2024 Diabetes Volume 68, October 2019

Early Detection of Peripheral Blood Cell Signature in Children Developing b-Cell Autoimmunity at a Young Age

Henna Kallionpää,1 Juhi Somani,2 Soile Tuomela,1 Ubaid Ullah,1 Rafael de Albuquerque,1 Tapio Lönnberg,1 Elina Komsi,1 Heli Siljander,3,4 Jarno Honkanen,3,4 Taina Härkönen,3,4 Aleksandr Peet,5,6 Vallo Tillmann,5,6 Vikash Chandra,3,7 Mahesh Kumar Anagandula,8 Gun Frisk,8 Timo Otonkoski,3,7 Omid Rasool,1 Riikka Lund,1 Harri Lähdesmäki,2 Mikael Knip,3,4,9,10 and Riitta Lahesmaa1

Diabetes 2019;68:2024–2034 | https://doi.org/10.2337/db19-0287

The appearance of type 1 diabetes (T1D)-associated function before T1D and suggest a potential role for IL32 autoantibodies is the first and only measurable param- in the pathogenesis of T1D. eter to predict progression toward T1D in genetically susceptible individuals. However, autoantibodies indi- cate an active autoimmune reaction, wherein the im- Family and sibling studies in type 1 diabetes (T1D) have mune tolerance is already broken. Therefore, there is implicated a firm genetic predisposition to a locus con- a clear and urgent need for new biomarkers that predict taining HLA class I and class II on the onset of the autoimmune reaction preceding auto- 6 suggesting a role for CD4+ as well as CD8+ T cells in T1D fl antibody positivity or re ect progressive b-cell destruc- pathogenesis (1–3). As much as 30–50% of the genetic risk – tion. Here we report the mRNA sequencing based is conferred by HLA class II molecules, which are crucial in analysis of 306 samples including fractionated samples antigen presentation to CD4+ T cells. Further, CD4+ cells of CD4+ and CD8+ T cells as well as CD42CD82 cell reactive to b-cell antigen peptides are found in peripheral fractions and unfractionated peripheral blood mono- blood and the pancreas and typically secrete the nuclear cell samples longitudinally collected from g + seven children who developed b-cell autoimmunity IFN (4,5). CD4 cells orchestrate adaptive immune responses, including that of antibody-secreting B cells as (case subjects) at a young age and matched control + subjects. We identified transcripts, including well as cytotoxic CD8 T cells. Indeed, circulating autoanti- 32 (IL32), that were upregulated before T1D-associated bodies against b-cell antigens may appear years before the + autoantibodies appeared. Single-cell RNA sequencing clinical onset. Further, a cytolytic CD4 subtype might studies revealed that high IL32 in case samples was directly contribute to target cell killing (6).

GENETICS/GENOMES/PROTEOMICS/METABOLOMICS contributed mainly by activated T cells and NK cells. Although HLA class II is associated with the develop- Further, we showed that IL32 expression can be in- ment of autoantibodies, HLA class I seems to be more duced by a virus and in pancreatic islets strongly linked to disease progression (7). Histological and b-cells, respectively. The results provide a basis analysis of pancreatic sections of cadaveric donors with for early detection of aberrations in the T1D revealed that HLA class I is highly expressed in islets

1Turku Bioscience Centre, University of Turku and Åbo Akademi University, Turku, 10Tampere Center for Child Health Research, Tampere University Hospital, Finland Tampere, Finland 2 Department of Computer Science, Aalto University School of Science, Espoo, Corresponding author: Riitta Lahesmaa, riitta.lahesmaa@utu.fi Finland Received 17 April 2019 and accepted 10 July 2019 3Children’s Hospital, University of Helsinki, and Helsinki University Hospital, Helsinki, Finland This article contains Supplementary Data online at http://diabetes 4Research Programs Unit, Diabetes and Obesity, University of Helsinki, Helsinki, .diabetesjournals.org/lookup/suppl/doi:10.2337/db19-0287/-/DC1. Finland H.K., J.S., S.T., and U.U. contributed equally to this work. 5Department of Pediatrics, University of Tartu, Tartu, Estonia H.K., M.K., and R.L. share senior authorship. 6Children’s Clinic of Tartu, Tartu University Hospital, Tartu, Estonia 7Research Programs Unit, Molecular Neurology and Biomedicum Stem Cell Centre, © 2019 by the American Diabetes Association. Readers may use this article as fi Faculty of Medicine, University of Helsinki, Helsinki, Finland long as the work is properly cited, the use is educational and not for pro t, and the 8Department of Immunology, Genetics and Pathology, Uppsala University, Sweden work is not altered. More information is available at http://www.diabetesjournals 9Folkhälsan Research Center, Helsinki, Finland .org/content/license. diabetes.diabetesjournals.org Kallionpää and Associates 2025

