Mining the transcriptome of target tissues of autoimmune and degenerative pancreatic β-cell and brain diseases to discover therapies

Summary Target tissues of autoimmune and degenerative diseases show signals of inflammation. We used publicly available RNA-seq data to study whether pancreatic β-cells in type 1 and type 2 diabetes and neuronal tissue in multiple sclerosis and Alzheimer’s disease share inflammatory gene signatures. We observed concordantly upregulated genes in pairwise diseases, many of them related to signaling by interleukins and interferons. We next mined these signatures to identify therapies that could be re-purposed/shared among the diseases and identified the bromodomain inhibitors as potential perturbagens to revert the transcriptional signatures. We experimentally confirmed in human β-cells that bromodomain inhibitors I-BET151 and GSK046 prevent the deleterious effects of the pro-inflammatory cytokines interleukin-1β and interferon-γ and at least some of the effects of the metabolic stressor palmitate. These results demonstrate that key inflammation-induced molecular mechanisms are shared between β-cells and brain in autoimmune and degenerative diseases and that these signatures can be mined for drug discovery.

Mining the transcriptome of target tissues of autoimmune and degenerative pancreatic b-cell and brain diseases to discover therapies

RNA-sequencing datasets
Scientific stractegy Type

INTRODUCTION
Autoimmune diseases are diseases of ''mistaken identity'' where the immune system -which is supposed to protect us against infectious diseases and neoplasias -attacks and destroys components of our body. There is no cure for autoimmune diseases and their incidence is increasing worldwide. These conditions -including type 1 diabetes (T1D) and multiple sclerosis (MS) -affect up to 5-8% of the population in different regions. 1 Although the immune targets of these diseases are distinct, they share several features, including up to 50% common genetic risk loci, chronic local inflammation, and consequently target tissue damage. 1,2 Other highly prevalent degenerative diseases, such as type 2 diabetes (T2D) and Alzheimer's disease (AD), show inflammatory but not autoimmune components. [3][4][5][6] Despite these common features, autoimmune disorders are traditionally studied independently and with a focus on the immune system rather than on target tissues. There is increasing evidence that the target tissues are not innocent bystanders of the autoimmune attack but participate in a deleterious dialogue with the immune system that contributes to their own demise as shown in a recent study by our group. 7 This dialogue is supported by changes in the proteome induced by inflammatory mediators that amplify autoimmune responses. 8 Furthermore, in T1D, several of the risk genes for the disease act at the target tissue level (i.e., pancreatic b-cells), regulating the responses to viral infections, 9 the dialogue with the immune system and apoptosis. 10, 11 We hypothesize that key inflammatory mechanisms, potentially shared between T1D, MS, T2D and AD, may induce similar molecular signatures at the target tissue level. Discovering similar (or, in some cases, divergent) signatures may allow the identification of key pathways that could be mined and then, based on the information obtained, targeted for therapy based for instance on the repurposing of drugs already in clinical use for other diseases.
The rationale for selecting these b-cell and brain diseases includes: (1) The striking gene expression similarity between pancreatic b-cells and neurons, including expression of splicing regulators and splice variants 12,13 ; (2) the fact that T1D and MS have several candidate genes in common, and express -at least to some extent -similar upregulated inflammatory pathways at the target tissue levels 7 ; and (3) the potential role for inflammation and amyloid deposition in T2D and AD. 5,6 The T1D samples consisted of FACS-purified pancreatic b-cells, whereas no purification was done for the other diseases, raising the possibility of significant infiltration by immune cells. To address this, we determined expression of the leukocyte marker CD45 in all samples. There was increased CD45 expression in T2D and AD, but it remained fairly low, particularly for T2D (Table S1). For comparison, although the mean transcript per million (TPM) for CD45 in T1D, T2D and their controls ranged from 1 to 15. Inflammatory genes are upregulated in target tissues of the four diseases Differential analysis of the modified genes indicated more up-than downregulated genes in T1D and T2D, whereas there were more downregulated genes in MS and AD ( Figure 1A). Gene set enrichment analysis (GSEA) based on the Reactome database 21 showed that interferon (IFN)-g-regulated pathways -an indicator of adaptive immunity -were augmented in T1D, MS and AD, but not in T2D ( Figures 1B-1E). Antigen

