 loop on the response of the T cell to K562 cells transduced with the DQ8-8E9E11ss complex. Also shown are the changes in the responses of T1D4 and T1D10 of to the same peptide after mutating CDR2 β50R or CDR1 β30R to A, respectively. Results are the average responses and SEM seen in three separate experiments. Results in the individual experiments are also shown (open circles). Also shown are P values (Student’s two-tailed t test rounded to a single significant place) comparing the responses of the mutated and nonmutated T cells.

Like other type B T cells (7), the response of 8F10 is inhibited by the natural p8E of the peptide, but the response improves markedly when the p8E side chain is removed by mutation to G (7, 9). This finding is explained by the structure, in which the 8F10 Vβ8.2 CDR3 loop was positioned over the C-terminal end of the peptide, in close contact with the peptide backbone at p6C, p7G, and p8G. The proximity of the CDR3 loop to the peptide leaves no room for a surface-exposed side chain at p8 (Fig. 2C). Consequently, leaving the natural p8E at this position (9) or replacing it with L or V (Fig. 2D) eliminated the ability of the peptide to stimulate the 8F10 T cell. The 8F10 TCR contains the well-studied Vβ8.2 element, which could explain its docking angle. Previously, β48Y in the CDR2 loop of Vβ8.2 has been shown repeatedly in TCR structures to favor a docking spot on the MHCII α-chain helix between α57Q and α61Q (25, 28). In the 8F10 structure, β48Y occupies this spot as well (Fig. 3A). Establishing this interaction requires the TCR to take the conventional docking angle seen in the 8F10 complex. Another property of the 8F10 TCR is its second-order binding kinetics to the IAg7-8G9E ligand, consistent with a fast-on/fast-off initial phase, followed by a conformational change that leads to a slower off-rate (9). Examination of the IAg7-8G9E structure before (9) and after 8F10 TCR binding shows large rotational changes in the α57Q and α61Q side chains, aligning these amino acids for interaction with Vβ48Y and the 8F10 CDR1 loop, respectively (Fig. 3A). These changes in the ligand to accommodate the TCR could explain the biphasic kinetics and the docking angle.

Fig. 3 Other features of the 8F10 and T1D3 TCRs.

(A) A ribbon representation of a portion of the IAg7 α1 helix (cyan) with the rotamer changes (arrows) in α57Q and α61Q side chains from their positions before (carbons, gray) and after (carbons, cyan) engagement by the 8F10 TCR. The subsequent interaction of these amino acids with the TCR Vβ8.2 CDR2 48Y (carbon, orange) and CDR1 28N and 29N (carbon, yellow) side chains are also shown with potential H-bonds in green. (B) Left: SPR data [resonance units (RU)] for binding of various concentrations of the soluble T1D3 TCR to IAg7-8E9E11ss immobilized via a biotin tag in a BIAcore streptavidin flow cell. A flow cell containing immobilized IAg7-8G9E was used to correct the data for the fluid phase SPR signal. Standard BIAcore BIAEval software was used to fit the data to a first-order kinetic model and to calculate the kinetic association (kd, 1/M•s) and dissociation (ka, 1/s) rates. The overall equilibrium dissociation constant (KD, μM) was calculated as the kd/ka. Right: Scatchard plot of the equilibrium SPR signal obtained with each concentration of the soluble T1D3 TCR tested. The KD was calculated as the negative slope of a straight line fit to the data. Rmax (predicted RU at infinite TCR concentration) is labeled. Results were shown from a single experiment. Similar results were obtained in a second experiment. t1/2, half-life.

