Extracellular High-Mobility Group Box 1 Acts as an Innate Immune Mediator to Enhance Autoimmune Progression and Diabetes Onset in NOD Mice

NOD/LTJ (H-2^g7), C57BL/6J (H-2^b), and SWR/J (H-2^q) mice were purchased from The Jackson Laboratories (Bar Harbor, ME) and housed in the specific pathogen-free facility. All studies were carried out in compliance with Medical College of Georgia Animal Care and Use Committee guidelines.

Recombinant HMGB1 (rHMGB1) was expressed and purified as reported and then used to raise neutralizing antibodies in five rabbits (Covance, Denver, PA) (23). The neutralizing activity for each antibody was determined by assay of tumor necrosis factor-α (TNF-α) release after stimulating macrophages with rHMGB1. Antibodies with neutralizing effect were first affinity-purified by protein-A columns and then passed through polymyxin B beads to remove possible contaminated endotoxin.

NIT-1 cells (3 × 10⁶) were treated either with cytokines (10 units/ml interleukin [IL]-1β, 100 units/ml TNF-α, and 100 units/ml interferon-γ [IFN-γ]; R&D Systems, Minneapolis, MN) for 72 h or with UV light (1 min) or 4 μmol/l camptothecin (Calibiochem, La Jolla, CA). NIT-1 cell apoptosis and secondary necrosis after treatment were confirmed using a TUNEL assay kit (Molecular Probes, Carlsbad, CA). Necrosis was induced by three cycles of freeze-thaw processes. Equal volumes of concentrated supernatants after each treatment were analyzed by Western blots for HMGB1 passive release (23). Similarly, primary islets were hand-picked under a dissection microscope after digestion and then dissociated into single cells for the same assay (24). For detection of HMGB1 active secretion, conditioned media from dendritic cells or macrophages were concentrated and analyzed as above.

The amount of TNF-α in the culture medium was determined using a sandwich ELISA kit (eBioscience, San Diego, CA). Intracellular Foxp3 or IFN-γ staining was carried out as reported (23). Flow cytometry data were analyzed using Quest v3.3 software (BD Bioscience, San Jose, CA).

Bone marrow cells were flushed from femurs with RPMI 1640 supplemented with 10% FCS. Cells (5 × 10⁵/ml) were plated in 150-mm Petri dishes supplemented with 5 ng/ml granulocyte macrophage–colony-stimulating factor and 1 ng/ml IL-4. Culture media were changed on days 4 and 7, stimulated with 0.5 μg/ml LPS (Sigma, St. Louis, MO) or 20 μg/ml rHMGB1 for 24 h, and then harvested on day 10 for flow cytometry. Splenic dendritic cells were purified using a mouse dendritic cell enrichment kit (StemCell, Seattle, WA). All antibodies used for flow cytometry were purchased from BD Biosciences (San Diego, CA). NOD dendritic cell allo-stimulatory capability was determined by mixed lymphocyte reaction as reported previously (25).

NOD bone marrow–derived dendritic cells (BMDCs) or macrophages were fixed with formaldehyde and permeabilized with saponin. After blocking with 10% BSA, the cells were sequentially incubated with a monoclonal HMGB1 antibody (Abnova, Taiwan) and a fluorescein isothiocyanate (FITC)-conjugated secondary antibody. Nuclei were counterstained with propidium iodide. The cells were visualized on a LSM510 confocal microscope (Zeiss, Oberkochen, Germany). HMGB1 immunostaining in the pancreatic sections was carried out as previously described (23).

Eight- or 12-week-old NOD female mice were used for the study. The HMGB1 neutralizing antibody (600 μg/mouse) was intraperitoneally injected into each mouse once a week starting from the indicated age until 25 weeks of age. Mice that received the same amount of normal rabbit IgG (Sigma) that underwent the same purification process served as controls. For monitoring diabetes incidence, the mice were housed in the Medical College of Georgia SPF facility for up to 35 weeks of age. Urine glucose was checked every other day with Diastix strips (Bayer, Elkhart, IN) when the mice reached 14 weeks of age. Once urine glucose was detected, the mice were tested for blood glucose, and they were diagnosed with diabetes when two consecutive blood glucose readings were >250 mg/dl.

