Hypoxia Inhibits G1/S Transition through Regulation of p27 Expression*

Mammalian cellular responses to hypoxia include adaptive metabolic changes and a G1 cell cycle arrest. Although transcriptional regulation of metabolic genes by the hypoxia-induced transcription factor (HIF-1) has been established, the mechanism for the hypoxia-induced G1 arrest is not known. By using genetically defined primary wild-type murine embryo fibroblasts and those nullizygous for regulators of the G1/S checkpoint, we observed that the retinoblastoma protein is essential for the G1/S hypoxia-induced checkpoint, whereas p53 and p21 are not required. In addition, we found that the cyclin-dependent kinase inhibitor p27 is induced by hypoxia, thereby inhibiting CDK2 activity and forestalling S phase entry through retinoblastoma protein hypophosphorylation. Reduction or absence of p27 abrogated the hypoxia-induced G1 checkpoint, suggesting that it is a key regulator of G1/S transition in hypoxic cells. Intriguingly, hypoxic induction of p27 appears to be transcriptional and through an HIF-1-independent region of its proximal promoter. This demonstration of the molecular mechanism of hypoxia-induced G1/S regulation provides insight into a fundamental response of mammalian cells to low oxygen tension.

Cells may also respond to hypoxia by diminishing their proliferative rates. Both invasive and noninvasive studies of a variety of normal tissues and tumors suggest that hypoxic cells may be viable but nonproliferating (8 -12). This low proliferative state may be related to the phenomenon of tumor dormancy, described in nonvascularized metastatic foci (13), and may also help explain why hypoxic tumors are relatively chemoresistant and radioresistant (14). Although some transformed cell lines undergo apoptosis in extreme hypoxia and an acidic environment, nontransformed hypoxic cells remain viable but arrest in G 1 (15)(16)(17).
The best characterized molecular event necessary for the G 1 /S phase transition is phosphorylation of the retinoblastoma protein (RB) by specific cyclin-dependent kinase (CDK)-cyclin complexes (18). CDK activity can be inhibited by cyclin-dependent kinase inhibitors (CDKIs), such as p27 and p21, which then promote RB hypophosphorylation. There is also evidence that CDKIs may promote a G 1 arrest that is RB-independent (19,20). Hypoxia-induced arrest is associated with hypophosphorylation of RB (16,21) and, as opposed to hypoxia-induced apoptosis, appears to be independent of p53 induction (17,22). A direct role for HIF-1␣ in regulating the cell cycle in hypoxia has not been clearly demonstrated (6,7), and the events leading to the hypoxic hypophosphorylation of RB, and indeed the very relevance of RB phosphorylation status in hypoxia-induced G 1 arrest, have not been well delineated.
We therefore sought to characterize the molecular mechanisms responsible for hypoxia-induced growth arrest, and the role of HIF-1 in this response. Many previous studies exploring the effect of hypoxia on the cell cycle have been limited by the use of transformed and/or immortalized cell lines, which may have altered cell cycle regulators and/or other mutations (23). In this ambiguous genetic background, and without the ability to manipulate regulators of the G 1 /S transition, conclusions regarding the significance of cell cycle regulators in hypoxiainduced G 1 arrest have not been definitive. We elected to first identify cell cycle regulatory elements that are altered by hypoxia in immortalized fibroblasts, and then study the effect of hypoxia on the cell cycle of wild-type murine embryo fibroblasts (MEFs) and primary fibroblasts deficient in key regulators of the G 1 /S checkpoint. We demonstrate that RB and p27 play important roles in the hypoxia-induced G 1 arrest of primary fibroblasts. We then used immortalized fibroblasts for further molecular manipulation to show that hypoxia transcriptionally induces p27 and causes a G 1 arrest in an HIF-1-independent manner. These observations suggest a molecular mechanism for hypoxia-induced cell cycle regulation.
For all experiments, 1 ϫ 10 5 cells were plated in 10-cm dishes and incubated in 20% O 2 at 37°C for 24 h. The media were then changed and supplemented with 25 mM HEPES (pH 7.55). To render cells hypoxic, dishes were placed in a modular incubator chamber (Billups-Rothenberg), flushed with 95% N 2 , 5% CO 2 , and incubated at 37°C. This resulted in ϳ0.1-0.5% O 2 after several hours. After 32 h, cells were released from hypoxia and quickly scraped in ice-cold phosphatebuffered saline, and analyses were performed as described below. For CoCl 2 experiments, cells were incubated with CoCl 2 (200 M) for 32 h prior to analysis.