The T1D3 T cell and the other human T cells in Table 1 also have a type A–like phenotype. They respond strongly to DQ8-8E9E11ss, and their responses are markedly reduced by changing the peptide p8E to A (11). The T1D3 TCR binds to DQ8-8E9E11ss with an affinity typical of CD4+ T cells (Fig. 3B) but has a docking angle like the I.29 TCR (Fig. 1C). However, the T1D3 TCR does not have an R in its βCDR2 (Table 1). Its dependence on p8E is due to β30R in its CDR1, which, because of the conventional flat docking of the T1D3 TCR, was brought within range of p8E to form a salt bridge (Fig. 2E). Mutation of this R to an A reduces the response to DQ8-8E9E11ss about 10-fold (Fig. 2F). Another human T cell specific for this insulin peptide bound to DQ8, T1D10, bears hVβ7.1, which has a CDR1 R similarly positioned to that in T1D3 (Table 1). Mutation of this R to an A results in a modest reduction in T1D10’s response to the DQ8-presented 8E9E11ss peptide (Fig. 2F), so the two TCRs may engage the DQ8-peptide complex similarly. Last, the human T1D4 T cell uses the hVβ8.3 element, which has an R in its CDR2 at β50 similarly positioned to the β51R of mouse Vβ2 (Table 1). Mutation of this R to an A eliminated the T1D4 response to DQ8 presented 8E9E11ss peptide (Fig. 2F), predicting that this TCR would use a docking angle on DQ8 similar to that of the I.29 TCR on IAg7.

These data confirm that the NOD and human T cells in Table 1 recognize the insulin peptide bound in R3, not in R2 and R1 as previously suggested (2, 3). Consequently, the differences among these T cells in the recognition of the insulin peptide are much subtler than previously proposed and are confined to how the TCR Vβ domain is oriented and how it interacts with the amino acid at p8 in the R3-bound peptide. The natural E at this position can be either helpful or harmful depending on the particular TCR Vβ, thus defining the type A phenotype versus type B phenotype.

There are shared features among the TCR-MHCII complexes

Despite the differences in orientation, footprint, and specificity of the TCRs on their MHCII-peptide ligands, the complexes have several features in common that can explain some of the previous observations about the response T cells to insulin in T1D. For example, in the mouse, both the type A and type B CD4+ T cell responses to the B:9-23 insulin peptide are dominated by CD4+ T cells, such as 8F10, whose TCRs bear members of the Vα13 (TRAV5) family or, in some cases, such as I.29, the related Vα15 (TRAV10) (Table 1). In the I.29 and 8F10 complex structures, the Vα CDR1 and CDR2 regions sit in similar positions on the IAg7 β1 helix. Particularly sticking is the fact that the CDR2 loops of these Vαs are nearly identical in sequence (Table 1) and bind very similarly to the IAg7 β1 helix at a site highly conserved among MHCII β chains of most mammalian species (Fig. 4A). Although the Vα3 of the human T1D3 TCR is not particularly related to mouse Vα13 or Vα15, its CDR2 is very similar (Table 1) and it sits on the same conserved site of the DQ8 β chain (Fig. 4A). Note that the human Vα8 of the T1D4 TCR is highly homologous to mouse Vα13, and its CDR2 is identical (Table 1). It seems likely that it would also have a similar docking of its αCDR2 on the DQ8 β chain.

Fig. 4 Conserved features of the three TCR complexes.

(A) Similar interactions of the TCR Vα CDR2 loops (cyan) of the I.29 (left), 8F10 (middle), and T1D3 (right) with conserved amino acids (β69E to β77T, magenta) of the IAg7 or DQ8 β-chain α helix. The view is from within the peptide-binding groove looking out to the β-chain α helix. (B) Importance of the p3Y (B:16, yellow) for interaction with the Vα CDR3s of the I.29 (left), 8F10 (middle), and T1D (right) TCRs. H-bonds are green. Also shown is how the introduced disulfide from p6 in the 8E9E6ss peptide to IAg7 α62 changes the position p3Y, realigning its hydroxyl (curved arrow) to make H-bonds both to the I.29 TCR Vα 100N and to α61Q of the IAg7 α chain. (C) The I.29 (left), 8F10 (middle), and T1D3 (right) TCRs all interact with the side chain of p-1E (B:13) of the peptide. Potential H-bonds or salt bridges are shown as green lines.

These conserved interactions between the α helix of the MHCII β chain and the TCR Vα CDR2 loop could explain the heavy selection for certain Vα elements during the response to this peptide. The interaction may create a flexible pivot point for initial TCR binding to the MHCII β chain via the Vα domain while allowing the different Vβ domains to take different pitches and angles before settling to complete the interaction with the peptide and MHCII α chain. This would be a version of the two-step TCR binding mechanism previously proposed (29).

