Anti-high mobility group box-1 (HMGB1) antibody inhibits hemorrhage-induced brain injury and improved neurological deficits in rats

As one of the most lethal stroke subtypes, intracerebral hemorrhage (ICH) is acknowledged as a serious clinical problem lacking effective treatment. Available evidence from preclinical and clinical studies suggests that inflammatory mechanisms are involved in the progression of ICH-induced secondary brain injury. High mobility group box-1 (HMGB1) is a ubiquitous and abundant nonhistone DNA-binding protein, and is also an important proinflammatory molecule once released into the extracellular space from the nuclei. Here, we show that treatment with neutralizing anti-HMGB1 mAb (1 mg/kg, i.v. twice) remarkably ameliorated ICH-injury induced by local injection of collagenase IV in the striatum of rats. Administration of anti-HMGB1 mAb inhibited the release of HMGB1 into the extracellular space in the peri-hematomal region, reduced serum HMGB1 levels and decreased brain edema by protecting blood-brain barrier integrity, in association with decreased activated microglia and the expression of inflammation-related factors at 24 h after ICH. Consequently, anti-HMGB1 mAb reduced the oxidative stress and improved the behavioral performance of rats. These results strongly indicate that HMGB1 plays a critical role in the development of ICH-induced secondary injury through the amplification of plural inflammatory responses. Intravenous injection of neutralizing anti-HMGB1 mAb has potential as a novel therapeutic strategy for ICH.

HMGB1 is involved in a diverse range of CNS diseases, including ischemic brain infarction, traumatic brain injury, Parkinson's disease and neuropathic pain [11][12][13][14][15] . To trigger the inflammation, the secreted HMGB1 stimulates plural receptors i.e., the receptor for advanced glycation end products (RAGE) and toll-like receptor-2 (TLR-2) and TLR-4, which are expressed in peripheral macrophages and vascular endothelial cells as well as microglia and neurons in the central nervous system 16,17 . Interestingly, the administration of anti-HMGB1 neutralizing mAb has been shown to protect the BBB and to inhibit the inflammation cascade in rat models of middle cerebral artery occlusion/reperfusion-induced infarction and fluid percussion-induced traumatic brain injury [11][12][13] . The recent studies also reported the increase in HMGB1 levels in peri-hematomal regions in subacute phase after ICH in rats [18][19][20] , however, there was little information about the acute dynamics of HMGB1 in the core area after ICH. Moreover, whether anti-HMGB1 mAb can also provide neuroprotective effects in a rat model of ICH remains to be seen.
In the present study, we demonstrated that anti-HMGB1 mAb remarkably ameliorated ICH injury induced by local injection of collagenase IV in the striatum of rats, and this effect was associated with a decrease in activated microglia and astrocytes and suppression of the expression of inflammation-related factors. In addition, the treatment with anti-HMGB1 improved neurological function, which may provide a new approach to potentially reduce ongoing edema and improve the neurological outcome after ICH.

