The ISR induces global mRNA translational initiation blockade.

The phosphorylation of eIF2α during the ISR leads to sequestration of the translation initiation factor eIF2B, suppressing mRNA translation initiation (for a review see ref. 27) and thereby reducing protein synthesis. To assess the effects of ISR on protein synthesis, we performed surface sensing of translation (SUnSET) (28) on isolated islets from 8-week-old CD1, NSG, and NOD mice. Incorporation of puromycin into elongating polypeptide chains, followed by immunoblotting with antipuromycin antibodies, allows for assessment of mRNA translation. Consistent with the activation of the ISR, we observed reduced puromycin incorporation into proteins of NSG and NOD islets, suggesting that global mRNA translation is reduced in prediabetic stages (Figure 1, I and J). ISR activation can be mediated by any 1 or more of the 4 kinases—PERK, PKR, GCN2, and HRI. Based on our observed increase in UPR in NOD islets over time (Figure 1, A–D), we surmised that PERK may be the relevant activated kinase in NOD islets. Immunoblot analysis demonstrated an increased trend in phosphorylated PERK (p-PERK) in islets of both NSG and NOD mice compared with CD1 controls, implicating PERK as the ISR kinase (Figure 1, K and L). We next performed SUnSET using mouse MIN6 β cells treated with proinflammatory cytokines (IFN-γ+IL-1β+TNF-α) to induce a state mimicking β cell ER stress seen in T1D (29). Like NOD islets, we observed reduced puromycin incorporation into proteins of proinflammatory cytokine-treated cells compared with vehicle control (Figure 1M). To directly correlate the block in protein synthesis with the activity of the ISR, we utilized 2 inhibitors of the ISR. HC-5770 is a highly specific inhibitor of PERK (previously characterized as Cmpd26 in ref. 30), and ISRIB is a previously described inhibitor of the p-eIF2α/eIF2B interaction (31). Treatment with 250 nM HC-5770 or 50 nM ISRIB partially reversed the block in protein synthesis induced by proinflammatory cytokines (Figure 1, M and N).

To interrogate the effects of the ISR on mRNA translation initiation, we performed polyribosome profiling (PRP) studies of total RNA from cadaveric human donor islets. PRP can distinguish global changes in mRNA translation initiation and elongation by assessment of the RNA sedimenting with polyribosomes versus monoribosomes (Figure 1O). Lower polysome sedimentation of RNA suggests a relative translation initiation blockade (32). Human islets treated with proinflammatory cytokines (IFN-γ+IL-1β) to mimic T1D inflammation (29) showed reduced association of RNA with polyribosomes (or translation initiation blockade) by PRP compared with control islets (Figure 1O). Concurrent treatment of human islets with either 250 nM HC-5770 or 50 nM ISRIB led to a recovery of the RNA sedimenting with polysomes, reversing the effects of proinflammatory cytokines (Figure 1O). Collectively, these data indicate that inflammation induces a translation-initiation blockade, which is reversed upon inhibition of the ISR.

PK and PD assessment of the PERK inhibitor HC-5770.

To evaluate the role of PERK in β cell dysfunction in vivo, we utilized the PERK inhibitor HC-5770. First, to ensure that HC-5770 does not cause a compensatory activation of GCN2 (as reported for other PERK inhibitors) (33), we performed a dose-ranging study in isolated CD1 mouse islets followed by immunoblotting for p-GCN2 and total GCN2. Between 8 and 1,000 nM concentrations of HC-5770, we observed no changes in p-GCN2 (Supplemental Figure 2A). To ensure that our antibody detects p-GCN2, we also performed immunoblotting of mouse embryonic fibroblasts (MEFs), which showed that 100 nM halofuginone (a GCN2-ISR activator) (34) robustly induces p-GCN2, while thapsigargin treatment suppresses it (Supplemental Figure 2A). Initial pharmacokinetic and pharmacodynamic (PK/PD) analyses were performed to confirm in vivo PERK inhibition in mouse pancreas and to identify appropriate doses for further study. The first PK analysis of HC-5770 followed a single oral administration of HC-5770 at doses ranging from 0.3 to 30 mg/kg in BALB/c mice, which revealed dose-proportionate increases in plasma exposure with a half-life of approximately 3 hours (Table 1). The unbound fraction (Fu) in mouse plasma was determined in vitro to be 0.3%, which enabled us to calculate the free, unbound drug plasma exposure across time in vivo (Supplemental Figure 2B). A second PK experiment in NOD mice followed a single oral administration of HC-5770 at 1 and 10 mg/kg and confirmed nearly identical exposure and clearance between NOD and BALB/c mouse strains (Supplemental Figure 2C).

