The EndoC-βH1 cell line present in this study was obtained from Human Cell Design. EndoC-βH1 cells were cultured as previously described (36). Briefly, cells were cultured on plates coated with matrigel (100 μg/mL, Corning, 356237) plus fibronectin (2 μg/mL), in DMEM (5.6 mM glucose) supplemented with 2% BSA, nicotinamide (10 mM), transferrin (5.5 μg/mL), sodium selenite (6.7 ng/mL), 2-mercaptoethanol (50 μM), and penicillin/streptomycin (100 U/mL, 100 μg/mL respectively). Passages were performed weekly, and cells were tested periodically for mycoplasma contamination (Applied Biological Materials, G238).

To generate the ATG7^Δβ-cell C57BL6/J mice, we employed a Cre-Lox system where constitutive Ins1^Cre mice were crossed with Atg7^fl/fl mice. Atg7^fl/fl mice (B6.Cg-Atg7<tm1Tchi>) were obtained with permission from the donating investigator from RIKEN (RRID: IMSR_RBRC02759). The Ins1^Cre mice (B6(Cg)-Ins1tm1.1(Cre)Thor/J) were obtained from Jackson Labs (RRID: IMSR_JAX:026801).

Random fed blood glucose levels were assessed by tail prick twice weekly starting at 6 weeks of age using an AlphaTrak2 glucometer (Zoetis). For Kaplan-Meier curves, diabetes was defined as consecutive readings of blood glucose >250 mg/dL (>13.9 mmol/L) on separate days. At 8 weeks of age, a glucose tolerance test was performed following an overnight fast. Each mouse was injected intraperitoneally with 2 mg glucose per gram body weight and blood glucose was measured just prior to injection, and at indicated time points up to 2 hours following glucose injections. For experiments utilizing isolated islets, ATG7^Δβ-cell and littermate control mice were sacrificed via cervical dislocation, and islets were harvested by the Indiana University School of Medicine (IUSM) Islet and Physiology Core as previously described (37). Mice were similarly sacrificed and pancreata harvested for immunofluorescence analysis. For mouse tissue harvest, islet isolations and pancreata harvest were performed several weeks prior to diabetes development (7-10 weeks of age) from ATG7^Δβ-cell and littermate control mice. Experimental mice were housed in a temperature-controlled facility with a 12-hour light, 12-hour dark cycle and had free access to food and water. All experiments were approved by the IUSM Institutional Animal Care and Use Committee.

Following islet isolation, islets were washed and incubated in lysis buffer containing 50 mM Tris-HCl (pH 8.0), 1 mM EDTA (pH 8.0), 150 mM NaCl, 5% glycerol, 0.05% sodium deoxycholate, 0.2% sarkosyl, 10 mM NaF, 0.1% IGEPAL, 0.1% SDS, 1mM DTT, 2 mM MgCl₂, 0.05% w/v benzonase, and 1x Protease Inhibitor Cocktail (Fisher Scientific, NC0818307). Lysis incubation occurred for 20 minutes on ice and samples were vortexed every 5 minutes to aid lysis. Samples were then passed through a 22-gauge needle 10x and stored at -80 °C until blotting was performed. Sample protein concentration was determined using Pierce™ Rapid Gold BCA Protein Assay Kit (Thermo Fisher Scientific, A53226) per manufacturer’s guidelines. Samples were mixed with 4x loading buffer (LI-COR, D00908-01) with 1:10 β-mercaptoethanol (50 μL total sample volume), boiled at 95 °C for 10 minutes, and run on pre-cast gels (BioRad, 4561094). Protein was then transferred to activated PVDF membranes for analysis using anti-ATG7 (Abcam, ab133528, 1:1,000) and anti-β-actin (CST, 3700S, 1:1,000) antibodies at 4 °C overnight followed by incubation with donkey anti-mouse IRDye 680RD (LI-COR Biosciences, 926-68072, 1:10,000) and donkey anti-rabbit IRDye 800CW (LI-COR Biosciences, 926-32213, 1:10,000) for 1 hour at room temperature. Blots were imaged on a LI-COR Odyssey Imager.

