Oxygen Sensing and the Transcriptional Regulation of Oxygen-Responsive Genes in Yeast

Baker’s yeast, Saccharomyces cerevisiae, is a facultative aerobe that responds to changes in oxygen availability by differentially expressing a large number of genes (reviewed by Zitomer and Lowry, 1992; Pinkham and Keng, 1994; Bunn and Poyton, 1996). This response, in conjunction with the regulation of gene expression by carbon substrates (e.g. repressing versus nonrepressing and fermentable versus nonfermentable sugars), is part of a biochemical and genetic program that regulates the efficiency of carbon-source utilization during aerobic growth, which is supported by both mitochondrial oxidative phosphorylation and glycolysis, and during anaerobic growth, which is supported exclusively by fermentative processes. The transcriptional regulation of genes by carbon substrates is complex and has been reviewed elsewhere (Trumbly, 1992; de Winde and Grivell, 1993; Thevelein, 1994).

Genes that respond to changes in oxygen availability can be placed into one of two broad categories: ‘aerobic’ genes, which are optimally expressed under normoxic conditions, and ‘hypoxic’ genes, which are optimally expressed under low-oxygen or anoxic conditions. Not surprisingly, many of the respiratory cytochromes and other proteins involved in strictly aerobic metabolism, including enzymes for controlling oxidative damage, are encoded by aerobic genes. However, most of the genes that are induced (derepressed) by oxygen deprivation also encode proteins that function in oxygen-utilizing pathways. These proteins include oxidases involved in electron transport and reductases and desaturases involved in the biosynthesis of heme, sterol and unsaturated fatty acids. Thus, the products of most hypoxic genes do not function in anaerobic (fermentative) metabolism per se and can be referred to in a functional sense as hypoxic genes rather than anaerobic genes. Why is the transcription of many of the hypoxic genes maximally upregulated (derepressed) during anaerobiosis, a condition in which their products serve no apparent function? Although direct experimental evidence is lacking in many cases, the derepression of hypoxic genes probably serves to increase flux through these biosynthetic pathways during oxygen-limiting conditions – by increasing the efficiency of oxygen usage (e.g. hypoxic isoenzymes) or simply by increasing protein levels – because many of these genes encode enzymes (or enzyme subunits) that are rate-limiting in their respective pathways. The importance of maintaining flux through these biosynthetic pathways is underscored by an absolute nutritional requirement for sterols and unsaturated fatty acids during anaerobic growth (Andreasen and Stier, 1953, 1954).

Fig. 1 conceptualizes oxygen-sensing pathways in S. cerevisiae. The transcriptional control of nuclear-encoded, oxygen-responsive genes in this organism is mediated by at least three trans-acting factors: Hap1p (heme activated protein), which activates the transcription of many of the aerobic and some hypoxic genes; Hap2/3/4/5p, which activates the expression of several aerobic genes, typically in a carbon-source-dependent manner; and Rox1p (regulation by oxygen), which represses the transcription of the hypoxic genes under aerobic conditions (reviewed by Zitomer and Lowry, 1992; Pinkham and Keng, 1994; Bunn and Poyton, 1996). Heme acts as an intermediary in regulating the expression of oxygen-responsive genes; it is required for the activation of the Hap proteins and for the transcription of the ROX1 gene (mediated by Hap1p). Because the biosynthesis of heme requires oxygen, it has been proposed that heme acts as a gauge of oxygen availability and dictates which set of genes will be transcribed. Under aerobic or heme-proficient conditions, transcription of the aerobic genes is activated (mediated, in part, by the Hap proteins), and under anoxic or heme-deficient conditions, transcription of the hypoxic genes is derepressed (owing to loss of repression by Rox1p).

Previous studies have focused primarily on the trans-acting factors and cis-sites that regulate the transcription of oxygen-responsive genes, while upstream events have gone largely unstudied. Currently, the signals to which the cell responds are unclear. In addition to O₂per se, byproducts and endproducts of oxygen-dependent metabolism (e.g. cellular redox, reactive oxygen species, heme) may act in signaling pathways that lead to changes in the transcription of specific genes. Similarly, little is known about the receptors (either their cellular location or identity) that are responsive to changes in these signals. Also unclear are how ‘oxygen sensors’ transduce this signal for the activation or repression of oxygen-responsive genes and how many pathways exist.

