Functional Redundancy of CXCR3/CXCL10 Signaling in the Recruitment of Diabetogenic Cytotoxic T Lymphocytes to Pancreatic Islets in a Virally Induced Autoimmune Diabetes Model

C57BL/6 (B6), NOD/ShiLtJ, CD45.1⁺ B6.SJL-Ptprc^a Pep3^b/BoyJ, CXCL10-deficient [B6.129S4-Cxcl10^tm1Adl/J (11)], and CXCR3-deficient animals (B6.129P2-Cxcr3^tm1Dgen/J) mice were originally obtained from The Jackson Laboratory (Bar Harbor, ME) and subsequently bred under specific pathogen-free conditions at our facilities. Transgenic P14 mice expressing a T-cell receptor (TCR) α-chain specific for the H-2D^b–restricted immunodominant LCMV epitope GP33–41 (12) were obtained as a gift from The Scripps Research Institute and maintained in our facilities. For tracking and imaging purposes, we crossed C57BL/6-Tg(ACTB-EGFP)1Osb/J (13) mice to the P14 mice. Mice expressing GP antigen from the LCMV-WE (14) or LCMV-Arm (15) strain under control of the RIP (RIP-GP) were bred to a CXCL10- or CXCR3-deficient background and to a strain with β-cell–restricted GFP expression (16) for imaging purposes. All animal procedures were approved by the Animal Care Committee of the La Jolla Institute for Allergy and Immunology (protocol number AP121-MvH3-0510). LCMV Armstrong clone 53b plaques were purified three times on Vero cells, and viral stocks were prepared by a single passage on BHK-1 cells. LCMV strain WE was a gift from Dr. Juan Carlos de la Torre (The Scripps Research Institute). The experiment in Fig. 4I showing CXCL10 neutralization in the RIP-GP model was performed in the Christen laboratory (Frankfurt am Main, Germany) using the same protocol, antibody reagents, and mouse and virus strains for diabetes induction.

Homogenized spleens from infected animals were incubated at 37°C, 5% CO₂, for 1 h with Vero cell monolayers grown in six-well plates (Costar). The plates were then overlaid with 1% agarose in minimal essential medium 199 (Invitrogen) containing 10% FBS and incubated at 37°C, 5% CO₂, for 5 d. The wells were treated with 25% formaldehyde and stained with 0.1% crystal violet for 2 min. The agarose overlay was removed, and infectious centers were counted. Additionally, viral LCMV stock was used as a positive control.

In the viral experiments, diabetes induction was achieved by infection of LCMV.GP-transgenic recipients with 10⁴ plaque-forming units (pfu) LCMV i.p. or 200 pfu LCMV.WE, where indicated. We recently developed a virus-free adoptive transfer model based on the conventional RIP-LCMV.GP system (17,18). Hereto, TCR-transgenic P14 CD8 T cells are transferred into RIP-LCMV.GP host animals followed by peripheral activation using a peptide/adjuvant mixture. Depending on the host strain and in particular its expression of the GP antigen on pancreatic β-cells, all mice develop diabetes in an acute fashion (RIP-LCMV/WE host) or in 50% of mice in a milder form of diabetes (RIP-LCMV/ARM host). Any effects due to specific viral signaling can be excluded in this variant model, and it should be perceived as a model to analyze the capacity of peripherally activated effectors to home to the site of antigen expression in the islets. Briefly, on day 0, a single-cell suspension of 1.5 × 10⁷ P14 splenocytes was transferred into recipient mice via retro-orbital injection. On days 1 and 3, mice were intraperitoneally injected with 2 mg of GP33 peptide (NH2-KAVYNFATM-COOH; Abgent) in ultrapure water containing 50 μg of CpG adjuvant (Sigma-Aldrich). Finally, on day 6, mice received 500 μg of polyinosinic-polycytidylic acid i.p. (Sigma-Aldrich). Blood glucose measurements were performed with a OneTouch blood glucose meter (LifeScan, Inc.). Mice with values >300 mg/dL were considered diabetic unless otherwise indicated. All two-photon imaging was done on days 7 and 8. Glucose tolerance test involved injection of 2 g/kg glucose i.p. into overnight-fasted animals and blood glucose measurements at the indicated time points.