Following islet isolation, islets were rested overnight prior to RNA extraction using RNAeasy Plus Mini Kit (Qiagen, 74134) per manufacturer’s guidelines. Quality of RNA was assessed at the IUSM Center for Medical Genomics using Agilent 2100 Bioanalyzer. Dual-indexed cDNA libraries were prepped using SMART-seq ultra-low input RNA kit (Takara Bio) and sequenced using NovaSeq6000 (Illumina). The sequenced libraries were mapped to mouse genome UCSC mm10 and differential expression was calculated at the IUSM Center for Medical Genomics and Bioinformatics core. ATG7^Δβ-cell mRNA transcript levels were plotted as fold change (FC) compared to control levels. Genes with a FC greater than 1.5 or less than -1.5 were considered up or downregulated respectively. A p-value threshold of ≤0.05 was also used for the selection of both up and downregulated genes. Differentially regulated genes were subjected to Gene Ontology and functional analysis using DAVID (https://davidbioinformatics.nih.gov/), a free online functional annotation tool (38, 39). The plots were generated using GraphPad Prism.

In addition, protein was isolated from ATG7^Δβ-cell and littermate control islets and prepared for mass spectrometry analysis. Islets were washed in PBS, spun down, decanted, pellet flash frozen in liquid nitrogen, and samples submitted to the IUSM Center for Proteome Analysis for further processing. Briefly, samples were resuspended in 8 M urea, reduced in 100 mM Tris (pH 8.5), alkylated, enzymatically digested (Promega trypsin/LysC mix, 1:70 ratio), and desalted (Waters SepPak). Samples were subsequently TMT labeled, mixed, dried, and desalted. The prepared sample was then loaded and separated on an ultra-high performance liquid chromatographer (Easynano LC1200) using a C18 Aurora 25cm nanocolumn coupled to an Eclipse Orbitrap mass spectrometer (Thermo Fisher). Following spectra generation, peptides were identified using Proteome Discoverer 2.5 searching the Mus musculus protein database, with common contaminants included. Proteins with a FC greater than 1 or less than -1 were considered up or downregulated respectively. A p-value threshold of ≤0.05 was also used for the selection of both up and downregulated proteins. Gene Ontology and functional analysis was performed using DAVID and plotted by GraphPad Prism.

Following tissue harvest, pancreata were embedded in optimal cutting temperature compound and frozen on dry ice prior to slicing tissue sections for immunofluorescence staining. Frozen 8 µm pancreatic sections were cut at -20°C (Leica CM1860 Cryostat) and adhered to poly-L-lysine coated slides. Tissue sections were then fixed in acetone for 10 min at RT, washed in PBS, incubated in protein blocking solution (Abcam, AB64226), followed by incubation in antibody diluent (Abcam, ab64211) with 40 μg/mL AffiniPure Fab Fragment (Jackson ImmunoResearch, 115-007-003) at 4 °C overnight. The next day slides were washed in PBS, PBS 0.1% Tween 20, then PBS prior to overnight incubation at 4 °C in antibody diluent containing primary antibodies at the following dilutions: anti-insulin 1:2 (Dako, a0564), anti-MHC-I 1:150 (Biolegend, 114602), and anti-CD45 1:50 (BD Biosciences, 550539). Slides were similarly washed as before prior to incubation in the dark at room temperature for 1 hour in antibody diluent containing secondary antibodies and nuclear stain at the following dilutions: Alexa Fluor 488 goat anti-guinea pig 1:500 (Invitrogen, A11073), Alexa Fluor 594 goat anti-mouse 1:500 (Invitrogen, A11032), Alexa Fluor 647 goat anti-rat 1:500 (Invitrogen, A21247), and DAPI 1:2,000. Slides were then washed and mounted with EverBrite Mounting Medium (Biotium, 23001) and imaged on a Leica Stellaris 5 confocal microscope using a 63x/1.40 oil objective.

Image analysis was then performed using ImageJ in a semi-automated fashion. Briefly, insulin positive area was used to determine a region of interest (ROI) wherein mean islet fluorescent intensity for MHC-I was measured. Within this ROI, CD45+ nuclei were also counted, as well as total nuclei, to determine the percentage of CD45+ cells within each islet. Background fluorescence was subtracted using the mean fluorescent intensity value of islets from serial sections treated with 2^nd antibodies only as a control. Data was plotted using GraphPad Prism.