OPEN ACCESS
processing and presentation and IFNa/b signaling pathways were induced in T1D and MS ( Figures 1B and  1D). These results were supported by GSEA based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, 22 which also indicated upregulation of antigen presentation in AD (Figures S1A, S1C, and S1D). The ''type 1 diabetes'' pathway was enriched not only in T1D but also in MS and AD, pointing to resemblance of disease-related genes in target tissues of T1D, MS and AD (Figures S1A, S1C, and S1D). We also found induction of ''apoptosis'' in T1D, T2D and MS, but not in AD (Figures S1A-S1D). Of , and AD (E) target tissues based on the Reactome database. Bars in red and blue represent enrichment or inhibition of pathways, respectively. The x-axis shows the normalized enrichment score (NES) of the fGSEA analysis, and the y-axis shows enriched pathways with an adjusted p-value <0.05. The full name of pathway ''Immunoregulatory interactions'' in (B, D and E) is ''Immunoregulatory interactions between a lymphoid and a non-lymphoid cell''; ''Runx1 regulates transcription of genes'' in A is ''Runx1 regulates transcription of genes involved in differentiation of HSCS''; ''Runx1 regulates genes involved in megakaryocyte differentiation'' in A is ''Runx1 regulates genes involved in megakaryocyte differentiation and platelet function''; ''Gene and protein expression by JAK stat signaling '' in B is ''Gene and protein expression by JAK stat signaling after interleukin-2 stimulation''; ''Respiratory electron transport ATP synthesis'' in D is ''Respiratory electron transport ATP synthesis and heat production'' (see also Figure S1).
Analysis of TF binding sites in promoter regions (transcription start site G 2 kb) of differentially expressed genes for each disease identified a clear enrichment of IFN-induced TFs in the upregulated genes of T1D, including IFN regulatory factor 1 (IRF1), IRF2, and IRF8 ( Figure S2A), which is in line with the marked induction of IFN-related pathways in T1D ( Figure 1B). These TFs also appeared in islets from T2D patients ( Figure S2B). REST-NRSF (neuron-restrictive silencer factor) was identified as the unique TF that binds to a set of downregulated genes in AD and as the top TF for T2D downregulated genes ( Figures S2B and S2D).
We next investigated the overlap between significantly modified genes (either up-or downregulated) of the four diseases, using a false discovery rate <0.1 cutoff ( Figure S3). There were 229 common genes between T1D and T2D, but less than 100 genes overlapped between the two types of diabetes and the other two diseases. Through a hypergeometric test using Reactome and KEGG databases as references, ''cytokine signaling'', ''interleukin signaling'', ''interferon-g signaling'' and ''type 1 diabetes'' pathways were commonly upregulated in two or three diseases (Table S2A). Of note, there were only two commonly upregulated genes (i.e., MS4A7 and MSR1) between the four diseases (Table S2A). The function of commonly downregulated genes between T1D and T2D were identified as neuronal function-related (e.g., ''neurotransmitter receptors'', ''postsynaptic signal transmission'' and ''GABA receptor activation''), b-cell function-related (e.g., ''regulation of gene expression in b-cells'') and energy metabolism-related (e.g., ''integration of energy metabolism'') (Table S2B). Because these genes were filtered by a fixed statistical threshold, this type of analysis largely depends on the number of samples analyzed. Despite the limitation of this approach, the enriched pathways (for either up-or downregulated genes) between two or three diseases generally agree with the above-mentioned GSEA results.

