Unexpected Acceleration of Type 1 Diabetes by Transgenic Expression of B7-H1 in NOD Mouse Peri-Islet Glia

B7-H1 cDNA was derived by RT-PCR from NOD mouse splenocyte RNA and inserted into a GFAP expression cassette (Fig. 1 A) (25). This construct was injected into fertilized NOD ova, generating two independent male founders. One died young, with typical overt type 1 diabetes, before successful breeding. Positional cloning and sequencing indicated that the surviving founder carried the transgene on chromosome 12 (chr12:3302043.0.3202044, Mus musculus, build 37.1) in a noncoding region, not near known NOD type 1 diabetes risk loci. This founder and offspring developed and bred normally but had an immune phenotype (see below). NOD, NOD.scid, NOD-derived GFAP promoter–driven transgenic (GFAP-tg), and GFAP-tg.scid mice were generated and maintained under approved protocols in our vivarium. GFAP-tg.scid resembled NOD.scid in every respect, including normal pancreas histology and confocal immunofluorescence.

NOD B7-H1, from transgenic brain, or wild-type spleen were purified with a B7-H1 antibody (Cedarlane, Bullington, ON, Canada), coupled to an affinity column (GE Healthcare, Mississauga, ON, Canada). A CM5 Biacore chip was used to couple 200 response units of PD-1 (R&D Systems, Minneapolis, MN), and serial dilutions of purified B7-H1 (4.25–100 nmol/l) were injected. BIAevaluation software was used for analysis.

For insulitis quantitation, frozen tissue was processed, hemotoxilin-eosin–stained, and scored as previously described (4). Three-color immunofluorescence was used for confocal imaging of GFAP (in pSC and astrocytes), CD3, and insulin as previously described (4). For GFAP quantification studies, Volocity software was used; 68–100 islets/group were counted. Each individual islet perimeter was identified through islet autofluorescence and traced, and the total islet area was calculated (micrometers squared). The GFAP staining signal (fluoroisothiocyanate) of pSC was quantified within each islet by selective identification of “green” pSC and calculated as a percentage of GFAP-positive staining (in pixels) per total area of each islet, measured in micrometers squared. Image-J software was used to image the steric distribution of pSC loss, comparing GFAP green pixel counts in wild-type and transgenic islets of 8-week-old mice.

Spleen cells were prepared for intravenous injection or for in vitro cultures as previously described (4). Splenocytes (10⁷) from NOD females with recent diabetes onset were injected intravenously into 8-week-old NOD.scid or GFAP-tg.scid recipients either treated or untreated with 200 μg i.p. mouse anti–B7-H1 antibody (R&D Systems) every second day. T-cell proliferation assays of splenocytes from 5- to 13-week-old NOD females were completed and analyzed as previously described (26). For DC antigen presentation analysis, 2 × 10⁴ CD11c⁺ GFAP-tg or wild-type NOD DCs were purified with a CD11c⁺ selection kit (Stem Cell Technologies, Vancouver, BC, Canada) from spleens and cocultured with 2 × 10⁵ T-cells purified with a CD3⁺ selection kit (Stem Cell Technologies) from 10-week-old wild-type NOD mice and pulsed with insulin. For pancreatic lymph node T-cell proliferation assays, 2 × 10⁵ lymphocytes from wild-type NOD or GFAP-tg animals were seeded with 2 × 10⁵ irradiated splenocytes, pulsed with GFAP, processed, and analyzed as previously described (2).

Pancreata were digested with collagenase (20′/37°C), dissociated, and applied to a Ficoll gradient. Islets were hand-picked and cultured in leukocyte complete medium with 18 ng/ml interleukin-2 for 8 days prior to staining (5). Cells were stained for flow cytometry with anti-CD8 or anti-CD11c, CD11b, and/or B7-H1 and gated as indicated (e.g., CD3⁺).

Blood glucose was measured twice weekly in animals ≥8 weeks old; measurements ≥13.8 mmol/l in two consecutive readings were considered diabetic. Insulin and glucose sensitivity and tolerance were measured as described (2).

hemotoxilin-eosin–stained submandibular salivary glands were scored by two blinded observers. Mononuclear foci were scored as “small” at <75 inflammatory cells/section and “large” at >75 inflammatory cells/section.

