Repression of Latent NF-κB Enhancers by PDX1 Regulates β-cell Functional Heterogeneity

SUMMARY

Interactions between lineage-determining and activity-dependent transcription factors determine single-cell identity and function within multicellular tissues through incompletely known mechanisms. By assembling a single-cell atlas of chromatin state within human islets, we identified β-cell subtypes governed by either high or low activity of the lineage-determining factor pancreatic duodenal homeobox-1 (PDX1). β cells with reduced PDX1 activity displayed increased chromatin accessibility at latent NF-κB enhancers. Pdx1 hypomorphic mice exhibited de-repression of NF-κB and impaired glucose tolerance at night. Three-dimensional analyses in tandem with ChIP-sequencing revealed that PDX1 silences NF-κB at circadian and inflammatory enhancers through long-range chromatin contacts involving SIN3A. Conversely, Bmal1 ablation in β cells disrupted genome-wide PDX1 and NF-κB DNA binding. Finally, antagonizing the IL-1β receptor, an NF-κB target, improved insulin secretion in Pdx1 hypomorphic islets. Our studies reveal functional subtypes of single β cells defined by a gradient in PDX1 levels and identify NF-κB as a target for insulinotropic therapy.

Graphical Abstract

eTOC BLURB

Transcription factor dynamics within individual cells drive β-cell identity and function. Weidemann et al. employ single-cell analyses within human and murine models to reveal reciprocal regulation of PDX1 and NF-κB by the clock in distinct β-cell sub-populations and provide logic for anti-inflammatory therapy to reverse β-cell failure caused by PDX1 deficiency.

INTRODUCTION

Collaboration between lineage-determining and signal-dependent transcription factors (LDTFs and SDTFs) sustains energy constancy in response to changes in the nutrient environment associated with the fasting-feeding cycle^1,2. The circadian clock also orchestrates anticipatory cycles of transcription within nearly all metabolic tissues, yet how circadian, nutrient, and lineage-dependent pathways interact within individual cells of multicellular tissues remains unknown^3–7.

The endocrine pancreas is a heterogenous organ that exhibits pronounced oscillation in function across the light-dark cycle and is composed of distinct glucoregulatory cell types. Identity of insulin-producing β-cells is initiated during the primary transition from the pancreatic endodermal bud through combinatorial interactions between pancreatic duodenal homeobox factor 1 (PDX1) and the NK6 homeodomain factor NKX6.1, the basic leucine-zipper MAFA, and hepatocyte nuclear factors (HNFs)^2,8,9. PDX1 also promotes β-cell ontogeny through repression of α-cell gene networks including those controlled by aristaless-related homeobox (ARX)^6,10 and repression of factors that impair aerobic metabolism such as the carboxylate-transporter(s)^11,12. Following weaning, PDX1 induces the expression of genes important in glucose-stimulated insulin secretion that are co-regulated by the core clock transcription factors CLOCK and BMAL1⁷. BMAL1 has also been shown to stabilize β-cell identity by preventing α-cell gene expression during regeneration in response to diptheria-induced cell damage¹³. A remaining question is whether impaired function of LDTFs may contribute to β cell failure in adulthood^14–18.

Here we use epigenomic analyses to identify and dissect the crosstalk between lineage- and signal-dependent transcriptional networks within human and murine islets. We identify distinct subtypes of β cells characterized by differences in the activity of the lineage factor PDX1, in addition to variation in the activity of both inflammatory and circadian gene networks. Our studies reveal that β cells with high PDX1 activity exhibit accessibility at gene networks involved in insulin secretion that are also controlled by the molecular clock. In contrast, β cells with low PDX1 activity exhibit greater chromatin accessibility within enhancers of NF-κB-induced genes. We also show that treatment of islets from Pdx1 hypomorphic animals with an IL-1β receptor blocking peptide enhances insulin secretory capacity. Our studies reveal that loss of PDX1-mediated repression of latent NF-κB enhancers leads to β-cell failure and provide rationale for targeting NF-κB to treat hypoin-sulinemia.

RESULTS

Single-cell chromatin accessibility sequencing identifies PDX1 and latent NF-κB enhancer signatures within distinct β-cell populations of the human islet

Pancreatic islets consist of diverse cell populations that produce physiologically-opposing hormones important in fasting and postprandial glucose metabolism. Recent single-cell sequencing analyses have suggested that within the islet, insulin-producing cells exhibit heterogeneity based upon distinct patterns of accessibility within cis-regulatory elements (CREs)^3,14,19. CREs that identify putative β-cell subtypes also encompass noncoding variants linked to risk of type 2 diabetes^20,21. To elucidate how combinatorial interactions between lineage- and signal-dependent transcription factors might determine β-cell heterogeneity and function, we profiled chromatin accessibility and TF binding motifs in human cadaveric islets obtained from the Integrated Islet Distribution Program (IIDP) using single-nuclei assay of transposase accessible chromatin-sequencing (snATAC-seq) (Figure 1A). We identified 334,116 accessible CREs across 20,519 single islet cells, which were grouped into 19 distinct clusters using the Uniform Manifold Approximation and Projection (UMAP) dimensional reduction approach (see STAR Methods). These 19 clusters included canonical endocrine cell types, with four α cell clusters (α1–4), three β cell clusters (β1–3), and one δ cell cluster (Figure 1B), which were characterized by chromatin accessibility nearby genes encoding glucagon (GCG), insulin/insulin-like growth factor 2 (INS/IGF2), and somatostatin (SST), respectively (Figure 1C). Other islet-associated cell subpopulations identified by accessibility at canonical cell markers included stellate (PDGFRB), neural crest (SOX10), lymphatic and myeloid immune (CCL4), endothelial (PECAM1), ductal (KRT19), and acinar (REG1A) cells (Figures 1B–C).

Figure 1. Single-cell chromatin accessibility sequencing identifies PDX1 and NF-κB enhancer signatures within distinct β-cell populations of the human islet.

(A) Profiling single nuclei and rhythmic bulk chromatin accessibility by ATAC-sequencing in human cadaveric islets. (B) Clustering and dimensional reduction analyses (UMAP) identified 19 distinct subpopulations of cells based on normalized reads at accessible cis-regulatory elements (20,519 single islet nuclei from 3 humans). (C) Aggregate sequencing fragments, segregated by cell-type, at lineage-specific marker genes: GCG (alpha), INS-IGF2 (beta), SST (delta), PDGFRB (stellate), SOX10 (neural crest), CCL4 (lymphatic and myeloid immune), PECAM1 (endothelial), KRT19 (ductal), and REG1A (acinar). Co-accessibility linkage analysis connects individual snATAC-seq peak accessibility scores within networks of coregulated regions. (D) Accessibility at PDX1 (left) and NF-κB (right) motifs was enriched in the β1 and β2 subpopulations of cells, respectively. Individual cell chromVAR z-scores (top) and subpopulation z-score distribution (bottom) across β1 and β2 subpopulations demonstrate significant variation in chromatin accessibility at PDX1 and NF-κB motifs. (E) Differential analysis of TF motif accessibility between β1 and β2 subpopulations revealed enrichment for basic helix-loop-helix (bHLH) and lineage-determining TFs in β1 cells and enrichment of immediate early bZIP and NF-κB TFs in β2 cells. (F) ATAC-seq performed in synchronized whole islets every 4 hrs for 24 hrs following a forskolin shock (n=10 human islet equivalents (IEQs) per sample per time point from 4 humans). Data was analyzed using eJTK_Cycle at a 2-hr resolution, revealing 24-hr patterns in chromatin opening at ATAC-seq peaks (left) and β-cell TF motifs associated with cell identity and rhythmic function (right). See also Figure S1 and Tables S1 and S3.

A key advantage of snATAC-seq over single-cell mRNA transcriptomic profiling lies in the ability to detect the regulatory activity of TFs at both gene promoters and distal enhancers^19,22–24. To examine whether differential gene regulation underlies β-cell heterogeneity, we first compared gene activity scores (the sum of accessible reads at fragments overlapping gene coordinates and the 2 kb region upstream) between the β1 and β2 β-cell subclusters, which comprised 47.2 and 45.1% of the total β-cell population, respectively. We identified 600 genes with significant variation (P < 0.05), including β-cell-defining genes such as INS (6.45e-87), MAFA (2.25e-50), and GLP1R (2.37e-36), and inflammatory genes such as PTGER3 (1.471551e-05), IL2 (5.968259e-15), and LPAR1 (8.646495e-15) (Figure S1A) (Table S1). Analysis of single-cell RNA-sequencing (scRNA-seq) from 12 non-diabetic human islets (GSE114297), using snATAC-seq gene activity scores as transfer anchors for clustering within Seurat, revealed significant correlation between log₂-fold changes in RNA expression and chromatin gene activity scores between the β1 and β2 subpopulations (P = 1.29e-30, Pearson’s r = 0.35) (Figure S1B). A similar correspondence between changes in gene expression and chromatin accessibility was observed when comparing α cell sub-populations (Table S2) and when comparing whole α and β cell populations directly (P = 1.0e-16, Pearson’s r = 0.75) (Figure S1B). These data reveal correspondence between single-cell transcript abundance and chromatin accessibility across cell types²⁵.

