Transient Changes in Oxygen Tension Inhibit Osteogenic Differentiation and Runx2 Expression in Osteoblasts*

Vascular disruption following bony injury results in a hypoxic gradient within the wound microenvironment. Nevertheless, the effects of low oxygen tension on osteogenic precursors remain to be fully elucidated. In the present study, we investigated in vitro osteoblast and mesenchymal stem cell differentiation following exposure to 21% O 2 (ambient oxygen), 2% O 2 (hypoxia), and < 0.02% O 2 (anoxia). Hypoxia had little effect on osteo - genic differentiation. In contrast, short-term anoxic treatment of primary osteoblasts and mesenchymal precursors inhibited in vitro bone nodule formation and extracellular calcium deposition. Cell viability assays revealed that this effect was not caused by immediate or delayed cell death. Microarray profiling implicated down-regulation of the key osteogenic transcription factor Runx2 as a potential mechanism for the anoxic inhibition of differentiation. Subsequent analysis revealed not only a short-term differential regulation of Runx2 and its targets by anoxia and hypoxia, but a long-term inhibition of Runx2 transcriptional and protein levels after only 12–24 h of anoxic insult. Furthermore, we present evidence that Runx2 inhibition may, at least in part, be because of anoxic repression of BMP2 , and that restoring Runx2 levels during anoxia by pretreatment with recombinant BMP2 rescued the anoxic inhibition of differentiation. Taken together, our findings indicate that The approach undertaken in this study to use the (cid:5) 2 S.D. criterion to identify a subset of genes with “significant” up- or down-regulation, and then to use comparisons of -fold change values to directly compare the responses at (cid:1) 0.02 and 2% O 2 for specific genes and/or functional groupings of genes. The overall effect of oxygen level on transcriptional response was assessed by comparing the variances in -fold change at (cid:1) 0.02 and 2% O 2 using Levine’s test for homogeneity of variances. For PCR data, -fold change values in individual genes were compared via analysis of variance or paired t tests (with Bonferroni adjustments to control the Type I error rate). All statistical analyses were performed using Quantitative Real-time PCR— For selective microarray confirmation and to investigate gene expression during the 28-day differentiation period, we performed quantitative real-time polymerase chain reaction (QRT-PCR). We obtained cDNA by reverse transcription of 1 (cid:3) g of DNase-treated total RNA from each sample using random hexamer priming in 50- (cid:3) l reactions according to the manufacturer’s recommen- dations (Taqman® Reverse Transcription Reagent Kit, Applied Biosystems, Foster City, CA). We proceeded with QRT-PCR using the Applied Biosystems Prism® 7900HT Sequence Detection System. A non-multi-plexed SYBR® Green assay in which each cDNA sample was evaluated at least in triplicate 20- (cid:3) l reactions was used for all target transcripts.

It has long been known that hypoxia is a prominent component of the microenvironment in both bony and soft tissue injury (1)(2)(3)(4). Interruption of vascular flow with fracture or surgical osteotomy results in a transient hypoxic gradient within the wound, with oxygen tension falling to 0 -2% in the central wound region (1)(2)(3)(4)(5)(6). Thus, the complex osteogenic regenerative response that ensues following injury begins in the setting of reduced oxygen (7). In addition to inflammatory cell recruitment, matrix processing, and angiogenesis, the activity of mesenchymal osteoblast precursors is a key component of this cascade (6,8). Both committed resident osteoblasts and recruited precursor cells are exposed to the hypoxic post-injury milieu. Given that normal osteoblast function is critical to the deposition of mineralized matrix, the hallmark of successful fracture repair, it is important to understand the response of osteoprogenitors to a hypoxic microenvironment.
Recent investigations have yielded significant insights into the transcriptional regulation of osteogenic differentiation (9 -11). Primary cell culture models and established cell lines that recapitulate osteoblast maturation in vitro have allowed the identification of closely regulated transcription factors (e.g. Runx2, Osx, and members of the Ets family) and osteogenic markers (e.g. collagens, alkaline phosphatase, and osteocalcin) specific to each stage of the osteoblast life cycle (12). Runx2, a member of the runt family of transcription factors, is both necessary and sufficient for osteoblast differentiation (9,13). Runx2 activity promotes the expression of a number of osteogenic markers including collagen I, osteopontin, and osteocalcin by binding to responsive elements within the promoters of these genes (11,12). Its clinical significance is highlighted by the observation that humans or mice heterozygous for Runx2 mutations develop cleidocranial dysplasia, whereas postnatal impairment of Runx2 results in osteopenia (10).
The effect of reduced oxygen on genes involved in osteoblast proliferation and differentiation remains unclear, in part because of the sometimes conflicting results observed for primary cultures versus established cell lines and for different oxygen tensions. An early study of the immortalized osteoblast-like cell line MC3T3-E1 exposed to chronic 10% O 2 demonstrated a decrease in both cell proliferation and alkaline phosphatase expression, an early to middle marker of osteoblast differentiation, relative to cells at 21% O 2 (14). Additional investigations in primary osteoblast-enriched cultures from rat calvaria and human periodontal ligament confirmed that hypoxic (5-10%) inhibition of alkaline phosphatase was not limited to the MC3T3-E1 line but was applicable to primary cells as well (15,16). However, in contrast to the effect seen in MC3T3-E1, an increase in proliferation was observed in both primary cell types, perhaps alluding to a reciprocal hypoxic effect on proliferation and differentiation in primary osteoblasts. Recently, a human osteosarcoma cell line continuously exposed to 2% O 2 for up to 96 h was shown to express diminished levels of the transcription factor RUNX2 and the late bone marker osteocalcin (17). Whereas this study on transformed cells provided important initial insight into the hypoxic regulation of osteogenic gene expression, a broad understanding of the hypoxic osteoblast transcriptional response remains to be elucidated. Furthermore, it may be difficult to extrapolate tumor cell line derived conclusions to primary osteoblasts given that Runx2 expression may be dissociated from osteogenic potential in immortalized cell lines such as MC3T3-E1 (18,19).