Description of Plasmids and Transient Transfection-Each 10-cm dish was plated with 1 ϫ 10 5 cells and 24 h later transfected with Lipofectin (Life Technologies, Inc.); after a 24-h incubation in 20% O 2 , cells were rendered hypoxic or incubated in 20% O 2 for an additional 24 h prior to analysis. The HIF-1␣ expression vector, pCEP 4 /HIF-1␣, was described previously (5). For the cotransfection experiments, Rat1a fibroblasts were transfected with 10 g of pCEP 4 /HIF-1␣ (or pCEP 4 ) and 1 g of green fluorescent protein (GFP) expression plasmid (pEGFP-N1, CLONTECH). HIF-1 function was assessed by a construct containing a hypoxia-responsive element from the human erythropoietin gene subcloned 5Ј to an SV40 promoter luciferase reporter (pGL2, Promega) (5). The murine p27 promoter and deletion constructs, subcloned into the pGL 2 basic luciferase vector (Promega), were the generous gift of Dr. Sehng-Cai Lin (27) and the 249-602 deletion construct was created by digesting the 1.1-kb promoter with EagI. The p53responsive p21 promoter/luciferase reporter construct was obtained from Dr. Bert Vogelstein (28). Luciferase assays (Promega) were performed, and the data were normalized for total protein.
Manipulation of Cell Cycle Regulators-Balb-3T3 cells were plated at 2 ϫ 10 5 /10-cm plate and 24 h later transfected with 30 nM of p27 antisense TGGCTCTCXTGCGCC or missense TGGCTCXCTTGCGCC oligonucleotides, where C ϭ 5Me C and X ϭ G-clamp (Gilead), in the presence of 2.5 g/ml GS3815 cytofectin (Glen Research). After 5 h buffered media were added, and cells were incubated in 20% O 2 or rendered hypoxic for an additional 32 h.
Recombinant adenoviruses containing full-length human p27 in the antisense direction and, as control, GFP alone were prepared using the AdEasy method (29) and titered so that Ͼ90% of infected cells expressed GFP and no toxicity was noted. AIN4 cells, plated at 1 ϫ 10 5 /10-cm plate, were infected with virus for 6 h in the presence of 2% fetal calf serum, and 24 h later media were exchanged and exposed rendered hypoxic or incubated in 20% O 2 for 48 h.
Cell Cycle Analysis-After 32 h, cells were released from hypoxia and quickly scraped in ice-cold phosphate-buffered saline, washed, and suspended in a buffer containing sucrose and trisodium citrate. Samples were then incubated for 10 min consecutively with trypsin/Nonidet P-40/spermine tetrahydrochloride, trypsin inhibitor/ribonuclease/and propidium iodide (PI) as described (30). PI and forward light scattering were detected by using a Coulter EPICS 752 flow cytometer equipped with MDADS 11 software, version 1.0. Cell cycle distribution profiles were determined with a curve fitting program ELITE version 3.0 (Coulter).
To assess directly the rate of DNA synthesis, cells were exposed to media containing BrdUrd (10 M) for 30 min. Cells were then trypsinized and removed from hypoxia. Nuclei were prepared and stained with a fluorescein isothiocyanate-labeled and BrdUrd antibody, and total DNA was stained with propidium iodide and analyzed as described (26). Wild-type MEF controls from RB, p53, and p27 wild-type animals all showed similar degrees of hypoxia-induced growth arrest and were averaged together.
Immunoblotting and Immunoprecipitation-Cells were released from 32 h of hypoxia, quickly washed with ice-cold phosphate-buffered saline, and then resuspended in a solution containing 1% SDS and boiled. Protein was quantitated by the BCA method (Pierce), and equal amounts of total cellular protein were resolved by SDS-polyacrylamide gel electrophoresis and subjected to immunoblot analysis (26). RB antibody was obtained from PharMingen, and cyclin A, D, and E were from Upstate Biotechnology, Inc.; for immunoblots, p27, p53, and p21 were from Santa Cruz Biotechnology, and HIF1␣ was from Novus Biologicals. Coomassie Blue staining of total cellular protein confirmed equal protein loading. For immunoprecipitation (IP) experiments, cyclin A (H432) and CDK2 (M2) antibodies were obtained from Santa Cruz Biotechnology, and cyclin E antibody (Ab1) was obtained from Neomarkers. 100 l of protein A-Sepharose beads (Amersham Pharmacia Biotech) were loaded with 20 -40 g of antibodies in IP buffer (50 mM Tris-HCl (pH 7.4), 250 mM NaCl, 0.5 mM Na 3 VO 5 , 20 mM ␤-glycerophosphate, 15 mM phosphatase substrate p-nitrophenyl phosphate, protease inhibitor Complete tablets (Roche Molecular Biochemicals), and 0.1% Nonidet P-40). Lysates from cells subjected to hypoxic or nonhypoxic culture conditions for 32 h were lysed with IP buffer, sonicated, and solubilized for 30 min at 4°C, and protein concentration was assayed with the BCA kit (Pierce). Washed protein A beads were incubated with 300 g of lysate for 2 h at 4°C, and the beads were then washed three times with IP buffer. Histone kinase assays were performed by incubating the beads in buffer containing 50 mM Tris (pH 7.4), 1 mM CaCl 2 , 5 mM MgCl 2 , 0.5 mM Na 3 VO 5 , 20 mM ␤-glycerophosphate, 15 mM phosphatase substrate p-nitrophenyl phosphate, protease inhibitor Complete tablets (Roche Molecular Biochemicals), 50 M ATP, 0.2 g of purified histone H1 (Roche Molecular Biochemicals), and 10 Ci of [␥-32 P]ATP at 30°C for 30 min. The reaction was stopped by the addition of 50 l of 2ϫ Laemmli buffer and heating for 95°C for 5 min. The products were separated by SDS-polyacrylamide gel electrophoresis, and autoradiography was performed. To assess the amount of CDK2 protein in the lysates, CDK2 was immunoprecipitated with a goat anti-CDK2 antibody, and the blot was probed with a rabbit anti-CDK2 antibody.