Table 1

PK analysis of HC-5770 in mouse plasma and pancreas

Open in a separate window

The PD effect of HC-5770 on p-PERK (T980) was evaluated in mouse pancreas. Whole-protein lysates from mouse pancreata isolated from BALB/c mice following single administration of HC-5770 at doses ranging from 0.3 to 30 mg/kg, as described above. At 10 and 30 mg/kg, HC-5770 achieved approximately 75% inhibition 1 hour after dose, whereas doses ranging from 0.3 to 3 mg/kg induced moderate effects on p-PERK/PERK levels that were sustained past 4 hours following administration (Supplemental Figure 2D). We next sought to evaluate the impact of PERK inhibition on insulitis by treating prediabetic NOD mice with HC-5770 at doses ranging from 0.3 to 30 mg/kg (twice daily) for 2 weeks. Following the treatment period, mouse pancreas sections were stained and scored for the level of islet immune infiltration (insulitis). HC-5770 decreased insulitis at all doses tested, with the greatest response noted at doses of 1 mg/kg and above (Supplemental Figure 2E). As complete and sustained PERK inhibition has previously been associated with pancreatic degeneration and dysfunction (35), we reasoned that the lowest efficacious doses should be selected for continued investigation in vivo and selected 0.3, 1, and 3 mg/kg twice daily as reasonable doses to advance. Flexibility in the dosing regimen was then evaluated by assessing the insulitis response in animals treated twice daily versus once daily with HC-5770. NOD mice were treated either twice daily at 0.3, 1, and 3 mg/kg or once daily at 0.6, 2, and 6 mg/kg HC-5770 for 2 weeks. The once-daily dosing schedules resulted in effects similar to those of twice-daily dosing on insulitis, significantly inhibiting insulitis at all doses tested (Supplemental Figure 2F). Based on these findings, once-daily doses ranging from 0.6 to 6 mg/kg (QD) were selected for continued evaluation in vivo.

Systemic inhibition of PERK delays autoimmune diabetes development in NOD mice.

We hypothesized that the β cell translational blockade induced by the ISR in the prediabetic state is maladaptive and contributes to the development of T1D. To test this hypothesis, we treated female NOD mice with HC-5770 for 4 weeks during the prediabetic stage when the ISR is active (6 to 10 weeks of age) and monitored for subsequent diabetes development until 25 weeks of age (see schematic in Figure 2A). NOD mice were treated with either vehicle or 3 different doses of HC-5770 (0.6, 2, or 6 mg/kg per day). Approximately 60%, 53%, and 68% of the mice treated with 0.6 mg/kg, 2 mg/kg, and 6 mg/kg of HC-5770, respectively, remained diabetes free by 25 weeks of age, whereas only 10% of the vehicle-treated mice remained diabetes free (Figure 2B). These findings suggest an enduring effect of early PERK inhibitor treatment. Notably, the exocrine pancreas of mice treated with HC-5770 (6 mg/kg) showed no gross pathological evidence of pancreatitis (Figure 2C), unlike findings observed upon more complete and sustained inhibition with other PERK inhibitors (36). This observation is also supported by gross pancreas weights that were unaffected by 2 weeks of treatment with either the twice-daily or once-daily dosing regimens (Supplemental Figure 2G).

Open in a separate window

Figure 2

PERK inhibition delays autoimmune diabetes in NOD mice.