B7-H1 expression was detected by transgene-specific RT-PCR in neuronal tissues, including brain and sciatic nerve, in pancreas but not in thymus or spleen (Fig. 1,B). Traces of transgene transcripts were also found in colon and salivary glands. B7-H1 protein expression did not completely correlate with transcription profiles, which is not uncommon (27). Total B7-H1 protein expression was similar in transgenic and age- and sex-matched wild-type mice in most tissues examined but comparatively elevated in transgenic pSC cells and brain (Fig. S1, available in an online appendix at http://diabetes.diabetesjournals.org/cgi/content/full/db09-1209/DC1). Confocal immunofluorescence colocalized GFAP and B7-H1 to transgenic pSC cells, with undetectable B7-H1 in wild-type pancreata (Fig. 1,C and D). There was considerable transgene expression throughout the central nervous system in transgenic but not wild-type mice (Fig. 1 E and F). Insular transgene expression in pSC did not generate endocrine malfunction, as analyzed by glucose and insulin sensitivity tests (Fig. S2). Transgene protein expression in colon and salivaries was at best very low and inconsistent in GFAP-tg mice (Fig. S1).

We compared binding affinities of purified wild-type PD-1 to wild-type and transgenic B7-H1 by surface plasmon resonance, with binding kinetics analyzed by 1:1 Langmuir binding isotherms (Fig. 1,G; Table 1). No differences were identified, suggesting that the transgene ligates PD-1 normally. Collectively, these observations argue strongly that GFAP-tg phenotypes reflect B7-H1 costimulatory function in the islet locale rather than nonspecific transgenesis effects—a conclusion further supported by in vivo B7-H1 blocking studies (discussed below).

As shown in Fig. 2,A, transgenic mice progressed more rapidly to type 1 diabetes than wild-type females (P = 0.0017, life tables). More rapid pre-diabetes progression peaked by 19 weeks, with two-thirds of GFAP-tg diabetic compared with one-tenth of wild-type control mice. To begin exploring the mechanism of this counterintuitive phenotype, we quantified the progression of islet inflammation. The early phase of islet infiltration (peri-insulitis) was unaffected in transgenics (P > 0.1 vs. wild type) (Fig. 2,B–D), but by 8 weeks of age, insulitis was much enhanced (Fig. 2,B) and was more invasive in GFAP-tg mice compared with controls (P < 0.001) (Fig. 2,C). Insulitis progression continued to accelerate (Fig. 2,C) up to overt diabetes declaration (P < 0.001) (Fig. 2 D).

To determine whether increased insulitis severity correlated with accelerated pSC loss, a quantitative scoring system was developed. pSC integrity was measured as GFAP pixel-counts/islet cross-section. There was marked variation of islet sizes, but the distribution was similar in 8-week-old transgenic and wild-type mice, wild-type and transgenic NOD.scid mice, and 3-week-old wild-type or transgenic mice (P > 0.1) (Fig. 3,A). However, the GFAP content (percentage of area) was much reduced in 8-week-old transgenics (P < 0.0001) (Fig. 3,B). This loss was autoimmune mediated, given that GFAP-tg.scid (and very young animals) did not have GFAP loss (P > 0.1) (Fig. 3,B). We used three-dimensional imaging to map the spatial distribution of pSC loss. As shown in Fig. 3,C–E, transgenics clearly depleted the pSC envelope by 8 weeks (Fig. 3,E), while pSC integrity was maintained in young (Fig. 3,C) as well as 8-week-old wild-type mice (Fig. 3 D). Thus, several independent approaches, from histology to quantitative imaging, fully supported each other, and the addition of wild-type and transgenic NOD.scid mice added credibility to the notion that the transgenic phenotype reflects pSC impact on pre-diabetic autoimmunity.

Splenocytes from diabetic GFAP-tg donors transferred diabetes to wild-type NOD.scid recipients dramatically faster to more recipients than did wild-type diabetic donor cells (P = 0.023) (Fig. 4,A; life tables), demonstrating that the transgene drives quantitative or qualitative immune changes that enhance the pathogenicity of systemic immune cells. Consistently, transfer of wild-type splenic grafts into transgenic NOD.scid recipients showed intermediate pathogenicity, with similar penetrance but slower pace of type 1 diabetes development in transgenic NOD.scid recipients (P = 0.02). Thus, these transfer data support those from natural history disease development (Fig. 2 A), directly demonstrating type 1 diabetes acceleration in and by the transgenic islet.