For monitoring insulitis, the selected mice were killed at 12, 15, and 18 weeks of age, respectively. Pancreatic sections were stained with hematoxylin-eosin using the established techniques (26). Insulitis severity was scored in a blinded fashion by two examiners using the following criteria: 0, intact islet, no cellular infiltrates; 1, peri-insulitis; 2, moderate insulitis, <50% of the islet was infiltrated; 3, severe insulitis, ≥50% of the islet was infiltrated. Eight mice (>200 islets) were examined at each time point for each study group.

Diabetes incidence curves were generated by the Kaplan-Meier method. The differences of diabetes incidence between groups were determined using the log-rank (Mantel-Cox) test. χ² test was used to determine the difference for insulitis severity at each time point. Comparisons between groups for flow cytometry, cytokine production, and intracellular cytokine expression were accomplished by one-way ANOVA using SPS 11.5 for Windows. Data are presented as means ± SD. P < 0.05 was considered statistically significant.

rHMGB1 was first purified using the Ni-NTA affinity columns followed by weak cation exchange chromatography. The purified protein was further passed over polymyxin B columns to remove any contaminated endotoxin. The purity of rHMGB1 was high, as determined on SDS-PAGE (Supplemental Fig. S1A, which is detailed in the online appendix [available at http://dx.doi.org/10.2337/db07-1499]). Contamination of endotoxin was <60 pg/μg protein.

A proportion of the purified rHMGB1 was used to raise antibodies in five rabbits (Covance). Antibodies produced from each rabbit showed high specificity and titers against HMGB1, as shown on the Western blots (Supplemental Fig. S1B). RAW264.7 cells were then used to test the neutralizing effect. A small scale of each antibody was purified by protein-A columns and then passed through polymyxin B beads to remove potential contaminated endotoxin. rHMGB1 is potent in stimulating RAW264.7 cell activation in a dose-dependent manner. When 20 μg/ml rHMGB1 was added into the cultures, RAW264.7 cells secreted 18-fold higher TNF-α (Fig. 1A). Next, 25 μg/ml of each purified antibody was added into the cultures 1 h before rHMGB1 stimulation. Only one antibody showed a strong neutralizing effect, which reduced TNF-α production by twofold (P < 0.001), whereas the rest of antibodies showed either weak or undetectable neutralizing effect, and the control rabbit IgG failed to inhibit the stimulatory effect, indicating the specificity of the neutralizing effect (Fig. 1A).

HMGB1 has been shown to be potent in stimulating dendritic cell activation in humans, C57BL/6, and BALB/c mice (27–29), but its effect on NOD dendritic cells remains unexplored. To this end, we first purified the antibody that showed the strongest neutralizing effect. To stimulate day-9 culture of NOD BMDCs for 24 h, 20 μg/ml rHMGB1 was used. It was found that rHMGB1 significantly stimulated NOD BMDC maturation, as manifested by increased surface marker expressions shown in Fig. 1B. To confirm this observation, 25 μg/ml neutralizing antibody was added into the cultures 1 h before rHMGB1 stimulation. As expected, addition of neutralizing antibody significantly inhibited rHMGB1-induced dendritic cell maturation, and as a result, they showed significantly lower surface marker expressions than the cells with rHMGB1 treatment alone (Fig. 1B). Consistently, these dendritic cells had significantly less allo-stimulatory capacity than those dendritic cells treated with rHMGB1 alone (Fig. 1C).