Results
Effects of anti-HMGB1 mAb on HMGB1 levels in the injured brain after ICH. We confirmed that the size of the hematoma in the control and anti-HMGB1-treated rats was the same based on the measurement of hemoglobin content in each group at 24 h after ICH (Fig. 1d).
Brain striatum samples of 3 × 3 × 3 mm 3 were excised, and HMGB1 levels were analyzed by Western blot (Fig. 1a). HMGB1 levels were decreased significantly 24 h after ICH in the control IgG-group compared with the sham group (Fig. 1b). However, anti-HMGB1 mAb administration markedly inhibited the ICH-induced decrease in HMGB1 levels compared with that in the control IgG group (Fig. 1b).
Effects of anti-HMGB1 mAb on HMGB1 levels in the plasma after ICH. The determination of plasma levels of HMGB1 in rats treated with control IgG revealed a clear increase in HMGB1 levels at 24 h after ICH (Fig. 1c). This increase was strongly inhibited by intravenous administration of anti-HMGB1 mAb.
Anti-HMGB1 mAb attenuates BBB disruption. We chose two time points, 6 h and 3 days after ICH, to evaluate the permeability of BBB, since it has been reported that brain edema persists for at least 3 days post-ICH 21 . Since BBB disruption very likely contributes to brain edema, changes in BBB integrity were determined by an Evans blue dye leakage assay after 6 h and 3 days, respectively. The concentration of Evans blue dye in the ipsilateral cerebrum of the control IgG group was significantly greater than that of the sham group after 6 h, indicating that ICH causes BBB disruption of the ipsilateral cerebrum (data not shown). However, the anti-HMGB1mAb group exhibited a significantly lower dye concentration than the control IgG group in the ipsilateral cerebrum at 6 h and even on day 3 post-ICH, indicating that anti-HMGB1 mAb significantly reduces the ICH-induced BBB disruption ( Fig. 1e and f).
Anti-HMGB1 mAb reduces brain edema. On day 3 post-ICH, the ipsilateral brain water content in the control mAb-treated group was 81.28 ± 0.22%, significantly larger than that in the sham group (79.03 ± 0.08%) (Fig. 1g). Anti-HMGB1 mAb significantly reduced the water content of the ipsilateral hemisphere. These findings indicate that anti-HMGB1 mAb treatment significantly reduced the cerebral edema on day 3.
Histological studies on the effects of anti-HMGB1 mAb. Hematoxylin-eosin staining of brain sections from the control mAb-treated rats revealed diffuse hyperchromatic cells in the perihematoma (about 200 μ m from the edge of the hematoma) and dentate gyrus in the hippocampus regions 24 h after ICH (Fig. 2a). In contrast, similar cells were rarely observed in the same areas in the rats treated with anti-HMGB1 mAb.
Release of HMGB1 from neurons, astrocytes, and microglia. The localization of HMGB1 in the rat brain after collagenase-induced ICH was observed by immunofluorescence. As shown in previous studies 12,13 , in the sham group, HMGB1 was localized in the nuclear compartment of cells including neurons (Fig. 2b). In the peri-hematomal regions, the nuclear immunoreactivities of HMGB1 were decreased significantly in most of the MAP-2-positive neurons 24 h after ICH induction (Fig. 2b). HMGB1 immunoreactivities completely disappeared in a small population of neurons (Fig. 2b), while most of the neurons exhibited lower levels of HMGB1 in the nuclear compartment along with dot-like immunoreactivity in the cytosolic and extracellular compartment (Fig. 2b). The intravenous administration of anti-HMGB1 mAb significantly inhibited the above-described translocation of HMGB1 (Fig. 2b).
There were two types of GFAP-positive astrocytes in peri-hematomal regions, HMGB1-negative (Fig. 3a) and HMGB1-positive cells (data not shown), suggesting that HMGB1 was released from both astrocytes and neurons. The number of HMGB1-negative astrocytes was reduced by the treatment with anti-HMGB1 (Fig. 3a). Thus, anti-HMGB1 strongly inhibited the translocation of HMGB1 in astrocytes.
In contrast to neurons and astrocytes, the immunoreactivities of HMGB1 in the nuclei of microglia in the peri-hematomal regions were retained in the control IgG-treated group, and there was no difference in the ratio of HMGB1-positive microglia against Iba1-positive cells among the sham, control IgG-treated and anti-HMGB1-treated groups (Fig. 3b). On the other hand, the number of Iba1-positive cells and the immunofluorescence intensity in a specific area were significantly increased in the control IgG-treated group compared to the sham group, suggesting that the activation of microglia largely occurred 24 h after ICH (Fig. 3c). The increase of the activation was strongly inhibited by treatment with anti-HMGB1 mAb (Fig. 3c).

