Hypoxia-inducible Factor-1 Mediates Neuronal Expression of the Receptor for Advanced Glycation End Products following Hypoxia/Ischemia*

Activation of the receptor for advanced glycation endproducts (RAGE) by its multiple ligands can trigger diverse signaling pathways with injurious or pro-survival consequences. In this study, we show that Rage mRNA and protein levels were stimulated in the mouse brain after experimental stroke and systemic hypoxia. In both cases, RAGE expression was primarily associated with neurons. Activation of RAGE-dependent pathway(s) post-ischemia appears to have a neuroprotective role because mice genetically deficient for RAGE exhibited increased infarct size 24 h after injury. Up-regulation of RAGE expression was also observed in primary neurons subjected to hypoxia or oxygen-glucose deprivation, an in vitro model of ischemia. Treatment of neurons with low concentrations of S100B decreased neuronal death after oxygen-glucose deprivation, and this effect was abolished by a neutralizing antibody against RAGE. Conversely, high concentrations of exogenous S100B had a cytotoxic effect that seems to be RAGE-independent. As an important novel finding, we demonstrate that hypoxic stimulation of RAGE expression is mediated by the transcription factor hypoxia-inducible factor-1. This conclusion is supported by the finding that HIF-1α down-regulation by Cre-mediated excision drastically decreased RAGE induction by hypoxia or desferrioxamine. In addition, we showed that the mouse RAGE promoter region contains at least one functional HIF-1 binding site, located upstream of the proposed transcription start site. A luciferase reporter construct containing this RAGE promoter fragment was activated by hypoxia, and mutation at the potential HIF-1 binding site decreased hypoxia-dependent promoter activation. Specific binding of HIF-1 to this putative HRE in hypoxic cells was detected by chromatin immunoprecipitation assay.

The receptor for advanced glycation end products (RAGE) 4 is a member of the immunoglobin superfamily of cell surface molecules. It was originally identified by its capacity to bind advanced glycation end products, adducts that accumulate during natural aging and are produced at an augmented rate during diabetes (1). Subsequently, several other ligands for this receptor have been reported including amyloid-␤ peptide, high-mobility group box 1, some members of the S100/calgranulin family, and Mac-1 (2)(3)(4)(5). Multiple studies indicate that RAGE signaling has profound stimulatory effects on gene expression of inflammatory mediators, a mechanism that has been implicated in the pathogenesis of diabetic complications in the periphery (for review, see Ref. 6). In the central nervous system, the expression of RAGE has been described in several cell types. During the embryonic and early postnatal period, RAGE is highly expressed by hippocampal, cortical, and cerebellar neurons (4). Expression of this receptor is limited in the normal adult brain, but enhanced in pathological conditions like Alzheimer disease (5,7). In this pathology, RAGE expression was observed not only in neurons but also in endothelial cells, and microglia (5,8). Several RAGE ligands are produced in the brain under normal and pathological conditions; interactions of these ligands with RAGE can lead to multiple molecular and cellular consequences, depending on the cell type involved and the nature and concentration of the ligand. S100B is one such ligand. In addition to its intracellular function, when secreted S100B can exert autocrine and paracrine effects that have trophic or toxic impact depending on its concentration. It has been proposed that activation of RAGE can mediate both sets of effects (9). In vitro studies using primary neurons or neuroblastoma cells showed that stimulation of RAGE by low levels of S100B leads to neurite outgrowth, and activation of pro-survival pathways during stress conditions like trophic factor deprivation, glutamate, N-methyl-D-aspartate, or amyloid-␤ toxicity (9 -12). On the other hand, treatment of primary neurons, astrocytes, or microglia with relatively high doses of S100B leads to oxidative stress, NF-B activation, expression of proinflammatory mediators, and cytotoxicity (9,11,13,14). Little is known about the in vivo synergistic or inhibitory effects of S100B and the other RAGE ligands, and for some of these proteins (i.e. high-mobility group box 1, S100B, and amyloid-␤) other cell surface interaction sites besides RAGE have been postulated (15,16).
Recently, it was reported that expression of RAGE is enhanced by ischemia both in the brain and heart (17,18). In addition, it is well established that the production of the glialderived protein S100B is substantially enhanced upon activation in response to cerebral ischemia (19,20). These reports support the possibility of an effective activation of RAGE-mediated signaling in the ischemic brain, the consequences of which have not been explored. Regulation of many cellular responses to ischemia requires the concerted activation of various transcription factors including hypoxia inducible factor 1 (HIF-1). HIF-1 is a heterodimer composed of HIF-1␣ and HIF-1␤ subunits; HIF-1␣ is the regulatory component of this complex and its expression is exquisitely regulated by an oxygendependent post-translational modification that targets HIF-1␣ for proteosomal degradation (for review, see Ref. 21). During hypoxia, HIF-1␣ is stabilized, translocates into the nucleus where it binds HIF-1␤ and forms the active HIF-1 complex. Interaction of HIF-1 with its consensus DNA binding site is required for the hypoxia-induced expression of a vast array of target genes, which are involved in various cellular and systemic adaptive responses to hypoxia, including erythropoiesis, angiogenesis, vasomotor regulation, cell proliferation and survival, cell death, and matrix metabolism, among others (22,23). HIF-1␣ is up-regulated in the hypoxic-ischemic brain (24 -26), however, potential interactions with S100B or RAGE have not been investigated. The purpose of this study was to investigate this potential signaling pathway in hypoxia-ischemia. We confirm earlier findings of neuronal RAGE up-regulation after cerebral hypoxia-ischemia (HI) in the mouse and further demonstrate that RAGE activation in the ischemic brain may have a neuroprotective role because RAGE-null mice displayed exacerbated brain damage when subjected to HI. In addition, by using a loss-of-function approach, we show that HIF-1 is required for RAGE expression during hypoxia. Chromatin immunoprecipitation experiments demonstrate that hypoxia induces the association of endogenous HIF-1␣ to the RAGE promoter. Finally, we show that activation of RAGE in primary neurons by low concentrations of S100B improves neuronal survival after oxygen-glucose deprivation. Conversely, high concentrations of S100B have a RAGE-independent cytotoxic effect. Together, our results indicate that HIF-1 mediates the transcriptional activation of RAGE expression in neurons, and activation of this receptor can be viewed as part of the glianeuron communication system that might be important for the neuronal response to ischemic insults, and explain the basis for the biphasic effects of S100B in the brain.

