Development of Cytosolic Hypoxia and Hypoxia-inducible Factor Stabilization Are Facilitated by Aquaporin-1 Expression*

O2 is essential for aerobic life, and the classic view is that it diffuses freely across the plasma membrane. However, measurements of O2 permeability of lipid bilayers have indicated that it is much lower than previously thought, and therefore, the existence of membrane O2 channels has been suggested. We hypothesized that, besides its role as a water channel, aquaporin-1 (AQP-1) could also work as an O2 transporter, because this transmembrane protein appears to be CO2-permeable and is highly expressed in cells with rapid O2 turnover (erythrocytes and microvessel endothelium). Here we show that in mammalian cells overexpressing AQP-1 and exposed to hypoxia, the loss of cytosolic O2, as well as stabilization of the O2-dependent hypoxia-inducible transcription factor and expression of its target genes, is accelerated. In normoxic endothelial cells, knocking down AQP-1 produces induction of hypoxia-inducible genes. Moreover, lung AQP-1 is markedly up-regulated in animals exposed to hypoxia. These data suggest that AQP-1 has O2 permeability and thus could facilitate O2 diffusion across the cell membrane.

Oxygen (O 2 ) is necessary for aerobic life because of its central role in mitochondrial ATP synthesis by oxidative phosphorylation. Traditionally, O 2 is considered to diffuse freely across the plasma membrane (1,2); however, recent studies have shown that O 2 permeability of lipid bilayers is some orders of magnitude lower than previously thought (3). Therefore, it has been suggested that there exist yet unknown plasmalemmal O 2 channels to ensure the fluxes required for O 2 uptake in conditions of high demand or limited O 2 availability. Good candidates are aquaporins, widely distributed intrinsic membrane proteins that form water-permeable complexes (3,4). Mammalian aquaporin-1 (AQP-1) 4 is highly expressed in cells with rapid gas (O 2 /CO 2 ) turnover such as erythrocytes (5) and microvessel endothelium (6,7), and experiments performed with recombinant AQP-1 expressed in Xenopus oocytes have suggested that it confers upon the cells increased membrane CO 2 permeability (8 -10). In addition, it has been shown that the AQP-1 tobacco plant homolog Nt-AQP-1 facilitates CO 2 transport, particularly in conditions of small transmembrane CO 2 gradient, and has a significant function in photosynthesis and in stomatal opening (11). Against a possible role of AQP-1 as a gas channel is that AQP-1 null mice do not show any obvious sign of respiratory distress or alteration of lung or erythrocyte CO 2 transport (12,13). This observation could, however, be explained if other aquaporins can compensate, at least partially, for the lack of AQP-1. In fact, AQP-1 functions as a well established water channel, but AQP-1 null humans (Colton-null blood group) (14) and AQP-1-deficient mice (12) have only subtle changes of erythrocyte water diffusion or renal urine concentration. A recent study shows that after subcutaneous or intracranial malignant cell implantation, AQP-1 null animals present impaired tumor growth and vascularity (15), alterations that are compatible with reduced O 2 uptake in extreme conditions by tumor cells. We performed experiments designed to test the hypothesis that AQP-1 regulates transmembrane O 2 transport. Here, we show that in mammalian cells stably transfected with AQP-1 and exposed to hypoxia, the loss of intracellular O 2 is accelerated, thus leading to faster stabilization of the hypoxia-inducible transcription factor (HIF) and upregulation of HIF-dependent genes. Redistribution of intracellular O 2 in hypoxia by mitochondrial inhibitors (16) is blunted in cells with high AQP-1 content. Moreover, inhibition of native AQP-1 expression with a small interfering RNA (siRNA) leads to up-regulation of hypoxia-inducible genes. Finally, we report that AQP-1 gene expression in lung in vivo is markedly induced by hypoxia. These data indicate that AQP-1 might have a role in O 2 homeostasis by facilitating O 2 diffusion across the cell membrane.