Figure 1. HMGB1 mobilization under ICH and effect of anti-HMGB1 mAb on HMGB1 dynamics and BBB permeability after ICH. (a)
Cerebral bleeding was induced by injection of 0.03 U bacterial type IV collagenase into the striatum, and the resultant bleeding areas with a volume of 3 × 3 × 3 mm 3 (as indicated by the white square in the picture) were sampled at 24 h after ICH for western blotting to determine brain HMGB1 levels. The representative results of western blotting are shown. (b) Quantitative analyses of the western blotting results were performed using NIH Image J software. F (2,7) = 23.419, p < 0.001. (c) Determination of plasma levels of HMGB1 by ELISA in rats with ICH. Blood samples were collected 24 h after the induction of bleeding. F (2,20) = 4.576, p = 0.023. (d) Evaluation of the hemorrhagic volume of rats subjected to the striatal ICH. (e,f) The permeability of brain capillary vessels was determined by Evans blue leakage at 6 h (e) and 3 days (f) after ICH. Representative images of Evans blue leakage at 3 h after dye injection are shown for each group. Bar graphs (lower panels) represent the results of quantification of Evans blue dye in the hemorrhagic hemisphere of the anti-HMGB1 mAb-or control IgG-treatment group. (g) Assessment of brain water content induced by ICH. At 3 days after ICH onset, brain water content was determined in the ipsilateral hemisphere. F (2,9) = 216.059, p < 0.001. Results are shown for the sham group (Sham, n = 3, 7, 3 in (b,c,g), respectively), the control IgG-treated group (Con IgG, n = 3, 8, 5, 5, 5, 4 in (b-g), respectively), and anti-HMGB1 mAb-treated group (α -HMGB1, n = 4, 8, 5, 5, 5, 5 to (b-g), respectively). Values represent the means ± SEM. *p < 0.05, **p < 0.01 compared with the sham group. # p < 0.05, ## p < 0.01 compared with the control IgG-treated group.   Anti-HMGB1 mAb inhibited the AQP4 levels in ICH rats. Since AQP4 expression has been suggested to be involved in increased BBB permeability, we next evaluated the effects of anti-HMGB1 mAb on AQP4 levels in ICH rats. We investigated AQP4 expression by immunohistochemistry 24 h after ICH, in the brain tissues of both control IgG-and anti-HMGB1 mAb-treated rats. As shown in Fig. 4, AQP4 immunoreactivities were observed in the cerebral cortex, striatum and hippocampus of control IgG-treated rats (white arrows in Fig. 4), whereas the immunoreactivity was much weaker in the sham group. The treatment with anti-HMGB1 led to a reduction in the immunoreactivity for AQP4 in ICH rats. We counted the number of AQP4-immunoreactive vessels (provided their lengths in the longitudinal direction were more than 10 μ m) in the cerebral cortex, striatum and hippocampus. As shown in Fig. 4, the antibody-treated group showed a significantly reduced number of AQP4-positive vessels on the ipsilateral side in all these areas compared with the control IgG group.

IL-1β induction and localization in peri-hematomal regions after ICH.
In the present study, we observed that IL-1β immunoreactivity was almost completely restricted to reactive astrocytes and neurons in the peri-hematomal regions at 24 h after ICH (Fig. 5a). The immunoreactivity of IL-1β in astrocytes was merged with that of GFAP, whereas the immunoreactivity of IL-1β in neurons was merged with DAPI staining. These findings suggested that the IL-1β localizations in astrocytes and neurons were cytosolic and nuclear, respectively. There were no IL-1β -positive cells in the Iba1-or MPO-positive populations. We frequently observed IL-1β + astrocytic processes surrounding and embracing blood vessels in the peri-hematomal region ( Supplementary Fig. S5).

IL-1β-positive cells within hematoma.
Within the hematoma, the IL-1β immunoreactivity was almost exclusively limited to neutrophils based on their morphology and the MPO expression (Fig. 5b). However, only a few Iba-1 + microglial cells showed IL-1β expression within the hematomal region at 24 h after ICH.

Effects of anti-HMGB1 on the expression of inflammation-related molecules and vasoconstriction-related receptors.
To analyze the anti-inflammatory mechanism for the effects of anti-HMGB1 mAb, we examined the expression of inflammation-and vasoconstriction-related molecules in the peri-hematomal region of the striatum using quantitative real-time RT-PCR (Fig. 6).
The expressions of TNF-α , iNOS, IL-1β , IL-6, IL-8R, COX-2, MMP2, MMP9 and VEGF 121 were all upregulated on the ipsilateral side of the control IgG-treated rats compared with the sham group at 24 h after ICH (Fig. 6a). These upregulations were significantly suppressed by the treatment with anti-HMGB1 mAb. The expressions of the vasoconstriction-related receptors AT-1 PAR-1, V1 and TxA2 were up-regulated, and the up-regulations were suppressed by anti-HMGB1 mAb treatment (Fig. 6b). The expression of α 1-adrenergic receptor was not changed after ICH (Fig. 6b). Endothelial NOS was induced in response to ICH and this induction was inhibited by anti-HMGB1 mAb (Fig. 6a).
Inhibition of IL-1β protein expression by anti-HMGB1. The total number of IL-1β -positive cells in the ipsilateral cerebral cortex and in the peri-hematomal region was reduced significantly by the administration of anti-HMGB1 (Fig. 7). The treatment with anti-HMGB1 mAb suppressed the expression significantly. Then, we counted the IL-1β -positive cells. As shown in Fig. 7c,d, the treatment with anti-HMGB1 mAb significantly reduced the number of IL-1β -positive cells on the ipsilateral side compared with the control IgG group.

