Molecular Pathways for Immune Recognition of Preproinsulin Signal Peptide in Type 1 Diabetes

In type 1 diabetes, the pathological process of immune-mediated destruction of insulin-producing β-cells leads to insulin deficiency and hyperglycemia (1). Multiple arms of the immune system are likely to contribute to this tissue-damaging process, with strong indications that CD8⁺ cytotoxic T lymphocytes (CTLs) are a dominant killing pathway. Evidence includes data from preclinical models showing dependence of disease development on intact CD8/MHC class I mechanisms (2), supported by compelling findings in human studies, including the existence of high-risk polymorphic HLA class I genes (3); enrichment of effector CTLs specific for β-cell targets in the circulation in new-onset disease (4,5); recapitulation of β-cell killing by patient-derived preproinsulin (PPI)-specific CTLs in vitro (4,6); CD8 T-cell dominance of islet infiltrates in patients, including the presence of CD8s bearing receptors specific for β-cell autoantigens (7,8); hyperexpression of HLA class I, both at the RNA and protein levels, in residual insulin-containing islets (ICIs) in type 1 diabetes pancreatic tissue (9); and recent success in halting β-cell loss using immunotherapy targeted at effector CD8 T cells (10).

The potential for the immunological dialogue between CTLs and β-cells to be a key component of the development of type 1 diabetes has led several groups to focus on the relevant molecular interactions that govern this interface, including studies of HLA class I gene polymorphisms carrying modified risk of disease (HLA-A*0201, -A*1101, -A*2402, -B*1801, -B*3801, and -B*3906 [3,11,12]) and antigenic targets within β-cells recognized by CTLs. These have especially focused on PPI, which is considered a primary target in β-cell autoimmunity because anti-insulin autoantibodies are frequently the first disease manifestation in high-risk children (12).

For immune recognition, processing and presentation of peptides by MHC class I commonly results from degradation of proteins by the proteasome (13), generating cytosolic peptides of 8–16 amino acids (14), which are transported into the endoplasmic reticulum (ER) lumen via the transporter associated with antigen processing (TAP) (15), part of the MHC class I peptide loading complex (16). ER aminopeptidase (ERAP) 1 trims peptides to a suitable length for MHC loading, if required (17), and peptide-MHC complexes are delivered to the cell surface for immune recognition. Proteasome-independent, cytoplasm-based noncanonical antigen presentation pathways have also been described (18). Whereas these and the canonical route collectively require TAP to delivery peptides into the ER, signal peptide epitopes from secretory proteins may be MHC-loaded independently of TAP (19) after signal peptidase cleavage and intramembrane proteolysis by signal peptide peptidase (SPP) (20–22). Interestingly, peptides originating from the ER luminal side of the signal peptide can access the peptide loading complex directly (23–25), whereas epitopes close to the cytosol may require proteasomal trimming and TAP for entry into the ER (26). However, the extent to which signal peptides and these noncanonical epitope-generation pathways fuel immune recognition of single antigens and play a role in physiological responses remains unclear.

This study was motivated by the need for a better understanding of autoantigen processing for immune recognition of β-cells via these different routes, which could provide novel insights into disease pathogenesis and highlight pathways susceptible to therapeutic manipulation. We previously reported that the predominant epitope species presented to CTLs by human cell lines cotransfected with the INS gene (encoding PPI) and the HLA class I genes HLA-A*0201 and A*2402 derived from the signal peptide region (4,6). These signal peptide–derived epitopes are recognized by patient CTLs and presented by β-cells bearing the relevant HLA class I molecule. Here we examine whether PPI signal peptide is a more general source of epitopes by studying additional HLA class I molecules associated with type 1 diabetes. We show that generation of epitopes from the PPI signal peptide is driven by the intramembrane protease SPP and that loading into nascent HLA class I molecules requires trimming by ERAP1 and is either direct or follows cytoplasmic translocation and TAP, with the selected pathway being determined by the HLA allele. We show that the key factors ERAP1 and TAP are expressed in insulin-containing islets of patients studied post mortem after type 1 diabetes diagnosis, indicating that the multiple pathways that are potentially critical in the CTL–β-cell dialogue are active in the disease setting.

