Synergistic Cooperation between Hypoxia and Transforming Growth Factor-β Pathways on Human Vascular Endothelial Growth Factor Gene Expression*

Signaling by transforming growth factor (TGF)-β family members is mediated by Smad proteins that regulate gene transcription through functional cooperativity and association with other DNA-binding proteins. The hypoxia-inducible factor (HIF)-1 is a transcriptional complex that plays a key role in oxygen-regulated gene expression. We demonstrate that hypoxia and TGF-β cooperate in the induction of the promoter activity of vascular endothelial growth factor (VEGF), which is a major stimulus in the promotion of angiogenesis. This cooperation has been mapped on the human VEGF promoter within a region at −1006 to −954 that contains functional DNA-binding sequences for HIF-1 and Smads. Optimal HIF-1α-dependent induction of the VEGF promoter was obtained in the presence of Smad3, suggesting an interaction between these proteins. Consistent with this, co-immunoprecipitation experiments revealed that HIF-1α physically associates with Smad3. These results demonstrate that both TGF-β and hypoxia signaling pathways can synergize in the regulation of VEGF gene expression at the transcriptional level.

Smad proteins are critical for transmitting transforming growth factor-␤ (TGF-␤) 1 superfamily signals from the cell surface to the nucleus (1)(2)(3)(4). Based on structural and functional characteristics, Smads fall into the following three different families: (i) Smad substrates of the signaling TGF-␤ receptor family of Ser/Thr kinases (Smad1-3, -5, or -8), also known as receptor-regulated Smads; (ii) common Smads that associate with the phosphorylated receptor-regulated Smads and then translocate into the nucleus as a heteromeric complex (Smad4); and (iii) inhibitory Smads that antagonize Smad sig-naling (Smad6 and -7). In the nucleus, Smads regulate transcriptional responses through functional cooperativity and physical interactions with different transcription factors, whose activity can be modulated by other signaling pathways. Thus, Smad2 associates with the winged helix proteins Fast-1 and Fast-2 to stimulate the Xenopus Mix-2 and mouse goosecoid promoter activities (5)(6)(7); Smads interact with the zinc finger OAZ in response to BMP2 to activate transcription of the Xvent-2 gene (8); Smad2 and Smad3 bind transcription factors AP-1 (9), Sp1 (10 -13), or TFE3 (14,15), which are able to bind independently and transactivate target gene promoters; Smad1 and STAT3, bridged by p300, form a complex that leads to the cooperative signal of leukemia inhibitory factor and BMP2 (16); and functional synergy of Smads with CREB and AML proteins results in enhanced TGF-␤-induced transcription (17)(18)(19). In addition, Smads are able to interact with transcriptional suppressors. Thus, TGF-␤ signaling can be inhibited through binding of Smads with the adenoviral oncoprotein E1A (20), with the TGIF factor (21), or with the SnoN and Ski protooncoproteins (22)(23)(24)(25). In summary, Smads regulate TGF-␤-dependent gene expression by recruiting co-activators and co-repressors to a wide array of DNA-binding partners, thus functioning as transcriptional co-modulators.
Hypoxia inducible factor (HIF)-1 is a transcriptional complex with a crucial role in oxygen-regulated gene expression (26 -28). This complex is a heterodimer formed by proteins HIF-1␣ and the aryl hydrocarbon receptor nuclear translocator (ARNT or HIF-1␤) (29). Both subunits are members of the PAS superfamily of basic helix-loop-helix proteins characterized by a high sequence homology with the Drosophila Periodic, the Aryl hydrocarbon receptor, the Aryl hydrocarbon receptor nuclear translocator, and Drosophila Single-minded (30 -32). HIF-1 binds to hypoxia-responsive elements and activates transcription of a wide variety of genes (27,28,33,34). Among hypoxiaregulated genes, vascular endothelial growth factor (VEGF) is a prototypic example because it plays a critical role in angiogenesis, a process that regulates oxygen access to tissues (35)(36)(37). A major hypoxia-responsive enhancer was identified as a 28-base pair sequence located at 900 base pairs upstream from the CAP site of the VEGF promoter region (38,39). Deletion of this element significantly inhibited hypoxic induction of transcription.
Independent lines of investigation have shown that TGF-␤ (40) or hypoxia (41) is able to regulate angiogenesis, a process where VEGF plays a critical role. By using the VEGF promoter as a model system, we have investigated the putative cooperation between hypoxia and TGF-␤. These two different signaling pathways were able to synergize in stimulating VEGF transcription, and accordingly, a functional cooperation and physical interaction between Smad and HIF-1␣ transcription factors could be demonstrated.

EXPERIMENTAL PROCEDURES
Cell Culture-The human endothelial HMEC-1, the rat myoblast L6E9, the human hepatoma HepG2 and Hep3B, the human epithelioid carcinoma HeLa, the monkey kidney COS, and the human colon adenocarcinoma SW480.7 cell lines were cultured in MCD131, Dulbecco's modified Eagle's medium, or ␣-minimum Eagle's medium supplemented with 10% heat-inactivated fetal calf serum in a 5% CO 2 atmosphere at 37°C, as described previously (9,(42)(43)(44). Hypoxic exposure was carried out under 1% oxygen, 5% CO 2 , and 94% nitrogen (AL Air Liquide Españ a) for 24 (promoter activity assays) or 4 h (DNA-protein and protein-protein interaction assays). In some cases, chemical hypoxia was carried out by an overnight treatment in the presence of 100 M deferrioxamine (Sigma). Unless otherwise indicated, treatment of cells with recombinant human TGF-␤1 (R & D Systems, Abingdon, UK) was performed at a concentration of 200 pM in culture medium supplemented with 0.2% fetal calf serum.