Prediabetic female NOD mice (6 weeks of age) were treated with varying doses of HC-5770 or ISRIB. (A) Experimental design for diabetes incidence study. (B) Diabetes incidence. n = 20 mice per group (Mantel-Cox). (C) Representative H&E stain of pancreata from nondiabetic mice at 25 weeks of age that were treated between 6 and 10 weeks of age. A, acinar; I, islet. Scale bars: 50 μm. (D) Experimental design mechanistic studies. (E) Representative images of pancreata from NOD mice following 2 weeks of HC-5770 administration (6 mg/kg) stained for insulin (brown) and counterstained with hematoxylin (blue). Scale bars: 500 μm. (F) β Cell mass of mice treated with HC-5770 (6 mg/kg) for 2 weeks; n = 4–5 mice per group (ANOVA). (G) Representative images of pancreata from NOD mice following 2 weeks of HC-5770 administration immunostained as indicated. Arrows indicate regions of insulitis. Scale bars: 50 μm. (H) Average insulitis score of mice treated with varying doses of HC-5770 for 2 weeks; n = 4–5 mice per group. NB: these data are replicated in Supplemental Figure 2F for comparative purposes (ANOVA). (I) Experimental design. (J) Representative images of pancreata from NOD mice following 2 weeks of ISRIB administration stained as indicated. Scale bars: 500 μm. (K) β Cell mass of mice treated with ISRIB for 2 weeks; n = 4–5 mice per group (ANOVA). (L) Representative images of pancreata from NOD mice following 2 weeks of ISRIB administration immunostained as indicated; arrows indicate regions of insulitis. Scale bars: 50 μm. (M) Average insulitis score of mice treated with varying doses of ISRIB for 2 weeks; n = 4–5 mice per group (ANOVA). Data are presented as mean ± SEM.

HC-5770 treatment engages molecular pathways related to PERK functions in β cells, reduces β cell death, and enhances β cell replication.

To identify proximal molecular effects of PERK inhibition and its impact on the islet microenvironment, we next performed a short-term, 2-week oral treatment of prediabetic NOD mice (beginning at 6 weeks of age) with differing doses of HC-5770 followed by assessment of glucose homeostasis, pancreas pathology, and islet single cell molecular analyses (see scheme in Figure 2D). Upon treatment at any of the HC-5770 doses, there was no statistical change in β cell mass (Figure 2, E and F) compared with controls. However, as noted previously, there was a significant decrease in insulitis in HC-5770–treated mice at all doses compared with vehicle controls (Figure 2, G and H), a finding preceding the eventual protection of these mice from diabetes. To confirm that the PERK inhibition effect occurs via blockade of p-eIF2α function, we next utilized ISRIB (an inhibitor of the p-eIF2α/eIF2B interaction) in NOD mice. Six-week-old NOD mice were treated with 2 different doses of ISRIB (0.25 or 2.5 mg/kg) or vehicle by intraperitoneal injection for 2 weeks (see schematic in Figure 2I). Consistent with the effects of HC-5770, mice receiving ISRIB exhibited no effect on β cell mass (Figure 2, J and K) and a significant reduction in insulitis (Figure 2, L and M).

To determine the effect of HC-5770 on molecular pathways in the cells of the islet microenvironment, we performed scRNA-Seq of islets following 2-week oral treatment of NOD mice beginning at 6 weeks of age. For these studies, we employed HC-5770 at 6 mg/kg, as the mice that received this dose in our diabetes outcome study had the lowest incidence of diabetes. We visualized cells based on expression profiles using uniform manifold approximation and projection (UMAP) for dimension reduction plots and identified clusters representing distinct pancreatic cell types (Figure 3A). The β, α, δ, PP, acinar, stellate, duct, T cells, B cells, and myeloid cell types were characterized based on expression of genes Ins1/2, Gcg, Sst, Ppy, Prss1, Col3a1, Krt19, Trbc2, Cd79a, H2-Eb1, respectively. Dot plots of the top 5 genes in each cell type confirm the correct identification of cell types (Supplemental Figure 3A).

Open in a separate window

Figure 3

PERK inhibition increases levels of PD-L1 in β cells of NOD mice.

Prediabetic female NOD mice (6 weeks of age) were treated with HC-5770 for 2 weeks, isolated islets were subjected to scRNA-Seq, and pancreas tissue was subjected to NanoString spatial proteomics. (A) UMAP embeddings of merged scRNA-Seq profiles from islets. n = 3 mice per group. (B) GO analysis of all cell clusters (pseudo-bulk analysis). (C) GSEA of β cell clusters showing UPR. (D) GSEA of β cell clusters showing PERK-meditated UPR. (E) Percentages of T, B, and myeloid cells identified within the immune cell clusters. (F) Percentages of α, β, δ, and PP cells identified within the islet cell clusters. (G) Example of the insulin-positive area and the insulitic area used to collect spatial tissue-based proteomics. (H) Heatmap of identified proteins in the insulitic area (left panel) and insulin-positive area (right panel); n = 10–11 ROI from 2 mice per group. *P < 0.05, t test. (I) Representative images of pancreata from mice following 2 weeks of treatment with vehicle or HC-5770 immunostained as indicated; dotted lines indicate islets. Scale bars: 50 μm. (J) Quantification of PD-L1 intensity in the β cells of I; each dot represents an islet from n = 4 mice (distinguished by color), with mean values of each mouse shown (t test for the means). (K) Six-week-old female NOD mice were treated as indicated with HC-5770 for 2 weeks, then administered either anti-PD-L1 or IgG control, followed by another 2 weeks of HC-5770 treatment; glucose was measured on alternate days after injection; n = 5 mice per group. (L) AUC analysis of the data in K (ANOVA). Data are represented as mean ± SEM.

