Adaptation to chronic ER stress enforces pancreatic β-cell plasticity

Pancreatic β-cells are prone to endoplasmic reticulum (ER) stress due to their role in insulin secretion. They require sustainable and efficient adaptive stress responses to cope with this stress. Whether episodes of chronic stress directly compromise β-cell identity is unknown. We show here under reversible, chronic stress conditions β-cells undergo transcriptional and translational reprogramming associated with impaired expression of regulators of β-cell function and identity. Upon recovery from stress, β-cells regain their identity and function, indicating a high degree of adaptive plasticity. Remarkably, while β-cells show resilience to episodic ER stress, when episodes exceed a threshold, β-cell identity is gradually lost. Single cell RNA-sequencing analysis of islets from type 1 diabetes patients indicates severe deregulation of the chronic stress-adaptation program and reveals novel biomarkers of diabetes progression. Our results suggest β-cell adaptive exhaustion contributes to diabetes pathogenesis.

The use of chemical stressors is critical to observe temporal changes during the progression from acute to chronic ER stress in cells 22 and in vivo [23][24][25][26] . Genetic chronic ER stress models [27][28][29] do not provide the necessary synchronized control over these transitions; however, numerous studies show that the biochemical principles of the ER stress response are shared between chemically-induced stress in vitro and in vivo [24][25][26] . Therefore, we used the reversible inhibitor of the sarcoendoplasmic reticulum pump Ca 2+ ATPase (SERCA) pump, cyclopiazonic acid (CPA), to model resolvable chronic ER stress in the mouse insulinoma cell line, MIN6. This inhibitor induces ER stress that can be tuned in duration by washing out 22 . Using CPA, we showed that β-cells undergo a state change associated with adaptive reprogramming of their transcriptome and translatome that compromises β-cell identity. A comparative analysis of adaptive reprogramming between β-cells (MIN6) and mouse embryonic fibroblasts (MEFs) revealed an elaborate, β-cell-specific network of genes that exhibit stress-induced expression changes. Adaptation to chronic ER stress in β-cells included changes in expression of 334 genes, with a predominant   downregulation of genes involved in Maturity Onset Diabetes of the Young (MODY), proinsulin expression, synthesis, processing, and secretion. MEFs did not show differential expression of these genes expressed during chronic ER stress 22 . In addition, β-cells induced expression of more than half of genes known to participate in ER protein processing (133) under chronic adaptation conditions. We define this subset of genes as the β-cell-specific ER adaptome. Upon relief from stress, β-cells were able to regain expression of mature identity markers, revealing the resilience of β-cells to near lethal ER stress. Remarkably, upon iterative cycles of ER stress, β-cells lost this plasticity. Laser capture-microdissected islets 30,31 and single-cell RNA sequencing (scRNA-seq) data from T1D patients highlighted gene expression changes in β-cells of T1D patients, in agreement with chronic ER stress adaptive exhaustion and deregulation of the β-cell ER adaptome. We propose cycles of ER stress exhaust the β-cell's adaptive plasticity and promote an irreversible loss of function that contributes to T1D pathogenesis. We call this novel mechanism the 'β-cell exhaustive adaptive response (βEAR).

Adaptation of MIN6 cells to chronic ER stress sustains insulin levels
MIN6 cells display β-cell-like insulin biosynthesis and secretion in response to glucose 32,33 . To test the effect of severe ER stress on MIN6 cell survival and on proinsulin synthesis, MIN6 cells were treated with CPA 34 , which disrupts Ca 2+ homeostasis in the ER to induce the UPR 35 . As in MEFs 22 , CPA treatment in MIN6 cells induced a time-dependent chronic ER stress response. The expression of hallmark proteins of the UPR, ATF4 and GADD34, showed a gradual increase (Fig. 1a), consistent with protein synthesis recovery 36 . Similar to MEFs 22 , MIN6 cells showed inhibition of protein synthesis followed by partial recovery (Fig. 1b) without recovery of eIF2B activity (Fig. 1c). The resumption of growth following recovery from 18 h of stress confirmed an adaptive stress response at that time point (Fig. 1d). The ability of MIN6 cells to recover from chronic stress was also accompanied by protein synthesis recovery (Fig. 1e) and restoration of eIF2B activity (Fig. 1c). Similarly, relief from ER stress inactivated stressresponse signaling and stress-induced gene expression (Fig. 1f), suggesting a return to homeostasis 22,37 . We conclude MIN6 cells and MEFs have the same core mechanisms of regulating global protein synthesis during adaptation to chronic ER stress 22 .
In MEFs, we have shown adaptive translational reprogramming involves a switch from eIF4E-dependent to eIF4E-independent, eIF3dependent mRNA translation 22 . Because the most efficient mechanism of mRNA translation initiation utilizes eIF4F, we proposed this switch allows inefficient translation of all mRNAs, specifically ER-associated mRNAs. However, we also found a subset of mRNAs encoding stressinduced and ER-synthesized proteins are translated during adaptive chronic ER stress 22 . Consistent with MEFs, we found accumulation of stress-induced proteins in MIN6 cells exposed to CPA was similar in control and eIF4E-depleted cells ( Supplementary Fig. 1a), supporting eIF4E-independent reprogramming of β-cells during chronic ER stress.
We next determined the regulation of proinsulin gene (Ins1 and Ins2) expression during the transition from acute to chronic ER stress. Levels of Ins1 and Ins2 mRNAs decreased in MIN6 cells during chronic stress, suggesting an adaptive response. This decrease was gradually reversed during recovery from stress ( Fig. 1g), supporting the regulation of proinsulin gene expression as part of the adaptive response. We also examined the levels of mature insulin and proinsulin, but neither the insulin monomer nor native insulin hexamer levels were altered during ER stress and recovery ( Supplementary  Fig. 1b). In contrast to the decreased levels of Ins2 mRNA, during adaptive chronic ER stress, proinsulin protein levels decreased dramatically early in the stress response, presumably due to eIF2α-pdependent inhibition of proinsulin mRNA translation 38 . Subsequently, proinsulin protein levels returned to normal at ≥12 h, despite decreased transcript levels during the chronic adaptive stress period (Fig. 1h). To demonstrate the regulation of proinsulin synthesis during progression of β-cells from acute to chronic ER stress, we showed that proinsulin synthesis was repressed at 1 h and de-repressed at 18 h of CPA treatment ( Supplementary Fig. 1c). Further, we demonstrated the efficiency of translation of proinsulin mRNA also decreased at 1 h and was recovered at 18 h of CPA treatment ( Supplementary Fig. 1d). Finally, in order to determine if proinsulin processing in chronic ER stress is similar to that in control cells, we examined the subcellular localization of proinsulin during adaptive chronic ER stress and subsequent recovery. Interestingly, intracellular and cell peripherylocalized proinsulin puncta were observed during stress, but not during recovery (Fig. 1i). The sequestration of proinsulin during adaptive chronic ER stress into intracellular structures suggests proinsulin processing may be inhibited 39,40 . Taken together, our results show MIN6 cells reprogram translation during chronic ER stress with reversible inhibition of protein synthesis and proinsulin mRNA levels.