Effect of anti-HMGB1 mAb on glial cell activation after ICH. Immunohistochemical staining of
Iba1 showed microglial activation in the peri-hematomal regions after ICH. Amoeba-shaped activated microglia were found around the hematoma at 24 h after ICH (Fig. 3c). The number of Iba1-positive cells in the control mAb-treated group was significantly higher than that in the sham group. Anti-HMGB1 mAb administration significantly reduced the numbers of activated microglia compared with the control IgG group. In contrast to neurons, HMGB1 in the microglia was retained in the nuclei (Fig. 3b). The increase in the number of Iba1-positive microglia was observed not only in the peri-hematomal striatum but also the cerebral cortex, corpus callosum and hippocampus (Supplementary Fig. S1a and b).
Anti-HMGB1 mAb decreased oxidative stress. It is well known that ROS are produced during normal oxidative metabolism, but high ROS levels may damage neurons and cause neuronal death. The serum hydroperoxide concentration was determined and the results are shown in Fig. 8a. Twenty-four hours after ICH, the hydroperoxide concentration increased in the control IgG-treated groups. However, the levels in the anti-HMGB1 group remained significantly lower than that in the control group. To evaluate the oxidant/antioxidant balance after ICH, the serum bioantioxidant potency (BAP) was measured (Fig. 8b). The pattern of serum bioantioxidant potency was roughly parallel to that of the ROS. When comparing the two groups, the BAP in the control group was significantly higher than that in the anti-HMGB1-treated group. ratio of HMGB1-positive microglia to total microglia (F (2,12) = 2.298, p = 0.143) are expressed as the means ± SEM. (c) The Iba1-positive cell numbers were counted at 24 h post-ICH in the peri-hematomal region of the sham, control IgG-and anti-HMGB1-treated groups. Iba1-positive cells in the control IgG-treated group had short processes and relatively larger cell bodies. The scale bar represents 100 μ m. F (2,15) = 20.537, p < 0.001 for area of iba1 + cell. F (2,15) = 14.090, p < 0.001 for immunofluorescence intensity of iba1 + cell. Results are shown for the sham group (Sham, n = 3, 3, 5, 6, 6 in (a-c), respectively), the control IgG-treated group (Con IgG, n = 5, 5, 5, 6, 6 in (a-c), respectively), and anti-HMGB1 mAb-treated group (α -HMGB1, n = 5, 5, 5, 6, 6 to (a-c), respectively). *p < 0.01, **p < 0.01 compared with the sham group. # p < 0.05, ## p < 0.01 compared with the control IgG-treated group.     IgG-treated rats (Fig. 8c). The number of TUNEL-positive cells at 24 h was much higher than that in sham rats. Few TUNEL-positive cells were observed in the contralateral brain in any of the three groups. The number of TUNEL-positive cells in the ipsilateral brain in the anti-HMGB1 group at 24 h was significantly lower than that in the ICH group, but significantly higher than that in the sham group (Fig. 8d).

Anti-HMGB1 mAb improved neurological deficits after ICH. To examine the effects of anti-HMGB1
mAb on neurobehavioral deficits after ICH, we assessed the neurologic function at baseline and at 6, 24, and 48 h after ICH. Compared with the baseline values, neurologic deficits were apparent in all rats at 6 h after ICH. Rats treated with anti-HMGB1 mAb showed a significant improvement in their recovery from initial deficits. In the grip strength test, the neurological score was decreased 6 h after the induction of ICH and remained declined over the 48-h observation period in the control IgG-treated group. Treatment with anti-HMGB1 mAb significantly facilitated the recovery, and the grip strength at 48 h was significantly higher than that in the control IgG-treated group (Fig. 8e). Anti-HMGB1 antibody was also effective in alleviating the deficit in performance in the cylinder test at 24 and 48 h after the induction of ICH (Fig. 8f).

The effects of anti-HMGB1 mAb treatment with regard to the therapeutic time window.
To examine the therapeutic time window for the effects of anti-HMGB1 mAb on ICH-induced brain injury, we started the first injection of anti-HMGB1 3 hours after the onset of ICH and determined the changes in HMGB1 contents and mRNA expression of inflammation-related molecules. The results were shown in Supplementary Fig. S6. In fact, we observed a prevention of HMGB1 decrease and inhibition of mRNA expression of inflammation-related molecules including IL-1β , iNOS, IL-8R, IL-6, TNF-α , VEGF 121, TLR-2 and TLR-4 by anti-HMGB1 similar to those obtained by immediate treatment with anti-HMGB1 mAb, indicating a therapeutic time window of 3 hours in this ICH model at least.