Nevertheless, evidence is accumulating that hypoxia regulates osteoblast-secreted growth factors implicated in the postinjury microenvironment. For example, cytokines potentially involved in osteoblast-endothelial interactions are oxygen-sensitive. Hypoxic osteoblasts exhibit increased expression of vascular endothelial growth factor-A (VEGF-A), 1 perhaps the most potent and specific of the angiogenic cytokines (20). VEGF-A appears to regulate osteoblast function by coupling ossification with cartilage resorption and angiogenesis during endochondral bone formation (21). Additional hypoxia-regulated modulators of osteoblast function that also act as endothelial mitogens include members of the transforming growth factor-␤ (TGF-␤), insulin-like growth factor, and fibroblast growth factor families (22)(23)(24)(25). Their elaboration by osteoblasts further suggests an important role for this cell type in regulating the angiogenic response that is critical to fracture repair (26).
Given that reduced oxygen is a major component of the fracture microenvironment, we sought to gain a better understanding of its effects on osteoblast gene expression and function. We first examined whether short-term hypoxic and anoxic insults, analogous to those observed in the post-fracture hypoxic gradient, would affect in vitro osteogenic differentiation of primary osteoblasts and multipotent mesenchymal precursors. We demonstrate that a 12-24-h reduction from 21 to Ͻ0.02% O 2 suppressed long-term osteogenic differentiation without altering cell viability. In addition, we present evidence implicating differential regulation of Runx2 expression by anoxia as a potential mechanism for this effect.

MATERIALS AND METHODS
Calvarial Osteoblast, MC3T3-E1, and Bone Marrow-derived Mesenchymal Cell Cultures-Primary mouse calvarial osteoblast cultures were established based on described techniques (27). Briefly, frontal and parietal bones from 6-day-old CD-1 mice were stripped of their pericranium and dura mater. Calvaria from five to six pups were pooled, minced, and washed with sterile phosphate-buffered saline. Osteoblasts were released by five sequential 20-min digestions with 0.1% collagenase (Invitrogen). Digestions were stopped with 5 volumes of ␣-minimal essential medium, 10% fetal bovine serum. Fractions 2-5 were collected, pelleted by centrifugation, and resuspended in ␣-minimal essential medium, 10% fetal bovine serum plus penicillin/streptomycin. Cells were allowed to attach and expand for 48 h, then replated for experimentation without further passaging.
The osteoblast-like immortalized cell line MC3T3-E1 (American Type Culture Collection, Manassas, VA) was maintained in subconfluent cultures in ␣-minimal essential medium, 10% fetal bovine serum plus pencillin/streptomycin. Human bone marrow-derived mesenchymal stem cells (MSCs) were purchased from Cambrex BioScience (Walkersville, MD) and maintained in an undifferentiated state in manufacturer recommended medium (Poietics TM MSCGM TM Mesenchymal Stem Cell Medium) that was replaced every 2 days. Early passage (Ͻ3) cells were utilized for experimentation. All cells were maintained at 21% O 2 and 5% CO 2 in humidified incubators at 37°C prior to hypoxia or anoxia experiments.
Cell viability was assessed by Hoescht and propidium iodide staining as previously described (28). Briefly, osteoblasts and MSCs were incubated for 15 min with 2 g each of bis-benzamide (Hoescht stain number 33342, Sigma) and propidium iodide (Sigma) per 1 ml of medium. The number of non-viable cells was determined by scoring three randomly selected low-magnification fields for double-stained cells (indicating loss of membrane integrity). The ratio of viable to total cells was then determined. For each sample, ϳ500 cells were scored in a blinded fashion, and experiments were repeated three times. Viability data were confirmed by trypan blue exclusion and cell counts of trypsinized MSCs and osteoblasts. In addition, MSCs released by trypsinization were incubated with propidium iodide in conjunction with the TACS TM Annexin V-FITC Apoptosis Detection kit (R&D Systems, Minneapolis, MN) for flow cytometry-based detection of apoptotis according to the manufacturer's directions.
Exposure to Ambient, Hypoxic, and Anoxic Conditions-At 95% confluence, calvarial osteoblasts and MC3T3-E1 cells underwent media change with fresh ␣-minimal essential medium, 10% fetal bovine serum. MSCs received fresh MSCGM TM . After 6 h (to minimize the effects of medium replacement upon gene expression), cells were exposed to 21% (ambient oxygen), 2% (hypoxia), or Ͻ0.02% O 2 (anoxia) for 0, 3, 6, 12, and 24 h. Oxygen was supplanted by infusion of a 5% CO 2 , 95% nitrogen gas in 37°C humidified hypoxia work stations (Bactron Anaerobic/Environmental Chamber, Sheldon Corporation, Cornelius, OR, and Ruskinn Microaerophilic Work station, Ruskinn Technology, Leeds, United Kingdom). These chambers employ an intermediate "pass-box" to minimize mixing of ambient air with the internal chamber environment as well as real-time temperature and oxygenation level monitoring.
After hypoxic or anoxic treatments, cells underwent total RNA isolation using TRIzol reagent (Invitrogen) or were returned to 21% O 2 for long-term differentiation. For samples derived from cells at 2 or Ͻ0.02% O 2 , RNA collections were done within the hypoxia chamber to limit reoxygenation effects on gene expression.