RNA Analysis-After 28 h of hypoxic or nonhypoxic conditions, medium was changed to contain 0.5 g/ml actinomycin D (Sigma), a concentration at which no toxicity was seen after 12 h of incubation. Total RNA was harvested at 0, 30, 120, and 240 min after actinomycin D addition under continued hypoxic or nonhypoxic conditions using Trizol (Life Technologies, Inc.) and the supplier's protocol. 15 g of RNA/lane was subjected to electrophoresis in 1% agarose-formaldehyde gels, transferred to nylon membranes, and hybridized with a p27 cDNA probe. Autoradiographic signals were quantitated with a PhosphorImager and normalized to 18 S RNA.

Hypoxia Induces a G 1 Cell Cycle Arrest in Immortalized
Fibroblasts and Is Associated with Hypoxia-mediated Changes in G 1 Regulators-Since contact inhibition and serum withdrawal are well documented to result in G 1 cell cycle arrest (31)(32)(33)(34), all experiments were performed at low (Ͻ50%) confluency in the presence of fresh, serum-containing, buffered media. We initially studied a well characterized, easily manipulated, immortalized fibroblast cell line to identify key cell cycle components that could be further investigated. When Rat1a fibroblasts were exposed to hypoxia for approximately two doubling times (32 h), propidium iodide staining and flow cytometric analysis of cell cycle distribution revealed a significant (p Ͻ 0.005) G 1 arrest (Table I). To assess directly the rate of DNA synthesis, fibroblasts were rendered hypoxic for 30 h, transferred to a large hypoxic atmospheric chamber (PlasLabs), and BrdUrd labeling was performed as described (26). Hypoxic cells incorporated significantly less BrdUrd (44 Ϯ 0.1% in normoxia versus 24 Ϯ 0.3% in hypoxia) confirming a decrease in S phase under hypoxic conditions.
We then sought to identify further components of the G 1 checkpoint that respond to hypoxia and might contribute to cell cycle arrest. RB phosphorylation is promoted by CDK2 and CDK4 when they complex with cyclins E, A, and D. A G 1 arrest occurs when these CDK activities are inhibited by the CDKIs p21, p27, or p16. After two doubling times in either normoxia or hypoxia, Rat1a fibroblasts were collected, and the expression of several proteins important for the G 1 -S phase transition was examined. As previously reported in other immortalized cell lines (16,21,(35)(36), hypoxia induced hypophosphorylation of RB in Rat1a cells, a hallmark of G 1 arrest in normal cells (Fig.  1A). This reduction in RB phosphorylation was associated with a decrease in CDK2 activity (Fig. 1A). Despite this decrease in kinase activity, the amount of immunoprecipitated CDK2 protein was unaffected by hypoxia ( Fig. 1A), indicating that a modifier of CDK activity (e.g. cyclin E, A, p21, and/or p27) is altered by hypoxia. Although p53 may be induced in hypoxia by an HIF-1␣-dependent manner (37,38), we observed only a minimal increase in p53 and actually observed a decrease in p21 under our experimental conditions (Fig. 1A). In contrast, protein lysates of hypoxic cells revealed a decrease in cyclin E and a minimal decrease in cyclin A (Fig. 1A). Also consistent with a decrease in CDK2 activity and RB phosphorylation, hypoxia led to a significant increase in the CDKI p27 (Fig. 1B).
Wild-type MEFs Also Undergo a Hypoxia-induced G 1 Arrest-The decrease in CDK2 activity and hypophosphorylation of RB associated with hypoxia that we and others observed in rat and other immortalized cell lines suggest, but do not conclusively support, a role for RB and p27 in hypoxia-induced G 1 arrest. Immortalized cells may have multiple genetic abnormalities. Alterations seen in one regulatory component (e.g. p27 induction) may be the result of hypoxia-induced growth arrest, not a cause of the G 1 arrest. In addition, both overexpression of p27 and serum withdrawal may induce G 1 arrest in an RB-independent manner (19,20,24). To evaluate the significance of these hypoxia-induced changes noted in immortalized cells, hypoxia-induced arrest was examined in mouse embryo fibroblasts (MEFs). This approach allows the identification of changes that occur in primary cells with minimal genetic alterations and permits the further examination of the effects of hypoxia on isogenic MEFs deficient in key regulators of the G 1 phase of the cell cycle.
Similar to the immortalized fibroblasts, wild-type MEFs underwent a G 1 arrest (Table I) and decreased BrdUrd uptake ( Fig. 2A) when rendered hypoxic. The extent of this hypoxiainduced arrest was most apparent when wild-type MEFs were synchronized with 32 h of serum withdrawal. When re-stimu-lated with serum, MEFs incubated in 20% O 2 showed a dramatic (Ͼ60%) percentage of S phase at 16 h, whereas serumstimulated hypoxic MEFs increased S phase only minimally (Ͻ12%) throughout the 32-h experiment (Fig. 2B). These observations indicate that hypoxia inhibits G 1 /S transition in normal primary fibroblasts.