becco's modified Eagle's medium (Invitrogen) supplemented with 5% fetal bovine serum, 10% horse serum, and 1% penicillin/streptomycin (Invitrogen) in a CO 2 (10%) incubator at 37°C. To obtain stable subclones, PC12 cells were transfected with 20 g of pcDNA3-AQP1 or pcDNA3-AQP3. Screening of positive clones was done by Northern blot and in situ hybridization analysis. From 40 clones analyzed, 20 were positive for either AQP-1 or AQP-3 with variable levels of expression. siRNA experiments were done in an endothelial murine cell line derived from a mixed hemangioendothelioma (EOMA cells, ATCC catalog No. CRL-2586). Transfections were performed using Lipofectamine 2000 and an AQP-1-designed oligonucleotide (100 nM; ID 48361, Ambion, Austin, TX), with a scrambled oligonucleotide as a negative control. After 48 h of inhibition, levels of distinct mRNAs were analyzed by real-time reverse transcription and polymerase chain reaction.
Hypoxia experiments performed to analyze mRNA levels of tyrosine hydroxylase (TH), phosphoglycerate kinase 1 (PGK1), or vascular endothelial growth factor (VEGF) were done on a cell incubator with 1-10% O 2 (CO 2 10%) (17). Experiments designed to measure stabilization of HIF-2␣ protein were done in a sealed glove box anaerobic workstation (Coy Laboratory Products Inc., Grass Lake, MI) with O 2 maintained at 1-2%. Mitochondrial respiratory inhibitor myxothiazol (Sigma-Aldrich) and the NO donor DETA-NO (Alexis Biochemicals, San Diego, CA) were prepared fresh and added to the culture medium.
In Vivo Hypoxic Treatments-Rats were maintained either in normal conditions or in a hypoxia incubator (Coy Laboratory Products) for 24 -48 h at 10% O 2 . Animals were anesthetized by injection of 350 mg/kg chloral hydrate and then sacrificed following the animal care protocols approved by our institution. Tissues were removed either to extract RNA or for immunohistochemistry studies.
Pimonidazole Staining of Hypoxic PC12 Clones-PC12 cells were grown on coverslips 24 h prior to the hypoxia treatment. Cells were then transferred to hypoxia (3% O 2 ) for 0.5, 1, or 2 h, or let in normoxia, and 200 M pimonidazole was added to the medium for 30 min. After washing with PBS, cells were fixed with paraformaldehyde (3%) for 10 min at room temperature and subsequently washed again overnight with PBS. After blocking with bovine serum albumin, coverslips were incubated for 30 min with the fluorescein isothiocyanate-labeled Hypoxyprobe TM -1 monoclonal antibody 1 (mAb1, 1:100 dilution) from the Hypoxyprobe TM -1 Plus kit (Chemicon International, Temecula, CA). As a secondary antibody, the anti-fluorescein isothiocyanate monoclonal antibody conjugated with horseradish peroxidase provided with the kit was used. Coverslips were mounted and immunostained cells were examined on an Olympus Provis (Tokyo) microscope. Densitometric measurements of photographs taken under Normaski optics were performed using the NIH Image software.
Immunohistochemistry-After removal from the animal, tissues were kept overnight in 10% phormol and included on paraffin. Five-m slices were cut with a microtome and mounted on microscope slides (Superfrost/plus, Fisher Sci-entific). Rabbit polyclonal anti-AQP-1 (1:500 dilution, Abcam, Cambridge, UK) and biotin peroxidase-conjugated secondary antibody (1:200, Pierce) were used. Sections were developed in diaminobenzidine and photographed using a BX61-Olympus microscope. For cytochemistry, cells were fixed for 10 min with 3% paraformaldehyde at room temperature. A rabbit polyclonal anti-AQP-1 antibody (1:100 dilution, Chemicon) and a fluorescent polyclonal goat anti-rabbit antibody (Alexa Fluor 568, Molecular Probes, Invitrogen) were used. Immunofluorescence was then analyzed with a Leica DM IRBE confocal microscope.