To assess the engagement of molecular processes by HC-5770, we performed a pseudo-bulk analysis followed by Gene Ontology (GO) analysis. We found that cytoplasmic translation, leukocyte proliferation, digestion, protein stabilization, and protein ubiquitination were among the top significantly regulated pathways in HC-5770–treated islets (Figure 3B). Furthermore, we performed GO analysis on β cell clusters (9 clusters). We found that cytoplasmic translation, protein folding, ER stress response, and antigen processing and presentation were among the top significantly regulated pathways in HC-5770–treated β cells (Supplemental Table 1). Consistent with these findings, GSEA showed that β cells of HC-5770–treated mice downregulated genes in the UPR pathway relative to vehicle controls, as indicated by a normalized enrichment score (NES) of –7.24 (Figure 3C). Because the UPR pathway encompasses 3 distinct arms — PERK, IRE1α, and ATF6 — we analyzed the PERK pathway by GSEA (prioritizing the genes Atf4, Eif2s1, Eif2ak3, Nck1, Nck2, Nfe2l2, Ptpn1, Ptpn2, Agr2, Abca7, Bok, Tmed2, Tmem33, and Qrich1). This analysis revealed a significant decrease in PERK-mediated UPR in the β cells of HC-5770–treated mice compared with vehicle controls (NES: –2.52) (Figure 3D). Collectively, these data support the suppression of PERK-related molecular processes by HC-5770, indicating appropriate target engagement.

Examination of cell clusters and numbers in the HC-5770 group compared with vehicle controls revealed several notable findings: (a) there was a decrease in the percentage of T and B cells, with an increase in the proportion of myeloid cells (Figure 3E); (b) GO analysis (Supplemental Figure 3, B–D) of T, B, and myeloid cells revealed alterations in cytoplasmic translation and T cell proliferation (T cells), cytoplasmic translation and the humoral immune response (B cells), and antigen processing and chemotaxis (myeloid cells) — all indicating phenotypic changes in these immune cell populations that likely influenced their function in response to PERK inhibition; and (c) with respect to T cells, in particular, GSEA analysis revealed a reduction in T cell activation (NES: –11.8) upon PERK inhibition (Supplemental Figure 3E). Accordingly, the β cell population demonstrated a reduced inflammatory response by GSEA (NES: –6.85) (Supplemental Figure 3F).

Regarding the endocrine cell population, there was an increase in the overall percentage of β cells and a decrease in α cell percentage (Figure 3F). The increased β cell numbers upon HC-5770 treatment led us to investigate β cell replication and death. Immunostaining of pancreata showed a trend toward an increased number of proliferating cell nuclear anthigen–positive (PCNA-positive) β cells upon HC-5770 treatment (Supplemental Figure 3, G and H) and no changes in β cell death by TUNEL assay (Supplemental Figure 3, G and I). In addition, scRNA-Seq revealed an increase in the percentage of β cells in S/G₂M phases consistent with the increased trend in PCNA-positive β cells (Supplemental Figure 3J).

PERK inhibition increases β cell PD-L1 levels.

