Multi-omic Mapping of Human Pancreatic Islet Endoplasmic Reticulum and Cytokine Stress Responses Provides Type 2 Diabetes Genetic Insights

SUMMARY

Endoplasmic reticulum (ER) and inflammatory stress responses contribute to islet dysfunction in type 2 diabetes (T2D). Comprehensive genomic understanding of these human islet stress responses, and whether T2D-associated genetic variants modulate them, is lacking. Here, comparative transcriptome and epigenome analyses of human islets exposed ex vivo to these stressors revealed 30% of expressed genes and 14% of islet cis-regulatory elements (CREs) as stress-responsive, modulated largely in ER- or cytokine-specific fashion. T2D variants overlapped 86 stress-responsive CREs, including 21 induced by ER stress. We linked rs6917676-T T2D risk allele to increased islet ER stress-responsive CRE accessibility and allele-specific β-cell nuclear factor binding. MAP3K5, the ER stress-responsive putative rs6917676 T2D effector gene, promoted stress-induced β-cell apoptosis. Supporting its pro-diabetogenic role, MAP3K5 expression correlated inversely with human islet β-cell abundance and was elevated in T2D β-cells. This study provides genome-wide insights into human islet stress responses and context-specific T2D variant effects.

eTOC Blurb:

This multi-omic study dissects the genome-wide effects of ER stress and pro-inflammatory cytokines on human islet transcriptional regulatory networks and motivates variant-to-function studies linking rs6917676 T2D risk allele to enhanced ER stress-responsive MAP3K5-mediated β-cell apoptosis. Findings suggest MAP3K5 inhibitors (e.g., Selonsertib) could therapeutically reduce stress-induced β-cell apoptosis.

Graphical Abstract

INTRODUCTION

Type 2 diabetes (T2D) is a complex metabolic disorder characterized by gene-environment factors contributing to pancreatic islet beta cell dysfunction and/or death, and inadequate insulin secretion in response to insulin resistance^1–5. Genome-wide association studies (GWAS) linked DNA sequence variants in >600 loci in the human genome with increased T2D risk^6–10. These variants’ non-coding location, combined with studies demonstrating their significant enrichment in islet cis-regulatory elements (CREs), suggests they contribute to islet dysfunction and failure by altering CRE function and effector gene expression^2,4,11–14. A subset of T2D-associated variants altered in vivo human islet CRE chromatin accessibility or effector gene expression under steady-state conditions^{4,11–13,15–17}. However, T2D pathogenesis involves dynamic interactions between genetic variants and environmental stressors^1,2,4,5, yet functional effects of putative regulatory non-coding variants on islet transcriptional stress responses to pathophysiologic endoplasmic reticulum (ER) stress and pro-inflammatory cytokine stressors¹, are largely unknown.

Proper ER trafficking and protein folding capacity is integral to β-cell protein quality control and insulin synthesis; both genetic and pathophysiologic ER functional perturbations (ER stress) are implicated in islet dysfunction across the diabetes spectrum^18,19. Multiple INS mutations in patients with permanent neonatal diabetes lead to ER retention and induce ER stress. Rare variants in genes encoding proteins that prevent and sense protein misfolding (DNAJC3^19–23) or activate and transduce unfolded protein response (UPR) signals (EIF2AK3²⁴, EIF2B1²⁵, ATF6²⁶, WFS1^27,28) are linked to syndromic, monogenic diabetes. Chronic hyperglycemia elicits sustained demand for insulin production, which can overwhelm the β-cell ER, increasing stress and activating UPR machinery¹⁸. Prolonged or excessive ER stress contributes to β-cell dysfunction and death^18,19. β-cell dysfunction has been further linked to elevated pro-inflammatory cytokine levels in the blood^19,29,30, which activates the NFKB pathway and impairs insulin secretion^30–32. Although ER and inflammatory stressors have been associated with T2D^33,34, our understanding of pancreatic islet responses to each, and how T2D variants modulate these transcriptional programs, is incomplete.

To fill these knowledge gaps, we defined robust transcriptional regulatory programs controlling human islet responses to ER stress and pro-inflammatory cytokines by mapping genome-wide CRE accessibility (ATAC-seq) and gene expression (RNA-seq) in islets exposed to ER stress-inducing agent thapsigargin (250 nM) or pro-inflammatory cytokines IL-1β (25 U/mL) and IFN-γ (1000 U/mL). These stressors and their respective concentrations were selected based upon previous studies using comparable concentrations of thapsigargin^15,35–38 or IL-1β + IFN-γ^39–41 to model islet ER stress or cytokine-induced inflammatory responses, respectively. Comparison of stress-responsive genes and CREs revealed complementary stress- and cell type-specific changes in transcriptional regulatory programs and expression of the factors mediating these responses. We identified T2D-associated variants in 38 signals overlapping ER stress- or cytokine-induced CREs as candidate causal variants and linked them to stress-responsive target (putative T2D effector) genes. Targeted variant-to-function analyses linked rs6917676-T T2D risk allele to increased ER stress-responsive CRE accessibility and demonstrated that its putative T2D effector MAP3K5, the only ER stress-responsive gene in the locus, promoted stress-responsive apoptosis in human β-cells and primary human islets.

RESULTS

Comprehensive comparative mapping of human islet ER stress- and cytokine-responsive genes

To define characteristic human pancreatic islet responses to ER stress and pro-inflammatory cytokines, we exposed primary human islets from multiple non-diabetic donors (Table S1) to 24-hour treatment with thapsigargin (vs. DMSO solvent control)^15,38,35 or IL-1β+IFN-γ cocktail (vs. untreated control)³⁹, respectively. We determined and compared genome-wide gene expression changes elicited by these T2D-relevant stressors using whole islet RNA sequencing (RNA-seq)⁴².

In total, ~30% (5,131/17,096) of human islet-expressed, autosomal protein-coding genes (Methods) were modulated by ≥1 stressor compared to their respective controls (Table S2; FC≥1.5 or ≤0.667, FDR<5%). 2,967 genes were differentially expressed (DE) upon ER stress (1,517 induced; 1,450 reduced), whereas 3,443 genes were DE upon cytokine-induced inflammation (1,893 induced; 1,550 reduced) (Table S2). ER stress and cytokine DE genes were largely distinct. For example, ~85% of induced genes were stressor-specific (1,064 ER stress-specific; 1,440 cytokine-specific (Figure 1A; Table S2). Pathway analysis confirmed that thapsigargin treatment activated genes facilitating both homeostatic (e.g., ATF4, ERN1/IRE1α, EIF2AK3/PERK, HERPUD1, HSPA5/BiP) and terminal (e.g., DDIT3/CHOP, MAP3K5) arms of UPR and ER protein processing pathways (Figure 1B; Table S2) central to beta cell function and survival^43–49 (Figures 1B-C). Importantly, thapsigargin engaged each UPR sensing (e.g., calreticulin/calnexin; BiP) and signaling (e.g., PERK, ATF6, and IRE1) branch (Table S2) linked to islet dysfunction and diabetes genetics, thereby providing a comprehensive perspective on the transcriptional regulatory networks and output of these processes. In addition to anticipated UPR-related signals, ER stress significantly induced cholesterol synthesis/metabolism and amino acid biosynthesis pathways (Table S2), including pharmacologic targets PCSK9 and LDLR with underappreciated roles in beta-cell insulin secretion^50,51 and ARG2, a gene with central roles in polyamine synthesis⁵² that is dysregulated in T2D islets⁵³, suggesting broader and more complex impact of ER stress on cellular functions extending beyond protein folding.

Figure 1. Induced transcriptional responses of human pancreatic islets to ER stress (ERS) and pro-inflammatory cytokines (CYT).

(A) Heatmap of genes induced by ERS and/or CYT treatment (FDR<5%; FC≥1.5). Induced gene categories are ERS-specific, CYT-specific, or shared between both conditions; gene numbers for each category are shown in parentheses. Note the majority of genes exhibit stress-specific induction. Expression values are z-score scaled. (B) Enriched pathways for induced genes; FDR values and example genes for each pathway are indicated. (C) Examples of enriched pathway genes induced by ERS, CYT, or both. Dot-and-box plots show normalized per-donor islet gene expression (CPM) in ERS (green), CYT (orange), or control samples (gray). ***=FDR<5%, FC≥1.5; ns=not significant. FDRs calculated using Benjamini-Hochberg p-value adjustment. FDR, false discovery rate; FC, fold change; CPM, counts per million. See also Tables S1-2, Data S1.

Consistent with previous reports^54–56, the most highly-enriched pathways for islet cytokine-induced genes were related to cytokine and chemokine signaling pathways (Figure 1A; Table S2), including NFKB complex members (e.g., NFKB1, NFKBIA) and signaling molecules (e.g., JAK2, STAT2) modulating inflammatory responses, promoting immune cell infiltration, and contributing to islet beta cell dysfunction (Figures 1B-C; Table S2)^57–63. In addition to canonical pathways, cytokines activated multiple innate immune pathways, including NOD- and RIG-I-like receptor signaling, CGAS/STING, NLRP3 inflammasome, and cytosolic DNA-sensing pathways (Table S2) that may contribute to sensing and signaling potentially immunogenic endogenous dsRNAs provocatively suggested as a non-viral source of autoimmune activation in early T1D⁶⁴. Finally, 453 genes were induced by both ER and pro-inflammatory cytokine stressors (Figure 1A; Table S2) and enriched in pathways related to: (i) metal ion response, including free radical-scavenging metallothioneins (MT1 genes) associated with reduced insulin secretion upon stress; and (ii) DNA double-strand break (DSB) processing, including RAD9A, involved in repairing DNA damage and DSBs associated with T2D^65–67 (Figures 1B-C; Table S2).

Among 2,470 genes with reduced stress-responsive expression, ~79% were stressor-specific, including 920 ER stress-specific and 1,020 cytokine-specific genes (Figure S1A; Table S2). PDX1, ADCY5, GLP1R and IGFBP5, which encode factors integral to islet identity and function^68–71, were reduced upon ER stress (Figures S1B-C; Table S2). In contrast, SLC1A1, COL2A1, NPNT, and ITGA10, which participate in protein digestion/absorption and extracellular matrix (ECM) receptor signaling related pathways important for β-cell function^72–74, were reduced upon cytokine-induced inflammation (Figures S1B-C; Table S2). Further, 530 genes were reduced by both stressors (Figure S1A; Table S2), including CDC20, CDC45, UGT2B11 and UGT2B15, which are involved in cell cycle and retinol metabolism and crucial for islet function^{73–75,75–80} (Figures S1B-C; Table S1).

