Repurposed JAK1/JAK2 Inhibitor Reverses Established Autoimmune Insulitis in NOD Mice

In type 1 diabetes, insulin-producing β-cells in the pancreas are destroyed by autoreactive T cells (1). NOD mice share key features of human type 1 diabetes pathogenesis and are used as a preclinical model for testing new therapies (2,3).

Interferon (IFN) signaling results in the upregulation of many genes in β-cells that result in the immune cell infiltration of the pancreatic islets and priming of β-cells for immune-mediated destruction. The genes upregulated include those that encode adhesion molecules and chemokines and MHC class I (4). High MHC class I expression on β-cells is a hallmark of human and mouse autoimmune diabetes (5–9). This has recently been shown to be associated with high levels of phosphorylated signal transducer and activator of transcription (p-STAT) 1 (8). IFN signaling through its receptor results in the activation of receptor-associated Janus kinases (JAKs) and phosphorylation of STAT proteins. A meta-analysis (10) of gene and protein interaction networks among 10 different autoimmune diseases showed JAK-STAT and IFN signaling as key converging pathways in autoimmune diseases. Surprisingly, the deficiency of IFN-γ or its receptor or the deficiency of IFN-α receptor does not prevent diabetes; however, the deficiency of STAT1, a downstream of IFN-γ signaling, or the overexpression of suppressor of cytokine signaling 1 in β-cells, a negative regulator of IFN signaling, resulted in protection from diabetes in NOD mice (11–13). This led us to look for established drugs that target this pathway. There are currently no inhibitors of STAT1 in clinical use, but there are a number of small-molecule inhibitors of the JAK proteins that are directly responsible for STAT1 phosphorylation/activation.

Small-molecule inhibitors of JAK proteins, such as ruxolitinib, have been approved for clinical use in the treatment of myeloproliferative disease (14) and have been used in clinical trials for the treatment of rheumatoid arthritis (15,16) and alopecia areata (17). We have tested the hypothesis that JAK1/JAK2 inhibitors could be repurposed to inhibit T-cell–β-cell interactions in type 1 diabetes. AZD1480 is an ATP-competitive inhibitor with specificity for JAK1 and JAK2 (18). It also inhibits TrkA, which predominantly functions in the nervous system and is not thought to have a major role in the immune system (19). Although no longer in clinical development, this drug targets the same active site as clinically approved drugs (ruxolitinib and baricitinib) but was less costly to obtain for our purposes, thus serving as a proof of principle for the use of JAK inhibitors in diabetes.

NOD/Lt mice were bred and maintained at St. Vincent’s Institute. C57BL/6 mice were obtained from the Animal Resources Centre (Canningvale, Western Australia, Australia). H-2K^d–restricted T cells from NOD8.3 mice (a gift from Professor P. Santamaria) recognize IGRP_206–214 (20), and IA-g7–restricted T cells from BDC2.5 mice recognize chromogranin A (21). All animal studies were approved by the institutional animal ethics committee.

AZD1480 (AstraZeneca) has an IC₅₀ of 0.26 nmol/L for JAK2 and 1.3 nmol/L for JAK1. It is more specific and chemically and metabolically more stable than first-generation tyrphostin compounds, such as AG490 (IC₅₀ >10 μmol/L for JAK2), that have previously been used (13,22). AZD1480 was readily soluble when diluted in 0.5% hydroxypropyl methylcellulose/0.1% Tween 80. Mice were treated with 30 mg/kg AZD1480 or vehicle twice daily by oral gavage. The duration of treatment is described for each experiment.

Mouse islets of Langerhans were isolated using collagenase P (Roche, Basel, Switzerland) and Histopaque-1077 density gradients (Sigma-Aldrich) as previously described (23). Human islets were purified from heart-beating, brain-dead donors, with written informed consent from next of kin, by intraductal perfusion and digestion of the pancreas with collagenase followed by purification using Ficoll density gradients (24).