K562 cells transfected with HLA-A*1101, HLA-B*1801, HLA-B*3801, or HLA-B*3906 and PPI cDNAs (K562-HLA-PPI cells) were generated as described previously (4). Expression of HLA class I was confirmed by flow cytometry (anti-HLA-ABC antibody W6/32; Serotec, Oxford, U.K.). Proinsulin was detected in supernatants by ELISA (DRG International, Marburg, Germany). Subsequently, ∼10¹⁰ cells from each cell line were harvested and immunoaffinity purification of HLA-A1101, HLA-B1801, HLA-B3801, and HLA-B3906, peptide extraction and nano high-performance liquid chromatography–mass spectrometry performed as described elsewhere (4,6,27,28).

K562-A24-PPI cells were transfected twice within 48 h with 20 nmol/L small interfering RNAs (Applied Biosystems, Foster City, CA) targeting B2M (s1852), ERAP1 (s28618), ERAP2 (s34520), TAP1 (s13778, s13780), and TAP2 (s13781, s13782, s13783) using Lipofectamine RNAiMAX (Invitrogen, Paisley, U.K.). Knockdown was assessed by RT-PCR using TaqMan-specific primers and relative cDNA content normalized to GAPDH gene expression. CD8 T-cell clones for PPI_3–11-HLA-A2402 and PPI_15–24-HLA-A0201 (4,6) were cocultured with K562-HLA-PPI target cells at indicated effector-to-target ratios (4 h), and response was measured as degranulation (CD107a expression [29]) or macrophage inflammatory protein 1β (MIP-1β) release (R&D Systems, Minneapolis, MN).

To alter single or multiple amino acids (highlighted in Supplementary Table 1), PPI-containing plasmid pcDNA3/PPI was amplified using altered primers (PPI9P→L, CCC→CTC; PPI12A→L, GCG→CTG; PPI15A→L, GCC→CTC; with 18- to 20-nucleotide overhang preceding and following the mutated site) and PfuTurbo DNA Polymerase AD (Agilent Technologies, Santa Clara, CA); sequences were confirmed before use.

Plasmid pcDNA3/PPI was linearized with EcoRI and transcribed in vitro with T7 RNA polymerase at 42°C using 500 μmol/L m7G(5′)ppp(5′)G CAP analog (New England Biolabs, Ipswich, MA) (30). mRNAs were translated in vitro in 25 μL rabbit reticulocyte lysate (Promega, Madison, WI) containing [³⁵S]-methionine and [³⁵S]-cysteine (PerkinElmer, Waltham, MA) and, where indicated, two equivalents of nuclease-treated dog pancreas rough microsomes (31). (Z-LL)₂-ketone (Calbiochem SPP inhibitor, 5 μmol/L; Merck, San Diego, CA) or DMSO control were added as indicated. After 30 min at 30°C, microsomes were extracted with 500 mmol/L KOAc, solubilized in SDS sample buffer (32) and analyzed by SDS-PAGE using Tris-bicine-urea acrylamide gels (15% Tris, 5% bicine; 8 mol/L urea) (33) and an FLA 7000 phosphorimager (Fuji) with Multi Gauge software (Fuji). The reference peptide comprising the 24–amino acid PPI signal sequence was translated in wheat germ extract (34).

Formalin-fixed, paraffin-embedded pancreas sections from six control and six case subjects with type 1 diabetes (Exeter Archival Diabetes Biobank, http://foulis.vub.ac.be/; Supplementary Table 2; ethical approval 15/W/0258) were studied with the use of immunohistochemistry. Sections were dewaxed, rehydrated, and heated in a microwave (800 W) for 20 min, then blocked in 5% normal goat serum before being incubated with primary and secondary antibodies (Supplementary Table 3), counterstained with hematoxylin, dehydrated, and mounted in a distyrene/xylene-based mountant (DPX). Multiple antigens within the same formalin-fixed, paraffin-embedded section were probed sequentially with up to three different antibodies (Supplementary Table 3) and images captured (AF6000 system; Leica Microsystems, Milton Keynes, U.K.). Mean fluorescence intensity of stained antigens was analyzed using ImageJ software, and isotype control antisera were used to confirm reagent specificity.

CRISPR guide sequences for exons of SPP (HM13 gene) were designed using CRISPR DESIGN tool (crispr.mit.edu) (Supplementary Table 4) and cloned into pLG2C vector containing EGFP linked to a Cas9 cassette via P2A. K562-A2-PPI was transfected with each of two pLG2C vectors (containing guide sequences for exon 2 or 3 and exon 10 or 11) or empty vector (mock treated, no guide sequence) using Effectene (Qiagen, Hilden, Germany). Single cells sorted for high HLA-A2 (W6/32; BioLegend, San Diego, CA) and EGFP were examined for gene truncation by RT-PCR and for SPP amplification using primers (forward: 5′-ATATATGAATTCGCACCCTCGCCATG-3′; reverse: 5′-ATATATCTCGAGGCACCAGCTGCATCATTTC-3′) (Eurofins Scientific, Ebersberg, Germany) and Phusion High-Fidelity DNA Polymerase (New England Biolabs) and by agarose gel electrophoresis.

