Induction of phosphoglycerate kinase 1 gene expression by hypoxia. Roles of Arnt and HIF1alpha.

To identify new dimerization partners for the aromatic hydrocarbon receptor nuclear translocator (Arnt), we used its N-terminal region (amino acids 1-470) as a target in a two-hybrid screening procedure, and we cloned the murine form of hypoxia-inducible factor 1alpha (HIF1alpha). Sequence comparisons reveal substantial identity between mouse and human HIF1alpha. Hypoxia induces a 10-fold accumulation of phosphoglycerate kinase 1 mRNA in wild type mouse hepatoma (Hepa 1c1c7) cells; the induction mechanism is Arnt dependent because induction does not occur in Arnt-defective cells. Furthermore, induction of phosphoglycerate kinase 1 mRNA requires Arnt's N-terminal region, which mediates DNA binding and heterodimerization; in contrast, induction does not require Arnt's C-terminal region, which mediates transactivation. We also show that a GAL4-HIF1alpha fusion protein transactivates a GAL4-dependent gene in the absence of Arnt, that HIF1alpha's transactivation capability is inducible by hypoxia, and that both hypoxia responsiveness and transactivation capability reside within the C-terminal 83 amino acids of HIF1alpha. Our findings generate new insights into the mechanism by which Arnt and HIF1alpha induce transcription in response to hypoxia.

To identify new dimerization partners for the aromatic hydrocarbon receptor nuclear translocator (Arnt), we used its N-terminal region (amino acids 1-470) as a target in a two-hybrid screening procedure, and we cloned the murine form of hypoxia-inducible factor 1␣ (HIF1␣). Sequence comparisons reveal substantial identity between mouse and human HIF1␣. Hypoxia induces a 10-fold accumulation of phosphoglycerate kinase 1 mRNA in wild type mouse hepatoma (Hepa 1c1c7) cells; the induction mechanism is Arnt dependent because induction does not occur in Arnt-defective cells. Furthermore, induction of phosphoglycerate kinase 1 mRNA requires Arnt's N-terminal region, which mediates DNA binding and heterodimerization; in contrast, induction does not require Arnt's C-terminal region, which mediates transactivation. We also show that a GAL4-HIF1␣ fusion protein transactivates a GAL4-dependent gene in the absence of Arnt, that HIF1␣'s transactivation capability is inducible by hypoxia, and that both hypoxia responsiveness and transactivation capability reside within the C-terminal 83 amino acids of HIF1␣. Our findings generate new insights into the mechanism by which Arnt and HIF1␣ induce transcription in response to hypoxia.
The aromatic hydrocarbon receptor nuclear translocator (Arnt) 1 was first identified and characterized as a component of a heterodimeric, DNA-binding protein that regulates cytochrome P4501A1 (CYP1A1) transcription in response to the environmental contaminant 2,3,7,8-tetrachlorodibenzo-p-dioxin (dioxin) (1)(2)(3). Induction of CYP1A1 gene expression by dioxin involves the binding of 2,3,7,8-tetrachlorodibenzo-p-dioxin to its intracellular target, the aromatic hydrocarbon receptor (AhR) (1, 4 -7). The liganded AhR enters the nucleus, where it interacts with Arnt, generating an AhR/Arnt heterodimer that recognizes a specific nucleotide sequence within an enhancer upstream of the CYP1A1 gene (8 -11). The binding of AhR/Arnt to the enhancer induces transcription, which is accompanied by alterations in chromatin structure and the binding of general transcription factors to the CYP1A1 promoter (12)(13)(14). Induction of cytochrome P4501A1 enzyme activity increases the ability of the cell to detoxify aromatic hydrocarbons (15). Thus, induction represents an interesting mechanism by which the cell adapts to changes in its external environment.
As is typical of many transcription factors, AhR and Arnt have modular organizations. For example, both AhR and Arnt contain basic helix-loop-helix (bHLH) motifs; the HLH domains mediate heterodimerization between AhR and Arnt, while the basic regions are responsible for the binding of the AhR/Arnt heterodimer to DNA (16 -18). Both AhR and Arnt exhibit regions of amino acid sequence homology with Per (a Drosophila circadian rhythm protein) and Sim (a protein involved in Drosophila central nervous system development) (19,20). These regions of homology (PAS domains) may contribute to proteinprotein interactions during heterodimerization (18,21). In addition, AhR and Arnt have transactivation domains, which are functionally distinct from the DNA-binding and heterodimerization domains (22)(23)(24)(25)(26).