To elucidate the epigenetic mechanisms underlying β-cell heterogeneity, we compared differential CREs between the INS-enriched (β1) and INS-low (β2) β-cell clusters (Figures S1A, S1C) and found that the majority of differentially-accessible sites (73.29% of 2,632 peaks) were located within enhancer regions (either greater than 3 kb upstream or 1 kb downstream of transcription start sites) (P_adj < 0.05) (Figure S1D). We next took advantage of our robust sequencing depth per cell (37,400 high-quality fragments per cell) to identify differential TF activity in the β1 versus β2 subpopulations using chromVAR to measure TF-associated chromatin accessibility^{26, 27}. Comparison of motif Z-scores between the β1 and β2 cell clusters identified differential accessibility at several β-cell lineage-determining transcription factor (LDTF) motifs that corresponded with increased accessibility nearby INS (Figures 1D, S1A). Among the established LDTFs, PDX1, a master regulator of pancreatic development²⁸, was one of the most significant differentially-enriched motifs (Padj=3.98e-88) that was also one of the most highly expressed TFs within adult β cells (Figures 1D–E, S1E). We further found that 58% of the differentially-accessible sites between β1 and β2 cells contained motifs for PDX1 compared to 40% in the background peak set (hypergeometric test, P < 0.001) (Figure S1D). We therefore refer to these β1 and β2 subpopulations as PDX1^high and PDX1^low cells, respectively. Other TFs that were enriched in the PDX1^high cells were ones known to be required for mature β-cell development, including neurogenic differentiation 1 (NEUROD1), NKX6.1, and the basic helix loop helix (bHLH) pancreas transcription factor 1A (PTF1A), as well as other basic helix-loop-helix (bHLH) TFs whose DNA-binding motif contains canonical E-box motifs that are targets of the molecular clock^29,30 (P_adj < 0.001) (Figure 1E) (Table S3). In contrast, the subcluster of β2 cells with low accessibility at the INS locus was highly enriched for accessible nuclear factor-kappa B1 (NF-κB1), the kruppel-like factor KLF14, JUNB, and the interferon regulatory factor-6 (IRF6) TF motifs (P_adj < 0.001) (Figure 1E) (Table S3), all of which function in signal-induced expression of inflammatory genes during stress, the immune response, and in cell proliferation and survival^31,32. These data support recent snATAC-seq studies that identified divergent hormone-high and hormone-low endocrine islet subtypes with unique immediate-early TF signatures^3,26 (GSE160472, GSE160473, GSE163610). We note a key advantage of our studies was the use of live tissue with the 10X Genomics platform³³, which enabled ~10-fold deeper sequencing read coverage per cell than prior studies and detection of rarer chromatin opening events including at NF-κB sites. In sum, our single-cell sequencing of chromatin activity identified bimodal populations of insulin-producing cells within the human islet – one INS-enriched cluster with high PDX1 activity and one INS-low cluster with low PDX1 activity but with significant enrichment in NF-κB TF activity. We also examined PDX1 RNA abundance in human islets using RNAscope, which revealed a bimodal distribution of PDX1 RNA levels in contrast to the normal distribution observed in various housekeeping genes (Figure S1F). These observations support the existence of two distinct populations of β cells with respect to PDX1 activity.

Collaborative interactions amongst TFs define individual subpopulations of the mature β cell, and our data suggests that known regulators of β-cell maturation and function are uniquely activated in PDX1^high β cells^34–36. The circadian clock activators CLOCK/BMAL1 have been shown to contribute to metabolic maturation and identity of insulin-producing cells, through the underlying mechanisms remain poorly understood^37,38. The findings above that PDX1 and clock regulatory elements co-exist in INS-enriched clusters support the possibility that PDX1 and rhythmic gene expression may play a role in β-cell function. To examine whether transcriptional regulators of β-cell identity are regulated in a circadian manner, we assessed 24-hr rhythmic opening of chromatin by performing ATAC-seq in human islets whose molecular clocks were first synchronized using a 1-hr forskolin pulse. Genomic analyses were performed every 4 hrs for 24 hrs starting >24 hrs after administration of forskolin, well beyond the ~1-hr half-life of forskolin action on chromatin³⁹ (Figure 1A). We generated within-donor z-scores across the 24-hr times series using normalized ATAC-seq counts (see STAR Methods) and identified 3,125 statistically rhythmic ATAC-seq peaks genome-wide using the empirical Jonckheere-Terpstra-Kendall algorithm (eJTK_CYCLE) (Bonferroni P < 0.05) (Figure 1F)⁴⁰. Rhythmic sites were identified near enhancers of genes implicated in islet cell identity (SIX3, PDX1, GCG), circadian expression (PER2, NPAS2), and inflammation (ETS2, TNFRSF21)^41–43. TF motif enrichment analysis at rhythmic ATAC-seq peaks identified significant enrichment for TFs (e.g., PDX1, NEUROD1, and PTF1A) (P_adj < 0.05) (Figure 1F) that were also enriched in PDX1^high cells (Figure 1E). Together, these studies indicate that rhythmic co-activation of chromatin by PDX1 and other TFs across the day may contribute to β-cell function.

PDX1 represses NF-κB enhancers to control β-cell function

Cross-talk between lineage and signal-dependent TFs within non-coding regulatory regions of the genome drive identity and function of the endocrine pancreas^24,44–46. Emerging studies suggest that LDTFs define distinct subpopulations of β cells important in both insulin secretion and type 2 diabetes^3,15. Based upon our finding of a PDX1 activity signature in distinct cell populations exhibiting divergent patterns of NF-κB and clock activity, we sought to dissect the function of PDX1 in control of inflammatory and circadian gene networks using snATAC-seq analyses of single-nuclei isolated from islets of mice with genetic deficiency of PDX1 (Figure 2A). Since ablation of Pdx1 causes embryonic lethality, we utilized mice heterozygous for a null allele of Pdx1 (Pdx1^+/−) in the presence or absence of second allele harboring a mutation within a key regulatory enhancer (area IV) of Pdx1 (mutant Pdx1^ΔIV/− vs control Pdx1^+/− mice, respectively) (Figure 2A). Pdx1^ΔIV/− mice develop post-weaning hypoinsulinemia, impaired glucose tolerance, and decreased β-cell proliferation^12,47. Using merged snATAC-seq data derived from nuclei harvested post-weaning (10–18 weeks of age) from Pdx1^ΔAIV/− and control Pdx1^+/− mice, we identified 187,343 accessible CREs from 9,480 single islet cells isolated and classified these into 15 unique islet cell populations, which included three β-cell populations (Figures 2B, S2A–C). These three murine β-cell populations varied according to PDX1 activity similar to our observations in human cadaveric islets (Figure 1B–E). Comparison of chromatin accessibility in the ensemble of all three β-cell populations of both Pdx1^ΔAIV/− mutant and Pdx1^+/− genotypes revealed significant alterations in accessibility at 1,902 chromatin peaks across 1,373 unique genes (P_adj < 0.05). The Pdx1 area IV enhancer region was the most significantly downregulated non-coding cis-regulatory element (Figure S2D), consistent with hemizygosity at this enhancer, followed by the Ins1 promoter (Figure 2C) (P < 0.001). Motif analyses from the pooled β-cell populations revealed reduced chromatin accessibility in Pdx1^ΔAIV/− mutant mice at binding sites for key TFs that promote β-cell function, including NEUROD1, NEUROG2, PTF1A, and bHLH factors (Figure 2D), providing genetic evidence that supports our findings that in human islets, PDX1^high subpopulations correlated with TFs associated with β-cell and circadian function (Figures 1E–F). In contrast, sites that displayed increased chromatin accessibility in Pdx1-deficient mutants contained binding sites for TFs involved in proinflammatory signaling, including NF-κB1, NF-κB2, and RELA (Figure 2D). The finding of an altered chromatin landscape in islets from Pdx1 hypomorphic animals with increased pro-inflammatory and reduced insulin secretory networks was analogous to our observation of unique enhancer signatures in human islets that were enriched or depleted in PDX1 and NF-κB, respectively (Figures 1E–F). Gene ontology pathway analysis further indicated a role for PDX1 in the induction of genes underlying glucose responsive insulin secretion in the β cell while at the same time repressing pro-inflammatory regulatory elements (Figure 2E). Peaks near closed chromatin in Pdx1^ΔAIV/− mutant β cells (i.e., increased accessibility in controls) were linked to pathways involved in hormone exocytosis, including MODY (e.g. Iapp, Ins1, Pdx1, Mafa) and type 2 diabetes mellitus (e.g. Irs1, Mapk10, Mtor, Pparg), as well as negative regulation of the acute inflammatory response (Figure 2E). In contrast, gene sets near chromatin that was more accessible in the mutant β cells were enriched within morphogenesis (e.g. Egfr, Igf1r, Mapt), negative regulation of synaptic transmission (e.g. Asic1, Mgll, Pla2g6, Gabrb3), and gene networks de-repressed following inhibition of the circadian clock (e.g., Bhlhe40, Ptger3, Cipc) (Figure 2E)^{48, 49}. However, we did not detect significant alterations in accessibility at cell-defining genes Gcg or Sst comparing α, β, or δ cells between Pdx1 mutant and control endocrine cell populations (Figure S2E). Finally, RNA-sequencing revealed significantly reduced expression of genes involved in glucose-stimulated insulin secretion, including Vamp2, Glp1r, and Snap25, as well as those involved in circadian rhythms, including the Per and Cry genes, in islets isolated from Pdx1^ΔAIV/− mutant compared to control mice (Figure 2F), consistent with impaired β-cell function and disrupted circadian rhythms in Pdx1^ΔAIV/− mutants. Furthermore, we observed a significant enrichment and upregulation of genes involved in pro-inflammatory NF-κB-signaling and apoptosis (Figure 2F), supporting the de-repression of NF-κB signaling in the absence of PDX1. Therefore, studies in both Pdx1 hypomorphic mice and in human islets indicate that PDX1 establishes a balance between the expression of circadian regulatory programs and NF-κB-mediated inflammatory pathways within the pancreatic β cells.