Islets were cultured with AZD1480 or vehicle control for 1 h at 37°C in 5% CO₂ and, then mouse IFN-γ (100 units/mL; BioLegend) or human IFN-γ (50 ng/mL; Peprotech) was added for 30 min. Islets were lysed in radioimmunoprecipitation assay buffer. Immunoblotting was performed with anti–p-STAT1 or anti–total STAT1 (BD Biosciences, San Jose, CA), normalized to β-actin (Santa Cruz Biotechnology, Dallas, TX), and shown as the fold induction versus the control sample. For the analysis of MHC class I/HLA expression, islets were treated with vehicle or AZD1480, cultured with IFN-γ for 48 h, and analyzed using flow cytometry.

Whole blood was stimulated with 100 units/mL IFN-γ for 30 min at 37°C. Red blood cells were lysed (155 mmol/L NH₄CL, 10 mmol/L Tris-HCl, pH 7.5). Leukocytes were analyzed by Western blot for p-STAT1 as described above.

In vitro cytotoxicity assays were performed with whole islets from 4- to 6-week-old NOD mice and activated NOD8.3 T cells as previously described (25). For AZD1480 treatment, ⁵¹Cr-loaded islets were treated with 2 μmol/L AZD1480 for 60–90 min before adding NOD8.3 T cells.

Eight-well μ-Slides (Ibidi) were coated overnight with 10 µg/mL poly-d-lysine. A single-cell suspension of mouse islets was prepared, and 70,000 cells were dispersed into each well in 300 μL of complete CMRL. Vehicle or AZD1480 (2 μmol/L) was added after 2 h. After 60–90 min, 100 units/mL IFN-γ was added. Cells were then kept for 36–48 h at 37°C. Activated NOD8.3 T cells were labeled with Fluo-4 (1 μmol/L) for 10 min (26). The 1 × 10⁵ Fluo-4–labeled, preactivated NOD8.3 T cells (25) were added to each well with 100 μmol/L propidium iodide. Slides were imaged using a Leica SP5 confocal microscope (26). Image analysis was performed using MetaMorph Imaging Series 7 software (Universal Imaging).

CD8⁺ T cells (CTLs) from NOD8.3 mice were labeled with carboxyfluorescein succinimidyl ester (CFSE) (23), resuspended at 2.5 × 10⁷ cells/mL in PBS and 200 µL was injected intravenously into the tail vein of AZD1480- or vehicle-treated mice. Treatment with AZD1480 or vehicle started 1 day before transfer and continued for 5 days, then the pancreatic lymph nodes (PLNs) and islets were analyzed by flow cytometry.

Isolated islets were dispersed using trypsin (23). Mouse islet cells were stained with anti-CD45 and biotinylated anti-H2K^d or biotinylated anti-H2D^b (all from BD Pharmingen) followed by streptavidin-allophycocyanin (BioLegend). β-Cells were identified by autofluorescence (23). Human islet cells were stained with phycoerythrin-conjugated anti-HLA class I (BioLegend). Propidium iodide was used to exclude dead cells. Single-cell suspensions of PLNs were stained with anti-CD8a (BioLegend). Data were collected on the FACSFortessa cell analyzer (BD Bioscience, San Jose, CA) and were analyzed using FloJo Software (TreeStar, Inc., Ashland, OR).

The 5-µm frozen sections were prepared from three levels (200 µm apart), acetone fixed, and stained with guinea pig anti-insulin followed by horseradish peroxidase–conjugated anti–guinea pig Ig (Dako Cytomation, Carpenteria, CA) (9). Insulitis was scored using the following scale: 0 = no infiltrate, 1 = peri-islet infiltrate, 2 = extensive (>50%) peri-islet infiltrate, 3 = intraislet infiltrate, and 4 = extensive intraislet infiltrate (>80%) or total β-cell loss. The percentage of islets with each score per pancreas was calculated by addition of the scores for the three sections.

Fifty islets per tube from AZD1480- or vehicle-treated mice were resuspended in 200 μL of PBS with protease inhibitor cocktail (Sigma-Aldrich). The islets underwent six freeze/thaw cycles, then centrifugation. The supernatants were analyzed using a mouse chemokine (C-X-C motif) ligand 10 (CXCL10) ELISA (R&D Systems).

For anti-CD3 monoclonal antibody (mAb) treatment, C57BL/6 mice or NOD mice were treated with AZD1480 or vehicle by oral gavage for 2 days. On day 2, mice were injected intraperitoneally with 10 µg anti-mouse CD3 Ab (clone 145–2C11; Bio X Cell, West Lebanon, NH). Islets were isolated on day 3 and analyzed by flow cytometry.