RNA Analysis-Extraction of RNA from PC12 cells was done following the method described by Cathala et al. (18). To extract RNA from tissues, the TRIzol reagent (Invitrogen) was used as indicated by the manufacturer. For Northern analysis, 10 g of total RNA were resolved in agarose/formaldehyde gel and transferred to Hybond-Nϩ nylon membrane. The Ultrahyb solution (Ambion) was used for hybridization with the specific 32 P probes (19,20). Results were visualized using a PhosphorImager (Typhoon 9400, Amersham Biosciences). The reverse transcription reaction was performed immediately after the mRNA isolation using SuperScript II RNase H Ϫ reverse transcriptase (Invitrogen).
Real-time PCR analysis was performed in an ABI Prism 7500 Sequence Detection System (Applied Biosystems, Warrington, UK) using SYBR Green PCR Master mix (Applied Biosystems) and the thermocycler conditions recommended by the manufacturer. Amplification of 18 S ribosomal or cyclophilin RNA was done to normalize for RNA input amounts. Primers were designed using the Primer Express software (Applied Biosystems). Western Blotting-Cells were washed with cold PBS, scrapcollected in 1 ml of cold PBS, and centrifuged at 165 ϫ g for 5 min at 4°C. For whole-cell protein extract, the pellet was lysed in 150 -300 l of homogenization buffer: 50 mM Hepes (pH 7.3), 5 mM EDTA, 250 mM NaCl, 5 mM dithiothreitol, 0.2% (v/v) Nonidet P-40 (Sigma-Aldrich), and 1% (v/v) of complete protease inhibitors mixture (Sigma-Aldrich). The resuspended pellet was left on ice for 5 min, vortexed, and then centrifuged at 16,000 ϫ g for 15 min at 4°C. For cell-membrane protein extract, cells were scraped and collected in 1 ml of homogenization buffer containing 320 mM sucrose, 5 mM Hepes, and 1% (v/v) of complete protease inhibitors mixture. Samples were homogenized with a Polytron and centrifuged at 2000 ϫ g for 5 min at 4°C, and then supernatant was centrifuged at 30,000 ϫ g for 30 min at 4°C. The pellet was resuspended in homogenization buffer plus 1% SDS. Protein concentration was analyzed with the Bradford method (Bio-Rad Protein Assay) for wholecell extract and with the Lowry method (Bio-Rad DC Protein Assay) for cell membrane protein extraction, and kept at Ϫ20°C until Western blot assay (21). Twenty g of wholecell extracts were resolved by SDS-PAGE (6%) for HIF2␣. For AQP-1, 40 g of cell membrane proteins were resolved in 10% SDS-PAGE. After electrophoresis, proteins were transferred into polyvinylidene difluoride membranes (Hybond-P, Amersham Biosciences) using a Novex apparatus (Novel Experimental Technology, San Diego, CA). Membranes were probed with 1:1000 anti-HIF2␣ (Abcam), 1:2000 anti HIF1␤ (Abcam), 1:1000 anti-AQP-1 (Chemicon), and 1:10,000 anti-␤-tubulin (Sigma). Immunoreactive bands were developed with the ECL system (Amersham Biosciences) and visualized using a PhosphorImager (Typhoon 9400, Amersham Biosciences).
Water Permeability Measurements-Cells grown on coverslips were loaded for 5 min with 1 M calcein acetoxymethyl ester (Molecular Probes) and mounted in a small (ϳ250 l) perfusion chamber that allowed rapid exchange of a solution with 300 (isosmotic) to another with 150 (hyposmotic) mmol/kg. The rate of change in calcein fluorescence was monitored as described (22). The isosmotic solution used for these experiments contained in mM: 140 NaCl, 4.5 KCl, 2.5 CaCl 2 , 1 MgCl 2 , 10 Hepes, and 10 glucose (pH: 7.4). The hyposmotic solution was obtained by water dilution.
Statistical Analysis-Data were presented as mean Ϯ S.E. and were analyzed with either paired Student's t test or the one-way analysis of variance followed by Tukey's test.

