A mitochondrial ROS–dependent antiviral response promotes β cell resilience and is diminished in donors with type 1 diabetes

Abstract

Type 1 diabetes (T1D) is a multifactorial disease driven by genetic and environmental factors, including, potentially, viral infection. However, the mechanisms linking infection-associated cytokines to human β cell loss are poorly understood. Here, we coupled in vivo and in vitro imaging with genetic analysis to investigate the impact of interferon α (IFN-α), a cytokine produced during the immune response to viral infection or detection of unedited endogenous double-stranded RNAs, on human β cell physiology. We identified a subset of human β cells that acutely produce reactive oxygen species (ROS) in response to IFN-α and were more prevalent in islets from donors with lower body mass index and HbA1c. RNA sequencing of flow-sorted ROS⁺ and ROS⁻ populations identified a gene signature predisposing some cells to IFN-α–stimulated ROS production, including genes involved in inflammatory and immune response. IFN-α treatment of human islets in vitro similarly elicited a heterogeneous increase in superoxide production. Parallel analysis of a human β cell line demonstrated that this ROS originated in the mitochondria. Rapid stimulation of key antiviral response genes by IFN-α in human islets was dependent on mitochondrial ROS elevation. Comparison with single-cell RNA sequencing datasets showed that genes up-regulated in ROS-producing cells were enriched in β cells from nondiabetic and autoantibody-positive donors rather than donors with T1D. Overall, our data demonstrate that IFN-α–induced mitochondrial ROS production in healthy human β cells is critical for the acute antiviral response, and impairment of this heterogeneous adaptive response may predict β cell loss during T1D pathogenesis.

Editor’s summary

Despite being an autoimmune disease, type 1 diabetes (T1D) is not solely driven by the adaptive immune response. Rather, insulin-producing # cells themselves can mount inflammatory responses and respond to stressors. Here, Wagner et al. demonstrated that a subset of # cells from healthy human donors respond to the antiviral cytokine interferon-# (IFN-#) by producing mitochondrial reactive oxygen species (ROS), which supports activation of antiviral response pathways. This ROS-dependent response was absent in # cells from individuals with T1D. Together, these data identify a # cell-intrinsic inflammatory response that may contribute to # cell resilience and is lost in the context of autoimmune disease. –Courtney Malo

INTRODUCTION

Type 1 diabetes (T1D) has long been described as a multifactorial disease, consisting of both a genetic predisposition and environmental triggering event. A leading hypothesis for this environmental trigger is early childhood viral infection leading to viral mimicry and spread of T cell immunity to β cell antigens. Interferon-α (IFN-α), a type I IFN, is a pluripotent inflammatory cytokine produced by cells involved in the innate immune response during viral infection or in response to unedited endogenous double-stranded RNAs. IFN-α and other type I IFNs likely play a role in T1D pathophysiology. For example, children with a genetic predisposition for T1D development have a type I IFN–inducible transcriptional signature in their serum preceding the detection of autoantibodies (1, 2). Likewise, islets isolated from newly diagnosed donors have been shown to have both IFN-α expression and an increase in type I IFN–stimulated genes compared with nondiabetic donors (3–5). In addition, studies conducted with nondiabetic human islets have shown that IFN-α alone induces human leukocyte antigen I (HLA-I) overexpression and endoplasmic reticulum (ER) stress, two hallmarks of early T1D, further suggesting its role in disease development (6).

We and others have shown that pro-inflammatory cytokines increase reactive oxygen species (ROS) in β cells (7–10). ROS are chemically reactive, unstable compounds derived from oxygen. Most ROS are naturally produced in β cells as by-products of metabolic reactions and can stimulate important cellular processes, such as β cell proliferation (11) and glucose-stimulated insulin secretion (12, 13). However, when the levels of ROS increase above what can be detoxified by scavenging enzymes in the antioxidant response, oxidative stress can lead to cellular dysfunction and destruction (14–16).

In recent years, β cell heterogeneity has been a well-explored topic, with β cells displaying heterogeneity in terms of both gene expression (17–20) and cellular function, including functional heterogeneity in insulin secretion (21) and calcium signaling (22–24), suggesting that not all β cells respond to stimuli in the same manner. In addition, although T1D is an autoimmune disorder characterized by immune-mediated destruction of the insulin-producing β cells in the pancreatic islet, recent evidence shows residual insulin-positive β cells in the islets of people with long-standing T1D, suggesting that cell survival heterogeneity also exists in this population (25–27). The nonobese diabetic (NOD) mouse model of autoimmune diabetes has also been studied in this context, and a similar subpopulation of persistent β cells resisting sustained immune attack has been identified (28). Together, this suggests that a small subpopulation of uncharacterized β cells exists that are innately resistant to immune assault.

In this study, we aimed to evaluate mechanisms that might link the acute IFN-α response in β cells to downstream functional outcomes. We investigated whether mitochondrial ROS contributes to antiviral signaling during acute IFN-α stimulation in human β cells and whether mitochondrial superoxide might act as a signaling intermediate in the induction of antiviral gene programs. We further examined the extent of heterogeneity in these responses across human donors as well as individual islets and assessed whether features of this signaling pathway differ in residual β cells from donors with T1D. Through these approaches, we sought to define mechanisms that shape intrinsic β cell antiviral responses and may influence islet resilience to inflammatory stress.

RESULTS

IFN-α induces heterogeneous ROS accumulation in human islets in vivo

Human islets (Table 1 and table S1) were infected with RIP1-Grx1-roGFP2, a previously established ratiometric ROS biosensor that we placed under the insulin promoter (29), before transplantation under the kidney capsules of NOD scid gamma (NSG) mice. In the presence of ROS in the cytoplasm, a conformational change in green fluorescent protein (GFP) results in the sensor being more excitable at 405 nm. Therefore, cells where ROS production is increasing can be identified by an increase in the 405-nm signal and a decrease in the 488-nm signal (Fig. 1A). Two weeks after transplantation, intravital imaging showed robust vascularization of the engrafted islets (fig. S1A) along with the presence of human C-peptide in all animals (fig. S1B). After baseline image acquisition, human IFN-α or saline was retro-orbitally administered, yielding measurable circulating IFN-α levels in all IFN-α–injected animals (fig. S1C). The biosensor response was measured according to the expected biosensor profile (fig. S2A) after treatment injection, and β cells were individually traced to define regions of interest (ROIs) for analysis and followed throughout the time of imaging.

Table 1. Human islet donors used for experiments.

Phenotypic information from each donor, including sex, age, BMI, and HbA1C, is provided. HC, healthy control.