Figure 2. Genetic deficiency of PDX1 leads to de-repression of NF-κB.

(A) Profiling single-cell chromatin accessibility using single-nuclei ATAC-sequencing in islets isolated from control (Pdx1^+/−) and Pdx1 mutant (Pdx1^ΔAIV/−) mice (n=8 per genotype). (B) Clustering and dimensional reduction analyses (UMAP) of control and mutant islets cells identified 15 distinct subpopulations of cells based on normalized reads at accessible cis-regulatory elements. (C) Cumulative (top) and single-cell (bottom) accessible reads at the Ins1 gene in pooled control versus Pdx1 mutant β cells. (D) TF binding motifs differentially-enriched within accessible chromatin regions of pooled control versus Pdx1 mutant β cells. Negative values indicate the motif is more enriched in controls, while positive values indicate motif is more enriched in the Pdx1 mutants. (E) Gene ontology analysis of annotated differentially-accessible peaks identified as upregulated in control or Pdx1 mutant β cells. (F) RNA-sequencing reveals dysregulation of circadian, insulin secretion, and NF-κB signaling in Pdx1A4 mutant islets. RNA-sequencing and pathway analyses using pathfindR reveal enrichment for circadian rhythm, insulin secretion, and NF-κB signaling KEGG pathways, as displayed by shared gene enrichment (top) and hierarchal clustering with fold-enrichment and multiple comparison P-values (bottom), among differentially-expressed transcripts (Adjusted P-value < 0.05) in islets isolated from Pdx1A4 (n=5) versus Pdx1 (n=4) heterozygous mice. See also Figure S2.

Circadian analyses reveal rhythmic regulation of NF-κB enhancer activity by PDX1

Circadian rhythms are first established within the β cell postnatally⁵⁰ and become more pronounced following weaning, where both a mixed macronutrient diet and induction of rhythmic feeding cycles lead to functional maturation of the β cell^51–53. In the adult β cell, robust circadian clock activity promotes rhythmic insulin release through activation of pre-established enhancers that colocalize with PDX1 binding sites^2,7,51,54. The observation that islets from Pdx1-deficient animals display compact chromatin within enhancers bound by circadian TFs (Figure 2E) prompted us to examine the effect of Pdx1-deficiency on the temporal regulation of β-cell enhancers. To examine chromatin accessibility in Pdx1^ΔAIV/− mutant and Pdx1^+/− control islets at two different phases of the intrinsic circadian cycle, we exposed islets (n=3 biologically-independent pools of islets per genotype) to a 1-hr forskolin pulse and collected the islets at the nadir (36 hr time post-synchronization, TPS) and zenith (48 hr TPS) of rhythmic insulin secretion (Figure 3A)⁷. We performed ATAC-seq adapted for low-input samples following the Omni-ATAC-seq protocol⁵⁵ and identified 132,195 unique accessible sites across all pooled samples (Figure 3A). Likelihood ratio tests (LRTs) identified 1,207 significantly (P_adj < 0.05) altered CREs (Figures 3B–C) with respect to either circadian time or the Pdx1 genotype using normalized ATAC-seq count data in DESeq2^56,57. Unbiased principal component analysis using ATAC-seq normalized counts at these sites revealed distinct patterns of chromatin opening with respect to genotype and circadian time (Figure S3A).

Figure 3. Circadian analyses reveal rhythmic regulation of NF-κB enhancer activity by PDX1.

(A) Pancreatic islet preparations from control (Pdx1^+/−) and Pdx1 mutant (Pdx1^ΔAIV/−) mice were subjected to a circadian synchronization pulse (forskolin), followed by ATAC-seq at two different time points, 36 and 48 hrs post-synchronization (the trough and peak of insulin secretion, respectively) (n=3 per timepoint per genotype). (B) Log2 fold-change in normalized accessible reads with respect to circadian time or genotype following identification of dynamic peaks using likelihood ratio tests. Significant peaks were segregated into different groups of peaks with distinct accessibility patterns using k-means clustering: Group 1 (orange) - sites with increased accessibility in control islets at the 48 hr time point (peak of insulin secretion) but with reduced accessibility in the mutants. Group 2 (green) - sites with increased accessibility at the 36 hr time point (trough of insulin secretion), which are also increased in the mutants. Group 3 sites (yellow) - sites only dependent on circadian time and not genotype. (C) Normalized accessible read Z-scores (left, middle) across all samples reveal increased accessibility nearby insulin secretion genes in control islets (Group 1) and near stress and inflammatory genes in mutant islets (Group 2). Motif analyses (right) within differentially-accessible peaks demonstrate enrichment for β-cell TFs and inflammatory and immediate early TFs in Group 1 and Group 2 chromatin peaks, respectively. (D) Increased normalized chromatin reads in Pdx1^ΔAIV/− mutant islets neighboring the TNF-related gene Tnfrsf11b at 36 hrs post-synchronization. (E) Glucose clearance, insulin secretion, C-peptide levels, and area under curves following oral glucose administration (2g/kg) in control (n=14 mice per time point for glucose, n=13 for insulin and C-peptide) and Pdx1 mutant (n=12 mice per time point for glucose, n=9 for insulin and C-peptide) in either the morning (ZT2) or evening (ZT14) analyzed by mixed-effects models. Main and interaction term significance determined by mixed-effects model. *P < 0.05, **P < 0.01, ***P < 0.001, n.s. not significant by Wald (ATAC-seq) or Holm-Šídák test. Data are represented as mean ± SEM. See also Figure S3.

To identify chromatin accessibility regulatory patterns among these data, we performed k-means clustering and identified three distinct clusters of genes that were dependent upon either circadian time and/or genotype. Group 1 sites had reduced accessibility at both time points in the Pdx1 mutant islets but exhibited more robust opening at CT48 in the control islets, while group 2 sites gained accessibility specifically at CT36 and were more accessible in the Pdx1 mutant islets (Figures 3B–C). Group 3 sites were regulated only by circadian time and not genotype (Figure S3C–D). Consistent with a collaborative role for the clock and PDX1 in promoting insulin secretion, Group 1 sites were enriched for PDX1, MAFA, and NKX6.1 binding motifs, including canonical PDX1 CREs that promote Ins1 and Mafa expression (P_adj < 0.001) (Figures 3B–C)⁷. In contrast, distinct ATAC-seq peaks within the Group 2 peak set enriched in the Pdx1-deficient cells occurred at sites near the inflammation mediators Tnfrsf11b, Nfkbiz, Il-1b, and the E26 transformation-specific (ETS) factors Etv1 and Ets2 (Figures 3B–D) and were enriched in motifs for NF-κB, RELA, and the immediate early complex FOS::JUNB (Figure 3C). These findings suggest that the loss of Pdx1 results in both the de-repression and rhythmic reprogramming of inflammatory and stress response genes. Genes encoding the α-cell-enriched adrenergic receptors ADRB1 and ADRA1b in Pdx1 mutant islets also exhibited increased circadian variation in accessibility, suggesting PDX1 normally represses islet α-cell transcription through mechanisms that may involve circadian factors (Figures 3B–F, S3B). Of note, lymphatic, macrophage, and exocrine cells in the Pdx1^ΔAIV/− mutants did not display differential PDX1 activity, indicating that increased chromatin accessibility of Group 2 genes could not be attributed to signal from non-β-cell types (Figure S2C). Finally, in contrast to Groups 1 and 2, which were dependent upon both genotype and circadian time, chromatin accessibility of Group 3 was only dependent upon circadian time (Figures 3B, S3C–D). These sites were enriched for DBP and NFIL3/E4BP4 TF motifs (Figure S3D), consistent with previous studies demonstrating that DBP regulates chromatin dynamics in the islet⁵⁸. In sum, levels of PDX1 determine the chromatin landscape within subpopulations of the β cell characterized by unique patterns in the accessibility of enhancer regions controlling circadian and inflammatory gene programs.

To determine whether the divergent chromatin signatures identified in our synchronized ATAC-seq datasets correspond with diurnal insulin secretory dynamics in the whole animals, we performed oral glucose tolerance tests (oGTTs) in Pdx1^ΔAIV/− mutant and Pdx1^+/− heterozygous control mice during both the day (ZT2) and night (ZT14). Pdx1^ΔAIV/− mutant mice displayed significantly impaired glucose tolerance compared to control mice (P < 0.0001) that was more pronounced during the nighttime (Figure 3E). Analysis of the whole glucose excursion curve by mixed effects modeling also revealed a significant interaction between the Pdx1 genotype, time of day (i.e., zeitgeber time), and glucose levels (P = 0.015). These mice also had reduced insulin and C-peptide levels at both time points compared to controls (P < 0.05 by two-way ANOVA) (Figure 3E), and insulin and C-peptide levels were both reduced to a greater extent in the evening in the mutants relative to control animals⁷. Together, these data reveal a requirement for PDX1 in regulation of chromatin compaction, β-cell maturation, and glucose homeostasis across the day through alterations in the control of secretory and pro-inflammatory gene networks.