The rapid adoptive transfer model facilitated studies of the role of B7-H1 function in wild-type and transgenic mice. Systemic blockade of B7-H1 by antibodies mediated acceleration and higher penetrance of type 1 diabetes development (P = 0.02) (Fig. 4 B). In contrast, antibody blockade had no effects on the enhanced pathogenicity of transgenic grafts, indicating that the transferred pool of cells was no longer B7-H1 dependent (P > 0.1) (Fig. 4C). These results indicate that transgenic pSC cells influence local T-cells at the islet level, which are then recirculated into systemic distribution, refractory to further B7-H1 signals. In transgenics, this change is associated with cognate expansion of CD8⁺ effector T-cells (as discussed below).

To better understand the enhanced pathogenicity of autoreactive T-cell pools in transgenics, we compared in vitro proliferative recall responses to GFAP (pSC-cell specific) and insulin (β-cell specific; B9–23) (Fig. 5). T-cell reactivity to both autoantigens was observed in transgenic and wild-type mice. While insulin autoimmunity amplified, with pre-diabetes progression in both groups equally, transgenic GFAP recall responses peaked early and then declined and in fact differed significantly at every time point measured, with the decline in transgenics a good correlate of earlier pSC demise (Fig. 3,E). Similarly, GFAP responses to the main NOD target epitope, GFAP p253–261 (5), were more pronounced at very low antigen concentrations, suggesting cognate selection of high-affinity T-cell receptors—again consistent with the more massive, early pSC damage (0.1 μg/ml, P = 0.020; 0.5 μg/ml, P = 0.015) (Fig. 5 C). These observations suggest that the pSC transgene enhances the effectiveness of progressive pSC damage through more effective pSC-specific T-cell selection.

These data prompted us to compare T sublineage distribution in transgenic and wild-type mice. The different pre-diabetic tissues analyzed had similar numeric and proportionate T sublineage distributions in both mouse lines (Fig. 6,A and B). However, a striking, transgene-driven phenotype was the islet-exclusive expansion of the proportions (P < 0.05) (Fig. 6 C) and numbers of CD8⁺ intraislet T-cell pool (P < 0.002). This independently supports the above conclusion that the pSC transgene imparts more pathogenic T-cell selection. There were no other changes in the cellular distributions, including B7-H1⁺ dendritic cells (Fig. S3).

Autoimmune priming of (ultimately) pathogenic T-cells is thought to occur at a young age in pancreatic lymph nodes, where dendritic cells present fragments of apoptotic β-cell debris (28). There was not any evidence for more pronounced pSC cell (or β-cell) death in 3-week-old transgenic versus wild-type mice (Fig. 7,A and B), with terminal deoxynucleotidyl transferase–mediated dUTP nick-end labeling–stained thymus tissue serving as a positive apoptosis control tissue (Fig. 7,C). Absent GFAP-specific T-cell responses in pancreatic lymph node cells of the same animals provided a functional read-out for autoantigen unavailability at that time (Fig. 7 D).

Early stages of the pre-diabetes progression program in these NOD females were thus not affected by the transgene, and enhanced number and pathogenicity of local autoreactive T-cells in transgenics likely reflected the local impact of positive costimulation by transgenic B7-H1. To rule out a dysfunction of DCs in this process, we purified DCs from transgenic and wild-type donors 3 weeks of age and used them to present antigen (insulin) to purified pre-diabetic (10 weeks old) wild-type T-cells (Fig. 7 E). As shown, DCs from transgenic and wild-type donors performed as well as antigen-presenting cells.

The NOD mouse develops two genetically independent, penetrant autoimmune disorders: type 1 diabetes and a primary Sjögren syndrome (29). Only if the transgene phenotype exclusively involved local costimulation in the islet would we expect the “the other autoimmune disease” (sialitis) to be unaffected. As shown in Fig. S4, sialitis proceeded at similar rates and with similar features in wild-type and GFAP-tg mice.