Donor ID	Source	Sex	Age	BMI	HbA1C	Disease Status	Experiment
SAM N10920290	IIDP	F	46	18.2	4.4	HC	Intravital imaging
SAMN41425660 (R309)	Alberta	F	47	27.4	5.5	HC	Intravital imaging
SAMN17762820 (R395)	Alberta	F	28	31.4	3.6	HC	Intravital imaging
SAMN16427178	IIDP	F	42	31.2	5.5	HC	Intravital imaging
SAM N16423389	IIDP	M	55	25.8	4.9	HC	Intravital imaging
SAMN17367763	IIDP	M	56	33	5.3	HC	Intravital imaging
SAMN28450743	IIDP	M	47	25.5	5.1	HC	Intravital imaging
SAMN29094695 (R448)	Alberta	F	61	36.1	5.8	HC	Intravital imaging
SAMN38094145	IIDP	F	52	33.6	5.7	HC	DHE in vitro imaging
SAMN37217050 (R505)	Alberta	F	53	33.1	5.4	HC	DHE in vitro imaging
SAMN37844676 (R511)	Alberta	F	47	19.2	5.9	HC	DHE in vitro imaging
SAMN37844810 (R512)	Alberta	M	63	34.7	5.9	HC	DHE in vitro imaging
SAMN38760239 (R522)	Alberta	M	56	29.4	5.0	HC	DHE in vitro imaging
SAMN37217101 (R506)	Alberta	M	66	32.4	7.6	T2D	DHE in vitro imaging
SAM N40974624	IIDP	M	44	34.8	6.5	T2D	DHE in vitro imaging
SAMN41264680	IIDP	M	56	27.8	8.5	T2D	DHE in vitro imaging
SAMN46746834	IIDP	M	58	39.8	6.3	T2D	DHE in vitro imaging
SAMN40137906 (R532)	Alberta	F	28	24.3	5.5	HC	DCFDA in vitro imaging
SAMN40723343 (R538)	Alberta	M	61	32.9	5.5	HC	DCFDA in vitro imaging
SAMN40709610	IIDP	F	55	28.5	5.7	HC	DCFDA in vitro imaging
SAMN47300602 (R575)	Alberta	F	59	43.9	6.1	HC	DCFDA in vitro imaging
SAMN47290391	IIDP	F	54	30.9	4.7	HC	DCFDA in vitro imaging
SAMN49540983 (R589)	Alberta	M	55	24.2	5.4	HC	MitoSOX Red Plate Reader
SAMN49824837 (R591)	Alberta	M	56	23.1	5.6	HC	MitoSOX Red Plate Reader
SAMN50225209	IIDP	F	53	33.4	5.8	HC	MitoSOX Red Plate Reader
SAMN38891395	IIDP	F	71	29	5.1	HC	RNA Sequencing
SAMN38872001	IIDP	F	54	39.7	5.9	HC	RNA Sequencing
SAMN39461369	IIDP	F	53	24.7	5.6	HC	RNA Sequencing
SAMN20926064	IIDP	F	62	26.8	6.2	HC	Immunofluorescence staining
SAMN20923891	IIDP	M	45	50.6	5.4	HC	Immunofluorescence staining
SAMN33456515 (R474)	Alberta	M	48	24.9	5.8	HC	Immunofluorescence staining
SAMN35985448 (R496)	Alberta	F	68	22.3	5.6	HC	Immunofluorescence staining
SAM N15879091 (6034)	nPOD	F	32	25.2		HC	Immunofluorescence staining
SAMN15879216 (6160)	nPOD	M	22.2	23.9	5.2	HC	Immunofluorescence staining
SAMN15879218 (6162)	nPOD	M	22	28.9		HC	Immunofluorescence staining
SAM N15879234 (6178)	nPOD	F	24.5	27.5	5.0	HC	Immunofluorescence staining
SAMN15879385 (6331)	nPOD	F	27	24	5.4	HC	Immunofluorescence staining
SAMN15879277 (6221)	nPOD	F	61	33.7		T2D	Immunofluorescence staining
SAMN25652252 (6541)	nPOD	M	57	26.9	11.1	T2D	Immunofluorescence staining
SAMN33284287 (6568)	nPOD	F	65	28.6	6.8	T2D	Immunofluorescence staining
SAMN53792063	IIDP	M	13	21.8	5.2	HC	qPCR
SAMN54553956	IIDP	F	43	26.5	6	HC	qPCR
SAMN54572192	IIDP	M	40	30.9	4.2	HC	qPCR

Fig. 1. Human β cells accumulate ROS in vivo after IFN-α stimulation.

(A) Representative intravital images of a responding β cell in the kidney capsule of an NSG mouse at the baseline and 60 min after IFN-α stimulation (white arrow). Using the RIP1-GRX1-roGFP2 biosensor, responding β cells can be identified by increased 405-nm channel intensity (magenta) and decreased 488-nm channel intensity (green). (B) Representative heatmaps showing fold change in the 405-nm/488-nm ratio of single cells after stimulation with saline or IFN-α relative to the baseline. Biosensor data, as in (A), are shown for each cell during the course of imaging. Cells with a fold change ≥ 2 are represented in magenta. (C) Graphs depicting the fold change in single cells from combined donors at times indicated. The dashed line indicates the cutoff value of 2. ****P < 0.0001 as determined by an unpaired t test. Data are presented as a scatterplot with each dot representing an individual β cell.

Approximately 30 min after IFN-α administration, we observed that a subset (0 to 63%, donor-dependent) of human β cells exhibits robust production of ROS, as measured by a >1.8-fold change in biosensor ratio (405 nm/488 nm) compared with the baseline, with some cells accumulating ROS as early as 15 min after IFN-α administration (Fig. 1B). Across multiple donors, β cell ROS accumulation peaks at around 45 min after IFN-α stimulation (Fig. 1C). Representative time course images tracking the redox state of single β cells, treated with saline or IFN-α, from one donor are shown in fig. S2B. The number and percentage of responding cells for each donor are shown in table S2.

Donors with a higher proportion of responders are healthier

We observed heterogeneity between donors, where roughly half of donors did not produce ROS in response to IFN-α production (Fig. 2A). Therefore, we asked whether this variable response was associated with phenotypic differences. Using the Human Islet Phenotyping Program data, we integrated our data with phenotypic information including sex, age, body mass index (BMI), HbA1c, ethnicity/race, dynamic insulin secretion data (perifusion), islet insulin content, and endocrine cell composition. Plotting these variables against the percentage of responders from each donor at the 30-min time point, we identified a negative correlation between the percentage of responders and both the BMI [Fig. 2B, coefficient of determination (R²) = 0.8845, P = 0.0016] and HbA1c (Fig. 2C, R² = 0.9096, P = 0.0032). We did not observe correlations with any other variables provided (fig. S3).

Fig. 2. Healthier donors have a higher proportion of responders.

(A) The percentage of responders among total β cells is plotted for each donor whose cells were used in intravital imaging at indicated time points. (B) BMI of each donor plotted against the donor’s frequency of responding cells at 30 min. (C) HbA1C of each donor plotted against the donor’s frequency of responding cells at 30 min. R² and P values from linear regression analysis are provided (n = 7).

IFN-α treatment induces acute elevation in JAK/STAT signaling, the antioxidant response, and oxidative damage in β cells

To validate that β cells also exhibit evidence of expected signaling downstream of the IFN-α receptor, kidneys containing transplanted islets were analyzed. We observed elevated nuclear pSTAT2 expression, a transcription factor in the IFN-α signaling pathway, in islets from animals stimulated with IFN-α (fig. S4). This is consistent with the species specificity of cytokines expected with injection of human IFN-α and with our observation of human IFN-α only in the circulation of the IFN-α–injected animals (fig. S1C). We proceeded to evaluate the mitochondrial antioxidant superoxide dismutase 2 (SOD2) and the oxidative DNA damage marker 8-hydroxy-2′-deoxyguanosine (8-OHdG) (fig. S5A), which should exhibit a negative correlation given that SOD2 is expected to reduce the ROS associated with DNA damage. We quantified 23 images containing 12,370 cells, including 4802 β cells (classified as insulin⁺), which revealed no elevation of either SOD2 or nuclear 8-OHdG protein expression in β cells stimulated with IFN-α (fig. S5, B and C) and no overall correlation between 8-OHdG and SOD2 signal (fig. S5D) (30). However, consistent with in vivo analyses, we noted donor-dependent variability in staining, with explants exhibiting heterogeneous expression of both SOD2 and nuclear 8-OHdG in β cells.