Together, the profiles indicate that islet thapsigargin and cytokine exposures engage multiple transcriptional responses as anticipated and provide comprehensive genome-wide views of genes and pathways modulated by ER stress and pro-inflammatory cytokines. Comparative analyses suggest these stressors elicit largely distinct, complementary transcriptional responses, inducing multiple stress-specific pathways and repressing islet cell type-specific critical functions.

ER stress induces robust, heterogeneous β-cell responses

To decipher cell type specificity of islet ER and cytokine stress responses, we completed single cell (sc) transcriptome profiling of islets treated with thapsigargin or pro-inflammatory cytokines (Table S1, n=3 donors/condition), yielding 18,945 total single cell transcriptomes (Figure S2A; Table S3). Unsupervised clustering identified each cell type (Figure 2A; Table S3) annotated by classic marker gene expression (Figures S2A-B; Table S3). α- (~38%) and β-cells (~37%) constituted the majority of islet cells (Figure S2C; Table S3). In striking contrast to α- and other islet cell types, β-cells distinctly exhibited increased sensitivity to ER and cytokine stressors, represented by presence of a distinct cluster (marked by *) comprised almost exclusively of stressed β-cells (Figure 2A; cluster composed of 6.53% thapsigargin; 92.91% cytokine; 0.31%; DMSO, 0.25% untreated β-cells).

Figure 2. Single-cell transcriptome analysis of human pancreatic islet ER stress (ERS) and pro-inflammatory cytokines (CYT) responses.

(A) Uniform Manifold Approximation and Projections (UMAPs) of aggregated single cell transcriptomes from islets exposed to ERS (thapsigargin), CYT (IL-1β + IFNγ) or respective control conditions for 24 hours Table S1, 3 donors/condition). UMAPs are color-coded by cell type annotations (top) or condition (bottom). n=number of cells per cell type (top) or condition (bottom). Black asterisk denotes beta cluster comprised almost exclusively of stressed cells. (B) Scaled fold-change of alpha or beta cell expression of genes induced by ERS or CYT in whole islets. Genes are grouped into genes whose induction is ERS-specific, CYT-specific, or shared between conditions. (C) Response scores for the islet-induced genes in α- or β-cells. ***p<1.0E-10; ns=not significant, two-sided Wilcoxon test. (D) α- or β-cell expression of genes representing induced gene sets in panel B. ***FDR<5%, FC≥1.5; ns=not significant. (E) UMAP of islet scRNA-seq profiles (left) reveals two beta cell clusters (BC) in ER stressed islets (middle), ERS-Beta Cluster 1 (ERS-BC1) or ERS-Beta Cluster 2 (ERS-BC2), respectively (right). Cell numbers are indicated in parentheses. (F) DEGs in ERS-BC1 (top) or ERS-BC2 (bottom) versus DMSO control. DEG numbers are indicated in parentheses. Expression values are z-score scaled. (G) Left, Venn diagram of induced DEGs (FDR<5%, FC≥1.5) in ERS-BC1 or ERS-BC2; right, KEGG, Reactome, and WikiPathways pathways enriched for intersecting vs. unique induced DEGs. FDRs are reported beneath each pathway. Note specific enrichment of apoptosis-related pathways in ERS-BC2. (H) Examples of unfolded protein response (UPR) or apoptosis gene expression in ERS-BC1, ERS-BC2, or DMSO control conditions. ***FDR<5%, FC≥1.5. False discovery rates (FDR) calculated as in Figure 1. FC, fold-change; DEGs, differentially expressed genes. See also Tables S1, S3, Data S1.

Both stressors induced α- and β-cell expression changes (Table S3). To determine relative α- and β-cell contributions to whole islet transcriptional responses, we assessed their expression of 1,020 ER stress-specific, 1,395 cytokine-specific, and 437 shared islet stress response genes (Figure 2B; Table S3). We compared fold changes of the 1,020 ER stress-specific islet induced genes in α- vs. β-cells and found that β-cells display stronger gene induction to ER stress than α-cells (Table S3; p<1.0E-03, two-sided Wilcoxon test). In addition, ER stress induced expression of 98 genes in β-cells and 35 genes in α-cells, 30 of which were shared between the two cell types (Figure S2D; Table S3). Moreover, the magnitude of ER stress-induced induction for these shared genes was higher in β- vs. α-cells (Figure S2E; Table S3; p<1.0E-03, two-sided Wilcoxon test). Together, these data suggest β-cells respond more vigorously to ER stress than α-cells.

To quantify cumulative α- and β-cell responses to ER stress and pro-inflammatory cytokines, we built and assessed three ‘response scores’ representing aggregate expression of (i) 1,020 ER stress-specific; (ii) 1,395 cytokine-specific; or (iii) 437 stress-shared induced genes (Methods). These showed that, while both α- and β-cells contribute to ER stress and cytokine responses, β-cells were more likely to yield a response to both ER stress and cytokines than α-cells (Figure 2C; p<1.0E-10; two-sided Wilcoxon test). For example, DDIT3, S100A6, and MT1F were significantly induced in β-but not α-cells (Figure 2D; Table S3). We observed the same β- vs. α-cell trend for genes with reduced expression in ER-stressed islets (Figures S2F-G). For example, β-cell identity/function genes (MAFB, SCG2, and SHISAL2B^81–88) were robustly reduced in β-cells (Figure S2H; Table S3).

Further inspection revealed two ER-stressed β-cell subpopulations (Figure 2E), comprising ~94% (Beta Cluster 1 (BC1); 1,700 cells) and ~6% (Beta Cluster 2 (BC2); 105 cells) of the total (Table S3). Response heterogeneity was both ER stress- and β-cell specific (Figure 2E, Figure S2I). ER stress BC1/BC2 subpopulations were detected in all 3 donors (Figure S2J; Table S3), suggesting these are robust, reproducible transcriptional states. We compared BC1 and BC2 transcriptomes to solvent control to identify 113 (96 induced; 17 reduced) or 170 (147 induced; 23 reduced) DEGs, respectively (Figure 2F, Table S3). 89 induced DEGs (~58%) were shared between BC1 and BC2, including canonical ER stress/UPR genes such as DDIT3, ATF4, and HERPUD1 (Figures 2G-H). BC2-specific induced DEGs (n=58) were enriched in cellular death-related pathways comprised by proteasome superfamily genes that degrade misfolded proteins and regulate apoptosis⁸⁷, such as PSMB8, PSMB9, and PSMB10 (Figures 2G-H). These ubiquitination, degradation, and apoptosis signatures were specific to the BC2 cluster (Figures S2K-L, Table S3).

In summary, scRNA-seq profiling revealed more extensive and vigorous ER stress responses in β- vs. α-cells. β-cell responses comprised of two distinct transcriptional states, including a smaller ER-stressed subset that highly expresses apoptosis-related genes. This β-cell subset may represent a distinct subpopulation more sensitive or vulnerable to ER stress-induced cell death or an inherent heterogeneity in the temporal dynamics of β-cell ER stress responses.

Identification of ER and inflammatory stress-responsive islet cis-regulatory architecture

To determine cis-regulatory elements (CREs) mediating islet ER and cytokine stress responses, we mapped and compared genome-wide CRE accessibility from six donors each in ER-stressed or cytokine-stressed islets vs. their respective DMSO or untreated controls (Table S1) using whole islet assay for transposase-accessible chromatin sequencing (ATAC-seq)⁸⁸ (Methods). Approximately 14% of CREs (14,968/109,399) were significantly (FDR<5%) remodeled in response to stress; 7,171 CREs were ER stress-responsive (3,375 opening; 3,796 closing) and 8,819 CREs were cytokine-responsive (5,768 opening; 3,051 closing) (Table S4). Consistent with stress-responsive gene expression patterns, the majority of modulated CREs exhibited stress-specific accessibility changes (Figure 3A; Figure S3A; Table S4): among 8,750 opening CREs, 2,982 were ER stress-specific, 5,375 cytokine-specific, with only 393 shared between stressors.

Figure 3. Comprehensive mapping of islet ER stress- (ERS) and pro-inflammatory cytokine- (CYT) induced cis-regulatory elements (CREs).

(A) Heatmap of human islet cis-regulatory elements (CREs) whose accessibility is increased by ERS and/or CYT treatment (FDR<5%). n=number of CREs per category. Accessibility values scaled using z-scores. (B) Percent of proximal vs. distal opening CREs (≤1kb vs. >1kb to nearest TSS, respectively). Bar chart (right) indicates percentage of opening CREs per condition. (C) Integrative Genomics Viewer (IGV) browser tracks of representative loci highlighting CREs with ERS-specific, CYT-specific, or shared accessibility increases. (D) Per control conditions for responsive genes in representative loci from panel C. ***FDR<5%, FC≥1.5; ns=not significant. (E) Heatmap of enriched transcription factor (TF) motifs in ERS-specific, CYT-specific, or shared opening distal CREs. Color gradient indicates the scaled fold-change of the motif (i.e., motif instances found in target vs. background sequences). *****FDR<1.0E-200; ****FDR<1.0E-100; ***FDR<1.0E-50; **FDR<1.0E-10; *FDR<1.0E-01; ns=not significant. (F) Chromatin footprint analyses indicating average islet chromatin accessibility in IRF8 (left), ATF4 (middle), or STAT1:STAT2 (right) TF binding sites of CYT-specific (left), ERS-specific (middle), or shared opening CREs (right). Number of footprints is indicated at the bottom of each plot. (G) Islet RNA-seq expression levels (CPM) in ERS, CYT, or control conditions for genes encoding TFs with enriched motifs or chromatin footprints in panels E or F. ***FDR<5%, FC≥1.5; ns=not significant. False discovery rates (FDR) calculated as in Figure 1. FC, fold-change; CPM, counts per million. See also Tables S1-2, Table S4 and Data S1.