EXPERIMENTAL PROCEDURES
Animals-The generation of RAGE-null mice has been described in detail elsewhere (27,28). The RAGE-null animals used for the present study were provided by Dr. A. M. Schmidt, these animals have been backcrossed into the C57Bl/6J strain for Ͼ10 generations at the Columbia University animal facility.
Age-matched wild-type C57Bl/6J mice were used as controls (The Jackson Laboratories, Bar Harbor, ME). RAGE-null mice are viable, and display normal development and reproductive capacity. We found no differences in pre-ischemic body weight and blood glucose levels among RAGE-null and control animals. Toth et al. (29) recently described normal brain weight and no white matter abnormalities in RAGE-null mice. Mice carrying conditional Hif-1␣ floxed alleles (Hif-1␣ F/F ) were generously provided by Dr. Randal Johnson (University of California, San Diego). These animals (C57Bl/6 genetic background) were generated by engineering loxP sites flanking exon 2 of the Hif-1␣ gene as described previously (30).
Induction of Unilateral Cerebral Hypoxia-ischemia Damage-Hypoxia-ischemia was induced in male RAGE-null or wild-type mice (12-15 weeks) as described previously (31,32). Animals were anesthetized with isoflurane (2.5% induction, 1.5% maintenance), and the right common carotid artery was isolated and double ligated with 4-0 surgical silk. The incision was sutured, and animals were allowed to recover with access to food and water for 2 h. Then, animals were exposed to hypoxia (30 min) in custom-designed plexiglass chambers individually equipped with oxygen and temperature sensors (OxyCycler A-series, BioSpherix, Redfield, NY). Oxygen levels during hypoxic exposure were monitored and controlled at a constant concentration of 8% O 2 balanced with nitrogen. Chamber temperature was maintained at 35°C; preliminary experiments showed that at this temperature, the animals rectal temperature was 37°C. After hypoxia, animals were allowed to recover at normoxia for 30 min in the chambers, and then returned to their cages. Animals were euthanized at 6 h to 5 days of recovery. The duration of systemic hypoxia was chosen to assure consistent injury in 80% of the animals and low mortality rate by 48 h of recovery. Animals with right common carotid artery ligation without hypoxia were used as sham-operated controls.
Systemic Hypoxia-C57Bl6J male mice (12 weeks old) were exposed to normobaric hypoxia (10% O 2 balanced with nitrogen) for 24 and 48 h (BioSpherix) with access to food and water. Littermate controls were kept under normoxia in a similar chamber for the duration of the experiment. Animals were sacrificed immediately after hypoxia.
Infarct Volume and Immunohistochemistry-Brains were quickly removed from the skull and frozen in dry ice-chilled isopentane (Ϫ30°C). Coronal sections (25 m) were cut serially in a Leica cryostat and mounted on superfrost slides (VWR). The extent of brain infarction was identified using cresyl violet staining; the infarcted area was determined indirectly by subtracting the area of healthy tissue in the ipsilateral hemisphere from the area of the contralateral hemisphere. Infarction volume was calculated by integration of infarct areas measured in 12 equidistant sections (at 250-m intervals) encompassing the entire lesion. For immunohistochemistry, sections were fixed with ice-cold methanol (Ϫ80°C), incubated with PBS containing 0.4% (v/v) Triton X-100 and 10% (w/v) normal serum for 2 h. Subsequently, sections were incubated overnight at 4°C with rat polyclonal antibody against RAGE (1:100; R&D Systems). RAGE-positive cells were visualized with secondary biotinylated anti-rat antibody (1:200; Vector Laboratories, Burlingame, CA) and Cy3-conjugated streptavi-din (1:500, Molecular Probes). Sections were subsequently double-stained for the neuronal specific nuclear protein NeuN (1:200, Chemicon) using the Vector M.O.M. Immunodetection kit. An additional group of sections were double stained with the astrocyte marker glial fibrillary acidic protein, using a polyclonal rabbit anti-glial fibrillary acidic protein antibody (1:400; Dako) and a secondary anti-rabbit antibody conjugated with fluorescein (1:100; Invitrogen). Staining was analyzed and documented using an Axiovert 200M microscope equipped with AxioVision software. The specificity of RAGE immunohistochemistry was confirmed by showing negligible staining in the ischemic RAGE-null mouse brain (not shown).