Several of the changes that we (Fig. 1A) and others (16,21,35,36) noted in immortalized cells were not apparent in primary cell lines. Specifically, we noted only mild decreases in cyclin E and cyclin A in MEFs ( Fig. 1C and data not shown). Also, in contrast to Rat1a cells, we noted minimal kinase activities associated with CDK4 and cyclin E in both hypoxic and normoxic MEFs (data not shown). However, similar to the immortalized cells that underwent a G 1 arrest in hypoxia, we noted that hypoxia increased p27, decreased CDK2 activity, and led to a hypophosphorylation of RB in hypoxic MEFs ( Fig.  1C and Fig. 3, B and C). These data suggest that these changes may be key in regulating hypoxia-induced growth arrest.
Isogenic p27 and RB-deficient MEFs Demonstrate the Importance of These Proteins in Hypoxia-induced G 1 Arrest-To better assess the significance of RB hypophosphorylation and p27 induction in hypoxia, MEFs deficient in G 1 cell cycle regulators were subjected to 32 h of hypoxia. Previous studies have shown that hypoxia-induced growth arrest occurs in transformed cells with mutant or null p53 status (17,22). Consistent with these reports, MEFs null for p53 or p21 arrested in hypoxia to the same degree as wild-type cells (Table I and Fig. 2A). Therefore, neither p53 nor p21 is required for hypoxia-induced G 1 arrest. However, as compared with these cells, the ability of RB-null MEFs to arrest in G 1 in response to hypoxia was markedly diminished (Table I and Fig. 2A), indicating that RB participates in this arrest.
Although the decrease in cyclin E in wild-type MEFs was less dramatic than that in immortalized fibroblasts (Fig. 1, A and  C), either a decrease in cyclin E and/or an increase in p27 could contribute to diminished CDK2 activity leading to RB hypophosphorylation in hypoxia. Previous studies have shown that p27 induction is necessary for the growth arrest observed in Balb-3T3 fibroblasts subjected to serum withdrawal (33). To determine the extent that p27 induction is responsible for hypoxia-induced growth arrest, p27 null MEFs were rendered hypoxic for two doubling times. There was no significant change in cell cycle profile or incorporation of BrdUrd (Table I and Fig. 2A) despite hypoxia, indicating that p27 is necessary for hypoxia-induced G 1 arrest. The contribution of p27 to hypoxia-induced G 1 arrest was also apparent when CDK2 activity and RB phosphorylation status were examined in p27 null fibroblasts rendered hypoxic (Fig. 3, B and C). The base-line RB phosphorylation appears enhanced in the p27 null fibroblasts. Whereas hypoxia led to a 42% decrease in CDK2 activity as well as RB hypophosphorylation in wild-type MEF cells, these changes were not evident in p27 null fibroblasts.
It could be argued that the increase of p27 in hypoxic wildtype MEFs is a secondary effect of cell cycle arrest and not a mediator of hypoxia-induced G 1 arrest. However, p27 expression was also induced in hypoxic yet cycling RB null MEFs (Fig.  1C), which indicates that the induction of p27 is a direct effect of hypoxia rather than a secondary feature of RB hypophosphorylation and/or G 1 cell cycle arrest. We also observed that cyclin E decreased in the hypoxic RB null MEFs, although the absolute levels of cyclin E in both hypoxic and normoxic RB null MEFs were greater than those in wild-type MEFs (Fig. 1C). Therefore, we cannot exclude the possibility that the resistance of RB null cells to hypoxia-induced arrest may be due, in part, to increased cyclin E, which in turn could then titrate hypoxiainduced p27. Transient Decreases in p27 Also Abrogate the Hypoxia-induced G 1 Arrest-Because a total absence of p27 may result in abnormalities of cyclin expression and CDK-cyclin complex formation (39,40), we assessed whether a transient decrease in p27 would also diminish the hypoxia-induced G 1 checkpoint. p27 antisense and missense phosphorothioate oligonucleotides containing both the "G-clamp" and 5-methylcytosine analogues (41) were transfected with cytofectin under conditions that have been shown to result in Ͼ90% efficiency of oligonucleotide delivery to the nucleus (42). Whereas hypoxia led to a G 1 arrest and p27 induction in nontransfected Balb-3T3 cells (data not shown) and missense transfected cells, the antisense-treated cells had markedly less p27 induction with hypoxia (Fig. 3D) and failed to arrest in G 1 (Table II). Antisense treatment also diminished RB hypophosphorylation (data not shown), again indicating that the hypoxic induction of p27 is an upstream event of RB hypophosphorylation. To determine whether our finding of a p27-mediated hypoxia-induced G 1 arrest may be generalized to other cell types, immortalized human mammary epithelial cells (A1N4) were infected with a full-length antisense p27-expressing adenovirus. Whereas A1N4 cells infected with control adenovirus expressing GFP arrested in hypoxia, AIN4 cells treated with the antisense p27 virus showed a decrease in inducible p27 and cell cycle progression despite hypoxia (Table II and Fig. 3D). These observations further support the hypothesis that an increase in p27 is required for hypoxia-induced cell cycle arrest.