Approximately 91% of stress-responsive CREs were distal, i.e., >1kb from transcription start site (TSS)^89,90 of the nearest expressed gene (Figure 3B; Figure S3B; Table S4), revealing an extensive role for putative enhancers as mediators of both transcriptional stress responses. We associated opening and closing distal CREs with the nearest islet-expressed genes and completed enrichment analyses (Table S4). As anticipated, stress induced concordant islet chromatin accessibility and gene expression changes (ER stress-specific: p=4.5E-16; cytokine-specific: p=1E-62; shared: p=5.9E-12; Fisher’s exact test) (Table S4). For example, we identified intronic ER stress-specific opening CREs in ERN1 and AOPEP (Figure 3C; Table S4), genes significantly induced by ER stress (Figure 3D; Table S2). ERN1 encodes IRE1α, a central ER stress sensor that initiates UPR and catalyzes unconventional splicing of the ER stress factor XBP1, while AOPEP catalyzes N-terminal peptide and amino acid hydrolysis^91–93. Cytokine-responsive CREs included those within introns of NFKB1 and CALCOCO2 (Figure 3C; Table S4), two genes induced by cytokine-induced inflammation (Figure 3D; Table S2). NFKB1 is a central mediator of inflammatory responses including in β-cells^94–96, whereas CALCOCO2 encodes a selective autophagy receptor recently identified as a putative T2D effector gene that maintains proper β-cell mitochondrial morphology, insulin granule homeostasis, and insulin content^97,98. Further, we identified stress-induced intronic CRE chromatin accessibility increases in NCKAP5 and ARID5B, genes induced by both stressors (Figure 3C-D; Tables S2-3).

To define regulatory drivers of islet ER and cytokine stress responses, we assessed transcription factor (TF) binding motif enrichment in opening distal peaks (Table S4). Motifs for ATF4 (FDR=2.20E-151), CHOP (FDR=4.40E-167), and NFIL3 (FDR=2.75E-38), transcriptional UPR mediators^99–102, were enriched in ER stress-specific opening distal peaks (Figure 3E; Table S4). In contrast, cytokine-specific opening distal peaks were enriched in motifs for interferon response factors IRF8 (FDR<2.23E-308) and IRF3 (FDR<2.23E-308), as well as the NFKB family member NFKB-p65 (alias RELA) (FDR=3.67E-68) (Figure 3E; Table S4). STAT1 (FDR=4.40E-33), BCL6 (FDR=1.52E-06), and CEBPB (FDR=1.76E-10) TF motifs were enriched in distal peaks opening in response to both islet stressors (Figure 3E). TF footprinting analyses, which integrate TF binding motifs with chromatin accessibility maps¹⁰³, confirmed significant genome-wide chromatin accessibility increases for ATF4 binding sites upon ER stress (p=2.56E-02) and IRF8 upon cytokine-induced inflammation (p=2.95E-04) (Figure 3F; Table S4). Increased accessibility at these TF binding sites was concordant with expression changes for the genes encoding these TFs: ATF4, DDIT3, and NFIL3 were induced by ER stress; IRF8, IRF3 and RELA were induced by cytokine-induced inflammation; and STAT1, BCL6, and CEBPB were induced by both stressors (Figure 3G).

We also identified concordant ER stress- and cytokine-responsive reductions in CRE accessibility and nearest gene expression (ER stress-specific: p=9E-53; cytokine-specific: p=3.9E-09; shared: p=3.5E-08; Fisher’s exact test) (Table S4). RAB27B and SLC6A17 play key roles in insulin granule exocytosis and amino acid vesicular trafficking, respectively^104–108. These genes were reduced and linked with closing CREs upon ER stress (Figures S3C-D; Table S4). Similarly, IGF1R and PCSK1, which play pivotal roles in glucose homeostasis and proinsulin-to-insulin processing, respectively^109–111, were reduced and linked with chromatin closing upon cytokine-induced inflammation (Figures S3C-D; Table S4). SORL1, involved in insulin receptor sorting¹¹² and PAX4, crucial for islet development^113,114 were reduced and linked to chromatin closing upon both stressors (Figures S3C-D; Table S4).

TF motif enrichment analysis for closing distal CREs (Table S4) revealed enrichment of PDX1 (FDR=1.33E-05) and MAFA (FDR=2.09E-04) motifs in the ER stress-specific and shared closing distal CREs, respectively (Figure S3E). We observed concordant downregulation of PDX1 and MAFA under these stress conditions (Figure S3F). Given PDX1 and MAFA are TFs central to islet identity, development, and function^68,115,116, their downregulation may signify a compensatory response to stress that impairs islet function, thereby altering glucose homeostasis and contributing to T2D.

Together, integrated comparative ATAC-seq profiling and analysis of these islet stress responses revealed that: (i) ER stress and cytokine responses in islets substantially remodel the islet epigenome, particularly modulating distal non-coding CREs (i.e., enhancers); and (ii) each stressor evokes a distinct epigenetic restructuring, mediated by different TFs (e.g., CHOP and ATF4 in ER stress; IRF8 and NFkB-p65 in cytokines) whose own expression (e.g., CHOP-encoding DDIT3, ATF4 upon ER stress) are themselves modulated by that stressor.

T2D-associated genetic variants overlap stress-responsive cis-regulatory elements

After identifying and comparing ER and cytokine stress-responsive cis-regulatory networks, we sought to identify genetic variants associated with diabetes (T2D/type 1 diabetes (T1D) GWAS) or related glycemic traits that might modulate the CREs and processes. Using a set of index and proxy variants collected from multiple GWAS studies and meta-analyses^{7–10,117–122} (Table S5; Methods), we identified 161 T2D, T1D, or related glycemic trait-associated variants that overlap stress responsive (opening or closing) CREs (Figure 4A; Figure S4A; Table S5). We identified 21 and 24 T2D-associated variants overlapping ER stress- or cytokine-specific opening CREs, respectively (Figure 4A; Table S5). Among these, 11 variants overlapped ER stress-specific opening CREs located <500kb from ER stress-responsive genes (Figure 4B; Table S5), such as AOPEP - a key gene involved in peptide processing⁹³ and robustly induced by ER stress in beta cells (Figure 4C; Tables S2-3). We detected an ER stress-specific induced intronic CRE in AOPEP that harbors T2D variant rs4744423 (Figure 4D; Tables S4-5). CRE chromatin accessibility increases correlated with rs4744423 T2D risk allele dosage (plus strand: T) (Figure 4E). The risk allele is predicted to significantly (p=3.28E-02) increase the binding affinity of TF BATF (Figure 4F; Table S5), but BATF is not islet-expressed. Further scrutiny revealed strong similarity between BATF and FOSB/JUNB heterodimer motifs (HOMER¹²³ similarity score: 0.98), and FOSB is itself an ER stress-induced islet gene (Figure 4G; Table S2). Thus, these data suggest the rs4744423 T2D risk allele may enhance FOSB/JUNB binding in ER stressed islets and concordant AOPEP induction. This is supported by increased AOPEP expression (p<1.0E-02; two-sided Wilcoxon test) in β-cells from donors with T2D (T2D) vs. without (ND) (Figure 4H) from single cell transcriptome data generated in a parallel study (see Data S1).

Figure 4. Type 2 Diabetes (T2D)-associated variants overlapping stress-induced islet CREs.

(A) Number of T2D- or glycemic trait-associated genome-wide association study (GWAS) variants overlapping opening cis-regulatory elements (CREs). (B) T2D-associated variants overlapping ER stress (ERS)-specific opening CREs <500 kb from TSS of ERS-specific induced genes. T2D association p-values were obtained from Mahajan et al. (2022)¹⁰ or Vujkovic et al. (2020)⁷ for each GWAS Source population listed. NR=Not Reported. (C) Normalized expression of putative T2D variant rs4744423 effector gene AOPEP in ERS, CYT, or control conditions in human islet RNA-seq (left) or scRNA-seq (right) profiles. Left, per-donor islet gene expression (CPM) in treated vs. control conditions. ***FDR<5%, FC≥1.5; ns=not significant. Right, AOPEP expression in α- vs. β-cell scRNA-seq profiles in ERS- or CYT-treated human islets. Dot size indicates percent of AOPEP-expressing cells per cell type; color denotes scaled average AOPEP expression in each. (D) Integrative Genomics Viewer (IGV) browser track of ERS-specific opening CRE containing T2D variant rs4744423. (E) Islet chromatin accessibility (CPM) in donors with rs4744423 TC or TT genotypes (plus strand). Homozygous T2D risk allele (TT) genotype is associated with the highest in vivo chromatin accessibility. (F) Composite logo plot illustrating atSNP¹⁷⁴ prediction that rs4744423 T2D risk allele (plus strand T, minus strand A) strengthens a BATF TF binding motif (atSNP ‘SNP impact p-value=3.28E-02). (G) Normalized per-donor islet FOSB expression (CPM) in ERS, CYT or control conditions. ***FDR<5%, FC≥1.5; ns=not significant. (H) Normalized AOPEP expression (CPM) in human islet α- or -cell pseudobulk scRNA-seq profiles of donors with T2D (T2D) vs. without (ND). **p<1.0E-02; ns=not significant, two-sided wilcoxon test. (I) Left, normalized expression (CPM) of putative rs6444081 T2D effector gene in ERS, CYT, or control conditions ***FDR<5%, FC≥1.5; ns=not significant. expression in ERS or CYT-treated islets. Dot size indicates percent -expressing cells; color denotes scaled average ETV5 expression in them. (J) IGV browser track of ERS-specific opening CRE containing T2D variants rs6444081, rs146872661, rs937563893, and rs150111048. (K) Islet chromatin accessibility (CPM) in donors with rs6444081 TT, TC or CC genotypes (plus strand). Homozygous T2D risk allele (CC) genotype is associated with the lowest in vivo chromatin accessibility. (L) Composite logo plot illustrating atSNP¹⁷⁴ prediction that rs6444081 T2D risk allele (plus strand C, minus strand G) disrupts an NFE2L2 (indicated by the position weight matrix) TF binding motif (atSNP ‘SNP impact p-value=3.88E-03). (M) Normalized per-donor islet NFE2L2 (protein alias NRF2) expression (CPM) in ERS, CYT or control conditions. **FDR<5%, FC>1. False discovery rates (FDR) calculated as in Figure 1. FC, fold-change; CPM, counts per million. See also Tables S1-5 and Data S1.