CPA alters intracellular localization of proinsulin
Due to the essential role of ER-Golgi transport in insulin maturation, we examined proinsulin localization in MIN6 cells following CPA treatment. In CPA-treated cells, proinsulin was associated with the ER, shown by the colocalization of proinsulin and P4HB, an ER-resident molecular chaperone. CPA treatment (6 h) induced punctate staining of proinsulin in most cells. While the majority of proinsulin still colocalized with P4HB, the fraction of proinsulin in puncta did not stain positive for P4HB (Fig. 2a, b), suggesting enrichment of proinsulin in granule-like structures outside the ER near the cell periphery. After 18 h treatment, more than a third of cells had significant proinsulin staining in these peripheral granule-like structures, suggesting proinsulin largely migrated from the ER and became stable in these structures, likely due to decreased processing. We further tested this hypothesis by comparing the degradation of proinsulin in non-stressed MIN6 cells and cells exposed to chronic ER stress via treatment with the protein synthesis inhibitor cycloheximide (CHX). Proinsulin in control cells is very unstable; 41 in contrast, we found proinsulin was very stable during chronic ER stress ( Supplementary Fig. 1e). Furthermore, proinsulin degradation was insensitive to proteasome inhibition. Finally, functional recovery of CPA-treated MIN6 cells following CPA washout indicated partial recovery of the CPA-inhibited glucose-stimulated insulin secretion (GSIS) (Supplementary Fig. 1f). Taken together, we posit the resilience of β-cells to chronic ER stress includes positive and negative mechanisms of regulation of proinsulin gene expression.
Because plasma proinsulin is a risk factor for diabetes 42 , we also determined if proinsulin is secreted from MIN6 cells treated with CPA 43 ( Supplementary Fig. 2a, b). We show proinsulin is secreted during treatment with CPA (1 h and 18 h) and secretion decreases following washout of CPA from cells (after 18 h treatment). These data support ER stress-mediated accumulation of proinsulin in β-cells can lead to its secretion.
We next examined whether proinsulin colocalized with the ER-Golgi intermediate compartment (ERGIC), a transient intracellular structure between ER and Golgi 44 . We found no colocalization between proinsulin punctate staining and the ERGIC marker ERGIC53 in MIN6 cells in CPA-treated conditions (Fig. 2c). Next, we examined proinsulin and ERGIC53 in INS1E, a rat insulinoma cell line, and EndoC βH3, a human β-cell line. Large vesicle-like proinsulin-enriched structures appeared following 6 h of CPA treatment in INS1E cells (Fig. 2d, e). This early response of INS1E cells is consistent with the higher sensitivity of INS1E cells to ER stress. In human EndoC βH3 cells, CPA-induced punctate proinsulin immunostaining was observed near the cell periphery after 72 h of treatment and also did not colocalize with ERGIC53 (Fig. 2f). Similar proinsulin puncta at the cell periphery were observed in β-cells in mouse islets treated with CPA for 18 h (Fig. 2g). Collectively, our data suggest during ER stress adaptation, proinsulin maturation is impaired and proinsulin accumulates in structures outside the ER (Fig. 2h).

MIN6 cells attenuate expression of genes associated with β-cell identity during chronic stress
The high demand for ER proteostasis in β-cells motivated us to examine β-cell-specific changes in gene expression in MIN6 cells during the progression from acute to chronic stress ( Supplementary  Fig. 3a, b). To this end, we analyzed effects of ER stress on genomewide perturbations in transcriptomes and translatomes using total RNA sequencing (RNA-seq) and ribosome profiling. We analyzed the data for the Ins2 gene for validation (about 50% of the mRNA in β-cells maps to Ins genes); while mRNA-seq reading frames in Ins2 mRNA were unchanged, the distribution of ribosome footprints was enriched at reading frame 0 of the mRNA ( Supplementary Fig. 3c), in agreement with ribosome footprints being at the ribosome P site 45,46 . Data quality was further confirmed by the selective enrichment of Ribo-seq reads on Ins2 mRNA within the coding region, in contrast to the mRNA-seq reads distributed throughout the gene (Supplementary Fig 3d). We showed a high correlation among independent experiments for both transcriptome and translatome datasets (Supplementary Fig. 3e). Finally, we confirmed UPR pathway genes were significantly upregulated in all independent experiments (Supplementary Fig. 3f), validated by qRT-PCR ( Supplementary Fig. 3g) and Western blotting (Fig. 1a).
Next, we determined the regulation of the global transcriptome and translatome during acute and chronic ER stress. For the translatomes, we normalized ribosome footprints to mRNA-seq datasets and assigned the normalized values the term 'ribosome occupancy' (ribo ocp ), as an indicator of the translation efficiency. We detected 10,163 transcripts encoding 6,938 genes in both transcriptome and translatome datasets. A significant change in mRNA abundance was observed in chronic (CPA18), but not acutely (CPA1) stressed cells compared to control cells (Fig. 3a), suggesting that reprogramming of the transcriptome is associated with adaptive chronic stress. We next assessed differences in ribo ocp in MIN6 cells treated with CPA ( Fig. 3b). At CPA18, total ribo ocp decreased compared to the untreated control, suggesting reprogramming of mRNA translation. We then classified the differentially regulated transcripts into six groups by trends of changes in mRNA abundance (Fig. 3c) and ribo ocp (Fig. 3d) in response to CPA (Supplementary Data 1). Although a few mRNAs showed a transient increase or decrease in abundance in the acute response (G2 and G5), the predominant effect was observed in chronic stress, with 2036 genes being upregulated and 385 genes being downregulated compared to control cells (G3 and G6). Translational control in the acute response was evident by 847 mRNAs being ribo ocp -up (G8) as compared to 246 mRNAs with continuous upregulation during progression of stress from acute to chronic (G7). In agreement with partial recovery of protein synthesis, 703 mRNAs decreased their ribo ocp during progression from acute to chronic ER stress (G10). To better understand establishment of adaptive proteostasis in chronic ER stress, we performed pathway analysis of these groups using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Transcripts associated with ribosome biosynthesis, protein processing, degradation and transport, as well as RNA transport and splicing were upregulated during chronic ER stress. In contrast, mRNAs related to the ribosome, cell cycle, and DNA replication were markedly decreased in the acute or chronic stress states (Fig. 3e). Importantly, transcripts associated with β-cell identity (Nkx2-2, Pdx1, Glut2, Ins1, and Ins2) and   function (Glut2) were substantially decreased between CPA1 and CPA18. These results suggest MIN6 cells can withstand acute ER stress and maintain β-cell-specific gene expression. However, β-cell transcripts are progressively reduced upon establishment of adaptive homeostasis. These data further suggest loss of β-cells in diabetic islets may be caused by adaptive responses to chronic ER stress without recovery. We refer to this cellular response as the "β-cell exhaustive adaptive response" to chronic ER stress, or βEAR.
Transcriptome reprogramming during chronic stress involves impaired expression of β-cell genes To further assess how temporal changes of reprogramming relate to β-cell identity, we performed additional analysis of transcriptomes and translatomes of CPA1 (Fig. 4a) and CPA18 (Fig. 4b) vs control. In the acute response, we identified 798 mRNAs (11.5%) with increased and 300 mRNAs (4.32%) with decreased ribo ocp , in agreement with translational reprogramming during acute stress. Changes in mRNA abundance were less prominent, with levels of 138 mRNAs increasing and 108 mRNAs decreasing (Fig. 4a). In contrast, there were more changes in mRNA abundance (1795) than translational control (1183) during the chronic response ( Fig. 4b). It was notable, that changes in mRNA abundance increased from 1.99% in the acute phase to 20.41% in the chronic, while changes in ribo ocp were similar between the two. An early increase in mRNA abundance and translation was observed for Atf4, which was sustained during the chronic phase, but generally the percent/fold increase in ribo ocp was larger than the percent/fold increase in mRNA levels (Figs. 4c and 1a). Similarly, downstream of Atf4, Chop and Gadd34 showed more than 2-fold increases in mRNA abundance and 3-fold increases in ribo ocp in the acute phase. We also examined changes in expression of 49 UPR-related genes ( Fig. 4c and Supplementary Data 2). A shift in mRNA abundance was observed for most well-known UPR genes, as well as prosurvival genes, such as Wolframin ER Transmembrane Glycoprotein (Wfs1) 47 . In conclusion, β-cells show an enhanced response during the acute phase, with elevated mRNA abundance and ribo ocp for adaptation proteins.
The β-cell-specific transcriptome is important to maintain β-cell functions 48 , misregulation of which can result in diabetes 49,50 . We found the mRNA abundance of β-cell-related genes, such as MAF BZIP Transcription Factor A (MafA) 51,52 and Glut2 (Fig. 4d), a glucose transporter essential for glucose import in rodent β-cells 53-55 , severely decreased during chronic ER stress and rapidly increased to pretreatment levels after CPA washout (Fig. 4e, f). Since MAFA is a transcriptional activator of insulin 56 , we expected the positive association observed between MafA and Ins2 mRNA levels ( Fig. 4g).
Genes associated with ER protein processing were upregulated in chronic stress, including those encoding molecular chaperones and factors involved in ER-associated degradation (ERAD) and ubiquitination (Fig. 4h). We identified an increase of VCP-interacting membrane selenoprotein (Vimp), Suppressor/Enhancer of Lin-12like (Sel1l), and Derlin-2 (Derl2) in chronic ER stress, suggesting an active role for ERAD in the degradation of misfolded proinsulin 57 . We confirmed mRNA levels were increased by CPA treatment, and after washout, showed that Sel1l mRNA levels returned to basal levels ( Fig. 4i). We further noticed that genes encoding ER molecular chaperones were differentially regulated during chronic ER stress and recovery from stress (Fig. 4j). Furthermore, we found chronic stress impaired both mRNA abundance and maturation of the convertase subtilisin/Kexin Type 2 (Pcsk2), a peptidase essential for insulin maturation (Fig. 4k, l). Because molecular chaperones are important in proinsulin disulfide bond formation and maturation, we investigated proinsulin misfolding in β-cells during ER stress. Using non-reducing SDS-PAGE gel electrophoresis, we found a gradual increase of high-molecular weight disulfide-linked proinsulin complexes upon CPA treatment; CPA washout was sufficient to rescue this phenotype over time (Fig. 4m). Collectively, we show the balanced regulation of ER quality control genes results in reversible ER stress, but also compromises proinsulin gene expression and protein maturation.