To examine whether the molecular clock contributes to rhythmic insulin secretion through the control of PDX1 chromatin binding, we performed PDX1 ChIP-seq in Bmal1 knockout mouse β-cell lines (Beta-TC-6). We found that loss of BMAL1 led to significantly reduced PDX1 DNA binding genome wide within β cells (P < 0.001) (Figure S3F), including at regions controlling the expression of genes related to regulation of insulin secretion, including Ins1, Neurod1, and Six2 (Figure S3F). Therefore, loss of the molecular clock leads to reciprocal impairment in PDX1-mediated regulation of genes important in β-cell function.

PDX1 forms long-range chromatin loops to regulate NF-κB and circadian clock enhancer activity

Given our observations that PDX1 and NF-κB regulatory elements distinguish distinct subpopulations of the adult β cell (Figures 1–2), we sought to determine whether PDX1 directly represses NF-κB genome-wide. We first examined how PDX1 depletion affects the activity of p65, a canonical NF-κB subunit, using p65 ChIP-seq in Beta-TC-6 treated for 48 hrs with siRNA targeting PDX1 compared to a control siRNA (Figure 4A). We observed that loss of PDX1 increased genome-wide occupancy of p65 at 5,268 unique sites, despite similar levels of p65 protein abundance (Figures 4A, S3E). This gain in binding included NF-κB/inflammation targets, such as Tnfrs11a, and the gene encoding the circadian repressor PER2 (Figure 4A), consistent with previous studies⁵⁹. Luminometry analyses revealed that Pdx1 hypomorphic islets exhibit a shortened period length (Figure S4A). Thus, loss of PDX1 in β cells increases NF-κB binding genome-wide and alters the expression of core clock repressors, suggesting that PDX1 repression of NF-κB in turn alters clock function.

Figure 4. PDX1 represses NF-κB through long-range chromatin contacts that co-localize with IL-1β-response elements controlling insulin secretion.

(A) p65 ChIP-seq in Beta-TC-6 cells treated with siRNA targeting Pdx1 (siPdx1) reveals increased genome-wide occupancy at inflammatory and circadian (Per2/Hes6) gene regions. (B) Chromatin conformational sequencing coupled with ChIP-seq (HiChIP-seq) targeting PDX1 identifies three-dimensional chromatin contacts involving PDX1 binding sites. (C) Significant chromatin interactions identified between the PDX1-SIN3A binding sites which loop to contact p65 binding sites. siPdx1-treated β cells exhibit increased p65 binding following loss of PDX1 protein, suggesting PDX1-SIN3A chromatin loops normally repress p65 binding near Per2. (D) IL-1β inhibits β-cell gene expression and activates p65-mediated expression of inflammatory genes in Beta-TC-6 cells. (E) Acute (2 hr) administration of IL-1β to β cells inhibits glucose-responsive insulin release specifically in control siRNA-treated cells, whereas siPdx1-treated cells do not respond to IL-1β (n=10 samples per siRNA treatment). Data are represented as mean ± SEM. (F) Antagonism of islet IL-1β signaling using the IL-1β receptor antagonist (IL-1RA) restores glucose-stimulated insulin secretion in Pdx1^ΔAIV/− islets (n=8 Pdx1^+/− mice, n=5 Pdx1^ΔAIV/− mice). Two-way ANOVA, *P < 0.05 by Holm-Šídák test. Data are represented as mean ± SEM. See also Figure S4 and Tables S4–S6.

Given our findings that ablation of Bmal1 impaired PDX1 DNA binding (Figure S3F), we next sought to determine whether loss of the clock in turn alters the genome-wide binding of p65 by performing p65 ChIP-seq in Bmal1 KO Beta-TC-6 cells. We observed significantly fewer p65 binding events in the Bmal1 KO β cells (177 peaks) compared to controls (2430 peaks) (Figure S3G), suggesting BMAL1 is also required for the ability of p65 to bind chromatin. Sites depleted of p65 binding in the Bmal1 KO cells included cell survival and anti-apoptotic genes, including the apoptosis antagonizing TF Aatf, janus kinase 1 Jak1, inositol polyphosphate-4-phosphatase Inpp4b, and the phosphodiesterase Pde10a (Figure S3G). This data suggests a three-way interaction in which activity of both PDX1 and p65 depends upon a functional circadian clock.

Three-dimensional chromatin analyses have revealed that LDTFs form contacts between widely-separated genomic regions to control β-cell function¹⁵, yet whether PDX1 controls SDTFs through long-range contacts has not been determined (Figure 4B). Within Pdx1-knockdown β cells, p65 binding sites were enriched at positions far away (>500 bp) from sites of DNA binding by PDX1 observed in control cell types (Figure S4B). Therefore, we utilized three-dimensional chromosome conformation capture coupled with PDX1 ChIP-seq (HiChIP-seq) to map long-range interactions between PDX1 and distant regulatory elements (Figure 4B). We identified 4,155 long-range (> 5 kb) chromatin interactions between PDX1 and distant cis-regulatory regions (see STAR Methods, FDR < 0.05) (Figures 4C, S4B–C, Table S4). These included 497 cytokine-inducible NF-κB binding sites that were identified as NF-κB targets by p65 ChIP-seq in IL-1β-treated β cells (Table S5). Of note, the proportion of p65 peaks induced by IL-1β within loops was threefold higher than the background occurrence of NF-κB motifs found in all accessible cis-regulatory elements identified in our mouse islet ATAC-seq peakset (12% versus 4%). A prominent PDX1 chromatin loop was observed between a PDX1 binding site 2.1 kb upstream of TNF receptor-associated factor 3 interacting protein 1 (Traf3ip1) and an NF-κB binding site within the 5’ UTR of the Per2 gene (Figure 4C). We also observed co-localization between PDX1 and NF-κB sites at genes involved in Ca²⁺ signaling (Inpp4b) and inflammation (Irf2) (Figure S4C), as well as at other IL-1β-induced NF-κB sites such as the IL-1β receptor (Il1r1), tumor necrosis factor receptor-associated factor interacting protein (Traf3ip2), and the type 2 diabetes associated allele Slc30a8 (Table S5). To test whether PDX1 inhibits NF-κB at these distal enhancers through recruitment of co-repressor complexes, we performed chromatin-immunoprecipitation with the transcriptional corepressor SIN3A, which establishes β-cell identity together with PDX1⁶⁰. Of the 4,925 sites bound by SIN3A, 41% co-localized with PDX1-regulated genes, including Traf3ip1 and TNF-related genes such as Tnfsf4 and Traf5 (Table S6, Figures 4C, S4C–D)⁶¹. Moreover, we found that loss of PDX1 disrupts SIN3A chromatin binding genome-wide (Figure S4E), including reduced SIN3A binding at IL-1β inducible NF-κB binding sites near the gene encoding Nfkbie (Figure S4F). Of note, this region also undergoes long-range chromatin contacts with PDX1 binding sites (Figure S4F), consistent with PDX1 repression of NF-κB across long distances via three-dimensional interactions and co-recruitment of SIN3A.

Inhibition of NF-κB signaling using an IL-1β-receptor antagonist enhances β-cell function in Pdx1-deficient β cells

Human PDX1^low and murine Pdx1^ΔAIV/− β cells both exhibit signatures of increased chromatin openness at pro-inflammatory networks and NF-κB targets such as interleukin-1β (IL-1β) (Figures 1–2). The IL-1 receptor (IL-1R) is one of the most highly expressed cytokine receptors within pancreatic islet β cells⁶². Exposure to IL-1β remodels the β-cell chromatin landscape⁶³, and antagonism of IL-1β represents a well-established anti-inflammatory strategy in humans^64,65. Inflammation also appears to play a role in β-cell dysfunction in type 2 diabetes through mechanisms that remain uncertain⁶⁶. We found that IL-1β stimulation of β cells from wild-type mice induced the expression of canonical NF-κB targets (including Nfkbiz and Nfkbia) and repressed the expression of PDX1 targets (including Mafa, Glp1r, and Pdx1 itself) (Figures 4D, S4G)⁶⁷. IL-1β also inhibited glucose-stimulated insulin release in wild-type Beta-TC-6 cells (Figure 4E). Conversely, treatment of islets isolated from Pdx1^ΔAIV/− mutant mice with an IL-1β receptor antagonist (IL-1RA) robustly enhanced insulin secretion (Figure 4G), consistent with therapeutic amelioration of PDX1 deficiency through blockade of IL-1β-induced NF-κB signaling. We used hyperglycemic clamping to directly investigate whether blocking IL-1β might overcome hypoinsulinemia in Pdx1 hypomorphic animals in vivo⁶⁸. We evaluated differences in insulin-mediated glucose-infusion rates (GIRs) following stabilization of GIR (>50 min post-initiation of infusion) and assessed genotype-drug interactions. Our analysis revealed genotype-dependent differences in IL-1β receptor antagonist-treated animals versus vehicle (PBS) controls, with a trend towards an increased GIR in the IL-1β receptor antagonist-treated vs PBS-treated mutants. Hyperglycemic clamp results therefore suggest that antagonism of IL-1β has genotype-specific effects on glucose clearance in Pdx1 hypomorphic animals (Figure S4H). Finally, anti-inflammatory treatment using the IL-1β receptor antagonist enhanced insulin secretion in Bmal1 KO cells (Figure S4I). Altogether, our epigenomic and genetic studies indicate that blocking inflammation upstream of NF-κB may provide a novel insulinotropic therapy in states of β-cell failure related to PDX1 and/or clock dysfunction.