ELISA Measurements of Secreted VEGF-Hep3B cells before confluency were preincubated for 48 h with serum-free medium. Then the culture medium was replaced by fresh serum-free medium containing or not containing 200 pM TGF-␤1. At time 0, cells were incubated under normoxic (21% oxygen) or hypoxic (1% oxygen) conditions for 3-72 h. The medium was removed at the indicated time points and stored at Ϫ80°C until assayed. VEGF concentrations were determined using a commercial ELISA kit (R & D Systems), following the manufacturer's instructions. Samples from three different experiments were analyzed in triplicate, and the mean and S.D. were calculated.
Northern Blot Analysis-Total RNA from human hepatoma Hep3B and endothelial HMEC-1 cells was isolated using an RNAeasy kit (Qiagen). RNA samples (10 g) were denatured, fractionated in 1.1% agarose/formaldehyde gels, and blotted onto nitrocellulose. Membranes were hybridized in 50% formamide at 42°C with excess 32 P-labeled probe and washed under high stringency conditions (0.2ϫ SSC and 0.5% SDS at 52°C). The probe used was a 0.5-kilobase pair HindIII insert of the human VEGF 165 cDNA, labeled with [␣-32 P]dCTP using the Rediprime II kit (Amersham Pharmacia Biotech). Radiolabeled bands were detected by autoradiography.
Plasmids-The reporter vector VEGF(Ϫ1910/ϩ379)-LUC was kindly provided by Dr. G. Semenza (35). It is derived from the pGL2 plasmid and contains the human VEGF promoter region fused to the firefly luciferase gene. The TATA-pXP2 vector was generated by inserting the rat minimal prolactin promoter (Ϫ36 to ϩ37) into the promoterless pXP2 vector and will be described elsewhere. The 2HRE/WT-LUC construct was obtained by inserting a dimer of the oligonucleotide Ϫ1006/ Ϫ954 upstream of the minimal TATA box of TATA-pXP2 plasmid. The reporter vectors 1HRE/WT-LUC, 1HRE/HM-LUC, and 1HRE/SM-LUC were obtained by inserting, respectively, the wild type oligonucleotide Ϫ1006/Ϫ954, the oligonucleotide Ϫ1006/Ϫ954-HM mutated at the consensus HIF-1 site Ϫ973/Ϫ971 (AAA by CGT), and the oligonucleotide Ϫ1006/Ϫ954-SM mutated at the consensus Smad site Ϫ992/Ϫ987 (AAAAA by CAGAC), into the pGL2-P reporter vector (Promega). The reporter plasmid 3TP-Lux, containing TGF-␤-responsive elements of the PAI-1 promoter was kindly provided by Dr. Joan Massague (Memorial Sloan-Kettering Cancer Center, New York).
Transfections-Transfection of L6E9, HepG2, HeLa, and COS cells was carried out using Superfect (Qiagen) according to the manufacturer's protocol. SW480.7 cells were transfected by using the calcium phosphate DNA precipitation method. Cells in 24-well plates were transfected with the appropriate reporter and/or expression vectors at densities of 5 ϫ 10 4 cells/well. When the reporter vector was co-transfected with expression vectors, the amount of DNA in each transfection was normalized by using the corresponding insertless expression vector as carrier. Relative luciferase units from duplicate samples were determined in a TD20/20 Luminometer (Promega, Madison, WI) with a sensitivity range of 0.05-10,000. Each transfection experiment was performed at least three times with different DNA preparations. Correction for transfection efficiency was made by co-transfection with the ␤-galactosidase expression vector pCMV-␤-galactosidase, and the corresponding enzymatic activity was determined using the Galacto-Light kit (Tropix). The mean and S.D. were calculated, and experimental results of the promoter constructs were displayed either as arbitrary units of luciferase activity or as a fold induction to the corresponding untreated sample.
For DNAP experiments, COS cells were transfected with the expression vectors encoding HIF-1␣, FLAG-Smad3, or the constitutively activated form of the TGF-␤ receptor type I (T␤RI). Transfected cells were resuspended in a buffer containing 50 mM HEPES, pH 7.5, 50 mM NaCl, 0.1% Tween 20, and 10% glycerol, supplemented with protease and phosphatase inhibitors and lysed by sonication. Cell extracts were incubated with 200 ng of double-stranded biotinylated Ϫ1006/Ϫ954 oligonucleotide, and DNA-protein complexes were isolated by centrifugation with ImmunoPure streptavidin-agarose (Pierce). Complexed proteins were separated by SDS-PAGE and transferred onto nitrocellulose membranes, and the presence of HIF-1␣ and Smad3 was revealed with specific monoclonal antibodies using a chemiluminescence assay. DNAP experiments were repeated at least four times with similar results, and a representative experiment is shown in the corresponding figure.