Mice were monitored for diabetes by measurement of urinary glucose levels with Diastix (Bayer Diagnostics). Mice suspected of hyperglycemia were further tested on 2 consecutive days, and those with three positive tests were considered diabetic. Blood glucose levels (≥15 mmol/L) were confirmed using Advantage II Glucose Strips (Roche).

BDC2.5 CD4⁺ T cells were sorted from the spleens and PLNs of transgenic mice (27) and 1 × 10⁶ cells/mouse were injected intravenously into NOD.Rag1^−/− mice that were monitored for the development of diabetes.

Female NOD mice (15–17 weeks of age) were treated with 300 mg/kg cyclophosphamide. At 4 days postinjection, mice received either AZD1480 or vehicle twice daily by oral gavage. Mice were monitored for diabetes development.

Female NOD mice with two consecutive blood glucose readings of ≥13.9 mmol/L (250 mg/dL) were implanted on the second day with a LinBit insulin pellet (LinShin Canada, Inc.) (28). The average blood glucose readings at diabetes diagnosis of AZD1480-treated mice was 17.5 ± 2.8 mmol/L, and of vehicle-treated mice was 18.6 ± 2.3 mmol/L. Mice were treated with AZD1480 or vehicle twice daily by oral gavage until the return of hyperglycemia or 35 days after diabetes diagnosis. Blood glucose levels were measured two times per week.

Statistical analysis was performed using GraphPad Prism 6 Software (GraphPad, San Diego, CA). All data shown as bar graphs are represented as the mean ± SEM. A P value of <0.05 was considered to be significant.

Isolated mouse or human islets were treated with AZD1480 or vehicle and IFN-γ. Western blot analysis shows that IFN-γ–induced phosphorylation of STAT1 was inhibited by AZD1480 in islets isolated from mouse (Fig. 1A) or human (Fig. 1B). IFN-γ induces upregulation of MHC class I on the surface of β-cells, and AZD1480 prevented this in mouse (Fig. 1C and D) and human β-cells (Fig. 1E and F), as determined by flow cytometry.

CTLs directly recognize β-cells through T-cell receptor (TCR)/peptide-MHC interactions. IFN-γ signaling on β-cells is required for the upregulation of MHC class I, and this is an essential step in β-cell destruction by cytotoxic CTLs (25). We used time-lapse microscopy to visualize the effect of AZD1480 on this interaction and in particular immune synapse formation (26,29). Dispersed NOD mouse islet cells cultured with vehicle or AZD1480 and IFN-γ were used as target cells. Activated CTLs from TCR transgenic NOD8.3 mice that recognize IGRP_206–214 were used as effector cells (20). Calcium influx in the CTLs, measured with the calcium fluorophore Fluo-4 AM, indicated the recognition of β-cell antigen on vehicle-treated β-cells (Fig. 2A and C and Supplementary Movie 1, green fluorescence). In contrast, CTL interaction with AZD1480-treated β-cells resulted in significantly reduced calcium influx in T cells (Fig. 2B and C and Supplementary Movie 2). The duration of synapse formation, measured as the time from calcium flux to propidium iodide uptake in the target, was also significantly decreased when CTL interacted with AZD1480-treated β-cells compared with vehicle-treated β-cells (Fig. 2D). Consistent with reduced immune synapse formation, ⁵¹Cr-loaded islets cultured with AZD1480 were significantly protected from CTL cytotoxicity compared with vehicle-treated islets (Fig. 2E).

Peripheral blood was collected at 6, 8, 16, and 20 h after oral gavage of 30 mg/kg AZD1480 or vehicle and stimulated with IFN-γ to induce phosphorylation of STAT1, a pharmacodynamic readout of drug action. At 6 and 8 h after administration, there was a ≥50% reduction in STAT1 phosphorylation (Fig. 3A and B). However, by 16 and 20 h after gavage, IFN-γ–induced STAT1 phosphorylation was not inhibited, indicating that clearance of the AZD1480 from the circulation had occurred by these later time points (Fig. 3A and B). On the basis of these data and consistent with previous studies, we treated mice twice daily with oral gavage at 30 mg/kg/dose in all subsequent experiments.