Discussion
In this study, we used an ICH model induced by local injection of collagenase IV into the striatum of rats to mimic spontaneous clinical ICH. Our results indicated that administration of anti-HMGB1 mAb inhibited (1) HMGB1 translocation and release, (2) brain edema, (3) microglia activation, (4) mRNA expression of pro-inflammatory cytokines, and (5) apoptotic cell death in the peri-hematomal areas, and improved the neurological performance of rats suffering from ICH. Anti-HMGB1 mAb also reduced the plasma levels of HMGB1. All of these results support the view that HMGB1 was released mainly from neuronal nuclei and partly from astrocytes in the peri-hematomal areas in the acute phase of ICH, yielding a diverse range of secondary inflammation responses that was inhibited by anti-HMGB1 mAb. As a result, the treatment with anti-HMGB1 mAb ameliorated the neurological symptoms. To the best of our knowledge, these findings are the first to demonstrate that intravenous treatment with a neutralizing anti-HMGB1 mAb had neuroprotective effects on this ICH model.
A recent clinical study reported that the serum levels of HMGB1 were dramatically increased in patients with acute ICH, and this increase was significantly correlated with stroke severity 22 . In related studies, ICH animal models were used to evaluate two pharmacological inhibitors of HMGB1 secretion, ethyl pyruvate and glycyrrhizin 17,23 . The administration of ethyl pyruvate improved the functional outcome, reduced brain edema and decreased the number of apoptotic cells in the rat intracerebral hemorrhage and rat traumatic brain injury models 23,24 . Similarly, glycyrrhizin suppressed brain edema and improved behavioral performance in ICH rats, probably by binding to HMGB1 and thereby inhibiting the interaction between HMGB1 and RAGE 17,25 . Together, these findings suggest that HMGB1 may be one of the key inflammatory factors in ICH pathology and that therapeutic targeting for HMGB1 produces beneficial effects in ICH. In the present study, anti-HMGB1 mAb was shown to bind to HMGB1 directly and suppress the secondary brain injury efficiently. Thus, administration of intravenous anti-HMGB1 mAb may be a quite effective therapy for ICH in addition to brain ischemia, brain trauma, and neuropathic pain 11-13,15,26 . In the previous investigations on ischemic brain injury and traumatic brain injury, the translocation and release of HMGB1 from the nuclei to extracellular space through the cytosolic compartment occurred mainly in neurons [11][12][13] . Similarly, in a Parkinson's disease model, HMGB1 was found to translocate from the nucleus to the cytoplasm and then to the extracellular space in neurons and astrocytes at days 1 and 7 post-lesion, respectively 14 . Consistent with previous studies, we also observed the translocation of HMGB1 into the cytoplasm not only in neurons but also in astrocytes in an ICH model at 24 h after bleeding. In addition, our western blotting and ELISA analyses showed that the level of HMGB1 was significantly decreased in the ipsilateral striatum (the core area of bleeding), and somewhat increased in the serum, respectively. This suggested that HMGB1 was released into the bloodstream from the damaged brain area. Lei et al. found elevated HMGB1 levels in the peri-hematomal region at 24 h and 72 h after ICH by western blotting 23 . The apparent discrepancy of HMGB1 dynamics between these previous studies and our present experiments may be ascribable to the difference in sampling areas and lesion size. In the present study, we sampled the hematoma core for the determination of HMGB1 with a relatively small amount of surrounding tissue, while Lei et al. collected the peri-hematomal region for analysis. Accordingly, we speculated that the most severe inflammation mediated by HMGB1 might occur in the center of hemorrhage due to the massive release and disappearance of HMGB1 occurring within the hematomal area.
Three receptors of HMGB1 (RAGE, TLR-2 and TLR-4) have been reported in earlier studies 16,27,28 . Although ICH was shown to significantly up-regulate the expression of all three receptors of HMGB1, only treatment with the RAGE antagonist (FPS-ZM1) significantly reduced ICH-induced infiltration of inflammatory cells and the expressions of IL-1β and MMP-9 in the peri-hematomal region 25,29 . In contrast, TLR2/4 antagonists had no influence on the ICH-induced infiltration of inflammatory cells, even though they did slightly reduce the expressions of IL-1 and MMP-9 25 . In the present experiments, the mRNA level of RAGE was also decreased by treatment with anti-HMGB1 mAb after ICH. Taken together, these results suggest that RAGE may be the predominant receptor for HMGB1 in the pathogenesis of inflammation after ICH, and anti-HMGB1 mAb might exert its effects through the HMGB1/RAGE signaling pathway.