DISCUSSION

We have identified a unique β-cell subtype of adult islet cells with inflammatory transcriptional activity characterized by genome-wide enrichment of NF-κB networks concurrent with reduced activity of the master lineage regulator PDX1. Identification of this new islet subtype at single-cell resolution was possible because of the high mapping density of chromatin accessibility and resolution of specific regulatory elements across human islet subpopulations. Using three-dimensional chromosome conformation capture, we further showed that PDX1 represses inflammatory signaling through long-range inhibition of NF-κB and recruitment of the canonical co-repressor SIN3A. Conversely, we showed that de-repression of NF-κB in single islet cells from genetic models of Pdx1 deficiency leads to β-cell failure and diabetes. These results indicate that PDX1 protects the β cell from stress and inflammation and provide evidence for the cross-regulation of PDX1 and pro-inflammatory/NF-κB signaling as a key element in β-cell function.

Our studies also revealed that PDX1 induces circadian gene networks important in metabolic homeostasis in mature β cells. We observed that PDX1 localizes to CLOCK/BMAL1 regulatory elements through a mechanism that varied across the circadian time-scale in both mice and humans. The observation that PDX1 binding sites were enriched within CLOCK/BMAL1 enhancers is consistent with prior studies that clock factors are involved in islet maturation and response to injury^13,24,69 and the repression of latent NF-κB-mediated inflammatory networks^59,70,71. Our analyses further suggest that PDX1 promotes time-of-day-dependent activity of insulin exocytic gene networks while simultaneously repressing inflammatory signatures⁷⁰. Varied conditions, such as overnutrition or aging itself, have been shown to alter circadian programming in a tissue-specific manner^{72, 73}. Similarly, our genomic studies in Bmal1 KO β cells revealed mixed effects of the clock factor BMAL1 on NF-κB-mediated gene expression. Specifically, ablation of Bmal1 altered p65 binding to DNA at cell survival and anti-apoptotic genes (Figure S3G), whereas ablation of BMAL1 did not alter the binding of p65 to cytokine-induced inflammatory genes (Figure S3G). Previous work has demonstrated that p65/NF-κΒ is required for β-cell function, so it is possible that loss of p65 binding in BMAL1-ablated β cells contributes to the hypoinsulinemia in these animals, perhaps due to anti-apoptotic actions of the NF-κB pathway^74,75. Future studies will be required to understand β-cell-specific responses to inflammatory stimuli throughout the circadian cycle and how core circadian activators or downstream clock-controlled genes interact with PDX1.

Human and animal studies have shown that disrupted clock function resulting from obesogenic diet and/or mistiming of feeding contributes to metabolic disease⁷⁶. The data presented here suggests that IL-1β induction, which has been shown to increase with diet-induced obesity^59,77,78, may in turn trigger β-cell failure. Genetic PDX1 deficiency led to de-repression of latent NF-κB enhancers and provided rationale to inhibit IL-1β as an insulinotropic therapy. While prior studies have shown that IL-1β control of glucose homeostasis involves the central nervous system^66,79,80, our results in β-cell lines indicate that blocking IL-1β in enhances insulin secretion directly within the islet. Since human islets used in our studies were harvested from glucose-responsive donors, future studies will be necessary to determine whether the antagonist relationship between PDX1 and NF-κB is disrupted in type 1 or type 2 diabetes.

Limitations of Study

Our experiments in human islets identified a link between PDX1, NF-κΒ, and the clock, though additional snATAC-seq and snRNA-seq studies will be required to determine whether the daily rhythms in chromatin accessibility in whole islets arise due to variation within individual β cells versus non β cells and whether the chromatin changes correspond with gene expression changes at the level of single cells. Further, while we found evidence of β-cell heterogeneity in both male and female subjects in our human islet studies, further studies will be required to specifically examine the effects of both sex and age in the Pdx1 mutant animals. Our glucose tolerance experiments likewise provided evidence of a significant interaction between circadian time, Pdx1 genotype, and glucose excursion, with a maximal difference in insulin release during the dark period. Given the controls used for these experiments were Pdx1 heterozygotes, additional studies will be needed to compare how the Pdx1 heterozygotes and Area IV mutants compare to wild-type controls^81,7. Finally, future studies in preclinical context should seek to address whether loss of the clock or of lineage-determining factors predisposes animals to an inappropriate inflammatory response through a potential immunoendocrine mechanism.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact:

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Joseph Bass (j-bass@northwestern.edu).

Materials Availability:

This study did not generate new unique reagents.

Data and Code Availability:

• Data in this study is publicly available in the GEO repository (GSE 214678). All unprocessed data can be found in Data S1 – Source Data.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper can be found in the Key Resources Table and is available from the lead contact upon request.

Key Resources Table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
anti-p65	Cell Signaling	Cat#8242
anti-PDX1	Cell Signaling	Cat#5679
anti-BMAL1	J. Bass, Millipore	Cat#ABE2599
anti-SIN3A	Cell Signaling	Cat#8056S
anti-Histone H3	Cell Signaling	Cat#9715
anti-GAPDH	Cell Signaling	Cat#5174
HRP-linked anti-rabbit IgG	Cell Signaling	Cat#7074
RNAScope Reagents
PDX1 RNAscope probe	ACD Bio	Cat#43081
INS1 RNAscope probe	ACD Bio	Cat#313571-C4
Human RNAscope 4-plex Positive Control Probes for RNAscope Multiplex Fluorescent Assay (POLR2A, PPIB, UBC and HPRT1	ACD Bio	Cat#321801
OPAL fluorophore 520	Akoya Biosciences	Cat#FP1487001KT
OPAL fluorophore 620	Akoya Biosciences	Cat#FP1495001KT
OPAL fluorophore 690	Akoya Biosciences	Cat#FP1497001KT
OPAL Polaris fluorophore 780	Akoya Biosciences	Cat#FP1501001KT
RNAscope 4-plex Multiplex Negative Control Probe	ACD Bio	Cat#321831
Bacterial and virus strains
Biological samples
Human islets	Integrated Islet Distribution Program (IIDP)	RRIDS: SAMN17928660, SAMN25519947, SAMN25980818, SAMN13134368, SAMN13254972, SAMN13515839, SAMN13739565, SAMN12227196, SAMN17928660, SAMN25519947
Chemicals, peptides, and recombinant proteins
Forskolin	Sigma-Aldrich	Cat#F3917
Glucose	Sigma-Aldrich	Cat#G7528
Lipofectamine 3000	Thermo Fisher Scientific	Cat#L3000015
Recombinant IL-1B	Biolegend	Cat#575102
Recombinant mouse interleukin-1 receptor antagonist	Sigma-Aldrich	Cat#SRP6006
Recombinant human interleukin-1 receptor antagonist	MedChemExpress	Cat#AMG-719
TRI reagent	Molecular Research	Cat#NC9277980
Collagenase from Clostridium histolyticum	Sigma-Aldrich	Cat#C7657
Tagment DNA Enzyme (TE) and Buffer (TD)	Illumina	Cat#FC-121–1030
Formaldehyde	Polysciences	Cat#18814
Glycine	Fisher Scientific	Cat#BP381–5
DMSO	Sigma	Cat#D8418
Sodium chloride	Fischer Scientific	Cat#BP35–10
Ethylenediaminetetraacetic acid (EDTA)	Lonza	Cat#51234
NP-40	Sigma	Cat#NP40s
SDS	Sigma	Cat#L3771
Triton X-100	Sigma	Cat#X100
Trypsin	Fisher Scientific	Cat#25300054
Mini EDTA-free protease inhibitor cocktail	Roche	Cat#11836170001
HALT protease and phosphatase inhibitors	Thermo Fisher Scientific	Cat#78444
Anti-rabbit IgG dynabeads	Thermo Fisher Scientific	Cat#11204D
Streptavidin M280 dynabeads	Thermo Fisher Scientific	Cat#11205D
Penicillin-streptomycin	Gibco	Cat#15–140-122
L-glutamine	Fischer Scientific	Cat#25–030-081
Direct-zol miniprep columns	Zymo Research	Cat#R2052
NEBNext Poly(A) mRNA magnetic isolation module	New England Biolabs	Cat#E7490S
NEBNext Ultra II directional RNA library kit	New England Biolabs	Cat#E7760
NEBNext single cell/low input RNA library prep kit for Illumina	New England Biolabs	Cat#E6420S
Ampure XP reagent	Beckman	Cat#A63881
MinElute PCR purification kit	Qiagen	Cat#28004
Q5 High-Fidelity 2X master mix	New England Biolabs	Cat#M0492L
Di(N-succinimidyl) glutarate (DSG)	Proteochem	Cat#C1104
Biocoll	Millipore	Cat#L6155
Critical commercial assays
Insulin ELISA Assays	Crystal Chem Inc	Cat#90082
C-peptide ELISA Assays	Crystal Chem Inc	Cat#90050
NextSeq 500/550 High Output Kit v2.5 (75 Cycles)	Illumina	Cat#20024906
NextSeq 2000 P3 Reagents (100 Cycles)	Illumina	Cat#20040559
NE-PER Nuclear and Cytoplasmic Reagent Kit	Thermo Fisher Scientific	Cat#78835
Deposited data
Sequence data, analyses, and resources related to the sequencing of human and mouse tissues and cells	Gene Expression Omnibus	GSE214678
Experimental models: Cell lines
Beta-TC-6 cells	American Type Culture Collection (ATCC)	CRL-11506
BMAL1 KO cells	Dr. Joseph Bass, Northwestern University
Experimental models: Organisms/strains
Pdx1+/− mice	Dr. Roland Stein, Vanderbilt University
Pdx1ΔAIV/− mice	Dr. Roland Stein, Vanderbilt University
PerLuc mice	Dr. Joseph Takahashi, UT Southwestern
Oligonucleotides
Anti-Pdx1 siRNA	ThermoFisher	Cat#4390771
Select Negative Control No. 1 siRNA	ThermoFisher	Cat#AM4611
ATAC-seq indexing primers 1–12 92	Integrated DNA Technologies	Custom order
Recombinant DNA
Software and algorithms
R (v4.1.2)	The R Foundation
DESeq2 package (v1.36.0)	Love, 201457	Bioconductor
Seurat package (v4.1.0)	Satija, 201593	Bioconductor
pathfindR (v2.3.0)	Ulgen94
Lme package (v1.1)	DOI: 10.18637/jss.v067.i01	Bioconductor
Cell Ranger ATAC (v2.0.0)	10X Genomics	https://www.10xgenomics.com/software
Cell Ranger (v6.1.2)	10X Genomics	https://www.10xgenomics.com/software
ChIPseeker package (v1.31.3)	Yu, 201595	Bioconductor
ChromVAR package (v1.16.0)	Schep, 201727	Bioconductor
Cicero package (v1.3.5)	Pliner, 201896	Bioconductor
STAR (v2.5.0)	Dobin, 201397	https://github.com/alexdobin/STAR
RSEM (v1.3.3)	Li, 201198	https://github.com/deweylab/RSEM
Prism (v9.0)	GraphPad
HiC-Pro (v3.1.0)	Servant 20187	https://github.com/nservant/HiC-Pro
Juicer (v1.6)	Durand, 20 1 699	https://github.com/aidenlab/juicer
HiC-Explorer (v2.2)	Ramirez, 2018100	https://github.com/deeptools/HiCExplorer
FitHiChIP (v8.0)	Bhattacharyya, 201989	https://github.com/ay-lab/FitHiChIP
ClockLab Data Collection and Analysis System (v6.0)	Actimetrics	https://actimetrics.com
Other
Islet snATAC-seq dataset	Chiou, 20213	GSE160472, GSE160473, GSE163610
Islet snRNA-seq dataset	Xin, 201825	GSE114297
R scripts and objects		https://github.com/bjw032/PDX1_scATAC

