Loss of immune tolerance to IL-2 in type 1 diabetes

Type 1 diabetes (T1D) is characterized by a chronic, progressive autoimmune attack against pancreas-specific antigens, effecting the destruction of insulin-producing β-cells. Here we show interleukin-2 (IL-2) is a non-pancreatic autoimmune target in T1D. Anti-IL-2 autoantibodies, as well as T cells specific for a single orthologous epitope of IL-2, are present in the peripheral blood of non-obese diabetic (NOD) mice and patients with T1D. In NOD mice, the generation of anti-IL-2 autoantibodies is genetically determined and their titre increases with age and disease onset. In T1D patients, circulating IgG memory B cells specific for IL-2 or insulin are present at similar frequencies. Anti-IL-2 autoantibodies cloned from T1D patients demonstrate clonality, a high degree of somatic hypermutation and nanomolar affinities, indicating a germinal centre origin and underscoring the synergy between cognate autoreactive T and B cells leading to defective immune tolerance.

A nti-cytokine antibodies have been reported in healthy individuals as well as in patients with infectious and autoimmune diseases, for example, anti-interferon (IFN)g antibodies in mycobacterial infections, anti-granulocyte-macrophage colony-stimulating factor (GM-CSF) antibodies in severe autoimmune pulmonary alveolar proteinosis, and anti-interleukin (IL)-17 antibodies in mucocutaneous candidiasis 1,2 . However, the stimuli eliciting anti-cytokine antibody responses, and whether these antibodies are pathologically causative in vivo, remains unknown. IL-2 is pleiotropic but indispensable for proper T reg cell function 3 , making it an attractive target in autoimmune disease. In type 1 diabetes (T1D), circulating, pancreas-specific autoantibodies are correlated with the onset of a chronic, pancreatic autoimmune attack leading to the progressive loss of insulin-producing b-cells 4 . Defects in the induction of central and peripheral tolerance checkpoints 5 are notable contributors to T1D pathology. Illustrating this point, non-obese diabetic (NOD) mice, which recapitulate many characteristics of the complex pathogenesis of human T1D 6 , and T1D patients develop both islet-specific autoantibodies and autoreactive T cells, and feature syntenic genetic linkage to disease 7 .

Results
We have previously shown that administration of low doses of recombinant human IL-2 (rhIL-2) at onset reverts disease in about half of the treated NOD mice 8 . To analyse if doses 10-, 20-, or 40-fold higher than what we have previously shown to be effective to revert hyperglycaemia in new onset diabetic NOD mice could increase treatment efficacy, we administered 2.5 Â 10 5 , 5 Â 10 5 , or 10 6 IU rhIL-2 to pre-diabetic NOD mice 9 . We observed that these higher rhIL-2 doses were, in a dosedependent manner: (i) lethally toxic in half of the mice; (ii) precipitated diabetes onset in around 25% of them; or intriguingly, (iii) induced no apparent clinical signs in around 25% of the mice, even after a 30-day administration (Fig. 1a-c). Interestingly, after 5 days of treatment, mice responded to all doses of administered rhIL-2 by increasing T reg cell frequencies, which returned to pre-treatment levels by day 30 after IL-2 administration (Fig. 1d). We reasoned that mice surviving 30 days post-high-dose rhIL-2 treatment may have developed antibodies capable of neutralizing the injected rhIL-2. Indeed, only sera from the surviving rhIL-2-treated NOD mice demonstrated high titres of rhIL-2 immunoglobulin-g IgG) as detected by enzyme-linked immunosorbent assay (ELISA) (Fig. 1e). Furthermore, those sera efficiently neutralized rhIL-2 biological activity in an in vitro assay using IL-2-dependent CTLL-2 cells (Fig. 1f), suggesting that they were responsible for the in vivo resistance to the side effects of high rhIL-2 doses. Interestingly, sera from untreated NOD mice, but not autoimmunity-resistant B6 mice, also showed detectable anti-rhIL-2 antibodies (Fig. 1e). These results suggested the existence of pre-formed antibodies capable of binding to rhIL-2, possibly representing naturally occurring, cross-reactive autoantibodies against murine IL-2. Indeed, only sera from untreated prediabetic and diabetic NOD mice, but not from B6 and BALB/c, reacted to mIL-2. Notably, IgG anti-mIL-2 autoantibody titres were significantly higher in overtly diabetic NOD mice as compared to their pre-diabetic counterparts (Fig. 2a). In NOD mice, IgG anti-mIL-2 autoantibodies were mostly of the IgG2b subclass (Fig. 2b). We confirmed the specificity of anti-mIL-2 autoantibodies by competitive binding to IL-2-coated beads ( Supplementary Fig. 1a,b, Supplementary Methods), and observed that anti-mIL-2 autoantibodies showed in vitro neutralizing activity, inhibiting CTLL-2 cell growth in a dose-dependent manner (Fig. 2c). Interestingly, anti-mIL-2 autoantibody titres increase with age and, consequently, with T1D progression (Fig. 2a,d). Moreover, NOD females generate higher titres of anti-mIL-2 autoantibodies than males of the same age, correlating with the higher frequency of spontaneous T1D incidence in females (Fig. 2e).