K562 cells transfected with INS encoding human PPI and one of HLA-A*1101, HLA-B*1801, HLA-B*3801, and HLA-B*3906 generated surrogate β-cells secreting proinsulin and expressing relevant HLA class I molecules (Supplementary Fig. 1). The immunopeptidome eluted from affinity-purified HLA-A1101 identified 905 peptides; from HLA-B1801, 615 peptides; from HLA-B3801, 455 peptides; and from HLA-B3906, 298 peptides (all Mascot scores ≥40), and corresponded to published HLA-binding motifs and peptide ligandomes (35). Of interest, PPI epitopes from the signal peptide region were identified for HLA-B3801, PPI_5–14 MRLLPLLALL, and HLA-B3906, PPI_5–12 MRLLPLLA (Fig. 1 and Supplementary Fig. 2). In addition, we identified a B-chain epitope for HLA-B3801, PPI_33–41 SHLVEALYL, and a C-peptide epitope for HLA-A1101, PPI_80–88 LALEGSLQK. Peptide identities were confirmed by tandem mass spectrometry profiling of the synthetic compound (Supplementary Fig. 2). No other peptides from PPI were identified.

When considered with our previous reports of immunodominant PPI signal peptide–derived epitopes in PPI_15–24-HLA-A0201 (6) and PPI_3–11-HLA-A2402 (4), these new discoveries indicate that the signal peptide region of PPI is a rich source for processing for immune recognition (Fig. 1 and Table 1). To examine whether this arises because signal peptide regions are immunogenic per se, as has been reported (36), the available peptidome data were mined in greater depth, showing that the degree to which signal peptides are presented is dependent on the HLA allele. While, for example, HLA-A0201 presents a signal peptide–derived epitope from 46% of source proteins that contain a signal peptide, for HLA-A1101 this figure is only 6.9% (Supplementary Table 5). Probing for any signal peptide bias using in silico prediction algorithms provided similar results (Supplementary Table 6). We conclude that presentation of signal peptides of PPI is not likely to result from a generalized propensity of HLA molecules to select this region for presentation, but the number of HLA molecules studied to date remains limited.

Based on previous studies of intramembrane cleavage of signal peptides (37), we hypothesized that SPP is involved in processing and cleavage of PPI signal peptide. To test this, mRNAs encoding PPI were translated in vitro with ER-derived microsomes and [³⁵S]-labeled methionine and cysteine. Upon insertion of PPI into microsomes, signal peptidase cleaves the PPI signal sequence, liberating translocated proinsulin from its signal peptide (Fig. 2A). Moreover, traces of a peptide that comigrated with an in vitro translated reference peptide comprising the PPI signal sequence were detected in the membrane fraction (Fig. 2A; compare lanes 2 and 4). Because processing of nascent chains by signal peptidase is a well-known activity in ER-derived microsomes (31), and because no low–molecular weight peptides were observed in the translation reaction lacking microsomes (lane 1), we conclude that the identified peptide corresponds to traces of PPI signal peptide that remain in the ER membrane fraction. By contrast, the PPI signal peptide is markedly stabilized and retained in the microsome fraction upon treatment with the SPP inhibitor (Z-LL)₂-ketone (Fig. 2A, lane 3). Overall, this shows that PPI behaves as a canonical nascent chain processed by signal peptidase, liberating a signal peptide released from the ER membrane by SPP-catalyzed intramembrane cleavage.

To further study the role of intramembrane proteolysis in turnover of PPI signal peptide, amino acid residues destabilizing the helical transmembrane span surrounding the scissile peptide bond were mutated to leucine, which blocks SPP-catalyzed cleavage (21). Consistent with our previous analysis of model signal peptides, single mutation of proline (at position 9) and alanine (position 12 and position 15) shows marginal effects on PPI signal peptide processing, whereas double mutants (9L/12L, 9L/15L, and 12L/15L) and the triple mutant (9L/12L/15L) show marked inhibition of SPP-catalyzed processing (Fig. 2B–D, Supplementary Fig. 3, and Supplementary Table 1). This identifies PPI signal peptide as a bona fide SPP substrate and confirms the requirement for helix breaks and limited hydrophobicity for intramembrane cleavage. Replacement of the classic helix-break residue proline at position 9 of PPI signal peptide alone is not sufficient to completely block SPP-catalyzed processing and only shows effect when at least one of the nearby alanine residues, which show an intermediate stability within transmembrane helices (38), is also mutated to leucine.