(8,9). Moreover, CD8+ cells are the most abundant cell type Sample Collections during insulitis (10), and the islets contain CD8+ cells At each study visit, 8 mL blood was drawn in sodium- specific for T1D autoantigens (11). Thus, the autoimmune heparin tubes (368480, Vacutainer; BD Biosciences). Pe- cascade in T1D might be initiated by self-reactive CD4+ ripheral blood mononuclear cells (PBMCs) were isolated by cells that activate B cells to produce autoantibodies that Ficoll-Paque centrifugation (17-1440-03; GE Healthcare) target the b-cells and unleash the cytotoxic activity of the and were suspended in RPMI-1640 medium (42401-018; autoreactive CD8+ cells. The environmental factors trig- Gibco) supplemented with 10% DMSO (cat. no. 0231, gering and driving the autoimmunity in T1D are poorly 500 mL, Thermo Fisher Scientific), 5% Human AB Serum defined, but the disease has been associated with viral (cat. no. IPLA-SERAB-OTC; Innovative Research), 2 (12), diet in early childhood (13), and reduced mmol/L L-glutamine (G7513; Sigma-Aldrich), and 25 mmol/L diversity of gut microbiota (14). gentamicin (G-1397; Sigma-Aldrich). After overnight Currently, the appearance of T1D-associated autoanti- incubation at 280°C, samples were stored in liquid nitro- bodies is the first and only measurable parameter to gen (2180°C). For fractionation, PBMC samples were predict progression toward T1D in genetically susceptible thawed quickly in a 37°C water bath and quantitated individuals. Although the disease progression rate varies for cell numbers and viability. On average, 90% of cells considerably, children with genetic HLA risk expressing at were viable. Magnetic antibody-coupled beads were used least two T1D autoantibodies will very likely progress to for sequential positive enrichment of CD4+ and CD8+ cells clinical disease during the next 15 years (15). However, (11331D and 11333D; Invitrogen). RNA was isolated autoantibodies are poor prognostic markers for the timing from the samples with an AllPrep (80224; QIAGEN), of the clinical presentation of T1D. The appearance of and quantity and quality were determined using a Qubit autoantibodies indicates an active autoimmune reaction, RNA assay (Q32852; Invitrogen) and Bioanalyzer 2100 wherein the immune tolerance is already broken. There- (Agilent), respectively. fore, there is a clear and urgent need for new biomarkers that predict the onset of the autoimmune reaction pre- Bulk RNA Sequencing of PBMCs and Other Fractions ceding autoantibody positivity or reflect progressive b-cell At least 80 ng total RNA was processed for RNA sequenc- destruction. Such markers would present a window for ing (RNA-seq) with the TruSeq Stranded mRNA Library early intervention aimed at complete disease prevention. Prep kit (RS-122-2101; Illumina). The sequencing was Previously, we reported changes in whole-blood transcripts carried out with the Illumina HiSeq2500 instrument using and serum before the detection of diabetes- TruSeq v3 (2 3 100 base pairs [bp] chemistry). The average associated antibodies in children who later progressed to T1D sequencing depth was ;51 million reads. Quality control (16,17). Therefore, we hypothesized that a comprehensive was performed using FastQC (version 0.10.0). All the analysis of the transcriptome of longitudinal cellular sam- samples passed the quality criteria. The reads were aligned ples including CD4+ and CD8+ T cells will lead to the to the human reference transcriptome, GRCh37 assembly identification of new early biomarkers. version 75, using TopHat (version 2.0.10) (20). Average mapping percentage was 93. The concordant pairs per- RESEARCH DESIGN AND METHODS centage was ;89. The aligned reads were counted with Study Cohort htseq-count (HTSEq. 0.6.1; overlap mode of “intersection- Samples were collected as part of the DIABIMMUNE study strict”) (21). The read counts of genes were normalized from Finnish (n = 10) and Estonian (n = 4) participants using the trimmed means of the M values (TMM imple- (Supplementary Table 1). The HLA-DR-DQ genotypes were mented in edgeR [22]). Coding, noncoding information analyzed as previously described (18). A total of 836 children were taken from Ensembl. Differential expression analyses with HLA-DR-DQ risk allele were monitored and sampled at were conducted separately for coding and noncoding 3, 6, 12, 18, 24, and 36 months of age. The study protocols genes, using edgeR (22). The variance of the data was were approved by the ethics committees of the participating estimated using the trended dispersion method. A further hospitals, and the parents gave written informed consent. filtering step retained only those genes as differentially Autoantibodies against insulin (IAA), glutamic acid decar- expressed (DE) that had |median log2 fold change (FC)| boxylase (GADA), islet antigen-2 (IA-2A), and zinc trans- .0.5 and had .65% samples across all individuals regu- porter 8 (ZnT8A) were measured from serum with specific lated in the same direction (i.e., up- or downregulated). radiobinding assays (19). Islet cell antibodies (ICAs) were These filtering steps were added to discard false positives analyzed with immunofluorescence in autoantibody-positive that may arise due to the heterogeneity of the samples subjects. The cutoff values were based on the 99th per- resulting from normal variation, which is unrelated to T1D centile in children without diabetes, which were 2.80 as well as to discard the outliers. A flowchart of the scheme relative units (RU) for IAA, 5.36 RU for GADA, 0.78 RU for of analysis has been shown in Supplementary Fig. 1. IA-2A, and 0.61 RU for ZnT8A. The detection limit in the ICA assay was 2.5 Juvenile Diabetes Foundation units Single-Cell RNA-seq (JDFU). A sample was considered seropositive when any The concentrations of the PBMC samples varied from 0.55 of the autoantibodies exceeded the thresholds. to 1.80 3 106 cells/mL. From each sample, we aimed at the 2026 Early Signs of T1D Autoimmune Reaction Diabetes Volume 68, October 2019 recovery of 5,000 single cells, loading ;9,000 cells on the GAPDH. The amplification was monitored with the Quant- Chromium Controller using Single Cell 39 Solution v2 Studio 12K Flex Real-Time PCR System under the following reagents and following the manufacturer’s instructions PCR conditions: 10 min at 95°C, followed by 40 cycles of (CG00052, Rev B; 10x Genomics). Single-cell RNA-seq 15 s at 95°C and 60 s at 60°C, and analysis with QuantStudio (scRNA-seq) sample processing was carried out in three software on Thermo Fisher Cloud. batches on consecutive days using the same lot of reagents For EndoC-bH1 cell data, cDNA was synthesized using and chips for all samples. The cDNA was further amplified the Maxima First Strand cDNA Synthesis Kit (Thermo using a Veriti Thermal Cycler (Applied Biosystems/Thermo Fisher Scientific) as per the manufacturer’s recommenda- Fisher Scientific), followed by cleanup (SPRIselect kit; tions. All reactions were performed in duplicates on at Beckman Coulter). Finally, enzymatic fragmentation, end least three biological replicates. Cyclophilin-A was used as repair, A-tailing, adaptor ligation, and PCR were performed an endogenous control. Primer sequences are presented in to produce indexed libraries, which were sequenced with Supplementary Table 2. Illumina HiSeq 3000 (one sample per lane) using paired end sequencing and 26 + 98 bp read-length configuration. ELISA The data were processed using the Cell Ranger pipeline, To measure secreted IL32 levels, we used IL-32 DuoSet version 2.0.0, yielding on average 2,546 viable cells per ELISA (DY3040-05 and DY008; R&D Systems) following sample and 114,309 reads per cell. the manufacturer’s instructions. The reads were aligned to the human reference genome (hg19) using STAR (23). The mean raw reads per cell varied Intracellular Staining and Flow Cytometry from 57,000 to 200,000. Quality control analysis and The cells were fixed for 10 min in Fix Buffer I (557870; BD further exploration were done using Seurat (24). After Biosciences), followed