Immunoprecipitation, Pull-down, and Western Blot Analysis-For immunoprecipitation experiments, HeLa cells were transfected with the appropriate expression vectors, and 48 h later, cells were collected by centrifugation, lysed, and subjected to immunoprecipitation with anti-human HIF-1␣ (mAb OZ 12ϩ15, Lab Vision Corp.) or anti-FLAG (Sigma) antibodies using protein G-Sepharose (Amersham Pharmacia Biotech). For GST pull-downs, GST fusion constructs of full-length Smad3, Smad3-MH1 (amino acids 1-144), Smad3-MH2 (amino acids 199 -440), and Smad3-NC (non-conserved linker region; amino acids 145-234) were purified using glutathione-Sepharose 4B beads (Amersham Pharmacia Biotech). In vitro transcription/translation GST pulldowns with in vitro transcribed and translated HIF-1␣ or HIF-1␤ in pcDNA3 were conducted as described (9). Basically, HIF-1␣ or HIF-1␤ labeled with [ 35 S]Met were incubated with glutathione-Sepharosebound fusion proteins on ice for 2 h. Beads were washed five times in wash buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 0.1% Triton X-100, 10% glycerol), and specifically bound proteins were detected by SDS-PAGE and autoradiography. For Western blot analysis, cell extracts, DNAP, and immunoprecipitates were subjected to SDS-PAGE under reducing conditions, and proteins were electrophoretically transferred to nitrocellulose membranes (Millipore Corp., Bedford, MA). Filters were blocked with phosphate-buffered saline containing 5% non-fat dry milk for 1 h. Specific immunodetection was carried out by incubation with anti-HIF-1␣ (mAb 54, Becton Dickinson; mAb H1␣67, Novus Biologicals), anti-Smad3 (Santa Cruz Biotechnology), or anti-FLAG antibodies, followed by peroxidase-conjugated rabbit anti-mouse Ig at room temperature. The presence of antigens was revealed using a chemiluminescence assay (Supersignal detection kit, Pierce). Experi-ments were repeated at least three times with similar results, and representative experiments are shown in the corresponding figures.

TGF-␤ and Hypoxia Cooperate to Induce Transcription of
Human VEGF-To determine whether TGF-␤ and hypoxiatriggered pathways could collaborate to induce VEGF gene expression, human Hep3B cells were chosen because they have been shown to stimulate production of VEGF in response to hypoxic conditions (35). Thus, Hep3B cells were incubated in hypoxic or normoxic conditions, with or without TGF-␤, and the culture supernatants were assayed for secreted VEGF protein by ELISA (Fig. 1A). Hypoxia, and to a lower extent TGF-␤, enhanced the production of VEGF, particularly to untreated cells under normoxic conditions. Thus, at 72 h, VEGF secretion was increased 44% with hypoxia and 20% with TGF-␤, whereas the combined treatment of hypoxia/TGF-␤ resulted in a much higher production (146% increase) of VEGF. To investigate whether this collaboration was also present at the VEGF transcript level, Hep3B and endothelial cells were incubated with TGF-␤1 under normoxic or hypoxic (1% oxygen) atmospheres; total RNA was extracted, and VEGF transcripts were analyzed by Northern blot (Fig. 1B). VEGF mRNA levels were found to be almost unaffected in response to TGF-␤ and moderately increased upon hypoxia treatment, whereas the simultaneous stimulation with both TGF-␤ and hypoxia showed a marked synergistic effect. To assess whether this collaborative effect was taking place at the transcriptional level, we analyzed the activity of the human VEGF promoter region (38,39). For this purpose, we used a reporter construct containing the 5Ј-flanking region of the human VEGF gene promoter (Ϫ1910/ϩ379), fused to the luciferase gene. This VEGF promoter construct contains the hypoxia-response element (HRE) located at Ϫ975/ Ϫ968, where HIF-1␣ binds to mediate hypoxia induction of VEGF (35). For transfection experiments with promoter constructs, the myoblast cell line L6E9 was chosen because preliminary experiments in our laboratory demonstrated a relatively high efficiency of transfection, as well as reproducible and comparable responses to hypoxia and TGF-␤ as individual stimuli; also, muscle cells are known to be a major source VEGF in vivo (45). As shown in Fig. 1C, transient transfection experiments demonstrated induction of the VEGF promoter activity in the presence of either TGF-␤ (1.8-fold) or hypoxia (2.7-fold), whereas the simultaneous presence of both stimuli resulted in a significant cooperative effect (7.3-fold induction). In a parallel experiment, protein levels of HIF-1␣ were stimulated by hypoxia but were not affected by the TGF-␤ treatment (Fig. 1C, lower panel), excluding the possibility of a TGF-␤-dependent induction of HIF-1␣ as responsible for the VEGF activation. These results suggest that the collaboration between TGF-␤ and hypoxia has a transcriptional control basis.