Rank-rank hypergeometric overlap-based pairwise analysis demonstrates similarities between the four diseases mostly related to inflammation
We next used the rank-rank hypergeometric overlap (RRHO) analysis 23 to compare global transcriptomic signatures between the four diseases without the limitation of a fixed threshold (see STAR methods). We observed generally similar pairwise transcriptomic signatures between the four diseases, particularly for upregulated genes. The highest correlation was observed between T1D and T2D, both for up-and downregulated genes ( Figure 2A). There was an unexpected and highly significant correlation between upregulated, but not downregulated, genes of T1D and AD, which is in line with the identification of ''type 1 diabetes'' as an upregulated pathway in AD ( Figure S1D). In contrast, T2D presented a larger number of downregulated than upregulated genes in common with AD (Figures 2A and 2B).   We next investigated the potential upstream TFs orchestrating the inter-disease similarities by predicting the TF binding sites in the promoter region (transcription start site G 2 kb) of commonly upregulated genes from the pairwise diseases comparison by RRHO (Figure 2). In line with the presence of IFN-related pathways ( Figures 3E-3H), there was a clear enrichment of motifs for IFN-induced TFs, including IFN-stimulated response element (ISRE), IRF1, IRF3, IRF8, and type I IFN-stimulated response element (T1ISRE), when comparing MS to T1D or T2D and AD to MS or T1D (Figures S5E-S5H). Binding sites for NFkB-p65 were enriched for commonly upregulated genes between T1D and T2D, which is in line with the identification of NF-kB and TNF signaling in islet cells in these diseases ( Figure S4A). For commonly downregulated genes from T2D vs T1D and T2D vs AD, we again identified REST-NRSF ( Figures S5B and S5D), a transcriptional repressor of neural genes, which is in line with the downregulation of neuronal pathways ( Figures 3B  and 3D).
We next compared the differential expression of selected genes between the four diseases allocated visually by one of us (DLE) to potentially relevant functional groups. Many key genes belonging to antigen presentation were predominantly modified in the target tissues of T1D and to a lesser extent in AD and MS, but not in T2D (Table S3). Chemokines, cytokines, complement and IFNs were markedly upregulated in b-cells from T1D patients. Some genes related to granule release and synaptic cycle were downregulated in T2D and AD target tissues. There were few changes in autophagy, lysosomal degradation, free radical scavenging and DNA damage response genes. Several genes critical for b-cell function and belonging to glucose and lipid metabolism, protein translation and modification were affected in T2D and T1D. These observations were generally supported by functional enrichment performed individually or pairwise (Figures 1, S1, 3, and S4).