To interrogate the nature of the immune cell populations in the islet microenvironment in PERK inhibitor–treated NOD mice, we performed spatial tissue-based proteomics after 2 weeks of HC-5770 treatment. We used insulin immunostaining and nuclei staining to identify β cells and the surrounding insulitic regions, respectively (Figure 3G). Prevalidated antibodies in the GeoMx mouse immune panel were used to probe for immune cell subtypes in the peri-islet insulitic region and within the islet. Whereas there were no statistical differences in the immune cell subtype populations in the insulitic regions of HC-5770–treated mice versus vehicle controls (Figure 3H), within the β cell region, there was a striking and significant upregulation of programmed death–ligand 1 (PD-L1) as well as elevations of its cognate receptor PD-1, CD3e, CD8a, and CD11b (Figure 3H) following PERK inhibition. The increase in PD-L1 levels on β cells was confirmed by immunofluorescence staining of pancreatic tissues (Figure 3, I and J). The interaction of PD-L1 on β cells with PD-1 on immune cells is known to skew immune cell populations to a more immunosuppressive phenotype (37). To assess whether increased PD-L1 levels after PERK inhibition mediate protection against T1D development, we treated female NOD mice that received HC-5770 for 2 weeks (6 to 8 weeks of age) with a single dose of monoclonal antibody against PD-L1 or the corresponding isotype IgG control, then continued HC-5770 treatment for an additional 2 weeks. As a control for diabetes development, 8-week-old female NOD mice were treated with a single dose of anti–PD-L1. HC-5770–treated mice that received isotype IgG control remained normoglycemic, while mice that received only anti–PD-L1 became hyperglycemic within 7 days (Figure 3, K and L). The mice receiving both HC-5770 and anti–PD-L1 displayed average glucose levels intermediate between those of the 2 controls (Figure 3, K and L), emphasizing that anti–PD-L1 administration partially antagonized the effect of HC-5770.

Animals and procedures.

Mouse experiments were performed under specific pathogen–free conditions, and mice were maintained on a 12-hour light/12-hour dark cycle with free access to food and water. CD1 mice were purchased from Charles River (Charles River, catalog 022), and NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) (catalog 5557) and NOD/ShiLtJ (NOD) (catalog 1976) mice were purchased from The Jackson Laboratory. PK studies using BALB/c mice were performed under a contract with Pharmaron.

For diabetes incidence, 6-week-old female NOD mice were orally gavaged with vehicle (0.5% methylcellulose) or 0.6, 2, or 6 mg/kg HC-5770 (30) for either 2 or 4 weeks. Six-week-old female NOD mice were injected intraperitoneally with vehicle (5% DMSO, 2% Tween 80, 20% PEG400, and saline) or 0.25 or 2.5 mg/kg ISRIB (trans-isomer) (MedChemExpress: HY-12495) for 2 weeks (71). Blood glucose was measured by the tail vein using a glucometer (AlphaTrak). For diabetes incidence, diabetes was classified as 2 consecutive blood glucose values greater than 250 mg/dL. At the end of each study, mice were euthanized, and tissue and blood were collected. To isolate islets, collagenase was injected into the pancreatic bile duct to inflate the pancreas before removal, as previously described (72). Briefly, a Histopaque-HBSS gradient was applied to the dissociated pancreas, followed by centrifugation at 900 g for 18 minutes. The mouse islets were then removed from the center of the gradient and cultured in RPMI medium. Islets were handpicked and allowed to recover overnight before experimentation.

For experiments involving anti–PD-L1, 6-week-old female NOD mice were orally gavaged with 6 mg/kg HC-5770 for a total of 4 weeks. When the mice were 8 weeks old, a single intraperitoneal injection of neutralizing rat monoclonal antibody against PD-L1 (dosage: 200 mg/kg; BioXCell; BE0101) or rat IgG2b isotype control (dosage: 200 mg/kg; BioXCell; BE0090) was administered, while continuing the HC-5770 treatment for an additional 2 weeks. Control 8-week-old female NOD mice were administered a single dose of 200 mg/kg of anti–PD-L1 intraperitoneally. Glucose was monitored over the subsequent 16 days after injection of anti–PD-L1 or IgG control.

PK analysis of HC-5770 in mouse plasma followed a methodology described elsewhere (30). HC-5770 was suspended in a vehicle consisting of 0.5% methylcellulose (400 cP) and 0.1% Tween 80 in water and administered to female BALB/c nude mice by oral gavage at 0.3, 1, 3, 10, and 30 mg/kg. Plasma was sampled from 5 mice per group following a single oral administration at 1, 4, 8, 12, and 24 hours after dosing. The plasma concentration of the compound was determined by protein precipitation with acetonitrile and liquid chromatography with tandem mass spectrometric detection (LC-MS/MS). Parameters were estimated using Phoenix (WinNonlin) PK software, version 6.1.0, using a noncompartmental approach consistent with the oral route of administration. PK analysis in female NOD mice followed a similar methodology, with the exception that only the 1 and 10 mg/kg doses were evaluated. Methods for the determination of mouse protein plasma binding were described previously (73). Frozen pancreata from mice treated with HC-5770 were homogenized and protein isolated in preparation for Simple Western analysis, as described previously (30). In brief, protein detection was performed on the Jess Simple Western high-throughput protein analysis platform (ProteinSimple) according to the manufacturer’s protocol using a 12-230 kDa Separation Module (ProteinSimple, SM-W004) and Total Protein Detection Module (ProteinSimple, DM-TP01). The following antibodies were used: p-PERK (Eli Lilly; 1:50) (74) and PERK (Cell Signaling Technology, 1:200, catalog 3192).