Because HMGB1 can be passively released from damaged cells, we sought to examine whether it can also be released from damaged β-cells. We first checked HMGB1 passive release in NIT-1 cells, a NOD-derived β-cell line. To induce apoptosis, 3 × 10⁶ NIT-1 cells were treated with combination of IL-1β, TNF-α, and IFN-γ. In the normal condition, NIT-1 cells showed >92% viability, whereas >73% of the cells became apoptotic after cytokine treatment, of which >79% of the apoptotic cells were undergoing secondary necrosis (Fig. 2A). The culture supernatants were then collected and concentrated by the Centricon 10-kDa filter devices (Millipore, Billerica, MA). Equal volume of each supernatant was used for Western blot analysis of HMGB1 passive release. NIT-1 cell lysates and supernatant from necrotic NIT-1 cells were used as positive controls (Fig. 2B, lanes 1 and 5). HMGB1 was present in the supernatants of NIT-1 cells after cytokine (Fig. 2B, lane 2), UV light (Fig. 2B, lane 3), and camptothecin (Fig. 2B, lane 4) treatment, whereas it was undetectable in the control supernatant (Fig. 2B, lane 6).

Next, we pretreated NIT-1 cells with 100 μmol/l Z-VAD-FMK (Calbiochem), a broad-spectrum caspase inhibitor, before cytokine treatment. Z-VAD-FMK significantly prevented cytokine-induced NIT-1 cell apoptosis (data not shown) and resulted in significantly lower HMGB1 passive release than the cells treated with cytokine alone (Fig. 2C, lane 3 vs. 1). We further investigated HMGB1 passive release from apoptotic primary islet cells, and similar results were obtained as shown in Fig. 2D.

In addition to passive release, HMGB1 can also be secreted by activated immune cells, such as dendritic cells (28,30). To check the capacity of NOD dendritic cells secretion of HMGB1, we first examined HMGB1 subcellular localization in NOD BMDCs. HMGB1 was stained with a monoclonal antibody followed by a FITC-labeled secondary antibody. Nuclei were counterstained by propidium iodide. Before stimulation, HMGB1 was solely localized in the nucleus, while a large proportion of HMGB1 translocated into the cytoplasm after LPS or TNF-α/IFN-γ stimulation, suggesting HMGB1 active secretion (Fig. 3A). Western blot analysis of culture supernatants confirmed NOD BMDC secretion of HMGB1 in a time-dependent manner, with the highest levels observed after 24 h of stimulation (Fig. 3B). We further stained for BMDCs with propidium iodide and annexin V and failed to detect a significant difference for cell apoptosis and secondary necrosis before/after cytokine stimulation (data not shown). These results excluded the possibility that the detected HMGB1 in the BMDC culture supernatants originated from damaged BMDCs. To further confirm our results, we performed similar studies in macrophages, and consistent results were obtained (data not shown).

To explore HMGB1 active secretion by NOD islet infiltrated immune cells, we performed in situ HMGB1 immunostaining using pancreatic sections originated from NOD mice with insulitis (14 weeks old). HMGB1 (visualized by diaminobenzidine) was exclusively localized in the nuclei of cells in the normal islets (Fig. 3C, left). HMGB1 was also localized in the nuclei in the majority of cells of islets with insulitis. However, a large proportion of HMGB1 had translocated into the cytoplasm in some of the infiltrated cells (Fig. 3C, ring-like cells indicated by arrows), suggesting HMGB1 active secretion by infiltrated immune cells. To confirm this notion, we double-stained pancreatic islets for HMGB1 (green) and CD11c (red) and found that some of these ring-like cells were CD11c⁺ dendritic cells (Fig. 3C, inset). Taken together, our data suggest that during autoimmunity, HMGB1 can be either passively released from damaged β-cells or actively secreted by islet infiltrated immune cells, such as dendritic cells, which could then enhance autoimmune progression.

To determine the implication of HMGB1 in autoimmune progression, we first performed studies in the early stage of autoimmunity in 8-week-old NOD mice. Forty-one mice were selected for HMGB1 neutralizing antibody treatment, 17 of which were housed for monitoring diabetes incidence, and the rest were used for monitoring insulitis. For controls, 36 mice were treated with normal IgG, among which 12 were housed for monitoring diabetes incidence, and 24 were used for examining insulitis progression. The mice were treated once a week starting from 8 to 25 weeks of age and housed up to 35 weeks of age for diabetes onset. Diabetes incidence was plotted as Kaplan-Meier survival curves. Of important note, HMGB1 antibody treatment significantly reduced diabetes incidence (Fig. 4A, 47 vs. 75%, P < 0.01). Furthermore, the treatment significantly delayed the onset of diabetes. In average, the age for onset of diabetes in HMGB1 antibody–treated mice was 28.7 ± 3.4 weeks, whereas the control IgG–treated mice was only 18.4 ± 3.1 weeks (P < 0.0001).