Anti-CXCL10 antibodies for in vivo neutralization were a kind gift from Dr. Andrew D. Luster and were validated previously (19) or purified from hybridoma supernatant (gift from Andrew D. Luster, Harvard Medical School) and tested by an in-house chemotaxis assay. It was administered in an identical fashion as described by Christen et al. (2) (i.e., 100 μg in 100 μL of PBS per injection, 6 h before infection and on days 1, 2, 4, and 6 after LCMV infection).

Single-cell suspensions were prepared from spleen and pancreas when indicated. Erythrocyte lysis was performed on splenocyte preparations. NRP-V7 tetramer was obtained from Dr. Rusung Tan (University of British Columbia). For intracellular cytokine staining, cells were restimulated with GP33 antigen for 5 h in the presence of brefeldin A (GolgiPlug; BD Biosciences), stained for surface markers, and subsequently fixed and permeabilized (Cytofix/Cytoperm; BD Biosciences) for intracellular staining. All antibodies were obtained from BD Biosciences. Acquisition was performed on an LSRII flow cytometer (BD Biosciences).

For experiments presented in Fig. 1, P14 CD8 T cells were purified by depletion prior to transfer using sheep antirat magnetic DYNAL beads (Invitrogen) in conjunction with rat anti-CD4, anti-B220, anti-CD16/CD32, and anti–I-A/E antibodies (all from BD Biosciences). Purity was verified by flow cytometry and was typically within the 90–95% range. Purified cells were labeled with 5- (and 6-)carboxyfluorescein diacetate succinimidyl ester (CFSE) according to the manufacturer’s instruction,where indicated (Invitrogen).

Pancreata were harvested and snap-frozen in OCT medium (Sakura), and 6-μm consecutive frozen sections were cut, acetone-fixed, and stained with anti-CD8a or anti-CD4 and insulin. Immunohistochemical detection was done using biotinylated secondary antibodies and Avidin-D coupled to horseradish peroxidase or alkaline phosphatase, followed by enzymatic development with diaminobenzidine and Vector Blue chromogens (Vector Laboratories). Immunofluorescent quantitation of T-cell islet infiltration was done in analogous fashion except for the use of Alexa Fluor 488–coupled secondaries and Prolong antifade, 4′,6′-diamidino-2-phenylindole–enriched mounting medium (both from Invitrogen). Images were acquired on a Nikon Eclipse 80i microscope with fluorescent attachment (Nikon), and islets surface, based on 4′,6′-diamidino-2-phenylindole morphology, was calculated using ImageJ (National Institutes of Health). Finally, a technician unaware of sample identity counted the infiltrating T cells using the Cell Counter plugin.

In vivo two-photon imaging of the pancreas was described in detail in Coppieters et al. (18).

Organs were harvested and subjected to RNA isolation using the RNeasy Midi kit (Qiagen), cDNA was prepared using the Quantitect Reverse Transcription kit (Qiagen), and quantitative SYBR Green-based real-time PCR was performed with validated Quantitect primer pairs specific for CXCL10 and 18S ribosomal RNA. Resulting threshold cycle values were used to determine relative expression values using the ΔΔ threshold cycle method. Data in Supplementary Fig. 2C used validated CXCL10 and glyceraldehyde-3-phosphate dehydrogenase housekeeping gene primers from SABiosciences.