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

Male Pdx1^+/− control and Pdx1^ΔAIV/− mutant mice (provided by Dr. Roland Stein, Vanderbilt University), as well as PdxCre;Bmal1^fx/fx mice, were maintained on a C57BL/6J background at the Northwestern University Center for Comparative Medicine^7,12. Pdx1^+/− and Pdx1^ΔAIV/− mice were crossed with Per2^Luc mice⁸² to generate Pdx1^+/−;Per2^Luc and Pdx1^ΔAIV/−;Per2^Luc mice. All animals were maintained on a 12:12 light:dark cycle and allowed free access to water and regular chow unless otherwise indicated. All animal care and use procedures were conducted in accordance with regulations of the Institutional Animal Care and Use Committee at Northwestern University.

Beta-TC-6 cell culture

Beta-TC-6 cells (ATCC) and Bmal1 KO cells⁸³ were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 15% fetal bovine serum, 1% penicillin-streptomycin, and 1% L-glutamine at 37°C with 5% CO₂. Culture medium was exchanged every two to three days. All cells used in these experiments were <15 passages and were routinely checked for mycoplasma contamination.

METHOD DETAILS

Single-nuclei ATAC-seq sample prep

Human islets were obtained from the Integrated Islet Distribution Program (IIDP) and are listed under RRIDS: SAMN17928660, SAMN25519947, and SAMN25980818 (Table S7). Approximately 200 human islet equivalents (IEQs) were cultured in RPMI supplemented with 10% human AB serum, 1% penicillin/streptomycin, and 1% glutamine at 5% CO₂ for two days prior to 10X processing. Mouse islets were obtained and pooled from 2–3 animals to acquire ~200 IEQs and maintained in in RPMI supplemented with 10% heat-inactivated fetal bovine serum (FBS), 1% penicillin/streptomycin, and 1% glutamine at 5% CO₂. Both human and mouse islets were dissociated into single cells using gentle trituration at 3 min and 6 min for a total of 6 minutes of trypsin-EDTA treatment at 37°C with 600 rpm rotation. Trypsin digestion was then quenched with media and passed through a 40 μm Flowmi cell strainer to obtain a single cell suspension. Cells were then pelleted at 500g for 5 min followed by two washes with 1X PBS + 0.04 % BSA. Cells were resuspended in 100 μL 1X lysis buffer (10 mM pH 7.4 Tris-HCl, 10 mM NaCl, 3 mM MgCl₂, 0.1% Tween-20, 0.1% Tergitol, 0.01% digitonin, and 1% BSA) and incubated on ice for 3 min, immediately followed by 5x passages through a 1 mL insulin needle (31 gauge) to fully lyse membranes. Nuclei were then dispensed into 1 mL lysis wash buffer (10 mM pH 7.4 Tris-HCl, 10 mM NaCl, 3 mM MgCl₂, 0.1% Tween-20, and 1% BSA) and pipetted to mix. Nuclei were pelleted at 500g for 5 min and resuspended in 1X nuclei buffer (10X Genomics) for viability staining and counting. Nuclei were then subjected to transposition, gel bead-in-emulation (GEM) generation, and cell-barcoding according to manufacturer’s recommendations.

Single-nuclei-sequencing library sequencing, alignment, and quantitation

Libraries were amplified and barcoded using 9–11 cycles of PCR. Average base pair size and library quality were determined using Bioanalyzer, and library concentrations quantity were determined using Qubit. Libraries were then subjected to paired-end sequencing using 34×34 paired-end reads, 8 cycles for i7, and 16 cycles for i5 index for barcode and UMI demultiplexing. snATAC samples were demultiplexed and aligned to reference genomes (Hg38 or mm10) using Cell Ranger ATAC (v2.0.0) for ATAC or Cell Ranger (v6.1.2) for 10X RNA-seq (GSE114297) using the “mkfastq” followed by “count” then “aggr” commands to generate nonnormalized individual and aggregate feature count by cell-barcode matrices for input into R.

Single-cell clustering and variable feature analysis

The analytical pipeline has been made publicly available and was performed using R version 4.1.2. In brief, snATAC peak matrices and cell metadata were read into Seurat (v4.1.0). Following low-quality cell removal (cells with either <15% reads in peaks, blacklist ratio > 0.05, nucleosome signal > 4, or transcription start site enrichment < 2), UMAP clustering was performed. For human preparations, Harmony⁸⁴ integration was first performed on a PCA reduction. For both mouse and human snATAC samples, UMAP reduction and clustering was then performed to identify distinct cell clusters. Variable features for cell type identification and analysis were used following peak annotation (ChIPseeker, v1.31.3) and quantification of gene activity was scored as the sum of accessible reads at fragments overlapping gene coordinates and the 2 kb region upstream (hg38 and mm10). To identify variable TF signatures, we assessed TF activity using chromVAR (v1.16.0) to measure TF-associated chromatin accessibility using position weight matrices from both mouse and human TF motifs annotated in the JASPAR 2020 database²⁶. UMAP coordinates were used in performing coaccessbility analysis via cicero (v1.3.5). Significant features enriched between cell clusters or genotypes were determined using a likelihood ratio test from comparing a logistic regression for gene activity, peak accessibility, or TF accessibility as appropriate⁸⁵. Pathway enrichment analysis was performed for differential gene activities between mouse genotypes was performed using Metascape with terms “KEGG” and “Gene Ontology Pathways” included for analysis.

