Interferon-gamma drives programmed death-ligand 1 expression on islet β cells to limit T cell function during autoimmune diabetes

Type 1 diabetes is caused by autoreactive T cell-mediated β cell destruction. Even though co-inhibitory receptor programmed death-1 (PD-1) restrains autoimmunity, the expression and regulation of its cognate ligands on β cell remains unknown. Here, we interrogated β cell-intrinsic programmed death ligand-1 (PD-L1) expression in mouse and human islets. We measured a significant increase in the level of PD-L1 surface expression and the frequency of PD-L1+ β cells as non-obese diabetic (NOD) mice aged and developed diabetes. Increased β cell PD-L1 expression was dependent on T cell infiltration, as β cells from Rag1-deficient mice lacked PD-L1. Using Rag1-deficient NOD mouse islets, we determined that IFN-γ promotes β cell PD-L1 expression. We performed analogous experiments using human samples, and found a significant increase in β cell PD-L1 expression in type 1 diabetic samples compared to type 2 diabetic, autoantibody positive, and non-diabetic samples. Among type 1 diabetic samples, β cell PD-L1 expression correlated with insulitis. In vitro experiments with human islets from non-diabetic individuals showed that IFN-γ promoted β cell PD-L1 expression. These results suggest that insulin-producing β cells respond to pancreatic inflammation and IFN-γ production by upregulating PD-L1 expression to limit self-reactive T cells.

Despite this body of knowledge, the timing, location, and specific cellular interactions that are regulated by PD-1:PD-L1 in T1D remain unclear. While previous reports have shown intra-islet PD-L1 expression on infiltrating mononuclear cells 6,27 , and suggest a role for non-hematopoietic PD-L1 expression to limit diabetes, it is unclear if β cells themselves express PD-L1 and how this expression is regulated during diabetes progression. Additionally, enforcing PD-L1 expression on β cells under the insulin promoter has shown conflicting results, as NOD mice were protected from disease 28 while diabetes-resistant mice were rendered susceptible with insulin promoter-driven PD-L1 expression 29 .
In this study, we measured islet β cell PD-L1 expression and regulation during diabetes pathogenesis. The goals of this study were to improve upon previous strategies for flow cytometric analysis of individual, insulin-positive, live β cells, and determine the specific regulators, location, and timing of PD-L1 expression in both mouse and human β cells. We utilized multicolor flow cytometry and epifluorescent microscopy to measure PD-L1 expression on islet β cells during spontaneous diabetes in NOD mice, and found that PD-L1 expression increased as mice approach diabetes onset, and was associated with islet infiltration. We also investigated the effect of cytokines on PD-L1 expression. The CD274 promoter contains two interferon regulatory factor-1 (IRF-1) binding sites, and previous work has shown that type 1 and type 2 interferons (IFN) induce PD-L1 expression on T cells, B cells, endothelial cells, epithelial cells, and tumor cells 1,22 . We found that IFN-γ and to a lesser extent, IFN-α, promoted increased frequency of PD-L1 + β cells, and increased expression on a per cell basis. Similar to our findings in mice, within human pancreas we found that increased PD-L1 expression correlated with increased inflammatory T cell infiltration in pancreatic lesions. Interestingly, we observed a minor increase in PD-L1 staining in autoantibody positive patients in the absence of overt autoimmune diabetes and found that Th1-associated cytokine IFN-γ modulated PD-L1 expression in vitro on isolated human islets. Taken together, this work illustrates that both mouse and human islet β cells express PD-L1 in response to the same inflammatory cues, which may help delay islet destruction, but is ultimately insufficient to prevent β cell death.

PD-L1 expression on islet β cells. We first performed a time course analysis of Programmed Death
Ligand 1 (PD-L1) expression on islet β cells from the pancreas of NOD mice during type 1 diabetes. In NOD mice, β cells were identified as side and forward scatter high, CD45.1 negative, CD4 negative, lineage marker negative (CD8 − , CD11c − , CD11b − , B220 − , F4/80 − ), live cells, that were positive for intracellular insulin (Fig. 1). Using this strategy, we quantified PD-L1 expression on live β cells directly ex vivo from 5-23 week old non-diabetic NOD female mice and compared them to diabetic NOD female mice. Shown are representative plots from non-infiltrated 5 week old or 13-week old infiltrated non-diabetic NOD mice, and NOD.Rag1 −/− mice ( Fig. 2A). NOD.PD-L1 −/− mice were used as negative controls for PD-L1 staining. We measured an increase in the percentage of PD-L1 + β cells as mice aged, with the highest frequency of PD-L1 in diabetic mice (Fig. 2B). Female NOD mice develop spontaneous autoimmune diabetes beginning at 12 weeks of age in our colony, with 70% incidence by 30 weeks of age (data not shown). Using 12 weeks as an age of demarcation, we demonstrate a statistically significant increase in the percentage of β cells expressing PD-L1 in NOD mice ≥12-weeks old (32.1 ± 2.8%) compared to young NOD mice (7.6 ± 1.5%, p < 0.0001, Fig. 2C). We also measured an increase in the level of PD-L1 expression on β cells as determined by the geometric mean fluorescent intensity (gMFI) on β cells from diabetic mice compared to all other ages except for 5 week old NOD mice (Fig. 2D). Again, using 12 weeks as an age of demarcation, we determined there was not a significant difference between the level of PD-L1 on islet β cells in NOD mice < or ≥12-weeks of age (Fig. 2E). Taken together, these results demonstrate a temporal correlation of PD-L1 expression with diabetes progression with the highest percentage of β cells expressing the highest levels of PD-L1 at diabetes onset.