Cytokines can increase the production of other cytokines via a positive feedback loop. For example, HMGB1 acting as an early pro-inflammatory cytokine promotes the production of many cytokines during ischemic brain injury, including IL-1β and TNF-α 30 . Conversely, increased levels of cytokines such as IL-1β and TNF-α have also been shown to stimulate HMGB1 secretion in different types of cells in in vitro experiments 31 . Here, we reported an up-regulated mRNA expression of proinflammatory molecules such as TNF-α , IL-1β , IL-6, IL-8R, iNOS and COX-2 in the peri-hematomal region after ICH induction, and a significant reduction in the expression of these molecules in the peri-hematomal region at 24 h after intravenous injection of anti-HMGB1 mAb. Hence, HMGB1 release may be one of the causative factors to enhance the expression of these proinflammatory molecules after ICH. Up-regulation of these inflammatory molecules probably contributes to astroglial proliferation and neutrophil recruitment in the region surrounding the hematoma, and causes BBB damage and brain edema as in cerebral ischemia, traumatic brain injury, and neurodegenerative disorders [32][33][34] . In addition, a recent study showed that IL-1β was bound to HMGB1 with high affinity and the association of HMGB1 with IL-1β further potentiated the inflammatory responses, as the cellular activation induced by the HMGB1-IL-1β complex was significantly greater than that found with equivalent amounts of IL-1β or HMGB1 alone 31 . In regard to the protein expression of IL-1β , we clearly showed that it was strongly up-regulated in control ICH rats, and that this up-regulation was considerably inhibited by anti-HMGB1 mAb therapy. It is therefore possible that anti-HMGB1 mAb, which has the ability to neutralize released HMGB1, diminished the synergistic proinflammatory effects between HMGB1 and IL-1β . Interestingly, it has been reported that IL-1β is expressed in the astrocytic processes surrounding and embracing blood vessels in the peri-hematomal region (Fig. S5). In addition, COX-2 inhibition treatment by celecoxib can reduce the infiltration of inflammatory cells and brain edema, resulting in a reduction of TUNEL-positive peri-hematomal cell death 35 . Notably, anti-HMGB1 mAb therapy may suppress plural pathways in the secondary inflammatory responses after ICH.
Blood components (e.g., thrombin, hemoglobin, iron) and the inflammatory responses they induce play a major role in ICH-induced BBB dysfunction 8 . Earlier studies showed up-regulation of MMP-2 and MMP-9, as well as acute brain injury, after ICH 36 . Although collagenase IV is itself one of the MMPs, it has no effect on the expression of other MMPs 37 . These findings suggest that the increase in MMP-2 and MMP-9 after ICH in our study probably contributed to further BBB disruption after collagenase. Indeed, the synthetic inhibitors of MMPs (BB-1101 and GM6001) have been shown to reduce BBB permeability and hemorrhage after ICH in animal models 8 . Here, we demonstrated that anti-HMGB1 mAb achieved a considerable reduction in collagenase IV-induced BBB disruption in the striatum, in association with a decrease of MMP2/9 expression. In addition, the mRNA expression of pro-inflammatory molecules (TNF-α , IL-1β , IL-6, IL-8R, iNOS and COX-2) and the protein expression of IL-1β were significantly increased after ICH induction. The increased levels of these molecules should be associated with a breakdown of BBB and recruitment of neutrophils into the CNS 38,39 . Since our previous investigations using an in vitro BBB system clearly demonstrated that recombinant human HMGB1 increased the vascular permeability of BBB in association with morphological changes in endothelial cells and pericytes 12 , in the present study, HMGB1 probably facilitated BBB disruption both directly and indirectly through induction of the expression of cytokines, MMPs and chemokines.
In conclusion, our findings lend support to the idea that HMGB1 contributes to ICH-induced secondary inflammatory responses, which in turn cause the microglial activation, cytokine expression and BBB disruption. Our results also suggest that anti-HMGB1 mAb could be a valuable neuroprotective agent for the treatment of ICH, even if the treatment is initiated at 3 h after the onset of hemorrhage. We propose that anti-HMGB1 mAb treatment could be a novel therapeutic strategy applicable for three types of stroke: ICH, brain infarction 11,12 and subarachnoid hemorrhage 40 .