Currently, it is assumed that heme plays a central role in oxygen-sensing and signal-transduction pathways by acting as a redox-insensitive, metabolic cofactor or ligand for transcription factors (e.g. Hap1p) (reviewed by Zitomer and Lowry, 1992). However, the following recent findings are difficult to reconcile with such a model. First, transcript levels of many aerobic genes decrease with declining oxygen concentration over a range (200–1 μmol l⁻¹ O₂) that is well above the K_m for oxygen of the rate-limiting step in heme biosynthesis (Burke et al. 1997). Cellular concentrations of heme are not thought to vary appreciably in this range of oxygen concentration (Labbe-Bois and Labbe, 1990). Similarly, transcripts of some of the hypoxic genes also respond (increase) over this range of oxygen concentration (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). Second, after cells are shifted from aerobic to anaerobic conditions, the time course for induction and transcript profiles differ markedly among the hypoxic genes (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). These data are difficult to reconcile with models invoking the derepression of hypoxic genes simply by the loss of common repressor (e.g. Rox1p) or effector (e.g. heme) elements. Third, clamping the redox state of hemoproteins under different conditions of oxygen availability (e.g. carbon monoxide + anoxia, transition metals + air) modulates the expression of several hypoxic genes (K. E. Kwast, P. V. Burke, B. Staahl, S. Fontaine and R. O. Poyton, in preparation), suggesting that one of the signaling pathways involves changes in the redox state of a hemoprotein oxygen sensor. Taken together, these data suggest that multiple mechanisms/pathways regulate the expression of these genes; some probably involve control by the concentration of heme and others probably involve the redox state of hemoproteins. In this paper, we provide an overview of the transcriptional regulation of oxygen-responsive genes and discuss the role that both heme and hemoproteins play in oxygen-sensing pathways.

In Saccharomyces cerevisiae, the expression of a large number of proteins is oxygen-responsive, many at the level of transcription. Table 1 is a compilation of genes whose transcription is oxygen and/or heme-regulated. All of the genes listed are nuclear-encoded. Genes are grouped according to function, and known trans-acting activators and repressors are indicated. As can be seen, the effects of oxygen and heme act in parallel, i.e. those genes that are positively regulated by oxygen are also positively regulated by heme and vice versa.

The first group of genes (group I) in Table 1 contains a large number that encode proteins involved in mitochondrial respiration and oxidative phosphorylation. It includes many of the nuclear-encoded proteins that constitute the terminal portion of the mitochondrial respiratory chain, including complex III (ubiquinol cytochrome c reductase), complex IV (cytochrome c oxidase) and cytochrome c, the mobile electron carrier acting between these complexes. In addition, this group contains the gene encoding cytochrome b₂ and three genes encoding isoforms of the mitochondrial adenine translocase. Given the roles that these proteins play in oxidative metabolism, it is not surprising that all of their genes – with the exceptions of COX5b, CYC7 and AAC3 (discussed below) – are positively regulated by oxygen/heme. In addition to affecting the expression of the nuclear-encoded subunits of cytochrome c oxidase listed in Table 1, oxygen also positively affects the expression of the mitochondrial genes COX1 and COX2, encoding the catalytic core of cytochrome c oxidase; this effect has been shown to be mediated post-transcriptionally (Groot and Poyton, 1975). Thus, the availability of oxygen/heme regulates the expression of a large portion of the mitochondrial respiratory chain.

COX5a/COX5b, CYC1/CYC7 and AAC2/AAC3, as well as TIF51a/ANB1 (see group IV genes below), are part of a family of genetically unlinked but functionally paired genes in which oxygen/heme activates the transcription of the aerobic isoform and represses the transcription of the hypoxic isoform. In all cases, the aerobic and hypoxic isoforms are functionally interchangeable. The primary sequence in the coding regions of these gene pairs is remarkably similar, for example 66% for COX5a/COX5b (Cumsky et al. 1987) and 79% for CYC1/CYC7 (Montgomery et al. 1980). Given the high degree of sequence homology, it is likely that these pairs arose by gene duplication and have subsequently diverged to function in different oxygen environments. Indeed, both the CYC1/CYC7 and TIF51a/ANB1 gene pairs have been shown to be part of a large cluster of duplicated genes (Kang et al. 1992; Melnick and Sherman, 1993). Although functional differences have not been documented for some of these pairs, the products of the gene pairs COX5a/COX5b and CYC1/CYC7 have been shown to influence the maximal turnover number of holocytochrome c oxidase, with the hypoxic isoforms increasing this rate several-fold (Allen et al. 1995; Burke and Poyton, 1998).