Hypoxia-induced G 1 Arrest Affects the CDK2-Cyclin A Complex-CDK2 is activated when complexed with either cyclin E or cyclin A; both of these complexes are inhibited by p27 (39). Hypoxia led to an increase in the amount of p27 associated with CDK2 in wild-type MEFs (Fig. 3A) as has been been reported in immortalized cell lines (21). We confirmed that radiolabeled in vitro translated cyclin E and cyclin A associate with CDK2 present in MEF lysates (data not shown). Despite the fact that our protocol immunoprecipitated in vitro translated cyclin E, the CDK2 activity associated with immunoprecipitated cyclin E was minimal in both wild-type and p27 null MEFs, and hypoxia decreased this minimal cyclin E/CDK2 activity in both wild-type and p27 null cells (data not shown). Additionally, immunodepletion of cyclin E had a negligible effect on total CDK2 activity in these MEFs (data not shown). In contrast, when CDK2-and cyclin A-associated kinase activities were measured from the same lysates, Ͼ90% of total CDK2 activity was recovered in cyclin A-associated complexes (Fig. 3B). Cyclin A/CDK2 activity decreased in hypoxic wild-type MEFs but was unchanged in p27 null cells. Although cyclin A is a target of E2F (43), is directly inhibited by hypophosphorylated RB (44), and has also been reported to be diminished by hypoxia in immortalized kidney cells (36), we did not observe cyclin A levels to change appreciably with hypoxia in either wild-type or p27 null MEFs (data not shown). Therefore, our data suggest that the predominant active CDK2 complex in MEFs is CDK2cyclin A and that increased p27 associated with this complex in hypoxia is responsible for inhibiting CDK2 activity and inhibiting exit from G 1 .
Hypoxia Induces p27 Transcription via an HIF-1-independ-ent Hypoxia-responsive Proximal Promoter Region-Because of the importance of p27 inhibition of CDK2-cyclin A complexes in hypoxia-induced G 1 arrest, we sought to characterize better the mechanism of p27 induction in hypoxia. Although the regulation of p27 during the cell cycle and tumor progression is thought to occur through translational and post-translational mechanisms (45,46), we observed a unique hypoxic regulation of p27 at the RNA level. When Rat1a fibroblasts were exposed to hypoxia for 32 h, p27 RNA levels increased 3-fold (Fig. 4A). When cells were treated with the RNA synthesis inhibitor actinomycin D in either hypoxia or normoxia, the p27 mRNA half-life did not increase in hypoxia (220 min in hypoxia versus 260 min in normoxia) (Fig. 4). These observations suggest that hypoxia induces p27 at the transcriptional level, which contrasts with the increased stability of some mRNAs, including vascular endothelial growth factor, in hypoxia (47). The best described hypoxia-induced transcription factor is HIF-1. HIF-1 has also been reported to have a role in cellular proliferation and is overexpressed in some tumors (6,7,48). We thus examined the role of HIF-1 in hypoxia-induced growth arrest and, more specifically, on p27 induction. Rat1a fibroblasts were cotransfected with an HIF1␣ expression vector and GFP expression plasmid in a 10:1 w/w ratio, and 48 h later cell cycle analysis was performed on GFP-positive cells. In addition, Rat1a fibroblasts were treated for 32 h with 200 M CoCl 2 , a known inducer of HIF-1␣ (5). For both conditions HIF-1␣ expression was assayed by protein immunoblot, and HIF-1 function was assessed by means of cotransfection experiments with an HIF-1Ϫresponsive luciferase reporter construct containing a hypoxia response element from the human erythropoietin gene (5). HIF-1␣ expression and HIF-1 activity were equivalent in both conditions to those found in hypoxic cells (data not shown). Although transfection of Rat1a cells with control plasmid caused a reproducible shift of cells distributed in the G 1 fraction, high expression of functional HIF-1 did not FIG. 3. A, p27 associates with CDK2 in hypoxia. CDK2 was immunoprecipitated from MEF after 32 h of hypoxia or normoxia, and the amount of p27 associated with this CDK2 was assessed by immunoblotting with an anti-p27. 10% of the total protein used for the CDK2 immunoprecipitation was run on a gel subsequently stained with Coomassie Blue to ensure equal loading (input). B, cyclin A and CDK2 histone kinase (HK) activity in wild-type MEFs and p27 null MEFs. MEFs were normoxic or rendered for 32 h, and kinase activity associated with immunoprecipitated cyclin A and CDK2 was assessed. The bottom panel displays kinase activity of wild-type and p27 null MEFs in both normoxia and after 32 h of hypoxia (dark bars). Hypoxic values reflect mean percent Ϯ S.E. of normoxic values, all averaged from three separate experiments. Kinase activities of normoxic p27 null lysates were normalized to normoxic wild-type levels. C, phosphorylation status of RB in hypoxic MEFs. Cell lysates from normoxic and hypoxic wild-type and p27 null MEFs were harvested. and RB was immunoblotted. PP represents the hyperphosphorylated forms of RB, and P represents the hypophosphorylated forms of RB. Coomassie Blue staining of total cellular protein confirms equal protein loading. D, p27 antisense oligonucleotides and p27 antisense adenovirus decrease the hypoxia-induced accumulation of p27 and hypoxia-induced G 1 arrest. Balb-3T3 cells were transfected with 30 nM p27 antisense or missense. A1N4 cells were infected with either full-length human p27 antisense virus or, as control, GFP-expressing adenovirus. After a brief recovery period, cells were exposed to 32 (Balb-3T3) or 48 h (A1N4) of hypoxia and compared with cells incubated in normoxia. Coomassie Blue staining of total cellular protein confirms equal protein loading.