Long-term Differentiation Experiments and Bone Nodule Staining-Following exposure to 21, 2, or Ͻ0.02% O 2 , cells underwent in vitro osteogenic differentiation in 21% O 2 via basal medium supplementation with 1 M dexamethasone, 5 mM ␤-glycerophosphate, and 100 g/ml ascorbic acid. Medium was replaced immediately after initial treatment with reduced or ambient oxygen, and every 2 days thereafter. For experiments in which osteoblasts were cytokine stimulated, we used a single dose of BMP2 (100 ng/ml), TGF-␤1 (5 ng/ml), TGF-␤2 (5 ng/ml), insulin-like growth factor-1 (20 ng/ml) or vehicle immediately prior to anoxic exposure for 24 h (all cytokines from R&D Systems).
Cultures were maintained for 28 days for gene expression analysis or assessment of mineralized matrix deposition by bone nodule staining via 0.5% alizarin red or the von Kossa method for calcium phosphates, and digitally photographed at low and high magnifications. Low magnification (whole plate) von Kossa images were analyzed using Scion Image Software version 4.0.2 (Scion Corp., Frederick, MD) to determine the area of bone nodule staining. The values reported for 2 and Ͻ0.02% O 2 -treated plates represent relative staining compared with parallel cultures at 21% O 2 that were normalized to a value of "one." In the same manner, we quantified staining of the three cell types after 3, 6, 12, and 24 h of Ͻ0.02% O 2 relative to parallel cultures that had not been anoxia treated. We also performed a colorimetric determination of calcium concentration using a Calcium Reagent Set (Biotron Diagnostics, Hemet, CA) following the manufacturer's instructions. This kit provides a quantitative assessment of calcium concentration based on the ability of calcium to react with cresolphthalein complexone in 8-hydroxyquinolone to form a complex purple in color that absorbs at 570 nm.
Gene Profiling Using Microarrays-RNA from mouse calvarial osteoblasts exposed to 21, 2, or Ͻ0.02% O 2 for 24 h was purified and reverse transcribed for microarray analysis on cDNA chips from the Stanford Functional Genomics Facility. Each array contains ϳ42,000 mouse elements, representing over 30,000 unique genes. Detailed protocols for probe synthesis and hybridization are available on-line. Briefly, 30 g of total RNA was used for single-stranded cDNA probe synthesis incorporating aminoallyl-dUTP, which was coupled with either Cy3 (for 21% O 2 samples) or Cy5 (for 2 or Ͻ0.02% O 2 samples). Type I experiments were performed in which 2 and Ͻ0.02% O 2 samples were each hybridized with pooled 21% O 2 reference samples for 16 h at 65°C. Arrays were washed, scanned, and analyzed using a GenePix Scanner and software (Axon Instruments, Foster City, CA). After gridding, array data were uploaded to the Stanford Microarray Data base or SMD (29). Total RNA from three independent experiments was pooled for each array probe, and array hybridizations were performed in duplicate.
Microarray Data Analysis-Using on-line software from SMD, data points that met the following spot quality criteria were selected for analysis: spot regression correlation Ͼ0.7, mean channel intensity Ͼ2.5 median background, and "spot flag" and "failed" filters ϭ 0. These user-selectable criteria are intended to exclude spots with non-uniform, dim, or otherwise unreliable signals. Relative changes in gene expression were evaluated by -fold change (i.e. Ͻ0.02 or 2% O 2 relative to 21% O 2 reference) as determined from the log 2 of red/green normalized ratio reported by SMD, as previously described (30). Both the -fold change values and the standard deviation (S.D.) of -fold changes for a particular array have been used to determine whether expression changes are significantly different from array background (29,30). Describing the number of genes whose -fold changes fall outside the range of the mean Ϯ 2 S.D. allows comparisons between populations that are not identical, e.g. those with different mean -fold changes or distributions. In contrast, reporting absolute -fold change allows comparison of the same gene across two populations (i.e. relative -fold change of gene A at Ͻ0.02 versus 2% O 2 ) or two or more genes within the same populations (i.e. gene A versus gene B at 2 or Ͻ0.02% O 2 ).
The approach undertaken in this study was to use the Ϯ2 S.D. criterion to identify a subset of genes with "significant" up-or downregulation, and then to use comparisons of -fold change values to directly compare the responses at Ͻ0.02 and 2% O 2 for specific genes and/or functional groupings of genes. The overall effect of oxygen level on transcriptional response was assessed by comparing the variances in -fold change at Ͻ0.02 and 2% O 2 using Levine's test for homogeneity of variances. For PCR data, -fold change values in individual genes were compared via analysis of variance or paired t tests (with Bonferroni adjustments to control the Type I error rate). All statistical analyses were performed using JMP 5.0.1 (SAS Institute, Cary, NC).
Quantitative Real-time PCR-For selective microarray confirmation and to investigate gene expression during the 28-day differentiation period, we performed quantitative real-time polymerase chain reaction (QRT-PCR). We obtained cDNA by reverse transcription of 1 g of DNase-treated total RNA from each sample using random hexamer priming in 50-l reactions according to the manufacturer's recommendations (Taqman® Reverse Transcription Reagent Kit, Applied Biosystems, Foster City, CA). We proceeded with QRT-PCR using the Applied Biosystems Prism® 7900HT Sequence Detection System. A non-multiplexed SYBR® Green assay in which each cDNA sample was evaluated at least in triplicate 20-l reactions was used for all target transcripts. Expression values were normalized to 18S or ␤-actin. QRT-PCR primers were designed using Primer Express version 2.0.0 (Applied Biosystems) and tested to confirm appropriate product size and optimal concentrations.
Western Blot Analysis-Mouse osteoblasts underwent osteogenic differentiation following exposure to 21 or Ͻ0.02% O 2 as described above. Cells were lysed in 9 M urea, 75 mM Tris-HCl, pH 7.5, and 0.15 M ␤-mercaptoethanol. After a brief sonication, lysates were centrifuged to remove debris and quantitated. Forty micrograms of protein were electrophoresed on 3-8% gradient SDS-polyacrylamide gels and transferred to nitrocellulose membranes. Membranes were probed using commercial Runx2 and ␤-actin antibodies (Santa Cruz Biotechnology) and the ECL detection kit (Amersham Biosciences).