by 45 min permeabilization using fi ltering steps, 18,396 cells expressing 20,830 genes were ice-cold permeabilization buffer III (558050; BD Bioscien- fi retained. For details on the ltering steps please see Sup- ces). The cells were stained using allophycocyanin- ’ plementary Data. The data were normalized using Seurat s conjugated IL32⍺ antibody (IC30402A; R&D Systems) default. Highly variable genes were selected for principal and FITC-conjugated IFNg antibody (MHCIFG01; Invitro- component analysis. The top 20 PCs were used in the graph- gen) in PBS containing 0.5% FCS. The data were acquired based clustering. To identify marker genes for each cluster, in BD Fortessa and analyzed using FlowJo (version 10.4.2). cells of a single cluster were compared with the cells of all other clusters combined. A was considered a marker of EndoC-bH1 Cell Culture a cluster if it was expressed in at least 25% of the cells of The EndoC-bH1 human b-cell line was obtained from either of the two groups and the logFC between the cluster Univercell Biosolution S.A.S., Toulouse, France. The cells and all other clusters was at least 0.25. were cultured as previously described (26). EndoC-bH1 For trajectory analysis, the pooled cells were ordered in cells were stimulated with IL32g either alone (100 ng/mL; pseudotime (i.e., placed along a trajectory corresponding R&D Systems) or in combination with a cocktail of IL1b with a type of biological transition, such as differentiation) (5 ng/mL; R&D Systems) and IFNg (50 ng/mL; R&D using Monocle 2 (25). The analysis was performed on cells Systems) for 24 h. RNA samples were collected at the fi + + speci cally from CD4 and CD8 T-cell clusters. For the end of each treatment and analyzed by RT-qPCR. details on the trajectory analyses, please see Supplemen- tary Data. Human CD4 T-Cell Isolation and Culturing CD4+ T cells were isolated from cord blood collected from RT-PCR Analysis neonates born in Turku University Hospital and were For PBMC samples, 50 ng total RNA was treated with cultured in Iscove’s modified Dulbecco’s medium contain- DNaseI (Invitrogen), and cDNA was synthesized with the ing 1% AB serum in absence (Th0) or presence (Th1) of Transcriptor First Strand cDNA Synthesis Kit (Roche Life 2.5 ng/mL IL12 (R&D Systems). Cells were activated with Science). For isoform-specific(IL32a, b,andg) assay, plate-bound CD3 (0.5 mg/well of a 24 well-plate) and quantitative PCR (qPCR) analysis was performed in tripli- soluble CD28 (0.5 mg/mL), both from Immunotech, cate runs using SYBR Select Master Mix (Applied Biosys- with or without 50 ng/mL rIL32g (R&D Systems). + tems). DCt values was calculated relative to EF1a. For CD4 12 ng/mL IL2 was added at 48 h. For IFNg neutralization, T cells and pancreatic islets, RNA was isolated using the anti-IFNg antibody (MAB285, 10 mg/mL; R&D Systems) RNeasy Mini Kit (74106; QIAGEN) and RNeasy Plus Mini was used. For reactivation, cells were treated with 5 ng/mL Kit (74134; QIAGEN), respectively. Purified RNA was trea- phorbol myristic acid (Calbiochem) and 0.5 pg/mL iono- ted with DNaseI and cDNA was synthesized with Super- mycin (Sigma-Aldrich) for 5 h. Script II Reverse Transcriptase (18064014; Invitrogen). For the detection of global IL32, qPCR reactions were Human Pancreatic Islets and Their With run using a custom TaqMan Assay reagent Coxsackie B Virus (AJ5IQA9; Thermo Fisher Scientific) in duplicate and in Human islets were isolated from pancreases obtained from two separate runs. DCt values were calculated relative to brain dead organ donors and purified by handpicking to diabetes.diabetesjournals.org Kallionpää and Associates 2027 apurityof.90%. Islet culturing and virus infection with referred to as a fraction henceforth), 889, 399, and 1,002 Coxsackie B virus-1 (CBV-1-7-10796 [CBV-1-7]) was per- genes were DE specifically in CD4+ (e.g., CD28, CTLA4), 2 2 formed as previously described (27). Islets were collected at CD8+ (e.g., CD8A, CD8B, KLRK1), and CD4 CD8 (e.g., the day 4 time point, and RNA was extracted using the IL1A, IL1B, IL6) fractions, respectively (Fig. 1C and Sup- RNeasy Plus Mini Kit or the AllPrep DNA/RNA Mini Kit plementary Table 3). CD4+ and CD8+ fractions shared (QIAGEN). For RNA-seq, 100 ng total RNA from three 1,815 DE genes, of which 1,803 genes (99%) were con- donors was used for library preparation according to Illu- cordant (either up or down in both fractions) (Supple- mina TruSeq RNA Sample Preparation v2 Guide (part no. mentary Fig. 2C and Supplementary Table 3). In summary, 15026495). The high quality of the libraries was confirmed fractionation of the PBMC population based on the T-cell with the Agilent Bioanalyzer 2100 and Qubit Fluorometric phenotype allowed improved detection of DE genes and Quantitation (Life Technologies). The libraries were pooled enabled identification of cell subset–specific gene expres- in two pools and run in two lanes on the Illumina HiSeq sion signatures. 2500 instrument using 2 3 100 bp. RNA-seq Analysis Identifies Transcriptomic Changes Data and Resource Availability Associated With b-Cell Autoimmunity All the raw data will be deposited to European Genome- Comparison of case samples with their respective controls phenome Archive for access. The study does not involve identified 51, 69, 143, and 85 genes as DE (false discovery 2 2 any noncommerical reagents or tools. rate ,0.05) in CD4+,CD8+,CD4 CD8 ,andPBMCfrac- tions, respectively (Supplementary Table 4), with a total of RESULTS 278 unique DE genes in one or more fractions (Fig. 2A). Six + + Fractionation of PBMC Sample Into CD4 , CD8 , and genes, AMICA1, BTN3A2, IL32, RPSAP15, RPSAP58,and 2 2 CD4 CD8 Cellular Subsets Reveals Distinct and WASH7P, were upregulated in the case subjects in all four Overlapping Gene Expression Signatures fractions (Fig. 2A). Only 16% of the DE genes have pre- We performed RNA-seq of 306 longitudinal samples in- viously been reported as DE in genetically susceptible cluding unfractionated PBMCs, as well as CD4-enriched + + children with prediabetes, using microarrays (16,28,29) (CD4 ), CD8-enriched (CD8 ), and CD4 and CD8 cell– – fi 2 2 or RT-PCR (30 32), con rming dysregulation of these genes depleted (CD4 CD8 ) cell fractions from seven case-control in children progressing to T1D. Besides -coding pairs (Table 1). The seven case children who developed + genes, 54 noncoding genes, including 3 antisense, 2 sense T1D-related autoantibodies (Aab ) were selected from the intronic, 7 enhancer, and 18 promoter-associated lncRNAs, DIABIMMUNE birth cohort (18), where HLA-susceptible were DE. To our knowledge, none of these lncRNAs have – A children are sampled at 3 36 months of age (Fig. 1 ). All been linked to the etiology of T1D (16,28–32). seven children developed T1D-associated autoantibodies by the age of 2 years (Table 1), and four of them developed Hierarchical Clustering Identifies Coregulated Gene clinical T1D between the ages of 2.4 and 3.7 years. For each Expression Clusters Associated With T1D case subject, an autoantibody-negative control child was Autoimmunity matched for sex, date and place of birth, and HLA- Gene- and sample-wise hierarchical clustering for each cell conferred risk category. fraction, including PBMCs, identified a cluster upregulated The samples clustered according to the cell fraction (Fig. in the case samples in all four fractions (Fig. 2B and 1B), and the clustering was not affected by case-control Supplementary Fig. 3A–D). Interestingly, this cluster con- status or sampling age, indicating that cell fraction– sistently contained IL32 and BTN3A2, along with other specific differences dominated over variation derived from fraction-specific genes (Fig. 2C). In the CD8+ fraction, other factors (Supplementary Fig. 2A and B). When CD4+, expression of a distinct cluster, including IFNG, was lower 2 2 CD8+,andCD4 CD8 samples from control subjects were in most of the case samples than in control samples compared with the unfractionated PBMC samples (also (Supplementary Fig. 3B). Surprisingly, in the PBMC