We next examined whether abrogation of SPP-catalyzed cleavage of PPI signal peptide affects immune recognition of signal peptide–derived epitopes. Double-targeting CRISPR-Cas9 technology generated three independent K562-A2-PPI cell lines with SPP knockout (K562-A2-PPI-SPPko; SPP knockout validation shown in Supplementary Fig. 4). When cultured with the PPI_15–24-specific, HLA-A0201-restricted CD8 T-cell clone, degranulation (which measures clone activation via T-cell receptor ligation by peptide-HLA) frequency and magnitude were markedly reduced in the presence of K562-A2-PPI-SPPko cells compared with mock-treated lines (Fig. 3). The difference in T-cell activation is not due to differences in HLA class I expression levels (Supplementary Fig. 5), and the SPP knockout phenotype of reduced T-cell activation is rescued when cells are pulsed with cognate peptide (Supplementary Fig. 6). These data show that SPP-catalyzed processing of PPI signal peptide is an essential step to generate an immunologically relevant epitope that is implicated as having a pathogenic role in type 1 diabetes through activation of β-cell-specific autoreactive CD8 T cells (6,8).

We previously showed that PPI_15–24 processing and presentation by HLA-A0201 does not require proteasome cleavage or import into the ER via TAP (6). In contrast with PPI_15–24-HLA-A0201, however, PPI_3–11-HLA-A2402 presentation is TAP dependent. RNA interference (RNAi) inhibition of TAP1 (94.5 ± 2.3% mRNA knockdown) and TAP2 (95.4 ±⁻ 0.3% mRNA knockdown) expression in K562-A24-PPI cells markedly reduces MIP-1β production (60.7% and 30.2%, respectively) by the PPI_3–11-HLA-A2402-specific CD8 clone upon coculture with TAP RNAi–treated K562-A24-PPI cells (Fig. 4A). This is attributable to a reduction in surface density of the specific peptide HLA ligand PPI_3–11-HLA-A2402 (TAP RNAi–treated K562-A24-PPI cells show only minimal reduction in HLA class I expression; Supplementary Fig. 7).

Next we investigated whether aminopeptidases residing in the ER are involved in the processing and presentation of PPI signal peptide epitopes. Knockdown of ERAP1 (87.2 ± 0.5% mRNA knockdown) in K562-A24-PPI cells by RNAi markedly reduced MIP-1β production (62.6%) by PPI_3–11-HLA-A2402-specific clone 4C6, whereas no effect was seen in the presence of ERAP2 knockdown (73.7 ± 6.1% mRNA knockdown; 8.0% change in MIP-1β production) (Fig. 4B). Similar effects were seen in target killing assays (data not shown). RNAi for ERAP1 and ERAP2 had only minimal effects on total surface HLA class I expression (Supplementary Fig. 7), indicating that the ERAP1-mediated interference in clone 4C6 activation is likely to be an effect on target cell surface density of the specific PPI_3–11-HLA-A2402 ligand.

Expression of the processing proteins TAP1 and ERAP1—both of which we have shown to potentially contribute to the generation of PPI epitopes that target β-cells for killing by PPI-specific cytotoxic CD8 T cells—were investigated in human pancreas recovered from healthy control subjects and those with type 1 diabetes. TAP1 was present at low levels in the pancreatic islets of healthy control subjects and was detected at similarly low levels in the insulin-deficient islets (IDIs) of patients with type 1 diabetes (Fig. 5A). By contrast, TAP1 was markedly upregulated in the ICIs of these patients (Fig. 5A). Colocalization studies revealed that the elevation in TAP1 expression was most evident in β-cells, although it was also increased in other islet cells. The immunostaining in six age-matched control subjects and six subjects with type 1 diabetes was quantified by measuring the mean fluorescence intensity of TAP1 labeling in three islets from each control subject and in six islets from each patient (three ICIs and three IDIs). This confirmed significant elevation of TAP1 expression in ICIs from patients with type 1 diabetes compared with IDIs or control islets (P < 0.001; Fig. 5B). Application of similar approaches demonstrated that, as previously described (9), HLA-ABC was also markedly elevated in the ICIs of patients with type 1 diabetes (Fig. 5C). Moreover, a strong positive correlation existed between the expression of TAP1 and HLA-ABC in ICIs (R² = 0.519, P < 0.001; Fig. 5D).