Anoxia Inhibits Bone Nodule Formation by Calvarial Osteoblasts and Bone Marrow-derived Mesenchymal Stem Cells-
Both resident and recruited precursor populations involved in post-injury repair are transiently exposed to wound oxygen levels as low as 0 -2% O 2 (1-7). Without the re-establishment of nutrient and O 2 delivery, tissue regeneration is not possible and necrosis ensues. To investigate the effect of oxygen deprivation on osteoprogenitor cells, we exposed primary mouse calvarial osteoblasts to 24 h of 21, 2, or Ͻ0.02% O 2 . The cells were subsequently maintained in osteogenic differentiation media for 28 days at 21% O 2 and assayed for bone nodule formation. As demonstrated in Fig. 1, a and b, von Kossa and alizarin red staining for the Ͻ0.02% O 2 group were diminished relative to the 21 and 2% O 2 groups. High magnification views revealed fewer and smaller mineralized nodules in the Ͻ0.02% O 2 group than the two other groups (Fig. 1, c and d).
Because primary cultures of calvarial osteoblasts likely include osteoprogenitors at different stages of commitment (31), we next sought to determine whether reduced O 2 altered the long-term differentiation of uncommitted cells with osteogenic potential. Primary human bone marrow-derived MSCs have previously been shown to be capable of osteogenic, adipogenic, chondrogenic, and myogenic differentiation (32). We exposed MSCs to 24 h of 21, 2, or Ͻ0.02% O 2 followed by osteogenic differentiation in 21% O 2 for 28 days. Similar to the results obtained for primary osteoblasts, bone nodule formation as assayed by von Kossa, and alizarin red staining was diminished for the Ͻ0.02% O 2 group relative to the 21 and 2% O 2 groups (Fig. 2, a and b).
Given that differences in response to hypoxia have been observed for primary versus immortalized cells, we exposed primary osteoblasts, MSCs, and the osteoblast-like cell line MC3T3-E1 to 24 h of 21, 2, and Ͻ0.02% O 2 , followed by 28 days of differentiation at 21% O 2 . For each cell type, we estimated the degree of von Kossa staining for extracellular matrix deposition at 28 days using image analysis software. Relative staining of anoxia-treated cells was determined by dividing the calculated bone nodule area by that of parallel cultures that had not received anoxic treatment. In marked contrast to the anoxic inhibition of osteogenic differentiation observed for pri- mary osteoblasts and MSCs, the MC3T3-E1 cell line did not reveal a significant decrease in bone nodule formation following anoxic treatment (Fig. 3a).
As Little as 12 h of Anoxia Inhibits Long-term Osteogenic Differentiation-Because the severity of oxygen deprivation is a function of both duration and oxygen tension, we next examined whether a more brief period of anoxia would inhibit in vitro osteogenic differentiation. Osteoblasts, MSCs, and MC3T3-E1s were exposed to 0, 3, 6, 12, and 24 h of Ͻ0.02% O 2 and subsequently differentiated for 28 days at 21% O 2 . We then estimated the bone nodule area as above (relative to cultures that had not been anoxia treated). Osteoblasts and MSCs exposed to 24 h of Ͻ0.02% O 2 demonstrated an ϳ70 -80% reduc-tion in bone nodule formation compared with cultures at 21% O 2 , whereas 12 h of anoxia resulted in a ϳ60% reduction (Fig.  3b). Shorter anoxic exposure periods (3 to 6 h) did not inhibit bone nodule formation for osteoblasts and MSCs. Consistent with the results obtained above, bone nodule formation by MC3T3-E1 cells was not affected at any of the durations tested (Fig. 3b). In addition, using a cresolphthalein complexonebased colorimetric assay we quantitatively determined the calcium concentration in MSC cultures treated for 24 h with 21 or Ͻ0.02% O 2 followed by up to 28 days of differentiation. Consistent with the bone nodule quantitation data, calcium concentration was 5-fold lower in anoxia-treated MSC cultures at 28 days relative to 21% O 2 cultures (Fig. 3c).
Anoxic Inhibition of Osteogenic Differentiation Is Not Caused by Increased Cell Death-As a first step to determine if observed differences in long-term differentiation between 21 and Ͻ0.02% O 2 groups were a result of alterations in cell viability at each oxygen concentration, we performed Hoescht and propidium iodide staining of MSCs to determine the proportion of nonviable (double-stained) cells. No significant differences in viability were observed immediately after 3, 6, 12, or 24 h of anoxic exposure, or after 2 days of reoxygenation following 24 h of anoxia (Fig. 4a). To determine whether there was a change in viability in long-term culture, we exposed MSCs to 24 h of Ͻ0.02% O 2 , then returned them to 21% O 2 for up to 28 days of differentiation. We observed no alterations in cell viability by either visual inspection of cultures stained as above, or by FACS analysis of cells incubated with propidium iodide and the early apoptosis marker annexin V (Fig. 4b). Viability assessment by trypan blue exclusion and cell count confirmed these results, which were similar for primary calvarial osteoblasts (data not shown). These data suggested that the differences in bone nodule formation described above could not be accounted for by alterations in cell viability immediately following anoxic treatment, in the first 2 days of reoxygenation, or during longterm culture.
Anoxia and Hypoxia Result in Distinct Transcriptional Profiles in Osteoblasts-Having determined that brief exposure to anoxia, but not hypoxia, inhibited in vitro osteogenic differentiation, we sought to gain a better understanding of the transcriptional changes underlying this effect. Given that 24 h of anoxic treatment exhibited the most marked inhibition of bone nodule formation, we used cDNA microarrays to compare the gene expression profiles of mouse osteoblasts cultured for 24 h at Ͻ0.02 or 2% O 2 relative to identical cells maintained at 21% O 2 . Examination of overall gene expression patterns revealed that exposure to Ͻ0.02% O 2 induced a larger range of relative expression as compared with 2% O 2 (31.8-fold down-regulation to 55.4-fold up-regulation for Ͻ0.02% O 2 , versus 5.0 to 13.1 for 2% O 2 ). This was reflected by a greater variance in -fold changes in the Ͻ0.02% than 2% O 2 groups (variance ϭ 3.62 for Ͻ0.02% O 2 versus 0.17 for 2% O 2 , p Ͻ 0.001).