Synergistic Action of Smads and HIF-1␣ on the Activity of the Human VEGF Promoter-TGF-␤ and hypoxia signaling pathways mediate their transcriptional regulation mainly through Smads (2-4) and HIF-1 (26, 28) transcription complexes, respectively. Thus, the involvement of these transcription factors in the TGF-␤/hypoxia collaboration was analyzed. In order to elucidate the participation of Smad proteins, co-transfection experiments of the human VEGF promoter construct with expression vectors coding for different Smad members that could mediate this effect, namely Smad2, Smad3, and the co-smad Smad4, were carried out ( Fig. 2A). The VEGF promoter activity was found to be increased significantly by Smads under normoxic conditions (between 1.4-and 3-fold induction), and it was markedly augmented in hypoxia. Smad3 (4.7-fold), and to a lower extent Smad2 (2.5-fold) and Smad4 (2.7-fold), clearly increased the hypoxia-dependent induction, either in the presence or in the absence of TGF-␤. As expected, the strongest activation was observed when Smad3 and Smad4 were coexpressed under hypoxic conditions, yielding a 6.3-fold stimulation. Thus, this Smad3/Smad4 combination was selected for future studies. Next, we analyzed the participation of HIF-1 by co-transfection of the VEGF promoter construct with an expression vector coding for HIF-1␣ (Fig. 2B). HIF-1␣ expression was able to synergize with TGF-␤ resulting in an enhanced VEGF promoter activity. Furthermore, in the presence of several combinations of Smad3/Smad4, HIF-1␣ expression mark-edlyincreasedtheVEGFpromoteractivity.ThestrongestHIF1␣dependent response was observed in the presence of Smad3/ Smad4 with or without TGF-␤. These results demonstrate that cooperation between TGF-␤ and hypoxia is mediated at the FIG. 1. Effect of TGF-␤ and hypoxia on human VEGF expression. A, Hep3B cells were cultured in the absence of serum in either normoxic or hypoxic (1% oxygen) conditions, with or without TGF-␤, for the times indicated. The medium was collected from triplicate wells, and the levels of secreted VEGF were determined by ELISA as described under "Experimental Procedures." The data were corrected for cell number, and the mean concentration of VEGF is shown. This is a representative experiment out of three different ones. B, human hepatoma Hep3B and endothelial HMEC-1 cells were exposed to normoxia or hypoxia (1% oxygen) and incubated in the presence or in the absence of TGF-␤ for 16 h, as indicated. Total RNA was extracted, and VEGF transcripts were detected by Northern blot analysis. The blots were stained by ethidium bromide to visualize the 28 S ribosomal RNA. C, L6E9 myoblasts were transiently transfected with the promoter construct VEGF(Ϫ1910/ϩ379)-LUC. After 24 h, cells were exposed to normoxia or hypoxia (1% oxygen) and incubated in the presence or in the absence of TGF-␤, as indicated. Transcriptional activity was measured using the luciferase reporter assay (upper panel). For comparative purposes, the activity of the VEGF promoter construct in the absence of treatment was given the arbitrary value of 1. As a negative control, the activity of the promoter-less vector pGL2 was not affected by the treatment. Cell extracts were also used to determine the levels of HIF-1␣ by Western blot (WB) analysis using the specific mAb H1␣67 (lower panel). The presence of a nonspecific (NS) band of higher electrophoretic mobility than HIF-1␣ is indicated. A representative experiment out of four different ones is shown. transcriptional level by the cooperation between Smads and HIF-1␣ proteins.
Cooperation between Smads and HIF-1␣ Occurs at the Fragment Ϫ1006/Ϫ954 of the Human VEGF Promoter-Within the human VEGF promoter, the major HRE motif, responsible for HIF-1 binding, has been localized at Ϫ975/Ϫ968 (35). By contrast, to our knowledge, no specific binding motifs for Smad proteins have been reported within the VEGF promoter. To analyze further the mechanism of synergy between Smads and HIF-1␣, we focused our attention on the HRE element at Ϫ975/ Ϫ968 and its flanking sequences. Thus, fragment Ϫ1006/Ϫ954 containing not only the HRE element at Ϫ975/Ϫ968 but also the flanking sequence GCCAGACT encoding the putative Smad-binding motifs GNCNGNCN (1) and AGAC box (46) was used to synthesize the promoter construct 2HRE-LUC (Fig.  3A). This construct contains the dimerized Ϫ1006/Ϫ954 fragment from the VEGF promoter fused to a minimal TATA promoter and the luciferase gene. As shown in Fig. 3B, the promoter activity of 2HRE-LUC clearly reproduced in myoblasts the synergy between Smads and HIF-1␣, previously observed with the full-length VEGF promoter. In addition, similar experiments were carried out in SW480.7 cells, which lack Smad4 and consequently are unresponsive to TGF-␤1 (9). Accordingly, TGF-␤1 treatment did not significantly transactivate the TGF-␤ reporter plasmid 3TPlux in SW480.7 cells (Fig. 3C), whereas a clear transactivation was obtained in HepG2 cells (Fig. 3D). Also, no TGF-␤ induction of 2HRE-LUC promoter activity, whether in normoxia or hypoxia, could be observed in SW480.7 cells (Fig. 3C). On the contrary, the 2HRE-LUC promoter activity was clearly enhanced upon TGF-␤ treatment in HepG2 cells under hypoxic or normoxic conditions (Fig. 3D). Further analysis of the TGF-␤-and hypoxia-responsive elements within the Ϫ1006/Ϫ954 sequence was carried out in SW480.7 cells by co-transfection of HIF-1␣, Smad3, and Smad4 (Fig. 4). Thus, the 2HRE-LUC construct did not show any synergistic effect between Smad3 and HIF-1␣ in SW480.7 cells (Fig. 4), as compared with the marked cooperation observed in myoblasts (Fig. 3B) and hepatocytes (data not shown). As a control of the experiment, the cooperation between Smad3 and HIF-1␣ could be rescued upon co-transfection of Smad4. After 24 h, cells were exposed to normoxia or hypoxia (1% oxygen) and incubated in the presence or in the absence of TGF-␤. Transcriptional activity was measured using the luciferase reporter assay. A representative experiment out of five different ones is shown in each panel.