Others (1) have suggested that Arnt might play a more general role in mediating cellular responses to environmental stimuli and, therefore, that cells might contain additional proteins that serve as heterodimerization partners for Arnt. In support of this idea, studies of human hypoxia-inducible factor 1, which regulates cellular responses to low oxygen tension, reveal that it contains two bHLH/PAS proteins, hypoxia-inducible factor 1␣ (HIF1␣) and Arnt (27). Such findings raise the intriguing possibility that cellular adaptation to hypoxia involves a mechanism that is Arnt dependent. Arnt could be very influential in such a role because adaptation to hypoxia involves increases in erythropoiesis, vascularization, and glycolytic enzyme activity, as well as other changes (28 -31).
We have undertaken a two-hybrid screening approach to find new proteins that interact with Arnt. We have used as the target Arnt's N-terminal region, which participates in Arnt's protein-protein interactions with AhR. Using this approach, we have identified mouse HIF1␣ as a heterodimerization partner for mouse Arnt. We demonstrate the functional importance of this observation by showing that Arnt is required for induction of the glycolytic enzyme phosphoglycerate kinase 1 (PGK1) in response to hypoxia. Our studies provide new insights into the mechanism by which cells adapt to low oxygen tension, and they emphasize the importance of Arnt as a regulator of cellular responses to environmental stimuli. was from U. S. Biochemical Corp. Reagents for chloramphenicol acetyltransferase and ␤-galactosidase assays were from Promega. The reagent for measuring protein concentration was from Bio-Rad. Cell culture materials were from Life Technologies Inc. 2,3,7,8-Tetrachlorodibenzo-p-dioxin was from the National Cancer Institute Chemical Carcinogen Reference Standard Repository. Polybrene (hexadimethrine bromide) was from Aldrich.
cDNA Cloning-A mouse liver cDNA library and the two-hybrid system were purchased from Clontech. The cDNA library was cloned into an expression vector (pGAD424) containing a GAL4 transactivation domain. The N-terminal portion of Arnt (amino acids 1-470, designated here as ArntNT) was cloned into an expression vector (pGBT9) containing a GAL4 DNA-binding domain. Using protocols suggested by Clontech, the ArntNT/pGBT9 plasmid was cotransfected together with the cDNA library/pGAD424 plasmids into yeast strain HF7c, which carries a lacZ reporter gene under the control of GAL4-binding sites. The interaction between the two fusion proteins generates a functional transcription factor that activates lacZ expression. ␤-Galactosidase activity was measured 3 days after transfection. Using this approach, we isolated a cDNA fragment (ϳ1.2 kb), which was sequenced using Sequenase 2.0 from U. S. Biochemical Corp. To obtain a full-length cDNA, the 1.2-kb fragment was used to screen a mouse hepatoma (Hepa 1c1c7) cDNA library. A cDNA (ϳ4 kb) was isolated and sequenced as described previously (23).
Construction of Arnt Deletion Mutants-We used PCR to generate fragments of Arnt. Primer sets containing an NcoI site and start codon (forward primers) or a BamHI site and stop codon (reverse primers) were designed to correspond to the ends of the desired fragments, as indicated in Fig. 3A. The PCR reaction was performed as described previously (12). The PCR products were digested with NcoI and BamHI and were ligated between the NcoI and BamHI sites of the retroviral expression vector pMFG (12).
Retroviral Expression of Full-length and Mutant Arnt cDNAs-3 g of recombinant retrovirus were transfected into the ecotropic packaging cell line BOSC23 (33). 48 h after transfection, 1 ml of the virus-containing medium was removed and mixed with 2 ml of hepatoma cell culture medium containing Polybrene (5 g/ml). This virus mixture was placed in 60-mm plates containing approximately 2 ϫ 10 5 cells; the plates were centrifuged at 1,000 ϫ g for 90 min at 32°C (34). Infection efficiency was estimated from the percentage of blue cells in populations infected with virus containing lacZ cDNA and was always greater than 95%.