Cell Culture and Treatments-Primary neuronal cultures were prepared from cerebral cortices of wild-type C57Bl/6J or homozygote conditional floxed HIF-1␣ (HIF-1␣ F/F ) mouse embryos (E15), according to the protocol described by Chavez et al. (33). Briefly, dissected cortices were dissociated in Earl's balance salt solution containing papain (50 units/ml) and DNase I (100 units/ml). Cells were seeded in poly-D-lysinecoated plates under serum-free conditions using Neurobasal medium supplemented with B27, glutamine (2 mM), glutamate (25 M), and ␤-mercapthoethanol (25 mM) (Invitrogen). On the fourth day of plating, one-half of the medium was replaced with glutamate-free B27/Neurobasal medium, and subsequently cultures were fed every 4 days with glutamate-free medium. These cultures contain Ͼ90% neurons as determined by microtubule-associated protein-2 cytochemistry. Experiments were performed in neurons at days 10 -15 in vitro unless otherwise indicated. 3T3 NIH cells (ATCC) were grown in Dulbecco's modified Eagle's medium supplemented with 10% bovine fetal serum. Primary neurons were stimulated with bovine brain S100B protein (Calbiochem), desferrioxamine (Sigma), or dimethyloxalglycine (Sigma) freshly dissolved in PBS at the indicated concentrations. RAGE-blocking antibody (Ab-RAGE) was provided by Dr. A. M. Schmidt (Columbia University, New York).
Hypoxia and Oxygen Glucose Deprivation-A custom-made temperature controlled hypoxic/anaerobic glove-box system was used (Coy Laboratories, MI). This system is equipped with an inverted microscope that allows visual inspection of cell viability before terminating each experiment. For hypoxia treatments, the system was set up at 37°C with an atmosphere of 0.5% O 2 , 5% CO 2 , and 94% N 2 , and all solutions were pre-equilibrated for at least 12 h before each experiment. Cells were transferred into the chamber, washed with PBS, and incubated with fresh media for up to 24 h in a humidified internal incubator. At the end of hypoxia, cell harvesting and lysis were performed within the chamber. In agreement with a previous report (33), this degree of hypoxia did not produce cell death (data not shown). For oxygen-glucose deprivation, the glovebox system was set up at 37°C with an atmosphere of 5% CO 2 , 5% H 2 , 90% N 2 (anaerobic). Neurons were transferred into the chamber, washed with PBS, and incubated with a pre-equilibrated glucose-free balance salt solution for up to 60 min. At the end of the procedure, cells were removed from the chamber, fresh Neurobasal media was added (reperfusion), and cultures were returned to a regular incubator. At different periods of reperfusion, neurons were harvested for immunoblot analysis, and cell death was determined using 3-(4,5-dimethylthiazol-2yl)-2,5-diphenyltetrazolium bromide assay following the manufacturer's protocol (CellTiter 96 Assay, Promega, Madison, WI). For experimental controls, cultures were subjected to the same procedures but maintained at normoxia with glucosecontaining media in a standard cell culture incubator.
Adenoviral Vector Construction and Transduction-A nonreplicative adenovirus in which the Cre recombinase gene is under the regulation of the cytomegalovirus promoter was obtained from Vector Biolabs. Reporter adenovirus encoding green fluorescence protein (AdGFP) was used as control. For adenoviral transduction, cortical neurons derived from homozygous Hif-1␣ F/F mice were prepared as described above. At day 6 in vitro, cells were infected with AdCre or AdGFP at a multiplicity of infection of 100. Significant deletion of floxed HIF-1␣ alleles by AdCre infection was accomplished at 7 days post-infection as determined by Western blot. Adenoviral infection with either AdCre or AdGFP did not affect neuronal viability (data not shown).
Preparation of Tissue and Whole Cell Lysates-Right and left hemispheres were dissected from ischemic brains and frozen in liquid nitrogen. Samples were homogenized with a Polytron homogenizer using ice-cold lysis RIPA buffer (1ϫ PBS, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS) supplemented with protease inhibitors (Complete, Roche). Homogenates were centrifuged at 10,000 ϫ g for 10 min (4°C), and supernatants were collected for immunoblots. For the preparation of whole cell lysates, cells were harvested, washed with PBS, and centrifuged (2000 ϫ g for 10 min). The resulting cell pellet was subsequently processed as described above. In all cases, protein concentrations were determined by Bradford protein assay with bovine serum albumin as standard (Bio-Rad).
Western Blot Analysis-Samples were electrophoresed on SDS-PAGE under reducing conditions and transferred to nitrocellulose membranes (Bio-Rad) by standard procedures. Membranes were blocked with 10% nonfat dry milk and incubated with the following primary antibodies: rabbit anti-RAGE (Santa Cruz), anti-␤-actin (Santa Cruz), anti-HIF-1␣ (R&D), and anti-HIF-1␤ (Novus Biologicals Littleton, CO). After washing, membranes were incubated with the appropriate horseradish peroxidase-conjugated secondary antibodies. Antigen-antibody complexes were visualized by enhanced chemiluminescence detection (ECL, Amersham Biosciences). Results were visualized and quantified with Kodak Image Station 4000. Membranes were stripped and reprobed as needed.
Real Time PCR Analysis-Total RNA from cell cultures or brain samples was extracted using RNgents Total RNA Isolation System (Promega). Complementary DNA was synthesized from 2.5 g of total RNA using the SuperScript III system with oligo(dT) (18) primer (Invitrogen, CA). Real time PCR analysis was performed using 0.5 l of the final cDNA synthesis mixture and mouse specific TaqMan-based gene expression assays (Applied Biosystems). The following assays were employed: Advanced glycation end products (Ager) (Mm00545815_m1), vascular endothelial growth factor (Vegf) (Mm00437304_m1), erythropoietin (Epo) (Mm00433126_ m1), and ␤-actin (Mn00607939_s1). The PCR was carried out in an ABI 7500 real time PCR thermocycler (Applied Biosystems, CA). All reactions were performed in duplicate and independently repeated at least three times. Results were normalized to ␤-actin, and expressed as -fold increase relative to control.