IFN-α induces heterogeneous ROS accumulation in human islets in vitro

Once we determined that IFN-α stimulates ROS production in a subset of human β cells, we proceeded to identify the specific ROS produced. After acute in vitro treatment of human islets with IFN-α and appropriate controls, islets were placed in fresh medium containing Hoechst and dihydroethidium (DHE) or 5 μM 2′,7′-dichlorodihy drofluorescein diacetate (DCFDA) for traditional detection of superoxide or hydrogen peroxide (H₂O₂) production, respectively. In the presence of superoxide or other ROS, a conformational change in DHE occurs, allowing intercalation with nucleic acids and resulting in a strong, nuclear signal (31). In the presence of H₂O₂ or other ROS, DCFDA undergoes a conformational change, allowing it to become fluorescent (32). We found that IFN-α induces a heterogeneous accumulation of superoxide or other ROS from healthy control donors in vitro, demonstrated by a nonuniform increase in nuclear DHE signal (Fig. 3A). Overall, IFN-α significantly increased the density of DHE-positive nuclei (P = 0.0002) and the percentage of DHE-positive nuclei (P < 0.0001) (Fig. 3B). In contrast, we observed no statistically significant changes in DCFDA intensity (Fig. 3C), suggesting that H₂O₂ is not a major factor in this process. To validate the species of ROS being produced in response to IFN-α, we stained human islets with MitoSOX Red, a mitochondrial superoxide indicator. We found that IFN-α stimulates mitochondrial superoxide production in human islets (Fig. 3D).

Fig. 3. Human islets accumulate ROS in vitro after IFN-α stimulation.

(A) Representative images from nondiabetic donor islets treated with PBS, IFN-α, or antimycin A and stained with Hoechst (blue) and either DHE or DCFDA (green) to assess ROS production. Scale bars, 50 μm. (B) Quantification of DHE⁺ nuclei per square micrometer (left) and percent positivity (right) in nondiabetic islets (n = 5). (C and D) Mean DCFDA (n = 5) (C) and MitoSOX intensity quantified and plotted as arbitrary units (A.U.) (n = 3) (D). (E and F) Representative images (E) and quantification of DHE-positive nuclei per square micrometer and percent positivity (F) in donor islets from individuals with T2D (n = 4) as shown in (A) and (B). Data are shown as fold change from vehicle and are presented as the means ± SD. ns, not significant. **P < 0.01, ***P < 0.001, and ****P < 0.0001 as determined by a one-way ANOVA.

Evaluation of islets from donors with type 2 diabetes (T2D) under the same treatment conditions in vitro showed no response to IFN-α (Fig. 3E), evidenced by no change in the occurrence and intensity of nuclear DHE staining, despite a demonstrated response to the positive control antimycin A (13). Quantitative analysis revealed that IFN-α does not stimulate an increase in the number or percentage of DHE-positive nuclei in T2D donor islets (Fig. 3F). Given our previous observation that donors with lower BMI and HbA1c showed a greater proportion of ROS-producing β cells in response to IFN-α, we extended this analysis to pancreatic tissue from donors with T2D (fig. S6A). We quantified 40 images containing 22,465 cells, including 9629 insulin⁺ β cells (fig. S6B). In T2D samples, we observed a significant increase in the total islet expression of SOD2 (fig. S6C, P = 0.0179) and of both SOD2 (P = 0.0357) and 8-OHdG (P = 0.0357) specifically in the β cells (fig. S6D), suggesting elevated basal oxidative stress, which may similarly contribute to the blunted ROS response to IFN-α in donors with higher BMI and HbA1c values (Fig. 2). In β cells of nondiabetic donors, the correlation between nuclear 8-OHdG and cellular SOD2 signal was negligible in the presence of saline (fig. S5D). Even though there was heterogeneity between donors, higher levels of 8-OHdG in T2D donors significantly correlated with a reduction in SOD2 levels [correlation coefficient (r) = −0.4133, P < 0.0001; fig. S6E]. This finding further suggests elevated basal oxidative stress in T2D islet β cells that may preclude ROS elevation in response to IFN-α exposure.

IFN-α induces an inflammatory transcriptional response in cells producing ROS

To identify transcriptional network differences that might predispose a subpopulation of islet cells to ROS accumulation in response to IFN-α, we acutely treated human islets from three donors with saline or IFN-α as above. After treatment, we dissociated islets, stained with DHE, and used fluorescence-activated cell sorting (FACS) to isolate populations that were negative or positive for ROS using antimycin A as a positive control for gating (Fig. 4A). We isolated RNA from these populations and conducted ultralow-mRNA sequencing. First, we compared the differences between the vehicle-treated cells exhibiting basal ROS production and the responder population to identify the effects of IFN-α stimulation on ROS-producing cells. We identified 296 up-regulated and 386 down-regulated genes from this analysis (Fig. 4B). We conducted gene ontology (GO) analysis on these differentially expressed genes and identified several pathways, including several antiviral defense pathways, up-regulated in the IFN-α responder population as compared with cells that produced ROS at the baseline (Fig. 4C). Several of the genes were involved in the MDA5 (melanoma differentiation-associated gene 5) and RIG-I (retinoic acid-inducible gene–I) pathways, which combat viral infection and respond to insufficiently edited endogenous double-stranded RNAs, including the genes encoding both RIG-I [DDX58 (adj. P: 6.075 × 10⁻¹¹)] and MDA5 [IFIH1 (adj. P: 2.026 × 10⁻⁷)] themselves.

Fig. 4. IFN-α induces an antiviral transcriptional response in cells producing ROS.

(A) Representative flow cytometry plots demonstrating how treated cells were sorted into ROS⁻ and ROS⁺ populations and collected for subsequent analysis. SSC-A, side scatter area; PE-A, phycoerythrin area. The red ovals were added to more easily visually distinguish between the two sorted populations. (B) Minus average plot of differentially expressed genes between the PBS ROS⁺ and IFN-α ROS⁺ responder populations. Significantly up-regulated (red) and significantly down-regulated (cyan) genes in the IFN-α–treated ROS⁺ responder cells are shown. (C) Significantly up-regulated pathways in the IFN-α ROS⁺ responder population when compared with the vehicle ROS⁺ population.

Cells accumulating ROS in response to IFN-α have a distinct gene expression signature

We proceeded to identify differences in gene expression of human islet cells that produce ROS in response to IFN-α stimulation from those that do not, even when exposed to IFN-α. We compared differentially expressed genes from the IFN-α–treated group negative for ROS (termed “nonresponders”) and the IFN-α responder population (Fig. 5A). Because the treatment time was only 1 hour, we rationalized that most of the differentially expressed genes (aside from the immediate response genes described in Fig. 4) would represent a signature that predisposes cells to ROS accumulation. Our analysis revealed 1459 up-regulated genes and 1287 down-regulated genes in the responder population (Fig. 5B). Neither of the IFN-α receptors (IFNAR1 and IFNAR2) were differentially expressed between these two populations (fig. S7). Ontology analysis of the differentially expressed genes revealed differences in several pathways in the responder population, including immune response, inflammatory response, negative regulation of apoptosis, and modulation of calcium signaling pathways (Fig. 5C). Because we conducted this analysis on whole human islets and DHE is not β cell–selective, we asked whether the responding population was enriched in any specific islet cell types. We found that responders had a higher expression of several β cell identity markers, such as insulin (INS; P = 0.001), pancreatic and duodenal homeobox 1 (PDX1; P = 1.11 × 10⁻⁵), and NK2 homeobox 2 (NKX2–2; P = 0.0003), suggesting that this population is enriched for β cells (Fig. 5D). In addition, the responder cells had significantly lower expression of glucagon (GCG; P = 6.03 × 10⁻⁷), an α cell marker (Fig. 5D).

Fig. 5. Cells accumulating ROS in response to IFN-α have a distinct gene expression pattern.