Table 1—Summary of the case and control children sampled at the age of 3–36 months Case no. Sex Seroconversion age* First autoantibodies Age at T1D diagnosis Matched control no. Case 1 Female 12 months IAA, GADA 3.2 years Control 1 Case 2 Male 12 months IAA — Control 2 Case 3 Male 18 months IAA, ICA 3.7 years Control 3 Case 5 Female 24 months IAA, IA-2A, ZnT8A, ICA 2.6 years Control 5 Case 9 Male 18 months IAA, GADA, ICA — Control 9 Case 10 Male 12 months IAA, GADA — Control 10.1, control 10.2 Case 11 Female 18 months GADA 2.4 years Control 11 For further details, see Supplementary Table 1. *First detection of T1D-associated autoantibodies. 2028 Early Signs of T1D Autoimmune Reaction Diabetes Volume 68, October 2019

Figure 1—Fractionation of PBMC sample into CD4+, CD8+, and CD42CD82 cellular subsets reveals distinct and overlapping gene expression signatures. A: Outline of the sample collection and cell fractionation. B: t-SNE (t-distributed stochastic neighbor embedding) visualization of the log2-transformed expression data (without any filtering steps) colored according to cell fraction information. C: Number of DE genes when CD4+, CD8+, and CD42CD82 fractionated samples were compared with their original PBMC aliquots. The functionally important fraction-specific upregulated genes are highlighted in red. Analysis was restricted to healthy control subjects only. For the gene lists, see Supplementary Table 3.

fraction, we detected case-specific upregulation of a cluster, profiles in CD4+ samples (Fig. 2D). In at least two of including insulin (INS), glucagon (CGC), and regulin 1a four fractions, this cluster also comprised TRBV4-1, (REG1A), transcripts (Supplementary Fig. 3D), which are TMEM14C, UROS, WASH7P, BTN3A3, CARD8, CCDC167, predominantly expressed in the pancreas. and LINC01184. The profile of these and other interesting To explicitly define coregulated genes in these clusters, genes is shown in Supplementary Fig. 4. Upon examination we calculated Euclidean distances for IL32 (in each frac- of the overrepresented transcription factor binding sites tion), IFNG (in CD8+ fraction), and INS (in PBMC fraction) on the promoters of IL32 cluster genes, the V$IK_Q5_01 and considered the genes with a median Euclidean dis- motif bound by Ikaros (IKZF1) was revealed to be among tance ,2.5 across all case-control pairs to be cocluster- the enriched transcription factor binding sites shared in ing with the gene of interest (Supplementary Table 5). both the CD4+ and PBMC fractions (Supplementary Table In three of the four fractions, the IL32 cluster included 5). IKZF1 has been genetically associated with T1D (33). BTN3A2, AMICA1, LARS, and RSU1 (Fig. 2C). IL32, The T1D-associated risk allele rs10272724 (T) increases AMICA1, and BNT3A2 show concerted gene expression IKZF1 transcript level (34). diabetes.diabetesjournals.org Kallionpää and Associates 2029

Figure 2—RNA-seq analysis identifies transcriptomic changes associated with b-cell autoimmunity. A: Number and overlap of DE genes between case and control subjects identified in cell fractions analyzed. Genes shared between all four fractions are highlighted. B: Heat map + of the genes DE in CD4 T cells between the case and control subjects. Values are presented as log2FC (truncated between [22, 2]) between each case-control pair at each time point (3–36 months) and standardized to the mean of each gene. Genes coregulated with IL32 (,2.5 Euclidean distance) are marked with red box and text. Sample collection time with respect to (w.r.t) seroconversion, sample pairing information, and clinical status have been indicated with colors on top of the heat map. “Before/After SC” informs whether the case sample was collected before or after seroconversion. “Pair Info” provides the case-control pair information. The “SC / T1D” annotation indicates whether the case subject has progressed to clinical T1D diagnosis (T1D) or not (SC). C: Number and overlap of IL32 coclustered genes in indicated cell fractions. Genes regulated at least in two fractions are highlighted. D: Profiles of IL32, AMICA1, and BNT3A2 in CD4+ samples, presented in log2 RPKM (reads per kilo base per million mapped reads) scale. For individual profiles, see Supplementary Fig. 4. The case- control pairs are grouped according to the diagnosis of the case subjects. T1D, case subject has been diagnosed with clinical T1D; SC, case has seroconverted to autoantibody positivity.

IFNG cluster of the CD8+ cells included TBX21 (codes functions with very similar expression profiles (Supple- for TBET), BHLHE40, and ZEB2, transcription factors mentary Fig. 4M–T). expressed in CD8+ T cells (35), as well as NKG7, OASL, Higher IL32 expression in case subjects was validated and KLRD1 (Supplementary Table 5). ZEB2 has been using qRT-PCR. Interestingly, genes encoding all three reported to drive terminal effector CD8+ cell differentia- major isoforms (IL32 a, b,andg) were upregulated in tion together with T-bet (36). In the PBMC fraction, GCG PBMC samples in all the case children at each of the time and REG1A were coregulated with INS (Supplementary points, including 3 months (Fig. 3A and Supplementary Fig. Table 4 and Supplementary Fig. 5). 7). Among these isoforms, the gene encoding IL32g was expressed at the highest level, followed by IL32b and IL32a. Transcriptional Changes Preceding the Appearance of T1D-Related Autoantibodies Are Enriched in the CD8+ scRNA-seq Identifies T and NK Cells as the IL32-High T-Cell Fraction Population To identify changes that occur immediately before the first To specify the cell populations responsible for the IL32 and detection of T1D-related autoantibodies (i.e., seroconver- INS signatures, we performed scRNA-seq on four selected sion), we performed a separate differential expression case and their nearest matched control PBMC samples analysis for the samples drawn at most 12 months before where the expression of IL32 or INS was high (or low) seroconversion. Altogether, 121 coding and noncoding based on the bulk RNA-seq data (Supplementary Table 6). genes were DE in case subjects compared with matched Unsupervised clustering of 18,396 single cells from all control subjects (Supplementary Table 4 and Supplemen- eight PBMC scRNA-seq runs identified 13 clusters (Fig. tary Fig. 6). Notably, more than half of these (58%) were 3B and Supplementary Fig. 8). The 2 largest clusters detected only in the CD8+ fraction. Besides IL32, only two expressing high CCR7 were merged as one cluster of naive other genes were common to all fractions, RPSAP58 and T cells, reducing the number of clusters to 12. Clusters RPSAP15—both being the pseudogenes with unknown named as “RGCC+ T cells,”“CD62L+ T cells,” and “activated 2030 Early Signs of T1D Autoimmune Reaction Diabetes Volume 68, October 2019

Figure 3—scRNA-seq of PBMCs identifies T and NK cells as IL32 high populations. A: Expression of IL32g isoform in longitudinal PBMC samples of case subjects and their control subjects (n = 7 + 7), assayed by qRT-PCR (for a and b isoforms, see Supplementary Fig. 7). B: t-SNE (t-distributed stochastic neighbor embedding) clusters from the pooled data from all scRNA-seq samples (4 case and 4 control subjects [in total 18,396 cells]). Clusters are named according to the expression of classical marker genes, such as CD8A (for details and marker gene list, see Supplementary Fig. 8; for contribution of each sample per cluster, refer to Supplementary Figs. 9 and 10). C: Expression of IL32 in the 12 cell clusters (natural logarithm transformation with addition of 1). For case-control comparison, please see Supplementary Fig. 11. D–F: Trajectories emerging when using the data from CD4+ cells and the precursor cells. G-I: Trajectories emerging when using the data from CD8+ and the precursor cells. Here, “precursor cells” refer to cells from the naive and RGCC+ T-cell clusters. For the trajectory analysis of all the cells from all clusters as well as the breakdown of each individual cluster, see Supplementary Fig. 12. In D and G, cells are colored based on the contributions from different t-SNE clusters. In E and H, cells are colored by case (orange) or control (gray) status. In F and I, cells are colored by the intensity of IL32 expression (log10 transformation with addition of 0.1). Act., activated; prolif., proliferating.