Construction of GAL4-HIF1␣ Fusion cDNAs-We used PCR to generate fragments of HIF1␣. Primer sets containing a BamHI site and start codon (forward primers) or a ClaI site and stop codon (reverse primers) were designed to correspond to the desired ends, as indicated in Figs. 4A and 5A. The PCR reaction contained (in 100 l): 50 pmol of primers, 4 fmol of HIF1␣ template, 25 nmol of dNTPs, and 4 units of Vent polymerase (New England Biolabs). The following PCR conditions were used: 94°C, 5 min; 1 cycle, then 94°C, 1 min; 55°C, 1 min; 72°C, 2 min; 35 cycles, then 94°C, 1 min; 55°C, 1 min; 72°C, 4 min; 1 cycle. The PCR products were digested with BamHI and ClaI and were ligated between the BamHI and ClaI sites of the pGALO expression vector (a gift from C. V. Dang).
Transfection and CAT Assay-Transfection was carried out using Arnt-defective cells, as described previously (8,23). The GAL4-HIF1␣ cDNAs were cotransfected with a reporter gene, pG5E4TCAT, which contains a CAT gene downstream of five copies of the GAL4 DNAbinding site (36). A ␤-galactosidase expression vector pCH110 (Promega) was also cotransfected to control for transfection efficiency. 24 h after the transfection, cells were exposed to hypoxic conditions, as indicated. After 18 h of hypoxia (1% O 2 , 5% CO 2 , 94% N 2 ), cell lysates were prepared and assayed for CAT activity, ␤-galactosidase activity, and protein concentration, as described previously (12,23).

Cloning of a Dimerization Partner for Arnt-Previous observations indicate that Arnt forms heterodimers with AhR and
Sim, both of which contain bHLH and PAS domains (37). Such findings suggested the existence of additional bHLH/PAS partners for Arnt. We tested this hypothesis by using a yeast two-hybrid system to screen a mouse liver cDNA library; this approach identifies potential partners by virtue of their ability to interact directly with the target protein (38). As the target, we used the N-terminal region of Arnt (amino acids 1-470) because it contains the HLH and PAS domains, which are primarily responsible for Arnt's interactions with AhR and, we presumed, for Arnt's interactions with other proteins. In addition, Arnt's C terminus had to be deleted from the target because the C terminus contains a transactivation function that introduces false-positive artifacts into the two-hybrid screening procedure.
Using two-hybrid screening, we isolated a 1.2-kb cDNA fragment, whose nucleotide sequence indicated that it was homologous to HIF1␣, previously identified in human cells (27). We used this 1.2-kb fragment as a hybridization probe to isolate full-length HIF1␣ cDNA from a mouse hepatoma (Hepa 1c1c7) cDNA library. Fig. 1 reveals the nucleotide and deduced amino acid sequence of mouse HIF1␣. The mouse protein exhibits 90% amino acid identity (89% at the nucleotide level) with the human protein. The N-terminal region (amino acids 1-330) is somewhat more conserved and exhibits 96% amino acid identity (91% at the nucleotide level) between the species; the C-terminal region (amino acids 331-822) exhibits 86% identity at both the amino acid and nucleotide levels between the mouse and human proteins. Like the human protein, murine HIF1␣ is a bHLH/PAS protein that exhibits similarities to Per, AhR, Arnt, and Sim (27).
Role of Arnt in the Response to Hypoxia-The identification of HIF1␣ as a potential heterodimerization partner for Arnt raised the possibility that Arnt may participate in cellular responses to hypoxia. To test this hypothesis, we first identified a hypoxia-responsive gene suitable for analysis in mouse hepatoma cells. PGK1 is a glycolytic enzyme that is induced by hypoxia in other cell systems (30,39). We observed that exposure of wild type mouse hepatoma cells to hypoxic conditions induced the accumulation of PGK1 mRNA by about 10-fold (Fig. 2). In contrast, hypoxia failed to induce PGK1 mRNA in Arnt-defective cells; however, reconstitution of Arnt-defective cells with full-length Arnt cDNA restored the response of the PGK1 gene to hypoxia (Fig. 2). These findings demonstrate that Arnt is required for the induction of PGK1 mRNA by hypoxia in this cell system.