(A) Representative image of how islet cells were sorted into ROS⁻ and ROS⁺ populations and collected for subsequent analysis. The flow cytometry plot is the same as in Fig. 4A. The red ovals were added to more easily visually distinguish between the two sorted populations. (B) Minus average plot of differentially expressed genes between the IFN-α ROS⁻ and IFN-α ROS⁺ populations. Significantly up-regulated (red) and significantly down-regulated (cyan) genes are shown. (C) Significantly up-regulated (red) and down-regulated (cyan) pathways in the IFN-α ROS⁺ (responder) population when compared with the IFN-α ROS⁻ population. (D to F) Heatmaps detailing the expression of islet cell markers (D), antioxidant enzymes (E), and cell survival and stress response markers (F). Each column represents an individual donor. Heatmaps represent the fold change of gene expression where yellow is increased expression and blue is decreased expression compared with the PBS-treated ROS⁻ population.

Because ROS levels are mitigated in the cell by the antioxidant response, we investigated the expression levels of several antioxidants between the nonresponder and responder populations (Fig. 5E). The expression of several genes encoding antioxidants was decreased in the responder population, including cytoplasmic antioxidant enzymes {GPX3 (P = 0.0001), GST [A1 (P = 4.04 × 10⁻⁵), A2 (P = 0.001), and A4 (P = 0.001)], and GLRX (P = 0.002)} and the mitochondrial antioxidant enzyme SOD2 (P = 0.0007). Because our results at this point suggested that ROS accumulation in response to IFN-α might be beneficial to the cell, we asked whether responders exhibited any difference in genes involved in cell survival and stress response pathways. We observed increased expression of several prosurvival (33) and stress response genes, including several genes involved in the cellular stress response to viral pathogens (Fig. 5F). This included several genes involved in the management of ER stress [HSPA1B (P = 4.05 × 10⁻⁷), HSPA5 (P = 1.13 × 10⁻⁵), HSP90AA1 (P = 6.41× 10⁻⁶), HERPUD1 (P = 2.96 × 10⁻⁶), and TFEB (P = 9.54 × 10⁻⁶)], regulation of cytokine production and inflammation [IFRD1 (P = 3.45 × 10⁻⁵) and CARD10 (P = 0.001)], and mitochondrial quality control [SPATA18 (P = 0.001) and EFHD1 (P = 0.001)]. In addition, responders have a decreased gene expression of HLA-A (P = 1.50 × 10⁻⁶), HLA-B (P = 2.79 × 10⁻⁸), and HLA-C (P = 4.86 × 10⁻⁵), genes encoding HLA-I molecules implicated in the T cell–mediated destruction of β cells in T1D pathogenesis.

Responder cells are enriched in genes negatively correlated with T1D

To determine whether there was any relationship between differentially expressed genes in the IFN-α responder population and gene expression in the T1D disease state, we compared genes differentially expressed in ROS accumulating IFN-α–treated cells with single-cell sequencing data collected by the Human Pancreas Analysis Program (HPAP; https://hpap.pmacs.upenn.edu) (34, 35). Gene set enrichment analysis [GSEA; (36)] of islet single-cell RNA sequencing (scRNA-seq) data from donors with T1D and respective control donors without diabetes or detectable autoantibodies revealed that the gene set up-regulated in the IFN-α responder population were significantly enriched in genes with higher β cell expression in the nondiabetic controls (Fig. 6A, P = 0.004). GO analysis of the associated leading-edge genes (listed in Fig. 6B) revealed genes involved in ER stress response and protein folding/modification (Fig. 6C). Collectively, these data suggest that either T1D donors have fewer IFN-α responder cells to begin with or these cells are eliminated selectively during disease development.

Fig. 6. Genes associated with IFN-α responsiveness are depleted in β cells from individuals with T1D.

(A) GSEA enrichment plot for the up-regulated genes in responders across genes ranked by their shrunk log₂ fold change of β cell expression in donors with T1D versus control donors from the HPAP scRNA-seq dataset. The leading edge (72 genes) is indicated in cyan. The normalized enrichment score (NES) and nominal P value are indicated. (B) Heatmap showing the DESEq2-normalized HPAP β cell expression of the leading-edge genes (rows) for each donor (column). Donor demographic information and disease status are shown above the heatmap as per the legend to the right. (C) GO analysis of the leading-edge genes up-regulated in β cells from donors with T1D.

A similar GSEA of islet scRNA-seq from nondiabetic donors with autoantibodies and respective controls revealed that the responder populations were enriched in genes with higher β cell expression in the autoantibody-positive population (fig. S8A). A GSEA of genes up-regulated in the responder population comparing donors with T2D to their respective age- and BMI-matched controls did not reveal any enrichment associated with disease state (fig. S8B).

Mitochondrial ROS accumulation after IFN-α treatment is required for activation of antiviral response genes

To determine whether the cytoplasmic ROS increase observed both in vivo and in vitro in response to IFN-α first originated in mitochondria, an organelle that is known for high amounts of superoxide production, we stained a human β cell line (EndoC-βH1) with MitoSOX Green (fig. S9, A and B). MitoSOX Green is a DHE-based fluorescent dye that detects superoxide levels in mitochondria (37). We observed a significant increase in mitochondrial superoxide after 1-hour IFN-α treatment (fig. S9C, P = 0.00100). Pretreatment of cells with MitoQ (38), a mitochondrially targeted antioxidant, abolished this effect (fig. S9C). To further investigate the timing of cytoplasmic and mitochondrial ROS production in human β cells, we transfected EndoC-βH1 cells with a plasmid containing RIP1-GRX1-roGFP2 to measure cytosolic ROS or TOMM20-GRX1-roGFP2 to measure mitochondrial ROS. Two days after transfection, cells were treated with vehicle, human IFΝ-α, human IFN-γ, or antimycin A. Images were collected from each RIP1-GRX1-roGFP2 dish at multiple time points after addition of treatment. At all time points after addition of IFN-α, there was a significant increase in cytoplasmic ROS production (Fig. 7, A to C, P < 0.05 to P < 0.001). The TOMM20-GRX1-roGFP2 biosensor yielded similar results, with IFN-α stimulating a significant increase in mitochondrial ROS at all time points evaluated (Fig. 7, D to F, P < 0.05 to P < 0.001). These data demonstrate that the ROS produced in a human β cell line in response to IFN-α likely originates in the mitochondria and then is rapidly released into the cytosol.

Fig. 7. Human β cells accumulate cytoplasmic and mitochondrial ROS in vitro after IFN-α stimulation.

(A to C) Representative images and quantification of EndoC-βH1 cells transfected with the RIP1-GRX1-roGFP2 plasmid and treated with IFN-α for 15 (A), 30 (B), or 60 (C) min. Scale bars, 10 μm. (D to F) Representative images and quantification of EndoC-βH1 cells transfected with the TOMM20-GRX1-roGFP2 plasmid and treated with IFN-α for 15 (D), 30 (E), or 60 (F) min. Scale bars, 10 μm. For quantification, the fold changes in 405-nm/488-nm intensity relative to average PBS 405-nm/488-nm ratio are shown. n = 3 replicates; *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 as determined by a one-way ANOVA. Data are presented as the means ± SD.

Last, to directly test whether mitochondrial ROS contributes to IFN-α–driven antiviral gene induction, we pretreated human islets with MitoQ, as above, or MnP, a cell-permeable manganese porphyrin and broad-spectrum, nonmitochondrial antioxidant and SOD mimetic (39, 40), followed by IFN-α treatment. After IFN-α exposure, we quantified the expression of DDX58 (RIG1) and IFIH1 (MDA5) by quantitative polymerase chain reaction (qPCR). Scavenging mitochondrial ROS significantly attenuated the IFN-α induction of both DDX58 (Fig. 8A, P = 0.0123) and IFIH1 (Fig. 8B, P = 0.0132). In contrast, this attenuation was not observed after pretreatment with MnP (Fig. 8, A and B).

Fig. 8. IFN-α stimulation of human islet antiviral response genes is dependent on mitochondrial ROS.