Th cells” expressed lower levels of CCR7. Activated CD8+ cells expressing CD14 or FCGR3A, LYZ, and TYROBP. In- T cells cluster expressed high levels of CD8A and CD8B as terestingly, the expression of many HLA class II molecules well as NKG7, and two separate clusters of CD8+ T cells was as high in B cells as in , suggesting high expressing either granulysin or granzyme A were observed antigen-presentation potential. (“activated GNLY+ CD8+ T cells” and “activated GZMA+ The contribution of different case or control samples to CD8+ T cells,” respectively). A subcluster of activated the cells in a given cellular population (cluster) varied from GZMA+ CD8+ cells had higher expression of cell-cycle genes cluster to cluster (Supplementary Figs. 9 and 10A and B). (e.g., STMN1, TUBA1B) and was named “activated pro- The naive T-cell cluster was dominated by the cells from the liferating GZMA+ CD8+ T cells.” An NK cell cluster was control samples (P , 0.05) whereas the /DC positive for expression of CD56, NKG7, and GNLY and cluster had more cells from case subjects (P , 0.005) negative for CD8A and CD3E. A B-cell cluster was identified (Supplementary Fig. 10B). Case 9, with the highest IL32 by the expression of MS4A1, CD79A, and CD79B, whereas expression levels in the bulk RNA-seq data, dominated the the monocyte/dendritic cells (DC) cluster was composed of CD62L+ T-cell cluster, activated NK cell cluster, and, most diabetes.diabetesjournals.org Kallionpää and Associates 2031

Figure 4—Virus- and cytokine-induced IL32 expression by pancreatic b-cells. A: Representative FACS dot plots showing IFNg and IL32 double staining in Th0 and Th1 polarized CD4+ cells. Staining controls and two other replicates are shown in Supplementary Fig. 13A. Percent IL32-positive cells as well as median fluorescence intensity (MFI) data (mean 6 SD) from all the three replicates are shown in B and C, respectively. Statistical significance was determined by paired two-tailed t test. D: IL32 in culture supernatant as measured by ELISA. Cells were cultured in Th0/1 condition for 72 h in the presence (+) or absence (2) of anti-IFNg. The expression plotted is relative to Th0 (2). Statistical significance was determined by paired two-tailed t test. E: IL32 expression in nonpolarized Th0 cells and cells differentiated to Th1 for 72 h in the presence (+) or absence (2) of IL32g as measured by the Taqman assay. The expression is calculated relative to EEF1A. Statistical significance was determined by unpaired two-tailed t test. F: IL32 secretion in culture supernatant as measured by ELISA. Cells were cultured in Th0/1 condition for 7 days in the presence (+) or absence (2) of IL32g, followed by washing and restimulation by phorbol myristic acid and ionomycin for 48 h. The expression plotted is relative to Th0 (2). Statistical significance was determined by paired two-tailed t test. G: Expression of the TNFA and IL6 or IL8 and IL32 genes when the EndoC-bH1 cells were stimulated with IL32g alone or in combination with other inflammatory cytokines for 24 h. The fold change is calculated compared with nontreated cells. The results shown here are from four independent biological replicates (mean 6 SD). Statistical significance for the effects on IL32 expression was determined by paired two- tailed t test. ns, not significant. H: IL32 expression as measured in an RNA-seq experiment where pancreatic islets were infected with CBV1-7. Statistical significance was determined by edgeR. I: IL32 expression in virus-infected pancreatic islets as measured by RT-qPCR Taqman assay. The expression is calculated as 2^2(dCt). The statistical significance is determined by paired two-tailed t test. RPKM, reads per kilo base per million mapped reads. *P , 0.05; **P , 0.01; ***false discovery rate ,0.001.