Identification of potential therapeutic targets based on top concordant genes identified between diseases
To identify potential therapeutic targets, we compared the top 150 commonly up-or downregulated genes from the RRHO analysis against the cell signatures induced by chemical perturbations in the Connectivity Map (See STAR methods). We identified perturbagen classes driving opposite signatures to the ones we submitted, indicating that these chemical perturbations could reverse commonly altered pathways and have potential therapeutic use ( Figures 4A-4H). The most consistently highly ranked perturbagen (often with |median tau scores| >90) was ''Bromodomain inhibitor'' when analyzing upregulated genes from the comparisons T2D versus T1D, T2D versus AD, T2D versus MS, MS versus T1D and T1D versus AD ( Figures 4A, 4C, 4E-4F, and 4H). We have recently shown that two broad-action bromodomain inhibitors, namely I-BET151 and JQ1, prevent some of the deleterious effects of IFNa (a cytokine involved in the early steps of islet inflammation in T1D 10,25 ) on human b-cells. 26 SRC inhibitors and JAK inhibitors were further predicted as potential drugs for commonly perturbed genes in T2D and T1D (Figures 4A and 4B). Importantly, JAK inhibitors have been shown by us and others to protect human b-cells against pro-inflammatory cytokines 26,27 and to prevent diabetes in mouse models. 27 One of these JAK inhibitors, baricitinib, is presently being tested as a potential therapy for T1D (Clinical Trials.govNCT04774224).
Based on these findings, we evaluated the impact of a broad-action (I-BET151) and a more specifically inflammation-targeting (GSK046) 28 bromodomain inhibitor in two models of human b-cell dysfunction in T1D and T2D, respectively the cytokines IFNg + IL1b that contribute to b-cell apoptosis at more advanced stages of islet inflammation in T1D and the metabolic stressor palmitate that contributes to b-cell dysfunction and death in T2D. 10,16,29-31 Exposure of human islets to IFNg + IL1b for 48 h induced the mRNAs encoding for HLA-ABC, the chemokine CXCL10, the cytokines IL6 and IL8 and the endoplasmic reticulum (ER) stress markers CHOP and BiP (Figures 5A-5F). This was confirmed at the protein level, by measuring CXCL10 and IL6 accumulation in the medium (Figures 6A and B). IFNg + IL1b also induced apoptosis ( Figure 5G). The bromodomain inhibitors I-BET 151 and GSK046 reduced the pro-inflammatory effects of the cytokines and lowered CHOP expression ( Figures 5A-5E, 6A, and 6B) but did not prevent apoptosis ( Figure 5G). To investigate whether these effects take place at least in part at the b-cell level, we exposed the human b-cell line EndoC-bH1 to the same cytokines with or without the bromodomain inhibitors for 24 ( Figures S6A-S6E) or 48 h (Figures S7A-S7E) iScience Article bromodomain inhibitors partially prevented most cytokine-induced pro-inflammatory gene expression but did not prevent apoptosis.
We next evaluated whether bromodomain inhibitors could protect human islets against the metabolic stressor palmitate (Figures 7A-7G). These experiments were not undertaken in EndoC-bH1 cells because these cells are resistant to palmitate due to their high stearoyl CoA desaturase expression. 32 In human islets, palmitate induced the chemokine CXCL1, IL6 and IL8 and CHOP, BiP and spliced XBP1, and apoptosis ( Figures 7A-7G). The beneficial effects of the bromodomain inhibitors were less marked in the context of palmitate than with cytokines ( Figures 5A-5G). There was less palmitate induction of CXCL1 and IL8 ( Figures 7A and 7C), and for the latter, the protection was observed with iBET-151 but not GSK046. The bromodomain inhibitors did not protect the human islet cells from ER stress ( Figures 7D-7F) or apoptosis ( Figure 7G).  iScience Article