It has been shown that anti-insulin autoantibody (IAA) titres in 8-week-old mice are predictive of diabetes onset 10 . Strikingly, anti-mIL-2 autoantibody titres display a similar positive correlation with shorter time to T1D onset at just 6 weeks after birth (Fig. 2f). Thus, the spontaneously produced anti-IL-2 autoantibodies in young NOD mice could be used for predicting diabetes well before disease onset, either alone or in combination with IAA. In contrast to IAA titres, which can oscillate during disease progression in NOD mice 11 , anti-mIL-2 autoantibody titres appear at a stable trajectory preceding or concomitant with disease progression. This may be due to the cyclical appearance of insulin as an autoantigen during waves of pancreatic destruction, whereas IL-2 is more persistently present to drive a response.
T1D susceptibility and resistance alleles on mouse chromosome 3 (Idd3) correlate with differential expression of IL-2 (ref. 12). NOD mice carrying the Idd3 locus from B6 mice (NOD.Idd3 B6 ) are resistant to T1D development and their T cells produce two-fold more IL-2 than NOD mice. Conversely, IL-2-haploinsufficient NOD mice (NOD.IL-2 þ / À ) produce half as much IL-2 as NOD mice and have accelerated diabetes 12 . To study the effect of fluctuation in constitutive IL-2 production levels on the generation of anti-mIL-2 autoantibodies, we quantified anti-mIL-2 autoantibodies in the sera of these different NOD congenic strains. Female NOD.IL-2 þ / À mice had higher anti-mIL-2 autoantibody titres than NOD mice, and NOD.Idd3 B6 mice had the lowest levels (Fig. 2g). As controls for disease resistance independent of the Il2 locus, we used NOR mice, which represent a major histocompatibility complexmatched diabetes-resistant control strain for NOD mice that share the Idd3 NOD locus, but carry Idd5.2, Idd9/11 and Idd13 B6 protective loci, and NOD.Idd6 C3H congenic mice, which share the Idd3 locus but are less susceptible to T1D development 13 . In these two strains, although insulitis and diabetes are reduced or absent, anti-mIL-2 autoantibodies are present, indicating that, while their presence is associated with T1D development, they are not sufficient to induce T1D. The different congenic strains produce different amounts of IL-2 that could be complexed with circulating anti-mIL-2 autoantibodies and therefore undetected by a typical ELISA, which only detects free anti-mIL-2 autoantibody. The levels of free anti-mIL-2 autoantibody and IL-2/anti-mIL-2-autoantibody immune complexes follow similar distributions across the congenic strains (Fig. 2h), confirming that increased genetically determined levels of IL-2 correlate with lower anti-mIL-2 autoantibody production. These data support the hypothesis that reduced functional IL-2 levels are responsible for broken self-tolerance. Along these lines, injection of anti-mIL-2-autoantibody-depleted versus undepleted serum in NOD mice induced a reduction of CD25 expression on blood T reg cells (similar to the injection of equivalent amounts of a control anti-IL-2 neutralizing antibody (S4B6), suggesting that anti-mIL-2 autoantibodies could contribute to reduced IL-2 availability and impact T reg fitness in vivo ( Supplementary Fig. 2, Supplementary Methods). Interestingly, while IL-2-deficient mice are defective for thymic formation of a subset of islet-specific T reg cells 14 , we do not observe a correlation between Foxp3-positive T-cell numbers and anti-mIL-2 autoantibodies in NOD mice, suggesting this effect is peripheral, and not thymus-driven.