Materials and Methods
Animals and treatment groups. All procedures for these experiments was approved by the Committee on Animal Studies of Okayama University and carried out following the guidelines of Okayama University for animal studies. Male Wistar rats (250-300 g, n = 150) were randomly divided into 3 groups: (1) an anti-HMGB1mAb group, in which ICH injury was induced by local injection of collagenase IV in the striatum and then the rats were treated with i.v. injection of anti-HMGB1mAb; (2) a class-matched control mAb group, in which ICH injury was induced and then the rats were treated with class-matched control mAb; and (3) a sham group, in which saline was injected into the striatum instead of collagenase IV and the rats were treated with saline.

Surgical induction of ICH and mAb administration.
Rats were placed in a stereotaxic frame after anesthesia with isoflurane. Intracerebral hemorrhage was induced by the local injection of collagenase IV in the striatum of the rats, as described previously 41 . Briefly, a 30-gauge needle was inserted into the right striatum, 0.2 mm anterior to the coronal suture, 3 mm lateral to the bregma and 6 mm below the skull. Bacterial type IV collagenase (0.03 U) in 2 μ l saline was delivered at a constant rate of 0.5 μ l/min. The needle was left in place for an additional 5 min to prevent backflow. In the sham group, we performed the same procedures, except that we used saline injection instead of collagenase IV. Reproducible lesions and hemorrhage were produced in the striatum by this procedure (Fig. 1). After injection, we removed the needle, filled the burr hole with bone wax and sutured the wound. None of the rats died during the experiments.
Spectrophotometric assay for hemoglobin. Brain hemorrhage was detected by quantifying the hemoglobin content with Drabkin's reagent (Sigma, St. Louis, MO) as described previously 42,43 . Briefly, after removing the blood through transcardial perfusion, the ipsilateral brain tissue containing the hemorrhage region was obtained at 24 h after ICH, and PBS (3 ml) was added to each hemisphere, followed by homogenization for 1 min, sonication on ice with an ultrasonicator for 30 seconds, and centrifugation at 15,000 g for 30 min. The supernatant (40 μ l) was reacted with 160 μ l of Darbkin's solution for 15 min. Fifteen minutes later, the optical density (OD) was measured at 540-nm wavelength. To derive the standard curve, aliquots of 0 μ l, 10 μ l, 20 μ l, 30 μ l, 40 μ l, and 50 μ l blood were added into the normal brain hemisphere, and the OD values were measured as a reflection of the amount of hemoglobin present. The OD values of the ICH-induced brains were then compared with this standard curve to obtain the data on hemorrhage volume.
Western blot analysis. Brain samples were collected from the striatum hematoma core (3 × 3 × 3 mm 3 ) and homogenized for 5 min in RIPA lysis buffer that contained a cocktail of protease inhibitors (Sigma, St. Louis, MO). The protein samples were homogenized and separated by 12% SDS-polyacrylamide gel electrophoresis, then transferred onto nitrocellulose membranes. After blocking with 10% skim milk, the membranes were incubated at 4 °C overnight with polyclonal rabbit anti-HMGB1 antibody labeled with horseradish peroxidase (made by our laboratory). β -actin, as a reference protein, was probed with a mouse anti-β -actin mAb (Santa Cruz Biotechnology, Santa Cruz, CA) followed by goat anti-mouse Ab. Finally, the bands were visualized using an ECL system (Thermo Fisher Scientific Inc., Bend, OR), and analyzed by Image J software (NIH).

Enzyme-linked immunosorbent assay.
To determine HMGB1 levels in plasma, blood samples (1 ml) were collected through the rat heart under deep anesthesia, then centrifuged for 15 min at 800 g. HMGB1 was detected by using an ELISA kit (Shino-Test Co, Sagamihara, Japan), according to the manufacturer's instructions.