clearly, activated and proliferating GZMA+ CD8+ Tcell the control samples were enriched (Fig. 3E and H and clusters (Supplementary Fig. 10B). Conversely, control chil- Supplementary Fig. 12). In contrast, the highest levels of dren 5 and 9 seemed to dominate the cluster of developing IL32 were expressed by cells close to the end points of T cells expressing pre–T-cell receptor PTCRA,suggestingthe branches II and III, corresponding to more advanced stages presence of immature T cells in those samples. of differentiation (Fig. 3F and I and Supplementary Fig. 12). Insulin, glucagon, and REG1A expression was not detected even in the INS-high samples of cases 5 and 9, IL32 and IFNg Are Coexpressed by Th1 Cells leaving the origin of these transcripts in bulk RNA-seq as To further study IL32 expression, we measured intracel- an open question. In contrast, IL32 expression was clear, lular IL32 expression at the protein level in CD4+ T cells and as expected, it was explicitly overexpressed in the case isolated from human umbilical cord blood. Cells were samples (Supplementary Fig. 11). IL32 was expressed at either activated through CD3/CD28 in the absence of a very low level in monocyte/DC, , and developing cytokine (Th0) or were differentiated toward a Th1 cell clusters; however, it was expressed at higher levels by lineage for 72 h. IL32 was induced upon activation and, both the T cells and the NK cells (Fig. 3C). unlike IFNg, was expressed in both Th0 and Th1 cells (Fig. To further define the relationship of IL32 expression and 4A). Interestingly, in Th1 cells, most IFNg-producing cells T-cell activation status, we performed separate trajectory were also positive for IL32 (Fig. 4A and Supplementary Fig. analyses for the CD4+ and CD8+ T cells. The less activated 13A) and the proportions of IL32-positive cells and the per precursor populations (naive and RGCC+ T cells), which cell IL32 levels were higher in IFNg-producing Th1 cells detect CD4 and CD8 encoding transcripts in low abundances, than in Th0 cells (Fig. 4B and C). Furthermore, neutral- were used as a starting point for the trajectory analyses. ization of IFNg significantly reduced IL32 secretion by Th1 The results revealed three major cellular branches (I–III) in cells (Fig. 4D), confirming that IFNg positively regulates the data in both CD4+ and CD8+ T cells (Fig. 3D–I). Branch IL32 expression. IL32 expression was also induced by IL32 I consisted mainly of naive T cells, among which cells from itself in Th1 cells, both at the RNA level (Fig. 4E) and in the 2032 Early Signs of T1D Autoimmune Reaction Diabetes Volume 68, October 2019 culture supernatant upon 48-h restimulation after 7 days specifically in the CD8+ cells. Further studies are needed to of polarization in Th1 condition (Fig. 4F). understand whether at-risk children have defects in formu- lating effector CD8+ response or their effector CD8+ cells Pancreatic b-Cells Can Express IL32 in Response to have homed to the sites of inflammation in the pancreas. Cytokine Stimulation and Viral Infection We selected IL32 as our candidate for functional studies fl To study how the elevated IL32 levels may in uence b-cell because it has not been linked to seroconversion before, it function, we treated human EndoC-bH1 b-cell line for is easy to measure with available assays from clinical 24 h with recombinant IL32g either alone or in combina- samples, and as a secreted molecule it can potentially fl tion with the proin ammatory cytokines IL1b and IFNg. affect the function of several cell types in paracrine and In agreement with earlier published data on pancreatic systemic fashion. Increased expression of IL32 in case fi ductal cell lines (37), IL1b and IFNg signi cantly subjects across many cell types before seroconversion IL32 induced expression in human EndoC-bH1 cells (Fig. suggest that IL32 is a critical member of the immunological G 4 ). However, addition of IL32g did not further enhance signature characteristic for children developing b-cell 1 - IL32 2 ) the IL1b- and IFNg induced expression; ) the autoimmunity. fl TNFA, IL6 IL8 expression of in ammatory cytokines , and IL32 is expressed by many immune and epithelial cells G 3 (Fig. 4 ); or ) the expression of ER stress marker genes and has been described to be proinflammatory (45). How- ATF3 ATF4 ATF6 HSPA5 CHOP sXBP1 ( , , , , , and ) (Supple- ever, to our knowledge, it has not been associated with B mentary Fig. 13 ) in EndoC-bH1 cells. Furthermore, the human b-cell autoimmunity. In contrast, IL32 is down- – IL32g treatment did not affect the expression of b-cell regulated in CD4+ T cells from recently diagnosed adult fi INS MAFA PDX1 speci c genes, such as , , and (Supplemen- T1D patients (46), which, along with our findings, suggests C tary Fig. 13 ). These results suggest that, while IL32 does dynamic changes in immune during the not appear to directly affect the survival or the differen- pathogenesis of the disease. On the other hand, IL32 tiation status of the b-cells, b-cells actively contribute to overexpression was observed in synovial biopsies of fl in ammation in the islets by secreting IL32 upon stimu- patients with rheumatoid arthritis (47), in inflamed mu- lation by cytokines. cosa of inflammatory bowel disease patients (48), and in Coxsackie B viruses are b-cell trophic viruses that have the serum of myasthenia gravis patients (49), indicating – been linked to the development of T1D (38 43). To study a connection between IL32 and autoimmunity in general. IL32 the possible trigger of expression in b-cells, we In T cells, IL32 is induced by T-cell activation, and it fi infected puri ed human pancreatic islets of three cadaveric modulates human CD4+ T-cell effector function by pro- donors with CBV1-7 strain. Infection by the virus led to moting Th1 and Th17 responses (50). Both Th1 and Th17 IL32 H the induction of expression in the islets (Fig. 4 ). cells have been linked to T1D pathogenesis in both human fi We further validated this nding in the three islet sam- and mouse (51). The IL32 gene has been identified only in ples used for RNA-seq as well as one additional islet higher , excluding rodents. Nonetheless, human sample using qRT-PCR assays and found a consistent IL32g transgenic mice exhibit impaired glucose tolerance IL32 increase in the expression upon CBV1-7 infection and increased levels of IFNg and other proinflammatory I (Fig. 4 ). Taken together, these results suggest that upon cytokines in the pancreas, as well as accelerated streptozotocin- H I a viral infection (Fig. 4 and )oracytokinerush(Fig. induced experimental T1D (52). No specific cell-surface H 4 ), b-cells may upregulate IL32 secretion, contributing receptor for IL32 has been identified, but it may act fl to in ammation. through cell-surface integrins or proteinase-3 (53). Our results showed that IL32 was often coregulated DISCUSSION with genes previously linked to autoimmunity. For exam- We identified a panel of novel molecular players detected ple, the BTN3 gene cluster resides in the extended MHC early in children who developed T1D-associated autoanti- class I locus. Further, BTN3 genes have been associated bodies or even the clinical disease at a young age. Since the with T1D in a genetic screen, especially in the case of immunological changes related to T1D are known to be BTN3A2 (54). AMICA1 is a plasma strongest among the T1D cases diagnosed at an early age involved in lymphocyte migration through its interaction (44), focusing on this age-group should enhance the possi- with Coxsackie-adenovirus receptor (CAR) expressed in bility of detecting aberrations in the immune system pre- epithelial cells and has been associated with multiple disposing to the disease. In this study, unbiased RNA-seq sclerosis (55). An analogous scenario could be envisaged of CD4+ and CD8+ cells revealed many T1D-associated DE for T1D: CAR is expressed by the pancreatic islet cells, transcripts not previously reported. Analysis of the PBMC including b-cells (42), and its expression is elevated in population offers an excellent overview of stable gene ex- autoantibody-positive individuals and patients with T1D pression patterns but, at the same time, appears to mask (56), suggesting that it might help recruit T cells to the some of the subtle fraction-specificchanges.Suchchanges islets. Interestingly, the findings point to human-specific included upregulation of CD52 detected only in the CD4+ cell phenomena not detectable in mouse models, as IL32 and fraction and downregulation of the IFNG and associated the BTN3 are not encoded by the mouse transcription factors ZEB2, TBX21,andZNF683 detected genome. diabetes.diabetesjournals.org Kallionpää and Associates 2033