ER-Golgi transport in adaptation to ER stress in MIN6 cells
It is currently unknown whether secretory cells utilize specific cellular machinery to achieve adaptive homeostasis compared to nonsecretory cells. Thus, we compared the transcriptomes of acute and chronic ER stress adaptation in MIN6 cells and MEFs. This analysis in MIN6 cells identified 1534 mRNAs with increased and 307 mRNAs with decreased abundance between the acute and the adaptive timepoints, respectively ( Supplementary Fig. 4a). Similarly, we identified 603 and 296 mRNAs subjected to positive or negative translational control, respectively. mRNAs encoding proteins of the ribosome biosynthesis pathway were enriched in the positive group, while the ribo ocp of cell cycle related mRNAs was reduced. Analysis of pre-existing data from MEFs 22 identified 567 and 1247 mRNAs which showed upregulation or downregulation of mRNA abundance, respectively. KEGG pathway analysis of the downregulated genes between MIN6 cells and MEFs showed the stress responses were cell type-specific and tailored to cellular function: a decrease of β-cell-related functions was observed in MIN6 cells, while mRNAs of cell adhesion and ECM proteins were decreased in MEFs during the transition from acute to chronic stress adaptation ( Supplementary Fig. 4b).
In corroboration with common adaptation mechanisms 58 , we identified an enrichment of genes in the ER protein processing pathway among mRNAs with increased abundance ( Supplementary  Fig. 4c). 68 mRNAs showed increased abundance and 10 mRNAs showed translational upregulation of this cellular pathway in MIN6 cells. In MEFs, a smaller proportion of mRNAs (37 of 78), were regulated, suggestive of similar, but less elaborate adaptation machinery. We analyzed the 37 overlapping genes between MIN6 and MEFs by Gene Ontology (GO) 59 , and observed these genes were highly associated with ER stress amelioration, including molecular chaperones and ERAD. The 41 MIN6-specific genes were enriched for functions in ER-to-Golgi vesicle transport ( Supplementary  Fig. 4d). This analysis revealed a unique adaptive event in MIN6 cells, i.e. an enhancement of the first step of vesicular transport, suggesting stress adaptation is influenced by cellular function. We termed this group of 78 genes in MIN6 cells the 'β-cell-specific adaptome'.

Lysosome function is impaired during chronic ER stress
Formation of proinsulin granules was reversible, supporting their formation as an adaptive response to chronic ER stress (Fig. 1i). To shed light on the mechanism of proinsulin accumulation, we examined gene regulation in chronic ER stress (Fig. 5a), and noted that genes encoding subunits of the proteasome were upregulated (Fig. 5b). By pathway analysis of the 497 downregulated mRNAs, we found lysosome biogenesis genes, such as Arsb, a lysosomal arylsulfatase, and Lamp2, a lysosomal associated membrane protein, were significantly repressed (Fig. 5c), suggesting impaired lysosome function. Accordingly, we observed a time-dependent reduction of lysosomal enzymatic activity during CPA treatment which gradually returned to levels of untreated cells 6 h after CPA washout (Fig. 5d). We also monitored the protein levels of the autophagy substrate ubiquitin-binding protein p62 (Sqstm1), and the conversion of microtubule-associated protein 1 A/1B-light chain 3 (LC3-I) to LC3-II in MIN6 cells. We observed a gradual increase of p62 and LC3-II levels during CPA treatment (Fig. 5e). Lastly, we observed an increase of punctate LC3 staining in MIN6 cells after 18 h CPA treatment (Fig. 5f). Taken together, we conclude the pathway of autophagosome-to-lysosomal degradation was inhibited during chronic ER stress in MIN6 cells.

Impairment of lysosome activity correlates with misfolded proinsulin accumulation
Our analysis demonstrated lysosomal activity gradually declined during chronic ER stress in MIN6 cells, while the ubiquitinproteasome system (UPS) was upregulated (Figs. 3e, 4h and 5c). Moreover, we revealed an accumulation of misfolded proinsulin outside the ER upon CPA treatment (Figs. 2 and 4m), implying disrupted proinsulin proteostasis in chronic stress. To determine if the UPS or the lysosome were involved in this regulation, we treated MIN6 cells with CPA for 18 h together with proteasome inhibitor MG132 or Baf A1 and monitored proinsulin folding status. Lysosomal impairment by Baf A1 over the last 4 h of the 18 h of CPA treatment strongly increased the amount of misfolded proinsulin (Fig. 5g). Moreover, CPA washout in the presence of Baf A1 led to h.
Ins2 j.   dramatic accumulation of misfolded proinsulin, while MG132 had only a modest effect. Therefore, our data suggest CPA-mediated chronic ER stress impairs clearance of misfolded proinsulin by disrupting autophagosome-to-lysosomal degradation.