Generation of Promoter Constructs, Transient Transfection, and Reporter Gene Assays-A fragment containing the 5Ј-flanking region (ϳ2000 bp) of the mouse Rage gene was generated from mouse genomic DNA by PCR using the following primers: forward, 5Ј-attgctagcgggaggtcagatatacagtc-3Ј and reverse, 5Ј-attaagcttccatctcccatctcgttc-3Ј. This product was cloned into the NheI and HindIII sites of the pGL3-basic vector (Promega), and the generated plasmid was designated pRAGE-luc1. Three additional RAGE promoter constructs (pRAGE-luc2, pRAGE-luc3, and pRAGE-luc4) were generated using the same downstream primer as for pRAGE-luc1 and the following NheI site-containing upstream primers: F2 (attgctagcccagagatgccaaaaatgg), F3 (attgctagcttgagaagtaagagccaaa), and F4 (attgctagctgaactcagtgattttgaa). In the pRAGE-mutHRE construct, the putative HRE of pRAGE-luc1 was replaced from 5Ј-ACGTG-3Ј to 5Ј-AAAAG-3Ј using the QuikChange sitedirected mutagenesis kit (Stratagene). All constructs were verified by DNA sequencing. 3T3 NIH cells at about 90% confluence in 24-well plates were transiently transfected with reporter plasmid (0.5 g) using Lipofectamine 2000 (Invitrogen) according to the manufacturer's directions. To correct for variable transfection efficiency, cells were cotransfected with the pRL-SV40 vector (0.05 g) encoding the Renilla luciferase gene. Transfected cells were allowed to recover for 24 h in fresh medium, and then treated with desferrioxamine (DFO) (100 M), dimethyloxalglycine (DMOG) (300 M), or subjected to 0.5% O 2 . Cells were lysed and luciferase activity was determined with a multiwell luminescence reader (Molecular Devices), by using the Dual-Luciferase Reporter Assay System (Promega).
Statistical Analysis-Data are presented as mean Ϯ S.D. Statistical comparisons among groups were made using a one-way analysis of variance test with Tukey correction. A p Ͻ 0.05 was considered statistically significant.

Induction of RAGE Expression after Cerebral Ischemia-
The effect of brain ischemia on RAGE expression was examined using a model of unilateral HI brain damage in the adult mouse (31,32). This methodology consists of a permanent unilateral ligation of the right common carotid artery followed by a short period of systemic hypoxia (8% oxygen, 30 min), which causes a reduction of blood flow in affected brain areas by 50 -60% (31). Reperfusion of the ipsilateral hemisphere commences with the return to normoxia. As previously described, this model produces ischemic brain damage that is restricted to the hemisphere ipsilateral to the ligation (Fig. 1A). There is no obvious neuropathology on the contralateral side of the brain, or after unilateral carotid occlusion alone (sham-operated animals).
At various intervals of reperfusion after HI, RAGE mRNA and protein levels were evaluated in the ipsilateral and contralateral cerebral hemispheres. Real time RT-PCR showed a significant increase of Rage mRNA levels in the ischemic but not in the contralateral hemisphere, as compared with the sham-operated animals (Fig. 1B). This up-regulation was evident at 12 h of recovery (ϳ70% increase), and continued at 2 and 5 days of recovery (ϳ150% increase). Similarly, immunoblot analysis showed a transient increase of RAGE protein levels in the ischemic hemisphere, with maximal induction observed at 24 and 48 h of recovery (Fig. 1C). The polyclonal RAGE antibody used in this assay recognized a major immunoreactive band at ϳ50 kDa and additional closely migrating bands that may represent spliced or less glycosylated forms reported previously (34 -36). Immunostaining performed at 24 h after HI revealed that induction of RAGE occurred mainly in neurons surrounding the infarction, as indicated by double staining with the neuronal marker NeuN (Fig. 1A). None of the RAGE-positive cells were positive for glial fibrillary acidic protein (Fig. 1A). Conversely, very little staining was evident in the non-damaged hemisphere (Fig. 1A). To ascertain whether RAGE-dependent signaling contributes to ischemic brain damage, we subjected wild-type and RAGE-null mice to HI. At 12 h of reperfusion, there was no significant difference in infarct volume between wild-type and RAGE-null mice. However, RAGE-null mice showed larger infarct volumes at 24 h of recovery (ϳ24% increase, Fig. 1D) suggesting progressive injury, whereas the infarct volume in the wild-type was not different between 12 and 24 h.
Systemic Hypoxia Increases RAGE Expression in the Brain-We then asked whether RAGE expression in the brain can also be modulated by prolonged systemic hypoxia, a stress that unlike HI does not produce significant neuronal death. For this purpose, animals were exposed to 10% O 2 for 24 and 48 h, and DECEMBER 14, 2007 • VOLUME 282 • NUMBER 50

JOURNAL OF BIOLOGICAL CHEMISTRY 36333
Rage mRNA levels evaluated in cortex by real time RT-PCR. Our results showed a significant induction of RAGE mRNA levels in cortex after hypoxia (ϳ80 and 140% by 24 and 48 h of hypoxia, respectively) ( Fig. 2A). The kinetics of RAGE mRNA stimulation closely followed that of VEGF and Epo, which have been previously reported to be hypoxia-inducible genes in the brain (25, 37) ( Fig. 2A). Immunohistochemistry performed at 48 h of hypoxia showed that RAGE staining co-localized primarily with NeuN-positive neurons and was detected throughout the forebrain (Fig. 2B). No significant levels of RAGE protein were detected in the normoxic brains.