Next, we checked insulitis severity. Eight mice were killed for each group at each time point, and a total of three time points (12, 15, and 18 weeks of age) were examined. Sections of each pancreas were stained with hematoxylin-eosin, and insulitis was scored as described above. As can be seen in Fig. 4B, the extent of insulitis in HMGB1 antibody–treated mice was significantly less severe than the control IgG–treated mice examined at 12 (P < 0.0005), 15 (P < 0.0006), and 18 (P < 0.0001) weeks old. These data demonstrate that blockade of HMGB1 prevented insulitis progression.

To examine the implication of HMGB1 in the late stage of autoimmunity, we performed a prevention study in 12-week-old NOD mice, because previous reports suggest that autoimmune infiltration at this stage has already progressed for at least 5 weeks (31). As above, the mice were treated with HMGB1 antibody once a week starting from 12 to 25 weeks of age. Consistently, HMGB1 antibody treatment significantly decreased diabetes incidence (Fig. 4C, 33 vs. 78%, P < 0.01). However, there was no significant delay for diabetes onset (18.3 ± 5.8 vs. 20.9 ± 5.5 weeks) as observed in 8-week-old NOD mice. All together, blockade of HMGB1 significantly inhibited insulitis progression and, as a result, decreased diabetes incidence in NOD mice.

To dissect the mechanisms underlying the protective effects, we first examined the impact on dendritic cells. Dendritic cells pooled from pancreatic lymph nodes (PLNs), inguinal lymph nodes (ILNs), and spleens of four mice from each study group were used for the study, and three pools were analyzed. PLN- or ILN-derived dendritic cells were analyzed without pre-enrichment to prevent any loss of potentially important cells. We gated CD11c⁺ cells and analyzed dendritic cell subpopulations by CD11c and CD11b expression levels. Interestingly, HMGB1 antibody treatment significantly reduced PLN CD11c⁺⁺CD11b⁺ dendritic cells (Fig. 5A, 14.38 ± 1.61 vs. 8.25 ± 2.01%, P < 0.01). Because CD11c⁺⁺CD11b⁺ dendritic cells are probably implicated in islet antigen presentation to autoreactive T-cells (32), we then examined their maturation status by analyzing surface marker expressions. Surprisingly, HMGB1 antibody treatment not only reduced PLN CD11c⁺⁺CD11b⁺ dendritic cells but also suppressed their maturation, as shown in Fig. 5B. We next did similar study for PLN and ILN dendritic cell subpopulations in HMGB1 antibody–treated animals. Only very few CD11c⁺⁺CD11b⁺ dendritic cells were present in the ILN; however, the number and maturation status for ILN CD11c⁺CD11b⁺ dendritic cells were significantly lower than that in the PLN (15.33 ± 1.49 vs. 22.09 ± 2.22%, P < 0.05; Supplemental Fig. S2). We finally compared ILN dendritic cell subpopulations between HMGB1 antibody–and control IgG–treated animals and failed to detect significant differences in terms of subpopulations and maturation status (data not shown).

Unlike PLN dendritic cells, HMGB1 antibody treatment did not show significant effect on splenic dendritic cell maturation (Supplemental Fig. S3). However, the treatment significantly inhibited splenic dendritic cell allo-stimulatory capability (Fig. 6A). To dissect its possible mechanism, we analyzed splenic dendritic cell subpopulations based on CD11c and CD8α expressions. Interestingly, antibody treatment significantly increased the number of splenic CD11c⁺CD8α⁺ dendritic cells (23.4 ± 2.8 vs. 11.6 ± 1.0%, P < 0.001; Fig. 6B). Because previous studies suggest that they are probably a subset of tolergenic dendritic cells (33), the reduced allo-stimulatory capability of splenic dendritic cells could be caused by the increase of CD11c⁺CD8α⁺ dendritic cells.