Group II (Table 1) includes several hypoxic genes whose products are involved in the synthesis of heme, sterol and unsaturated fatty acids. These biosynthetic pathways require molecular oxygen as an electron acceptor in redox reactions, and most of these genes encode enzymes that utilize oxygen directly. Unlike transcripts of the genes encoding hypoxic isoforms, transcripts of these hypoxic genes are detectable under normoxic conditions, and levels increase further in response to declining oxygen concentration (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). Although it has been reported that HMG1 and HMG2 represent an aerobic and hypoxic (respectively) gene pair, whose transcription is regulated in opposite directions by oxygen/heme (Thorsness et al. 1989), we have found that both of these genes are optimally expressed in a number of wild-type strains under low-oxygen or anoxic conditions in both glucose-repressed and nonrepressed conditions (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). Moreover, aerobic transcript levels of both HMG1 and HMG2 are derepressed in Δrox1 and rox1 mutant strains, indicating that both of these genes are repressed by Rox1p (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). In support of this finding, a search of the promoter region of these genes reveals putative Rox1p-binding sequences (see Rox1 section below). Thus, it appears that both HMG1 and HMG2 are hypoxic genes that are regulated by Rox1p. There are striking differences in the post-translational regulation of these isoforms that indicate different functional roles for these isoforms (Hampton et al. 1996).

Group III (Table 1) includes a number of genes that encode enzymes involved in the oxidative stress response. Not surprisingly, their transcription is positively regulated by oxygen/heme as well as, for some, reactive oxygen species. Lastly, group IV (Table 1) contains an aerobic and hypoxic gene pair of the putative translational initiation factor eIF-5 encoded by TIF51a/ANB1. Functional differences between these isoforms have not been documented. In addition, group IV contains the aerobic gene ROX1, which encodes a transcriptional repressor of all of the hypoxic genes listed in Table 1.

The transcription of most of the aerobic genes is activated by Hap1p, the Hap2/3/4/5p complex, or both Hap1p and Hap2/3/4/5p. For some of the aerobic genes (COX6, COX7, COX9, COX8, CTA1 and SOD1), the trans-acting factor(s) responsible for their upregulation in response to heme/oxygen has not been identified. It makes intuitive sense that genes encoding enzymes that utilize oxygen are positively regulated by heme/oxygen and, in some cases, induced by growth on a nonfermentable carbon substrate (Hap2/3/4/5p activation) that must be oxidized in the mitochondrion. It is important to realize, however, that each of these genes (including the hypoxic genes) is independently regulated by these and possibly other trans-acting factors. Thus, as a set, the aerobic (and hypoxic) genes do not constitute a coordinately expressed regulon or operon.

The negative effect of heme/oxygen on the transcription of the hypoxic genes has been shown to be mediated by the transcriptional repressor Rox1p. For a subset of these hypoxic genes (CYC7, HEM13, HMG1 and ERG11), heme can act in both a positive (mediated by Hap1p) and a negative (mediated by Rox1p) manner. This combinatorial regulation by heme may allow for the fine-tuning of transcript levels of these genes in response to heme/oxygen. However, the overall effect of heme/oxygen proficiency on the transcription of hypoxic genes is negative. As stated above, given that most of the hypoxic genes encode products that are involved in the direct use of oxygen, their high levels of expression under strict anoxia are puzzling unless this upregulation (derepression) represents an adaptive response to make better use of a limiting substrate (oxygen).

As discussed above, the expression of most of the oxygen-responsive genes is regulated by three transcription factors: Hap1p, Hap2/3/4/5p and Rox1p. In this section, we provide a more detailed review of transcriptional regulation by these factors. A discussion of earlier work on these components can be found in prior reviews (Forsburg and Guarente, 1989b; de Winde and Grivell, 1993; Zitomer and Lowry, 1992; Pinkham and Keng, 1994; Bunn and Poyton, 1996).