Another feature common to the three complexes is the placement of insulin B:16Y in the center of the TCR footprints (Fig. 2C). This amino acid is very prominently exposed on the MHCII-peptide surfaces at p3 when the peptide is bound in R3 (11). Nakayama et al. (10) established the importance of this amino acid in mouse T1D, demonstrating that replacing it with an A in NOD mice prevents the development of diabetes. Mutation of this Y prevents TCR recognition by both the mouse and human CD4+ T cells reactive to B:9-23. (10, 11). In agreement with these results, in our structures, all three TCRs strongly interact with this amino acid in the complexes with multiple van der Waals and H-bond interactions (Fig. 4B and table S2). Particularly noteworthy is the I.29 TCR complex with IAg7-8E9E6ss. Our previous work showed that the disulfide introduced between p6C to IAg7 α62C caused a small shift in the position of the p3Y side chain in the 8E9E6ss structure while improving recognition of the complex by the I.29 TCR (7, 9). The I.29 complex structure suggests an explanation for this improved response (Fig. 3D). This shift perfectly aligns the p3Y hydroxyl to create bidentate H bonds with the I.29 Vα 100 N and IAg7 α61Q.

Last, the focus of the TCRs on p3Y shifts their footprints toward the N-terminal part of the peptide. Consequently, all three TCRs now make a strong contact with the peptide E at p-1 (Fig. 4C). The p-1 amino acid is only occasionally contacted by TCRs interacting with MHCII ligands in the dozens of published structures of these complexes.

Could there be natural modifications of the B:9-23 peptide that create similar agonists?

The data we present here reinforce our conclusions from previous studies that the B:9-23 peptide is recognized by NOD and human CD4+ T cells bound to IAg7 or DQ8 in R3 and that mutations at the C-terminal end of the peptide are required to create an effective epitope (69, 11, 30). We have suggested that a C-terminal posttranslational modification in the peptide may be required in vivo for induction of T1D and have proposed that these modifications could be accomplished by the process of transpeptidation (9, 16). Transpeptidation is a posttranslational mechanism in which proteases can fuse two peptides to form a new chimeric peptide [reviewed in (14, 15)]. As shown schematically in Fig. 5A, during proteolysis with serine, threonine, or cysteine proteases, a transient covalent bond is formed between the enzyme and the newly generated carboxylate at the cleavage site. This bond can be resolved by water to complete the proteolysis or by the N terminus of a nearby peptide, thus reforming a peptide bond and producing a new, chimeric peptide. This reaction has long been studied in vitro and is known to play a role in the natural processing of proteins in vivo in microorganisms, plants, and animals. It can be particularly efficient when the donor and acceptor for the fusion are part of the same protein (15).

Fig. 5 How transpeptidation-mediated internal deletions in proinsulin could form superagonists similar to the mutant insulin peptides.

(A) Schematic representation of how transpeptidation could cause internal deletions in proinsulin (see text for description). (B) Some potential transpeptidation protease-mediated deletions in mouse or human proinsulin that could generate superagonist fused epitopes with properties of 8E9E and 8G9E mutant peptides. (C) Peptides were synthesized, joining the highlighted donor C-peptide sequences in (B) to the appropriate truncated B-chain acceptors shown. The natural B:19C at p6 was mutated to A to prevent peptide dimerization in vitro. The figure shows the IL-2 produced in representative titrations of the mouse or human peptides with two mouse type A T cells (I.29 and 12-4.1), two mouse type B mouse T cells (8F10 and 12-4.4), and two human T cells (T1D3 and T1D4) as described in Materials and Methods. (D) All of the mouse and human T cells in Table 1 (except 7E6) were tested with the synthetic WT and fused peptides as shown in (C). Potencies (i.e., fold increase in activity of the fused peptide over the WT peptide) were calculated as described in Materials and Methods. The mouse data are from three experiments containing single titrations. The average increases in potency are shown with the SEM of the result from each experiment (open circles). The P values (Student’s two-tailed t test rounded to a single significant place) are shown for peptides with average potencies >1. The human data are from two experiments, each with triplicate titrations, which were averaged before calculating the potencies. The average overall increases in potency are shown with the result in each of the two experiments (open circles).