(A and B) Human islets from three independent donors were treated with IFN-α for 60 min in the absence or presence of MitoQ or MnP; gene expression of DDX58 (A) and IFIH1 (B) was measured. ACTB was used as the housekeeping control gene. Data are presented relative to the PBS control. n = 3 donors run in duplicates; *P < 0.05 as determined by a one-way ANOVA. Data are presented as the means ± SD.

DISCUSSION

T1D is a complex autoimmune disease, with both a genetic predisposition and unknown environmental triggering event, leading to the progressive loss of insulin-producing β cells in the pancreatic islet. Prior literature suggests that childhood viral infections may play a key role in driving T1D pathogenesis, possibly via viral mimicry, or similarity of viral epitopes to those presented by β cells (41, 42). Enteroviruses have been reported in the pancreata of patients with T1D (43), although recent data do not support persistent infection of the pancreas as a common occurrence in T1D (44). Insufficient editing of endogenous double-stranded RNA molecules has been suggested as another possible trigger for the inflammatory response in autoimmune diseases, including T1D (45, 46). Type I IFNs, including IFN-α, produced during the innate immune response to viral infection or insufficiently edited endogenous RNAs are implicated in T1D pathogenesis. For example, canonical interaction of type I IFNs with their receptors activates JAK (Janus kinase)/STAT (signal transducer and activator of transcription) signaling, where STAT1/STAT2 transcription factors are phosphorylated and translocated to the nucleus (47), activating IFN-stimulated genes to mount a robust antiviral response. Multiple case studies report induction of insulin-dependent diabetes in patients prescribed type I IFNs as treatments for other medical conditions (48), and a strong type I IFN signature has been identified in the sera and tissues of individuals with new-onset T1D (3–5) and children at risk for T1D development (1, 2). This has prompted interest in small-molecule inhibitors of downstream IFN-α effectors such as TYK2 (tyrosine protein-kinase 2), with recent results demonstrating that they can elicit a protective effect in a mouse model of T1D (49). However, a critical knowledge gap still exists as to how type I IFN signaling might facilitate autoimmune attack and β cell dysfunction or destruction.

Prior studies have pointed toward IFN-α–induced islet HLA-I overexpression, making β cells more visible to the adaptive immune system, and ER stress as a mechanism by which IFN-α signaling could lead to autoimmune diabetes (6). Studies in other cell types have also shown that IFN-α induces ROS production (50–52). Although ROS have been implicated in detrimental conditions such as oxidative stress, ROS are also key signaling molecules, promoting beneficial cellular processes like β cell proliferation and glucose-stimulated insulin secretion (11–13). ROS are mostly naturally generated as by-products of the electron transport chain in the mitochondria, an essential process for glucose metabolism (53). As complex signaling molecules, ROS have also been identified as essential molecules in promoting the antiviral response (54–57).

Here, we used intravital microscopy to image human islets in a pseudophysiological condition after their transplantation under the kidney capsules of immunodeficient mice (58). By coupling use of a β cell–specific redox biosensor (RIP1-Grx1-roGFP2) with this technique, we could monitor transient and reversible changes in the cellular glutathione redox state of human β cells in vivo in real time. Using this approach, we demonstrated a rapid, heterogeneous production of ROS in human β cells in vivo after IFN-α stimulation. Specifically, we identified heterogeneous ROS accumulation in ~20% of β cells across donors studied. However, the response was highly variable across individual donors, with some donors not having any responding cells and other donors having as much as 63% responding cells.

To dive deeper into this heterogeneity among donors, we identified a negative correlation between IFN-α–induced ROS production and either HbA1C or BMI, suggesting that donors with a lower BMI and HbA1c have a higher number of β cells that produce ROS in the presence of IFN-α. Selective analysis of the subset of ROS-producing cells indicated that genes up-regulated in responders are enriched in the β cells of control donors as compared with T1D donors in the HPAP scRNA-seq datasets. A similar enrichment of this gene set was observed in the β cells of autoantibody-positive donors. Whereas we cannot be certain of the functional meaning of this observation, especially considering that we do not know when or even whether the autoantibody-positive donors will go on to develop T1D, we hypothesize that this shift in transcriptional signature could represent an attempt to protect the cells from the damaging effects of ROS. Combined with the fact that ROS have been implicated in immune activation, we hypothesize that the increase in ROS production in the presence of acute IFN-α exposure is a protective response.

Our in vivo observations were recapitulated in vitro using the fluorescent ROS probes DHE and DCFDA. We observed an increase in nuclear DHE intensity but no change in DCFDA intensity after IFN-α stimulation, suggesting that superoxide might be the ROS being produced. DHE fluorescence can reflect a mixture of oxidation products, given that superoxide-specific detection via quantification of 2-hydroxyethidium was not performed (59). In addition, DCFDA can detect a broad range of oxidants, not only H₂O₂, because of the rapid pace of conversion of products such as superoxide and the modification of DCFDA itself by hydroxyl radicals and reactive nitrogen species. Therefore, we analyzed the superoxide-specific spectra of MitoSOX Red, finding a significant increase in mitochondrial superoxide in human islets treated with IFN-α for as little as 30 min. Thus, the moderate increase in DCFDA signal observed in IFN-α–treated islets likely represents an increase in superoxide that is rapidly converted to H₂O₂ during ROS metabolism.

Given that several studies have shown that, specifically, mitochondrial ROS promotes activation of the immune response (54, 56, 57), we further validated our finding that IFN-α induces mitochondrial superoxide production in human islets and that these events are associated with the induction of an antiviral response. We observed that IFN-α induced mitochondrial ROS production in a human β cell line and that this ROS elevation was reversible with MitoQ pretreatment. Selective inhibition of mitochondrial ROS with MitoQ in intact human islets inhibited the IFN-α–mediated induction of antiviral response genes, whereas more broad cytosolic antioxidant stimulation with MnP did not have the same effect. The ability of MitoQ to suppress IFN-α–induced antiviral gene activation is likely mediated by its modulation of mitochondrial redox potential and electron transport, which limits superoxide formation. This mechanism is consistent with its inner mitochondrial membrane localization and the failure of the nonmitochondrial antioxidant MnP to produce similar effects. These data support a functional requirement for mitochondrial ROS in mediating an IFN-α–driven antiviral response. In addition, when comparing the gene signatures of nonresponders and responders, we found that there is increased expression of prosurvival and mitochondrial quality control genes coupled with a decrease in the gene expression of the mitochondrial superoxide antioxidant SOD2, supporting that there are differences in these genes that correlate with phenotypic stress response behavior. These findings suggest that intrinsic heterogeneity in antioxidant capacity underlies the selective ROS observed in only a subset of islet cells exposed to IFN-α. This interpretation is consistent with prior literature demonstrating that variability in antioxidant defenses contributes to differential islet cell susceptibility to oxidative stress (60–62). Alongside down-regulation of SOD2, we observed increased expression of SOD3 and GPX3 in responding β cells. This suggests a change in redox dynamics that could favor extracellular superoxide conversion and intracellular H₂O₂ accumulation, enhancing redox signaling in certain β cell populations.

There are several limitations remaining in our study that should be addressed moving forward. Although our results are strongly suggestive of functional outcomes, because our sample size was too small to incorporate multivariate analyses, we acknowledge that this is an exploratory look into correlation between donor characteristics and the percentage of responders. Furthermore, our focus on the acute response and correlations with a lack of survival does not definitively demonstrate the functional outcomes of heterogeneous β cell ROS production in response to extrinsic cytokine stress. Whereas our experimental design models short-term IFN-α exposure to study early dynamic redox responses, in T1D pathogenesis, islets may be subjected to sustained or fluctuating cytokine environments, including prolonged exposure to type I IFN. Thus, our findings may represent an early-phase response to activate an antiviral defense mechanism that may evolve under chronic stimulation. Future studies incorporating extended cytokine treatments and longitudinal follow-up experiments will better model the chronic inflammatory milieu of T1D to determine whether the redox-regulatory pathways identified here are maintained, amplified, or dysregulated over time and whether responders with enhanced redox signaling are more resistant to pro-inflammatory cytokines and immune-mediated injury.