Islet β cell PD-L1 expression is associated with insulitis. The absence of PD-L1 expression on islet β cells from NOD.Rag1 −/− mice and a significant increase in percent PD-L1 positive β cells after 12 weeks of age prompted us to determine if there was a correlation between PD-L1 expression and insulitis. It is well established that islet infiltration increases with age as mice approach diabetes onset 30,31 . We measured co-expression of PD-L1 and insulin in islets from 5 and 13-week old NOD mice by immunofluorescence (Fig. 3). The expression of islet PD-L1 was not detectable above background in 5-week old NOD mice, but widespread PD-L1 expression was present within islets from 13-week old NOD mice with notable overlap with insulin staining (represented as  (E) PD-L1 gMFI of PD-L1 + insulin-positive β cells from younger (<12 wks) and older (≥12 wks) NOD mice from (C). The lines in (D,E) represent the mean ± standard deviation. Significance was determined as compared to diabetic mice: *p < 0.05, **p < 0.01, ***p < 0.001 using One-Way ANOVA with Tukey correction. Differences between the two age groups were analyzed using Student's t-test (****p < 0.0001). Mice that were diabetic at the time of islet harvest are depicted in red (panels B-E). yellow, Fig. 3A). We detected few CD3ε + T cells in the islets of 5 wk old NOD mice, whereas significant T cell pancreatic insulitis was seen in 13 wk old NOD mice (Fig. 3A). Again using 12 weeks of age as the cut off, we quantified the severity of insulitis using a five tier grading scale 30 . Results presented in Fig. 3B illustrate that approximately 25% of the islets in younger NOD mice (<12 wk) were infiltrated, with less than 10% having severe insulitis (score of 4). However, older NOD mice (≥12 wk) had significant and severe insulitis, with 88% of all islets classified as infiltrated and more than 50% of islets having severe insulitis (score of 4) (Fig. 3B). Despite the presence of insulitis in some mice younger than 12 weeks, PD-L1 was expressed at a low level and by few β cells (Fig. 3A). Islet β cell PD-L1 expression is driven by IFN-γ. Given the correlation between PD-L1 expression and T cell infiltration, we next tested the role of cytokines in regulating PD-L1 expression on islet β cells. It has been reported that CD4 + T cells are required for the initiation of diabetes and that the Th1 cytokines IFN-γ and TNF-α may contribute to islet β cell death [32][33][34][35][36] . We hypothesized that exposing NOD islets to inflammatory cytokines or chemokines could provoke PD-L1 expression. We harvested NOD.Rag1 −/− islets and treated intact islets in vitro for 24 hours with IFN-α, IFN-γ, IL-1β, IL-4, IL-10, IL-12, IL-21, TGFβ, TNF-α, and CXCL10 and measured PD-L1 expression 1,30,31 . Following culture, islets were dissociated into single cells and PD-L1 expression on live beta cells was determined by flow cytometry according to the gating strategy shown in Fig. 1. Representative histogram plots from NOD.PD-L1 −/− mice are shown as negative controls (Fig. 4A, left). Rag1 −/− islets cultured in media alone do not express PD-L1 (Fig. 4A, right), while β cells cultured with IFN-γ show significant PD-L1 expression (Fig. 4A, right). Examination of the cytokine panel revealed that three cytokines induced PD-L1 expression. IFN-α and IL-4 caused a small, but significant increase in β cell PD-L1 expression with 11.9 ± 6.9% and 3.5 ± 0.65% β cells being PD-L1 + after culture, respectively (Fig. 4B). IFN-γ caused the most β cells to express PD-L1, with 78.92 ± 9.92% PD-L1 + β cells after culture ( Fig. 4A and B, p < 0.0001). In addition, PD-L1 expression levels as measured by gMFI were significantly increased over media control after culture with IFN-γ (Fig. 4C, p = 0.0033). Taken together, these data demonstrate the Th1 cytokine IFN-γ caused β cells to express high levels of PD-L1. β cells from infiltrated diabetic patient pancreas sections express PD-L1. The strong response to inflammation combined with Th1 cytokines in mouse islets lead us to hypothesize that diabetic human pancreatic sections would also exhibit significantly higher PD-L1 than non-diabetic control samples. We also predicted that sections from patients with type 2 diabetes would lack T cell infiltration and therefore lack PD-L1 expression given the proposed differences in mechanisms of β cell death in type 