In recent years, transpeptidation by proteasomal threonine proteases to create neopeptides presented by MHCI has been well documented (12, 13, 18, 19). We have proposed that the lysosome should also be an ideal site for the creation of chimeric neoepitopes feeding into the peptide-loading pathway for MHCII (9, 16). The high concentration of many different cysteine cathespsins and serine proteases in the confined space of the lysosome creates an ideal milieu for transpeptidation such as that found in the proteasome. We and others have suggested that neoantigens formed in this way could create the epitopes driving T1D and other autoimmune diseases (9, 1417). Because transpeptidation is enhanced by the proximity of donor and acceptor peptides, we have considered whether transpeptidation-mediated deletions within proinsulin that fuse a B-chain–derived acceptor with a C-peptide–derived donor could create superagonist versions of the B:9-23 similar to our mutated peptides.

Examples of potentially functional chimeric peptides, in mouse and in human, that could be generated by internal transpeptidation in proinsulin are shown in Fig. 5B. For the mouse type A and human type A–like T cells listed in Table 1, the acceptor would be an insulin B-chain fragment cleaved at B:21E to remove B:22R and donors would be peptide fragments from C-peptide with an N-terminal E to replace the p9R. There are many possibilities within mouse and human C-peptide, a few of which are shown. For mouse type B T cells, the acceptor would be formed by a cleavage of the B chain at B:20G to remove both B:21E and B:22R. C-peptide donors would be peptides that have an E or D at the second position for p9 and some amino acid other than E at the N terminus for p8. The possibilities are G, V, or L, thus replacing B:21-22 ER with GD, VE, or LE. To test the predicted potencies of these chimeric peptides, we synthesized six of them (highlighted in Fig. 5B) and compared their stimulating activity to that of the wild-type (WT) B:11-23 peptide with 11 of the 12 T cells listed in Table 1. Figure 5C shows sample titrations of the peptides with two mouse type A, two mouse type B, and two human type A–like T cells, showing data representative of the patterns of stimulation we obtained. Titrations with all eight mouse T cells and the three human T cells were performed, and the results are summarized in Fig. 5D.

As predicted, the B:12-21 acceptor peptide fused to the N-terminal fragment of mouse C-peptide, EVE, was every bit as effective with the NOD type A T cells as the B:22R to E mutant peptide used in our previous studies (9), increasing potency about 100- to 500-fold. In addition, as expected, this fused peptide was virtually inactive with all but one of the NOD type B T cells. Likewise, for the human type A–like T cells, fusing either the EAE or ELG fragments of human C-peptide to B:12-21 improved the potency of the peptide with the human type A–like T cells >1000-fold.

For all of the NOD type B T cells, fusion of B:12-20 to the mouse C-peptide fragment, GDLQ, increased its potency 100- to 1000-fold. The results with the VEQL and LELG were very different. 8F10 failed to respond to these peptides, consistent with the structure of the 8F10 TCR bound to the 8G9E peptide bound to IAg7 (Fig. 2C) and the 8F10 stimulation data in Fig. 2D. Similarly, 8-1.1 failed to respond to these peptides. On the other hand, the 12-4.4 and AS91 T cells both responded to these peptides in several cases much better than they did to the GDLQ-fused peptide or, in our previous experiments, with the 8G9E mutated peptide. For 12-4.4, the potency of the LELG-fused peptide increased >500,000-fold, and for AS91, the potency of the VEQL fused peptide increased ~100,000-fold (Fig. 5D). The GDLQ, VELQ, and LELG stimulated all of the type A NOD T cells worse than the unmutated control peptide. Together, these results point out how the appropriate modifications of the C terminus of the insulin B-chain peptide can lead to enormous increases in the stimulating activity of these fused peptides for diabetogenic T cells.


For decades, the NOD mouse and certain rat strains have been used as models for human T1D [reviewed in (31)]. Although some have questioned whether these animal models are relevant to the human disease, the similarities in the genetic risk and in the CD4+ autoimmune T cell response between humans and rodents are notable. One problem in using the rodent CD4+ T cell responses in T1D to understand the corresponding human CD4+ T cell response has been a lack of consensus on the molecular nature of the diabetogenic MHCII-peptide complexes driving the disease in these species. Nowhere has this problem been more apparent than in defining the CD4+ T cell response to insulin, a major autoantigen in both humans and rodents. In NOD mice, the insulin B-chain peptide B:9-23 was reported decades ago to contain epitopes recognized by a variety of diabetogenic CD4+ T cells (32), but how this peptide binds to IAg7 to form the pathogenic complex has remained a contentious issue. Different studies have proposed different binding positions or registers even for the same T cells (2, 3, 6, 7, 9).