ERAP1 was detected in the islets of healthy control subjects and in the islets of donors with type 1 diabetes (Supplementary Table 2). In contrast to TAP1, however, ERAP1 expression was not noticeably altered between the islets of control subjects and those of patients with type 1 diabetes (Supplementary Fig. 8).

Collectively, these findings imply a model of PPI signal peptide processing for HLA class I presentation in which the fate that follows intramembrane cleavage depends on location within the ER membrane and HLA binding potential (Fig. 6). On the one hand, N-terminal peptides may be released into the cytoplasm, where they are dependent on TAP transport into the ER and ERAP1 trimming before loading into nascent HLA molecules. As an alternative, distally generated ER luminal peptides are directly released into the ER lumen, omitting the need for TAP, but these may require N-terminal trimming by ERAP1 for optimal presentation.

In this study we extend our previous PPI epitope discovery effort to encompass new HLA class I alleles associated with risk for/protection against type 1 diabetes. This reveals PPI signal peptide as a frequent source of epitopes for multiple HLA-A and HLA-B alleles. Our previous finding of proteasome independence for HLA-A0201 PPI presentation implied a noncanonical pathway to generate signal peptide–derived epitopes (6). This concept is extended in the current study, in which we highlight the requirement for intramembrane cleavage and indicate the importance of SPP in this role. We further show that after intramembrane cleavage, the pathway of peptide loading varies according to whether a signal peptide–derived epitope is N-terminal (requiring TAP) or COOH-terminal (independent of TAP). Loading is dictated by the HLA allele and optimized in the presence of ERAP1. These findings highlight a distinct set of processing principles for PPI (stoichiometry with translation and proximity to the site of HLA molecule synthesis and loading) that contrast with those in canonical endogenous antigen processing, which relies on proteasome degradation of effete or damaged cytoplasmic proteins. Together with evidence that TAP expression is upregulated in relevant tissues in the disease setting, this study offers important insight into the molecular processes that are a key underpinnings of interactions between cytotoxic CD8 T cells and β-cells during disease pathogenesis.

One caveat in our study relates to using tumor cells transfected with the INS gene and selected HLA alleles as “surrogate β-cells” to represent and understand the endogenous pathway of PPI presentation as it may occur in a human β-cell in vivo. Whether this approach biases epitope discovery toward a specific PPI region remains open to interpretation, although our finding of epitopes in the signal peptide, B chain, and C-peptide across the different HLA molecules suggests that this is unlikely. K562 cells predominantly express the constitutive proteasome (39), while some evidence indicates that the immunoproteasome can be upregulated in β-cells under inflammatory conditions (40), and the balance of these two could influence the spectrum of epitopes generated in human islets. Our approach is pragmatic in that obtaining sufficient, pure β-cells for such work presents a severe technical and logistical challenge. We therefore elected to conduct epitope discovery using “surrogate β-cells” and then confirm findings using human tissue. Our previous experience has been that epitopes identified in this way for HLA-A0201 and HLA-A2402 are faithful phenocopies of PPI presentation by β-cells. In both cases we were able to generate CD8 T-cell clones that recognize PPI epitopes presented by both the surrogate and donor-derived β-cells. Providing similar proof for the less common alleles remains challenging, however (41), although studies examining the frequency and antigen experience of, for example, PPI_5–12-B3906 and PPI_5–14-B3801 restricted T cells in the blood using peptide-HLA multimer technology, will provide supportive evidence of a pathogenic role.

One of our findings is the identification of a noncanonical pathway of endogenous antigen processing for an epitope relevant in the recognition of β-cells by the immune system. Central to this is an intramembrane cleavage step. Our studies conducted in vitro using microsomes indicate that SPP is capable of cleaving the PPI signal peptide, which is released from the ER membrane by SPP-catalyzed cleavage. CRISPR-Cas9-mediated SPP knockout indicates that this enzyme is also rate limiting for epitope generation in vivo. Using this and the (Z-LL)₂-ketone inhibitor approaches we are able to argue that SPP knockout affects the processing of signal peptide, leading to reduced presentation of PPI_15–24 and, in turn, reduced activation of the specific T-cell clone. Overall, this seems to be compelling evidence for a direct role of SPP in the processing of PPI signal peptide in a manner that is critical to the generation of this important epitope.