HIF-1-dependent Regulation of Hypoxic Expression of RAGE in Primary Neurons-The in vivo up-regulation of RAGE expression in the brain after prolonged hypoxia or ischemia led us to explore the possibility that this gene might be regulated by the conserved HIF-1-dependent pathway. For this purpose, we first evaluated the response of RAGE in cortical primary neurons to DFO (100 M, 24 h) or DMOG treatment (300 M, 24 h), two well characterized pharmacological HIF-1 inducers under normoxic conditions (38,39). Similar to hypoxia, DFO and DMOG led to an increase in HIF-1␣ accumulation, whereas HIF-1␤ levels were not affected. Interestingly, Rage mRNA levels were induced not only by hypoxia (0.5% O 2 , 24 h) but also by DFO and DMOG treatment (ϳ2.3-fold increase, p Ͻ 0.05). These results resembled those obtained for Vegf mRNA levels (Fig. 3), an established HIF-1 target.
To further study the role of HIF-1 in the hypoxic induction of RAGE, the expression of HIF-1␣ was down-regulated in primary neurons using Cre-mediated deletion. For this approach, primary neurons were isolated from homozygous mice harboring lox-P sites flanking exon 2 of the Hif-1␣ gene (referred to as Expression of RAGE was only detected in the ipsilateral (b) but not in the contralateral hemisphere (f). B, real time RT-PCR of Rage in the ipsilateral (ischemic) and contralateral cerebral hemispheres of mice subjected to HI and different durations of recovery/reperfusion (6 h to 5 days). Data were normalized to ␤-actin and expressed relative to the contralateral hemisphere of sham-control (C) that was arbitrarily defined as 1. Data are expressed as mean Ϯ S.D. from 4 animals per group; *, p Ͻ 0.05 versus sham. C, representative immunoblot of RAGE in the ipsilateral and contralateral hemispheres of animals subjected to HI followed by recovery. C, sham-operated animal control; ϩ, mouse lung lysate positive control (ϩ). Co-detection of ␤-actin was performed to assess equal loading. D, evaluation of infarct volume at 12 and 24 h after the onset of ischemia. Data are presented as mean Ϯ S.D. (n ϭ 5-6); *, p Ͻ 0.05 compared with wild-type. HIF-1␣ F/F ). These HIF-1␣ F/F neurons were then transduced with adenovirus encoding Cre-recombinase (AdCre) to generate HIF-1-deficient neurons (referred as HIF-1 ⌬/⌬ ). Untreated cells or cells infected with AdGFP were used as controls. Using this approach, previous reports have shown that infection of HIF-1␣ F/F neurons or astrocytes with AdCre results in efficient excision of floxed HIF-1␣ allele (33). Accordingly, immunoblot analysis of AdCre-infected neurons subjected to hypoxia (0.5% O 2 , 24 h) confirmed a substantial decrease of hypoxic HIF-1␣ protein levels starting at 4 days after the infection, whereas HIF-1␤ was not affected (Fig. 4A). Conversely, AdGFP-infected neurons showed normal HIF-1␣ induction after hypoxia. Based on these results, subsequent experiments with HIF-1␣ F/F cells were performed at 7 days post-infection. Interestingly, exposure of non-infected or AdGFP-infected neurons to hypoxia (0.5% O 2 , 24 h) or DFO led to up-regulation of Rage mRNA and protein levels, whereas HIF-1 ⌬/⌬ neurons presented abrogated RAGE induction (Fig. 4B). As expected, HIF-1 ⌬/⌬ neurons also showed a significant attenuation of the induction of Vegf after hypoxia or DFO (Fig. 4, B and C).
Hypoxia Increases RAGE Promoter Activity-The transcription factor HIF-1 binds to a conserved hypoxia-response element (HRE) of target genes containing a HIF DNA binding site referred to as the core HRE (23). Sequence analysis of the 5Ј-flanking region of the mouse Rage gene revealed the presence of a putative binding site for HIF with the characteristic motif, 5Ј-RCGTG-3Ј, located at position Ϫ1063/Ϫ1059 upstream of the transcription start site. To determine whether this promoter region mediates RAGE activation in response to hypoxia, we constructed a series of deletion reporter plasmids in which fragments of RAGE 5Ј-flanking sequences were fused to the firefly luciferase gene (Fig. 5A). For this purpose, a ϳ2-kilobase pair RAGE promoter construct (pRAGEluc1, Ϫ2168/ Ϫ59) was used as a template for the generation of shorter plas-  . Cre recombinase-mediated deletion of HIF-1␣ affects hypoxiainduced RAGE expression. A, primary neurons derived from HIF-1␣ F/F transgenic mice were infected with AdGFP or AdCre. AdCre-infected cells were exposed to hypoxia at several days post-infection (1-7 days). Representative immunoblot analysis demonstrating Cre recombinase-mediated reduction of HIF-1␣ but not HIF-1␤ levels in hypoxic AdCre-infected cells. Conversely, AdGFP-infected cells exposed to normoxia (N) or hypoxia (H, 0.5% O 2 , 24 h) at 7 days post-infection showed normal HIF-1␣ response. B and C, RT-PCR analysis of Rage and Vegf transcripts performed in AdGFP or AdCre-infected neurons exposed to normoxia (N), hypoxia (H), vehicle (C), or DFO (100 M). Noninfected wild-type (not shown) and non-infected HIF-1␣ F/F neurons were used for comparison. All RT-PCR results were normalized to ␤-actin and expressed as -fold induction compared with wild-type cells treated with vehicle or kept at normoxia (n ϭ 6); *, p Ͻ 0.05 compared with normoxic or vehicle-treated cells. DECEMBER 14, 2007 • VOLUME 282 • NUMBER 50 mids with truncated 5Ј ends: pRAGEluc2 (Ϫ1650/Ϫ59), pRAGEluc3 (Ϫ1175/Ϫ59), and pRAGEluc4 (Ϫ950/Ϫ59). In addition, a mutant plasmid was generated from pRAGEluc1 by introducing a 4-bp substitution through site-directed mutagenesis at the putative HIF binding site (pmutRAGE-luc). Due to the low transfection efficacy of primary neurons, 3T3 NIH cells were used for these experiments. 3T3 NIH cells were transiently transfected with these constructs and exposed to normoxia, hypoxia (0.5% O 2 ), DFO (100 M), or DMOG (300 M) for 24 h. Our results showed that luciferase activity of pRAGE-luc1, pRAGEluc2, and pRAGEluc3 was stimulated by hypoxia, DFO, and DMOG by ϳ2-fold relative to normoxic controls (Fig. 5A). However, pRAGEluc4 or pmutRAGE-Luc constructs were insensitive to hypoxia, DFO, or DMOG (Fig. 5A). These experiments indicate that a hypoxia-sensitive region extending between Ϫ1175 and Ϫ950 bp in the Rage promoter harbors regulatory elements including a potential HRE that was sufficient for transcriptional stimulation of the reporter gene.