Our studies, presented here and previously, resolve this confusion. We show that, for the T cells listed in Table 1, which include those previously reported to recognize R1- and R2-presented epitopes, the functional register is R3, which places ALYLVCGER (B:14-22) in the core p1 to p9 positions in the IAg7 or DQ8 peptide-binding groove. The extremely poor binding of this peptide in this register is due to the incompatibility of the B:22R with the p9 pocket of IAg7 and DQ8. Binding can be improved about 100-fold by changing this amino acid to the E, which is optimal for peptide binding to either MHCII allele (6, 7, 33). Disulfides engineered between the peptide and the IAg7 or DQ8 α-chain helix, which can form only when the peptide is bound in R3, show that this mutation is sufficient to force R3 binding (6, 7, 9, 11). Furthermore, we have shown that an additional mutation at B:14A to R for IAg7 or to E for DQ8, although not required for optimal binding and T cell recognition, creates an optimal R3 p1 anchor for these MHCII alleles and does not interfere with peptide recognition by the set of T cells shown in Table 1 (6, 7, 9, 11). However, if the functional presenting register for these T cells were in fact R1 or R2, then these B:14 mutated amino acids would lie on the surface at p3 (R1) or p2 (R2) where they would be expected to interfere with TCR recognition.

Our structures presented also show how mouse and human TCRs interact with these R3-presented epitopes. We conclude that, for the human T cells and mouse type A T cells in Table 1, the natural B:21E at p8 is an important part of the ligand interacting via a salt bridge with an R in the TCR. On the other hand, for mouse type B T cells, this E is very inhibitory and recognition is greatly improved by the removal of its side chain by changing it to G or, for some T cells, a V or L. Moreover, although the TCR Vα and Vβ segments and CDR3 sequences are quite different among the three T cells studied here, there are notable similarities in their complexes with their ligands. These include virtually identical interactions between the VαCDR2 loops with a highly conserved site on the MHCII β-chain α helix and a focus by the TCR on the surface exposed B:16Y at p3, which was previously shown to be important for T1D development in the NOD mouse. These findings clear up much of the uncertainty in the literature about how this insulin epitope is bound and recognized by CD4+ T cells.

However, our findings raised the question: if these C-terminal modifications of this peptide are needed to create strong R3 MHCII binding and T cell stimulation, what is the nature of the real peptide driving the disease in vivo? We have suggested that there may be a role for posttranslational modifications of this region of the peptide in T1D development (8, 9, 16). There are now multiple examples of how posttranslational modifications can convert weak T cell autoantigens into strong or “heteroclitic” epitopes [reviewed in (8, 34)]. Specific examples are the conversion of arginine to citrulline in epitopes driving rheumatoid arthritis (35), conversion of glutamine to glutamic acid in celiac disease (36), and modifications to a myelin basic protein peptide in a mouse model of multiple sclerosis (37). As with our results with the B:9-23 peptide, in these three examples, the improving modification can be in either an anchor or a TCR contact residue.

On the basis of the recent findings that transpeptidation in the proteasome creates chimeric peptides for loading into MHCI (14, 15, 18, 19), we have proposed peptide fusion by transpeptidation as a means of creating the required modifications to the B:9-23 peptide (9, 16). Here, we have synthesized several versions of proposed transpeptidation-mediated deletions within mouse and human proinsulin, creating peptides every bit as active and, in some cases, markedly more active than our original mutated peptides. A limitation to our hypothesis is that we have not yet identified the presence of these proinsulin chimeric peptides in vivo in the pancreas, a formidable task on which we continue to work. However, the requirement for fusing a B-chain acceptor to a C-peptide donor to complete these CD4 epitopes could explain why patients with T1D commonly do not show hypersensitivity responses to daily injections of fully processed insulin, because mature insulin lacks the attached C-peptide, which might be required to complete the fusion.