We hypothesized that the markedly larger range of relative expression for Ͻ0.02 versus 2% O 2 corresponded with the more severe deprivation of oxygen with anoxic exposure. Directly comparing the effects of anoxia versus hypoxia on the relative expression of known oxygen-sensitive genes confirmed this. For example, in a subset of genes with a change in relative expression greater than Ϯ2 S.D. from the array sample mean (Table  I), the largest functional grouping comprised 10 glycolytic enzymes (represented as 12 isoforms). This finding was consistent with a shift to a primarily anaerobic glycolytic mode of metabolism observed in many cell types with oxygen deprivation (33). We noted a more marked activation of all glycolytic pathway genes at Ͻ0.02 than 2% O 2 (Table I). This effect was also true for VEGF-A, adrenomedullin, and ␤ 1 -integrin, all of which have For each cell type, the degree of von Kossa staining was estimated using image analysis software and expressed relative to parallel cultures kept at 21% O 2 (as described under "Materials and Methods"). Osteoblasts and MSC cultures, but not MC3T3, demonstrated diminished bone nodule formation. b, relative bone nodule staining of each cell type after exposure to 0, 3, 6, 12, or 24 h of Ͻ0.02% O 2 . For osteoblast and MSC cultures, anoxia for 12 or 24 h significantly reduced bone nodule staining, whereas 3 or 6 h exposure did not. c, relative calcium concentration in MSC cultures exposed to 24 h of 21% O 2 (squares) or Ͻ0.02% O 2 (circles), then differentiated in osteogenic medium for the indicated days (d). Values represent the calcium concentration as compared with day 1 (i.e. immediately following the initial 24-h treatment period) measured by a colorimetric assay as described under "Materials and Methods." Each bar or point represents the mean Ϯ S.D. for three separate experiments. *, p Ͻ 0.05 for the indicated anoxic groups compared with the same cell type maintained at 21% O 2 (analysis of variance or paired t tests). been implicated in angiogenesis or modulating endothelial cell activity (Table I) (34 -37). Consistent with previous reports (38), VEGF-B and -C isoforms were not induced in our microarrays (data not shown).
Runx2 Transcriptional Levels Are Inhibited by Anoxia but Not 2% O 2 (Hypoxia)-To investigate the mechanism underlying the anoxic inhibition of osteogenic differentiation, we interrogated the array data sets for changes in osteogenic marker expression. Analysis of seven genes associated with various stages of osteoblast growth and differentiation revealed that they were all inhibited with anoxia (38 to 86% down-regulation relative to 21% O 2 ), with little or no overall change at 2% O 2 (33 down-regulation to 30% up-regulation) (Fig. 5a). Anoxia but not 2% O 2 down-regulated Runx2, three bone-associated collagens (ColI, ColIII, and ColVI), and two additional markers of osteogenic differentiation (Osteopontin and Osteoglycin) (Fig.  5a). These findings were consistent with our observation that anoxia (but not 2% hypoxia) inhibited osteogenic differentiation (Figs. 1-3).
Given that the transcription factor Runx2 is required for normal osteoblast differentiation, that anoxic insult was associated with down-regulation of Runx2 and several downstream targets, and that anoxia but not hypoxia inhibited bone nodule formation, we hypothesized that anoxic down-regulation of Runx2 could account for the observed inhibition of differentiation. Therefore, we sought to further define how oxygen regulates Runx2 expression in osteoblasts. To confirm our microarray findings, we treated calvarial osteoblasts in basal or differentiation media with 24 h of Ͻ0.02% O 2 and assayed Runx2 expression by QRT-PCR immediately after treatment (Fig. 5b, 24 h) as well as after 6 and 12 h of reoxygenation (Fig.  5b, 30 and 36 h). Runx2 expression was markedly down-regulated by anoxia, and only minimally recovered during the reoxygenation period examined. In contrast, osteoblasts exposed to 2% O 2 for 24 h did not demonstrate Runx2 inhibition during this period (Fig. 5c). Of note, our QRT-PCR data demonstrated an ϳ80% down-regulation of Runx2 following 24 h of Ͻ0.02% O 2 (Fig. 5b). The more conservative down-regulation noted by microarray (Fig. 5a) is consistent with prior analyses demonstrating that cDNA array data may underestimate the degree of change in expression when compared with QRT-PCR (39,40).

Anoxic Treatment Down-regulates Long-term Runx2 Expression and Alters Patterns of Collagen I and Osteocalcin
Expression-Because osteoblast differentiation is a multistage process that lasts up to one month in vitro, we next determined if transient hypoxia or anoxia resulted in long-term transcrip-tional changes that could affect osteogenic differentiation. Furthermore, we sought to determine whether Runx2 protein levels were affected by brief anoxic exposure. Following induction of Runx2, proliferating osteoblasts deposit collagen I into the extracellular space. Osteoblasts in later stages of differentiation orchestrate matrix calcification and produce the late osteogenic marker osteocalcin (12,13). We therefore used QRT-PCR to assay osteoblast gene expression for Runx2, ColI, and Osteocalcin during long-term differentiation. Osteoblasts were exposed for 24 h to 21 or Ͻ0.02% O 2 then placed in fresh maintenance or differentiation media for 28 days as described under "Materials and Methods." Gene expression was analyzed immediately following treatment (24 h), and at 7, 10, 14, and 28 days (Fig. 6).