Chronic ER stress adaptation genes are downregulated in T1D
Pancreatic β-cells encounter numerous microenvironment changes over their life span, and studies have highlighted the importance of the ER stress response in diabetes 17,60 . We hypothesized that β-cell loss/   dysfunction in diabetes could be associated with the homeostatic state of ER stress adaptation or its failure. To test this hypothesis, we selected genes that were affected in MIN6 cells during the chronic adaptive state, and filtered out those that exhibited similar behavior in MEFs 22 . This allowed us to establish a β-cell-specific ER stress gene dataset. This gene set, which we called the 'regulome', contains 334 genes enriched for β-cell-associated functions by KEGG pathway analysis (Fig. 6a). We examined the expression levels of this gene set in a pre-existing microarray database (Supplementary Data 4), generated from laser-excised islets 30,31 and the scRNA-seq data produced by the Human Pancreas Analysis Program (HPAP) 61 from patients with T1D and T2D and healthy donors (Fig. 6b). We identified 11 genes that showed downregulation in T1D samples (Fig. 6b), but not in T2D samples ( Supplementary Fig. 5a). One gene, PPP1R1A, was identified in both T1D and T2D, but showed less significant changes in T2D. 10 of the 11 downregulated genes were present and decreased in MIN6 cells during chronic ER stress adaptation. Pancreatic islets comprise several cell types, including α-cells (30-50%), β-cells (50-60%), δ-cells (<10%), pp-cells (<5%) and ε-cells (<1%). In T1D patients, there is a specific loss of β-cells; 62,63 thus, the downregulation of the 11 downregulated genes in the T1D microarray dataset could be a result of differential expression in multiple islet cell types. To further investigate β-cell-specific responses in T1D, we first compared the expression profiles of healthy donor and T1D patient islets 61 for the 11 genes in the scRNA-seq dataset ( Supplementary  Fig. 5b). We found that the expression of all 11 genes was decreased in T1D β-cells ( Supplementary Fig. 5c), and using a more stringent cut off (FDR≤10 −3 ), 6 of the 11 genes were significantly downregulated in βcells of T1D patients (Fig. 6c, d). In addition to these 11 genes, INS and MAFA mRNA abundance was also decreased in β-cells of patients with Article https://doi.org/10.1038/s41467-022-32425-7 T1D (Fig. 6d), while G6PC2, ABCC8 and NKX2.2 showed a significant decrease in expression compared to healthy donors (Fig. 6c). All 8 genes shown in Fig. 6d were also downregulated in MIN6 cells during chronic ER stress (Fig. 6e). Lastly, we examined how the regulome was regulated in the scRNA-seq dataset of T1D islets (Fig. 6f). We observed a larger decrease in expression of these genes in β-cells in comparison to other cell types in T1D (β-cells/all other cell types = 57.23%/ 29.13 ± 0.79%). Only 7.51% of the gene set showed an increase in β-cells, lower than other cell types (12.95 ± 2.23%). Collectively, our analysis revealed reduced abundance of distinct mRNAs of β-cell chronic ER stress genes in T1D. This finding supports the hypothesis that chronic ER stress adaptation likely fails in β-cells in T1D, leading to targeted downregulation of β-cell function-and identity-specific mRNAs. We propose failure to adapt to chronic ER stress, or βEAR (Supplementary Data 3), may contribute to the pathogenesis of T1D.
Cycles of chronic ER stress lead to decreased plasticity and loss of β-cell-specific gene expression Decreased expression of β-cell-specific genes in T1D islets and β-cells from T1D patient islets suggested inadequate adaptation to chronic ER stress. To test this, we compared the expression profile of the MIN6 βcell-specific adaptome (78 genes) with scRNA-seq data from T1D and healthy donor β-cells. We found overlap of 46 of these genes expressed in both healthy donors and T1D patients. In contrast to upregulation during chronic adaptation in MIN6 cells ( Supplementary Fig. 4c), most adaptome genes were downregulated in T1D patients compared to healthy donors (Fig. 7a). Notably, the expression of the UPR sensor BiP 64 , was significantly increased in T1D patients. Strikingly, the relative expression of these genes in αand δ-cells is different than in β-cells; many genes were upregulated in αand δ-cells, including BiP, suggesting partial adaptation to ER stress in these cell types (Supplementary Fig. 6a, b). Our data indicate β-cells in T1D subjects are likely more sensitive to ER stress and lose their adaptive ER proteostasis faster than other cell types during disease progression.
Next, we looked at the relevant distance in gene expression between (i) the scRNA-seq dataset from all human β-cells (healthy and T1D, 8,349) and (ii) the expression of UPR signature genes (GO:0030968) 29 and β-cell-specific gene markers (INS, PAX6, NKX2.2, NKX6.1, MAFA, and PDX1). Similarity was determined via the t-distributed stochastic neighbor embedding algorithm, or t-SNE 65 (Fig. 7b). This analysis identified clusters of β-cells with the UPR signature but heterogeneous UPR marker gene expression. By mapping the T1D disease progression ( Supplementary Fig. 5b) on the UPR signature clusters of β-cells, we identified clusters corresponding to healthy donors and others derived from patients with more advanced T1D (Fig. 7b). This further supported differential clustering of gene expression patterns for healthy and T1D subjects (Supplementary Fig. 6c). We then determined the expression of INS mRNA in each cluster to define β-cell identity, and found good correlation between high levels of INS expression and healthy donors. We also noticed a decline of INS mRNA abundance with T1D pathogenesis and an anti-correlation between INS and BiP gene expression with T1D pathogenesis (Fig. 7b). In agreement with unresolved ER stress, expression of the β-cell adaptome marker WFS1 was decreased in T1D pathogenesis (Fig. 7a, b), and was more abundant in the healthy β-cell clusters. Finally, in order to define the maturity of β-cells in T1D subjects, we performed pseudo-timing analysis by in silico pseudotime reconstruction in single-cell RNA-seq analysis using TSCAN 66 , and Monocle 3 67 packages in R. We showed a gradual decrease of βcell markers and increase of BiP (Fig. 7c). Taken together, these data suggest β-cells from T1D patients experience an unresolved UPR.
It has been recently reported that heterogeneity in β-cell maturity in islets is essential for their function in vivo 68 . Thus, our data showing ER stress decreases expression of genes involved in β-cell maturity, combined with the documented heterogeneity of β-cells in islets in vivo, suggests that β-cells undergo cycles of ER stress and recovery in response to metabolic stress. We therefore reasoned the collapsing chronic ER stress adaptation response in T1D islets could result from exposure to frequent ER stress cycles or after prolonged ER stress. We therefore tested these two hypotheses in MIN6 cells. The UPR markers showed complete reversibility after prolonged stress ( Supplementary  Fig. 7a, b). Although prolonged ER stress decreased cell survival ( Supplementary Fig. 7c, d), the surviving cells did not indicate increased cleaved caspase 3 ( Supplementary Fig. 7e). Our data suggest β-cells can endure prolonged ER stress and still recover.
We next determined the effect of multiple cycles of recovery from chronic ER stress exposure on the levels of stress-induced and β-cell specific proteins. MIN6 cells were challenged with four cycles of 18 h CPA treatment followed by 6 h recovery in fresh medium (Fig. 7d). We showed an increase of cell loss with cycle number ( Supplementary  Fig. 7f); however, no cleaved Caspase 3 was detected at the end of the fourth stress cycle, suggesting the attached cells were resilient to cyclic ER stress (Supplementary Fig. 7g). We next examined the expression of UPR genes and β-cell identity markers in response to the stress/ recovery cycles. We found a gradual reduction of UPR proteins with increasing cycle number. Expression of BiP was induced in the first cycle of CPA treatment, but expression was not increased in subsequent cycles (Fig. 7e). In agreement with basal UPR being essential for β-cell health 69 , BiP and XBP1s were expressed in control MIN6 cells (Fig. 7e). Although these data suggest BiP regulation is lost with stress/ recovery cycling, protein levels remained higher in cycling than in untreated cells (Fig. 7e). Notably, the β-cell marker NKX2.2 showed a decrease and partial recovery during the second and third stress/ recovery cycle, but was unchanged following the first cycle, and NKX2.2 protein levels gradually decreased with cycles (Fig. 7e). MAFA, on the other hand, was affected by early stress cycles; MAFA showed a significant decrease, with partial recovery at the end of the first and subsequent cycles (Fig. 7e). Lastly, we found a significant irreversible reduction in Ins1 and Ins2 gene expression in the last two stress cycles (Fig. 7f). The pattern of changes in MafA and Herpud1 mRNA levels in the last two cycles were in good correlation with the changes in their protein levels (Fig. 7e, f). Taken together, these results show increasing frequency of stress/recovery cycling progressively diminish β-cell plasticity and the return to homeostasis. Since T1D is a condition of increased β-cell sensitivity to environmental and nutritional factors, we propose frequent episodes of chronic ER stress in β-cells may contribute to diminution of β-cell identity and promote β-cell loss (Fig. 7g).
Although we focused on molecular adaptive plasticity here, metabolic plasticity is also expected to be part of the adaptation mechanism to chronic ER stress 70,71 . We found biosynthesis of cholesterol, owing to its importance for GSIS 72 , decreased during chronic ER stress and gradually recovered upon stress removal ( Supplementary  Fig. 8a); the inhibition of cholesterol biosynthesis was consistent with the adaptive response of β-cells to chronic ER stress. While Ins mRNA levels decrease, mechanisms of inhibition of secretion of proinsulin or insulin such as decreased cholesterol biosynthesis can generate a reserve pool of insulin for recovery from stress.