Human islets.

Deidentified nondiabetic male and female human donor islets were obtained from the Integrated Islet Distribution Program (IIDP) and the University of Alberta Diabetes Institute Islet Core (Edmonton, Alberta, Canada) (Supplemental Table 2).

SUnSET assay.

New protein production was determined using the SUnSET technique (28). Briefly, at the end of the treatments, cells were incubated with 10 μg/mL puromycin for 15 minutes for MIN6 cells and 30 minutes for mouse islets, and protein was isolated using RIPA lysis buffer and used for immunoblotting. After detection of puromycin, the blots were stripped using RESTORE PLUS Stripping Buffer (Thermo Fisher. 46430) and then stained with REVERT 700 Total Protein Stain (Licor, 926-11016). Puromycin lanes were normalized against corresponding lanes in total protein stain to account for loading differences.

scRNA-Seq.

Islets were submitted to the University of Chicago Genomics Facility for library generation using 10x Chromium Single Cell 3′ v 3.1, as previously described (75). Approximately 12,800 cells were loaded to achieve 8,000 captured cells per sample to be sequenced. Sequencing was performed on Illumina NovaSeq 6000. Raw sequencing files were processed through the Rosalind (https://rosalind.bio/) pipeline with a HyperScale architecture (Rosalind). Quality scores were assessed using the FastQC tool (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Cell Ranger was used to align reads to the Mus musculus genome build GRCm38, count unique molecular identifiers (UMIs), call cell barcodes, and perform default clustering. After initial processing, raw RNA matrices from each sample were then analyzed for quality control, “ambient” mRNA was removed using SoupX, version 1.6.1 (76), and clustering was performed utilizing Seurat (77), version 4.3.0, in R, version 4.2.2. Basic filtering parameters included cells with unique features of a minimum of 200 and a maximum of 7,500. Cells expressing less than 25% mitochondrial-related genes were included. The cell-cycle effect was regressed using previously established methods in Seurat. After filtering, replicates from each condition were merged using standard sctransform Seurat protocols using 3000 integration features. After clustering, cells were visualized using UMAP (78) and color customized using ggplot2 (79). Marker genes were determined using the FindAllMarkers function (Wilcoxon’s rank-sum test) in Seurat. Contaminating endothelial and neuronal cells were distinct from the other clusters and were subsequently removed from the final analysis.

GSEA was performed using the msigdbr package (https://cran.r-project.org/web/packages/msigdbr/vignettes/msigdbr-intro.html) to determine lists of genes from the following gene sets: Hallmark Gene Set (Unfolded Protein Response, Inflammatory Response), GO Biological Processes (PERK Mediated Unfolded Protein Response), and Reactome (Antigen processing: Ubiquitin & Proteasome degradation). Pseudo-bulk differential gene expression analysis was performed where gene counts in all the cells for each biological replicate were aggregated and pathways identified by GO (80). Previously published single-cell transcriptional data (25, 26) deposited in the NCBI’s Gene Expression Omnibus database (GEO GSE141786 and GSE117770) were reanalyzed using the above-mentioned protocol using R. GSEA for the Hallmark UPR pathway was performed on the endocrine subpopulation from 4-week-old, 8-week-old, and 15-week-old NOD islets from the GSE141786 data set and from the β cell population from 8-week-old, 14-week-old, and 16-week-old NOD islets from GSE117770 data set.