Interestingly, in addition to inhibition of Runx2 expression immediately following 24 h of Ͻ0.02% O 2 , we noted a persistent attenuation of Runx2 in differentiating osteoblasts that had undergone initial anoxic treatment (Fig. 6a). In contrast, Runx2 transcription was induced strongly in differentiating osteoblasts that were maintained in 21% O 2 throughout the experiment. As expected, Runx2 expression remained relatively low in undifferentiated groups. Consistent with the tran-FIG. 5. Effect of anoxia versus hypoxia on expression of Runx2 in calvarial osteoblasts. a, calvarial osteoblast microarray analysis revealed uniform down-regulation of Runx2 and other markers of osteogenesis by anoxia but not 2% hypoxia (including alkaline phosphatase and bone-associated collagens, which are downstream effectors of the Runx2 pathway). y axes represent the percentage change in gene expression from 21% for 2% O 2 (gray bars) and Ͻ0.02% O 2 (black bars). In a, OP, osteopontin; and OG, osteoglycin. To confirm our microarray findings, we performed QRT-PCR using mouse osteoblasts following 24 h of anoxic (b) or hypoxic (c) treatment, and after 6 -12 h of reoxygenation. In b, N,U indicates normoxic (21% O 2 ) undifferentiated cells, whereas A,U indicates anoxia-treated undifferentiated cells. In contrast, N,D and A,D cells were placed in osteogenic differentiation medium from the beginning of the experiment. In c, H,U denotes 2% hypoxia-treated undifferentiated osteoblasts, whereas H,D reflects hypoxia-treated cells in differentiation medium (described under "Materials and Methods"). N,U and N,D are as described above. *, p Ͻ 0.05 for A,D compared with N,D groups at the same time point (paired t tests). Each bar represents the mean Ϯ S.D. of three separate experiments. All values were normalized to 18 S expression.

FIG. 6. Relative expression in calvarial osteoblasts of Runx2 by QRT-PCR and Western blot (a) and collagen I (col I) (b) and osteocalcin (c) by QRT-PCR.
Mouse osteoblasts were exposed to 24 h of normoxic/ambient air (21% O 2 ) or anoxia (Ͻ0.02% O 2 ) and cultured for 28 days in the presence or absence of osteogenic differentiation media. N,U denotes normoxic (21% O 2 ) undifferentiated cells, whereas A,U indicates anoxia-treated undifferentiated cells. In contrast, N,D and A,D cells underwent 24 h of normoxic or anoxic exposure followed by osteogenic differentiation. †, p Ͻ 0.05 for A,U compared with N,U at 24 h (immediately following treatment). *, p Ͻ 0.05 for A,D compared with N,D groups at 7, 10, 14, and 28 days (paired t tests). Each bar represents the mean Ϯ S.D. for three separate experiments. All QRT-PCR values were normalized to 18 S expression. For Runx2 Western blot images, the normoxic band is shown on the left and the anoxic band on the right for each indicated time point pair (with the corresponding ␤-actin control). scriptional data, Runx2 protein levels were decreased following anoxic treatment, and remained low throughout the 28-day differentiation period (Fig. 6a).
Given that Runx2 binds cis-acting elements that activate ColI and Osteocalcin promoters, we sought to determine whether anoxic exposure altered the expression profiles of these well characterized genes expressed by differentiating osteoblasts (12,13). Analysis of ColI expression revealed an immediate anoxic inhibitory effect at 24 h (Fig. 6b). Furthermore, similar to the results obtained for Runx2, differentiating osteoblasts exposed to 24 h of anoxia demonstrated attenuated ColI expression at 10, 14, and 28 days relative to cells that had not been anoxia-treated (Fig.  6b). Type I collagen expression in undifferentiated cells gradually decreased over 28 days, consistent with long-term culture of post-confluent osteoblasts (41).
Transcription of the late osteogenic marker osteocalcin was also inhibited in differentiating osteoblasts throughout the 28 days following anoxic exposure (Fig. 6c, A, D). Whereas lower than their normoxic counterparts, osteocalcin levels, nevertheless, increased in anoxia-treated osteoblasts between 14 and 28 days. This was consistent with the diminished (but not absent) bone nodule staining after anoxic treatment observed in Figs. 1 and 2.

Short-term Restoration of Runx2 Expression by BMP2 Administration Rescues Anoxic Inhibition of Differentiation-Be-
cause Runx2 is essential to osteogenic differentiation, several growth factors that influence osteoblast differentiation have, accordingly, been implicated in regulating its expression, including TGF-␤, insulin-like growth factor, and fibroblast growth factor family members (12,42). We therefore examined our microarray data set for mediators of osteogenesis to define the expression profile of potential upstream regulators of Runx2. Interestingly, we noted differential regulation of members of the transforming growth factor superfamily, including several bone morphogenic proteins (Fig. 7a). Specifically, we observed a 72% anoxic down-regulation of BMP2, an activator of Runx2 transcription (via Smad-mediated signaling) with a well established role in bone development and repair (40,43,44). TGF-␤2 was also significantly repressed by anoxia (by 80%) (Fig. 7a). In contrast to BMP2, TGF-␤ signaling has been shown to repress Runx2 transcription in differentiating primary calvarial osteoblasts (45). Additional paracrine and autocrine growth factors implicated in osteoblast differentiation (including insulin-like growth factor-1 and -2, and fibroblast growth factor-1 and -2 genes) were not significantly changed in our 2 and Ͻ0.02% microarray data sets (data not shown).