Diabetes in the RIP-LCMV.GP diabetes model results from the activation and expansion of T cells in response to LCMV and their subsequent redirection to the β-cells, which transgenically express the viral glycoprotein (14,15). Analyzing effects on diabetogenicity in this model requires differentiation between effects on activation and expansion during the antiviral phase and on effector recruitment to the islets during the diabetogenic phase. To verify that systemic CXCL10 or CXCR3 deficiency does not interfere with antiviral CD8 T-cell functionality, antigen-stimulated cytokine production was analyzed in gene-deficient mice that were infected 8 days prior (Fig. 1A). Analogous frequencies of antigen-experienced, cytokine-producing cytotoxic T lymphocytes (CTLs) were found as compared with wild-type, indicating that CXCL10 or CXCR3 deficiency does not affect the antiviral response. In agreement with these results, complete viral clearance was consistently achieved (Fig. 1B). Likewise, transferred LCMV-specific CD8 T cells responded similarly to infection in CXCL10-deficient hosts as compared with wild-type (Fig. 1C). Intrinsic CXCR3 deficiency within the transferred CD8 T-cell population did not interfere with activation and expansion of the cells (Fig. 1D). Collectively, these data indicate that neither CXCL10 nor CXCR3 deficiency preclude the development of antiviral CTL immunity in this system and thus validate its use for recruitment studies during the diabetogenic phase.

It was shown earlier that CXCL10 is upregulated in the pancreas shortly after LCMV infection (2) and that diabetogenic T cells preferably home to sites containing high levels of this chemokine (20). However, quantitative RT-PCR analysis across various organs harvested after LCMV infection revealed even higher expression in kidney, liver, and spleen (Fig. 1E). Thus, LCMV-induced CXCL10 hyperexpression is not a pancreas-restricted response. This observation thus questions the importance of CXCL10 in the specific recruitment of CTL to pancreatic islets.

It was shown previously that LCMV infection results in the expression of CXCR3 on ∼93% of activated endogenous GP33–41-specific CD8 T cells (2). In order to confirm CXCR3 expression on TRC-transgenic P14 cells, which are of particular relevance to our transfer model data, we performed a detailed kinetic analysis (Supplementary Fig. 1). The outcome clearly shows that CXCR3 upregulation in the transferred P14 population surpasses the frequencies observed in endogenous CD8 T-cell responders after viral infection. These data therefore indicate that the transferred cells should be responsive to CXCL10-mediated chemotaxis and validate the use of the transfer model for our purposes.

In view of published reports (21), we also performed parallel analyses for CCR5 and CCR2 expression across different organs during the development of diabetes in the RIP-LCMV model and the related transfer model (Supplementary Fig. 1). Our data show that CXCR3 expression on the dominant diabetogenic CD8 T-cell population is clearly highest and increases over time in all compartments including the pancreas. CCR2 expression is close to undetectable, whereas CCR5 expression significantly increases, particularly in the pancreas. The finding that CCR5 is overexpressed in the diabetic pancreas in our model is in line with NOD data by Carvalho-Pinto et al. (21), although effector influx per se was not affected in that paper. Of note, we tried to block the CCR5 ligand CCL5 (regulated on activation, normal T cell expressed and secreted) with a neutralizing antibody before in the RIP-LCMV model but found no effect (2). More recent data using CCL5-knockout mice indicate that there is no difference in viral clearance, T1D incidence or islet infiltration (S.L., E.H., U.C., unpublished observations).

We next evaluated whether CXCR3 expression is also present on spontaneously arising, low-affinity diabetogenic CD8 T cells (Supplementary Fig. 2A and B). Hereto, we used the NRP-V7 clone, which is known to react against the islet antigen IGRP in NOD mice in which this clone constitutes a major effector population (22). From our studies, it is clear that around diabetes development, the CXCR3 expression on these low-affinity effectors is increased. Thus, spontaneously occurring, low-affinity diabetogenic clones upregulate CXCR3 as diabetes progresses and, in this regard, mirror the P14 cells used in our present report. We also verified that the CXCR3-binding chemokine of interest, CXCL10, is expressed in the pancreas of NOD mice (Supplementary Fig. 2C). We thus conclude that both ligands (in the target organ) and receptors (on low-affinity effectors) are present within the NOD system, yet a lack of CXCR3 does not confer protection in this model (5). We therefore conclude that it is unlikely that the degree of dependence on CXCR3/CXCL10 signaling required for effector influx into pancreatic islets differs between high- and low-affinity T-cell clones in these diabetes models. Similar data as in the RIP-LCMV model were also obtained with regard to the relative levels of CXCR3 expression versus CCR5 and CCR2 (Supplementary Fig. 2A and B).