HIF-1 Regulates RAGE Expression during Hypoxia
HIF-1 Binding to the HRE Present in the RAGE Promoter-To assess the ability of HIF-1␣ to interact with the HRE consensus sequence identified in the Rage promoter, we performed EMSA and chromatin immunoprecipitation analysis. EMSA was performed using radiolabeled oligonucleotides harboring the core HRE and flanking sequence of the Rage promoter (wtRAGE). Fig. 5B shows that this oligomer bound constitutive factors present in nuclear extracts of both normoxic and hypoxic neurons. Importantly, an induced binding activity was present in the hypoxic but not normoxic extracts. This hypoxia-induced complex was supershifted by antibodies raised against HIF-1␣, suggesting that HIF-1␣ was a component of this complex (Fig. 5B). Disruption of the HIF-1 recognition sequence in the probe (mutRAGE) resulted in loss of binding to nuclear extracts of hypoxic neurons. Similar results were obtained when the assay was performed with an oligonucleotide corresponding to the HRE of the 3Ј enhancer of Epo gene, an established HIF-1 target (Fig. 5B).
For chromatin immunoprecipitation assays, HIF-1␣ F/F and HIF-1␣ ⌬/⌬ primary neurons were exposed to normoxia or hypoxia (0.5% O 2 , 1-8 h). Following crosslinking, an antibody directed against HIF-1␣ was used to immunoprecipitate the cross-linked DNA-protein complex. PCR analysis of isolated DNA fragments indicates that hypoxia causes HIF-1␣ recruitment to the endogenous HRE-containing Rage promoter region. Confirming the specificity of these results, decreased association of HIF-1␣ with the Rage promoter was observed in hypoxic HIF-1␣ ⌬/⌬ neurons. In neurons maintained under normoxic conditions, HIF-1␣ was not significantly associated with the Rage promoter (Fig. 5C).
Effect of RAGE Activation by S100B on Oxygen-Glucose Deprivation-induced Neuronal Death-The production and release of the glia-derived protein S100B is greatly augmented by cerebral ischemia (19,20). We next proceeded to study the cellular effects of RAGE activation by S100B in primary neurons subjected to oxygen-glucose deprivation (OGD), an in vitro model of ischemia-reperfusion injury. Exposure of cortical neurons to OGD (60 min) followed by recovery resulted in a progressive decrease of cell viability; by 24 h there was a ϳ70% reduction of cell survival compared with control (Fig. 6A). Western blot analysis of RAGE after OGD showed stimulation of RAGE at 3-12 h of recovery (Fig. 6B). Treatment of neurons with S100B (24 h), at a con- FIGURE 5. RAGE is a target gene of HIF-1. A, schematic representation of the 5Ј-deleted RAGE promoterluciferase reporter constructs. Potential transcription factor binding sites identified in this area are represented by boxes. Numbering refers to position relative to the transcription initiation site (ϩ1). NIH 3T3 cells were co-transfected with the indicated RAGE-luciferase reporter constructs (pGL3-basic vectors) together with pRL-SV40 as a transfection control. pRAGE-mutHRE is the mutant form of pRAGE-luc1, with point mutations at the putative core HRE. After transfection, cells were exposed to normoxia (N), 1% O 2 (H), DFO (100 M), or DMOG (300 M) for 24 h. Promotor activity was expressed as a ratio of firefly/Renilla luciferase activities (RLU, relative luciferase units, mean Ϯ S.D., n ϭ 4; *, p Ͻ 0.05 versus normoxia). B, EMSA performed with nuclear extracts from primary neurons exposed to normoxia or 0.5% O 2 and 32 P-labeled probes containing either wild-type (Wt) or mutated (Mut) HRE from Rage or Epo genes. For supershift assay, binding reactions contained antibody against HIF-1␣. The positions of the putative HIF, constitutive (C), and supershift (SS) complexes are indicated by arrowheads. C, chromatin immunoprecipitation carried out with HIF-1␣ F/F and HIF-1␣ ⌬/⌬ neurons. Cells were harvested at 0, 1, 4, and 8 h of hypoxia (0.5% O 2 ). Immunoprecipitation (IP) was performed with monoclonal anti-HIF-1␣ antibody. Coprecipitated genomic DNA fragments were evaluated by PCR using primers flanking the HRE-containing region of the Rage promoter. HIF-1␣ ⌬/⌬ neurons were generated by AdCre infection as described under "Experimental Procedures." centration range of 0.1-200 nM, under normal oxygen/glucose conditions did not have any effect on cell viability (Fig.  6C). Additional experiments with higher doses of S100B (250 nM and 500 nM) showed mild toxicity, as cell viability decreased by 20%. Treatment of neurons with S100B (100 nM) for 1-24 h prior to OGD did not have any protective effect against this stress (Fig. 6D). In contrast, treatment of neurons with S100B immediately after OGD (0.1-200 nM, 24 h) led to protection (Fig. 6E). Pro-survival effects were noted with 10 nM, and maximal effects were observed with 50 -100 nM S100B (ϳ40% increase of cell survival). Higher S100B concentrations (250 or 500 nM) failed to exert any protective effect against OGD. To determine whether the neuroprotective effects of S100B require the RAGE receptor, neurons were co-treated with a specific RAGE neutralizing antibody (50 g/ml) and S100B immediately after OGD. This led to a complete blockade of the pro-survival effect of S100B confirming that this protective outcome requires RAGE signaling. In contrast, the toxic effect of relatively high doses of S100B (250 or 500 nM) was not affected by this antibody (Fig. 6C).