1 versus type 2 diabetes 37 . To test this, we obtained fresh frozen pancreas samples from the Network for Pancreatic Organ Donors with Diabetes (nPOD). These included sections from patients with type 1 diabetes, autoantibody positive individuals, patients with type 2 diabetes, and non-diabetic controls. Evaluation of these samples by epifluorescent microscopy allowed us to simultaneously evaluate insulitis, defined by the intra-islet presence of one or more CD4 + or CD8 + T cells, and determine the intensity of PD-L1 expression on islet β cells defined by insulin positive cells 38 (Fig. 5). We  Table 1). In all three experiments, islet beta cells from diabetic patients exhibited significantly higher PD-L1 expression on a per cell basis (Fig. 6A-C) compared to autoantibody positive individuals, patients with type 2 diabetes and non-diabetic individuals (representative images shown in Fig. 5). Given that only the islets from patients with T1D had significant insulitis, these results suggest that PD-L1 expression in human islets is associated with pancreas T cell infiltration, consistent with our analysis of NOD mouse islets.
Human islet β cells upregulate PD-L1 in response to IFN-γ. Finally, we tested whether inflammatory cytokines could promote PD-L1 expression on human islet β cells. Given our cytokine results from mouse islets, and limited human islet availability, we restricted our analysis to the Th1 cytokine IFN-γ. Following 24 hours of culture, 1000 hand-counted human islets per well were dispersed to single cells and stained with a similar antibody panel as in the mouse studies to evaluate PD-L1 expression on β cells (Fig. 7A). Specifically, β cells were identified as forward and side scatter high cells, live cells, CD45RA/RO − , CD11c − lineage negative (CD3 − , CD11b − , CD11c − , CD20 − ), intracellular insulin-positive cells. Similar to our results using murine islets, Fig. 7B illustrates human β cells cultured in media alone did not express PD-L1 (4.3 ± 0.7%). We found that IFN-γ promoted a strong increase in the frequency of PD-L1 + β cells (23.2 ± 1.8%, p < 0.0001; Fig. 7B) along with enhanced PD-L1 expression on a per β cell basis (p = 0.0084; Fig. 7C). These data illustrate that NOD mouse and human islet β cells both respond to inflammatory cytokines and suggest that there may be common pro-inflammatory cytokine cue that regulates PD-L1 expression, in particular IFN-γ.

Discussion
Previous work has demonstrated PD-L1 expression within the islets 6,27 , but the exact cellular source, timing, and regulation of PD-L1 remained unclear. In the current study, we designed a flow cytometric approach to detect live, insulin-producing β cells to evaluate PD-L1 expression during diabetes pathogenesis. Importantly, this flow cytometry-based approach allowed for exclusion of potentially contaminating intra-islet cell types and facilitated analysis of both murine and human β cell PD-L1 expression either directly ex vivo or following in vitro culture with cytokines. Using this approach, we determined that β cell PD-L1 expression increased as diabetes progressed. We also confirmed that increased PD-L1 expression was associated with lymphocyte infiltration in the context of spontaneous diabetes and following adoptive transfer of in vitro activated BDC2.5 CD4 + T cells Figure 6. PD-L1 expression from pancreas β cell tissue is highest in T1D patients compared to autoantibody positive or T2D patients. β cell-intrinsic PD-L1 expression was quantified by epifluorescent microscopy for one subject in each group (non-diabetic, type 2 diabetic, autoantibody positive, type 1 diabetic) over three independent staining experiments (A-C). Each pancreatic islet from each subject is represented by one symbol, with red symbols representing islets containing T cell infiltration. Significant increase in islet PD-L1 mean fluorescence intensity (MFI) was determined by Student's t-test within each experiment and is represented with: **p < 0.01, ***p < 0.001, ****p < 0.0001 as compared to non-T1D individuals. (data not shown). We extended our analysis to human islets from non-diabetic, type 2 diabetic, autoantibody positive, and type 1 diabetic human patients and determined that β cells from type 1 diabetic and autoantibody positive patient pancreas sections also have significant PD-L1 associated with T cell infiltration.