A similar story has emerged for another NOD T1D antigen involving the WE14 peptide of chromogranin A (38). We have shown that the WE14 peptide binds very poorly to IAg7 because it fills only the p5 to p9 positions of the IAg7-binding groove. Its binding and recognition can be improved tremendously by adding the appropriate four amino acids to its N terminus that fill the rest of the peptide-binding groove (7, 16). These amino acids also contribute to TCR binding at p2 and p3 while providing an optimal anchor amino acid at p4 (16). We suggested that this modification could be mimicked in vivo by transpeptidation and proposed a list of potential acceptors for a WE14 donor among various beta cell granule proteins (16). One of these, from proinsulin C-peptide, would place TLAL at the p1 to p4 positions when fused to WE14. The Haskins group also proposed this fusion and were able to find the fused peptide in lysates of a primary insulinoma tumor (17) and mouse pancreatic cells (20). They also found the same C-peptide fragment fused to a peptide from islet amyloid polypeptide precursor, completing the epitope for another diabetogenic CD4+ T cell (17, 20, 39). These remain the only CD4 functional chimeric epitopes reported to be found so far from an in vivo source.

An open question is still where these transpeptidation reactions might take place. We suggest that the relevant site is the lysosome, because it feeds into the MHCII loading pathway and has the right proteases and environment for the reaction. One possibility is that transpeptidation could be a by-product of crinophagy in beta cells, in which granule turnover is regulated by fusion with lysosomes, a process that is enhanced by beta cell stress (40). Alternately, lysosomes in antigen-presenting cells (APCs), such as dendritic cells, macrophages, or insulin-specific B cells, present either in the islet or in the pancreatic lymph node, could be the site of the reaction.

Last, the growing list of T cell autoantigen epitopes created by various mechanisms of peptide posttranslational modification raises the question of whether these processes are restricted to the periphery and absent in the thymic medullary epithelial cells responsible for T cell–negative selection. If so, then this could explain the escape of the pathogenic T cells from the thymus. Exploring this idea requires a much deeper understanding of the antigen-processing and MHC-presentation pathways of these proteins in the thymus than is currently available.


Study design

This study was designed to use x-ray crystallography to examine the interface between TCRs from human and mouse CD4+ TCRs and ligands consisting of superagonist-mutated insulin B-chain peptides bound to MHCII. On the basis of these results, we also analyzed synthetic chimeric peptides designed by fusing fragments of insulin C-peptide to the C termini of B-chain fragments to see whether they have the same superagonist activities as the mutant peptides. These experiments begin to test the idea that transpeptidation in vivo might create superagonist-chimeric epitopes to drive T1D.

T cell hybridomas and T cell avatars

The origins, constructions, and properties of the 12-4.1, 12-4.4, 8-1.1, PCR1-10, I.29, 8F10, AS150, and AS91 NOD mouse T cell hybridomas and TCR-transduced avatars, as well as the avatar versions of the human T1D3, T1D4, and T1D10 T cell clones used in these studies, were described previously (2, 3, 6, 7, 9, 11, 2123). Briefly, hybridomas were produced by fusion to the TCR-negative version of the mouse T cell tumor line, BW5147 (41). T cell avatars were produced by cloning a sequence encoding the TCR V domains fused to mouse Cα or Cβ into MSCV (mouse stem cell virus)–based vectors bearing an internal ribosomal entry site, followed by green fluorescent protein or human nerve growth factor as surrogate markers. Virus prepared from these constructions were used to transduce a mouse T cell hybridoma (5KC) that had previously been selected for the loss of its original functional TCR genes (42). The human avatars were transduced with human CD4, and the 12-4.1 T cell was transduced with a mutant version of human CD4 with high affinity for MHCII (43, 44). This system was also used to produce avatars expressing the I.29, T1D3, T1D4, or T1D10 TCRs with single-point mutations in their CDR loops. For testing the effects of these mutations, the cells were selected for comparable levels of surface TCR. All cells were cultured in supplemented minimum essential medium with 10% fetal calf serum as previously described (41).