We next tested whether AZD1480 could inhibit cytokine-induced MHC class I upregulation on β-cells in vivo using a rapid non-autoimmune model. Injection of anti-CD3 mAbs into mice induces systemic secretion of IFNs and the upregulation of MHC class I on β-cells (30,31). NOD mice were treated with AZD1480 twice daily by oral gavage, and on the second day of treatment 10 μg of anti-CD3 mAb was administered. The following day, MHC class I expression on β-cells was analyzed by flow cytometry (Fig. 3C–F). β-Cells isolated from mice receiving a combination of AZD1480 and anti-CD3 mAbs expressed significantly less MHC class I on their surface compared with mice that received vehicle and anti-CD3 or anti-CD3 mAbs alone (Fig. 3C–F). The inhibition of MHC class I upregulation on the β-cell surface in vivo suggests that AZD1480 blocks proinflammatory cytokine signaling in β-cells. An alternative explanation could be that AZD1480 inhibits the production of cytokines in response to TCR stimulation by anti-CD3 Ab.

At 90–110 days of age, NOD mice have established islet infiltration, high levels of IFN-γ detectable within islets, and ongoing β-cell destruction. This is a time point near or at the diagnosis of diabetes. Treatment of 90- to 110-day-old NOD mice with 30 mg/kg AZD1480 or vehicle twice daily by oral gavage for 2 weeks resulted in a significant reduction in MHC class I expression on β-cells compared with vehicle-treated mice (Fig. 4A and Supplementary Fig. 1). We also observed a significant reduction in the proportion of CD45⁺ immune cells in the islets of AZD1480-treated NOD mice compared with vehicle-treated mice (Fig. 4B and Supplementary Fig. 1).

The reduced proportion of CD45⁺ cells in islets of AZD1480-treated mice suggested that blocking JAK-STAT signaling may affect the migration and retention of T cells in islets (32). CFSE-labeled NOD8.3 splenocytes were adoptively transferred into AZD1480 or vehicle-treated NOD mice. Treatment was started 1 day before transfer and maintained for another 5 days. Islets and PLNs were harvested for the analysis of T-cell migration and proliferation (Fig. 4C–E). Although proliferation was similar between AZD1480- and vehicle-treated groups in the PLN, AZD1480 treatment significantly inhibited the proportion of CFSE-labeled CTLs trafficking from the PLN into the islets (Fig. 4C and D). The proportion of endogenous CTLs in the islets was also reduced in the AZD1480-treated mice compared with controls (Fig. 4B and E). These data support the idea that inhibiting JAK1/JAK2 alters the migration and accumulation of CTLs in the islets from the site of priming in the PLN. We have previously shown that reducing the expression of MHC class I on the β-cell surface reduces the ability of CTLs to proliferate within the islet (23). However, because of the relatively short duration of AZD1480 treatment (1 day) of the adoptively transferred CTLs that were observed in the islets, proliferation was similar in CTLs from AZD1480- and vehicle-treated mice (Fig. 4F and G). It is possible that a longer length of treatment would affect CTL proliferation.

The reduced proportion of CTLs in the islets suggested that JAK1/JAK2 inhibition may impact on the secretion of chemokines within islets and prevent immune-cell migration. STAT1 is required for IFN-γ–induced expression of the CXCL10 (33), which is required for trafficking of CTLs to the site of inflammation (34). There was significantly less CXCL10 detectable in islet lysates prepared from AZD1480-treated mice (Fig. 4H). Therefore, a possible explanation for reduced insulitis in JAK1/JAK2 inhibitor–treated mice is reduced chemoattraction, migration, and accumulation of autoreactive CTLs in islets. It is also possible that AZD1480 promotes apoptosis of T cells within islets.

NOD8.3 mice develop diabetes at an accelerated rate, with 70–80% of mice developing disease by 100 days of age (20). Mice were treated with AZD1480 or vehicle from 32 days of age, when insulitis has begun but the β-cell mass remains. After 28 days, the majority of islets in the vehicle-treated NOD8.3 mice had extensive immune cell infiltration and β-cell destruction (Fig. 5A and B). In contrast AZD1480-treated mice were significantly protected from developing insulitis (Fig. 5A and B). AZD1480-treated NOD8.3 mice were significantly protected from diabetes with delayed onset and reduced incidence compared with vehicle-treated NOD8.3 mice (Fig. 5C). These results show that JAK1/JAK2 inhibition prevents the initiation of insulitis in previously uninfiltrated islets, inhibits the progression of destructive insulitis, and protects β-cells from destruction that is CTL dependent in the NOD8.3 model.