Evaluation of BBB integrity.
To evaluate BBB permeability, Evans blue extravasation was measured as described previously 44 . Briefly, 2% Evans blue dye (2 ml/kg) was injected intravenously at 6 h or 3 days after ICH. Three hours later the rats were perfused with 100 ml of saline via the left ventricle of the heart under deep anesthesia with pentobarbital sodium. The brains were immediately removed and the ipsilateral hemispheres were homogenized in 1 M potassium hydroxide. Then the homogenized mixture was placed in 50% trichloroacetic acid and centrifuged at 800 g for 30 min. The absorbance of the supernatant solution was measured at 620 nm. The levels of Evans blue (ng/g) were calculated by means of a standard curve.
Brain water content. Brain edema of the ipsilateral brains was determined by the wet-dry weight ratio method at 72 h after ICH, as described previously 12 . Brain water content was calculated as [(wet weight − dry weight)/wet weight] × 100%.
Immunohistochemical studies. For immunofluorescence staining, paraffin-embedded brain sections were prepared as described in our previous studies 11 . The primary antibodies used in the experiments included anti-HMGB1 Ab (R&D Systems Inc., Minneapolis, MN), anti-AQP4 Ab (Abcam Plc, Cambridge, UK), anti-IL-1β Ab (R&D Systems Inc.), anti-microtubule-associated protein 2 (MAP2) Ab (Abcam Plc), anti-glial fibrillary acid protein (GFAP) Ab (Abcam Plc), anti-ionized calcium-binding adaptor molecule 1 (Iba1) Ab (Wako, Osaka, Japan) and anti-myeloperoxidase (MPO) Ab (Abcam Plc). In order to investigate the cellular source as well as the localization of IL-1β , double immunohistochemical staining was carried out with cell marker antibodies including MAP2, GFAP, iba1 or MPO antibodies and an antibody against IL-1β . The sections were then incubated with secondary Abs conjugated with Alexa-488, Alexa-555 or Alex594, which were purchased from Invitrogen (Tokyo, Japan). Finally, the sections were mounted using VECTASHIELD Hard Set Mounting Medium with DAPI (Vector Laboratories Inc., Burlingame, CA) and observed under an LSM 780 confocal microscopic system (Carl Zeiss Inc., Jena, Germany). The counting was performed in a blinded manner.

Assay of reactive oxygen metabolites and antioxidant potency (biological antioxidant potency). Derivatives of reactive oxygen metabolites (d-ROM) and biological antioxidant potency (BAP)
were measured in plasma at 24 h after ICH using a free radical electron evaluator (FREE; Health & Diagnostic Limited Co., Naples, Italy) according to the manufacturer's instructions. In brief, the plasma was added to an acetate buffer with FeCl 2 at 37 °C, followed by the addition of a chromogenic mixture including aromatic alkyl-amine. After incubation at 37 °C for 5 min, the colored radical derivative was measured at 505 nm. The results were expressed in Carratelli units (U.CARR); One U.CARR is equal to 0.08 mg/dl of H 2 O 2 . For BAP assay, the plasma was dissolved in a colored solution that was prepared previously by mixing FeCl 3 with a thiocyanate derivative.
After 5 min of incubation at 37 °C, the solution gradually lost its color, and the intensity was directly proportional to the ability of the plasma to reduce ferric ions to ferrous ions. Finally, the chromatic change was read at 505 nm.
TUNEL staining and analysis. Paraffin-embedded brain sections were fixed at 24 h after ICH, then processed for terminal deoxynucleotidyl transferase-mediated dUTP-biotin nick end labeling assay (TUNEL; Takara, Tokyo, Japan) in accordance with the manufacturer's protocol. Briefly, the sections were treated with proteinase K (20 μ g/ml) for 20 min at room temperature, and then 50 μ l of an FITC-labeling reaction mixture (consisting of TdT Enzyme 5 μ l with FITC-Labeling Safe Buffer 45 μ l) was applied to the slide and allowed to react for 90 min in a 37 °C humidified chamber. After that, the sections were counterstained with DAPI. TUNEL-positive cells were analyzed by fluorescent microscopy (n = 5). The number of TUNEL-positive cells under a high-power (x200) field was counted in four separate regions surrounding the hematoma (Fig. 2a), and the ratios of the positive cells to the total cell number were calculated and analyzed by computer. The counting was performed by an experimenter blinded to the sample information.

Assessment of motor function (behavioral testing).
A grip strength test (GPM-100B; MELQUEST, Toyama, Japan) was performed before injury and at 6, 24 and 48 h after brain injury according to the manufacturer's instructions. Briefly, the rats were held by their body, and allowed to grip the grid with the contralateral forelimb of the hemorrhaged side. Then, the rats were gently pulled downwards and the peak of the grip strength was recorded (in grams). Three trials were performed for each rat and the average was used as the animal's forelimb grip force at that particular time point. For the forelimb use asymmetry test, forelimb use was analyzed by observing the rats in a transparent cylinder (20 cm in diameter and 40 cm in height) before and after ICH according to the method described previously 44 . All behavior tests were done by an experimenter blinded to the treatment. Statistical analysis. Statistical significance was evaluated using two-tailed Student's t-test to compare two groups. For multiple comparisons, one-way ANOVA followed by the post hoc Bonferroni test or the post hoc Fisher's LSD test was used. Excel Tohkei software (Social Survey Research Information, Tokyo, Japan) was used for the statistical analyses. A probability value of less than 0.05 was considered to indicate statistical significance. The results were expressed as the means ± SEM.