Long-lasting serum antibody production is mainly driven by long-lived plasma cells located in bone marrow and memory B cells resident in secondary lymphoid organs 15 . Indeed, the bone marrow of NOD, but not B6 mice, contained anti-IL2 IgG secreting cells (Fig. 3a), and mIL-2-specific memory B cells were present in the spleen of most NOD mice (Fig. 3b).
To evaluate the presence of IL-2-reactive T cells, we generated a library of 53 peptides (15-mers with 12 aa overlap, Supplementary Table 1) derived from mIL-2 and measured IFN-g production by B6 and NOD splenocytes upon in vitro stimulation. No IFN-g production was detected from B6 splenocytes (Fig. 3c), but cells from NOD mice specifically recognized two mIL-2-derived peptides (in bold in Fig. 3d). The mIL-2 10-24 epitope spans the signal peptide and may thus be preferentially processed and presented by the major histocompatibility complex Class I pathway, as described for the preproinsulin signal peptide in T1D 16 . The other peptide recognized, mIL-2 67-81 , likely represents an immunodominant target of IL-2 autoreactivity. As a control, we included the class II-restricted BDC2.5 mimotopic peptide (P31) 17 that induced IFN-g production only in NOD mice and to similar levels as those induced by the IL-2-derived peptides. These experiments establish the existence of IL-2-specific auto-reactive B and T cells in NOD mice.
Extrapolating the discovery of IL-2 as a novel autoantigen in mouse models of T1D to humans, we found that a significantly higher fraction of T1D patient sera (from three independent cohorts) contain anti-rhIL-2 autoantibodies than either healthy donors or type 2 diabetic patients (who present chronic hyperglycaemia in the absence of islet autoimmunity) ( Fig. 4a and Supplementary Tables 2 and 3). Interestingly, as in the NOD mice, there was a positive correlation between anti-rhIL-2 autoantibody titres and patient age ( Supplementary Fig. 3). The reduced penetrance of anti-rhIL-2 autoantibodies in T1D patients relative to the NOD mice may be attributed to the genetic and temporal heterogeneity of the human disease, such that only a subset of diabetic patients are phenotypically similar to the NOD model with respect to anti-IL-2 autoantibodies. To resolve the composition of the anti-IL-2 IgG autoantibodies, we analysed the IgG subclasses in the sera of T1D patients for anti-IL-2 specificity, and found that most samples contain IgG1 and IgG2 anti-rhIL-2 autoantibodies (Fig. 4b). To estimate the EC 50 of the anti-rhIL-2 autoantibody, we purified IgG from the sera of T1D patients. Purified IgG samples with higher affinity for IL-2 showed a   (f) Proliferation of CTLL-2 cells cultured for 3 days with 3 IU ml À 1 rhIL-2 and serially diluted serum from B6 (closed circles) or NOD mice treated for 30 days with high-dose rhIL-2 (open circles). Proliferation is expressed as percentage of control (CTLL-2 cultured for 3 days with 3 IU ml À 1 rhIL-2 without mouse serum). Data are cumulative of at least two independent experiments. ns, not significant. ***Po0.001 (non-parametric Mann-Whitney test).
To understand the ontogeny of anti-hIL-2 autoantibodies in type 1 diabetics, we isolated circulating human IgG memory B cells from four T1D patients and clonally expanded them in vitro at very low density (1-4 cells/well) to drive their differentiation into antibody-secreting cells 18 . After 12 days, we analysed culture supernatants for the presence of anti-influenza, anti-insulin, and anti-IL-2 IgG antibodies. Similar frequencies of autoreactive anti-insulin and anti-IL-2 IgG memory B cells were present (albeit at lower frequencies than for, for example, influenza virus as a stereotypical recall antigen), indicating the existence of a specific, persistent immune response against IL-2 in T1D (Fig. 5a). Heavy and light chain sequences of anti-IL-2specific BCRs (B cell receptor) were cloned from three independently expanded B cells from a single patient. Sequence analysis revealed a clonal origin (V H 3-30/V K 1-39) of all three hits, as well as extensive somatic hypermutation from the germline sequence (Fig. 5b). Recombinantly produced antibodies were found to have an affinity for rhIL-2 between 6 and 32 nM (Fig. 5b). Taken together, this evidence indicates an affinity matured, germinal centre origin for anti-rhIL-2 autoantibodies.