The JAK-STAT pathway has also been linked to the pathogenesis of CD4⁺ T cell–mediated diabetes (35). BDC2.5 T cells were adoptively transferred into NOD.Rag1^−/− recipients. In this accelerated model, BDC2.5 T cells migrate into the islets as early as 24 h after transfer and mice develop diabetes in 1–2 weeks, independent of cytotoxic CTLs (27). NOD.Rag1^−/− recipients were treated with AZD1480 or vehicle twice daily after adoptive transfer of BDC2.5 T cells. The percentage of CD45⁺ cells present in the islets of vehicle-treated mice increased over the 13 days until all mice had diabetes (Fig. 5D). AZD1480 treatment significantly reduced islet infiltration of CD45⁺ cells at day 11 and at day 13 (Fig. 5D). All vehicle-treated mice developed diabetes within 10–12 days. By contrast, diabetes onset in the AZD1480-treated mice was significantly delayed until 19–22 days after transfer (Fig. 5E). Thus, JAK1/JAK2 inhibition impairs the ability of BDC2.5 CD4⁺ T cells to cause insulitis and diabetes.

Treating NOD mice that have established insulitis with cyclophosphamide results in accelerated diabetes development (36). Four days after treatment of 105- to 120-day-old NOD mice with cyclophosphamide, mice were treated with AZD1480 or vehicle. The expression of MHC class I on β-cells was reduced in AZD1480-treated mice, and the proportion of CD45⁺ cells in the islets was also reduced (Fig. 5F and G). AZD1480-treated mice were protected from diabetes development during the 3 weeks of treatment. By contrast, >80% of vehicle-treated mice developed diabetes (Fig. 5H).

Finally, we tested the ability of AZD1480 to reverse already established diabetes in female NOD mice. At the time of diabetes diagnosis, mice were implanted with an insulin pellet and treated twice daily with AZD1480 or vehicle. Although hyperglycemia returned in 89% of vehicle-treated mice when the effects of the insulin pellet wore off 15–25 days after diabetes diagnosis, 78% of AZD1480-treated mice remained normoglycemic for the study period of 35 days (Fig. 6). After treatment ceased, some mice were kept to assess the durability of the diabetes reversal. All mice became hyperglycemic after 1 week (n = 5), indicating that after 35 days of treatment the effect of the drug wore off relatively rapidly. It has recently been reported (37) that gene expression changes induced by a JAK1 inhibitor remain even after a washout period of 18 h. Our data suggest that these transcriptional changes might not be observed a week after drug removal.

Previous studies using genetically modified mice have identified the JAK-STAT pathway as being important in the development of autoimmune diabetes (11–13,38). We used a small-molecule inhibitor of JAK1/JAK2 (AZD1480) to inhibit this pathway as a way to translate these older findings to a clinically applicable setting. AZD1480 blocked IFN-γ–induced MHC class I upregulation on β-cells, resulting in a weakened interaction and shorter synapse dwell time between autoreactive CTLs and JAK1/JAK2 inhibitor–treated β-cells and reduced overall killing. Furthermore, the JAK1/JAK2 inhibitor prevented or delayed β-cell destruction by T cells in three different mouse models of autoimmune diabetes and reversed newly diagnosed diabetes in NOD mice. Our data suggest that JAK1/JAK2 inhibitors block cytotoxicity of β-cells and reduce insulitis. It is unlikely that these drugs target the autoimmune process itself (e.g., by removing or disabling the autoreactive T cells).