Responsiveness to Hypoxia Is
Accelerated in Cells with High Levels of AQP-1-To investigate the role of AQP-1 in plasmalemmal O 2 transport, we generated several subclones of PC12 cells stably transfected with either AQP-1 or AQP-3, representative members of the two major classes of aquaporins (23,24). Because direct measurement of O 2 fluxes at the cellular level is not technically feasible, in our initial experiments the expression of HIF-dependent genes was monitored as a readout of cytosolic O 2 concentration (Fig. 1A). HIF, a master regulator of the responses of tissues to low O 2 tension (25)(26)(27), is composed of ␣and ␤-subunits; the stability of the ␣-subunit is regulated by a family of O 2 -dependent prolyl hydroxylases (26,28). At low O 2 tensions,  OCTOBER 12, 2007 • VOLUME 282 • NUMBER 41 cellular hydroxylase activity is inhibited, and HIF␣ accumulates to heterodimerize with HIF␤. The heterodimeric HIF activates the expression of numerous genes. In AQP-1 or AQP-3-transfected clones we tested the hypoxic induction of TH, an O 2 -dependent gene up-regulated by even mild hypoxia in PC12 cells (29). The rationale behind these experiments was that if AQP-1 is O 2 -permeant, development of cytosolic hypoxia and subsequent cumulative induction of TH mRNA should be faster in cells expressing high levels of AQP-1 than in wild type PC12 cells. Hypoxic TH mRNA induction in the seven clones studied besides wild type PC12 cells is shown in Fig. 1B (top). Although the basal TH mRNA levels varied among the different clones and in separate experiments on a given cell clone, we compared for each cell type the relation of TH mRNA in hypoxia versus the values in normoxia. A quantitative summary of TH mRNA induction after 12 h of hypoxic (6% O 2 ) exposure in several clones of AQP-1-and AQP-3-transfected cells is shown in Fig.  1B (bottom) and C. Although for AQP-1-expressing clones there was a correlation between hypoxic TH mRNA induction and the level of AQP-1 expression, the response of AQP-3 clones (either with high or low expression levels) was indistinguishable from that of wild type PC12 cells. Differences between AQP-1-expressing clones were also seen with shorter (6 h) hypoxic (6%) exposure (supplemental Fig. 1A). In the case of AQP-1 clones there was a lineal correlation between TH mRNA expression and the level of AQP-1 in the various cell types as determined by Western blot (Fig. 1D). As result of this initial screening, two AQP-1-transfected clones, one with high level of AQP-1 and highly responsive to hypoxia (clone HAQP) and another with low level of AQP-1 and low responsiveness to hypoxia (clone LAQP) similar to wild type cells (Fig. 1D), were selected for further analysis. The high expression of AQP-1 in HAQP in comparison with LAQP or wild type cells was further confirmed by immunocytochemistry ( Fig. 2A). In addition, we showed that HAQP cells responded more rapidly to an hyposmotic challenge than LAQP cells (Fig. 2B), thus indicating that recombinant AQP-1 expressed in the PC12 cell membrane was functional as a water-permeable channel. We also used in some experiments a highly expressing AQP-3 clone (clone 3.3 in Fig.  1, B and C), which behaves similar to the wild type PC12 cells (supplemental Fig. 1, B-D), to make comparisons with the highly expressing AQP-1 cells.

Aquaporin-1 and Membrane O 2 Transport
The differential behavior of HAQP and LAQP cells that was clear when they were assayed in mild hypoxia (6% O 2 ) (Fig. 3A, left) became even more apparent in conditions of extreme hypoxia. After 12 h in 1% O 2 , TH mRNA accumulation in HAQP was ϳ3-4-fold higher than in LAQP cells (Fig. 3A,  right), indicating that the loss of cytosolic O 2 was accelerated by AQP-1. Contrarily, TH mRNA induction by cobalt, an agent that stabilizes HIF independent of O 2 tension (25,27), was the same in the two cell clones (Fig. 3B). Similar to TH, the hypoxic mRNA up-regulation of phosphoglycerate kinase 1 (PGK1) and vascular endothelial growth factor (VEGF), two other O 2 -sensitive genes (25,27), was higher in HAQP than in LAQP or wild type cells (Fig. 3, C and D). Cobalt TH induction and hypoxic induction of PGK1 in the AQP-3 clone was similar to the values obtained in PC12 cells (supplemental Fig. 1, C and  D). Altogether these data supported the hypothesis that AQP-1 expression confers increased O 2 permeability upon PC12 cell membrane.