Antigen presentation assays and soluble peptides

For APCs for soluble peptides, we used two versions of the M12.C3 B cell lymphoma, one expressing IAg7 (M12.C3.g7) (45) and the other expressing DQ8 (M12.C3-DQ8-8) (46). In addition, for the presentation of the peptides to T cell avatars expressing mutated versions of T1D3, T1D4, or T1D10 human TCRs (Fig. 2F), the K562 human leukemia cell line (47) was transduced with MSCV expressing HLA-DQ8 bearing the covalently attached insulin peptide shown in Fig. 1A (11). For soluble peptides, T cell hybridomas or TCR-transduced avatars (105 cells) were mixed with 105 paraformaldehyde-fixed (7) APCs and cultured overnight with various concentrations of peptide in a volume of 250 μl. When using the K562 cells bearing HLA-DQ8 with a covalent peptide, the APCs were not fixed and no peptide was added. Secreted interleukin-2 (IL-2) was assayed with either a functional assay, after the growth and survival of the HT-2 IL-2–dependent cell line (41), or with an enzyme-linked immunosorbent assay–based assay for IL-2 (48). The relative potencies of the mutant and chimeric peptides relative to the WT insulin peptide were calculated from the shift in the IL-2 production versus peptide titration curves along the peptide log10 dose axis (41). Soluble peptides (>95% pure) were obtained from either CHI Scientific (Maynard, MA) or Schafer-N (Copenhagen, Denmark).

MHCII-peptide expression and purification

As previously described (7, 9), acid-base leucine zipper–stabilized, soluble IAg7 and human HLA-DQ8 molecules with covalently attached peptides were produced in baculovirus-infected insect cells and purified by immunoaffinity chromatography. For surface plasmon resonance (SPR) or flow cytometry experiments, a biotinylation peptide tag attached to the C terminus of the acidic half of the zipper was enzymatically biotinylated with BirA enzyme produced in our own laboratory, after purification of the molecule. For crystallization, the zippers and biotinylation tag were removed with papain.

TCR expression and purification

Sequence-encoding V domains of WT mouse TCR I.29, 8F10, and human TCR T1D3 were cloned separately into the pET30 bacterial expression system, in which the TCR V domains were fused to sequence encoding the extracellular domains of human Cα or Cβ (24, 25). Mutated I.29 TCR (β56 K to A) and 8F10 TCR (α169R to A) were prepared similarly. After isopropyl-β-d-thiogalactopyranoside induction of overexpression, the resultant proteins were solubilized from inclusion bodies in 8 M urea buffer, mixed, and refolded by gradual dialysis. The refolded TCR was further purified with a HiLoad Superdex 200 26/600 size chromatography column, followed by a Mono Q ion exchange chromatography. The mutated TCRs were tested for binding to their ligands to confirm that mutations had no impact on TCR binding to their IAg7-peptide ligands.

SPR measurements

About 2000 RU of biotinylated DQ8-8E9E11ss or IAg7-8E9E were captured in the two separate flow cells of a BIAcore streptavidin BIAsensor chip. Various concentrations of the soluble refolded T1D3 TCR were injected, and the association and dissociation kinetics were recorded. The data for the DQ8-8E9E11ss flow cell was corrected for the fluid-phase SPR signal using the data from the IAg7-8E9E flow cell as baseline. Kinetics were analyzed using BIAcore BIAEval 4 software.

Protein crystallization

Crystallization was performed by the hanging drop vapor diffusion method. WT TCR I.29 and 8F10 proteins were concentrated to 5 mg/ml. I.29 crystals were grown from 20% 2-methyl-2,4-pentanediol, 100 mM sodium cacodylate at pH 6.0, and 50 mM calcium acetate. 8F10 crystals were grown in 14% polyethylene glycol (PEG) 3350 and 100 mM sodium acetate at pH 6.0. Both I.29 and 8F10 crystals were cryoprotected by a well solution plus 25% glycerol. To obtain the I.29/IAg7-8E9E6ss complex crystals, IAg7-8E9E6ss and I.29(β56A) mutant TCR were mixed at a molar ratio of 1:1 at the concentration of 10 mg/ml. The crystals were grown in 12% PEG 20000, 100 mM bicine (pH 9.0), and 20% ethylene glycol and then cryoprotected by the well solution plus 15% ethylene glycol. Similarly, IAg7-8G9E/8F10 complex crystals were obtained by mixing IAg7-8G9E and 8F10(α169A) mutant TCR at a 1:1 molar ratio at the concentration of 10 mg/ml. Crystals were grown in 11% PEG 3350 and 100 mM formate (pH 7.0) at room temperature and then cryoprotected by the well solution plus 30% glycerol. Human T1D3/DQ8-8E9E11ss complex was crystallized in 10% PEG 8000, 100 mM cacodylate (pH 6.2), and 400 mM NaCl at 4°C by mixing TCR T1D3 and DQ8-8E9E11ss at a 1:1 molar ratio at the concentration of 12 mg/ml and then cryoprotected by the well solution plus 30% glycerol.