To assess T-cell reactivity to rhIL-2 by IFN-g ELISPOT following in vitro antigen stimulation 19 , we generated a library of 30 peptides covering the entire human IL-2 sequence and included four peptides covering regions comprising the two mutations introduced in therapeutic recombinant rhIL-2 (Proleukin) relative to native IL-2 (hIL-2) (Supplementary Table 4). Relative to age-matched healthy donors, a higher frequency of T1D patients displays T-cell reactivity against hIL-2, with two significantly enriched peptides (in bold in Fig. 5c). The T1D-associated intracellular insulinoma-associated antigen 2 (IA-2) included as a control was targeted by T cells at the same frequency in T1D patients (38%) as the hIL-2 56-70 peptide; of the five T1D patients responding to IA-2, three also responded to hIL-2 56-70 . Intriguingly, this hIL-2 56-70 peptide (LTRMLTFKFYMPKKA) overlaps with the orthologous mIL-2 67-81 (NLKLPRMLTFKFYLP) peptide found targeted in NOD mice ( Supplementary Fig. 4), suggesting that this epitope is immunodominant across species.
This concerted dual T-and B-cell adaptive autoimmune response strongly supports the hypothesis that tolerance to IL-2 is broken in T1D, demonstrating uniquely that autoimmunity is also directed against non-islet antigens in T1D. The physiopathology of T1D has previously been linked to genetic defects in the IL-2/IL-2R pathway that indirectly affect immune tolerance, partly by disrupting T reg cell function 12,20,21 . Adaptive autoimmune responses against IL-2 add another piece to this puzzle, further reinforcing the link between impaired IL-2 bioavailability and T1D. Interestingly, significantly increased percentages of sera from rheumatoid arthritis (RA), Sjögren syndrome, systemic lupus erythematosus and autoimmune myositis patients were also anti-rhIL-2-autoantibodies þ , and anti-mIL-2 autoantibodies were found in lupus-prone mice (Fig. 6); indicating that IL-2 autoreactivity plays a previously unsuspected role in the pathogenesis of different autoimmune conditions in human and mouse.

Discussion
Steady-state levels of IL-2 are too low to stimulate effector T and NK cells but are critical for the maintenance of T reg cells 3 . Thus, a reduction in peripheral IL-2 bioavailability may interfere with T reg homeostasis and function, and lead to broken immune tolerance. This effect is well illustrated by the administration of neutralizing anti-IL-2 antibodies, which triggers autoimmune gastritis in BALB/c mice and accelerates T1D in NOD mice. The effect in NOD mice is accompanied by numerous autoimmune presentations, including gastritis, thyroiditis, sialadenitis, and neuropathy. Of note, short-term low-dose IL-2 administration induces variable T reg cell expansion in T1D 22,23 and autoimmune vasculitis patients 24 . Thus, it would be interesting to verify whether the de novo presence or treatment-induced production of anti-rhIL-2 autoantibodies underlies these variable responses.
While the highly significant association between free soluble anti-rhIL-2 autoantibodies and T1D is remarkable, the functional capacity of these antibodies-at the monoclonal level-to neutralize or otherwise modulate IL-2 function remains to be assessed. Such antibodies could be useful diagnostic, experimental or therapeutic agents in the field of autoimmune disease. Intriguingly, the presence of low levels of IL-2-complexed anti-rhIL-2 IgG in healthy individuals has been described some time ago 25 ; however, the origin and function of these Ab have not been addressed in the years since. In light of our results, it will be especially interesting to revisit this topic and determine the prevalence of these autoantibodies in a healthy population, as well as whether their functionality differentiates them from T1D-derived anti-rhIL-2 autoantibodies. With the proper tools now in hand to characterize the BCR repertoire underlying these autoreactive antibodies, we can return to long-standing questions with a modern perspective and insight into clinically relevant aspects of human health.