In summary, we describe a previously unrecognized, mitochondrial ROS–dependent antiviral response to acute IFN-α stimulation in a subset of human β cells in vivo. Rather than serving solely as a mediator of oxidative damage, mitochondrial superoxide acted as a signaling intermediate required for rapid induction of key antiviral genes. This response was heterogeneous across donors and enriched in β cells from individuals without diabetes, whereas transcriptional features of ROS-producing cells were diminished in residual β cells from donors with T1D. These findings identify intrinsic β cell heterogeneity in antiviral signaling capacity downstream of ROS production as a potential determinant of islet resilience to inflammatory stress. A recent meta-analysis and systematic review suggested that increased early-life contact with microbes in children may be associated with a reduced risk of T1D (63). It is also known that the NOD mouse model of autoimmune diabetes exhibits reduced incidence of disease when not housed in a pathogen-free environment (64, 65). These data are aligned with the hygiene hypothesis that has been historically attributed primarily to an immune cell–mediated effect (66, 67). Given the broad interest in the contribution of β cells to T1D pathogenesis (59, 68–71), it will be imperative moving forward to determine the role of β cell heterogeneity in intrinsic adaptive stress response and to understand the functional role that mitochondrial redox signaling plays in normal and diseased states.

MATERIALS AND METHODS

Study design

The overall goal of our study was to evaluate the oxidative stress response to IFN-α in human islet β cells. We coupled in vivo analysis to in vitro analysis using both imaging and standard molecular biology approaches to gain insight into the acute response of human β cells to this T1D-associated cytokine with the hypothesis that the acute response may better predict long-term functional response during disease pathogenesis. Sex was considered as a biological variable during the planning of all experiments involving both animals and human specimens. Both male and female mice were used for transplantation studies. Human islets used for experiments were accepted from both male and female donors, as indicated in Table 1. All animal experiments detailed here were approved by the Indiana University School of Medicine Institutional Animal Care and Use Committee. Mice were housed in a temperature-controlled facility with a 12-hour light/12-hour dark cycle and were given free access to standard food and water. For xenograft studies, male and female NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ [NOD/SCIDγ(−/−)] (NSG) mice, aged 8 to 18 weeks, were used. These mice were obtained from the on-site breeding colony at the Preclinical Modeling and Therapeutics Core at Indiana University Simon Comprehensive Cancer Center at 6 weeks of age. All animals for which images were collected were included in the final analysis. All human samples used in this study were deidentified specimens exempt from Institutional Review Board approval. Human islets were obtained via the Integrated Islet Distribution Program (IIDP) or Alberta Diabetes Institute (ADI) islet core. Pancreas sections from organ donors were obtained from the Network for Pancreatic Organ Donors with Diabetes (nPOD). All experiments were performed in at least biological triplicate. For islet studies, a biological replicate was defined as islets from an individual donor, whereas for cell line–based studies, a biological replicate indicates an individual passage of cells. Several experiments, as indicated below, also included technical triplicates to assess consistency in individual donor islets. However, all significance was assessed on the basis of the biological replication of our experiments. Image analysis was performed in a blinded fashion to prevent bias in the analysis process.

Transduction and transplantation of islets

Human islets were cultured in non–tissue culture–treated dishes containing 0.22-μm filter sterilized complete islet medium consisting of standard Prodo islet media (Prodo, cat. no. PIM-S001GMP) supplemented with 5% human AB serum (Prodo, cat. no. PIM-ABS001GMP), 1% glutamine and glutathione (Prodo, cat. no. PIM-G001GMP), and ciprofloxacin (10 mg/ml; Thermo Fisher Scientific, cat. no. MT61277RG). Upon arrival, islets were gently resuspended in fresh complete islet medium to recover in a dish overnight before experimentation. For transduction, islets were partially dissociated with Accutase (Innovative Cell Technologies, cat. no. AT-104) for 1 min at 37°C. The Accutase enzyme was inactivated with an equal volume of complete islet media as defined above. Islets were then transduced by adding 1 × 10¹¹ PFU (plaque-forming units) of adenoviral RIP1-Grx1-roGFP2 per 1000 islet equivalents (IEQs; where the standard definition of 1 IEQ equaling 1 spherical islet with a diameter of 150 μm was applied to account for the nonuniform sizes of donor-derived islets) to islets in complete islet medium and incubating for 16 to 18 hours at 37°C. After this step, islets were gently pelleted and resuspended in complete islet medium, and then ~1000 IEQs were transplanted under the left kidney capsule of each anesthetized NSG mouse, with transplantation into four separate mice per donor conducted to account for variability. Briefly, mice were anesthetized with isoflurane inhalation, and hair was removed from an area on the left flank region approximately twice as large as the intended surgery site. Skin was disinfected using three alternating rounds of betadine surgical disinfectant solution with 70% isopropyl alcohol or sterile saline rinses. The body wall was incised with an incision no more than 1 cm and the kidney exposed. Islets were placed in sterile polyethylene (PE-50) tubing that was distally attached to a Hamilton screw syringe equipped with a blunt needle tip. Tubing containing islets was beveled with surgical scissors. A small nick was made on the renal capsule using a 30-gauge needle, and the beveled end of tubing was carefully placed under the kidney capsule. The islets were slowly delivered under the capsule using the Hamilton syringe. The kidney was then gently placed back into its location in the peritoneal cavity, and the incisions were noncontinuously sutured using absorbable 5–0 vicryl filament. Postsurgery, the animals were rehydrated with saline and administered analgesics [extended-release buprenorphine (1.3 mg/ml)] for pain relief. After surgery, animals were monitored until they regained reflexes, then singly housed in clean cages, and provided with wet feed.

Intravital imaging

Two weeks posttransplant, animals were anesthetized with a 2.5 to 3% isoflurane circuit for imaging. The kidney was gently externalized onto a 40-mm coverslip glass-bottom dish. The animals were placed so that the kidney was beneath the animal, pressed against the coverslip for imaging. Islets were identified using GFP fluorescence and then imaged using a Leica SP8 DIVE microscope equipped with a HCX IR APO 25×/0.95 W lens. During confocal imaging, RIP1-Grx1-roGFP2 was sequentially excited by 405-nm/488-nm laser stimulation, and emission was collected between 490 and 600 nm using internal hybrid detectors, as described previously (29). After baseline acquisition, mice were injected with saline or recombinant, carrier-free human IFN-α (2.75 × 10⁵ IU/ml; Novus Biologicals, cat. no. NBP2–26551) via retro-orbital injection. Z-stack images (30- to 50-μm total thickness, 5-μm step size) were collected every 2 to 3 min for 30 to 60 min. After imaging, animals were injected with a vasculature dye (Alexa Fluor 647–conjugated Albumin, produced in house) to visualize blood flow, and a final image was collected with excitation at 650 nm and collection at 670 to 700 nm. After imaging, terminal blood and tissues were collected. Sera were used to assess human C-peptide levels to determine the functionality of transplanted human islets and serum IFN-α levels using a Mesoscale Discovery U-PLEX custom biomarker assay (cat. no. K15067M) according to the manufacturer’s instructions.