Similarly, we identified an ER stress-induced CRE harboring T2D variant rs6444081 whose putative effector gene is ETV5, an insulin secretion modulator^124–126 induced by ER stress in β-cells (Figures 4I-J; Tables S2-5). T2D rs6444081 risk allele (plus strand: C) was associated with reduced CRE accessibility (Figure 4K) and predicted to significantly disrupt an NFE2L2/NRF2 TF binding motif (p=3.88E-03, Figure 4L; Table S5), which we previously identified as a putative islet chromatin accessibility regulator¹²⁷. NFE2L2 was ER stress-induced (Figure 4M), and together with KEAP1^128–130, may facilitate stress-responsive ETV5 activation. These data, therefore, suggest that ER stress increases rs6444081-harboring CRE accessibility and regulates ETV5 activation. However, our data indicate T2D rs6444081-C risk allele reduces chromatin accessibility, presumably by disrupting NRF2 binding, which would contribute to diminished ETV5 induction and islet dysfunction or death. Indeed, Etv5^−/− mice exhibit impaired insulin secretion and glucose tolerance defects¹²⁴. Etv5^−/− islets are smaller and contain smaller β-cells than those from wildtype littermates¹²⁴, and reduced ETV5 expression was previously reported in T2D vs. ND islets¹²⁵.

We also found 14 T2D-associated variants overlapping 11 cytokine-specific opening CREs <500kb from cytokine-induced genes (Figure S4B; Table S5), including β-cell cytokine-induced GALNT15 (Figure S4C; Tables S2-3) - a member of the GALNT family involved in protein metabolism^131,132. In addition, we identified a cytokine-induced intronic CRE in ANKRD28 containing T2D variant rs4685264 (Figure S4D). T2D rs4685264 risk allele (plus strand: G) was associated with increased CRE chromatin accessibility (Figure S4E) and significantly (p=2.84E-03) increased predicted binding affinity for the TF MAX (Figure S4F; Table S5), which was itself a cytokine-induced gene (Figure S4G; Table S2). These data suggest that cytokine exposure increases accessibility of the CRE harboring rs4685264, enabling MAX binding. We posit that the T2D risk allele MAX binding and induction of the putative T2D effector gene GALNT15 to alter protein metabolism in cytokine-exposed islets. Together, these analyses nominated putative T2D variants with context-specific effects on transcriptional processes modulating islet responses to ER stress and cytokine-induced inflammation.

Variant-to-function dissection of ER stress-responsive T2D variant in the SLC35D3 locus

Integrated islet multi-omic data analysis converged to provide variant-to-function insights for T2D-associated variant rs6917676, which overlapped an ER stress-induced CRE residing in an islet enhancer hub¹³³ spanning ~479 kb on chromosome 6 (Figure 5A, magenta bracket). The rs6917676 T2D risk allele (plus strand: T) was associated with increased chromatin accessibility of this ER stress-responsive CRE (Figure 5B, compare GG, TG, and TT islet donors). We previously identified rs6917676 as the expression-modulating variant (emVar) in this CRE using massively parallel reporter assays (MPRA) in mouse MIN6 β-cells, with the T risk allele increasing MPRA activity¹⁵. The rs6917676-T risk allele is predicted to strengthen binding of NFIL3 (Figure 5C; Table S5), which was itself induced by ER stress in β-cells (Figure 5D; Tables S2-3). To test if rs6917676-T risk allele was differentially bound by different beta cell nuclear/transcription factor(s) as predicted, we completed electrophoretic mobility shift assays (EMSAs)¹³⁴ using human EndoC-βH3 nuclear extracts (Figure 5E). EMSA revealed robust T allele-specific binding (red arrows) in untreated, ER stressed, and DMSO solvent control β-cell extracts, consistent with detectable NFIL3 protein in EndoC-βH3 cell nuclear extracts for each of these conditions (Figure S5B).

Figure 5. Variant-to-function analyses link rs6917676 to ER stress-responsive CRE accessibility, MAP3K5 expression, and increased β-cell apoptosis.

(A) Integrated Genomics Viewer (IGV) browser track of 1 Mb window (blue gene annotations) centered on T2D-associated variant rs6917676 and enhancer hub (magenta) identified by Miguel-Escalada et al. (2019)¹³³. Enhancer hub encompasses multiple enhancers, with one (inset) mapping to an ERS-specific, opening distal CRE containing T2D variants rs6937795 and rs6917676. All genes within the 1Mb interval are shown; ERS-induced (green check), non-responsive (red “X”), or non-expressed/non-protein coding (gray text) are indicated. (B) Normalized islet chromatin accessibility levels (CPM) in donors, stratified by rs6917676 plus strand genotype (GG, TG or TT). Note in vivo chromatin accessibility increases with T2D risk allele (T). (C) Composite logo plot illustrating atSNP¹⁷⁴ prediction that rs6917676 T2D risk allele (plus strand T) creates an NFIL3 TF binding motif ( atSNP ‘SNP impact p-value=2.37E-03). (D) Left, normalized per-donor NFIL3 expression (CPM) in ERS- or CYT-treated islets versus their respective controls. ***FDR<5%, FC≥1.5; ns=not significant. Right, α- or β-cell scRNA-seq NFIL3 expression in ERS- or CYT-treated islets (right). Dot size indicates percent NFIL3-expressing cells per cell type; color indicates scaled average NFIL3 expression in each. (E) Electrophoretic mobility shift assay (EMSA) using untreated (Untx), DMSO solvent control, or thapsigargin (Tg)-treated human EndoC-βH3 cells nuclear extracts (NE). Red arrows highlight nuclear factors specifically binding T2D risk allele rs6917676-T. Representative image from n=3 EMSAs. (F) Normalized per-donor expression (CPM) of putative rs6917676 T2D effector gene MAP3K5 in human islets in ERS, pro-inflammatory cytokine (CYT), or control conditions. ***FDR<5%, FC≥1.5; ns=not significant. (G) α- vs. β-cell MAP3K5 scRNA-seq in ERS- or CYT-treated human islets. Dot size indicates percent MAP3K5-expressing cells in each cell type; color indicates scaled average MAP3K5 expression in each. (H) Normalized MAP3K5 expression (CPM) in human islet α- vs. β-cell pseudobulk scRNA-seq profiles from donors with T2D (T2D) vs. without (ND). **p<1.0E-02; ns=not significant, two-sided Wilcoxon test. (I) Correlation plot demonstrating significant (p=9.30E-04) inverse relationship between normalized MAP3K5 expression (CPM) and β-cell-to-endocrine cell proportions for each of 48 human islet donors from a separate study (see Data S1). (J) Percent apoptosis (Annexin V+ cells) detected in human EndoC-βH3 cells exposed to 500 nM thapsigargin vs. DMSO solvent control after MAP3K5 (shMAP3K5) or non-targeting shRNA control (nt-shCTRL) knockdown. n=5 biological replicates per condition. **p<1.0E-02; ns=not significant, unpaired two-tailed Student’s t-test. (K) Percent apoptotic (Annexin V+) cells detected in human islets from five donors incubated with MAP3K5 inhibitor Selonsertib (2 μM) or left untreated (0 μM) and exposed to glucolipotoxic (+Palmitate; 0.4 mM Palmitate +25 mM glucose) or control (-Palmitate; 25mM glucose) conditions for 72 hours. *p<0.05; ns=not significant, paired two-tailed Student’s t-test. False discovery rates (FDR) calculated as in Figure 1. FC, fold-change. See also Tables S1-5 and Data S1.

Islet promoter capture Hi-C (pcHi-C) data¹³³ link this CRE to MAP3K5, PEX7, and IL20RA promoters (Figure 5A). MAP3K5 was the only of these and other genes in the locus (MAP7, SLC35D3, and IFNGR1) induced by ER stress – specifically in beta cells – thereby nominating it as the likely effector gene of this variant (Figures 5F-G; Figure S5A; Tables S2-3). These data suggest T2D risk allele rs6917676-T contributes to islet dysfunction or death by increasing ER stress-responsive MAP3K5 expression via increased NFIL3 binding at this ER stress-responsive opening CRE. Consistently, we detected increased MAP3K5 expression (p<1.0E-02; two-sided Wilcoxon test) in T2D vs. ND donor β-cells (Figure 5H) from a parallel islet single cell transcriptome study (see Data S1).

MAP3K5 encodes ASK1, a pro-apoptotic MAPK kinase reported as ER stress-responsive in MIN6 β cells¹³⁵ that phosphorylates JNK and p38¹³⁶. Ask1/Map3k5 knockdown or germline deletion increases MIN6 survival and reduces islet caspase activity, respectively¹³⁵. Higher MAP3K5 expression in ER stress-BC1 vs. apoptosis-related ER stress-BC2 clusters suggest that elevated MAP3K5 expression may promote or enhance an apoptotic fate in ER-stressed β-cells, after which it is not required (Figure S5C; Table S3). Consistent with this pro-apoptotic role, MAP3K5 expression was significantly associated with reduced beta cell percentages in human islets (Figure 5I). Therefore, to test if MAP3K5 modulates ER stress-responsive apoptosis in human beta cells, we assessed if MAP3K5 shRNA knockdown altered apoptosis in EndoC-βH3 cells exposed to a (patho)physiologic range of thapsigargin concentrations (125–2000 nM). We achieved ~80% knockdown of MAP3K5 (Figures S5D-E), and western blot analyses confirmed that this knockdown nearly abrogated ER stress-induced phosphorylation of p38, one of its targets^137,138, in EndoC-βH3 cells (Figures S5F-G). Importantly, MAP3K5-deficient cells exhibited significantly fewer apoptotic (Annexin V+) cells compared to the non-targeting shRNA control cells exposed to pathophysiologic thapsigargin concentrations (Figure 5J; Figure S5E). These data align with in vivo models of diabetes, where germline Map3k5/Ask1 deletion significantly reduced beta cell apoptosis in Akita mice carrying a permanent neonatal diabetes- (PNDM) linked and islet ER stress-inducing Ins2^C96Y mutation¹³⁹.