Discussion
T1D is preceded by β-cell dysfunction in islets, but the mechanisms are not well understood 73,74 . ER stress could cause β-cell dysfunction and progression of T1D. In the T1D microenvironment, ER stress can develop via several factors inherent to β-cells or in response to cytokines secreted by local T cells and macrophages 74 . Here, we tested the intrinsic effect of ER stress and the impact of recovery from stress on βcell identity as a means to understand the pathogenesis of T1D. Using a chemical that reversibly elicits an ER stress response, we revealed plasticity of β-cells during their response to chronic stress. We showed during chronic stress, β-cells undergo a homeostatic change associated with reprogramming of their transcriptome and translatome. Chronic stress altered more than half of genes known to participate in ER protein processing, while also compromising β-cell identity. Upon relief from stress, β-cells regained their mature identity, indicating plasticity. This suggests that episodes of ER stress, even at high amplitude, are tolerated by β-cells in a manner designed to protect ER function.
Importantly, frequent repeated cycles of ER stress diminished βcell plasticity, as shown by an inability to recover expression of their identity genes, including Ins1 and Ins2. Laser-captured islets and scRNA-seq data from patients with T1D indicated gene expression alterations in T1D β-cells consistent with a model of ER stress adaptive exhaustion and deregulation of the β-cell ER adaptome during disease progression. Thus, it appears that frequent cycles of ER stress over time exhaust the adaptation machinery of β-cells to promote an ultimately irreversible loss of function that likely facilitates T1D pathogenesis.
Our findings on adaptive plasticity to chronic ER stress are reminiscent of recent studies on β-cell heterogeneity in vivo, as a requirement for islet function 68,75 . It was shown in a mouse model that adult islets are composed of mature and immature β-cells, and the balanced expression of these β-cell populations was important for the physiological metabolic response of the islets 68 . The authors indicated that βcell isolation from islets can eliminate differences in maturity. We can speculate that subpopulations of β-cells in islets represent cells in states of adaptation to ER stress (low levels of MafA and Pdx1) and recovery from ER stress (high levels of MafA and Pdx1). Therefore, the heterogeneity in β-cell maturity in normal islets may be a physiological response to cycles of ER stress and recovery in vivo. In our experimental system of CPA-induced ER stress, proinsulin exited the ER and formed granule-like structures. The correlation with the decreased maturation of PCSK2, the protease that cleaves proinsulin to produce the mature insulin 76 , and decreased lysosomal activity, a mechanism of clearance of insulin granules 77 , may implicate these events as part of that mechanism. Additional possibilities could include changes in luminal acidification 78 and decreased cholesterol in immature secretory granule membranes 79 . Furthermore, dysfunction of the zinc transporter SLC30A8, a known autoantigen in T1D, can also impact proinsulin processing 80 . Interestingly, cholesterol biosynthesis ( Supplementary Fig. 8a) and the RNA levels of SLC30A8 decreased during chronic ER stress in MIN6 cells, supporting multiple potential contributions to the inhibition of proinsulin processing.
Our studies in MIN6 cells can be considered a novel methodology to identify biomarkers for T1D and T2D pathogenesis. Most of the 11 genes identified here (Fig. 6) associate with T1D and T2D, suggesting a role of ER stress adaptation in the diabetic homeostatic state. Remarkably, seven of the 11 signature genes are associated with either insulin maturation or GSIS. PCSK1/3 and PCSK2 are involved in the C-peptide removal from proinsulin, a crucial step in insulin maturation. Deficiency of three plasma membrane proteins (GLP1R, ABCC8, and P2RY1) is tightly linked to abnormal GSIS and associated with diabetes [81][82][83] . Two secreted proteins, PPP1R1A and SCGN, also showed decreased levels in both ER stress adaptation in MIN6 cells and T1D patients. PPP1R1A is an inhibitor of the protein-phosphatase 1, and is positively correlated with insulin secretion and negatively with HbA1c. SCGN is an insulin-interacting calcium sensor protein that stabilizes insulin. Due to the secretion property of these two proteins, they have been proposed as biomarkers for diabetes detection 84,85 . SYT4, Synaptotagmin 4, localizes in mature insulin granules and is important for GSIS; SYT4 overexpression increases GSIS threshold, while ablation compromises GSIS 86 . Finally, we implicate a new diabetic candidate gene VAL1L, Vesicle Amine Transport 1-Like, associated with ER stress adaptation in MIN6 cells and T1D. Overall, our data suggest the genome-wide cellular response in MIN6 cells can lead to new discoveries in the pathogenesis in diabetes.
Although our current study modeled changes typical of pre-T1D, alternative experimental systems can study T2D pathogenesis. For example, lipotoxicity can induce ER stress in tissue cultured β-cells, and is a factor associated with β-cell loss in T2D. Similar to the studies presented here via the use of a chemical ER stressor, palmitic acidinduced ER stress in MIN6 cells results in a similar reversible decrease in MAFA protein levels ( Supplementary Fig. 8b, c). Future studies can determine the commonalities between different physiological and pathological stressors and the bona fide ER stress cellular response described here.
One limitation of our study is that while MIN6 cells provide an excellent system to discover factors involved in the resilience of β-cells to chronic ER stress, mimicking the range of stressors in human diabetes that include hyperglycemia, hyperlipidemia, cytokine action, etc. or paracrine interactions in the islet environment is not possible. We showed that in mouse islets treated with CPA, levels of MAFA in β-cells decreased, followed by quick recovery when CPA was washed out (Supplementary Fig. 9a, b). Furthermore, GSIS was blunted in CPAtreated cells and resumed recovery when CPA was washed out (Supplementary Fig. 9c). Hence, future experiments mimicking the islet microenvironment in cultured mouse and human islets would provide invaluable insight into states of reversibility, regulation, and function of βEAR and pave the way for novel approaches preventing the loss of functional β-cell mass in T1D.