Systemic models (blocking or gene deletion) for the study of PD-1 and its ligand, B7-H1, have unequivocally shown dramatic effects on T-cell development (30,,–33), activation in lymphoid tissue (34,35), and modification of effector and regulatory T-cell functions in various tissue sites (36,37). With very few exceptions (20), PD-1 ligation is associated with negative costimulatory function, and an elusive, alternative B7-H1 receptor has been proposed, but not identified, for the exceptions (23,24). We sought to employ the negative costimulatory activity of B7-H1 to protect pSC cells from autoimmune destruction and thus obtain a tool to understand the pSC role in pre-diabetes progression. This failed because B7-H1 transgenic pSC accelerated disease development through selection of CD8⁺-biased, more pathogenic, islet-resident T-cell pools. Our observations were more surprising because overexpression of B7-H1 in β-cells protected NOD mice from diabetes (19). However, a similar transgene in naturally type 1 diabetes–resistant C57BL/6 mice did generate diabetic mice (20). Current information is insufficient to explain these findings, but at the very least, one must conclude that expression of B7-H1 in nonhemopoietic tissue can bias local immune regulation and effector function either positively or negatively.

Irrespective of the molecular mechanisms that determine the outcome of tissue PD-1 ligation, our findings demonstrate an immunologically meaningful, disease-relevant interaction between T-cells and pSC cells in pre-diabetes. In GFAP-tg animals, severe pSC depletion by 8 weeks is the harbinger of rapid β-cell insufficiency. The selective numerical and proportionate expansion of CD8⁺ T-cells was exclusive to the islet locale, which is in line with the lack of major histocompatibility complex class II expression in pSC. The early peak of high-affinity, pSC-specific T-cell autoreactivity indicates that cognate events drive this T-cell expansion. Without concomitant rises in regulatory T-cells, these data delineate a cogent mechanism for type 1 diabetes acceleration, supported by the sharing of type 1 diabetes–associated target autoantigens in pSC and β-cells (2,4).

The enhanced pathogenicity at type 1 diabetes onset, observed in adoptive transfer experiments, is then explained by systemic recirculation of the small islet T-cell pools. In vivo B7-H1 blocking studies suggested that these recirculated cells themselves are no longer influenced by B7-H1. This might actually contribute to their pathogenicity given that endogenous, negatively costimulating B7-H1 otherwise should have lessened the pathogenicity of these T-cells.

Other contributors to rapid β-cell loss likely include depletion of pSC neurotropic factors, a long-recognized pSC cell function (6,–8). Thus, when transgenic and wild-type islets were compared, our data demonstrated that pSC actively modified the type 1 diabetes progression program—in the present case through the selection of more pathogenic anti-pSC T-cell reactivities.

One concern with any transgenic model is the creation or modification of a phenotype through insertional toxicity, postcloning transgene rearrangements, or nonphysiological transgene expression levels. To rule out the possibility that the transgenic B7-H1 induced pSC apoptosis or affected islet physiology independently of T-cells, the transgene was crossed to lymphocyte-deficient NOD.scid mice. On the scid background, spontaneous diabetes was not observed and islets were indistinguishable from NOD.scid controls. In addition, the transgenic phenotype was faithfully transmitted in adoptive transfer studies, indicating its immune nature. Thus, the present observations establish that pSC cell–T-cell interactions directly impact diabetogenesis.

Rapid disease development in GFAP-tg mice reflects the selection of more pathogenic T-cell pools, skewed toward CD8⁺ T-cells. Regulatory T-cells have been described to express the receptor for B7-H1 and PD-1 (37,–39) and might be affected by the transgene expressed in the islet. However, the cellularity of CD4⁺CD25⁺FoxP3⁺ T regulatory cells in GFAP-tg mice was not different by total numbers or proportions. Because of the very small absolute T regulatory numbers in the islet, we cannot exclude the possibility that they might be functionally suppressed by the islet transgene. This would then imply a coincidence of positive and negative costimulation in the same locale.

Collectively, the data now available cohesively support the conclusion that pSC cells are potent modulators of islet inflammation, able to shape the pre-diabetes progression program. Therapies that halt pSC damage should delay or reduce β-cell destruction, and innovative strategies for islet transplants, local β-cell regeneration, or de novo growth may benefit from the provision of pSC cell cografts derived from sural nerve fragments.

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

J.Y. was project leader, involved in all project aspects. H.T., S.W., and G.P. contributed to transfer and in vitro experiments. A.S., P.W., and J.E. generated transgenic mice. H.-M.D. conceived, supervised, and financed the study. J.Y., H.T., and H.-M.D. made major editorial contributions.