Next, we tested if pharmacologic MAP3K5 inhibition reduced stress-responsive apoptosis of primary human islets (Table S1) exposed to glucolipotoxicity, another T2D-associated pathophysiologic stress condition, using Selonsertib. Selonsertib is a MAP3K5 (alias ASK1) inhibitor¹⁴⁰, and its structural analog GS-444217^141,142 have been shown to improve diabetic nephropathy by targeting p38 in pre-clinical rodent models of diabetes^143,144. Randomized placebo-controlled double-blind Phase 2 clinical trials (Clinical Trial Identifier: NCT04026165)¹⁴⁵ for diabetic complications, such as diabetic kidney disease^146,147, have been completed¹⁴⁸, and Selonsertib has been approved for, and is currently undergoing, Phase 3 clinical trials to prevent/treat moderate to advanced diabetic nephropathy^149–151. As anticipated, glucolipotoxic exposure increased apoptosis in primary human islets from five donors (Figure 5K; Figures S5H-I; compare percent Annexin V+ cells in ‘+Palmitate, 0 μM Selonsertib’ vs. ‘-Palmitate 0 μM Selonsertib’, p=0.045). Importantly, Selonsertib treatment significantly reduced glucolipotoxicity-induced apoptosis in these donor islets (Figure 5K; Figures S5H-I; compare Annexin V+ cells in ‘+Palmitate, 0 μM Selonsertib’ vs. ‘+Palmitate, 2 μM Selonsertib’, p=0.047). Taken together, these variant-to-function analyses suggest that the T2D-associated rs6917676-T risk allele contributes to T2D risk or progression by enhancing ER stress-responsive islet CRE accessibility and MAP3K5 expression to promote, sensitize, or enhance stress-responsive islet beta cell apoptosis.

DISCUSSION

This study provides genome-wide insights into the transcriptional regulatory circuitry mediating pancreatic islet stress responses, particularly to ER stress and pro-inflammatory cytokines, two pathophysiologic stressors implicated in T2D pathogenesis. Through comprehensive RNA-seq and ATAC-seq analyses, we identified distinct sets of genes and CREs responsive to ER stress and cytokine-induced inflammation. The majority of stress-responsive genes and CREs were specific to either ER stress or cytokines. scRNA-seq analyses uncovered α- and β-cell specificity of these responses. Context-specific islet response to ER stress and cytokines are intriguing yet not entirely unexpected. The specificity likely reflects a finely tuned cellular mechanism allowing islets to adapt and tailor responses to diverse pathophysiologic stimuli. For example, ER stress predominantly triggered pathways related to protein folding and secretion, both crucial for β-cells’ insulin-producing function¹⁵². In contrast, cytokine treatment activated pro-inflammatory and signaling pathways that can interfere with islet identity and function^58,153.

scRNA-seq profiling of these human islet stress responses revealed cell type-specificity of responses to ER stress and cytokine-induced inflammation. Although these data suggest β-cells respond to stress more extensively than α-cells, they also indicate α-cells are not devoid of response when faced with stress. As previous studies have primarily focused on β-cells due to their direct role in insulin production and extensive links to diabetes pathology^15,153–164, comparative scarcity of α-cell pathophysiologic studies underscores the significance of our findings and suggests that they, too, are not unaffected bystanders to islet stress. Recent studies have begun to uncover potential α-cell roles in islet pathological responses in diabetes. For example, α-cells exhibit autoimmune-induced ER stress signatures in T1D¹⁶⁵, yet are protected from apoptosis under metabolic stress^166,167. It was suggested this could be due to higher α- vs. β-cell production of pro-survival proteins such as BCL2L1^166,168. Supporting these reports, basal BCL2L1 expression is higher in α- vs. β-cells (Table S3). These emerging data warrant further islet α-cell biologic studies and scrutiny of their putative role(s) in diabetes risk and progression. Our data also uncovered heterogeneity in β-cell responses to ER stress, marked by the presence of two transcriptionally distinct putative β-cell subpopulations. One subset (ER stress-BC1) reflected the activation of bona fide ER stress response genes and pathways (e.g., DDIT3 and ATF4), whereas the smaller subset (ER stress-BC2) featured induction of apoptosis-related genes (e.g., PSMB8 and PSMB9). Interestingly, this apoptotic β-cell subpopulation was detected in all donors. These findings suggest a fraction of β-cells are prone to ER stress-induced cell death, which could contribute to β-cell death associated with T2D^169,170.

Epigenetic stress responses were concentrated in distal regulatory regions of the genome, highlighting the noncoding genome’s importance in modulating cellular stress responses. Stress-responsive opening CREs were enriched in the binding sites of critical TFs (e.g., ATF4 upon ER stress, IRF8 upon cytokine-induced inflammation). Genes encoding these TFs were also activated upon these stressors, suggesting that cellular responses are tightly regulated at the epigenetic level by the activation of critical TFs as well as by the increased chromatin accessibility at their binding sites. By overlapping the locations of T2D-associated genetic variants and stress-responsive islet CREs, we uncovered 52 variants residing in 38 ER or cytokine stress-induced CREs which we hypothesize contribute to T2D etiology by altering their use or activity. Although the identification of stress-responsive CREs, their overlap with T2D-associated variants, and targeted allelic analyses implicate this subset of T2D variants as genetic modulators of these responses, larger sample sizes are needed to formally demonstrate their allelic effects on stress-responsive chromatin accessibility and gene expression using allelic imbalance or quantitative trait locus approaches. Additionally, the exploration of additional T2D-associated pathophysiologic stressors (e.g., glucolipotoxicity or oxidative stress) or various stimuli, and their interaction with genetic variants, could further help elucidate the complex molecular landscape of T2D, stratify T2D association variants/signals into functional bins, and identify therapeutic gene targets and pathways.

For a subset of loci, this study nominated putative causal T2D variants and provided a range of variant-to-function insights into their context-specific effects on islet ER stress- and cytokine-responsive programs. This includes rs6917676, which overlaps an ER stress-responsive CRE. Human islet pcHi-C data from Ferrer and colleagues converge with RNA-seq from this study to nominate MAP3K5 as the target gene of this CRE and the T2D effector gene for this genetic association signal. MAP3K5 encodes MAP3K5 (alias ASK1), a kinase that promotes apoptosis via activation of JNK and p38 signaling pathways^{135,171–173}. Our variant-to-function dissection of this locus suggests a mechanistic model in which the rs6917676 T2D risk allele (T) increases ER stress-induced CRE accessibility and MAP3K5 expression to promote excessive apoptosis and exacerbate β-cell loss in T2D. The apoptosis-promoting association of elevated MAP3K5 in stressed β-cells seems to be reflected in the scRNA-seq data, wherein the apoptosis-related ER stress-BC2 beta cluster had reduced MAP3K5 expression compared to the ER stress-BC1 cluster, and it fits with previous in vivo studies demonstrating that germline Map3k5/Ask1 deletion significantly reduces islet β-cell apoptosis in Ins2^Akita mice harboring a permanent neonatal diabetes (PNDM)-linked mutation that causes proinsulin misfolding and ER stress¹³⁹. Moreover, MAP3K5 expression inversely correlated with β-cell abundance in a 48-donor islet scRNA-seq cohort, and T2D donors in this cohort exhibited higher β-cell MAP3K5 expression and significantly fewer β-cells comprising their islets. Finally, pharmacologic MAP3K5 inhibition reduced glucolipotoxicity-induced islet apoptosis, suggesting that: (i) elevated MAP3K5 expression or activity may contribute more broadly to T2D-associated pathophysiologic stress-induced apoptosis; and (ii) Selonsertib, the MAP3K5 inhibitor in Phase II/III clinical trials to prevent or treat diabetic nephropathy, might be an effective primary intervention to combat progression to T2D by preserving mass and function of stressed β-cells. More broadly, these findings highlight the significance of studying GWAS variants in the context of stress conditions, which more closely reflect cellular state(s) during disease, including for T2D.

LIMITATIONS OF STUDY

Despite uncovering islet responses to T2D-associated pathophysiologic factors, there are potential limitations. First, the focus on a single time point may not represent fully these stress response dynamics nor capture the compendium of stress-responsive CREs and genes engaged by chronic stressors. Future studies could employ longitudinal designs for more comprehensive understanding of these temporal cellular adaptation mechanisms. Second, although thapsigargin^15,35–38 and IL1β+IFNγ^39–41 exposure to model T2D-relevant islet ER or cytokine responses, respectively, is based on precedent in the literature, they likely miss features of the complex, progressive nature of in vivo islet dysfunction in T2D. This may limit direct translational relevance of our findings. Future studies employing more sophisticated organoid, pancreatic slice, or animal models with transplanted human (pseudo)islets should help to better understand the intricate in vivo milieu contributing to islet dysfunction in T2D.

Despite these potential limitations, the comprehensive, comparative multi-omic mapping we describe provides important mechanistic insights into human islet responses to two (patho)physiologic stressors: ER stress and cytokine-induced inflammation. Importantly, these maps enabled nomination of candidate causal T2D variants that likely contribute to T2D risk or progression by modulating these responses. These findings support growing literature revealing importance of cell- and context-specific responses in the pathophysiology of, and approaches to combat, islet dysfunction in T2D. This study both enhances understanding of T2D pathogenesis and provides potential genetics-based avenues or insights, such as repurposing MAP3K5/ASK1 inhibitors (e.g., Selonsertib) to combat ER stress-induced β-cell apoptosis, as targeted interventions to preserve β-cell function under pathophysiologic ER stress.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by Lead Contact Michael L. Stitzel (michael.stitzel@jax.org).

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

• All cadaveric human islet ATAC-seq, RNA-seq, and scRNA-seq data generated in this study are available via the Gene Expression Omnibus under study accession GSE251913.

• Data used to generate manuscript figures are provided in Tables S1-5, Data S1, and at https://doi.org/10.5281/zenodo.13730232.

• Scripts used to generate manuscript figures can be found at https://github.com/UcarLab/ERS-vs-CYT-Islet-Stress-Responses.

• This paper does not report original code.