The strength of our study is that the children studied responsible for experiments on virus-infected pancreatic islets. R.Lu. was re- here comprise a homogeneous population with the early sponsible for study design, cell fractionation, sample analysis, and data pro- appearance of T1D-associated autoantibodies. Increasing duction. H.L. was responsible for computational data analysis, interpretation of the evidence suggests that T1D can be subdivided into differ- results, editing the manuscript, and supervising J.S. M.K. was responsible for the DIABIMMUNE study design, sample collection, sample storage, clinical information ent phenotypes, e.g., characterized by age-dependent B-cell fi for the children, directing of the clinical study, interpreting the results, and editing in ltration in the pancreas (57), defect in Coxsackievirus- the manuscript. R.La. was responsible for study design, sample and data analysis, induced antibody response in children with early insulin interpretation of the results, writing the manuscript, and supervision of the study. autoimmunity (58), or rapid versus slow progression to All authors contributed to the final version of the manuscript. H.L. and R.La. are the clinical disease (59). Thus, our results may not apply to guarantors of this work and, as such, had full access to all the data in the study and “late progressors,” adolescents, and adults. Although the take responsibility for the integrity of the data and the accuracy of the data analysis of the global transcriptome of T-cell subsets of analysis. children with prediabetes over the period of seroconver- sion is unique, a limitation of the current study is the References analysis of only seven Aab+ children. The results of this 1. Todd JA, Bell JI, McDevitt HO. HLA-DQ beta gene contributes to suscep- study need to be validated and expanded in a larger cohort tibility and resistance to insulin-dependent diabetes mellitus. Nature 1987;329: of children with prediabetes but serve as a starting point 599–604 for better understanding of immunological changes pre- 2. Nejentsev S, Howson JM, Walker NM, et al.; Wellcome Trust Case Control Consortium. Localization of type 1 diabetes susceptibility to the MHC class I genes ceding the clinical onset of the disease. In the future, we – fi HLA-B and HLA-A. Nature 2007;450:887 892 are interested in addressing whether our ndings on – fl 3. Todd JA. Etiology of type 1 diabetes. Immunity 2010;32:457 467 a cellular level are re ected also in IL32 levels in plasma 4. Babon JA, DeNicola ME, Blodgett DM, et al. Analysis of self-antigen as well as studying whether IL32 alone or in combination specificity of islet-infiltrating T cells from human donors with type 1 diabetes. Nat with other identified molecules would have sufficient Med 2016;22:1482–1487 sensitivity and specificity as an early indicator for T1D. 5. Delong T, Wiles TA, Baker RL, et al. Pathogenic CD4 T cells in type 1 diabetes recognize epitopes formed by peptide fusion. Science 2016;351:711–714 6. Takeuchi A, Saito T. CD4 CTL, a cytotoxic subset of CD4+ T cells, their Acknowledgments. The authors are grateful to the families for their differentiation and function. Front Immunol 2017;8:194 participation in the DIABIMMUNE study. The DIABIMMUNE study group is ac- 7. Lipponen K, Gombos Z, Kiviniemi M, et al. Effect of HLA class I and class II knowledged for excellent collaborations, work with the families, and collection of alleles on progression from autoantibody positivity to overt type 1 diabetes in the samples for the study. Marjo Hakkarainen, Sarita Heinonen, Päivi Junni, and children with risk-associated class II genotypes. Diabetes 2010;59:3253–3256 Elina Louramo (Turku Bioscience Centre, University of Turku and Åbo Akademi 8. Foulis AK, Farquharson MA, Hardman R. Aberrant expression of class II major University, Turku, Finland) are acknowledged for skillful assistance in the histocompatibility complex molecules by B cells and hyperexpression of class I laboratory. Next-generation sequencing was performed at the Finnish Functional major histocompatibility complex molecules by insulin containing islets in type Genomics Centre (FFGC), Turku, Finland, part of the Biocenter Finland network. 1 (insulin-dependent) diabetes mellitus. Diabetologia 1987;30:333–343 The authors thank Satu Mustjoki and her team at University of Helsinki for advice in 9. Richardson SJ, Rodriguez-Calvo T, Gerling IC, et al. Islet cell hyperexpression designing scRNA-seq experiments and Riina Kaukonen at the FFGC for the sample of HLA class I antigens: a defining feature in type 1 diabetes. Diabetologia 2016; preparation. 59:2448–2458 Funding. This work was financially supported by the JDRF; the Academy of 10. Willcox A, Richardson SJ, Bone AJ, Foulis AK, Morgan NG. Analysis of islet Finland (AoF) Centre of Excellence in Molecular Systems Immunology and inflammation in human type 1 diabetes. Clin Exp Immunol 2009;155:173–181 Physiology Research (SyMMyS) 2012–2017 (grant no. 250114); the AoF Person- 11. Coppieters KT, Dotta F, Amirian N, et al. Demonstration of islet-autoreactive alized Medicine Program (grant no. 292482); AoF grants 294337, 292335, CD8 T cells in insulitic lesions from recent onset and long-term type 1 diabetes 319280, and 314444; the Sigrid Jusélius Foundation; the Diabetes Research patients. J Exp Med 2012;209:51–60 Foundation (Diabetestutkimussäätiö); the Novo Nordisk Foundation Innovative 12. Rodriguez-Calvo T, Sabouri S, Anquetil F, von Herrath MG. The viral paradigm Medicines Initiative 2 Joint Undertaking under grant agreement no. 115797 in type 1 diabetes: who are the main suspects? Autoimmun Rev 2016;15:964–969 (INNODIA). This Joint Undertaking receives support from the Union’s Horizon 13. Virtanen SM. Dietary factors in the development of type 1 diabetes. Pediatr 2020 research and innovation programme and EFPIA, JDRF, and The Leona M. and Diabetes 2016;17(Suppl. 22):49–55 Harry B. Helmsley Charitable Trust. The DIABIMMUNE study was supported by the 14. Knip M, Siljander H. The role of the intestinal microbiota in type 1 diabetes European Union Seventh Framework Programme (grant no. 202063). T.L. was mellitus. Nat Rev Endocrinol 2016;12:154–167 supported by the AoF (311081). 15. Ziegler AG, Rewers M, Simell O, et al. Seroconversion to multiple islet Duality of Interest. No potential conflicts of interest relevant to this article autoantibodies and risk of progression to diabetes in children. JAMA 2013;309: were reported. 2473–2479 Author Contributions. H.K., S.T., and U.U. were responsible for the 16. Kallionpää H, Elo LL, Laajala E, et al. Innate immune activity is detected prior interpretation of the results. J.S. conducted bioinformatic analyses. H.K., J.S., S.T., to seroconversion in children with HLA-conferred type 1 diabetes susceptibility. and U.U. drafted the manuscript. H.K., J.S., and U.U. prepared the figures. H.K. Diabetes 2014;63:2402–2414 was responsible for supervising E.K. R.d.A., E.K., and O.R. were responsible for the 17. Moulder R, Bhosale SD, Erkkilä T, et al. Serum proteomes distinguish isoform-specific IL32 RT-PCR assay and the intracellular IL32 staining in T cells children developing type 1 diabetes in a cohort with HLA-conferred susceptibility. and interpretation of the results. T.L. provided expertise in scRNA-seq study Diabetes 2015;64:2265–2278 design, sample and data analysis, and interpretation of the results. H.S., J.H., T.H., 18. Peet A, Kool P, Ilonen J, Knip M, Tillmann V; DIABIMMUNE Study Group. Birth A.P., and V.T. were responsible for sample collection, sample storage, and further weight in newborn infants with different diabetes-associated HLA genotypes in clinical information of the children. V.C. and T.O. carried out the experiments and three neighbouring countries: Finland, Estonia and Russian Karelia. Diabetes interpreted the results of the studies in pancreatic b-cells. M.K.A. and G.F. were Metab Res Rev 2012;28:455–461 2034 Early Signs of T1D Autoimmune Reaction Diabetes Volume 68, October 2019