FASTQ files of 10x Genomics scRNA-Seq data for human islets were downloaded from the data portal of the HPAP (70) (https://hpap.pmacs.upenn.edu). The data sets were generated from pancreatic islet samples of the donors indicated in Supplemental Table 3. Raw reads were processed with Cell Ranger, version 6.1.2 (81), for quality control, alignment, and gene expression quantification. The reads were aligned to the human genome reference (GRCh38). The “ambient” mRNA was removed using SoupX, version 1.6.1 (76), employing genes (INS, GCG, SST, TTR, IAPP, PYY, KRT9, and TPH1) identified from the initial clustering by Seurat (77) as markers that represent the major cell types of human islets. The potential doublet cells were assessed and removed by scDblFinder, version 3.16 (82). The remaining cells were further filtered by the following criteria: number of genes detected greater than 200 and fewer than 9,000, percentage of mitochondrial reads less than 25%, and number of counts less than 10,000. Next, the SCTransform function (83) implemented in Seurat software was used to normalize the counts by removing the effects of library depth and regressing out the variation from the mitochondrial reads’ ratio. The top 3,000 variable genes were selected to perform the principal component analysis (PCA). Finally, software Harmony, version 0.1.1 (84), was employed to integrate all the samples. This integration was performed based on the top 50 PCA components, considering the donor identity and reagent kit batches as the primary confounding factors. scSorter, version 0.0.2 (85), which uses known marker genes of cell types, was utilized to annotate the cell types in human islets. After the above-described analyses, a total of 10,167 β cells from all the donors were identified, including 4,730 cells from nondiabetic donors, 2,372 cells from single-autoantibody–positive donors, 2,480 cells from double-autoantibody–positive donors, and 585 cells from type 1 diabetic donors.

NanoString spatial proteomics.

Paraffin-embedded pancreata were used for NanoString spatial proteomics analysis. Tissues were stained with morphology markers as follows: AF-647–conjugated insulin (Cell Signaling Technology; 9008s; 1:400) and nuclei marker (SYTO13). Tissues were hybridized using a prevalidated mouse GeoMx Immune Cell Panel (NanoString; GMX-PROCONCT-MICP) comprising the following markers: PD-1, CD11c, CD8a, PanCk, MHC II, CD19, CTLA4, SMA, CD11b, CD3e, fibronectin, Ki-67, CD4, GZMB, F4/80, CD45, PD-L1; housekeeping genes: histone H3, S6, GAPDH; and IgG antibodies: Rb IgG, Rat IgG2a, and Rat IgG2b for background subtraction. All the markers were conjugated to UV-photocleavable oligos for indexing. At least 5 to 6 islets with insulitis were chosen as regions of interest (ROI) per mouse based on the morphology markers (insulin and nuclei). The ROIs were segmented into insulitic region and insulin⁺ region for each islet. Oligos from the segmented ROIs were photocleaved and collected in a 96-well plate; reads were counted using nCounter (NanoString). Analysis was performed using NanoString software, version 2.4.0.421. Scaling was performed to normalize for any differences in tissue surface area and depth. After scaling, reads were normalized to housekeeping markers and background was subtracted using IgG markers. Normalized counts were visualized as heatmaps using GraphPad Prism, version 10.

Statistics.

All data are represented as mean ± SEM. For comparisons involving more than 2 conditions, 1-way ANOVA or repeated-measures ANOVA (with Tukey’s post hoc test or Dunnett’s post hoc test) was performed. For comparisons involving only 2 conditions, a 2-tailed Student’s unpaired t test was performed. Mantel-Cox log-rank test was performed to determine the differences between groups in the NOD diabetes outcome experiments. GraphPad Prism, version 10, was used for all statistical analyses and visualization. Statistical significance was assumed at P < 0.05.

Supplemental Methods.

Additional details are provided in Supplemental Methods.

Study approval.

Studies involving mice were performed under a protocol approved by the University of Chicago Institutional Animal Care and Use Committee. The use of deidentified human samples was approved by the Institutional Review Board at the University of Chicago and considered exempt from human subjects research.

We would like to thank Kara Orr (Indiana University School of Medicine) and Sarida Pratuangtham (University of Chicago) for their technical assistance. This work was supported in part by NIH grants R01 DK060581 (to RGM), U01 DK127786 and U01 DK127786S1 (to RGM and BJMWR), R01 DK133881 (to EKS and RGM), R01 DK135832 (to SAT), R01 CA219815 (to SAO), F31 DK134070 (to CMA), and T32 AI153020 (to JRE); an investigator-initiated award from HiberCell (to SAT and RGM); a Breakthrough T1D postdoctoral fellowship (3-PDF-2023-1326-A-N) and Diabetes Research Connection awards (both to CM); and a Quad Summer Scholarship award (to JEW). This study utilized Diabetes Center core resources supported by NIH grant P30 DK020595 (to the University of Chicago) and P30 DK097512 (to Indiana University) and utilized services of the University of Chicago Histology and Genomics Cores.