These results indicate that the cooperative effect between TGF-␤ and hypoxia maps within the Ϫ1006/Ϫ954 fragment of the VEGF promoter and is mediated by HIF-1␣ and Smad3/Smad4.
Identification of HIF-1 and Smad-binding Motifs within Fragment Ϫ1006/Ϫ954 of the Human VEGF Promoter-To analyze the interaction of HIF-1␣ and Smad3 proteins with the Ϫ1006/Ϫ954 DNA fragment, EMSA studies were conducted (Fig. 5, A and B). Specific complexes between HIF-1 and the DNA probe were detected in extracts from HeLa cells subjected to hypoxia (Fig. 5A). The retarded complex was triggered by hypoxia and could be supershifted by preincubation with antibodies against HIF-1␣, in agreement with previous results (35). No alteration of the hypoxia-triggered complex was detected upon treatment with TGF-␤, or by transfection with Smad3/ Smad4, or preincubation with anti-Smad antibodies (data not shown), pointing out the difficulty in observing the labile Smad-DNA interaction using nuclear extracts. This is similar to the difficulty in detecting Smad in association with the Sp1-DNA complex, even though Smad and Sp1 do synergize in the activity of several TGF-␤-inducible gene promoters (10,13,18). To study the interaction between Smads and the Ϫ1006/ Ϫ954 DNA fragment, EMSA studies with GST-Smad proteins were carried out (Fig. 5B). No specific band could be detected with GST alone. Specific complexes could be detected in the presence of GST-Smad3 or GST-Smad4, which were more intense and shifted up when both proteins were present.
Further proof for the direct interaction of HIF-1␣ and Smad3 with the Ϫ1006/Ϫ954 oligonucleotide was obtained by DNA affinity precipitation studies. Nuclear extracts from COStransfected cells were incubated with a biotinylated Ϫ1006/ Ϫ954 oligonucleotide, and isolated DNA-protein complexes were separated by SDS-PAGE, and specific protein bands were recognized by antibodies to HIF-1␣ or Smad3 (Fig. 5C). Both, HIF-1␣ and Smad3 could be detected in association with the Ϫ1006/Ϫ954 oligonucleotide. In the case of Smad3, the association with the DNA probe was revealed only in the presence of the activated TGF-␤ receptor type I. As a control, the expres-  D), as well as with expression vectors coding for Smad3, Smad4, and HIF-1␣ (B), as indicated. After 24 h, cells were exposed to normoxia or hypoxia (1% oxygen) and incubated in the presence or in the absence of TGF-␤, as indicated. Transcriptional activity was measured using the luciferase reporter assay. A representative experiment out of three different ones is shown in each panel. sion of recombinant Smad3 in total extracts from transfected cells was determined in the same experiment.
To confirm the formation of a ternary complex containing Smad, HIF-1, and DNA, electrophoretic mobility shift assays were carried out using recombinant proteins (Fig. 5D). HIF-1␣ and its DNA-binding partner HIF-1␤ were synthesized using an in vitro translation system and were incubated with the radiolabeled oligonucleotide Ϫ1006/Ϫ954 to allow the formation of a specific DNA-HIF-1 complex. As described above (Fig.  5B), small amounts of Smad3/Smad4 proteins were able to form a complex with the probe, which was visible upon overexposure of the gel (Fig. 5D, lane 3, data not shown). Also, the simultaneous presence of HIF-1␣ and HIF-1␤ yielded a specific complex, as expected (lane 4). When both HIF-1 and Smad proteins were present, a distinct band could be observed (see asterisk in lane 5), suggesting the formation of a ternary complex. Despite its predicted larger size, this ternary complex displays a higher mobility than the band corresponding to the HIF-1 complex alone. A similar behavior was reported previously for Smad3-Smad4 complexes that show a higher mobility than those of the individual Smad3 or Smad4 proteins, when synthetic DNA consensus sequences are used (47). The presence of HIF-1 and Smad proteins in this complex was further demonstrated by the supershift effect induced by anti-HIF-1␣ (lane 6), anti-HIF-1␤ (lane 9), or anti-Smad (lane 7) antibodies. Interestingly, the bands supershifted by anti-HIF-1␣ (lanes 6 and 8) are much more intense than those in the absence of the antibody, suggesting that anti-HIF-1␣ confers stability to the complex. Overall, these results demonstrate that Smad3/Smad4 and HIF-1␣ are able to interact with the Ϫ1006/Ϫ954 oligonucleotide.