TABLE II
The effect p27 manipulation on hypoxia-induced growth arrest Balb-3T3 cells were transfected with antisense (AS) or missense p27 oligonucleotides, rendered hypoxic for 32 h, and analyzed as described under "Materials and Methods." Human breast epithelial AIN4 cells were plated at low density and infected with GFP or GFP-human ASp27 as described. After a 24-h recovery, cells were rendered hypoxic for 48 h and then analyzed. Results represent the average Ϯ S.E. of three separate experiments performed in duplicate.  (6), another reported that HIF-1␣Ϫnull ES cells proliferate more slowly in hypoxia as compared with normoxia (7). When we subjected subconfluent ES cells (prepared and cultured as described (7)) to 32 h of hypoxia, similar degrees of G 1 arrest were observed in wildtype cells (G 1 29 Ϯ 0.1% in normoxia versus 34 Ϯ 0.1% in hypoxia) and in HIF-1␣-null ES cells (G 1 26 Ϯ 0.1% in normoxia versus 31 Ϯ 0.6% in hypoxia). Although aspects of cell cycle regulation may be significantly different in ES cells as compared with somatic cells (49,50), these observations suggest that, at least in ES cells, HIF-1␣ is not necessary for hypoxia-induced growth arrest.
Consistent with our finding that HIF-1 transactivation does not contribute to the hypoxia-induced G 1 arrest, induction of HIF-1 in normoxia did not increase p27 expression (Fig. 1B). We thus sought to determine whether the p27 promoter contains a hypoxia-responsive region. NIH-3T3 cells, which arrest in G 1 and induce p27 when hypoxic (data not shown), were transiently transfected with various fragments of the p27 promoter linked to a luciferase reporter and subsequently subjected to 24 h of hypoxia. Hypoxia increased the transcriptional activity of a 1.1-kb fragment of the p27 promoter almost 3-fold relative to normoxia (Fig. 4B). Consistent with our observation that p21 is not induced by hypoxia, transient transfections with a p53-responsive p21 promoter-luciferase reporter construct did not show an increase in luciferase activity in hypoxic cells (data not shown). This lack of activity also argues against nonspecific activation of G 1 CDKI promoters when cells are arrested in G 1 . Induction of the p27 promoter was dependent on DNA sequences located 1133 to 602 bp 5Ј to the transcription start site. Although it is necessary for the hypoxic response of the p27 promoter, this 531-bp region is not sufficient, since it alone subcloned 5Ј to an SV40 promoter-luciferase construct did not result in a hypoxia-induced increase in luciferase activity above that of the SV40 promoter alone (data not shown). Deletional analysis indicated that the 531-bp region, when joined with more proximal promoter sequences, is sufficient for hypoxic induction of luciferase activity (1.1 D249 -602 in Fig.  4B). Cotransfection of the 1.1-kb p27 promoter and an HIF-1␣ expression vector did not result in a significant increase in luciferase activity, even in hypoxia where HIF-1 transactivation of the erythropoietin promoter increased nearly 10-fold over base line (data not shown).

DISCUSSION
Cellular responses to hypoxia are fundamental adaptive changes that are required for normal mammalian physiology and may be exploited by cancer cells for a proliferative/survival advantage. Although the understanding of hypoxia-induced cellular metabolic changes is rapidly emerging, our understanding of the effects of oxygen deprivation on the control of the cell cycle is still rudimentary. In this study, we have observed that hypoxia induces an accumulation of cells in G 1 . This was best demonstrated when cells were synchronized prior to being rendered hypoxic. As we and others (24) have noted, even after prolonged serum withdrawal ϳ25% of MEFs continue to incorporate thymidine. The observation that cells in other stages of the cell cycle are also present after prolonged incubation in hypoxia suggests several possibilities. One is that some individual cells may escape G 1 arrest. Purging the external environment of oxygen may not necessarily lead to equivalent amounts of intracellular hypoxia, and cell cycle regulation may vary with the extent of hypoxia. Indeed, truly anoxic cells show an immediate and dramatic decrease in BrdUrd incorporation, possibly through a mechanism other than the one we observed at more moderate (ϳ0.1%) levels of hypoxia. 2 It is also possible that other stages of the cell cycle are also delayed in hypoxia. We observed a decrease in S phase in all cell lines that exhibited a hypoxia-induced G 1 arrest (Table I.); this hypoxia-induced decrease in S phase was greatly diminished or absent in RB null cells, in p27 null cells, or when p27 was transiently reduced with antisense techniques (Table II and Fig. 3D). A hypoxia-induced G 2 delay has been reported in some transformed cell lines (23), and indeed when hypoxic cells were manipulated to escape their G 1 arrest with antisense p27 oligonucleotides, a small accumulation in G 2 was noted (Table  II). Our data, however, are most consistent with the activation of a hypoxia-induced G 1 /S checkpoint.