19. Cianciaruso C, Phelps EA, Pasquier M, et al. Primary human and rat b-cells 38. Dotta F, Censini S, van Halteren AG, et al. Coxsackie B4 virus infection of beta release the intracellular autoantigens GAD65, IA-2, and proinsulin in exosomes cells and insulitis in recent-onset type 1 diabetic patients. Proc together with cytokine-induced enhancers of immunity. Diabetes 2017;66:460–473 Natl Acad Sci U S A 2007;104:5115–5120 20. Kim D, Pertea G, Trapnell C, Pimentel H, Kelley R, Salzberg SL. TopHat2: 39. Krogvold L, Edwin B, Buanes T, et al. Detection of a low-grade enteroviral accurate alignment of transcriptomes in the presence of insertions, deletions and infection in the islets of Langerhans of living patients newly diagnosed with type gene fusions. Genome Biol 2013;14:R36 1 diabetes. Diabetes 2015;64:1682–1687 21. Anders S, Pyl PT, Huber W. HTSeq–a Python framework to work with high- 40. Laitinen OH, Honkanen H, Pakkanen O, et al. Coxsackievirus B1 is associated throughput sequencing data. Bioinformatics 2015;31:166–169 with induction of b-cell autoimmunity that portends type 1 diabetes. Diabetes 22. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for 2014;63:446–455 differential expression analysis of digital gene expression data. Bioinformatics 41. Richardson SJ, Willcox A, Bone AJ, Foulis AK, Morgan NG. The prevalence of 2010;26:139–140 enteroviral capsid protein vp1 immunostaining in pancreatic islets in human type 23. Dobin A, Davis CA, Schlesinger F, et al. STAR: ultrafast universal RNA-seq 1 diabetes. Diabetologia 2009;52:1143–1151 aligner. Bioinformatics 2013;29:15–21 42. Ylipaasto P, Klingel K, Lindberg AM, et al. Enterovirus infection in human 24. Butler A, Hoffman P, Smibert P, Papalexi E, Satija R. Integrating single-cell pancreatic islet cells, islet tropism in vivo and receptor involvement in cultured islet transcriptomic data across different conditions, technologies, and species. Nat beta cells. Diabetologia 2004;47:225–239 Biotechnol 2018;36:411–420 43. Oikarinen S, Tauriainen S, Hober D, et al.; VirDiab Study Group. Virus antibody 25. Qiu X, Mao Q, Tang Y, et al. Reversed graph embedding resolves complex survey in different European populations indicates risk association between single-cell trajectories. Nat Methods 2017;14:979–982 coxsackievirus B1 and type 1 diabetes. Diabetes 2014;63:655–662 26. Ravassard P, Hazhouz Y, Pechberty S, et al. A genetically engineered human 44. Shields BM, McDonald TJ, Oram R, et al.; TIGI Consortium. C-peptide decline pancreatic b cell line exhibiting glucose-inducible insulin secretion. J Clin Invest in type 1 diabetes has two phases: an initial exponential fall and a subsequent 2011;121:3589–3597 stable phase. Diabetes Care 2018;41:1486–1492 27. Anagandula M, Richardson SJ, Oberste MS, et al. Infection of human islets of 45. Kim SH, Han SY, Azam T, Yoon DY, Dinarello CA. Interleukin-32: a cytokine Langerhans with two strains of Coxsackie B virus serotype 1: assessment of virus and inducer of TNFalpha. Immunity 2005;22:131–142 replication, degree of cell death and induction of genes involved in the innate 46. Orban T, Kis J, Szereday L, et al. Reduced CD4+ T-cell-specific gene ex- immunity pathway. J Med Virol 2014;86:1402–1411 pression in human type 1 diabetes mellitus. J Autoimmun 2007;28:177–187 28. Reynier F, Pachot A, Paye M, et al. Specific gene expression signature 47. Joosten LA, Netea MG, Kim SH, et al. IL-32, a proinflammatory cytokine in associated with development of autoimmune type-I diabetes using whole-blood rheumatoid arthritis. Proc Natl Acad Sci USA 2006;103:3298–3303 microarray analysis. Genes Immun 2010;11:269–278 48. Shioya M, Nishida A, Yagi Y, et al. Epithelial overexpression of interleukin- 29. Ferreira RC, Guo H, Coulson RM, et al. A type I transcriptional 32alpha in inflammatory bowel disease. Clin Exp Immunol 2007;149:480–486 signature precedes autoimmunity in children genetically at risk for type 1 diabetes. 49. Na SJ, So SH, Lee KO, Choi YC. Elevated serum level of interleukin-32a in the Diabetes 2014;63:2538–2550 patients with myasthenia gravis. J Neurol 2011;258:1865–1870 30. Jin Y, Sharma A, Bai S, et al. Risk of type 1 diabetes progression in islet 50. Jung MY, Son MH, Kim SH, Cho D, Kim TS. IL-32g induces the maturation of autoantibody-positive children can be further stratified using expression patterns dendritic cells with Th1- and Th17-polarizing ability through enhanced IL-12 and of multiple genes implicated in peripheral blood lymphocyte activation and IL-6 production. J Immunol 2011;186:6848–6859 function. Diabetes 2014;63:2506–2515 51. Walker LS, von Herrath M. CD4 T cell differentiation in type 1 diabetes. Clin 31. Reinert-Hartwall L, Honkanen J, Salo HM, et al.; DIABIMMUNE Study Group; Exp Immunol 2016;183:16–29 DIABIMMUNE Study Group. Th1/Th17 plasticity is a marker of advanced b cell 52. Jhun H, Choi J, Hong J, et al. IL-32g overexpression accelerates strepto- autoimmunity and impaired glucose tolerance in humans. J Immunol 2015;194: zotocin (STZ)-induced type 1 diabetes. Cytokine 2014;69:1–5 68–75 53. Xin T, Chen M, Duan L, Xu Y, Gao P. Interleukin-32: its role in asthma and 32. Heninger AK, Eugster A, Kuehn D, et al. A divergent population of potential as a therapeutic agent. Respir Res 2018;19:124 autoantigen-responsive CD4+ T cells in infants prior to b cell autoimmunity. Sci 54. Viken MK, Blomhoff A, Olsson M, et al. Reproducible association with type Transl Med 2017;9:eaaf8848 1 diabetes in the extended class I region of the major histocompatibility complex. 33. Swafford AD, Howson JM, Davison LJ, et al. An allele of IKZF1 (Ikaros) Genes Immun 2009;10:323–333 conferring susceptibility to childhood acute lymphoblastic leukemia protects 55. Alvarez JI, Kébir H, Cheslow L, et al. JAML mediates monocyte and CD8 T cell against type 1 diabetes. Diabetes 2011;60:1041–1044 migration across the brain endothelium. Ann Clin Transl Neurol 2015;2:1032–1037 34. Ram R, Mehta M, Nguyen QT, et al. Systematic evaluation of genes and 56. Hodik M, Anagandula M, Fuxe J, et al.; POD-V Consortium. Coxsackie- genetic variants associated with type 1 diabetes susceptibility. J Immunol 2016; adenovirus receptor expression is enhanced in pancreas from patients with type 196:3043–3053 1 diabetes. BMJ Open Diabetes Res Care 2016;4:e000219 35. Arsenio J, Kakaradov B, Metz PJ, Kim SH, Yeo GW, Chang JT. Early 57. Leete P, Willcox A, Krogvold L, et al. Differential insulitic profiles determine specification of CD8+ T lymphocyte fates during adaptive immunity revealed by the extent of b-cell destruction and the age at onset of type 1 diabetes. Diabetes single-cell gene-expression analyses. Nat Immunol 2014;15:365–372 2016;65:1362–1369 36. Dominguez CX, Amezquita RA, Guan T, et al. The transcription factors ZEB2 58. von Toerne C, Laimighofer M, Achenbach P, et al. Peptide serum markers in and T-bet cooperate to program terminal differentiation in response islet autoantibody-positive children. Diabetologia 2017;60:287–295 to LCMV viral infection. J Exp Med 2015;212:2041–2056 59. Achenbach P, Hummel M, Thümer L, Boerschmann H, Höfelmann D, Ziegler 37. Nishida A, Andoh A, Inatomi O, Fujiyama Y. Interleukin-32 expression in the AG. Characteristics of rapid vs slow progression to type 1 diabetes in multiple islet pancreas. J Biol Chem 2009;284:17868–17876 autoantibody-positive children. Diabetologia 2013;56:1615–1622