Protein sample preparation for Western blot analysis
For protein extraction, cells were washed twice with ice-cooled 1xPBS before lysis. Ice-cooled lysis buffer (50 mM Tris-HCl pH7.5, 150 mM NaCl, 2 mM EDTA, 1% NP-40, 0.1% SDS, 0.5% sodium deoxycholate, supplemented with EDTA-free protease inhibitor (Roche Applied Science) and PhosSTOP phosphatase inhibitor (Roche Applied Science) was added to cells. Cells were scraped off and sonicated on ice. Protein lysates were centrifuged for 5 min in 10,000 × g at 4°C. Supernatant was collected and quantified by DC Protein Assay Kit (Bio-Rad). The lysate was diluted to 1 μg/μL by using lysis buffer. The diluted lysates were mixed with 5x sampling buffer (300 mM Tris-HCl pH6.8, 50% glycerol, 10%(v/v) β-mercaptoethanol, 10%(w/v) SDS and 50 mg bromophenol blue) or Pierce Lane Marker Non-Reducing Sample Buffer (Thermo Scientific) in a 1-to-4 ratio for reducing or non-reducing Western blot analysis, respectively. For non-reducing analysis, gels were soaked with 25 mM DTT for 10 min before electrotransfer to Immobilon-P PVDF membrane (Sigma-Aldrich).

Proinsulin secretion and dot blotting
MIN6 cells were plated onto 10 cm plates and cultured in the cell growth medium until reaching to 60-70% confluency. Cells were treated with 200 μM CPA in a fresh medium until 1 h before the designated treating times. The cell was quickly washed with prewarmed 1x PBS, before refreshing 10 mL culture medium with or without CPA. Cells were cultured for 1 h at 37°C. 8 mL culture medium was collected in a 15 mL centrifuge tube and centrifuged at 100 × g for 3 min. Top 4 mL medium of the supernatant was collected for dot blotting analysis. The supernatant was stored in at −80°C and analyzed within 1 week.
Supernatants containing secreted proinsulin from the treated cells were transferred to PVDF membranes (Thermo Scientific) at increasing amounts (50, 100 and 150 μL) by using Minifold I microsample filtration manifold (Schleicher and Schuell). MIN6 cell lysates and fresh medium were used as positive and negative controls, respectively. The membrane was incubated in 1xTBST containing 5% skim milk for 30 min at 25°C with mild shaking. The membrane was washed twice with 1xTBST, followed by a standard Western blotting protocol.

Measurement of protein synthesis rate
Protein synthesis rates were measured as previously described 22 . In brief, cells were treated with CPA for the indicated times. At the end of treatments, [ 35 S]Met/Cys (30μCi/mL EXPRE 35 S Protein Labeling Mix (PerkinElmer)) was added to the cells for an additional 30 min. After labeling, cells were washed and lysed, and the radioactivity incorporated into proteins was determined by liquid scintillation counting. The protein synthesis rate was calculated as the rate of [ 35 S]Met/Cys incorporation to total cellular protein from the same lysate.
Measuring in vitro guanine nucleotide exchange factor (GEF) activity of eIF2B EIF2B activity was measured as previously described 22 . In brief, cells were washed and scraped off in homogenization buffer (45 mM HEPES-KOH pH 7.4, 0.375 mM MaOAc, 75 mM EDTA, 95 mM KOAc, 10% glycerol, 1 mM DTT, 2.5 mg/mL digitonin, supplemented with EDTA-free protease inhibitor (Roche Applied Science) and PhosSTOP phosphatase inhibitor (Roche Applied Science)). Cell lysates were homogenized and quantified for protein concentration. The EIF2B activity was calculated as the rate of exchange from eIF2α[ 3 H]GDP to nonradioactive GDP over the time points. Equal amount of protein lysates were immunoprecipitated with guinea pig polyclonal anti-insulin antibody and protein A agarose overnight at 4°C. Immunoprecipitates were washed and analyzed by reducing 4-12% NuPAGE Bis-Tris Gel SDS-PAGE, followed by phosphorimaging, and bands were quantified with ImageJ software.

Polysome profile and RT-qPCR
MIN6 cells were plated onto 15 cm plates and cultured in the cell growth medium until reaching to 60-70% confluency. After 0 h, 1 h or 18 h treatment with 200 μM CPA, cells were quickly washed with precooled 1x PBS twice, then lysed with RIPA buffer. The sample was layered on 10-60% sucrose gradients (20 mM Tris-HCl ph7.4, 150 mM NaCl, 10 mM MgCl2, 1 mM DTT). The tubes were centrifuged at 36,000 rpm (221632.5 × g) at 4°C for 3 h. The samples were fractionated into ten collection tubes. The first collection tube was discarded and total RNA of the rest of nine tubes was purified by Trizol (Invitrogen). Reverse transcription was performed with ProtoScript II (New England Biolabs) according to the company instructions. QPCR was done by using the StepOnePlus Real-Time PCR System (Applied Biosystems).

RNA preparation and qRT-PCR
Cells were washed twice with ice-cooled 1xPBS on ice after the designated treatments. Total RNA was extracted using TRIzol LS reagent (Invitrogen). The cDNA was synthesized using SuperScript III First-Strand Synthesis SuperMix (ThermoFisher). The relative quantity of specific mRNAs was measured by using VeriQuest SYBR Green qPCR Master Mix (ThermoFisher) with StepOnePlus Real-Time PCR System (Applied Biosystem). Primers used in this study are listed in SupplementaryTable 5.

Immunofluorescence staining
MIN6 and INS1E cells, were plated on glass microscope cover slips (ThermoFisher) in 6 cm culture dishes and were allowed to grow for 48 h. For Human EndoC-βH3 cells, cells were plated on coated (matrigel (1:500), Sigma-Aldrich and fibronectin (1 μg/cm 2 ), Sigma-Aldrich) glass microscope cover slips (ThermoFisher) in 6 cm culture dishes and allowed to grow for 1 week. After the designated treatments, cells were washed twice with ice-cooled 1xPBS on ice. Cells were fixed with 4% paraformaldehyde for 10 min. Fixed cells were washed twice with ice-cooled 1xPBS and incubated in PBST (1xPBS + 0.02% Triton X-100) for 15 min, PBST with 10% FBS for 30 mins, and PBST with 10% FBS and primary antibodies at 4°C for 16 h. After washing with icecooled PBST twice, cells were incubated in PBST with 10% FBS and secondary antibodies for 2 h in dark. This was followed by washing with ice-cooled PBST twice and nuclei staining with Hoechst 33342 for 5 min in dark. After washing with ice-cooled PBST twice, cells were mounted in Fluoromount-G (Electronic Microscope Sciences) and sealed with clear nail polish on microscope slides. The images were captured using a Leica SP8 confocal microscope. Imaging areas were randomly selected in a single-blind manner by a microscope specialist. Four imaging areas were fetched in each condition, while a representative image was shown.