Although several studies have demonstrated changes in cell cycle regulatory proteins associated with hypoxia-induced growth arrest, mechanistic insight was not forthcoming from these reports. Tumor cells lines with altered G 1 checkpoints have also been used in several studies. For example, previous studies have revealed that hypoxia-induced arrest neither occurs in osteosarcoma cell lines lacking functional RB nor in cells infected with HPV-E7 or E1a, in which RB is inactivated 2 S. L. Green and A. J. Giaccia, personal communications.

FIG. 4.
A, hypoxia induces p27 RNA expression increases with hypoxia but does not prolong p27 RNA half-life. After 28 h of hypoxic or nonhypoxic conditions, media were changed to contain 0.5 g/ml actinomycin D (Sigma). Total RNA was harvested at 0, 30, 120, and 240 min after actinomycin D addition under continued hypoxic or nonhypoxic conditions. The upper panel is a representative experiment, with ethidium bromide-stained 18 S RNA presented as a loading control. The lower panel shows the average of three experiments, normalized to the amount of p27 RNA present at time 0 in nonhypoxic cells, with error bars reflecting S.D. Filled bars represent hypoxic conditions. Autoradiographic signals were quantitated by PhosphorImager. B, increased p27 promoter activity in hypoxic cells. NIH-3T3 fibroblasts were transiently transfected with fragments of the murine p27 promoter linked to a luciferase reporter (1.5 g) and subsequently incubated in either 20% O 2 or hypoxia as described. Fold increase indicates the ratio of activity of the constructs in hypoxia to their activity in normoxia. Numbers represent the average Ϯ S.E. of three to ten experiments performed in duplicate. (17,21). A G 1 arrest, however, does occur in transformed and immortalized cells with mutant or null p53 status (17,22). The use of tumor cell lines provides some insight into cellular responses to hypoxia, but it is unclear if these cells have other abnormalities in their cell cycle regulatory mechanisms (51). Indeed we have noted subtle differences in the effects of hypoxia on immortalized fibroblasts and early MEFs. The immortalization process often leads to an inactivation of p53, RB, and/or p16/p19/ARF (52). Additionally, immortalized rat fibroblasts have been found to have a methylated p21 promoter that may not be responsive to p53 transactivation (53). We observed that immortalized fibroblasts have a greater reduction of cyclin E and cyclin A in response to hypoxia, when compared with wild-type MEFs (Fig. 1, A and C, and data not shown). Additionally, CDK2/cyclin E kinase activity was higher in immortalized rat fibroblasts than in MEFs, where it was negligible (data not shown). However, consistent changes in both immortalized and primary fibroblasts rendered hypoxic included an increase in p27, a decrease in CDK2 activity, and RB hypophosphorylation.
By using genetically defined primary fibroblasts, we demonstrate that hypoxia-induced G 1 arrest requires a functional RB but not p53 or p21. Indeed, neither the protein levels nor the promoter of the CDKI p21, which is transcriptionally regulated by p53, is induced under our hypoxic conditions. MEFs null for p27 also failed to arrest in hypoxia. The importance of p27 induction in hypoxia-induced arrest is underscored by the observation that the hypoxia-induced G 1 block can be overcome by decreasing p27 through antisense approaches. These observations suggest that the major regulator of the hypoxia-induced G 1 arrest is p27.
p27 was first noted to bind to cyclin E-CDK complexes and promote the dephosphorylation of RB (20,54). However, overexpression of p27 induces G 1 arrest in osteosarcoma cell lines with mutant RB (54), and serum starvation of RB null MEFs also induces a G 1 arrest (24). Together, these observations suggest that a p27 induced, RB-independent G 1 arrest pathway exists. Indeed, another CDKI, p16, has also been shown to induce both RB-dependent and RB-independent G 1 arrests in glioma cells (19). Our data also revealed a small G 1 arrest in hypoxic RB null MEFs, although it was greatly diminished when compared with wild-type MEFs (Table I). We also noticed diminished hypoxia-induced RB dephosphorylation in p27 null MEFs and when p27 was reduced with antisense oligonucleotides (Fig. 3D). The absence of RB led to increases in cyclin E, which may also play a role in titrating induced p27, decreasing CDK activity, and promoting entry into S phase even in hypoxia. However, the observation that cyclin E decreased minimally in wild-type MEFs, where a significant decrease in CDK2 activity was noted, and the fact that CDK2 activity changed minimally in hypoxic p27 null MEFs, argues for a predominant role of p27 induction in hypoxia-induced G 1 arrest. Thus, our data strongly suggest that p27 induces a G 1 arrest in hypoxia by modulating RB, although we cannot exclude the additional presence of an RB-independent mechanism or an additional contribution from the high levels of cyclin E noted in p27 null cells.