Our initial in vivo data under conditions of CXCR3 deficiency suggested that the receptor is not required for diabetes development (Fig. 2A). These data were obtained after intraperitoneal infection with 10⁴ pfu of LCMV, as commonly used in our laboratory. Immunohistochemical staining for CD8 and CD4 further revealed that T-cell recruitment to pancreatic islets was unaffected (Fig. 2B and C).

As we specifically aimed to replicate the dataset obtained by Frigerio et al. (1), we altered our model by performing intravenous infection with 200 pfu of LCMV-WE instead of the LCMV-Armstrong strain. In contrast to their report, we did not observe a potent suppression of >2 weeks, yet a statistically significant delay could be observed (Fig. 2D). This delay was more prominent when a conventional hyperglycemia limit was used (300 mg/dL; P = 0.003; Fig. 2D, top panel) rather than the limit used by Frigerio et al. (1) (upper meter limit of 600 mg/dL; Fig. 2D, bottom panel), which only exhibits borderline significance in our studies (P = 0.04). Our work also shows that >90% of mice eventually become diabetic, regardless of CXCR3 deficiency. We conclude that ablation of CXCR3 does not protect from diabetes development yet offers a modest delay in diabetes progression in the RIP-LCMV model.

Finally, we exploited a recently described virus-free transfer model to study the role of CXCR3 specifically on transferred diabetogenic T-cell effectors (17,18). In brief, naive effector P14 CD8 T cells are activated and expanded in vivo by antigen administration in conjunction with Toll-like receptor–ligating adjuvants and subsequently home to the islets of the RIP-LCMV host. Depending on the background of the host, the model is more or less acute and penetrant (17). It was also confirmed in this study that major histocompatability complex class I is upregulated in the transfer model, which is considered essential for CD8 T cell–mediated β-cell killing (Supplementary Fig. 3). In line with the data from the conventional viral model, we found that CXCR3-deficient CD8 T-cell effectors provoked hyperglycemia with a marginally reduced efficiency, regardless of host background (Fig. 2E and F). Histology data confirmed that CD8 T-cell migration to the islets was generally comparable to controls. We conclude that T cell–intrinsic CXCR3 expression is not required for diabetes development in this model yet modestly influences the course of disease progression.

Diabetes onset and progression in CXCL10-deficient RIP-LCMV mice was identical as in wild-type mice (Fig. 3A). Detailed histological analysis indicated that T-cell recruitment to the pancreatic islets was entirely preserved (Fig. 3B and C). In agreement with this finding, data using CXCL10-deficient hosts in the transfer model described above showed no significant differences (Fig. 3D). Next, we transferred labeled TCR-transgenic CD8 T cells into CXCL10-deficient RIP-LCMV mice and analyzed pancreatic recruitment after LCMV infection by two-photon microscopy as recently reported (23) (Fig. 4). In correspondence with our histology data, it was found that the ability of the effectors to extravasate and migrate through the pancreas tissue remains intact under conditions of CXCL10 deficiency.

Since these data contradicted the important role previously ascribed to CXCL10 in CD8 T-cell homing, we questioned whether timed neutralization, rather than systemic ablation, could indeed mediate protection. The first experiment was performed with CXCL10-neutralizing antibody donated by Dr. Andrew D. Luster (Harvard Medical School), the clone previously assessed in the RIP-LCMV model (2) (Fig. 5A). In this paper, a nonsignificant tendency toward acceleration was noted in the treated animals as compared with control. To evaluate the role on T-cell homing to the islets, we performed immunohistochemical (Fig. 5B and C) and immunofluorescent staining followed by quantification of CD8⁺ and CD4⁺ cell numbers within islets (Fig. 5D–F), which indicated no effect on T-cell accumulation. Normoglycemic mice were subjected to glucose tolerance testing, but no differences were observed (Fig. 5G).