To delineate in greater detail Arnt's role in regulating the response to hypoxia, we analyzed the induction of PGK1 mRNA in Arnt-defective cells that had been reconstituted with Arnt mutants (Fig. 3A). In these studies, the high efficiency of retroviral gene transfer allows us to analyze the response of the target PGK1 gene in its native chromosomal setting, thus avoiding potential artifacts associated with the use of an episomal reporter gene. Our results (Fig. 3B) reveal that hypoxia fails to induce PGK1 mRNA in cells reconstituted with Arnt⌬bHLH and Arnt⌬NT, mutants which lack Arnt's bHLH domain and N-terminal domain, respectively. These observations imply that Arnt's DNA-binding and heterodimerization functions are required for the induction of PGK1 gene transcription. We envision that heterodimerization generates an HIF1␣/Arnt complex that induces transcription by binding to its cognate DNA recognition motif in the vicinity of the PGK1 gene (30). In contrast, substantial induction of PGK1 mRNA occurs in cells reconstituted with Arnt⌬C1 and Arnt⌬C2, mutants which lack Arnt's transactivation domain (Fig. 3B). These findings imply that the response of the PGK1 gene to hypoxia does not require transactivation by Arnt.
Transactivation Function of HIF1␣-Because the induction of PGK1 gene expression by hypoxia does not require Arnt's transactivation capability, we hypothesized that HIF1␣ must itself contain a transactivation function; we tested this idea by measuring the ability of GAL4-HIF1␣ fusion proteins (Fig. 4A) to activate the expression of a GAL4-dependent reporter gene. We performed these studies in Arnt-defective cells to ensure that our findings reflect the transactivation capability of HIF1␣ and not the HIF1␣/Arnt heterodimer. Our results (Fig.  4B) reveal that HIF1␣ exhibits constitutive (basal) transacti- vation capability that is about 5-fold higher than background; in addition, we find that its transactivation function is inducible about 5-fold by hypoxia. Analyses of deletion mutants reveal that HIF1␣'s inducible transactivation function is located within the C-terminal 491 amino acids of the protein and not within its N-terminal portion, which contains its bHLH and PAS domains. These observations indicate that HIF1␣ contains a hypoxia-inducible transactivation capability that does not require Arnt for expression. Also, HIF1␣'s transactivation domain functions independently of its (presumed) DNA-binding and heterodimerization domains; thus, like other transcription factors, HIF1␣ appears to have a modular organization (40).
The inducible nature of its transactivation function raised the possibility that HIF1␣ contained a hypoxia-responsive region that was distinct from its transactivation domain. For example, Arnt's other dimerization partner, AhR, has a dioxinresponsive domain that is separate from its transactivation domain (22,24,26). To determine whether hypoxia responsiveness and transactivation were separable functions, we assayed progressively smaller fragments of HIF1␣'s C-terminal region for hypoxia-inducible transactivation. Our findings (Fig. 5) reveal that HIF1␣'s constitutive transactivation capability maps to an 83-amino acid segment at the protein's C terminus; furthermore, this segment also exhibits the ability to respond to hypoxia. Thus, the domains responsible for transactivation and hypoxia responsiveness occupy the same small region of HIF1␣, implying that they overlap or are congruent. FIG. 1.-continued   FIG. 2. Induction by hypoxia of PGK1 mRNA in mouse hepatoma cells. Cells were exposed to hypoxic conditions for 18 h. 10 g of total RNA was analyzed by 1% agarose gel electrophoresis, transfer to nitrocellulose, hybridization with 32 P-labeled PGK1 cDNA, and autoradiography. Lane 1, wild type cells; lane 2, Arnt-defective cells; lane 3, Arnt-defective cells reconstituted with full-length Arnt cDNA.

FIG. 3. Induction by hypoxia of PGK1 mRNA in Arnt-defective cells reconstituted with full-length or mutant Arnt cDNAs. Panel
A, structure of full-length and mutant Arnt cDNAs. Panel B, reconstitution experiments. Arnt-defective cells were reconstituted with the indicated cDNAs by retroviral infection. Cells were exposed to hypoxic conditions for 18 h. 10 g of total RNA was analyzed by 1% agarose gel electrophoresis, transfer to nitrocellulose, hybridization to 32 P-labeled PGK1 cDNA, and autoradiography. Lane 1, full-length Arnt; lane 2, Arnt⌬C1; lane 3, Arnt⌬C2; lane 4, Arnt⌬bHLH; lane 5, Arnt⌬NT. DISCUSSION We have used a yeast two-hybrid system to clone and identify mouse HIF1␣ as a dimerization partner for mouse Arnt. The functional importance of this protein-protein interaction is revealed by our observation that the PGK1 gene fails to respond to hypoxia in Arnt-defective cells. Our findings complement and extend those of Wang et al. (27) and Wang and Semenza (41), who identified HIF1␣ and Arnt as components of a DNAbinding protein complex that mediates cellular responses to hypoxia in human cells. The high amino acid sequence homology between mouse and human HIF1␣ suggests that the regulatory mechanism by which cells adapt to low oxygen tension is similar for the two species.