• Any additional information required to reanalyze data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Study Subjects and Primary Islet Culture

Fresh human cadaveric pancreatic islets were procured from Prodo Labs or the Integrated Islet Distribution Program (IIDP) and were required to have a minimum of 80% purity as measured by dithizone and 90% viability. Because the islets are from de-identified cadaveric organ donors, the study was determined to be human subjects exempt. The characteristics of islet donors are listed in Table S1. The mean and standard deviation in donor age was 47.6 ± 12.4 years. Mean and standard deviation in donor BMI was 30.1 ± 7.1. Islets were obtained from 22 males and 13 females. Upon arrival, cells were transferred into PIM(S) media (Prodo Labs) supplemented with PIM(ABS) (Prodo Labs) and PIM(G) (Prodo Labs) and incubated in a T-150 non-tissue culture treated flask (VWR) for recovery at 37°C and 5% CO₂ overnight. The following day, media was changed to CMRL (10% FBS, 1% Glutamax) supplemented with either 0.025%v DMSO, 250nM thapsigargin or 25 U/mL of IL1β + 1000 U/mL of IFNγ (R&D Systems). After 24-hr incubation at 37°C and 5% CO₂, nuclei and total RNA were isolated for RNA-seq and ATAC-seq library preparation as previously described¹²⁷.

EndoC-βH3 Human β-cell Line

For validation experiments, we used EndoC-βH3 (CVCL_IS72; age at sampling 9FW; sex unspecified) an engineered cell line resulting from the lentiviral transfection of SV40LT expressed under the control of the insulin promoter and hTERT that can be excised by the Cre recombinase upon addition of tamixofen¹⁷⁵. This cell line has been shown to respond to glucose stimulus by secreting insulin and serves as an adequate model for studying human β-cells. EndoC- βH3 cells were obtained from Human Cell Design (1 place Pierre Potier, 31106 Toulouse, France) and cultured as described previously¹⁷⁵. Briefly, EndoC- βH3 cells were maintained at 37C and 5% CO 2 atmosphere, in Advanced DMEM/F12 media (Gibco) consisting of 3g/L of D-glucose, supplemented with 2mM Glutamax (Gibco), 100 units/mL penicillin (Gibco), 100 ug/mL streptomycin (Gibco), 2% albumin from bovine fraction V (SIGMA), 50 uM 2-mercaptoethanol (SIGMA), 10 mM nicotinamide (SIGMA), 6.7 ng/mL sodium selenite (SIGMA) and 10ug/ml Puromycin (Calbiochem) on ECM (Sigma) and Fibronectin (Sigma) coated flasks¹⁷⁵.

METHOD DETAILS

RNA-seq Library Preparation and Sequencing

Human islet RNA-seq libraries were prepared from total RNA using the stranded TruSeq kit (Illumina). ERCC Mix 1 or Mix 2 spike-ins were randomly added to each sample (Thermo Fisher, catalog #4456740) and sequenced on Illumina NovaSeq S4 as previously described¹²⁷. The paired-end (2×150 bp) RNA-seq FASTQ files for each islet were aligned against the human genome (GRCh38/hg38) using STAR¹⁷⁶ and counts were generated using QoRTs¹⁷⁷ (Table S2; Data S1).

RNA-seq Analyses

Genes were annotated using Ensembl¹⁷⁸, and only genes in autosomal chromosomes were considered for downstream analysis. Non-protein coding genes in autosomal chromosomes, including RNA and pseudogenes (annotated as ‘transcribed_unprocessed_pseudogene’, ‘processed_pseudogene’, ‘lncRNA’, ‘unprocessed_pseudogene’, ‘TR_V_pseudogene, snRNA’, ‘misc_RNA’, ‘rRNA_pseudogene’, ‘IG_V_pseudogene’, ‘IG_C_pseudogene’, ‘TEC, scRNA’, ‘translated_processed_pseudogene’, ‘vault_RNA’, ‘sRNA’, ‘pseudogene’, ‘transcribed_unitary_pseudogene’, ‘transcribed_processed_pseudogene’, ‘unitary_pseudogene’, ‘miRNA’, ‘snoRNA’, ‘rRNA’, ‘TR_J_pseudogene’, ‘ribozyme’, ‘IG_J_pseudogene’, ‘scaRNA’, ‘translated_unprocessed_pseudogene’, and ‘IG_pseudogene’) were filtered out. The remaining (protein coding) genes were then filtered for expression by requiring >0 CPM in ≥8 samples, and ERCCs were filtered for expression by requiring >5 reads in ≥2 samples. Normalization of protein coding genes with ERCC was performed using RUVSeq¹⁷⁹, which also estimated unwanted variation (W_1) in the data. Surrogate variable analysis was then performed using svaseq¹⁸⁰, and the surrogate variables that explained >10% of variance in the data (n=3) were considered in downstream analysis. Genes were then tested for differential expression (FDR<5%; |LFC|≥0.585) between their respective control (DMSO; untreated) and treatment (thapsigargin; IL-1β+IFN-γ) conditions (Table S2), with gene expression adjusted for age, sex, batch, BMI, surrogate variables and W_1, using edgeR’s¹⁸¹ ‘tagwise’ and robust dispersion estimation parameter on TMM normalized counts. FDR was calculated using Benjamini-Hochberg p-value adjustment. Differentially expressed (DE) genes were classified as specific or shared using a Venn diagram, and were input into DAVID¹⁸² to find the enriched pathways (FDR<10%) using KEGG¹⁸³, Reactome¹⁸⁴, and WikiPathways¹⁸⁵ (Table S2).

scRNA-seq Library Preparation and Sequencing

After a 24-hr treatment, as described above, islets were treated with Accutase for 8–10 min at 37°C to generate a single cell suspension. Cells were then washed and suspended in Staining buffer (PBA, 2%BSA, 0.01%TweenS), and immediately processed as follows: Incubate with Fc Blocking reagent (FcX, BioLegend) 10 minutes at 4 °C before adding 0.5ug of a unique Cell Hashing antibody (TotalseqTM-A0251 to A0257 anti-human hashtag antibody, Biolegend) for 20 minutes at 4 °C, then wash cells 2 times with Staining buffer and once with PBS+0.04%BSA. Cell viability was assessed on a Countess II automated cell counter (ThermoFisher), and up to 30,000 cells (~5,000 cells from each hash-tagged (HTO) sample) were loaded onto one lane of a 10X Chromium Controller. One single cell suspension was loaded twice, i.e. added to two lanes of the 10x chip, and two gene expression and two HTO libraries were generated in total, which were processed into a single set of gene expression and HTO outputs. Single cell capture, barcoding, and library preparation were performed using the 10X Chromium platform V3 chemistry according to the manufacturer’s protocol (#GC000103). cDNA and libraries were checked for quality on Agilent 4200 Tapestation, quantified by KAPA qPCR, and pooled and sequenced on an Illumina NovaSeq 6000 S2/S4 flow cell lane, targeting an average sequencing depth of 50,000 reads per cell. Illumina base call files for all libraries were converted to FASTQ using Illumina’s bcl2fastq¹⁸⁶. FASTQ files were associated with the gene expression libraries, aligned to the GRCh38.93 reference genome and merged, including all transcribed unitary pseudogenes, using the 10x Genomics Cell Ranger’s count pipeline^187,188. FASTQ files representing the HTO libraries were then processed into hashtag-count matrices using CITE-seq-Count¹⁸⁹ (Table S3; Data S1).

scRNA-seq Clustering and Annotation

Sample identities were determined using demuxlet¹⁹⁰. Ambient RNA for each islet was removed using SoupX¹⁹¹ by setting contamination fraction to 20%. These SoupX ‘cleaned’ data were then demultiplexed based on enrichment of HTO using Seurat¹⁹². Only cells with >2000 genes <40% mitochondrial mapping were included in downstream analysis. Doublet cells were then identified using Scrublet¹⁹³ and removed. To remove any remaining potential doublets or multiplets, cells in the >0.95 quantile with respect to the number of genes expressed were discarded (Table S3). These data were then merged into a single object using Seurat¹⁹² and corrected for batch effect using Harmony¹⁹⁴. Seurat’s¹⁹² ‘FindClusters’ was implemented to identify cell clusters, which were then annotated for cell type identity using islet marker genes (Table S3). Seurat clusters expressing more than one marker gene were classified as doublets and removed from downstream analyses.

scRNA-seq Data Analyses

To generate response scores, differentially expressed (DE) genes from bulk data were curated into ER stress-specific, CYT-specific and shared response gene modules. UCell’s¹⁹⁵ ‘AddModuleScore_UCell’ was then used to calculate each module’s enrichment (i.e. response) scores (Table S3). To identify expressed genes between the control (DMSO; untreated) and treatment (thapsigargin; IL-1β+IFN-γ) conditions in alpha and β-cells, Seurat’s¹⁹² ‘FindMarkers’ was implemented using the MAST¹⁹⁶ test and adjusted with respect to batch and disease state (Table S3). This methodology was also implemented on only those genes that were detected in a minimum of 10% of cells in either of the populations being compared to identify DE genes between (i) ER stress-BC1 vs. DMSO (FDR<5% and |LFC|≥0.585), (ii) ER stress-BC2 vs. DMSO (FDR<5% and |LFC|≥0.585), (iii) ER stress-BC1 vs. ER stress-BC2 (FDR<5%), and (iv) thapsigargin vs. DMSO (FDR<5% and |LFC|≥0.585) in alpha and β-cells comparisons (Table S3). To find the alpha-specific, beta-specific, and common DE genes upon ER stress, a Venn diagram was created (Table S3). DE genes were input into DAVID¹⁸² to determine significantly enriched pathways (FDR<10%) using KEGG¹⁸³, Reactome¹⁸⁴, and WikiPathways¹⁸⁵ (Table S3).