Finally, two independent experiments were simultaneously conducted using the same antibody clone, freshly purified from hybridoma supernatant that was tested in our laboratory for complete blockade of CXCL10-mediated chemotaxis in vitro. The experiment performed in the von Herrath laboratory demonstrated no effect whatsoever (Fig. 5H), whereas the experiment in the Christen laboratory showed a tendency toward deceleration of diabetes development (P = 0.0504; Fig. 5I).

The present series of preclinical experiments was designed to assess whether CXCL10/CXCR3 signaling truly represents a key checkpoint in the recruitment of T cells to pancreatic islets in T1D. Our data reveal that neither chemokine nor receptor is required for full antiviral responsiveness and viral clearance in the LCMV system. In combination with the fact that CXCL10 and CXCR3 have been linked to the development of diabetes development in virus-infected patients (7), the RIP-LCMV viral system therefore appears adequate to study the role of this chemokine axis in T-cell recruitment during autoimmune diabetes.

Two publications previously ascribed a pivotal role to CXCL10/CXCR3 signaling in the RIP-LCMV diabetes model (1,2). Frigerio and coworkers (1) observed a substantial delay in diabetes development in the absence of CXCR3. However, an unconventionally high diabetes threshold was used (600 mg/dL vs. conventional 250–300), and it is uncertain what the clinical correlate thereof is. Our data now indicate that the effect of CXCR3 deficiency exists but is rather subtle and depends on factors such as route of infection, viral strain, and dose. By no means do our experiments indicate that CXCR3 is essential for diabetes development or T-cell recruitment in this model, although a certain contribution does seem plausible. These findings are entirely in agreement with recent data using small-molecule CXCR3 inhibitors in the RIP-LCMV model, which also showed marginal efficacy (6).

Of note, conventional B6 mice lack a functional Cxcl11 gene, another CXCR3 ligand with higher affinity for the CXCR3 receptor than CXCL10 and CXCL9 (24). We checked the history of B6.129S4-Cxcl10tm1Adl/J and B6.129P2-Cxcr3tm1Dgen/J mice, and, as with our RIP-LCMV mice, these animals were backcrossed for numerous generations to the conventional B6 genetic background. It is therefore probable that these strains lack functional CXCL11. Our earlier study in the RIP-LCMV model (2) indeed found very small amounts of CXCL11 transcript in the pancreas. The potential presence of functional CXCL11 in the (C57BL/6 × BALB/c) F1 mice used in the study by Frigerio et al. (1) could theoretically account for the difference in diabetes protection between our respective CXCR3-deficient RIP-LCMV strains. However, Frigerio et al. (1) reported that they “barely detected specific transcripts for this factor (CXCL11) in insulitic islets using quantitative PCR.” We can therefore conclude that it is unlikely that this potential strain-specific difference in CXCL11 expression will account for our divergent findings.

The current study investigated the issue from several different angles, using CXCL10 (never reported) and CXCR3-deficient [as in Frigerio et al. (1) but on a clean B6 background] RIP-LCMV mice to evaluate the impact of genetic deletion. A new virus-free transfer model was used to test the role of T cell–intrinsic CXCR3 signaling, while antibody-mediated CXCL10 treatment was performed to investigate the effect of timed chemokine neutralization. Thus, the present paper offers a thorough reassessment of the true importance of CXCR3/CXCL10 signaling in the RIP-LCMV system and provides conclusive evidence that therapeutic blockade of this axis does not curb autoimmune diabetes as previously reported. This reconciles negative findings from the small-molecule inhibitor study as well as data by others in the NOD model and conveys the important message that antichemokine monotherapy may not offer therapeutic benefit, likely due to the high degree of redundancy.