The histological data (both mouse and human) suggests that islet β cell expression of PD-L1 is highly dependent on T cell infiltration. We demonstrate that inflamed islets have the highest level of PD-L1 expression and neighboring islets without significant T cell infiltration do not have strong PD-L1 expression. This result suggests that the cytokines driving PD-L1 act locally and not through a diffuse network, reminiscent of observations in human malignancies where PD-L1 expression is often detected on tumor cells adjacent to immune cell infiltrates 39 . This intimate regulation may also limit effective T cell inhibition globally throughout the pancreas and account for continued β cell death and eventual loss of islets. It is also true that PD-L1 expression is not uniquely expressed on islet β cells within the pancreas. Results in Fig. 5 illustrate PD-L1 staining that does not entirely co-localize with insulin. In other words, PD-L1 may be expressed on β cells that no longer make insulin, or may be expressed by intra-islet T cells, B cells, macrophages and dendritic cells. Previous work in the field has demonstrated islet PD-L1 expression was important in the transplant setting, and suggested that intra-islet dendritic cells or islet β cells were responsible for enhanced graft survival 27,40,41 . Our work suggests that β cell-intrinsic PD-L1 expression is strongly associated with insulitis and driven by IFN-γ and to a lesser extent IFN-α, and likely has a functional significance during disease progression. It is possible that other cytokines involved in T1D but not tested herein also drive PD-L1 expression. For instance, IL-17 has been shown to induce PD-L1 on human colon and prostate cancer cell lines in an NF-κB-dependent manner 42 . Whether IL-17 has the same effect on β cells remains to be tested.
Checkpoint blockade targeting the PD-1/PD-L1 pathway has garnered international attention as a promising cancer immunotherapy 43 . PD-1 is highly expressed on tumor infiltrating T cells and checkpoint therapy in patients with PD-L1 + tumors has been shown to be effective in early clinical trials 43 . However, adverse events such as rapid autoimmunity including T1D can develop following checkpoint blockade, further suggesting a role for this inhibitory pathway in regulating autoimmunity 19 . Therefore, given the current information and PD-L1 expression within the islets, and our previous work showing that nearly all islet antigen-specific CD4 + T cells express PD-1 in the pancreas 44 , cautious application of PD-1:PD-L1 blockade is warranted. Targeting tumor-specific T cells with checkpoint therapy while restraining autoimmunity is of great importance. Our work suggests that regulation of inflammatory cytokines may be a new avenue of focus to enhance or diminish PD-L1 mediated T cell inhibition, especially when considering checkpoint therapy in individuals with HLA types associated with higher risk of developing autoimmunity or individuals that test positive for islet specific autoantibodies. In one of the three human patients with increased autoantibodies we found a significant increase in PD-L1 expression over control and T2D patients (Fig. 6). PD-L1 expression status therefore could be used as an additional criterion prior to checkpoint inhibition and biomarker of adverse events following therapy. PD-L1 expression by nonhematopoietic cells has been shown to limit immunopathology in the context of chronic viral infection [45][46][47][48] . The present study indicates that both mouse and human β cells may be able to reduce local effector T cell function and act in their own defense by expressing PD-L1 in response to IFN-γ. It is tempting to speculate that β cell-intrinsic PD-L1 expression plays a role in promoting T cell exhaustion in the pancreas. Our findings represent a significant advancement in our understanding of PD-L1 expression and regulation in response to inflammation in both animal models of diabetes and in autoantibody positive individuals or patients with T1D. However, even though PD-L1 can hold autoreactive T cells in check for some time, ultimately the majority of NOD female mice develop diabetes, indicating PD-L1 alone is not sufficient to prevent autoimmunity. Our findings in autoantibody positive and type 1 diabetic human samples support this hypothesis. Future studies using combinatorial approaches to limit autoreactive T cells together with enhancing β cell PD-L1 expression could provide a therapeutic opportunity to maintain β cell mass in patients with T1D.
NOD mice were purchased from Taconic (Hudson, NY). Diabetes was defined as 2 consecutive blood glucose readings >250 mg/dL. All mice were housed in specific-pathogen free barrier facilities and all animal experiments and methods were performed in accordance with the relevant guidelines and regulations approved by the Institutional Animal Care and Use Committee at the University of Minnesota.
Statistical Analysis. Data were analyzed using Prism 7 (GraphPad Software, La Jolla, CA). P-values shown for individual figures represent results of two-tailed Student's T-test or One-Way ANOVA with Tukey post hoc correction between cytokine treatment groups, or between different age groups for percent PD-L1-expressing β cells and PD-L1 geometric mean fluorescent intensity.
Data availability statement. The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.