HIF-2␣ Stabilization in Hypoxia Depends on the Level of AQP-1 Expression-Because the expression of the O 2 -sensitive genes depends on the stabilization of HIF, we investigated whether the regulation of the two components of this heterodimeric transcription factor (HIF␣ and HIF␤) was influenced by the presence of functional AQP-1. In fair agreement with the results on the O 2 -dependent genes studied (TH, PGK1, and VEGF), the accumulation of HIF-2␣ (the most abundant O 2 -dependent isoform in PC12 cells (21,30)) during the first few hours after exposure to hypoxia was significantly greater in HAQP than in LAQP or wild type cells. As an internal control, the levels of HIF-1␤ (a constitutive protein that dimerizes with all HIF isoforms (25)) were unaltered by hypoxia (Fig. 4, A and B). To check whether the presence of AQP-1 has a bidirectional impact on transmembrane O 2 transport (i.e. it also facilitates O 2 influx), we induced HIF-2␣ accumulation in hypoxia, and after rapidly switching to normoxia, HIF-2␣ degradation time course was monitored. As proteasomal degradation of accumulated HIF-2␣ could be nonspecifically influenced by the amount of proteins existing in the cells, we compared in these experiments clones with high levels of aquaporin expression (clones 1.22 and 3.3, Fig. 1 and supplemental Fig. 1B). Degradation of HIF-2␣ in normoxia was clearly faster in cells expressing AQP-1 (HAQP1) than in cells with AQP-3 (HAQP3), thus supporting the view that AQP-1 increases transmembrane O 2 transport (Fig. 4, C and D).
To further analyze the alterations in membrane O 2 transport caused by AQP-1 expression, we studied the differential response of HAQP and LAQP cells to redistribution of intra- cellular O 2 by mitochondrial inhibition. As reported before (16,31), inhibition of mitochondrial electron transport in wild type cells with myxothiazol (a complex III blocker) or DETA-NO (an NO donor that inhibits cytochrome c oxidase) lead to destabilization of HIF-2␣ in hypoxic (1% O 2 ) cells (Fig. 5). It has been proposed that HIF destabilization upon inhibition of mitochondrial respiration in hypoxic cells (16,31) is caused, among other mechanisms (32,33), by intracellular O 2 redistribution toward nonrespiratory O 2 -dependent targets such as prolyl hydroxylases, so that they do not register hypoxia (16). Quite interestingly, HIF-2␣ destabilization produced by myxothiazol and by DETA-NO was less pronounced in HAQP in comparison with LAQP or wild type cells (Fig. 5). These observations further support the view that AQP-1 increases O 2 permeability in HAQP cells, thereby facilitating the transmembrane efflux of intracellular O 2 spared by inhibition of respiration. Thus, upon mitochondrial inhibition hypoxic HAQP cells can maintain a more severe cytosolic hypoxia and higher HIF-2␣ content than LAQP or wild type cells.
Cytosolic Loss of O 2 in Hypoxia Is Facilitated by AQP-1 Expression-To obtain more direct evidence that AQP-1 facilitates the development of cytosolic hypoxia, and to rule out that overexpression of AQP-1 activates HIF through an oxygen-independent signal transduction pathway, we studied the transmembrane fluxes of O 2 in HAQP and LAQP cells using pimonidazole staining, a broadly used hypoxia marker that forms intracellular protein adducts at low O 2 tension (31,34,35). With this technique, the loss of cytosolic O 2 upon exposure to hypoxia was clearly faster in HAQP than in LAQP cells (Fig. 6). These differences estimated with the Hypoxyprobe method were similar to those inferred from the HIF-2␣ experiments (Fig. 4B), thus suggesting that the distinct time courses of HIF-2␣ stabilization observed in HAQP and LAQP cells reflect, indeed, differences in membrane O 2 transport between the two cell types.