CTLL-2-based neutralization assays. CTLL-2 cells (ATCC, mycoplasma-free) were cultured (10 4 cells per well) in 96-well plates in complete Roswell Park Memorial Institute (RPMI) medium (Gibco) containing no mIL-2, 1 ng ml À 1 mIL-2 or 3 IU ml À 1 rhIL-2 with or without heat-inactivated ( Murine T-cell studies. A peptide library of 15-mers overlapping by 12 amino acids covering the whole sequence of mIL-2 (including the signal peptide) was generated (GL-Biochem). Peptides (10 mM) were stored in dimethylsulfoxide (DMSO) at À 20°C until use. For initial screening, peptides were divided in 17 pools of 6 peptides/each with an overlap of 3 so that the final concentration of each peptide was 3 or 10 mM. A first screening using pools of six peptides (B6: n ¼ 4,   NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13027 ARTICLE NOD: n ¼ 4) allowed the identification of two potentially immunogenic regions. In a different set of mice, pooled peptides that showed a significant response in NOD compared to B6 mice were single-tested at a concentration of 3 and 10 mM. Splenocytes from 10-to 18-week-old female B6 or NOD mice were cultured in triplicate (4 Â 10 5 cells/150 ml per well) in X-Vivo 15 serum-free medium (Lonza) containing DMSO (negative control), class II-restricted BDC2.5 mimotope peptide (P31; YVRPLWVRME; 3-10 mmol l À 1 ), aCD3-CD28 beads (positive control, ratio 1bead:1cell, Life Technologies) or peptides. After 72 h, 50 ml of supernatant was saved for the analysis of cytokine production. IFN-g in culture supernatants were measured with IFN-g cytometric beads assay (CBA) Flex Set (BD Biosciences) according to the manufacturer's instructions. Quantification of anti-human IL-2 autoantibodies. Serum titres of anti-rhIL-2autoantibodies were assessed by ELISA. Microtitre 96-well plates (Medisorp, Nunc) were incubated overnight at 4°C with 100 ml per well of carbonate coating buffer containing 10 5 IU ml À 1 rhIL-2 ('IL-2 coated wells') or buffer alone ('uncoated wells', blank). After blocking with PBS/2% BSA for 2 h, plates were incubated with 50 ml serially diluted serum samples for 2 h at room temperature. After extensive washing with PBS/0.1% Tween 20, HRP-conjugated anti-human IgG (1:2,000; Dako) was added to each well and the plates were kept at room temperature for 1 h. Peroxidase activity was measured with TMB substrate as before. Standard curve was generated using two-fold serial dilutions of rat anti-human IL-2 (clone MQ1-17H12, eBioscience) revealed with an HRP-conjugated goat antirat Ig. Arbitrary Units for each sample were calculated using the O.D. value obtained after subtraction of the blank. For the IL-2 competition ELISA, purified IgG from T1D patient sera (at a constant concentration of 50 nM) were incubated in 384-well microtitre plates (Maxisorp, Nunc), coated overnight with 10 mg ml À 1 Proleukin or 1 mg ml À 1 each H1, H3, and B influenza variants and blocked with 0.5% w/v BSA, in competition with decreasing concentrations of soluble Proleukin (from 9 mM to 5 pM in 3-fold dilution series). Non-competed anti-IL-2 or antiinfluenza antibodies were detected with HRP-conjugated Goat anti-human IgG (SouthernBiotech). Human B-and T-cell cultures. PBMCs were separated by Ficoll-Paque PLUS centrifugation (GE Healthcare) from T1D patients and buffy coats obtained from samples from healthy blood donors (Saint Antoine-Crozatier Blood Bank, Paris). Circulating total human B cells were enriched with EasySep kit Stem cell technology (STEMCELL tech. 19054) followed by isolation of class switch memory B cells (CD20 þ IgG þ CD27 þ ) by cell sorting with FACSAria II cytometer using the following antibodies: CD27-v450, IgD-PE (clone M-7271 BD), CD20-Alexa488 (Clone 2H7 Biolegend), CD3-CD14-CD16-CD56-PE-Cy5 (Clone UCHT1, RMO52, 3G8, N901 Beckmann Coulter) and IgA-IgM-AlexaFluor647 (Jackson ImmunoResearch). Purified class-switched memory B cells were plated at 4 cells per well in 384-well plates and stimulated based on previously published protocols 18 . After 12 days, supernatants from human B-cell cultures were collected and analysed by ELISA for the presence of anti-IL-2, insulin and influenza IgG antibodies.