We hypothesized that anoxia-mediated inhibition of BMP2 expression may contribute to the anoxic down-regulation of Runx2. Because Bmp2 can up-regulate Runx2 expression in certain systems, we sought to determine whether restoring BMP2 levels during the period of brief anoxic exposure could "rescue" Runx2 expression in primary osteoblasts (9,46). We therefore administered recombinant BMP2 or vehicle to calvarial osteoblasts and exposed them immediately to 24 h of 21 or Ͻ0.02% O 2 . QRT-PCR analysis revealed that anoxic osteoblasts treated with vehicle exhibited a significant decrease in Runx2 mRNA level, consistent with our findings above (Fig.  7b). In contrast, anoxic osteoblasts pretreated with BMP2 exhibited similar Runx2 mRNA levels to normoxic vehicle-treated cells, indicating that BMP2 administration can "rescue" anoxia-mediated repression of Runx2 levels (Fig. 7b). Interestingly, Runx2 expression in anoxic BMP2-treated osteoblasts did not achieve levels observed in normoxic BMP-2-stimulated osteoblasts (Fig. 7b).
Given that our microarray analysis revealed a significant inhibition of TGF-␤2 with anoxia, we performed parallel cyto-kine stimulation experiments using both TGF-␤1 and TGF-␤2. TGF-␤1 treatment did not prevent inhibition of Runx2 by anoxia and, in fact, resulted in decreased Runx2 during normoxia as well, consistent with prior evidence implicating TGF-␤-mediated repression of Runx2 transcription (45). Also, TGF-␤2 administration resulted in anoxic inhibition of Runx2 similar to the vehicle-treated group (Fig. 7b). These findings confirmed that restoration of Runx2 mRNA levels was specific to Bmp2 pre-treatment.
Given that BMP2 administration "rescued" Runx2 mRNA levels despite anoxic exposure, we asked whether BMP2 could restore the ability of anoxia-treated osteoblasts to undergo long- FIG. 7. Exogenous administration of recombinant Bmp2 rescues the anoxic down-regulation of runx2 transcriptional levels and long-term inhibition of differentiation. a, microarray profiling of transforming growth factor-␤ and bone-morphogenic protein members at 2 and Ͻ0.02% O 2 . Relative to 21% O 2 , TGF-␤1 levels were not significantly altered, whereas TGF-␤2 was repressed under anoxia. Of the prosteogenic bmps, only BMP2 and (to a lesser degree) BMP4 were significantly down-regulated by anoxia. b, QRT-PCR of Runx2 following 24 h of calvarial osteoblast stimulation with the indicated cytokines or vehicle, in the presence or absence of anoxia. Only BMP2 resulted in sustained Runx2 mRNA levels despite anoxic exposure. All QRT-PCR values were normalized to 18 S expression. c, representative photomicrographs of von Kossa-stained osteoblasts that were pre-treated with the indicated cytokines during anoxic exposure, followed by 28 days of differentiation. Normoxic and anoxic vehicle-treated osteoblasts are also shown. Relative staining was quantitated as described under "Materials and Methods" using low-magnification (whole plate) images. term differentiation. Thus, we pre-treated osteoblasts with BMP2 prior to 24 h of anoxic exposure, then maintained them in osteogenic differentiation media for 28 days at 21% O 2 (without additional BMP2). Consistent with the aforementioned restoration of Runx2 mRNA levels by BMP2 (in the setting of anoxia), we observed normal bone nodule formation in BMP2 pre-treated osteoblasts (Fig. 7c). Administration of TGF-␤1 or TGF-␤2 (which had not prevented the anoxic down-regulation of Runx2) did not negate the anoxic inhibition of differentiation (Fig. 7c). These data further implicate Runx2 modulation, along with that of its upstream activator BMP2, as critical to the oxygen-dependent regulation of osteoblast differentiation. DISCUSSION As a first step to understanding osteoprogenitor behavior in the setting of a reduced oxygen microenvironment, we have investigated the effects of three oxygen levels (21, 2, and Ͻ0.02% O 2 ) on an immortalized cell line and two primary cell types involved in bone repair. We demonstrated that mouse primary osteoblasts and human bone marrow-derived mesenchymal cells, but not the cell line MC3T3-E1, were sensitive to anoxia. Anoxic (but not 2% hypoxic) insult was associated with diminished in vitro osteogenesis without inducing necrosis or apoptosis during 28 days of differentiation.
Differences in the long-term phenotypic response to 24 h of 2 versus Ͻ0.02% O 2 prompted us to explore the transcriptional profiles of osteoblasts following this brief treatment period. Anoxic osteoblasts not only had a more drastic transcriptional response (e.g. more marked glycolytic induction) but also activated pathways distinct from their hypoxic counterparts. For example, specific heat-shock elements were activated in anoxic cells, suggesting that Ͻ0.02% O 2 , but not 2% O 2 , induced a selective cell stress response. Of the numerous heat shock proteins examined, only Hspa5 and Hsp30 were up-regulated by anoxia. Whereas Hspa5 has been shown to be up-regulated under reduced oxygen, Hsp30 has not previously been implicated in hypoxic stress in mammalian cells. Interestingly, studies in Saccharomyces cerevisiae suggest that Hsp30 is unique among the heat shock protein family, for its induction does not appear to require the same transcriptional apparatus shared by other heat shock genes (47). Perhaps the selective activation of Hsp30 by anoxia in our system may suggest a separate pathway for activation in mammalian cells as well, although this would require subsequent investigation.
Perhaps the most interesting finding of our microarray analysis was a uniform down-regulation of Runx2, bone-associated collagens, and additional markers of osteoblast differentiation following 24 h of anoxia. As mentioned earlier, the behavior of osteoprogenitors in a low oxygen microenvironment has not been well characterized, and differences in proliferative and differentiation responses have been observed for primary versus immortalized osteogenic cells. This was true in our study as well; whereas primary calvarial osteoblasts and mesenchymal stem cells were quite sensitive to anoxic insult, the MC3T3-E1 cell line was not. Whereas the mechanism for this differential response to anoxia is unclear, this finding may allude to the inherent differences between primary and immortalized cells in their response to cell stress, and/or to the regulation of Runx2 expression, as discussed below (18,19).