EndoC-βH1 cells were treated overnight with 100 nM Bafilomycin A1 (Millipore Sigma, 508409), 100 μM C381 (MedChemExpress, HY-100347A), or vehicle control, DMSO. Subsequent experiments included pre-treatment for 24 hours with 2,000 IU/mL recombinant IFNα 2 protein (Novus Biologicals, NBP2-26551) followed by washout with PBS, replacement with EndoC-βH1 culture media, culture for 72 hours prior to overnight treatment with 100 nM Bafilomycin A1, 100 μM C381, or DMSO. Following overnight treatment, cells were washed, removed from the culture plate using 0.05% trypsin, and spun down at 500x g for 5 minutes. Cells were resuspended and washed in FACS buffer containing 1% bovine serum albumin in 1x dPBS, 0.1% sodium azide, and 0.02% w/v EDTA. Cells were then incubated in FACS buffer containing HLA-ABC Class I primary conjugated Alexa Fluor 488 (clone: W6/32) for 30 minutes at 4 °C (BD Biosciences 567851), washed, and fixed in 1% paraformaldehyde in PBS on ice. Cells were subsequently washed, resuspended in FACS buffer, and analyzed using an Attune Nxt acoustic flow cytometer.

Spleens were harvested from 8-week-old NOD.Cg-Tg(TcraBDC2.5, Strain-004460) homozygous mice, and splenocytes were isolated as previously described (40). Briefly, spleens from BDC2.5 mice were processed by mashing through 70 µm strainer (Corning, CLS352350) and then treated with red blood cell lysis buffer (Sigma Aldrich, R7757) for 2 minutes, followed by washing twice with complete DMEM media. For in vitro experiments, 500,000 splenocytes cells were plated per well in a 96 round bottom plate (Corning, 3799) and cultured in complete DMEM media. Concurrently, islets were isolated from ATG7^Δβ-cell and Cre+ control mice. Islets from 3-6 ATG7^Δβ-cell and 3-6 Cre+ control mice were pooled, and half the control islets were treated with 100 nM Bafilomycin A1 overnight prior to co-culture. To set up the co-culture, islets were spun down, washed with PBS, resuspended in Accutase (Fisher Scientific, NC9839010), and incubated in a water bath for 10-15 minutes at 37 °C. Islets were pipetted vigorously every few minutes and an aliquot was checked under a microscope to confirm islet dissociation into single cells. Islet cells were then counted using a hemocytometer to add the same amount of islet cells per condition (3,600 and 1,800 cells) to splenocytes previously plated. As a positive control, BDC2.5 splenocyte cells were stimulated with 0.05 µM of BDC2.5 peptide (2.5HIP: RGG-LQTLALWSRMD-GGR) and 2.5 µg/ml Concanavalin A (Sigma, C2272) respectively. Each condition was performed in triplicate. At 24, 48, and 72 hours, samples were spun down and the supernatant was collected to measure IFNγ secretion by ELISA.

Culture supernatant from different time points were measured for mouse IFNγ cytokine level. 96 well high affinity flat bottom plate (Corning, 3361) were coated with 25 ng/mL monoclonal rat anti-murine IFNγ (BD Pharmaceutical, 551216) by diluting in carbonate buffer (pH 9.6) overnight at 4 °C. For detection, biotin rat anti-mouse IFNγ (Clone XMG1-2, BD Pharmingen, 554410) and Streptavidin-HRP (Invitrogen, 434323) were used according to manufacturer’s protocol. The plate was read on Spectramax M5 microplate reader (Molecular Devices) at OD 450 nm and data analyzed using SoftMax Pro version 7.0.2. Data summarizes results from two independent experiments. Each condition was performed in triplicate and 9 Cre+ control and 9 ATG7^Δβ-cell mice were used between experiments.