Quantification of in vivo redox status of β cells

Raw images from the intravital experiments were processed using rolling ball background subtraction (radius of 50 pixels), and a Gaussian blur filter was applied. Maximum intensity projections were created for each islet. A three-dimensional drift correction (Image Stabilizer plug-in) was applied to datasets affected by motion artifacts from the animal’s breathing. ROIs were drawn manually around each cell-expressing sensor using the 488-nm channel. Mean fluorescence intensities (405 and 488 nm) were calculated for each ROI for every time point using ImageJ (72). The ratio of fluorescence intensity (405 nm/488 nm) was calculated for each ROI to determine the oxidative state. In addition, the fold change in ratio from the baseline was calculated for each ROI to determine changes in oxidative state. We identified responders as β cells that had a fold change in 405-nm/488-nm ratio of 2 or higher when compared with the baseline on the basis of our prior in vivo observations (29).

Tissue embedding and immunofluorescence staining

Tissues collected after intravital imaging of human islets (kidneys containing transplanted islets) were fixed with 4% paraformaldehyde and embedded in paraffin blocks for downstream immunofluorescence analysis. Tissues containing embedded islets, or nPOD tissues, were sectioned at a 5-μm thickness, deparaffinized with citrate and heat-based antigen retrieval, and blocked for 1 hour at room temperature with a serum-free blocking solution (Abcam, cat. no. 64226). Sections were then stained by immunofluorescence using the following primary antibodies overnight at 4°C in antibody diluent (Abcam, cat. no. 64211): anti-pSTAT2 (Cell Signaling Technology, cat. no. 88410; RRID: AB_2800123; 1:200), anti-insulin (Dako, cat. no. IR002; RRID: AB_2800361; 1:10), anti–4-HNE (Abcam, cat. no. ab48506; RRID: AB_867452; 1:100), anti-SOD2 (Cell Signaling Technology, cat. no. 13141S; RRID: AB_2636921; 1:500), and anti-8OHdG (Abcam, cat. no. ab62623; RRID: AB_940049; 1:100). Secondary antibodies were diluted 1:500 in antibody diluent (Abcam, cat. no. 64211) and incubated at room temperature for 1 hour: Alexa Fluor 647 donkey anti-rabbit immunoglobulin G (IgG; Invitrogen, cat. no. A31573), Alexa Fluor 594 goat anti-mouse IgG (Invitrogen, cat. no. A11032), and Alexa Fluor 488 goat anti–guinea pig IgG (Invitrogen, cat. no. A11073). Tissues were stained with DAPI (4′,6-diamidino-2-phenylindole) to mark nuclei. Images were acquired using a Zeiss LSM 800 confocal microscope or a Leica Stellaris confocal microscope.

For the analysis, maximum intensity projections of z-stacks were generated from images obtained with the confocal microscope. Semiautomated image analysis using supervised algorithms was conducted with QuPath version 0.5.1 (73) by adapting an image analysis pipeline previously published and validated in human pancreatic tissue (74, 75). Multiple quality control steps take place in the image analysis pipeline to guarantee the quality of the results, as detailed below. Channel names and image characteristics were standardized, the entire image was included in a rectangle annotation, and islet areas were manually selected on the basis of insulin staining. We optimized the built-in automatic cell detection on the basis of the DAPI signal, and we added intensity features into the cellular objects. We trained a machine learning–based object classifier for insulin⁺, SOD2⁺, and 8-OHdG⁺ cells; created a composite object classifier; and ran the composite object classifier in all of the cells in the samples. Each cell was classified as positive for any combination of one, two, or three of the markers or was classified as negative. We retrieved the total number of cells and how many of them expressed each of the proteins of interest. We also retrieved the fluorescence intensity data of 8-OHdG and SOD2 from the islet regions and from every cell (8-OHdG from the nucleus and SOD2 from the whole cell).

Live imaging of human islets with ROS probes

Roughly 500 IEQs were transferred into fresh complete islet medium in six-well plates for a 1-hour treatment with phosphate-buffered saline (PBS) vehicle [human IFN-α (2000 IU/ml); Novus Biologicals, cat. no. NBP2–26551), 10 μM antimycin A (Sigma-Aldrich, cat. no. A8674), or 20 μM H₂O₂ (Sigma-Aldrich, cat. no. 216763). Islets were transferred into new wells containing 2 μM Hoechst (Thermo Fisher Scientific, cat. no. 62249) and 10 μM DHE (Cayman Chemical, cat. no. 12013) or 10 μM DCFDA (Cayman Chemical, cat. no. 20656) for 25 min and then transferred to imaging dishes containing fresh complete islet medium for imaging. Z-stack images of at least three whole islets per condition were collected on a Zeiss LSM800 confocal microscope or a Leica Stellaris confocal microscope with the following excitations/emissions (ex/em): Hoechst (ex: 405 nm/em: 413 to 491 nm), DHE (ex: 488 nm/em: 590 to 640 nm), and DCFDA (ex: 488 nm/em: 498 to 610 nm).

Analysis of mitochondrial superoxide in human islets

Human islets were treated with PBS, human IFN-α (2000 IU/ml; Novus Biologicals, cat. no. NBP2–26551), or 10 μM antimycin A (Sigma-Aldrich, cat. no. A8674) for 30 min at 37°C. During treatment, islets were co-incubated with 1 μM MitoSOX Red (Invitrogen, cat. no. M36008) to detect mitochondrial superoxide. After treatment, islets were washed three times with PBS and transferred to a black-walled, clear-bottom 96-well plate for fluorescence analysis. To selectively measure superoxide, we used the 396/580-nm ex/em spectra specific to superoxide detection to measure MitoSOX Red intensity on a SpectraMax iD5 plate reader (Molecular Devices) equipped with an ultraviolet filter. All experiments were performed in technical duplicate or triplicate from three individual donors.

Preparing human islets for gene expression analysis

Islets were divided and treated for 1 hour with PBS vehicle (2000 IEQs), human IFN-α (2000 IU/ml; Novus Biologicals, cat. no. NBP2–26551) (2000 IEQs), or 10 μM antimycin A (Sigma-Aldrich, cat. no. A8674) (500 IEQs). A subset of islets remained untreated for a negative control (500 IEQs). Posttreatment, islets were washed in PBS and then dissociated in 500 μl of Accutase (Innovative Cell Technologies, cat. no. AT-104) for 15 min in a 37°C water bath under gentle agitation. An equal volume of PBS containing 20% heat-shocked fetal bovine serum was added to inactivate the Accutase. Cells were pelleted and washed with PBS and then resuspended in 250 μl of FACS buffer consisting of 1% bovine serum albumin (Millipore Sigma, cat. no. A9418) in PBS. Single cells were filtered through a 30-μm mesh filter, and the cell suspension was transferred to the FACS instrument. A blue laser equipped with a phycoerythrin filter was used for the sorting experiments on the FACS instrument. The unstained, untreated negative control population was used to gate true negative cells. Five minutes before running the antimycin A sample, 10 μΜ DHE (Cayman Chemical, cat. no. 12013) was added to the tube and incubated at room temperature in the dark. After the antimycin A sample was run to gate the positive population of cells, the vehicle- and IFN-α–treated samples were each stained with 10 μΜ DHE and sorted on the basis of the presence or absence of ROS, as determined by the established gates. RNA of the sorted populations was isolated using the Qiagen RNeasy Mini Kit (cat. no. 74134) protocol with addition of deoxyribonuclease (Qiagen, cat. no. 79254).

Another set of islets was treated overnight with 1 μM MitoQ or 68 μM MnP 30 min before stimulation with IFN-α (2000 IU/ml) for 1 hour. Total RNA was isolated after treatment using a Qiagen kit as above, and cDNA was generated using the Applied Biosystems High-Capacity cDNA Reverse Transcription kit (cat. no. 4374967). qPCR was performed to assess the expression of antiviral response genes using primer pairs specific for DDX58 (forward: CATGTCCACCTTCAGAAGTGTCT; reverse: CATAGCAGGCAAAGCAAGCTC) and IFIH1 (forward: GCCCGCTACATGAACCCTGA; reverse: TCTGCAGCAGCAATCCGGT). Relative gene expression levels were calculated using the 2^−ΔΔCt method normalized to the housekeeping gene ACTB (forward: TCGCCCTGGACTTCGAGCAAGA; reverse: TGCCCAGGAAGGAAGGCTGGAA). All experiments for RNA analyses were performed in biological triplicate on three individual donors.