To define more specifically the HIF-1␣-and Smad3-interacting sequences located within the fragment Ϫ1006/Ϫ954 of the human VEGF promoter, wild type and mutant oligonucleotides were generated (Fig. 6A). These oligonucleotides were used as competitors in EMSA experiments (Fig. 6, B and C). The hypoxiadependent formation of a specific complex was competed out by the wild type Ϫ1006/Ϫ954 oligonucleotide, the oligonucleotide Ϫ1006/Ϫ954 mutated at HIF-1 flanking positions Ϫ984 and Ϫ963, and the oligonucleotide Ϫ1006/Ϫ954 mutated at the consensus Smad sequence (Fig. 6B). However, the same complex was not competed out by the oligonucleotide Ϫ1006/Ϫ954 mutated at the HIF-1 site. This finding demonstrates that the motif at Ϫ975/Ϫ968 is responsible for HIF-1 binding, in agreement with a previous report (35). On the other hand, the specific DNA complex formed by Smad3/Smad4 (Fig. 6C) was competed out by excess of the unlabeled wild type probe, and by the oligonucleotide Ϫ1006/Ϫ954 mutated at HIF-1 flanking positions Ϫ984 and Ϫ963, but not by the oligonucleotide Ϫ1006/Ϫ954 mutated at the consensus Smad sequence, nor by the oligonucleotide Ϫ985/Ϫ954 devoid of the Smad motif, suggesting that the consensus Smad site at Ϫ992/Ϫ986 plays a critical role in the Smad protein binding. Taken together, these results demonstrate that HIF-1 and Smad proteins are able to bind consensus motifs within fragment Ϫ1006/Ϫ954 of the VEGF promoter.
To analyze the functional involvement of the HIF-1 and Smad consensus motifs, the wild type and two different mutant oligonucleotides were cloned into the luciferase reporter vector pGL2-p, and the activity of the resulting constructs was analyzed (Fig. 6, D and E). The responsiveness to hypoxia or TGF-␤ of the wild type plasmid (Fig. 6D) was found to be lower than that of the dimerized form shown in Fig. 3. Mutation of the HIF-1 consensus motif at Ϫ974 resulted in a significantly impaired response of 1HRE/HM-Luc to hypoxia, whereas the low TGF-␤ response was unaffected. On the other hand, mutation of the Smad consensus binding motif at Ϫ992 yielded a reduction of the 1HRE/SM-Luc activity to the TGF-␤ stimulus, whereas the hypoxia-dependent stimulation was not affected. Furthermore, both 1HRE/HM-Luc and 1HRE/SM-Luc constructs showed an impaired response to the simultaneous stim-FIG. 5. Characterization of HIF-1 and Smad binding to the ؊1006/؊954 VEGF fragment. A, electrophoretic mobility shift assay with HIF-1 present in nuclear extracts. HeLa cells were subjected or not to hypoxia for 4 h, as indicated. Nuclear extracts were incubated with the radiolabeled Ϫ1006/Ϫ954 oligonucleotide, used as a probe, either in the absence or in the presence of anti-HIF-1␣ (H) or nonspecific (C) antibodies. The hypoxia-dependent induction of a specific complex is indicated by an arrowhead on the left. This complex shows a supershift (SS) effect in the presence of anti-HIF-1␣. B, electrophoretic mobility shift assay with recombinant Smad proteins. The radiolabeled Ϫ1006/ Ϫ954 oligonucleotide, used as a probe, was incubated with 1 g of bacterially expressed GST, Smad3-GST, or Smad4-GST, as indicated. The asterisk indicates the presence of specific Smad-DNA complexes. C, DNAP analysis. COS cells were transfected with expression vectors encoding FLAG epitope-tagged Smad3, HIF-1␣, or a constitutively activated form of the T␤RI. Cell extracts were incubated with the biotinylated Ϫ1006/Ϫ954 oligonucleotide, and DNA-protein complexes were isolated by centrifugation with streptavidin-agarose. Complexed proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes, and the presence of HIF-1␣ and Smad3 was revealed with specific monoclonal antibodies using a chemiluminescence assay. As a control, the expression of Smad3 in the total lysates was also determined. D, electrophoretic mobility shift assay with recombinant HIF-1 and Smad proteins. HIF-1␣ and HIF-1␤ proteins, synthesized in vitro using a TNT kit (Promega), and 0.1 g of bacterially expressed GST-Smad3 (S3) and GST-Smad4 (S4) proteins, were incubated with the radiolabeled Ϫ1006/Ϫ954 oligonucleotide, used as a probe. The specific complexes containing Smad3/4, HIF-1, or HIF/Smad (*) are indicated. Supershifted complexes (SS) induced by anti-HIF-1␣ (H), anti-HIF-1␤ (H␤), or anti-Smad3/Smad4 (S) are also shown. ulation by hypoxia and TGF-␤. Similar results were obtained when using the HIF-1␣ and Smad expression vectors to mimic the hypoxia or TGF-␤ stimuli, respectively (Fig. 6E). These results allowed the identification, within fragment Ϫ1006/ Ϫ954, of the functional motifs necessary for hypoxia-and TGF-␤-dependent activation at Ϫ974 and Ϫ992, respectively. These motifs were also shown to be necessary for the collaboration between hypoxia and TGF-␤ responses.