With regard to CXCL10, previous results using antibody-mediated blockade suggested that >70% of mice may experience lasting protection (2). In an elaborate set of experiments, we now find that the therapeutic potential of CXCL10 inhibition is in fact rather limited, if existent. This finding is corroborated by the generation of a CXCL10-deficient RIP-LCMV strain that showed unaltered diabetes development. The cause for these conflicting antibody neutralization data are not clear, but the problem might potentially lie in the antibody batch used at the time, which for instance might have been contaminated or cross-reactive. We confirmed that the present antibody batch neutralized well in vitro but obviously cannot directly compare with the potency of the batch used >10 years ago. Interestingly, in some experiments, a tendency toward protection was noted while others showed almost matching incidence curves. We infer that laboratory-specific differences in mouse background, vivarium conditions, and viral stocks may account for the lack of a consistently penetrant, minor tendency toward protection. The overarching message, however, is that the effect of CXCR3 or CXCL10 ablation/inhibition is much more subtle in this model than previously assumed.

While CXCL10/CXCR3 signaling may well be essential in in vitro chemotaxis assays, we question whether the abundant, nonexclusive expression of chemokines across the entire organism in response to viral infection critically governs the specific recruitment of T cells to pancreatic islets in vivo. Indeed, we find in this study that upregulation of CXCL10 is not a pancreas-specific phenomenon during disease development. Furthermore, Sarkar et al. (10) recently showed that CXCL10 expression is not restricted to islet lesions in mice and human T1D patients, but is generally expressed across both endocrine and exocrine tissue. It is thus unlikely that β-cells are specifically responsible for the recruitment of CD8 T-cell effectors to pancreatic islets in vivo, as suggested by Frigerio et al. (1). This hypothesis was also tested by Rhode and colleagues (25), who created RIP-CXCL10 mice, which express massive amounts of the chemokine specifically in their pancreatic islets and show spontaneous insulitis and aggravated diabetes. It can thus be assumed that nonphysiological overexpression of CXCL10 results in its domination over other factors and accelerated pathogenesis, while during the natural disease course of T1D, other factors might be at least equally important.

Finally, in the cyclophosphamide-induced NOD mouse model, signaling is not required for the recruitment of diabetogenic effectors (3). Ablation of CXCR3 even leads to acceleration of diabetes in the NOD model system (5). Further supporting this notion are recent data demonstrating that prolonged, systemic CXCR3 blockade does not affect T-cell recruitment and diabetes development in the RIP-LCMV model (6). It can thus be inferred that there is no universal role for this chemokine axis in T-cell trafficking to pancreatic islets.

In conclusion, our present results challenge the notion that the CXCL10/CXCR3 chemokine axis is a potential target for therapeutic neutralization in recent-onset T1D. We argue that blockade of a single chemokine may be insufficient to abrogate T-cell recruitment to islets and prevent β-cell decay.

This study was supported by National Institutes of Health/National Institute of Allergy and Infectious Diseases Grant R01-DK-66346-05 (to M.G.v.H.).

K.T.C. and M.G.v.H. are currently employed by Novo Nordisk. No other potential conflicts of interest relevant to this article were reported.

K.T.C. performed and designed experiments, interpreted data, and wrote the manuscript. N.A., P.P.P., C.B.J., A.W., S.L., and E.H. performed experiments. U.C. designed experiments and interpreted data. M.G.v.H. designed experiments, interpreted data, and wrote the manuscript. M.G.v.H. 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.

The authors thank Mira Bel Hani and Yulia Manenkova (both of the La Jolla Institute for Allergy and Immunology) for technical assistance during revision of the manuscript. The authors also thank Dr. Andrew D. Luster (Massachusetts General Hospital, Harvard Medical School) for the donation of anti-CXCL10 antibody and Carol P. Leary (Massachusetts General Hospital, Harvard Medical School) for coordination of preparation and shipping.