Discovering new uses for approved drugs is one way to shorten the transition from the laboratory to the clinic and has many advantages. For example, pharmacokinetic, safety, efficacy, and mechanism of action data are already known from previous clinical trials (39,40). Successful transition of any new molecular entity from preclinical development to the clinic requires the drug to have a clear mechanism of action together with a high exposure of the target to the drug (41). Cytokine-induced HLA class I upregulation on β-cells is necessary for cytotoxicity of human CTLs (42). This is a potent step to protect β-cells because once an immune synapse has been formed between a cytotoxic T cell and β-cell there are at least two pathways of cell death to block and protection becomes more difficult (25). We showed that the JAK1/JAK2 inhibitor AZD1480 blocked the upregulation of MHC class I on β-cells in vivo, providing support for both mechanism of action and exposure of the target β-cells to the drug.

We observed a significant drop in insulitis in 12- to 15-week-old NOD mice treated with AZD1480. Reduced stimulation of T cells by antigen/MHC class I, for example when β-cells are all destroyed, results in the efflux of T cells from islets. Previous reports (43,44) have shown that the presence of cognate antigen is required for T-cell homing and retention in islets. Blocking β-cell responses to IFNs also results in reduced expression of the chemokines necessary for immune cell trafficking, and blocking β-cell chemokine expression can protect NOD mice from insulitis and diabetes (45,46). It is likely that AZD1480 had more effects than just those that we observed on β-cells and T cells. Indeed, JAK-STAT signaling is activated by cytokines in many immune cell types, including antigen-presenting cells, macrophages, and natural killer cells, all of which contribute to the pathogenesis of autoimmune diabetes in NOD mice.

Our data provide preliminary justification for testing JAK1/JAK2 inhibitors in type 1 diabetes clinical trials. Because of the low rate of success in translating findings from mice to humans, progress to a clinical trial needs careful consideration and perhaps a multicenter trial using a predefined protocol as has recently been proposed (28,47). Our diabetes reversal experiment followed the same protocol as that described by Gill et al. (28).

Because of their excellent safety profile, selective JAK1/JAK2 inhibitors, including ruxolitinb and baricitinib, have successfully been trialed in individuals with rheumatoid arthritis and alopecia areata (15,17). Although these drugs are very likely to have effects on the immune system, the effects do not appear to be profound enough to cause significant immunosuppression when used clinically. Common side effects of JAK inhibitors are anemia, neutropenia, and thrombocytopenia due to targeting JAK2 downstream of erythropoietin. Despite these side effects, ruxolitinib and baricitinib were well tolerated, including in individuals with diabetic kidney disease, and have generally not resulted in the discontinuation of treatment (14,15,48–50). New inhibitors with increased specificity and improved safety profiles are in development, such as JAK1-specific inhibitors that would reduce the risk of anemia (51). In conclusion, our results present an exciting opportunity to apply drugs that are approved for human use to the prevention and treatment of type 1 diabetes.

Acknowledgments. The authors thank Joshua Szanyi, Aimee Khoo, Lorraine Elkerbout, Rochelle Ayala, Evan Pappas, Michaela Waibel, and Esteban Gurzov (St. Vincent’s Institute) and Vicki Moshovakis and Emma Duff (St. Vincent’s Hospital) for technical assistance and animal husbandry and Dr. Thomas Loudovaris and Lina Mariana (St. Vincent’s Institute) for providing human islets. The authors also thank AstraZeneca for providing AZD1480.

Funding. This work was funded by the National Health and Medical Research Council (NHMRC) of Australia (grants 1037321, 1061961, 1043414, 1080321, and 1105209), the Leona M. and Harry B. Helmsley Charitable Trust, and JDRF (grants 17-2013-514 and 1-SRA-2016-224); fellowships from the NHMRC (M.R.J., A.M.L., and H.E.T.) and JDRF (B.K. and S.I.M.); and a Government of India, Ministry of Science and Technology, Department of Biotechnology Associateship (BT/20/NE/2011; S.M.). This work also received support from the Operational Infrastructure Support Scheme of the Government of Victoria.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. P.M.T., K.L.G., and N.A.S. designed the study, performed experiments, analyzed the data, and wrote and revised the manuscript. M.R.J. contributed to the conception, design, and interpretation of this work; performed experiments; analyzed the data; and critically revised the manuscript. S.M. and S.F. performed experiments, analyzed the data, and revised the manuscript. R.M.S., A.M.L., C.J.B., B.K., T.C.B., and S.I.M. contributed to the conception, design, and interpretation of this work and critically revised the manuscript. T.W.K. and H.E.T. designed the study and wrote the manuscript. H.E.T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.