DISCUSSION
Injuries to the mammalian brain, such as hypoxia and ischemia, can induce both neurodegenerative and neuroregenerative responses. An understanding of the specific pathways involved is essential to design therapeutic interventions that inhibit neurodegeneration/death pathways, while promoting neuroregenerative/plasticity pathways. In the present study, we employed in vivo and in vitro models of hypoxia and ischemia to investigate the response of RAGE to cerebral injury. We show that RAGE is expressed at very low levels under normal conditions in mouse brain, but is significantly up-regulated in response to prolonged, mild hypoxia or hypoxia-ischemia; in both conditions, RAGE expression colocalizes primarily with neurons. We also demonstrate that this response is mediated by the interaction of the transcription factor HIF-1 with an HRE sequence on the RAGE promoter, and that HIF-1 mediated stimulation of neuronal RAGE following hypoxia-ischemia is part of the neuroprotective/neuroregenerative response that also involves the glial-derived RAGE ligand S100B. To the best of our knowledge this may represent a direct ligand-independent activation of the RAGE promoter by HIF-1. It is also possible that hypoxia can affect the release of specific RAGE ligands that can subsequently modulate RAGE expression (i.e. high-mobility group box 1).
The biology of RAGE has been predominantly studied in the context of diabetic complications in the periphery, and Alzheimer disease in the brain. In these situations, evidence indicates that RAGE-mediated signaling leads to pro-inflammatory gene expression, a mechanism believed to be involved in cellular dysfunction (5,6,40). Very limited reports have explored the function of RAGE in acute injury such as cerebral ischemia, a stress that also leads to extensive release of glia-derived S100B and high-mobility group box 1, suggesting an effective stimulation of RAGE in this scenario (18 -20, 41). Excessive accumulation of other RAGE ligands such as advanced glycation end products and amyloid ␤-peptide are not likely in the ischemic adult brain; however, as a consequence of senescence or diabetes, advanced FIGURE 6. Effect of S100B on neuronal survival to OGD. A, cell viability was assessed in primary cortical neurons subjected to OGD (60 min) followed by recovery (0 to 24 h). B, representative RAGE immunoblot in neurons subjected to OGD followed by reperfusion (0 -12 h). C, primary neurons were incubated for 24 h with the indicated concentrations of S100B in the presence of anti-RAGE antibody or non-immune IgG under normal glucose-oxygen conditions. *, p Ͻ 0.05 versus untreated group. D, neurons were pretreated with S100B (100 nM) in the presence of anti-RAGE antibody, then they were subjected to OGD (60 min) followed by reperfusion (24 h). E, neurons were exposed to OGD (60 min) followed by 24 h of reperfusion. Treatment with S100B (0 -500 nM) and anti-RAGE antibody was performed at the onset of reperfusion; *, p Ͻ 0.05 versus OGD. In all cases, cell death was determined using 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assay, and data expressed as percentage of cell survival (mean Ϯ S.D., n ϭ 4 -6). C and N ϭ normal glucose/ oxygen conditions, H ϭ 0.5% O 2 for 24 h. DECEMBER 14, 2007 • VOLUME 282 • NUMBER 50 glycation end product levels do increase and can potentially elicit RAGE-dependent responses (42). In the present study, we found a sustained induction of Rage mRNA and protein levels after cerebral hypoxia/ischemia. This response was confined to the ischemic hemisphere, whereas the contralateral hemisphere showed unchanged RAGE expression. Additionally, our results showed increased RAGE immunolabeling in neurons located in the ischemic hemisphere at 24 and 48 h of recovery, which is in agreement with a previous study demonstrating increased Rage mRNA by in situ hybridization in ischemic neurons at 72 h of recovery from brain hypoxia-ischemia in rats (18). However, the possibility that other cell types express this receptor at different stages during the evolution of the infarct cannot be ruled out. In this regard, in contrast to our findings, Toth et al. (29) recently reported RAGE immunolabeling not only in neurons but also in endothelial and glial cells in diabetic and normal mouse brain. The reasons for this discrepancy are currently unknown but may be related to differences in primary antibodies and/or staining protocols that could lead to variations in labeling effectiveness. In our study, specificity of RAGE staining was confirmed by showing negligible staining in ischemic RAGE-null mice.