Statistical analyses were performed using GraphPad Prism v10.2. Pairwise comparisons between two groups were conducted using an unpaired two-tailed Student’s t-test. For experiments where HLA-I expression was normalized to control levels for each replicate, a one sample t-test was performed, with a hypothetical population mean equal to 1. One-way ANOVA was used to compare the means from multiple treatment groups, with Dunnett’s post hoc test adjusted for multiple comparisons. Histogram plots are represented as mean or median ± standard deviation, and dot plots are represented as mean ± standard error of the mean. A p-value <0.05 was considered statistically significant.

To better understand the role of autophagy in β-cell function and survival, we selectively knocked out the critical autophagy enzyme, autophagy related gene 7 (ATG7), in the β-cells of mice (ATG7^Δβ-cell mice, see Figures 1A, B ). ATG7 knockout in a variety of target tissues has elucidated the critical role of autophagy in various cell types (41). The ATG7 enzyme is essential for the expansion and maturation of autophagosomes, thus rendering ATG7 knockout cells severely autophagy deficient (42). In humans, recessive variants causing severe loss of the ATG7 enzyme, lead to neurodevelopmental disorders affecting the brain and muscle, with more subtle evidence of endocrine hypofunction also present in these individuals (42). Previous mouse models of β-cell autophagy deficiency have shown evidence of impaired glucose homeostasis and glucose stimulated insulin secretion, as well as reduced insulin content and β-cell mass (43, 44). Here, we conducted bi-weekly random fed blood glucose measurements, confirming previous findings of early onset progressive hyperglycemia leading to overt diabetes development in ATG7^Δβ-cell mice ( Figures 1C, D ) (44). There was also evidence of impaired glucose tolerance as early as 8 weeks in male mice, but not yet in females, as indicated by glucose tolerance challenge ( Supplementary Figures 1A, B ). However, we found that females develop similar levels of hyperglycemia and diabetes with a delayed onset compared to males, which was not previously appreciated ( Figures 1C, D ). All ATG7^Δβ-cell mice developed diabetes, with male ATG7^Δβ-cell mice having a median age of 11 weeks at diabetes onset and female ATG7^Δβ-cell mice a median age of 15 weeks at diabetes onset ( Figure 1D ).

Next, we aimed to understand the islet biological process pathways altered by dysfunctional β-cell autophagy in the ATG7^Δβ-cell model. To do this, we evaluated bulk islet mRNA and protein levels by RNA-sequencing and mass spectrometry, respectively. Islets were isolated several weeks prior to diabetes development to assess pathways that are altered leading to overt disease. At the transcript level, we observed an upregulation of transcripts involved in extracellular matrix and collagen fibril organization pathways ( Figure 2A ). This includes upregulation of metalloproteinases, which are involved in extracellular matrix remodeling. This remodeling could be indicative of, and necessary for, immune cell recruitment in ATG7^Δβ-cell islets, as has been observed in T1D pathology (45, 46). The upregulation of these pathways could also indicate a fibrotic response that has been observed with loss of ATG7 in other tissues (47).

In addition, we found a significant upregulation of transcripts involved in the inflammatory response and immune system processes ( Figures 2A, C ). Upregulation of these pathways suggested potential enhanced immune cell recruitment to ATG7^Δβ-cell islets, as both upregulation of chemotactic factors and receptors was observed ( Figure 2C ). We also observed upregulation of transcripts involved in tumor necrosis factor (TNF) and interleukin-6 (IL-6) production, cytokines that have been linked to early events in T1D development including immune cell recruitment and activation ( Figure 2A ) (48–50). This suggests that in the absence of a hyperactive immune system, due to absence of NOD genetic risk factors on the C57BL6/J background, defective autophagy and concurrent ER stress might be sufficient to promote key features of inflammation in diabetes development. This is supported by a recent study that showed that pharmacological inhibition of the maladaptive ER stress response can delay onset of autoimmune diabetes in NOD mice (51).

We also identified genes in several key pathways that were downregulated at the transcript level, including those associated with cholesterol and lipid biosynthesis and metabolism ( Figures 2B, D ). Evidence from the literature suggests ER stress negatively regulates cholesterol export and metabolism, which could explain the downregulation of transcripts involved in these pathways (52, 53). There was also downregulation of the Regenerating (Reg) family of genes that are associated with β-cell proliferation, protection from apoptosis, and protection from cytokine induced damage ( Figure 2D ) (54–56). This suggests that these processes are potentially impaired in ATG7^Δβ-cell islets.