Two observations imply that Arnt's HLH domain is important for its interaction with HIF1␣. First, successful cloning relied upon a protein-protein interaction between HIF1␣ and a fragment of Arnt that contains its bHLH and PAS domains, which have been implicated in Arnt's heterodimerization capability. Second, deletion of Arnt's bHLH domain abrogates the response of the PGK1 gene to oxygen deprivation. The simplest explanation for this finding is that Arnt⌬bHLH fails to heterodimerize with HIF1␣.
Our findings indicate that Arnt's transactivation domain is dispensable for the induction of PGK1 gene expression by hypoxia. Similarly, we reported previously that Arnt's transactivation capability is not required for the induction of CYP1A1 transcription by dioxin (12). Therefore, in both the hypoxiaresponsive and dioxin-responsive systems, Arnt appears to serve as a dimerization partner that confers DNA recognition capability upon HIF1␣/Arnt and AhR/Arnt, respectively.
Studies of dioxin-inducible transcription reveal that AhR/ Arnt binds to the consensus DNA sequence 5Ј-TNGCGTG-3Ј in the CYP1A1 enhancer (42,43). Cross-linking studies and binding site selection experiments suggest that Arnt's basic domain binds to the 5Ј-GTG-3Ј component of the CYP1A1 consensus sequence (37,44). By comparison, studies of hypoxia-responsive genes imply that HIF1␣/Arnt recognizes the consensus DNA sequence 5Ј-(G/Y)ACGTGC(G/T)-3Ј upstream of PGK1 (30). We note that this nucleotide sequence also contains a 5Ј-GTG-3Ј motif; we speculate that it interacts with Arnt's basic domain during the induction of PGK1 transcription by hypoxia. We also note that AhR/Arnt's recognition sequence and HIF1␣/Arnt's recognition sequence both contain a CpG dinucleotide, a motif that may undergo cytosine methylation. Cytosine methylation blocks the binding of AhR/Arnt to its recognition sequence and the response to dioxin (45). By analogy, we envision that cytosine methylation may also block DNA binding by HIF1␣/Arnt and the response to hypoxia. Thus, we hypothesize that DNA methylation has the potential to inhibit some cellular responses to decreased oxygen tension, possibly in tissue-specific fashion.
Our findings indicate that HIF1␣ and AhR differ with respect to their transactivation mechanisms. In particular, we show that HIF1␣'s transactivation function is expressed independently of Arnt; however, in the dioxin-responsive system, transactivation by AhR requires heterodimerization with Arnt. We envision that heterodimerization triggers a conformational change in AhR, which allows its latent transactivation function to be expressed (12,24). Transactivation by HIF1␣ must involve a different mechanism because it does not require Arnt for expression.
In several other receptor-dependent regulatory systems (i.e. those that respond to glucocorticoid hormones, heat shock, or dioxin), the receptor domain that mediates inducibility is distinct from the domain that mediates transactivation (24,46,47). Thus, the observation that hypoxia responsiveness and transactivation capability map to the same 83-amino acid region of HIF1␣ is unusual because it suggests that the functional domains are congruent. Its amino acid composition and primary sequence reveal that HIF1␣'s C-terminal 83-amino acid segment is not rich in acidic residues, glutamine, or proline; thus, it does not resemble certain transactivation domains described previously (40). The segment is enriched in leucine and isoleucine residues (ϳ20%), as well as in serine residues (ϳ11%), and it exhibits clusters of hydrophilic and hydrophobic amino acids. These atypical properties may reflect the fact that the segment mediates hypoxia responsiveness as well as transactivation and, therefore, that its structure reflects a combination of these two functions.
The fact that Arnt is a component of more than one regulatory pathway is interesting for several reasons. First, it increases the plausibility that Arnt mediates additional adaptive responses to environmental stimuli and that other Arnt-dependent signaling systems exist within the cell. Second, it raises the possibility that "cross-talk" exists between Arnt-dependent regulatory systems and that activation of one pathway (e.g. by dioxin) may affect the cell's ability to respond to a second stimulus (e.g. hypoxia). These appear to be potentially interesting issues for future research.