ATAC-seq Library Preparation and Sequencing

Human islet ATAC-seq libraries were prepared following the Active motif ATAC prep kit (Active motif catalog# 53150). Briefly, 50 islet equivalents (50,000 cells) per sample were transposed in triplicate, libraries were barcoded, pooled into three-islet batches, and sequenced using 2 × 150 bp Illumina NovaSeq S4 chemistry as previously described¹²⁷. Paired-end (2×150 bp) ATAC-seq FASTQ files for each islet were trimmed using Trimmomatic¹⁹⁷ and aligned against the human genome (GRCh38/hg38) using BWA-MEM¹⁹⁸. Duplicate reads were removed, and the remaining reads were shifted as previously described^199,200. Using SAMtools²⁰¹, technical replicates were merged and peaks were called using the MACS2²⁰² ‘BAMPE’ parameter. TDF files were generated using IGVTools²⁰³ to visualize peaks on IGV²⁰³. Separate consensus peak sets for ER stress and CYT samples were generated by considering peaks that were present in at least 2 samples, and those peaks that mapped to ENCODE DAC Exclusion List Regions²⁰⁴ were excluded using DiffBind²⁰⁵. The union of all peaks from ER stress and CYT samples was then determined using GenomicRanges²⁰⁶, and counts were normalized using CPM (Table S4; Data S1).

ATAC-seq Data Analyses

Only peaks in autosomal chromosomes were considered, which were then filtered for depth by requiring >0 CPM in ≥8 samples. Surrogate variable analysis was then performed using svaseq¹⁸⁰, and the surrogate variables that explained >10% of variance in the data (ER stress: n=2; cytokines: n=3) were considered in downstream analysis. Peaks were then tested for differential accessibility (FDR<5%) between their respective control (DMSO; untreated) and treatment (thapsigargin; IL-1β+IFN-γ) conditions (Table S4), with accessibility adjusted for age, sex, batch, BMI, and surrogate variables using edgeR’s¹⁸¹ ‘tagwise’ and robust dispersion estimation parameter on TMM normalized counts. FDR was calculated using Benjamini-Hochberg p-value adjustment. Peaks were then annotated to the nearest expressed protein-coding gene extracted from GENCODE v35²⁰⁷ in islets using HOMER’s¹²³ ‘annotatePeaks.pl’ command. Peaks with distance ≤1kb to the nearest expressed gene’s TSS were considered proximal, and the other peaks were considered distal^89,90. IRange’s²⁰⁶‘subsetByOverlaps’ function was then used to classify proximal and distal peaks as specific or shared (Table S4).

Enrichment and Footprinting Analysis

Nearest genes to differentially accessible (DA) peaks were used as input into DAVID¹⁸² to find the enriched pathways (FDR<10%) using KEGG¹⁸³, Reactome¹⁸⁴, and WikiPathways¹⁸⁵ (Table S4). TF motifs present in the DA peaks were found using HOMER’s¹²³ ‘findMotifsGenome.pl’ command. FDR of TFs was calculated using Benjamini-Hochberg p-value adjustment, and the fold change was calculated by dividing `% of Targets Sequences with Motif’ by the `% of Background Sequences with Motif` (Table S4). Using SAMtools²⁰¹, samples of the same control or treatment conditions were merged, and peaks were called using MACS2²⁰² with ‘BAMPE’ parameter. HINT-ATAC¹⁰³ was used to identify TF footprints and to calculate differences in TF activity between the respective control and treatment conditions (Table S4).

Overlapping Genetic Variants with Peaks

Index variants associated with T1D, T2D, and glycemic traits (fasting glucose, fasting insulin, HbA1c, 2-hour glucose, HOMA-B, HOMA-IR, proinsulin, modified Stumvoll insulin sensitivity index, and disposition index) were obtained from the largest and most recent genome-wide association meta-analyses for each trait^{7–10,117–122} (Table S5). Proxy variants in strong linkage disequilibrium (LD) with the index variant were defined as any variant with r²≥0.75 with the index variant calculated using the 1000 Genomes Phase 3 reference panel²⁰⁸, accessed through ‘https://ldlink.nih.gov’, and the global ancestry group that most closely matched the original GWAS meta-analysis. All individuals in the reference panel were used for GWAS meta-analyses of multi-ancestry populations. When necessary, index and proxy variants were lifted over to hg38 genome. Variants that did not lift over were excluded from further analyses (Table S5). Index and proxy variants with a reference SNP ID (rsID) assigned by dbSNP²⁰⁹, accessed through ‘https://www.ncbi.nlm.nih.gov/snp/’, were then overlapped with DA peaks using IRange’s²⁰⁶ ‘findOverlapPairs’ function to identify T2D variants that overlap stress-responsive peaks (Table S5). T2D variants overlapping stress-responsive peaks and located <500kb from the nearest up- or downregulated gene were then entered into atSNP¹⁷⁴, accessed through ‘http://atsnp.biostat.wisc.edu’, to identify all TF motifs being disrupted by the variant in the sense or antisense strand with a ‘SNP impact p-value’ <0.05. This list of motifs was compared to the list of enriched TF motifs identified by HOMER¹²³ (as described above) to determine relevant TF motifs (Table S5). ATAC-seq read pileups were used to infer the genotypes of donors for the T2D variants using pysam^201,210 (Table S5).

shRNA knockdown in EndoC-βH3

Plasmid pLKO-puro shRNA clones (Mission shRNA) were purchased from Sigma (SHC016 (shCTRL); TRCN0000000993 (shMAP3K5). Lentivirus was produced in HEK293T cells co-expressing the shRNA plasmid together with psPAX2 packaging plasmid and pVSVG envelope plasmid (Addgene). Virus was concentrated using Lenti-X Concentrator (Takara) and titer quantified using p24 ELISA antigen assay (Takara). MOI=5 was used to transduce 1×10²¹¹ EndoC-βH3 cells in culture media without pen/strep and puromycin. Cells were collected for RNA extraction 96 hrs post transduction using TRIZOL (Invitrogen), phase separation was achieved using Chloroform. Isopropanol was used for RNA precipitation using glycogen as a carrier, the pellets were washed using 75% ethanol, air-dried, and resuspended in DEPC water. RNA was measured using Qubit RNA HS Assay (Applied Biosystems). Total RNA was used to perform qPCR using RNA to CT kit (Applied Biosystems) and FAM-Taqman probes (Invitrogen) and analyzed on QuantStudio 7 (Applied Biosystems) normalized to TBP/HPRT1 Taqman probes (Figure S5D; Data S1).

Flow cytometry analysis of EndoC-βH3 cells

Eighteen hours post-transduction, media was changed to pen/strep and puromycin complete media with 0, 125, 250, 500, 1000, 1500, 2000 nM thapsigargin (Sigma Aldrich) dissolved in DMSO or 0.5% DMSO solvent control (VWR). Ninety hours after transduction, cells were collected using Trypsin (Gibco) and stained using PE-Annexin V Apoptosis Detection Kit (BioLegend) according to manufacturer’s instructions. The samples were assessed on Fortessa (BD Sciences) and Annexin V-positive cells were analyzed and quantified using FlowJo Software (BD Sciences) (Figure 5J; Figure S5E; Data S1).

Electrophoretic mobility shift assay (EMSA)

EMSAs were completed as previously described¹³⁴. Nuclear extracts were prepared from EndoC-βH3 cells using NE-PER Extraction kit (PIERCE), quantified using Pierce BCA protein assay kit (PIERCE), and stored in −80°C until use. 21-bp biotin end-labeled, and unlabeled complementary oligonucleotides were designed to the variant rs6917676 (5’Biosg/TAATGACTGT[G/T]TTCTTAAGAT-3’, Integrated DNA Technologies), and double stranded probes were generated for both alleles. The Lightshift EMSA optimization and control kit (Thermo Scientific) was used according to the manufacturer’s instructions. Each reaction consisted of a 10x binding buffer, Poly Di-Dc, 4μg of nuclear extract, and 200nM of labeled probe. Reactions were incubated at 25°C for 25 minutes and separated on a 6% TBE gel (Invitrogen). DNA-protein complexes were detected using Lightshift Chemiluminescent Nucleic Acid Detection kit (PIERCE) according to manufacturer’s protocol. EMSAs were repeated at least three times and yielded comparable results. See Data S1.

Cell extracts and western blotting

Nuclear extracts were prepared from EndoC-βH3 cells after 24-hour incubation with 500 nM thapsigargin or DMSO solvent control using the NE-PER Extraction kit (PIERCE) and quantified using BCA protein assay kit (PIERCE). Ten micrograms of EndoC-BH3 cytosolic or nuclear extracts were loaded on a 4–12% Tris-Glycine gel (Biorad). The proteins were transferred onto a 0.2 uM PVDF membrane (Biorad) using Turboblot (Biorad). The membrane was blocked with 5% Milk in TBST solution (Tris buffer saline + 0.005% Tween 20) for 1 hour at room temperature and incubated with primary antibodies (STAR Methods) in 5% BSA in TBST solution at 4°C overnight. The membrane was washed three times with TBST and incubated with the secondary HRP-linked antibody in 5% BSA+TBST solution for an hour. After incubation, the blot was washed with TBST and developed using Supersignal West Pico PLUS Chemiluminescent Substrate (PIERCE) on a Biorad ChemiDoc Touch imager (Figure S5B). For phosphorylated and total p38 western blots and analyses (Figures S5F-G), cell extracts were prepared using RIPA buffer (supplemented with PMSF and ROCHE protein inhibitor) from EndoC-βH3 cells exposed for 30 minutes to 500 nM thapsigargin or DMSO solvent control for 30 minutes. Protein concentration was quantified using BCA protein assay kit (PIERCE), and 20 μg of each extract was separated, transferred, blocked, and probed with primary antibodies as described above using phosphorylated or total p38 primary antibodies (STAR Methods). Western blot bands were quantified using Fiji (ImageJ).

Flow cytometry analysis of human islets

Human islets from five donors were incubated for 72h in CMRL (10% FBS, 1% Glutamax, 25mM Glucose) supplemented with or without 0.4mM palmitate (stock couplings: 18.4% BSA + 8mM palmitate diluted 1:20 in culture media) with or without 2 μM Selonsertib (Gilead Sciences). A single cell solution was prepared by incubating islets in Accutase (Gibco) for 10 minutes at 37C as previously described²¹² and stained using DAPI (Roche) and PE-Annexin V Apoptosis Detection Kit (BioLegend) according to manufacturer’s instructions. DAPI and Annexin V staining was analyzed and quantified (Data S1) using FloJo Software (BD Sciences) for each islet donor sample sorted on a Fortessa flow cytometer (BD Sciences) using gating strategy shown in Figure S5H. These data can be found in Data S1.