It is widely observed that transcript changes do not always directly correlate to protein changes, as was recently shown very broadly in human islets from numerous donors (57). At the protein level, we saw an upregulation of proteins involved in response to ER stress, with a concomitant increase in chaperone proteins involved in protein folding as well as proteins involved in the ER associated degradation (ERAD) pathway ( Figures 3A, C ). Downregulated proteins were associated with pathways involved in protein translation, mRNA splicing, and mRNA processing, which are consistent with the ER stress induced by ATG7^Δβ-cell knockout ( Figures 3B, D ). Several of the most strongly upregulated proteins were unsurprisingly associated with the autophagy pathway that are accumulating under conditions of impaired autophagic flux. This includes receptors that target proteins for autophagic degradation and aid autophagic vesicle maturation. Upregulation of compensatory protein degradation pathways to deal with accumulating cellular waste, typically targeted for autophagic degradation, included enhanced proteasome-mediated proteolysis and protein secretory pathways ( Figure 3A ) (57). We also found several of the upregulated ER stress response proteins are also involved in antigen production and MHC-I complex assembly. This included chaperones that stabilize the complex before peptide loading (ERp57/PDIA3, CNX), help facilitate peptide binding as well as recruit other members of the complex (ERp57/PDIA3, CRT), and transport it to the plasma membrane (BCAP31) (58). Many subunits of the proteasome, which processes proteins for MHC-I presentation, are also upregulated. A shunting of proteins typically targeted for autophagic degradation to alternate protein degradation pathways could have broad implications for the antigen repertoire presented by the β-cell to surveilling immune cells.

Next, we looked at the intersection of genes differentially regulated at both the transcript and protein level. Because of the limited number of genes identified from this intersection, the pathway analysis was not separated into upregulated and downregulated genes. Many of the pathways previously discussed appeared in our analysis, including response to ER stress, protein folding, and ERAD pathways ( Supplementary Figure 2A ). To determine whether these pathways were primarily up or down regulated, we plotted gene transcript fold change (FC) vs. protein FC ( Supplementary Figure 2B ). We observed that genes associated with the response to ER stress and protein folding pathways were almost exclusively downregulated at the transcript level and upregulated at the protein level when autophagy is defective ( Supplementary Figure 2B ). This suggests that reduced degradation of these proteins, and subsequent feedback inhibition on transcription, limits further accumulation of proteins in these pathways. Similarly, receptors that mediate selective autophagic degradation of the ER during ER stress, such as CCPG1, are upregulated at both the transcript and protein level (59). Finally, DMBT1 and ZIP8—among the most strongly downregulated factors—are specifically linked to immune defense and negatively regulate proinflammatory response (60, 61).

Due to alterations of proteins involved in antigen production and MHC-I complex assembly, we next asked whether MHC-I expression was altered in ATG7^Δβ-cell islets. Therefore, fresh frozen pancreas tissue sections were assessed for MHC-I islet expression. We observed a significant increase in mean MHC-I intensity in ATG7^Δβ-cell islet area compared to control islets (p = 0.0002, Figures 4A, B ). This result is consistent with other models of dysfunctional autophagy in immune cells that also demonstrated increased MHC-I expression (62, 63). In humans, there is also evidence of enhanced HLA Class I (HLA-I, human homolog of MHC-I) expression in individuals who harbor recessive variants causing severe loss of the ATG7 enzyme (42). Therefore, impaired autophagy could be a mechanism that contributes to HLA-I islet hyperexpression in T1D, a well-known feature of disease development (64–68).

In addition to potentially enhancing β-cell visibility to immune cells via MHC-I upregulation, we assessed whether the ATG7^Δβ-cell islets showed altered immune cell presence within the islet area. We immunostained for CD45+ cells in the pancreata of ATG7^Δβ-cell and control islets and observed a significant increase in the number of CD45+ immune cells in and around ATG7^Δβ-cell islets compared to littermate controls (p < 0.001, Figures 4A, C ). This discovery coincides with our transcriptomic data showing upregulated inflammatory response and cytokine production in ATG7^Δβ-cell islets.