We present evidence that down-regulation of Runx2, an essential mediator of osteoblast differentiation, may explain the inhibition of osteogenic differentiation by anoxia. Without normal Runx2 expression, osteoblasts cannot differentiate (9, 13), for Runx2 acts as a master gene that regulates the expression of a number of osteogenic markers including collagen I, osteopontin, and osteocalcin (11,12). The critical role of Runx2 is well established in vivo as well. Mice lacking both copies of Runx2 are deficient in functional osteoblasts, hypertrophic cartilage, and mineralized bone (48). Mutations in the Runx2 locus in humans results in cleidocranial dysplasia, characterized by the absence of clavicles, open fontanelles, and short stature, among other osteogenic defects (49).
In our in vitro system, calvarial osteoblast and MSC exposure to anoxia was associated with profound inhibition of Runx2, without evidence of significant recovery during 12 h of reoxygenation. Interestingly, we also observed a long-term down-regulation of Runx2 mRNA and protein (as well as transcriptional levels of two of its targets, ColI and Osteocalcin). In contrast, exposure to 2% O 2 neither inhibited bone nodule formation nor altered Runx2 expression. Furthermore, unlike in primary calvarial osteoblasts and MSCs, anoxia did not inhibit the differentiation of MC3T3-E1 cells. It has been shown in the MC3T3-E1 cell line that Runx2 expression may not correlate with differentiation potential or induction of osteogenic markers, i.e. MC3T3-E1 cells that express relatively high levels of Runx2 may not go onto form bone nodules or express osteocalcin (19). The dissociation between Runx2 expression and osteogenic differentiation in the MC3T3-E1 cell line (which was not inhibited by anoxia) underscores the importance of Runx2 as a potential mediator of anoxic inhibition of differentiation in primary osteoblasts.
To further investigate if Runx2 down-regulation is a critical step in the anoxic inhibition of differentiation, we examined whether abrogating the decrease in Runx2 expression that occurs with anoxia could restore the ability of osteoblasts to differentiate. Whereas Runx2 may be responsive to multiple signal transduction pathways, a complete understanding of up-stream regulators of Runx2 transcription remains to be determined (12,42). This prompted us to further focus our microarray analysis on known osteogenic mediators, and revealed differential regulation of members of the transforming growth factor superfamily. Specifically, we noted marked down-regulation of TGF-␤2 and BMP2, with no significant alterations in TGF-␤1 and several BMP family members. Whereas several isoforms of TGF-␤ are expressed by osteoblasts and have key roles in bone development and remodeling, the effect of TGF-␤ signaling upon Runx2 expression remains unclear. TGF-␤ has been shown to repress Runx2 transcription in differentiating primary calvarial osteoblasts while having the opposite effect in myoblasts (45,46). Whereas TGF-␤1 was not significantly altered by hypoxia or anoxia, we found that anoxia inhibited TGF-␤2 mRNA levels. Pre-treatment with either TGF-␤ isoform did not ameliorate the anoxic inhibition of Runx2 or osteogenic differentiation. In fact, TGF-␤1 administration resulted in decreased Runx2 levels in both normoxic and anoxic osteoblasts, and thus neither isoform was able to rescue the anoxic inhibition of differentiation. These data suggest that anoxic regulation of TGF-␤ is not a likely mediator of anoxic inhibition of Runx2 expression and osteoblast differentiation in our system.
In contrast to TGF-␤, BMP2 is a well established inducer of osteogenic differentiation in multiple cell types (12,50). Given our finding of significant anoxic down-regulation of BMP2, we further evaluated its potential role in anoxic regulation of Runx2 expression. Indeed, we found that treatment with BMP2 (but not TGF-␤1 or -␤2) just prior to anoxic exposure restored Runx2 expression to the levels observed with normoxic unstimulated osteoblasts. However, exposure of BMP2-treated osteoblasts to anoxia resulted in lower Runx2 expression levels compared with normoxic BMP2-stimulated cells. A possible explanation for the persistent anoxia-mediated decrease in Runx2 mRNA despite BMP2 administration is that anoxia may repress Runx2 transcription directly and indirectly, i.e. by in-hibiting the Runx2 promoter (through a transcriptional repressor) and by regulating important upstream inducers of its expression, such as BMP2.
Whether anoxic modulation of BMP2 transcription completely accounts for the reduction in Runx2 mRNA observed with anoxia (or is a distinct but contributing process) remains unknown. Nevertheless, rescuing Runx2 mRNA levels by exogenous BMP2 administration was sufficient to abrogate the long-term inhibition of differentiation. Thus, we propose that anoxic exposure has a deleterious affect on osteogenic differentiation by inhibiting Runx2 transcription directly and/or through a molecular pathway that involves BMP2 repression (Fig. 8).
In conclusion, anoxic insult resulted in diminished osteogenic differentiation in vitro, sustained inhibition of Runx2 levels, and a distinct transcriptional response compared with 2% O 2 . Why the hypoxic response differs greatly from that of anoxia may allude to the question of what constitutes physiologic or in vivo "normoxia." In vivo oxygenation has been shown to be closer to 5-6% oxygen than 21% in many normally perfused tissues, and even less in diseased states and injury (1)(2)(3)(4)(5)(6)(7)(8). Thus, 2% O 2 may represent a mild hypoxic insult such as that observed with a small injury or exercise-induced damage (in which tissue regeneration would be both desirable and possible). In contrast, cells facing the absence of oxygen secondary to complete vascular obstruction or more severe injury may negotiate a survival benefit by activating cell stress elements and anaerobic metabolism, while turning off potentially costly differentiation pathways such as those mediated by BMP2 and Runx2. Elucidating the osteoblast transcriptional response during transient periods of reduced oxygen provides a crucial step toward understanding osteoblast behavior in the context of compromised perfusion following injury.