QUANTIFICATION AND STATISTICAL ANALYSIS

Please refer to the figure legends for detailed information on the statistical tests used, what n represents, the exact value of n, and for how the significance was defined. Significant results have been indicated with asterisks, and the non-significant results have been indicated with ns.

Supplementary Material

1

Data S1. Data used to generate figures in this manuscript, related to Figures 1–5 and Figures S1-5

3

Table S1. Donor demographics, related to Figures 1–5

4

Table S2. RNA-seq analyses, related to Figures 1, 3, 4-5

5

Table S3. scRNA-seq analyses, related to Figures 2, 4-5

6

Table S4. ATAC-seq analyses, related to Figures 3–5

7

Table S5. Overlapping genetic variants with stress-responsive CREs, related to Figures 4–5

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
NFIL3	Proteintech	Cat# 11773–1-AP
ATF4	Cell-signalling	Cat# 11815S
Lamin A/C	BioLegend	Cat# 600002
bTubulin	Cell-signalling	Cat# 2128S
phospho-p38 MAPK (Thr180/Tyr182) Antibody	Cell-signalling	Cat# 9211S
p38 MAPK Antibody	Cell-signalling	Cat# 9212S
Anti-rabbit IgG, HRP-linked Antibody	Cell-signalling	Cat# 7074S
Anti-mouse IgG, HRP-linked Antibody	Cell-signalling	Cat# 7076S
Bacterial and virus strains
psPAX2	Addgene	Cat# 12259
pCMV-VSV-G	Addgene	Cat# 8454
Biological samples
Cadaveric human pancreatic islets	Integrated Islet Distribution Program (IIDP), Prodo Laboratories	Table S1
Chemicals, peptides, and recombinant proteins
Prodo Media	Prodo Laboratories	Cat# PIM-S001GMP Cat# PIM-ABS001GMP Cat# PIM-G001GMP
CMRL1066 Media	Gibco	Cat# 11530037
Glutamax	Gibco	Cat# 35050061
Human recombinant Interleukin beta 1 (IL1β)	R&D systems	Cat# 201-LB-005/CF
Human recombinant Interferon gamma (IFNy)	R&D systems	Cat# 285_IF-100/CF
Thapsigargin	Cayman Chemicals	Cat# NC1412406
DMSO	VWR	Cat# 97063–136
StemPro™ Accutase™ Cell Dissociation Reagent	Gibco	Cat# A1110501
Trizol	Life Technologies	Cat# 15596018
Trypsin .05% EDTA	Gibco	Cat# 25300120
FBS, USDA approved origin- 500mL	VWR	Cat# 89510–186
BSA	Sigma-Aldrich	Cat# A7030
Advanced DMEM/F-12	Gibco	Cat# 12634028
2-beta mercaptoethanol	Gibco	Cat# 21985023
nicotinamide	Sigma-Aldrich	Cat# 72340
Sodium selenite	Sigma-Aldrich	Cat# S1382
Puromycin	Takara	Cat# 631306
Penicillin/Streptomycin	Gibco	Cat# 15140163
Fibronectin	Sigma-Aldrich	Cat# F1141
ECM	Sigma-Aldrich	Cat# E1270
Sodium palmitate	Sigma-Aldrich	Cat# P9767
Selonsertib (GS-4997)	Selleckchem/Gilead Sciences	Cat# S8292
Critical commercial assays
10x snATACseq	10x Genomics	Cat# PN-1000175
10x scRNAseq	10x Genomics	Cat# PN-1000121
ATAC kit	Active Motif	Cat# 53150
ERCC Mix 1 & 2	Thermo Fisher	Cat#4456740
LightShift EMSA optimization and control kit	Thermo Scientific	Cat# 20148X
Chemiluminescent Nucleic acid detection kit	PIERCE	Cat# PI89880
6% DNA Retardation gel	Invitrogen	Cat# EC6365BOX
NE-PER Nuclear and Cytoplasmic Extraction Kit	PIERCE	Cat# PI78833
BCA protein assay kit	PIERCE	Cat# PI-23225
PE Annexin V Apoptosis Detection Kit	Biolegend	Cat# 640934
MAP3K5 shRNA	SIGMA	TRCN0000000993
MISSION® pLKO.1-puro Non-Target shRNA Control	SIGMA	SHC016
RNA to CT kit	Applied Biosystems	Cat# 4392938
TaqMan HPRT1	Thermo Fisher	Hs02800695_m1
TaqMan MAP3K5	Thermo Fisher	Hs00178726_m1
LentiX concentrator	Takara	Cat# 631232
LentiX Takara p24 ELISA	Takara	Cat# 632200
Quibit RNA HS Assay	Applied Biosystems	Cat# Q32852
A0251 = GTCAACTCTTTAGCG A0252 = TGATGGCCTATTGGG A0253 = TTCCGCCTCTCTTTG A0254 = AGTAAGTTCAGCGTA A0255 = AAGTATCGTTTCGCA A0256 = GGTTGCCAGATGTCA A0257 = TGTCTTTCCTGCCAG	Biolegend	Cat# 394601 Cat# 394603 Cat# 394605 Cat# 394607 Cat# 394609 Cat# 394611 Cat# 394613
4–20% Mini-PROTEAN® TGX™ Precast Protein Gels	Biorad	Cat#456–1095
Supersignal West Pico Plus Chemiluminescence Substrate	PIERCE	Cat# 34580
Deposited data
RNA-seq, scRNA-seq and ATAC-seq	This paper	GSE251913
Data used to generate figures	This paper	Data S1 and https://doi.org/10.5281/zenodo.13730232
Scripts used for generating figures	This paper	https://github.com/UcarLab/ERS-vs-CYT-Islet-Stress-Responses
Experimental models: Cell lines
EndoCbH3	Univercell Biosolutions	CVCL_IS72
HEK 293T	ATCC	Cat# CRL-3216
Experimental models: Organisms/strains
N/A	N/A	N/A
Oligonucleotides
rs6917676_G_FWD-/5Biosg/TAATGACTGTGTTCTTAAGAT	Integrated DNA Technologies (IDT)	N/A
rs6917676_G_REV-/5Biosg/ATCTTAAGAACACAGTCATTA	Integrated DNA Technologies (IDT)	N/A
rs6917676_T_FWD-/5Biosg/TAATGACTGTTTTCTTAAGAT	Integrated DNA Technologies (IDT)	N/A
rs6917676_T_REV-/5Biosg/ATCTTAAGAAAACAGTCATTA	Integrated DNA Technologies (IDT)	N/A
Recombinant DNA
N/A	N/A	N/A
Software and algorithms
SoupX	Young and Behjati191	https://github.com/constantAmateur/SoupX
STAR	Dobin et al.176	https://github.com/alexdobin/STAR
QoRTs	Hartley and Mullikin177	https://hartleys.github.io/QoRTs/
Ensembl	Howe et al.178	https://useast.ensembl.org/Homo_sapiens/Info/Index
RUVSeq	Risso et al.179	https://bioconductor.org/packages/release/bioc/html/RUVSeq.html
svaseq	Leek180	https://github.com/jtleek/svaseq
edgeR	Robinson et al.181	https://bioconductor.org/packages/release/bioc/html/edgeR.html
DAVID	Sherman et al.182	https://david.ncifcrf.gov/
KEGG	Kanehisa and Goto183	https://www.genome.jp/kegg/pathway.html
Reactome	Vastrik et al.184	https://reactome.org/
WikiPathways	Pico et al.185	https://www.wikipathways.org/
bcl2fastq	Illumina186	https://emea.support.illumina.com/sequencing/sequencing_software/bcl2fastq-conversion-software.html
Cell Ranger Count Pipeline	10X Genomics188	https://www.10xgenomics.com/support/software/cell-ranger/latest/analysis/running-pipelines/cr-gex-count
CITE-seq-Count	Roelli et al.189	https://github.com/Hoohm/CITE-seq-Count
Demuxlet	Kang et al.190	https://github.com/statgen/demuxlet
Seurat	Hao et al.192	https://satijalab.org/seurat/
Scrublet	Wolock et al.193	https://github.com/swolock/scrublet
Harmony	Korsunsky et al.193	https://github.com/immunogenomics/harmony
UCell	Andreatta et al.195	https://bioconductor.org/packages/release/bioc/html/UCell.html
MAST	Finak et al.196	https://www.bioconductor.org/packages/release/bioc/html/MAST.html
Trimmomatic	Bolger et al.197	http://www.usadellab.org/cms/?page=trimmomatic
BWA-MEM	Li and Durbin198	https://github.com/lh3/bwa
SAMtools	Li et al.201	https://github.com/samtools/samtools
MACS2	Zhang et al.202	https://pypi.org/project/MACS2/
IGV	Thorvaldsdóttir et al.203	https://igv.org/doc/desktop/
ENCODE Blacklist	Amemiya et al.204	https://github.com/Boyle-Lab/Blacklist/
DiffBind	Ross-Innes et al.205	https://bioconductor.org/packages/release/bioc/html/DiffBind.html
GenomicRanges	Lawrence et al.206	https://bioconductor.org/packages/release/bioc/html/GenomicRanges.html
GENCODE	Harrow et al.207	https://www.gencodegenes.org/human/release_35.html
HOMER	Heinz et al.123	http://homer.ucsd.edu/homer/
IRanges	Lawrence et al.206	https://bioconductor.org/packages/release/bioc/html/GenomicRanges.html
HINT-ATAC	Li et al.103	https://reg-gen.readthedocs.io/en/latest/hint/introduction.html
1000 Genomes Phase 3 Reference Panel	The 1000 Genomes Project Consortium208	https://ldlink.nih.gov
dbSNP	Sherry et al.209	https://www.ncbi.nlm.nih.gov/snp/
atSNP	Zuo et al.174	http://atsnp.biostat.wisc.edu
pysam	Li et al.201	https://github.com/pysam-developers/pysam
ImageJ	Schneider et al.213	https://imagej.net/
Other
N/A	N/A	N/A

Highlights:

1. Genome-wide comparative human islet ER/cytokine stress response maps

2. T2D variants overlap 86 stress-responsive cis-regulatory elements (CREs)

3. Variant-to-function links between T2D variant rs6917676, MAP3K5, islet apoptosis

4. Risk allele increased ER stress-induced CRE accessibility, MAP3K5-mediated death