HIF-1 Regulates RAGE Expression during Hypoxia
Insight into the potential role of RAGE in the pathophysiology of stroke was obtained using mice genetically deficient for RAGE. Our results showed that during early reperfusion (12 h), RAGE-null animals displayed comparable infarct size to wildtype mice. At 24 h of recovery, there was no increase in infarct size in wild-type; however, RAGE-null animals developed larger infarct volumes. Brain damage was not evaluated at later intervals and the possibility exists of a further evolution of the infarction in both RAGE-null and wild-type animals. Nevertheless, these initial results indicate that activation of RAGE-dependent processes in wild-type animals may prevent ongoing cell death in the peri-infarct area halting spread of the infarction. Our data do not exclude the possibility that germline deletion of RAGE affects systemic parameters like vascular tone that might indirectly account for the greater ischemic brain injury.
In stark contrast to our results in brain ischemia, in a rodent model of myocardial ischemic-reperfusion injury, Bucciarelli et al. (17) found that pharmacological blockade or genetic ablation of RAGE resulted in attenuation of heart ischemic damage and dysfunction, suggesting that in this context activation of RAGE-dependent signaling has a deleterious effect. Further reflecting the complexity of RAGE function, in a paradigm of peripheral nerve injury, augmented RAGE expression was found associated with axons and infiltrating macrophages, and in this context, inhibition of RAGE resulted in deficient nerve regeneration (43,44). However, in experimental diabetic neuropathy, genetic ablation of RAGE led to partial protection from diabetes-induced nerve functional deficits (45). Therefore, these studies suggest that the outcome of RAGE activation in vivo varies depending on the spatiotemporal expression of this receptor and the extent of ligand availability.
The diverse biological effects of ligand-RAGE interaction have also been uncovered in cell culture studies. Whereas evidence exists that at least in some cultured cells, activation of RAGE induces the expression of proinflammatory genes (i.e. Cox-2 and Tnf-1␣), RAGE can also regulate potentially neuroprotective genes such as bcl-2 and Creb (9, 13, 46 -48). Our in vitro experiments involving oxygen-glucose deprivation showed a dual effect of exogenous S100B on neuronal survival, being neuroprotective at low doses and neurotoxic at high doses. These results are in accordance with previous studies showing survival effects of S100B at relatively low concentrations against serum deprivation, glutamate, N-methyl-D-asparatate or amyloid-␤ peptide neurotoxicity (9,11,12). The requirement of RAGE for S100B protective effects was demonstrated in some of these paradigms (9,11). Consistent with these findings, we found that the neuroprotective effect of low concentrations of S100B against OGD was abrogated by RAGEblocking antibodies. However, discrepancies exist regarding the involvement of RAGE in S100B toxicity. Our results using RAGE neutralizing antibodies indicated no involvement of RAGE in S100B-induced cell death in primary neurons; however, Huttunen et al. (9) reported contrary results using neuroblastoma cells overexpressing signal-deficient RAGE. An important consideration regarding the nanomolar range of action reported for S100B by the present and previous reports (11,12,49) is its apparent discrepancy with other studies that indicate a moderate binding affinity (micromolar range) of RAGE for this ligand (50,51). This raises the uncertainty of whether low concentrations of S100B can effectively bind to RAGE, and leads to the consideration that low concentrations of S100B might mediate neuroprotection by pathways not involving direct RAGE signaling but still requiring RAGE indirectly. However, because these binding studies were performed using sRAGE immobilized on a sensor chip surface, it is possible that the kinetics of RAGE-S100B interaction are quite different at the physiological conditions of the plasma membrane.
The novel contribution of our studies regarding the underlying mechanism of hypoxia-induced RAGE expression is that, like many hypoxia inducible genes, this receptor is regulated by the transcriptional activator HIF-1. We showed that Rage mRNA levels were induced by hypoxia and pharmacologic activators of HIF-1, namely DFO and DMOG. In neurons with decreased HIF-1 activity generated by a Cre-lox based approach, the up-regulation of RAGE by hypoxia or DFO was abolished, and a similar effect was observed with Vegf, a well characterized HIF-1-regulated gene. Analysis of the proximal promoter region (ϳ2.5 kb) of the mouse Rage gene showed at least one putative HIF-1 binding site with the consensus sequence 5Ј-ACGTG-3Ј. When this region was cloned into a luciferase reporter vector and transfected into 3T3 fibroblast, luciferase expression was induced by hypoxia, DFO, or DMOG. Deletion analysis revealed a unique functional hypoxia response element located ϳ1000 bp upstream of the proposed transcription start site. Mutation of this potential HIF-1 binding site abrogated hypoxia-induced luciferase activation. Moreover, DNA binding activity of HIF-1 to this putative HRE was demonstrated by EMSA analysis and chromatin immunoprecipitation assay in hypoxic neurons. Although these data conclusively establish that HIF-1 regulates RAGE expression during hypoxia, a single HRE might not be sufficient to convey efficient hypoxia sensitivity to this promoter. Thus, it is conceivable that in vivo other transcription factors are involved by acting synergistically with HIF-1 to regulate hypoxic/ischemic RAGE expression. For instance, functional NF-B binding sites have been identified in the human RAGE promoter that can be also important for the transcriptional regulation of this gene especially in ischemic situations (52). The link between HIF-1 and RAGE described here may have ramifications for other aspects of hypoxia, such as tumor development, in which RAGE-mediated signaling events have been implicated (53). As many tumors show elevated expression of HIF-1␣, caused by hypoxia inherent to growing tumors and/or genetic mutations (54), it will be interesting to investigate whether the elevated levels of HIF-1 underlie the increased RAGE expression found in some tumors. In conclusion, the results of this study indicate that activation of the HIF-1/RAGE/S100B pathway in neurons results in a neuroprotective response to hypoxic/ischemic stress and could be a target for strategies promoting post-ischemic survival.