DISCUSSION
In the present study we investigated the hypothesis that key gene expression changes, potentially related to inflammation, are commonly present in the target tissues of autoimmune or degenerative diseases affecting pancreatic b-cells, namely T1D and T2D, and the brain, namely MS and AD. Exposure to different forms of stress leaves ''molecular signatures'' at the target tissue levels, and discovering similar gene signatures may allow the identification of key pathways to be targeted for therapy via drug repositioning or discovery. 14,25 The transcriptome analysis of the target tissues in the four diseases showed concordant upregulation of cytokine-cytokine receptor interaction, chemokine signaling, and JAK-STAT signaling, supporting the idea that these diseases contain similar inflammatory molecular components ( Figure S1). The IFNg-regulated pathways, an indicator of adaptive immunity, were found augmented in T1D, MS, and AD, but not in T2D. This is in line with previous studies, showing that both innate and adaptive immunity are essential for developing MS 33 and that neuroinflammation and innate immunity are hallmarks of AD. 34 Adaptive immunity and its interactions with microglia are important for restraining AD through limiting amyloid pathology 35 but at later phases of the disease may contribute to neuronal damage. Regarding T1D, the complex interaction between b-cells and innate/adaptive immune cells is critical for the development of the disease. 10,36 Of note, the most intense expression of inflammatory markers in target tissue was observed in T1D, suggesting a more important impact of inflammation in this disease as compared to MS, T2D and AD. The present data could be confounded by immune cells infiltrating the target tissues, which could impact gene expression of inflammation/adaptive immunity pathways. Expression of the leukocyte marker CD45 was, however, low in the four tissues studied (Table S1), and IFN signatures are present in purified b-cells and neurons in T1D and MS, respectively 7 . Specific b-cell function pathways were downregulated in T1D and T2D, whereas neuronal function pathways were downregulated in AD. These changes imply dysfunction of the target tissues.
The functional enrichment for RRHO-quadrant genes between diseases cross-validated these molecular alterations, especially for the highly concordant upregulation of inflammatory pathways. As potential upstream regulators, we predicted many inflammation-induced TFs (e.g., IRFs and NF-kB-p65) for commonly upregulated genes in most combinations of the four diseases. Of interest, the TF REST-NRSF was predicted as an important regulator for commonly downregulated genes between T2D, AD, and T1D, which is in line with the striking depletion of neuronal function pathways in the pairwise comparisons between the three diseases. NRSF represses pancreatic endocrine and neuronal development and function through the recruitment of multiple transcriptional and epigenetic co-regulators that restrict endocrine or neuron fate acquisition; its expression wanes as these cell types differentiate. [37][38][39] Based on the molecular changes shared between the four diseases, we mined in silico for drug repurposing strategies using the Connectivity Map L1000 platform, 14 which includes pre-computed differential gene expression signatures from nine core cell lines exposed to chemical or genetic perturbations. The anti-correlation between highly concordant signatures of target tissues and Connectivity Map libraries allowed us to identify perturbagen classes that could target more than one disease (Figure 4). Among them, JAK inhibitors, acting downstream of types I and II IFN receptors, block the activation of JAK1 and JAK2 kinases. The JAK inhibitor baricitinib protects pancreatic b-cells against the deleterious effects of type I IFNs, 26 and it is being tested for the prevention of T1D (Clinical Trials.govNCT04774224). Src family tyrosine kinase is expressed at high levels in cells specialized for exocytosis, such as neuronal and endocrine cells. It has been iScience Article shown that PP2, an Src family tyrosine kinase inhibitor, enhances neurotransmitter release from neuronal cells. 40 Two structurally different Src family kinase inhibitors, SU-6656 and PP2, enhanced Ca 2+ -dependent insulin secretion in rat pancreatic islets and INS-1 cells. 41 The enriched ''neurotransmitter release cycle'' pathway for commonly downregulated genes from T2D and AD ( Figure 3D) suggests that indeed the use of Src inhibitors could be beneficial for both diseases. The bromodomain and extra-terminal domain family of epigenetic reader proteins regulate inflammatory and cancer-related gene expression. 42 Bromodomain inhibitors are being tested as a potential therapy in experimental models of AD, 43  iScience Article MS. 45 Treatment with JQ1 in 3-month-old mice carrying 3 mutations associated with familial Alzheimer's disease reduced neuroinflammation, with decreased expression of pro-inflammatory modulators (e.g., IL-1b, Il-6 and TNFa). 43 A short treatment with iBET-151 in non-obese diabetic mice prevented both insulitis and diabetes. 44 Furthermore, JQ1 treatment in mice with experimental autoimmune encephalomyelitis (a model of MS) significantly protected them from encephalomyelitis by selectively preventing the generation of T H 17 cells, essential effectors of autoimmunity in this model. 45 Finally, the BET protein inhibitor Apabetalone decreased the ex vivo inflammatory responses of monocytes obtained from patients with type 2 diabetes/cardiovascular diseases. 46 We have shown that iBET-151 prevents IFNa-induced inflammatory pathways but not apoptosis in human islets. 26 GlaxoSmithKline has recently described the inhibitor GSK046 (targeting the second bromodomain) that is particularly effective in models of inflammatory and autoimmune diseases, including psoriasis, collagen-induced arthritis and non-alcoholic fatty liver disease. 28 iBET-151 and GSK046 significantly protected human b-cells from the pro-inflammatory but not from the pro-apoptotic effects of IFNg + IL1b (Figures 5, S6, and 7), cytokines that mimic advanced inflammatory features in T1D, present in the later stages of insulitis. 10 There was a milder protective effect of the bromodomain inhibitors against palmitate-induced chemokine production, but they did not alter palmitate-induced ER stress or apoptosis (Figure 7). These agents may hence be more useful in settings of autoimmune b-cell destruction.
In conclusion, we integrated the transcriptomes of target tissues from four major diseases affecting b-cells or the brain. We identified commonly dysregulated gene signatures and mined these for potential therapeutic candidates. We validated i-BET151 and GSK046 as promising drugs to rescue pancreatic b-cells from aggressive inflammation in diabetes.