Data collection, data processing, and structural analysis

All diffraction datasets were collected at synchrotron beamline ID-24C at the Advanced Photon Source, Argonne National Laboratory using the Pilatus Detector. The 8F10/IAg7-p8G9E complex data was processed by XDS software package at ID-24C beamline (49). All the rest of the collected data were processed with HKL2000 package (50). The structures were solved by molecular replacement method using Phaser (51) software and further refined by REFMAC5 (52) or Phenix (53). Rebuilding of the structure was performed with Coot (54). NCONT in CCP4 (55) was used to analyze the atom-to-atom contacts between the TCRs and their ligands. Atoms within 4.5 Å of each other were considered part of the interface. Contacts involving potential electron donors and acceptors (O or N) within 3.5 Å were considered potential hydrogen bonds. Negatively charged O’s within 4.0 Å of positively charged N’s were considered salt bridges. Other contacts were considered van der Waals contacts. Data collection and refinement statistics are shown in table S1. The contributions of the TCR CDR loops to the BSA in the interface with their ligands were calculated with the CCP4 PISA program (56). Molecular superimpositions were performed with the Swiss PDBViewer (57). Graphical representations were made with Discovery Studio 3 (Accelrys).


Fig. S1. Mutations to prevent I.29 and 8F10 TCR dimerization.

Fig. S2. Extension of the mutational analysis of the I.29 TCR shown in Fig. 2B in the main manuscript.

Table S1. Crystallography statistics.

Table S2. Supplementary spreadsheet listing TCR- to MHC-peptide contacts.

Table S3. Supplementary spreadsheet with raw data used to construct graphs in Figs. 2 and 5.


Acknowledgments: We thank E. Kushnir and C. Brown for technical support, as well as K. Bakke and M. Millstein for administrative support. P.M. and J.W.K. acknowledge their previous long-standing support from the Howard Hughes Medical Institute. Funding: This work was supported by NIH grants 5 T32-AI-074491 (to Y.W.), ES-025797 and ES-025885 (to S.D.), AI-018785 (to P.M. and J.W.K.), and DK-032083 (to T.S.); JDRF grant 1-2008-588 (to L.G.) and P01AI-118688 (J.W.K.), as well as JDRF grant 1-PNF-2015-126-A-R; University of Colorado CCTSI grant KL2 TR001080; and a Boettcher Foundation Investigator award (to S.D.). Additional financial support came from the National Jewish Health (to J.W.K. and P.M.), as well as the School of Medicine of University of Colorado at Denver and the Claire Friedlander Family Foundation (to J.W.K.). The synchrotron data were collected at the Advanced Photon Source Argonne National Laboratory, beamline 24 ID-C, supported by NIH grant GM103403 and NIH-ORIP HEI grant S10 RR029205. Author contributions: Y.W., T.S., J.W.K., and S.D. designed the studies. Y.W., T.S., A.N., F.C., J.W., N.J., Z.L., J.Z., J.W.K., and S.D. performed the experiments. H.W.D., L.G., J.W.K., and S.D. supervised the work. M.N. and W.W.K. provided essential reagents. D.N. assisted in collection of the x-ray data. Y.W., F.C., P.M., J.W.K., and S.D. wrote and edited the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All of the cell lines and expression plasmids used in this work are available on request to J.W.K., M.N., and S.D. The coordinate and structure factors files for the five structures presented here have been validated and deposited in the RCSB Protein Data Bank for release at publication. Accession numbers are 6DFQ, 6DFS, 6DFV, 6DFW, and 6DFX.
View Abstract

Stay Connected to Science Immunology

Navigate This Article