RNAscope multiplex analysis

Paraffin-embedded human pancreas slides (human RRIDs: SAMN12227196, SAMN17928660, SAMN25519947) were obtained from IIDP. Multiplex RNAscope hybridization was done by the Northwestern University Research Histology and Phenotyping Laboratory. Briefly, tissue sections were dewaxed, treated with 3% H₂O₂ for 10 min and incubated at 95°C for 15 min in Target Retrieval Reagent (ACD Bio). Cooled slides were rinsed twice in 100% ethanol, dried for 5 min at 60°C, and marked with ImmEdge Hydrophobic Barrier pen. Slides were treated with RNAscope Protease Plus (ACD) for 30 min at 40°C, hybridized with human RNAscope probes for PDX1 and INS (ACD, Cat#43081, Cat#313571-C4, respectively) or Human RNAscope^® 4-plex Positive Control Probes for RNAscope Multiplex Fluorescent Assay (POLR2A, PPIB, UBC and HPRT1) (ACD Cat#321801) and then with OPAL fluorophores OPAL 520, OPAL 620, OPAL 690 and OPAL Polaris 780 (Akoya Biosciences, Cat#FP1487001KT, Cat#FP1495001KT, Cat#FP1497001KT, Cat#FP1501001KT) for 2 hrs at 40°C, and processed for probe detection using RNA-scope Multiplex Fluorescent Detection Kit (ACD, Cat#323110). To control for background noise, slides were hybridized with RNAscope^® 4-plex Multiplex Negative Control Probe were used every run (ACD Cat#321831). Following hybridization, all slides were mounted in Vectashield VIBRANCE Plus DAPI (Vector Labs) and imaged using Nikon Ti2 Widefield fluorescence microscope and NIS Elements Software (Nikon). To assess the distribution of β cells according to RNA abundance, images of 13–35 islets/donor were analyzed by quantifying fluorescence for each probe in individual cells using Fiji image processing package (ImageJ). Raw values for PDX1, HPRT1, UBC, PPIB, and POLR2A) were normalized to DAPI staining. Best fit for single- and multiple-component distributions was determined using the fitdistrplus and mixtools packages in R, respectively. Multiple components were considered for normalized expression value distributions which exhibit a significant deviation (P-Value < 0.05 by Kolmogorov–Smirnov test) from a single component Gaussian distribution.

Bulk ATAC-seq preparation and analysis

Mouse or human islets (10 IEQs per sample per timepoint, human RRIDs: SAMN13134368, SAMN13254972, SAMN13515839, SAMN13739565) were synchronized using a 10 μM forskolin pulse for 1 hour, followed by culturing in normal media for 24–48 hours as indicated. Single cell suspensions were obtained as described above. Following a brief ice-cold PBS wash, cells were resuspended and pipetted three times in 50 μL ice-cold resuspension buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl, 3 mM MgCl2) containing 0.1% IGEPAL, 0.1% Tween-20, and 0.01% digitonin. Cells were then pelleted and washed twice in resuspension buffer without detergent. Cells were then resuspended in 50 μL of ATAC-seq reaction buffer containing 1.25 μL transposase enzyme (TE, Illumina, FC-121–1030) in a 50 μL 1X transposition mix (25 μL 1Χ TD, 16.5 μL PBS, 0.5 μL 1% digitonin, 0.5 μL 10% Tween-20, 6.25 μL water) and incubated at 37°C for 30 minutes in a thermomixer with 1000 RPM mixing. The transposition reaction was immediately quenched and DNA purified using the Qiagen MinElute Reaction Cleanup Kit. Samples were eluted followed by amplification and barcoded using Nextera indexes for 8–10 cycles Q5 High-Fidelity 2X Master Mix. PCR-amplified libraries were then cleaned up using two-sided size selection (0.5X followed by 1.3X Ampure XP, per manufacturer’s instruction). Library quality control and quantitation was performed using an Agilent 2100 Bioanalyzer. Libraries were sequenced using paired-end sequencing (42×42 bp) on a NextSeq 500. Following sequencing, library quality control and alignment to mm10 or hg38 was performed using the ENCODE pipeline. Peaks were called for individual merged samples using 150 bp peak width to identify open chromatin regions, followed by peak annotation and read quantitation using HOMER. Following removal of peaks with < 10 average counts per sample, we then performed differential peak analysis at filtered consensus peaks using the DESeq2 algorithm in R to identify significantly (FDR < 0.05) altered peak regions with respect to circadian time and/or genotype in a two-factor general linearized model using log-odd ratio testing. We compared significantly altered peaks against the filtered input peak list to identify enriched TF motifs using the findMotifsGenome.pl tool in HOMER. DESeq2 normalized counts were calculated and used for box- and barplots using the ggplot package within R. Rhythmic ATAC-seq analysis was performed using the approach previously utilized for analysis of human islet RNA-sequencing data⁷. In brief, normalized DESeq2 counts were transformed into Z-scores for each individual islet donor across the 24-hour timeseries. Z scores were then imported into eJTK_CYCLE using the provided 24-hour cosine wave parameter at a resolution of 2 hours to identify significantly rhythmic peaks (Bonferroni P-Value < 0.05).

ChIP-sequencing

Approximately 1 × 10⁷ cells per replicate were fixed on tissue culture treated-plates in 2 mM disuccinimidyl glutarate in PBS + 1% DMSO for 30 min, followed by formaldehyde (1% in PBS) cross-linking for 10 min, followed by quenching to a final concentration of 125 mM glycine. Nuclei were isolated in lysis buffer (150 mM NaCl, 5 mM EDTA at pH 8, 50 mM Tris-HCl at pH 8, 0.35% NP-40) using five passages through a needle and syringe in buffer in the presence of Roche Mini EDTA-free protease inhibitor cocktail. Nuclei were pelleted at 2000 × g for 5 min to obtain a rough nuclear isolate. Chromatin was sheared using a Covaris M220 with 75 W power; 10% duty factor, 200 cycle per burst, 4-minute total time at 4°C, in shearing buffer (1% SDS, 2.5 mM EDTA at pH 8, 50 mM Tris-HCl at pH 8). Dilution buffer (0.01% SDS, 1.1% Triton X-100, 167 mM NaCl, 1.2 mM EDTA at pH 8, 1.67 mM Tris-HCl at pH 8) was added 9:1 for dilution and antibodies added (10 μL anti-p65, CST#8242; 10 μL anti-PDX1, CST#5679; Millipore ABE2599, anti-SIN3A, CST#8056S), followed by rotation overnight at 4°C. Secondary antirabbit (for anti-p65 and anti-PDX1) magnetic beads were added (40 μL) and incubated for an additional 6 hours at 4°C with rotation. Magnetic beads were washed 6 times in lysis buffer followed by one additional wash in 1X TE. Two bead elutions were carried out at 25°C in 100 μL of 100 mM NaHCO₃ and 1% SDS for 30 min with 1000 rpm rotation. Eluted DNA was then reverse crosslinked with the addition of 8 μL NaCl (5 M) and rotation at 65°C overnight. ChIP DNA was then treated with 1 μL of 10 μg/mL RNAase A and 1 μL of 10 μg/mL of proteinase K for a total of 1 hour at 37°C and purified using the Qiagen MinElute kit. Libraries were then prepared for sequencing using the NEBNext Ultra II library preparation kit, size-selected using Sage Pippin PrepHT for fragments between 200 and 600 bp. Libraries were indexed and amplified for 11 cycles of PCR followed by DNA purification using 0.9 × Ampure XP bead size selection, analyzed for quality and base pair size using an Agilent 2100 Bioanalyzer, and sequenced on an Illumina Nextseq 500 using 75-bp single-end reads for more than 20 million unique reads aligned per sample. Raw sequencing reads were aligned to the mm10 genome using Bowtie2 (v2.2.4) with default parameters. Peaks were called for each sample relative to sheared DNA input using the “findPeaks” function in HOMER with the optional “-style factor” parameter. A differential peak threshold of >2-fold change above input (for peak identification) or relative to control (for siRNA experiments) was considered significant. The “mergePeaks” function in HOMER was used to identify unique peaks identified between SIN3A ChIP-seq peaks. Enriched TF motifs (taken from JASPAR2020) were detected using the findMotifsGenome.pl tool in HOMER using an FDR cut-off of 0.05. Differential peak analysis was performed at filtered consensus peaks with raw tag counts >10 using the DESeq2 algorithm in R to identify significantly altered binding (FDR < 0.05). HOMER normalized tag counts were imported into R to generate scatter and boxplots.