GSIS in MIN6 cells
MIN6 cells were plated onto 6 cm plates for GSIS analysis. To optimize the measurement, 500,000 cells were plated and medium was refreshed every day. After two days recovery, cells were treated with 200 μM CPA or 0.2% DMSO in a fresh medium of the designated treating times. Cells were quickly washed with pre-warmed KRB buffer (120 mM NaCl, 20 mM HEPES, 5 mM KCl, 2 mM CaCl2, mM MgSO4, 0.5% BSA, pH7.4) twice. After washing, cells were incubated with 1 mL of KRB buffer with 2.8 mM glucose for 1 h at 37 o C. After incubation, the buffer was replaced with 1 mL of KRB with 2.8 mM glucose or 1 mL of KRB with 16.8 mM glucose for 1 h at 37°C. The supernatant was collected into a 1.7 mL tube, then centrifuged at 100 × g for 3 min. The upper 1 mL of the supernatant was collected for evaluation of GSIS analysis. The analysis was done by using Mouse Insulin ELISA kit (10-1247-01 Mercodia) according to the manufacturer's instruction.

Mouse pancreatic islet of Langerhans isolation and GSIS evaluation
Islets were isolated from C57BL/6 J (the Jackson Laboratory, stock no.000664) six-week-old male mice, for ex vivo immunofluorescence and GSIS studies. The mouse work was approved and maintained in the Animal Resource Center in Case Western Reserve School of Medicine with all relevant ethical regulations. The mice were maintained on a normal mouse diet and a 12 h/12 h light/dark cycle, with the temperature ranging between 70°F-75°F, the humidity was ranging between 30%-40%. The islet purification protocol has been previously described. Briefly, mice were anesthetized by Nembutal. The pancreas was removed, briefly washed with 1xPBS twice. The pancreas was digested with 1000 U/mL collagenase for 15 min with shaking (60 rpm). The sample was centrifuged at 290 × g for 30 seconds. The pallet was transferred to a 10 cm plate with pre-warmed islet medium (DMEM supplemented with10% FBS and 1% PSQ) and cultured at 37°C for 1 h. Single islets were collected and incubated with pre-warmed fresh islet medium and allowed to recover for 24 h before experimentation. For immunofluorescence, 10 islets were collected into a 1.7 mL tube and incubated with 200 μM CPA or 0.2% DMSO for the designated times. Islets were washed with precooled 1xPBS twice and were used for subsequent immunofluorescence experiments. For GSIS, 20 islets were collected into a 1.7 mL tube and incubated with 200 μM CPA or 0.2% DMSO for the designated times. After washing with pre-warmed 1xPBS, islets were incubated with 1 mL of KRB buffer with 2.8 mM glucose for 1 h at 37°C. After incubation, the buffer was replaced with 1 mL of KRB with 2.8 mM glucose or 1 mL of KRB with 16.8 mM glucose for 1 h at 37°C. The supernatant was collected into a 1.7 mL tube, and then centrifuged at 100 × g for 3 min. The upper 0.5 mL of supernatant was collected for GSIS analysis. The analysis was done by using the Mouse Insulin ELISA kit (10-1247-01 Mercodia) according to the manufacturer's instructions.

Quantification of MAFA fluorescence intensity in mouse islets
The raw, multiple Z-section confocal images were used for the measurement. To avoid a multiple calculation from the same cell, five z-sections from the upper layers of the islet images were stacked. The fluorescence intensities of MAFA staining and Hoechst 33324 nucleus staining were measured by ImageJ. A related MAFA expression level per cell was calculated by MAFA value/nucleus staining intensity of the same islet cell. The value of MAFA intensity smaller than 50 (8-bit intensity scale) was removed to avoid an inclusion of other cell types in the same islet in the statistical analysis.

Deep sequencing library generation
The general procedures for generation of mRNA-seq and Ribo-seq libraries were previously reported by us and others. In brief, cells were washed twice by ice-cooled PBS with 100 μg/ml cycloheximide (Sigma-Aldrich) before harvest. Lysis buffer (10 mM Tris, pH 7.4, 100 mM NaCl, 140 mM KCl, 10 mM MgCl 2 , 1 mM DTT, 1% (v/v) Triton X-100, 500 U/mL RNasin and 100 μg/mL cycloheximide) was added and cells were scraped fast on ice. The lysate was centrifuged for 5 min in 500 × g at 4°C. The supernatant was collected and centrifuged for 10 minutes in 10,000 × g at 4°C. The supernatant was collected for mRNA and monosome enrichments. For mRNA enrichment, total RNA was extracted from lysates with TRIzol LS reagent (Invitrogen). The RNA with poly adenosine tract was purified by a Magnetic mRNA Isolation Kit (New England Biolabs) according to the user manual. The RNA with poly adenosine tract was fragmented with an alkaline buffer (2 mM EDTA, 10

Islet microdissection and microarray analysis
The existing microarray raw data was utilized in this study; the preparation was described previously 30,31 . In brief, frozen tissue was obtained from the Network for Pancreatic Organ donors with Diabetes (nPOD) (Supplementary Data 4) 87 . Optimal cutting temperature slides of pancreatic tissue were used for laser-capture microscopy that was conducted on an Arcturus Pixcell II laser capture microdissection system (Arcturus Bioscience). All islets visible in two to five sections from each sample were pooled, and RNA was extracted using the Arcturus PicoPure RNA Isolation Kit (Applied Biosystems). RNA quantity and quality were determined using a Bioanalyzer 2100 (Agilent Technologies). RNA samples were subjected to gene expression analysis using Affymetrix expression arrays (ThermoFisher). Fluorescence intensity was used to refer to gene expression. Data was aggregated to minimize the identifiable information before analysis.

Single-cell RNA sequencing analysis
The scRNA-seq raw data from T1D and healthy donors datasets were obtained from HPAP (https://hpap.pmacs.upenn.edu/) 61 . The raw data were pooled and normalization using SCtransform 88 in the Seurat package in R. No additional batch-effect removal step was applied. The normalized dataset contains gene expression profiles of the 35041 islet cells, along with 23509 gene expression levels. Cell types were determined by Seurat clustering results according to the expression of pancreatic marker genes 89 . In brief, we performed a non-supervised Uniform Manifold Approximation and Projection (UMAP) 90 of the dataset to cluster cells with similar expression profiles. Totally 8349 cells, which contain β-cell identities, were defined as β-cell for the study. For the UPR study, the cell profiling was done by a machine learning approach via the Rtsne package in R according to the expression of the UPR signature (GO:0030968) and 6 β-cell identity genes, INS, PAX6, NKX2.2, NKX6.1, MAFA and PDX1. Patients' information used in this study was listed in Supplementary Fig. 4b. in silico pseudo-time reconstruction of scRNA-seq The pseudo-time reconstruction of scRNA-seq was done by Monocle 3 67 and TSCAN 66 packages in R in a default setting. In brief, a dataset contained the expression information of the UPR signature (GO:0030968) and 6 β-cell identity genes of the 8349 cells. Data were preprocessed to reduce dimensionality, via UMAP for single-cell trajectory. The expressions of the average of β-cell identity genes and BiP along the trajectory were fetched for the analysis.