We noted that the predominant cyclin associated with CDK2 in both unsynchronized, cycling MEFs, as well as in hypoxic arrested wild-type MEFs, was cyclin A and not cyclin E. Cyclin E/CDK2 collaborates with cyclin D-dependent kinases to complete RB phosphorylation. Inhibition of cyclin E-associated CDK2 activity during G 1 inhibits entry into S phase, whereas inhibition near the G 1 /S transition does not delay S entry (55). Cyclin A is necessary for G 2 and S phase progression, is upregulated by E2F, and is repressed by hypophosphorylated RB (43,44). However, cyclin A-associated CDK2 activity is also vital for the G 1 /S transition. Microinjection of anti-cyclin A antibodies inhibits entry into S phase (56), and cyclin A can complex with CDK2 to phosphorylate RB (57). p27 can bind to and inhibit CDK2/cyclin A activity (57,58), and increases in cyclin A activity found in late G 1 do not correlate with increased cyclin A levels but rather decreased binding of p27 to the CDK2-cyclin A complex (57). These observations, combined with our data, suggest that that in mid to late G 1 CDK2-cyclin E and CDK2-cyclin A complexes are sequentially activated, and when cells are hypoxic, p27 inhibits cyclin A/CDK2 activity and RB phosphorylation and prevents entry into S phase (Fig. 5).
Although it might be argued that the hypoxia-induced increase in p27 protein levels is simply a secondary effect of G 1 arrest, our observations that changes in both proteins also occur in RB null cells, which continue to cycle in hypoxia, suggest that they are primary effects of hypoxia. The regulation of p27 during the cell cycle typically occurs at the protein level, regulated by translational control and ubiquitination (45,46). However, our experiments revealed a hypoxia-induced elevation of the p27 transcript that has an unaltered half-life in hypoxia. These results are consistent with a transcriptional activation of p27 by hypoxia. The existence of a physiological role for the transcriptional regulation of p27 is supported by the recent finding that dioxins can bind to the Ah receptor and increase transcription of p27 in hepatoma cells and thymocytes (59). Although the Ah receptor is closely related to HIF-1␤, we found that functional HIF-1 could not induce growth arrest, activate the p27 promoter, or elevate p27 expression. Although regulation of the cell cycle in somatic cells and ES cells may not be directly comparable (49,50), we found no abrogation of the hypoxia-induced growth arrest in HIF-1␣ null ES cells. In addition, p27 mRNA increases in hypoxic HIF-1␣-null ES cells (6). In aggregate, these studies suggest that p27 is induced by hypoxia at the transcriptional level independent of HIF-1, and the hypoxia-induced G 1 arrest is also independent of HIF-1. Although we have not formally excluded the other well characterized hypoxia-inducible transcription factor, endothelial periodic acid-Schiff protein-1 (EPAS-1/HLF/HIF-2␣) (60), this factor is also induced by CoCl 2 (61), a condition in which we noted no cell cycle arrest.
The p27 CDKI appears to play an important role in the response of cells to their environment. Induction of p27 expression is responsible for the growth arrest seen in serum-deprived fibroblasts (33). p27 protein is also induced by E-cadherin when transformed cells are grown in spheroid culture and is responsible for the growth arrest and chemotherapy resistance seen under these conditions (34,42). Our data suggest that hypoxia is another example of environmental control of p27. In the potentially damaging environment of hypoxia, elevation of p27 inhibits DNA replication and may prevent the inappropriate proliferation of genetically damaged cells.
Two of the most common genetic abnormalities found in human tumors are mutations or inactivation of the p53 and RB pathways (62). p53 expression in transformed hypoxic cells, possibly promoted by HIF-1 (37), leads to apoptosis and thus selects for cells with mutant p53 (15). Our data suggest that these cells will still be hypoproliferative in the background of a functional RB. Further studies will be required to determine whether other RB-related proteins such as p107 or p130 may also participate in the cellular response to hypoxia. Hypoxic proliferation of tumor cells could occur through inactivation of RB or, as shown by our studies, a decrease in inducible p27. Although p27 is not one of the many cell cycle regulators known to be commonly deleted in human cancers, its importance in neoplastic progression is underscored by the fact that even a heterozygous reduction of p27 leads to spontaneous and radiation-induced tumors in mice (63). Furthermore, p27 protein decreases during tumor development and progression in breast, colon, prostate, and other cancers (64). It is intriguing that some transformed cells have been reported to escape partially the hypoxia-induced cell arrest (23); the role of RB and/or p27 in this escape is not known. Further studies will be necessary to characterize better the trans-acting factor(s) that regulate p27 gene expression in hypoxia, to determine whether modulation of p27 expression accounts for the ability of some neoplastic cells to escape this G 1 arrest, and to determine whether modulation of p27 expression may alter chemosensitivity or radiosensitivity of hypoxic tumors.