Physical Interaction between Smad3 and HIF-1␣-Smad proteins regulate transcription in collaboration with other transcription factors through direct protein-protein interactions (2)(3)(4). To determine the molecular basis of the cooperation between Smad3 and HIF-1␣, we co-transfected HeLa cells with expression vectors encoding both transcription factors (Fig. 7A). Immunoprecipitation with anti-Smad antibodies followed by the immunodetection with anti-HIF-1␣ demonstrated the association between Smad3 and HIF-1␣. This interaction could be detected in the absence of TGF-␤ stimulation, although co-transfection with the activated T␤RI slightly improved the amount of HIF-1␣ co-precipitated with Smad3. Conversely, HeLa cells were subjected to hypoxia or TGF-␤ treatments, and the endogenous HIF-1␣ was immunoprecipitated from cellular lysates, followed by immunodetection of the endogenous Smad3 using specific antibodies, demonstrating again the association between HIF-1␣ and Smad3 (Fig. 7B). In these experiments (Fig. 7, A and B), the expression of either recombinant or endogenous Smad3 or HIF-1␣ proteins in total cell extracts was also revealed. To determine whether the interaction between Smad3 and HIF-1␣ was direct, we examined the association using an in vitro binding assay. Full-length cDNA encoding HIF-1␣ factor was transcribed and translated in vitro, labeled with [ 35 S]methionine, and incubated in the presence of recombinant GST-Smad3 (Fig. 7C). GST-Smad3, but not GST alone, showed binding to HIF-1␣. Furthermore, MH1 and MH2 domains, but not the linker domain of Smad3, were also able to bind HIF-1␣. Parallel studies with HIF-1␤ did not show any specific binding to the GST-Smad3 protein. These results suggest that HIF-1␣ interacts with Smad3 through the MH1 and MH2 domains.

DISCUSSION
Hypoxia and TGF-␤ are known to regulate angiogenesis, a process where VEGF plays a critical role promoting the formation of blood vessels, by inducing proliferation, migration, elongation, network formation, and branching of endothelial cells (40,41,48). Involvement of individual components of the TGF-␤ and hypoxia pathways in the angiogenic process has been demonstrated during recent years by gene ablation experiments in mice. Thus, deletion of genes encoding different components of the TGF-␤ system, including TGF-␤1 (49), activin receptor-like kinase-1 (50,51), TGF-␤ receptor type I (52), endoglin (53)(54)(55), or Smad5 (56), leads to defective angiogenic and vascular remodeling processes. Similarly, loss of HIF-1␣ in mice reduces hypoxia-induced expression of VEGF and impairs vascular formation and function (57).
VEGF expression has been shown to be induced by either hypoxia (35)(36)(37) or TGF-␤ (58,59) pathways. Here, we show that hypoxia and TGF-␤ cooperate to induce expression of human VEGF at the transcriptional level. This cooperation cannot be explained by the cross-stimulation of the respective signaling pathways as TGF-␤ alone does not induce HIF-1␣ protein levels (Fig. 1B) nor the formation of HIF-1-DNA complexes (data not shown). Conversely, hypoxia does not enhance the activity of the TGF-␤ reporter vector p3TPlux (Fig. 3,  C and D).
In our study we provide a molecular characterization of the mechanism for the cooperative effect. Fragment Ϫ1006/Ϫ954, within the human VEGF promoter, has been identified as a target for the HIF-1␣-and Smad-binding sites, sustaining the synergistic activity between TGF-␤ and hypoxia. Furthermore, the demonstration of a physical interaction between Smads and HIF-1␣ provides a solid basis for this cooperativity. This is in agreement with previous reports showing that Smads are recruited to specific regulatory elements of TGF-␤ target genes After 24 h, cells were exposed to normoxia or hypoxia (1% oxygen) and incubated in the presence or in the absence of TGF-␤1, as indicated. Transcriptional activity was measured using the luciferase reporter assay. For comparative purposes, the activity of the VEGF promoter constructs in the absence of treatment was given the arbitrary value of 1. This is a representative experiment out of three different ones. E, L6E9 myoblasts were transiently transfected with the reporter vectors 1HRE/WT-LUC, 1HRE/HM-LUC, and 1HRE/SM-LUC and expression vectors coding for Smad3, Smad4, or HIF-1␣, as indicated. After 48 h, the transcriptional activity was measured using the luciferase reporter assay. For comparative purposes, the activity of the VEGF promoter constructs in the absence of treatment were given the arbitrary value of 1. This is a representative experiment out of three different ones. through their association with different transcription factors (1)(2)(3)(4). This association, and the presence of Smad-binding motifs adjacent to those of the HIF-1␣ within the Ϫ1006/Ϫ954 fragment, are thought to result in the stabilization of the transcriptional complex (1)(2)(3)(4). Similar adjacent Smad-binding sites have been identified previously (1) for other Smad-interacting transcription factors in several TGF-␤ target genes. It is worth noting that although we have studied the Smad and HIF-1 sites on the Ϫ1006/Ϫ954 fragment, we cannot exclude the existence of additional TGF-␤ and/or hypoxia-responsive ele-ments within the rest of the VEGF promoter. The physiological relevance of the Ϫ1006/Ϫ954 fragment has been demonstrated recently (60) using a knock-in mice where deletion of this HRE element leads to chronic vascular insufficiency and motor neuron degeneration.