We next examined the effect of antibody treatment on Treg production. CD4⁺ T-cells pooled from PLNs and spleens were further analyzed for Foxp3 expressions. It was found that the proportion of CD4⁺Foxp3⁺ T-cells in the PLN CD4⁺ T-cells examined was significantly higher in HMGB1 antibody–treated mice than the control mice (4.85 ± 0.20 vs. 2.89 ± 0.15%, P < 0.05; Fig. 6C). In contrast, there was no difference for the splenic Tregs between the two groups of mice (data not shown).

To dissect the impact of HMGB1 antibody treatment on T-cell activation, PLN or splenic CD4⁺ and CD8⁺ T-cells were analyzed by flow cytomety for activation markers, such as CD69, CD62L^low, and CD44. To our surprise, antibody treatment showed no effect on the maturation status for CD4⁺ or CD8⁺ T-cells (data not shown). We then checked CD4⁺ and CD8⁺ T-cell subpopulations. Unexpectedly, antibody treatment significantly increased PLN CD8⁺ T-cells (16 ± 1.9 vs. 11.96 ± 2.5%, P < 0.05). Next, we examined the proportion of CD4⁺IFN-γ⁺ (Th1) and CD8⁺IFN-γ⁺ (Tc1) cells. PLN or splenic CD4 or CD8 T-cells were intracellularly stained for IFN-γ expression as described above. As above, a significantly higher proportion of PLN CD8⁺IFN-γ⁺ (Tc1) cells was observed in antibody-treated animals (13.85 ± 2.87 vs. 8.75 ± 1.32%, P < 0.01; Fig. 7). A similar trend was also observed in the splenic T-cells examined but did not reach statistical significance (17.69 ± 2.11 vs. 15.02 ± 1.69%). On the contrary, there was no significant difference detected for the proportion of PLN or splenic CD4⁺IFN-γ⁺ (Th1) cells between the two groups of mice.

Hmgb1 probably originated >500 million years ago before the split between the animal and plant kingdoms. It is among the most evolutionarily conserved proteins in the eukaryotic kingdom and shares 99% amino acid identity between rodents and humans. Recently, we have demonstrated a pivotal role for HMGB1 in the initiation and progression of allograft rejection by acting as an innate alarmin. Because allograft rejection and type 1 diabetes–associated autoimmune response share many similar features, we reasoned that HMGB1 could also be implicated in type 1 diabetes pathogenesis. Additionally, the nature of HMGB1 acting as an innate “danger signal” made it a perfect molecule for us to dissect the role of innate immunity in type 1 diabetes development.

We first demonstrated the potency for HMGB1 in inducing NOD dendritic cell and macrophage activation, and we investigated the origin of extracellular HMGB1 during autoimmunity in NOD mice. HMGB1 passive release from necrotic cells has been demonstrated in the previous studies (34). Recent studies also support that HMGB1 passive release is a feature of apoptosis at least in some cell types and likely occurs during late apoptosis (35,36). β-Cells do not seem to be exceptional in this regard, as shown in our study. Given that apoptosis and secondary necrosis may be the dominant form of β-cell death during insulitis course (37), we induced β-cell death by cytokines, UV, and camptothecin and then checked HMGB1 release. More than 79% of apoptotic β-cells were undergoing secondary necrosis after cytokine treatment, which resulted in HMGB1 passive release (Fig. 2). We also demonstrated NOD BMDC and macrophage secretion of HMGB1 after stimulation (Fig. 3A), which resembled the observations obtained from human dendritic cells. In situ immunostaining further indicated HMGB1 secretion by islet- infiltrated immune cells, such as dendritic cells (Fig. 3C). These findings suggest that extracellular HMGB1 in the pancreatic islets during autoimmunity could originate from both damaged β-cells and infiltrated immune cells, such as dendritic cells.