To investigate human T-cell reactivity against IL-2, a library of 15-mer peptides with 10 amino acid overlaps covering the whole native hIL-2 was used. We further included four peptides covering regions comprising the two mutations introduced in therapeutic recombinant rhIL-2 (Proleukin) relative to native IL-2, namely an alanine-methionine substitution at the N-terminus and a cysteine-serine replacement at position 145 (125 in the Proleukin sequence). Proleukin sequences were synthesized (GL-Biochem; Supplementary Table 4). Intracellular IA-2 (10 mg ml À 1 final concentration; amino acids 214-591, endotoxin 5 EU mg À 1 ; kindly provided by J.F. Elliott, University of Alberta, Edmonton, Canada), adenoviral (AdV) lysate and phytohemagglutinin (PHA) were included as positive controls. Frozen-thawed PBMCs from T1D and healthy subjects recruited in Paris were processed as described 19,34 . PBMCs (10 6 per well in 96-well flat-bottom plates) were stimulated for 48 h with peptides using the accelerated co-cultured dendritic cell (acDC) technology 19 . At the end of the 48 h culture, PBMCs were washed and 2 Â 10 5 cells per well distributed in triplicate wells of 96-well polyvinylidene difluoride ELISPOT plates coated with an IFN-g antibody. After 6 h of incubation, plates were developed as described 35 , counted on a BioSys 5000 Pro-SF Bioreader and results expressed as IFN-g spot-forming cells (SFC)/10 6 PBMCs after background subtraction. For initial screening, peptides were divided in pools of 3/each (2/each for paired native hIL-2 and Proleukin sequences) and used at a final concentration of 5 mM per each. T-cell responses were scored as positive for SFC counts 46 s.d. above spontaneous background responses in the presence of DMSO diluent alone. Individual peptides from positive pools were further tested (10 mM final concentration) and scored as positive for IFN-g SFC numbers 46.5/ 10 6 PBMCs after background subtraction (which was o10 SFC/10 6 PBMCs in all cases). Cut-offs were selected according to receiver-operator characteristic (ROC) analyses as described 35 .
Anti-IgG ELISA was used to determine which wells contained B cells that responded to the stimuli; the response efficiency (20-35% of all seeded B cells) was used to determine the number of BCRs under investigation for each of the singleantigen ELISAs (influenza, IL-2, or insulin). RNA from cell lysates of wells positive for anti-IL-2 IgG (but neither of the other antigens nor a BSA negative control) was reverse transcribed and this cDNA was used to clone the respective heavy and light chain genes in each well into mammalian expression vectors, whereby the V H and most of the C H 1 domain of the heavy chain, and the V K or V L domain were grafted onto backbones containing the remaining IgG1, IgK, or IgL C domains, respectively. After eliminating PCR errors, unique identified sequences were re-expressed as heavy/light chain pairs to find the correct sequence pairing reproducing the antibody in the cell culture supernatant. These sequences were input into the IgBLAST database to identify V and J segment usage, as well as quantify somatic hypermutation relative to the closest germline alignment.
Statistical analyses. For diabetes incidence experiments, the number of mice per group was calculated based on our experience 8,36 . For serum transfer experiments, NOD mice were randomized with GraphPad QuickCalcs (www.graphpad.com/ quickcalcs/) to receive NOD serum, anti-mIL-2-autoantibodies-depleted NOD serum or S4B6. No randomization method was used for the other experiments performed. Investigators were not blinded throughout the whole study. Correlations were performed using a non-parametric Spearman correlation test. As sample distribution throughout the manuscript was not normal (as determined by a D'Agostino and Pearson omnibus normality test), differences between separate groups or related groups were analysed using a two-sided Mann-Whitney test or a two-sided Wilcoxon matched-pairs signed rank test, respectively, with Po0.05 taken as statistical significance. We used the nonparametric Kruskal-Wallis procedure followed by the Dunn's multiple comparisons test to evaluate the difference in the anti-mIL-2 autoantibody titres among NOD mice according to their age and disease status. When comparing the percentage of anti-rhIL-2-autoantibodies þ patients within two different groups or the percentage of subjects responding to rhIL-2 peptides in two different groups, we used a Fisher's exact test, with Po0.05 taken as statistical significance. For ELISA tests quantifying anti-rhIL-2 autoantibodies, we fixed the threshold of positivity at a value of 24.3AU, which allowed discrimination of healthy donors and T1D subjects with 95% specificity. This cut-off was calculated with an ROC curve with a 95% confidence interval, using the T1D subjects of cohorts 1, 2 and 3 (n ¼ 75), as patients; and the healthy donors coming from these cohorts (n ¼ 103), as controls. This cut-off value was applied for all the samples. All statistical analyses were performed using GraphPad Prism v6 software.
Data availability. The data that support the findings of this study are available from the corresponding author on request. The authors declare that all other data supporting the findings of this study are available within the article and its supplementary information files.