Knocking Down Native AQP-1 with siRNA Results in Up-regulation of Hypoxia-inducible Genes-Besides the studies on PC12 cells overexpressing AQP-1, we also performed loss of function experiments using a nonmetastatic murine hemangioendothelioma cell line (EOMA cells), which behave in vitro in a manner similar to microvascular endothelial cells (36,37). First, we investigated whether, as in microvessel endothelium (6,7), EOMA cells also express AQP-1 as well as hypoxia-dependent  12 h (C, control). Average values from three separate experiments. C, Northern blot analysis of phosphoglycerate kinase 1 (PGK1) mRNA expression in AQP-1-expressing cell clones HAQP, LAQP, and wild type PC12 cells exposed to normoxia (N) or hypoxia (H, 6% O 2 for 12 h). Cyclophilin (Cp) mRNA is used as a load control. The bar diagram represents values (mean Ϯ S.E.) from 5 separate experiments. D, TH and VEGF mRNA analysis by real time PCR in the cell clones indicated exposed to normoxia (N) or hypoxia (H, 6% O 2 for 12 h). Note that the level of hypoxic TH mRNA induction is similar to that estimated by Northern blot (panel A, left). Mean Ϯ S.E. of 5 experiments. *, p Ͻ 0.05; ***, p Ͻ 0.001. OCTOBER 12, 2007 • VOLUME 282 • NUMBER 41 genes such as PGK1 or VEGF (Fig. 7A). In these cells the level of PGK1 mRNA expression was quite insensitive to moderate hypoxia; however, VEGF mRNA induction was clearly modulated by moderate levels of hypoxia in an O 2 tension-dependent manner (Fig. 7A). In normoxic EOMA cells treated with specific AQP-1 siRNA, the level of AQP-1 expression decreased to ϳ50 -60% of the value in control. In parallel with the decrease of AQP-1 expression we observed an almost 2-fold increase of VEGF mRNA (Fig. 7B). Knocking down AQP-1 with siRNA had little effect on the levels of either ␤-actin (a gene independent of O 2 tension) or PGK1, a gene that, as shown above, is less sensitive to hypoxia than VEGF in EOMA cells (Fig. 7). These data further suggest that the decrease of AQP-1 expression in endothelial cells diminishes the transmembrane O 2 fluxes, which results in reduced cytosolic O 2 tension and induction of HIFdependent genes.

Aquaporin-1 and Membrane O 2 Transport
AQP-1 Gene Expression Is Up-regulated by Hypoxia in Vivo-Because AQP-1 is expressed in microvessels (6, 7), we searched its up-regulation in lungs of animals subjected to hypoxia. A priori, we presumed that this could be an advantageous adaptive response if AQP-1 plays any role in membrane O 2 transport. Indeed, AQP-1 was expressed in lung endothelial cells (inset in Fig.  8), and the AQP-1 gene was also highly O 2 -sensitive in comparison with the more classically studied PGK1 gene. In animals exposed to mild hypoxia (10% O 2 ) for 24 h, the level of PGK1 mRNA was barely altered, but we observed a robust ϳ3-fold AQP-1 mRNA induction in lung tissue (Fig. 8).