Cholesterol synthesis measured by GC-MS
Cholesterol synthesis was determined following deuterium labeling incorporation from deuterium-enriched media. In brief, MIN6 cells were incubated in the medium containing 10% deuterated water ( 2 H 2 O, molar percent enrichment) with CPA for 18 h (CPA18); while without CPA served as control. For the recovery sample, after 18 h of CPA treatment, the cells were washed twice with 1xPBS and refreshed with the medium containing 10% deuterated water for 24 h incubation. After incubation, the supernatant was collected for determination of deuterium enrichment in the medium. After 1xPBS wash twice, cells were trypsinized and pelleted by centrifugation at 4°C at 650 × g for 5 min. Cell pellets were washed with 1xPBS and pelleted by centrifugation at 4°C at 650 × g for 5 min. Supernatants were discarded and pre-cooled methanol was added into the tubes containing cell pellets. After mixing by inversion for 5 times, cells were pelleted by centrifugation at 4°C at 650 × g for 5 min. The supernatant was discarded and pellets were stored at −80°C until extraction.
The 2 H-labeling of body water was determined by the exchange with acetone, as previously described by 91 and further modified. In brief, 5 µL of cellular medium or standard were reacted for 4 h at room temperature with 5 µL of 10 N KOH and 5 µL of acetone. Headspace samples were injected into GC/MS in split mode (1:40). Acetone m/z 58 and 59 were monitored. Isotopic enrichment was determined as ratio of m/z 59/(58 + 59) and corrected using a standard curve.
For cholesterol extraction, cell pellets in the tubes were homogenized frozen in 600 µl of Folch solution (chloroform:methanol, 2:1, v/v) on dry ice. After addition of 0.4 volumes of ice-cold water, cells were homogenized again and let incubate on ice for 30 min. Homogenates were centrifuged at 4°C at 20,000 × g for 10 min. Upper methanol/water layer was discarded. Internal standard (heptadecanoic acid) was added to the bottom chloroform layer and was evaporated to dryness. Cholesterol was then converted to TMS (trimethylsilyl) derivative by reacting with bis(trimethylsilyl) trifluoroacetamide with 10% trimethylchlorosilane (Regisil) at 60°C for 20 min. Resulting TMS derivatives were run in GC-MS.
GC-MS conditions used for the analysis were carried out on an Agilent 5973 mass spectrometer equipped with 6890 Gas Chromatograph. A DB17-MS capillary column (30 m × 0.25 mm × 0.25 μm) was used in all assays with a helium flow of 1 mL/min. Samples were analyzed in Selected Ion Monitoring (SIM) mode using electron impact ionization (EI). Ion dwell time was set to 10 msecs. Cholesterol m/z 368-372 were monitored to detect isotopic label incorporation.
The absolute cholesterol biosynthesis was computed as follows. Cholesterol GC-MS data were corrected for natural abundance and resulting enrichments were divided into precursor isotopic enrichment of media deuterium (10%) to follow precursor-product relationship. This resulted in fractional synthesis rate. Absolute cholesterol synthesis rate was determined by multiplying fractional rate by the cholesterol content which was determined using internal standards.

Biostatistics and bioinformatics
Mouse RNA reference (GRCm38.p6.rna) was used for sequencing reads alignment. The alignment was done by Bowtie2 92 in an in-houseestablished Galaxy platform. In brief, raw sequencing reads were filtered by quality, Q30 > 90%. The length of reads between 15 and 40 were selected. The selected reads were aligned to the RNA reference with parameters of equal or higher than 95% identity and less than one mismatch was allowed. Genes with at least one read aligned and identified in all three independent experiments were collected for further analysis. Transcript count was calculated as reads per kilo-base per million reads of total aligned reads (RPKM). Ribosome occupancy (Ribo ocp ) of a transcript was calculated as the mean of transcript count in Ribo-seq divided by the mean of transcript count in mRNA-seq. Gene functional annotations were done in KEGG: Kyoto Encyclopedia of Genes and Genomes (https://www.genome.jp/kegg/) and Gene Ontology Resource (http://geneontology.org/). Data analysis, statistics and presentations were done by R (https://www.r-project.org/) and Excel (Microsoft). Bar charts were done by Excel. Heatmaps were done by Heatmap2 package in R. Scatterplots, volcano plots and violin plots were done by the Ggplot2 package in R.
The statistics of the heatmaps in Fig. 3 was done by Excel. For  Fig. 3c, the changes of transcript expression in [log2] values of CPA1 over control (acute), CPA18 over CPA1 (chronic) and CPA18 over control (overall) were calculated. The cutoffs to define whether the transcript was upregulated or downregulated were set as 0.32 or −0.32, respectively. Transcripts that showed upregulation in both acute and chronic phases were grouped into G1. Transcripts that showed upregulation in the acute phase but downregulation in the chronic phase were grouped into G2. Transcripts that showed no change in the acute phase but upregulation in the chronic phase but overall upregulation were grouped into G3. Transcripts that showed downregulation in both, acute and chronic phases were grouped into G4. Transcripts that showed downregulation in the acute phase but upregulation in the chronic phase were grouped into G5. Transcripts that showed no change in the acute phase but downregulation in the chronic phase but overall downregulation were grouped into G6. For Fig. 3d, transcripts were grouped separately according to the Ribo ocp values. Transcripts that showed increased Ribo ocp in both acute and chronic phases were grouped into G7. Transcripts that showed increased Ribo ocp in the acute but decreased Ribo ocp in the chronic phase were grouped into G8. Transcripts that showed no change of Ribo ocp in the acute phase and an increase in the chronic phase, leading to overall increased Ribo ocp were grouped into G9. Transcripts that showed no change of Ribo ocp in the acute phase and a decrease in the chronic phase, leading to overall to decreased Ribo ocp were grouped into G10. Transcripts that showed decreased Ribo ocp in the acute, but increased Ribo ocp i in the chronic phase were grouped into G11. Transcripts that showed decreased Ribo ocp in both acute and chronic phases were grouped into G12. The functional enrichments of the 12 groups were analyzed by the KEGG pathway via the DAVID bioinformatics database. The number of genes identified in a single biological pathway was denoted.
For the transcriptome and translatome analysis the gene expression levels represent the means of RPKM of three independent experiments. The values of differential expression were determined as log2(expression in RPKM in condition 2expression in RPKM in condition 1) (L2DE). The cuff-off value for data filtering and statistics was set as log2 = 0.6 or −0.6, which represented at least 50% difference in RPKMs between two conditions. The definition of expression regulation was set as described previously (Guan et al. 2017). In brief, upregulation in both mRNA abundance and ribo ocp , −0.6≦(x-y)<0.6, x≧0.6, y≧0.6; up-regulation in ribo ocp only, (x-y)<−0.6, y > 0.6; downregulation in both mRNA abundance and ribo ocp , −0.6≦(x-y)<0.6, x≦−0.6, y≦−0.6; down-regulation in ribo ocp only, (x-y)>0.6, y < −0.6, where x represents the L2DE in transcriptomic data and y represents the L2DE in ribosome footprints.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The RNA-seq and Ribo-seq sequencing data of the CPA treated MIN6 cells and the control generated in this study are available in the Gene Expression Omnibus (GEO) database under accession code GSE174679. All data and resources supporting the findings described in this manuscript are available in the article and in the Supplementary Information and from the corresponding author upon reasonable request. Source data are provided with this paper.