Limitations of the study
A limitation of the present study is that the original RNA-seq datasets were generated in different studies, using different RNA-seq methods, and obtained from patients of different ages and sex. Due to differences in disease prevalence and the difficult access to target tissues, we only had 4-5 samples of target tissues of individuals affected by T1D or MS as compared to larger numbers for T2D (28 individuals) and AD (122). This may decrease the power of the analysis and lead to fewer modified genes being identified in T1D and MS. Despite these limitations, we identified disease-specific gene expression signatures -mostly related to downregulated genes -and commonly upregulated gene signatures -mostly related to inflammationin the four diseases.
We acknowledge that the study only tested i-BET151 and GSK046 on human beta-cells in vitro and that a follow-up in vivo validation study is needed. It also remains to be tested whether similar beneficial effects can be observed for neuronal protection in MS and AD.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following: . Quantitative real-time PCR was performed using SYBR Green and data were expressed as number of copies/mL using a standard curve. Gene expression was corrected by the geometric mean of the reference genes b-actin and VAPA, as their expression is not modified under the experimental conditions used here. 58 The highest value of each experiment was considered as 1. Primers sequences are listed in Table S5.

ELISA
Supernatants of dispersed human islet cells (30,000 cells/200 mL) pretreated with the bromodomain inhibitors IBET-151 or GSK046 and exposed or not to IFN-g + IL1b were used to determine CXCL10 and IL6 secretion to the medium by ELISA (Quantikine ELISA kit, R&D Systems, Minneapolis, MN, USA).

Assessment of apoptosis
The percentage of viable, apoptotic, and necrotic cells was assessed by microscopy after nuclear dye staining (propidium iodide, 10 mg/mL, and Hoechst 33342, 10 mg/mL, Sigma-Aldrich, St. Louis, MO, USA). A minimum of 500 cells was counted for each experimental condition by two different observers, one of them unaware of sample identity.

QUANTIFICATION AND STATISTICAL ANALYSIS
Quality control, quantification, and differential analysis of RNA-seq data Raw RNA-seq data of target tissues from T1D, 15 T2D, [16][17][18] and AD 59 were gathered from the Gene Expression Omnibus (GEO) Portal, European Genome-Phenome Archive (EGA) and Synapse Platform (Table 1). For each dataset, the raw RNA sequencing reads in Fastq format were processed with fastp 0.19.6 60 using the default parameters for quality control, adaptor trimming, and quality filtering to obtain clean reads for downstream analysis. Gene expression levels of target tissues were quantified as TPM with Salmon 1.4.0 47 using additional parameters ''-seqBias-gcBias-validateMappings'' to remove potential sequencing bias. The transcriptome reference was based on the indexed GENCODE version 36 (GRCh38.p13) 61 with the default k-mer values. Differential analysis was performed by DESeq2 1.28.1. 48 There was a relatively even distribution of age and sex between cases and controls. Because the T2D cohort was collected from three independent studies [16][17][18] and AD cohorts were sequenced from three separate batches of samples, 59 we applied batch correction in the general linear model used in DESeq2, taking batch as a confounding factor and formulated the design matrix (design = $ batch + condition) to estimate the dispersions and the log 2 fold changes of the model. All other parameters used in the differential analysis of the diseases were similar. After correction of batch effects, a log 2 fold change was computed and a Wald test was assessed with a p-value and an adjusted p-value (Benjamini-Hochberg correction) for differential analysis by DESeq2. The threshold to determine a gene as differentially expressed was adjusted p-value <0.05.