DISCUSSION
The results in this study provide evidence that AQP-1 can accelerate the establishment of cytosolic hypoxia possibly through facilitation of O 2 transport across the plasma membrane. We have also shown that AQP-1 gene expression is regulated exquisitely by O 2 tension. Although we did not directly measure transmembrane O 2 fluxes in PC12 cells because it is not technically feasible, our conclusions are supported by several independent observations. (i) In cells overexpressing AQP-1 (HAQP cells) and exposed to a hypoxic environment, the levels of HIF␣and O 2 -dependent genes (TH, PGK1, and VEGF) were higher than in LAQP cells, suggesting that the loss of cytosolic O 2 was accelerated by the presence of AQP-1. It must be noted that this experimental protocol estimates the cumulative stabilization of HIF (and mRNA induction of HIF-dependent genes) during the development of hypoxia rather than the level of cytosolic O 2 in steady state conditions. (ii) HIF-2␣ degradation elicited by reoxygenation of hypoxic cells was faster in cells expressing AQP-1 than in cells highly expressing AQP-3. (iii) The differential behavior exhibited by HAQP and LAQP cells during exposure to low O 2 tension was confirmed when the development of cytosolic hypoxia was estimated by the cumulative staining of pimonidazole, an HIF-independent hypoxia marker (34,35). (iv) In hypoxic PC12 cells, inhibition of mitochondrial respiration produced HIF-2␣ destabilization (16), which was less pronounced in HAQP than in LAQP or wild type cells. This finding is compatible with an increased O 2 permeability in HAQP cells, thereby facilitating the transmembrane efflux of intracellular O 2 spared by inhibition of respiration. (v) In normoxic endothelial cells, lowering AQP-1 expression with a siRNA blunted aquaporin mRNA expression by ϳ50 -60% and selectively upregulated VEGF, a gene induced in these cells by even moderate levels of hypoxia. Altogether, these data strongly support the view that AQP-1 is involved in O 2 homeostasis and facilitates transmembrane O 2 transport. Our proposal that AQP-1 could function as an O 2 channel is in fair agreement with previous reports suggesting that AQP-1 increases membrane CO 2 permeability (8 -11). Recently, it has been proposed that AQP-1 also transports NO across endothelial cells and contributes to the regulation of vascular tone (38,39). Although our experiments suggest that AQP-3 is not O 2 -permeable, there are several other aquaporins that overlap AQP-1 distribution and could partially compensate for the AQP-1 deficiency (24). For example, in alveolar epithelial type I    OCTOBER 12, 2007 • VOLUME 282 • NUMBER 41 cells the AQP-5 gene is up-regulated by HIF through interaction with hypoxia-responsive elements in the promoter (40), and AQP-4 mRNA has also been shown to be up-regulated by hypoxia in astrocytes (41). This could explain why humans (14) and mice (12,13) lacking AQP-1 do not exhibit major gross physiological alterations related with either water or O 2 metabolism. In addition, besides aquaporins, other membrane proteins could also contribute to membrane O 2 transport (3,10).

Aquaporin-1 and Membrane O 2 Transport
The novel role of AQP-1 in O 2 homeostasis proposed here is compatible with its expression in erythrocytes (5) or vascular endothelium (6,7). AQP-1 is also overexpressed in cells of proliferating microvessels (8), as well as in aberrant cells of human and rat tumors (42)(43)(44), situations in which a limited O 2 availability could have induced AQP-1 expression to facilitate O 2 uptake by the cells. In fact, AQP-1 null mice present impaired tumor growth after subcutaneous or intracranial malignant cell implantation, with reduced tumor vascularity and extensive necrosis (15). Migration of AQP-1-deficient aortic endothelial cells in vitro is greatly decreased, and it has been suggested that the lack of AQP-1 alters water fluxes required for rapid turnover of cell membrane protrusions at the leading edge of migrating cells (15). The presence of AQP-1 could also be critical for aerobic ATP synthesis, particularly in membranes of low gas permeability (4) or when transmembrane O 2 gradients are small, as it occurs in cells exposed to environments with extremely low O 2 concentrations or in multilayered diffusion barriers such as that existing between the lung alveoli and blood. In conclusion, our observations support a new physiologic role for aquaporins as O 2 transporters (4) besides their canonical function as water-permeable channels. AQP-1 pharmacology could be of potential use in conditions of altered O 2 exchange or to control O 2 -dependent angiogenesis and tumor cell growth.