HiChIP-sequencing

Cells were prepared according to the absolute quantification of architecture HiChIP (AQuA-HiChIP), with a few modifications as noted below⁸⁶. 1 × 107 cells per replicate were fixed on 15-cm tissue-culture treated plates in formaldehyde (1% in PBS), followed by quenching to a final concentration of 125 mM glycine. Cells were lysed in lysis buffer (150 mM NaCl, 5 mM EDTA at pH 8, 50 mM Tris-HCl at pH 8, 0.35% NP-40) using five passages through a needle and syringe in buffer in the presence of Roche Mini EDTA-free protease inhibitor cocktail. Rough nuclei were pelleted at 2000 × g for 5 min and then resuspended in 0.5% SDS in TE pH 7.4, incubated at 62°C, and quenched with 10% Triton X-100. MboI digestion was performed followed by overhang blunting and biotinylation using biotin-dATP. Ligation using T4 DNA ligase for 4 hours at room temperature followed by overnight at 4°C. Chromatin was pelleted and then lysed and sheared using a Covaris M220 with 75 W power; 10% duty factor, 200 cycle per burst, 4-minute total time at 4°C) in shearing buffer (1% SDS, 2.5 mM EDTA at pH 8, 50 mM Tris-HCl at pH 8). Dilution buffer (0.01% SDS, 1.1% Triton X-100, 167 mM NaCl, 1.2 mM EDTA at pH 8, 1.67 mM Tris-HCl at pH 8) was added 9:1 for dilution and anti-PDX1 added overnight at 4°C with rotation (10 μL anti-PDX1, CST#5679), followed by 6-hour secondary incubation using anti-rabbit IgG Dynabeads. Following bead washing, elution, decrosslinking and purification using MiniElute columns (see above), streptavidin M280 Dynabeads were prewashed twice in Tween Wash Buffer (5 mM Tris-HCl pH 8.0, 0.5 mM EDTA, 1 M NaCl, 0.05% Tween-20), resuspended in 2 × Biotin Binding Buffer, and added to reverse crosslinked DNA to bind biotinylated fragments. Following bead binding, beads were washed 3 × times in Tween Wash Buffer followed by 3 × washes with TE pH 7.4. Library preparation was performed as above for ChIP-seq, without size selection. Libraries were sequenced using 42×42 bp paired end sequencing using an Illumina NextSeq 500. Data analysis was performed using the AquA-HiChIP pipeline using HiC-Pro (v3.1.0) followed by Juicer (v1.6)⁸⁷. Valid pair files were processed using HiCExplorer (v2.2) followed by analysis in R using the package rtracklayer to generate genome browser tracks alongside ChIP-seq samples⁸⁸. Significant interactions were determined using mouse bulk ATAC-Seq peaks or IL-1β inducible p65 peaks using the “Peak2Peak” option via FitHiChIP (v8.0) and using an FDR P-Value cutoff < 0.05⁸⁹. Significant interactions were visualized using the WashU Epigenome Browser.

RNA isolation and RNA-sequencing

Cells (or islets) were rapidly lysed in 350 μL TRI reagent, vortexed for 10s at high speed, frozen, and processed for RNA isolation using using Direct-zol miniprep columns (Zymo) with DNA digestion. mRNA libraries were prepared using either NEBNext Ultra II Directional RNA Library Prep Kits (New England Biolabs) with poly(A) enrichment for cells, or the NEBNext Single Cell/Low Input RNA Library Prep Kit for islets. Libraries were sequenced to a depth > 15M reads per library using 75–100 bp single-end reads. Reads were mapped to the mm10 genome using annotations from gencode vM10 assembly with STAR version 2.5.0 followed by quantification using RSEM version 1.3.3 using default parameters. Differential expression was carried out using DESeq2 with default parameters, and pathway analysis was performed using Metascape and pathfindR (v2.3.0).

Pdx1 knockdown and IL-1β treatment in Beta-TC6 cells

Knockdown of Pdx1 was performed in Beta-TC6 cells using reverse transfection in OptiMEM by adding 12 pmol anti-Pdx1 siRNA (ThermoFisher) or control siRNA (ThermoFisher Select Negative Control No. 1 siRNA) per well and 7.5 μL of Lipofectamine 3000 per well of a 6-well dish for RNA. For ChIP, transfection was performed in 10 cm tissue-cultured plates using 70 pmol siRNA and 43.5 μL lipofectamine. Cells were then plated at 5 × 10⁵ cells/well for RNA and 5 × 10⁶ for ChIP. For Western blots, transfection was performed in 6 cm plates using 33 pmol siRNA again 15 μL of Lipofectamine 3000, and cells were plated at 4 × 10⁵ cells/dish. 48 hours post-transfection, cells were treated with recombinant IL-1β (BioLegend) for 2 hours at a final concentration of 0.1 ng/mL in normal media (DMEM supplemented with 5% FBS, 1% glutamine, and 1% penicillin/streptomycin) and harvested for RNA-seq, ChIP-seq, qPCR, or Western blots.

Nuclear and cytoplasmic fractionation and Western blotting

Cytoplasmic and nuclear proteins were harvested using NE-PER Nuclear and Cytoplasmic Reagent Kit (Thermo Fisher Scientific) with the following modifications: nuclear pellets were vortexed for 15 sec in the NER1 buffer with HALT Protease and Phosphatase Inhibitors (Thermo Fisher Scientific) and then sonicated at 4°C for 3–6 cycles of 30 sec pulses. Samples were spun at 16,000g for 10 min and supernatants were stored at −80°C. Protein levels were quantified using BioRad DC Protein Assay and protein expression was analyzed with Western blotting using 20 μg of each fraction. The Immobolin-P Membrane (Sigma) was hybridized to the following antibodies: PDX1, NF-κB p65, Histone H3, GAPDH, and HRP-linked anti-rabbit IgG (Cell Signaling, Cat#5679, Cat#8242, Cat#9715, Cat#5174, Cat#7074, respectively). Signal was detected using Super Signal West PICO PLUS Chem-Illuminescent Substrate (Thermo Fisher Signaling) and X-ray film.

Insulin secretion assays in pancreatic islets, pseudoislets, and cell lines

Mouse pancreatic islets were isolated via bile duct collagenase digestion (Collagenase P, Sigma) and Biocoll (Millipore) gradient separation and left to recover overnight at 37°C in RPMI 1640 with 10% FBS, 1% L-glutamine, and 1% penicillin/streptomycin. For insulin release assays, duplicates of 5 islets of equal size per mouse were statically incubated in Krebs-Ringer Buffer (KRB) at 2 mM glucose for 1 hr, then stimulated for 1 hr at 37°C with 2 mM or 20 mM glucose in the presence or absence of 0.5 μg/mL murine Interleukin-1 receptor antagonist (Sigma). Supernatant was collected and assayed for insulin content by ELISA (Crystal Chem Inc). Islets were then sonicated in acid-ethanol solution and solubilized overnight at 4°C before assaying total insulin content by ELISA. For insulin release assays from pseudoislets, Pdx1 knockdown in Beta-TC-6 cells was achieved as described above. Then, 3 × 10⁶ cells were plated for 3 days in 60 mm suspension dishes and allowed to form pseudoislets. Glucose-responsive insulin secretion was performed as described above, using 10 pseudoislets per sample and a basal glucose level of 0 mM glucose instead of 2 mM. When indicated, 0.1 ng/mL recombinant mouse IL-1β was used.

Oral glucose tolerance tests and hyperglycemic clamps

Mice were fasted for 14 hrs, and glucose tolerance tests were performed at ZT2 and ZT14 following glucose administration of 2 g/kg body weight through oral gavage. Blood glucose was measured at indicated times by hand-held meter, and plasma insulin and C-peptide levels were measured by ELISA (Crystal Chem Inc). Hyperglycemic clamps were performed in 5 hr fasted chronically catheterized (carotid artery and jugular vein) conscious mice at the Vanderbilt Mouse Metabolic Phenotyping Center^{90, 91}. Blood was obtained from the carotid artery and glucose was infused into the jugular vein to raise and maintain arterial glucose levels around 280 mg/dl. Washed red blood cells were infused to replace blood removed for serial sampling.

LumiCycle analysis

Approximately 100–150 pancreatic islets from Pdx1^+/−;Per2^Luc and Pdx1^ΔAIV/−; Per2^Luc mice were isolated as described above. Islets were then cultured on tissue culture membranes in modified DMEM⁷. Sealed cultures were placed at 37°C in a LumiCycle luminometer (Actimetrics), and sample bioluminescence was recorded continuously. After several days in culture, islets were synchronized by the addition of 10 μM forskolin for 1 hr, followed by incubation in fresh media. Period length was calculated via a modified best-fit sine wave analysis using LumiCycle analysis software (Actimetrics).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses

Data was analyzed using R (v4.1.2) and Prism (v9.0). Differential analysis of single-cell sequencing data was performed using the Seurat (v4.1.0) package in R (v4.0.1) using the likelihood-ratio test (“LRT”) method with minimum “percent expressed” set to 0.10 and “minimum difference in percent expressed” set to 0.05. Two-way ANOVAs were used to analyze insulin secretion and GTTs, followed by Holm-Šídák multiple comparison testing. The “lme4” package (v1.1) was used to assess the effect of genotype and drug in the hyperglycemic clamping data by using a mixed linear model and setting time as fixed random effect (https://www.jstatsoft.org/article/view/v067i01).

Supplementary Material

1

Data S1. Unprocessed data underlying the display items in the manuscript, related to Figures 3–4, S1, and S4.

3

Table S1. Genes with significant variation in gene activity scores between β1 and β2 subclusters, related to Figure 1

4

Table S2. Comparison of gene expression and chromatin activity between a cell subpopulations, related to Figure 1

5

Table S3. Differential motif accessibility between β1 and β2 subpopulations, related to Figure 1

6

Table S4. All significant PDX1 HiChIP loops, related to Figure 4

7

Table S5. PDX1 HiChIP loops called at IL-1β-inducible p65 sites, related to Figure 4

8

Table S6. Genes co-bound by SIN3A and PDX1, related to Figure 4

9

Table S7. Summary of human islet donor information, related to Star Methods

HIGHLIGHTS.

• snATAC-seq reveals critical role of PDX1 gradient in adult human β-cell function

• PDX1-SIN3A co-repression of latent NF-κB enhancers determines β-cell identity

• Crosstalk between PDX1-BMAL1-NF-κB controls secretory and apoptotic gene networks

• Blocking peptide to IL-1β improves insulin secretion in PDX1 hypomorphic mice