ll
OPEN ACCESS iScience 25, 105376, November 18, 2022 iScience Article Functional enrichment using GSEA or hypergeometric test GSEA was based on pre-ranked Wald statistics (the ratio of log 2 fold change and the SE of estimation) generated from the DESeq2 pipeline. The fGSEA algorithm 49 was performed against the Reactome 21 and KEGG 22 databases. The number of permutations was set as 50,000 for the most accurate p-values and the gene sets, including the number of genes between 15 and 500, were chosen as references.
Significantly enriched pathways (adjusted p-value <0.05, Benjamini-Hochberg correction) were then sorted according to their normalized enrichment score (NES). To decide the functional enrichment of genes significantly overlapped in RRHO pairwise analysis, we conducted a hypergeometric test incorporated in clusterProfiler (3.12.0) tool 62 against the Reactome and KEGG databases for genes with common up-or downregulation. Pathways with adjusted p-values <0.05 (Benjamini-Hochberg correction) were considered significantly enriched.

Rank-rank hypergeometric overlap (RRHO) pairwise analysis
To compare the global transcriptomic signatures of target tissues, we applied the RRHO algorithm, 23 an unbiased and threshold-free method to reveal similarities and dissimilarities between diseases. For each pair of diseases, genes measured in both experiments were ranked according to their log 2 -transformed fold-change generated by DESeq2, from the most up-to the most downregulated ones. A hypergeometric test was performed to assess the significance of the similarity of gene profile, using a sliding window with step size (i.e., 50) for each pair of diseases. A False Discovery Rate correction was applied to adjust for the multiple hypothesis testing. The visualization of the output of this analysis is the RRHO level map (Figure 2A), in which the most significant hypergeometric p-value (log 10 transformed and direction-signed) was labeled after computing all possible rank combinations, generating an index of the matrix for the most significant rank combination in each pair of diseases. Based on the hypergeometric test, we defined the most significant commonly regulated genes as the intersected genes above the most significant rank combination (with the most significant hypergeometric p-value). The RRHO level map is visualized as a heatmap displaying the degree of the similarities or dissimilarities in quadrants (e.g., commonly up-or downregulated in two diseases, upregulated in one disease and downregulated in the other).

TF binding site discovery
TF binding sites were searched with HOMER software 63 in the promoter regions of genes from up-or downregulated genes for each disease or the commonly up-or downregulated genes in a disease pair identified by RRHO. The promoter regions were defined as the G 2,000 base pairs from the transcription start sites of these genes. We used the script findMotifs.pl incorporated in HOMER with the parameters ''-start-2000-end 2000-length 8,10,12''. TFs predicted with an enrichment p-value <0.05 by a hypergeometric test were considered significant.

Identification of potential therapeutic targets
To identify potential therapeutic targets for pairs of diseases, we selected the top 150 most up-or downregulated genes from the RRHO common gene set and submitted them to the Connectivity Map L1000 platform 14 through the cloud-based CLUE platform (https://clue.io). The gene signatures revealed from our datasets were matched with the ones included in Connectivity Map libraries, which contain gene signatures for cells under many chemical or genetic perturbations. This allowed us to search for potential drugs that could restore the differential transcriptomes for more than one disease.

Statistical analysis for the human b-cell experiments
Data are expressed as means G SEM EndoC-bH1 cells from different passages or human islets from different donors were considered as independent experiments. Differences between experimental conditions were assessed by Student's paired t-test or one-way ANOVA or linear mixed model in case of missing values, followed by Bonferroni correction for multiple comparisons as indicated in the figure legends. Results with p-value %0.05 were considered significant. Analyses were performed using GraphPad Prism 9 software (GraphPad Software, La Jolla, CA, USA).