The presence of HMGB1 in the insulitis lesion suggests its importance in type 1 diabetes pathogenesis. To address this question, we first treated 8-week-old NOD mice with an HMGB1 blocking antibody. As expected, blockade of HMGB1 not only significantly prevented insulitis progression and decreased diabetes incidence but also delayed the onset of diabetes (Fig. 4). Similar results were also obtained for the studies in 12-week-old NOD mice. However, blockade of HMGB1 in 12-week-old NOD mice failed to delay the onset of diabetes, indicating that it is more difficult to block autoimmunity once it progressed to the late stage. However, this phenomenon may also be associated with shorter blocking antibody treatment, because blocking antibody administration did not start before 12 weeks of age and was stopped once the mice reached 25 weeks of age.

Given the critical role of dendritic cells in type 1 diabetes–associated autoimmunity, we first checked the impact for blockade of HMGB1 on dendritic cells, the most potent APCs known today. It was found that antibody treatment reduced the number and maturation of CD11c⁺⁺CD11b⁺ dendritic cells in the PLN (Fig. 5A and B). This observation is consistent with the functionality of these dendritic cells because they are probably associated with autoantigen presentation to naïve T-cells in the PLN (32). We further noticed higher number and more matured CD11c⁺CD11b⁺ dendritic cells in the PLN than that in the ILN in HMGB1 antibody-treated animals (Supplemental Fig. S2), suggesting that they might also be implicated in autoantigen presentation. Of note, HMGB1 antibody treatment apparently did not have an effect on these dendritic cells, because both HMGB1 antibody–and control IgG–treated animals showed similar characteristics for these dendritic cells. Our studies also suggested a role for HMGB1 regulation of splenic dendritic cells. After blockade of HMGB1, the mice showed a skewed population of CD11c⁺CD8a⁺ dendritic cells (Fig. 6B), which have been suggested to be responsible for inducing peripheral tolerance to tissue-associated antigens (33). Consistently, splenic dendritic cells originated from antibody-treated mice had a decreased allo-stimulatory capability (Fig. 6A).

We also noticed that HMGB1 antibody treatment resulted in increased CD4⁺Foxp3⁺ Tregs in the PLN (Fig. 6C). The basis for this observation is not clear from our studies. Previous studies in other model systems have shown that a subset of dendritic cells can induce Treg production (38–40). Therefore, blockade of HMGB1 may modulate dendritic cells to an appropriate maturation state that can induce Treg development. Also, blockade of HMGB1 may have an effect on Treg migration, which leads to Treg accumulation in the PLN.

Of note, blockade of HGMB1 did not show an observable effect on T-cell activation. However, blockade of HMGB1 unexpectedly increased the number of total CD8⁺ T-cells in the PLN. Consistent with this observation, the proportion of CD8⁺IFN-γ⁺ cells (Tc1) in the PLN and spleen was higher in antibody-treated mice compared with the control mice (Fig. 7). The exact mechanism responsible for this observation remains unclear. Based on the fact that HMGB1 antibody treatment significantly inhibited insulitis progression characterized by the reduction of islet-infiltrated immune cells (Fig. 4B), this phenomenon could be associated with retarded migration of activated autoreactive T-cells from PLN into the pancreatic islets after blockade of HMGB1. In fact, a similar result was observed in the experimental autoimmune encephalomyelitis model. Yan et al. (41) noticed that after blockade of receptors for advanced glycation end products, a co-receptor for HMGB1 prevented the entry of encephalitogenic T-cells into the central nervous system. Taken together, these results indicate that HMGB1 could affect the migration of activated autoreactive T-cells into the target organ.

In summary, we demonstrated strong evidence indicating the involvement of HMGB1 in type 1 diabetes pathogenesis. Blockade of extracellular HMGB1 significantly inhibited insulitis progression and diabetes development in NOD mice. The mechanisms underlying this protective effect are associated with the inhibition of inflammation and the regulation of autoimmune response.

C.-Y.W. has received grants from the American Diabetes Association and the Juvenile Diabetes Research Foundation International. J.Z. is an international graduate exchange student from the Tongji Medical College, Huazhong University of Science and Technology, PR China.