Bulk RNA sequencing and analysis of human islets

Sequencing was carried out in the Center for Medical Genomics at Indiana University School of Medicine using paired analysis. After passing quality control checks, 1 ng of RNA from all samples was run on an Illumina NovaSeq X Plus sequencer. Library preparation was conducted using the Takara SMART-seq mRNA kit. Raw sequencing data were processed and analyzed for differential gene expression using EdgeR, and reads were aligned to the hg38 genome using STAR aligner. Low-quality mapped reads (including reads mapped to multiple positions) were excluded, resulting in analysis of 17,192 genes. Differentially expressed genes were identified after a false discovery rate correction to have an adjusted P value of less than 0.05.

To further assess the β cell expression in donors with T1D versus control donors, we leveraged publicly available scRNA-seq data from HPAP. For each donor islet sample, we obtained counts with CellRanger version 7.1.0 (76) and used Seurat version 4.9.9.9041 (77) [complemented by SoupX (78) and scDblFinder (79)] for cleanup, normalization, preprocessing, and data integration. Cells were annotated with two approaches: scSorter (80) and manual cluster annotation based on selected pancreatic cell markers. Only cells for which the two approaches agreed were assigned to a final cell type. We then used HPAP samples with a minimum number of 100 β cells and whose libraries had been prepared with the same chemistry kit (SC3Pv3). For each donor, β cell pseudobulk counts were generated, and 20,621 genes with ≥10 counts in ≥4 HPAP donors were retained for analyses. DESeq2 (81) was used to compare the available 4 T1D and 26 control donors, generating a preranked gene list on the basis of shrunk log₂ fold change, which was then input into GSEA (36). GSEA was used to study the enrichment of the gene set of up-regulated genes in responders (441 genes with adj. P < 0.05 and fold change >1.5) in the HPAP datasets. The 72 leading edge genes from this analysis were further examined with GO analysis. Genes of interest were first identified via a P value of less than 0.05 and a log₂ fold change of 1.5. The official gene symbol of these identified genes was analyzed using the NIH Database for Annotation, Visualization and Integrated Discovery program. Briefly, the Functional Annotation Tool was used to conduct GO analysis of identified genes. These pathways were subsequently analyzed to be represented as a −log₁₀ (P value).

Live imaging with EndoC-βH1 cells

Human EndoC-βH1 cells (Human Cell Design) were cultured in a monolayer and maintained on Matrigel (100 μg/ml; Corning, cat. no. 356237)– and fibronectin (2 μg/ml; Sigma-Aldrich, cat. no. F1141)–precoated flasks and dishes. Cells were cultured in medium (complete cell medium) consisting of low-glucose Dulbecco’s modified Eagle’s media containing 2% albumin from bovine serum fraction V (Equitech-Bio, cat. no. BAH66), 10 mM nicotinamide (Sigma-Aldrich, cat. no. N0636), transferrin (5.5 μg/ml; Sigma-Aldrich, cat. no. T8158), sodium selenite (6.7 ng/ml; Sigma-Aldrich, cat. no. S5261), 50 μM 2-mercaptoethanol (Sigma-Aldrich, cat. no. M3148), and penicillin (100 units/ml)/streptomycin (100 μg/ml) (Thermo Fisher Scientific, cat. no. 15140122). For imaging, cells were split into imaging dishes (MatTek, cat. no. P35G-1.5–14-C) at a seeding density of roughly 100,000 cells per dish. One to three days later, a subset of dishes was pretreated with 1 μM MitoQ (Selleck, cat. no. S8978) for 6 hours. Cells were treated for 1 hour with PBS vehicle, human IFN-α (2000 IU/ml; Novus Biologicals, cat. no. NBP2 26551), or 10 μM antimycin A (Sigma-Aldrich, cat. no. A8674) in complete cell medium. The medium containing each treatment was removed and replaced with complete cell medium containing 2 μM Hoechst (Thermo Fisher Scientific, cat. no. 62249), 0.5 μM MitoSOX Green (Invitrogen, cat. no. M36006), and 0.1 μM MitoTracker Deep Red (Invitrogen, cat. no. M22426) for 25 min. Cells were then imaged in fresh medium using a Leica Stellaris confocal microscope with the following ex/em: Hoechst (ex: 405 nm/em: 413 to 491 nm), MitoSOX Green (ex: 488 nm/em: 498 to 610 nm), and MitoTracker Deep Red (ex: 638 nm/em: 646 to 811 nm).

For biosensor imaging, RIP1-GRX1-roGFP2 and TOMM20-GRX1-roGFP2 plasmids were used. EndoC-βH1 cells cultured in complete cell medium as above were split into imaging dishes (MatTek, cat. no. P35G-1.5–14-C) at a seeding density of roughly 150,000 cells per dish. One day later, cells were transfected with 3 μg of TOMM20-GRX1-roGFP2 or RIP1-GRX1-roGFP2 plasmid using Lipofectamine 2000 per the manufacturer’s protocol. Cells were treated for 1 hour with PBS; human IFN-α (2000 IU/ml; Novus Biologicals, cat. no. NBP2–26551); recombinant, carrier-free human IFN-γ (1000 IU/ml; R&D Systems, cat. no. 285–100/CF); or 10 μM antimycin A (Sigma-Aldrich, cat. no. A8674) 36 hours posttransfection. Z-stack images were collected using a Leica Stellaris confocal microscope with sequential excitation at 405 and 488 nm and emission at 498 to 610 nm. The fold change in the RIP1-GRX1-roGFP2 ratio (405 nm/488 nm) was calculated from the average vehicle condition for each time point.

Statistical analysis

Individual-level data are presented in data file S1. All data were analyzed using GraphPad Prism 10 software (GraphPad Prism version 10.6.0 for macOS, GraphPad Software; www.graphpad.com). For all analyses, groups were compared by Student’s t test or a one-way analysis of variance (ANOVA), followed by a Dunnett’s or Tukey’s test for multiple comparisons. When appropriate, a mixed-effects ANOVA was used to account for repeated measures, followed by Dunnett’s test for multiple comparisons. All results are presented as the means ± SD. For in vivo analyses, we performed a descriptive statistical analysis and plotted the data to identify outliers. We used the Shapiro-Wilk test for normality and determined that the data were not normally distributed. We then used the nonparametric Mann-Whitney U test to compare the means of the two groups (saline versus IFN-α or nondiabetic versus T2D). Graphs are represented as “SuperPlots” for accuracy of data representation (82). Correlation analysis for figs. S5 and S6 was performed using linear regression and significance assessed using Spearman’s correlation test. For correlation of in vivo data in Fig. 2 and fig. S3, data from each donor were recorded, and simple linear regression analysis was conducted. P values <0.05 were considered statistically significant for all analyses.

Supplementary Material

The PDF file includes:

Figs. S1 to S9

Tables S1 and S2

Other Supplementary Material for this manuscript includes the following:

Data file S1

MDAR Reproducibility Checklist

Data, code, and materials availability:

All data associated with this study are in the paper or supplementary materials. RNA sequencing datasets are available to the public via the Gene Expression Omnibus (GEO) accession number GSE318506. The HPAP scRNA-seq data and metadata were retrieved from the PancDB website (https://hpap.pmacs.upenn.edu/). All materials used or generated in this study are commercially available or will be supplied upon reasonable request.