To determine if our ATG7^Δβ-cell mouse model observations would replicate in human cells, we also assessed whether pharmacological modulation of autophagy in EndoC-βH1 cells would alter HLA-I expression. EndoC-βH1 cells were treated overnight with the lysosome acidification inhibitor Bafilomycin A1 (Baf A1), to inhibit the last stage of autophagic degradation, or vehicle control. Alternatively, cells were treated with a novel small molecule, C381, known to stimulate lysosome acidification/function (formerly known as SRI-011381) (69). After treatment, surface HLA-I expression was analyzed using flow cytometry. We found that treatment with the inhibitor Baf A1 enhanced human β-cell HLA-I expression, whereas stimulating lysosome function with C381 treatment had the opposite effect and reduced HLA-I expression (p = 0.0012 and p = 0.0094 respectively, Figures 5A, B ). Previously, our lab demonstrated evidence for impaired lysosome function in β-cells both in the setting of T1D and in autoantibody positive individuals, who have not yet developed T1D (18). Thus, impaired β-cell autophagy could be a mechanism that contributes to the HLA-I hyperexpression that is thought to be induced primarily by proinflammatory cytokines (64, 70).

Following this result, we wanted to determine whether autophagy modulation following IFNα induced HLA-I hyperexpression would alter β-cell HLA-I surface expression. To test this, EndoC-βH1 cells were treated for 24 hours with IFNα, the treatment was washed out, and then followed with overnight treatment with the autophagy inhibitor Baf A1, stimulator C381, or vehicle control. The washout period was determined by when the literature indicates β-cell HLA-I expression induced by IFNα treatment should begin declining to return to baseline levels (71). Following treatment, surface HLA-I expression was again assessed by flow cytometry. Our results demonstrate that Baf A1 treatment maintains the hyperexpression of HLA-I induced by IFNα above vehicle treatment following washout, whereas stimulating lysosome function by C381 treatment accelerates the disappearance of HLA-I following IFNα treatment (p = 0.0023 and p = 0.0034 respectively, Figures 5C, D ). This result, in combination with observations from the literature, suggests that impaired autophagy could exacerbate the effects of IFNα on β-cells tipping the balance from a protective to a maladaptive interferon response (29, 70).

Taken together, our results suggested that impaired β-cell autophagy may make β-cells more immunogenic and susceptible to immune attack and destruction. Impaired autophagy could increase β-cell visibility to CD8+ T cells through enhanced HLA-I expression, and may also allow for the buildup and/or release of misfolded and improperly modified peptides, including neoantigens. These peptides can then be taken up by antigen presenting cells for presentation on HLA-I and HLA-II complexes to prime effector cells in T1D pathogenesis. To begin testing this hypothesis, we co-cultured dispersed ATG7^Δβ-cell or wild type islet cells, pretreated with Baf A1 or vehicle control, with splenocytes from BDC2.5 T cell receptor transgenic mice that contain diabetogenic CD4+ T cells. Following 24-72 hours of co-culture, T cell stimulation was assessed by collecting the cell supernatant and measuring Interferon-γ (IFNγ) production by ELISA. We found that in the presence of dispersed cells from ATG7^Δβ-cell islets, there is sustained elevation of IFNγ production at 72 hours compared to littermate control islet cells (p = 0.0014, Figures 6A, B ). Furthermore, we found that in the presence of wild type islet cells pretreated with Baf A1, there is a similar sustained elevation of IFNγ production at 72 hours compared to vehicle treated control islet cells (p = 0.0373, Figures 6A, B ). We also obtained strikingly similar results when we performed the experiment with a dilution of islet cells ( Supplementary Figures 3A, B ). The IFNγ levels induced by islet cells with defective autophagy were comparable to levels induced by a positive control treated with Concanavalin A, a potent control of polyclonal activation of all T cells. This suggests that dysfunctional β-cell autophagy could make β-cells more immunogenic and susceptible to immune attack and destruction, as indicated by the enhanced BDC2.5 T cell activation. One could hypothesize that the delay in autophagic clearance of damaged or mis-folded proteins may lead to their accumulation and subsequent recognition by the β-cell reactive T cells, similar to previous observations in the NOD model (72).