Analysis of the human VEGF promoter activity indicates that HIF-1␣ cooperates with Smad3, and to a lower extent with Smad2, but the optimal cooperation was achieved in the presence of Smad3/Smad4. Thus, overexpression of Smad3/Smad4 seemed to be sufficient for synergism with either hypoxia or HIF-1␣ in the absence of ligand stimulation. The marked increase of the promoter activity induced by Smad4 is probably due to the hetero-oligomerization process between the receptorregulated Smads and the common Smad4 (2-4). Smad1, Smad5, or Smad8 seem to regulate BMP-dependent cellular responses, whereas both Smad3 and Smad2 have been involved in TGF-␤-mediated signal transduction. TGF-␤ is a potent growth inhibitor for most cell types and can also induce programmed cell death (61,62). Interestingly, reduced proliferation and apoptosis in response to hypoxia has also been found in HIF-1␣ϩ/ϩ, but not in HIF-1␣Ϫ/Ϫ embryonic stem cells (57), suggesting cellular proliferation/apoptosis as a synergistic meeting point for TGF-␤ and hypoxia. In agreement with this hypothesis, treatment of TGF-␤1 plus hypoxia induced the greatest levels of apoptosis in endothelial cells (63). Given the wide range of biological processes regulated by hypoxia and TGF-␤, it can be postulated that further cooperative examples remain to be discovered. Also, hypoxic treatments for long periods result in the late up-regulation of TGF-␤1, thereby suggesting the induction of an autocrine loop by hypoxia (64). Both stimuli are involved in several human diseases, including cardiovascular ischemia, pulmonary hypertension, cancer, or pregnancy disorders (62,65,66). With regard to this latter pathology, Caniggia et al. (67) have reported that HIF-1 mediates the biological effects of oxygen on human trophoblast differentiation through TGF-␤3, suggesting a collaboration of both stimuli. Therefore, it is possible that additional target genes are synergistically regulated in a similar fashion as VEGF.
The functional data suggest that Smad3 acts as a co-modulator of the HIF-1-mediated transcriptional activity. Several studies (68 -71) have shown that Smad proteins can bind to CBP/p300 to activate transcription. This association might involve third parties, as is the case of Smad1 and STAT3, bridged by CBP/p300, forming a complex which leads to the cooperative signal of leukemia inhibitory factor and BMP2 (16). Since HIF-1␣ has been also reported to associate with p300 to stimulate VEGF transcription (72), it is possible that the HIF-1/ Smad association on the VEGF promoter could lead to a more efficient recruitment of p300 into the DNA-protein complex (1).
We have characterized the interaction between Smad and HIF-1␣ as responsible for a cooperative effect in human VEGF transcription. This is in line with the direct Smad interaction described with other DNA-binding partners such as FAST, OAZ, AP-1, ATF2, LEF1/TCF, vitamin D receptor, or TFE3. Smad3 interacts with DNA through the MH1 domain, whereas Smad3 association with other proteins is mediated at least by the MH1 (TFE3, AP-1), the MH2 (FAST, OAZ, AML, Co-Smad), or both MH1 and MH2 (LEF1/TCF) domains (1,2,73). We find that Smad3 interacts with the transcription factor HIF-1␣ through the MH1 and MH2 domains, similarly to LEF1/TCF (73). However, there is no apparent sequence homology between HIF-1␣ and LEF1/TCF proteins. Interestingly, HIF-1␣ and TFE3 are members of the basic helix-loop-helix family of transcription factors, and they also have in common their capacity to interact with the MH1 domain of Smad3 (15). It can be FIG. 7. Interaction between Smad3 and HIF-1␣. A, HeLa cells were transfected with expression vectors encoding FLAG-Smad3, HIF-1␣, and the activated form of the T␤RI, as indicated. Cell lysates were immunoprecipitated with anti-FLAG antibodies. Immunoprecipitates (IP) and total extracts were separated by SDS-PAGE and blotted onto nitrocellulose, and filters were incubated with anti-HIF-1␣ or anti-FLAG antibodies. The presence of HIF-1␣ or Smad3 was revealed using a chemiluminescence assay. B, HeLa cells were subjected to hypoxia or TGF-␤ treatments as indicated. Cell lysates were immunoprecipitated with anti-HIF-1␣ or anti-FLAG (Control Ab) antibodies. Immunoprecipitates and total extracts were separated by SDS-PAGE and blotted onto nitrocellulose. The presence of endogenous Smad3 and HIF-1␣ was revealed by incubating the filters with anti-Smad3 or anti-HIF-1␣ monoclonal antibodies, followed by a chemiluminescence assay. C, GST pull-downs. HIF-1␣ or HIF-1␤ were transcribed/translated in vitro in the presence of [ 35 S]methionine. Radiolabeled HIF-1␣ or HIF-1␤ was incubated in the presence of bacterially expressed GST, GST fused to the full-length Smad3 (FL-S3), GST fused to the MH1 domain of Smad3 (MH1-S3), GST fused to the MH2 domain of Smad3 (MH2-S3), or GST fused to the non-conserved linker domain of Smad3 (NC-S3), and the associated protein was detected by autoradiography (upper panel). Aliquots with the total input (10%) of radiolabeled HIF-1␣ or HIF-1␤ are shown. As a control, the amount of recombinant GST protein used in each sample was visualized by Coomassie Blue staining and is shown in the lower panel. The bands corresponding to HIF-1␣, HIF-1␤, or the different GST-Smad constructs are indicated. speculated that the MH1 domain of Smad3 binds directly to the consensus SBE on the VEGF gene promoter, and thus promotes assembly of a transcription complex with HIF-1␣ involving the MH1 and MH2 domains. The specific residues of the MH1 domain involved in the interaction with DNA have been determined using crystallographic data (46). Thus, it will be of interest to determine the specific residues of the MH1 and MH2 domains involved in the interaction with the transcription factor HIF-1␣ and their spatial relation with those involved in DNA binding.