SPP belongs to the family of GxGD intramembrane proteases including presenilin/γ-secretase and related SPP-like proteases (37). SPP residing in the ER cleaves various signal peptides and type II membrane proteins in a number of physiological settings (42), including the generation of a regulatory peptide from the HLA-A0301 signal peptide that is required for HLA-E-mediated immunosurveillance (22). This is not the first study to draw attention to the potential requirement for intramembrane cleavage of an autoantigen in type 1 diabetes. Using algorithms to predict HLA-A0201 epitopes of islet amyloid polypeptide (IAPP), previous studies have shown CD8 T-cell reactivity to IAPP_9–17 and IAPP_5–13 in the blood (43,44) and, in the case of IAPP_5–13, also in islets of patients with type 1 diabetes (8). Whether these peptides are naturally processed and presented by β-cells remains unclear, however, and if confirmed for both it would imply a complex processing pathway, because the C-terminus of IAPP_5–13 overlaps and extends into the N-terminus of IAPP_9–17.

Signal sequences are essential for protein targeting to the secretory pathway via the ER (45,46), where, after insertion into the protein-conduction translocation channel, they are cleaved from the preprotein by signal peptidase (47); in the case of PPI_15–24-HLA-A0201, this also generates the epitope C-terminus. Signal peptides spanning the ER membrane require cleavage by SPP for efficient disposal (19,20). No consensus SPP cleavage motif has been identified, beyond a strong preference for helix-destabilizing residues in the membrane-spanning region (21,48,49), which are likely to be the proline and two alanine residues at positions 9, 12, and 15, respectively. Indeed, in our studies, mutation of any one of these has a mild effect on signal peptide cleavage; any two mutations together gives a moderate phenotype; and when all three are mutated, SPP is unable to cleave, indicating that these are the helix-destabilizing residues in PPI. Once liberated, the SPP-cleaved N-terminal signal peptide fragment gets access to the cytosol, where it is either trimmed by the proteasome or directly loaded onto TAP and the MHC peptide-loading complex. In the case of PPI_3–11-HLA-A2402, we show a requirement for TAP, whereas PPI_15–24-HLA-A0201 requires neither proteasome nor TAP (6), leading us to speculate that “untapped” loading of a COOH-terminal signal peptide fragment makes use of other chaperones, such as TAPBP.

We considered it important to try to link findings in relation to mechanisms of β-cell antigen presentation obtained in vitro with the tissue inflammatory process taking place in subjects with type 1 diabetes. Our analysis shows that TAP expression is significantly increased in ICIs and correlated in its hyperexpression with HLA class I molecule expression. That the β-cell contributes to its own destruction in various ways has become a much-vaunted metaphor. In the particular setting of PPI presentation to cytotoxic T cells, it would seem that hyperexpression of TAP is another example, certainly for HLA alleles such as HLA-A2402. In this way it may complement the effects of hyperexpression of HLA class I and hyperglycemic conditions (which upregulate PPI presentation [6]) in enhancing β-cell cytotoxicity under the inflammatory milieu that prevails in islets of Langerhans in patients with type 1 diabetes.

Acknowledgments. The authors are grateful to John Todd, University of Oxford, for providing the typed HLA class I cell lines and Dr. Pierre Vantourout, King’s College London, for providing the pLG2C vector.

Funding. This study was supported by the National Institute for Health Research Biomedical Research Centre at Guy’s and St Thomas’ Hospital Trusts and King’s College London (through a PhD studentship to D.K.-V.), JDRF (Centre Grant no. 1-2007-1803 to M.P. and a Career Development Award [5-CDA-2014-221-A-N] to S.J.R.), Diabetes UK (project grant 15/0005156 to N.G.M. and S.J.R.), and the Deutsche Forschungsgemeinschaft (project grant FOR2290-TP1 to M.K.L.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. D.K.-V. and M.E. designed and performed experiments, analyzed data, conceived ideas, oversaw research, and wrote the manuscript. M.A.R., A.d.R., B.H., and N.Y. performed experiments. P.A.v.V., S.J.R., N.G.M., M.K.L., and M.P. conceived ideas, oversaw research, and wrote the manuscript. M.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented at the Immunology of Diabetes 15th International Conference, San Francisco, CA, 19–23 January 2017.