The CYP1 gene was first identified as an activator of the hypoxic gene CYC7 (Clavilier et al. 1969) and later found to be the same gene as HAP1 (Verdiere et al. 1986), an activator of CYC1 (Guarente et al. 1984). For simplicity, we shall use the HAP1 designation here. Since the discovery of HAP1, the number of genes it has been found to regulate has steadily increased (Table 1). The HAP1 gene has been cloned and sequenced (Creusot et al. 1988; Verdiere et al. 1988; Pfeifer et al. 1989) and found to encode a 1483-amino-acid protein with at least three functional domains. The first is a zinc-cluster DNA-binding domain in the amino terminus, between residues 1 and 148. This domain is responsible for sequence-specific DNA binding and the formation of the Hap1p homodimer (Verdiere et al. 1988; Pfeifer et al. 1989; Timmerman et al. 1996), which is the transcriptionally active form. This C6 zinc-cluster motif (Zn[II]2Cys6 binuclear cluster) has a high level of structural identity with other yeast regulatory proteins, including Gal4p, Ppr1p, Leu3p, Put3p and Cha4p (Schjerling and Holmberg, 1996). A second domain is found in the carboxy terminus, between residues 1309 and 1483. This acidic activation domain is required for the transcriptional activation of Hap1p (Pfeifer et al. 1989) and is similar to other transcriptional activation domains such as that in Gal4p. Lastly, there are two regulatory domains: one between residues 244 and 444, and the other between residues 1192 and 1197 (Pfeifer et al. 1989; Zhang and Guarente, 1995). Both of these regions contain a Lys/Arg-Cys-Pro-Val/Ile-Asp-His motif that has been implicated as a metal- or heme-binding site (Creusot et al. 1988; Zhang and Guarente, 1996). This sequence motif occurs six times in the first regulatory domain and once in the second domain of Hap1p, and it is found in a number of heme-dependent regulatory proteins (Lathrop and Timko, 1993).

The upstream activator sites (UASs) in the promoter region of Hap1p-regulated genes have been studied extensively. The Hap1p homodimer binds to two different classes of sequences: one represented by the UAS1 of CYC1 and the other represented by the UAS of CYC7 (Pfeifer et al. 1987b). The sequences within these UASs are divergent, and hap1 mutants have been identified that bind to the UAS of CYC7 but not to the UAS1 of CYC1 (Kim and Guarente, 1989; Turcotte and Guarente, 1992; Verdiere et al. 1988). The DNA target for Hap1p binding has been further delineated through mutational analyses and DNAase I footprinting. The sequence consists of a direct repeat of a CGG triplet separated by a specific number of nucleotides (Zhang and Guarente, 1996); the consensus sequence has been determined to be CGG N3 TAN CGG N3 TA (Ha et al. 1996). This sequence motif is somewhat different from that recognized by other yeast regulatory proteins that contain the C6 zinc-cluster domain because most of these proteins recognize an inverted repeat rather than a direct repeat, as is the case for Hap1p. Recently, the three-dimensional structure of Hap1p bound to DNA has been examined using ¹H,¹⁵N resonance spectroscopy (Timmerman et al. 1996). Interestingly, the specificity of DNA binding is attributable, in part, to a fine tuning between the structure of the Hap1p linker peptide and/or dimerization helix and the number of bases separating the two CGGs (Timmerman et al. 1996).

The role of heme in Hap1p activation has been the focus of considerable research. Although it was originally thought that heme was required for Hap1p to bind to its DNA target sequence (via a heme-dependent unmasking of the DNA-binding domain; Pfeifer et al. 1989), this model has been recently modified and refined (Fytlovich et al. 1993; Zhang and Guarente, 1994, 1995, 1996). Using in vitro DNA-binding assays with crude cellular extracts, it has been shown that Hap1p binds to its target DNA in a heme-independent manner, albeit probably more weakly, and forms a large complex with other, as yet unidentified, cellular protein(s) (Fytlovich et al. 1993; Zhang and Guarente, 1996). Residues 244–444, containing the heme regulatory motifs (HRMs) 1–6, are required for the formation of this larger complex (Fytlovich et al. 1993; Zhang and Guarente, 1996; Lodi et al. 1996). When hemin (oxidized heme) is added, a smaller complex is formed.

These observations have led to the following model of heme-regulated Hap1p activity. In the absence of heme, Hap1p is weakly bound to its target DNA and is transcriptionally inactive because of the binding of repressor proteins(s) to the region between residues 244 and 444. When heme is present, a smaller, presumably transcriptionally active, complex is formed because of heme binding to HRM1–6; the binding of heme to this region may mask the binding site for the transcriptional repressor(s) (Zhang and Guarente, 1995). Heme also plays a role in Hap1p activation through its interaction with the second regulatory domain (residues 1192–1197), which probably indirectly influences the acidic activation domain of Hap1p (Zhang and Guarente, 1995). Thus, heme may regulate the activity of Hap1p by (1) masking the binding sites for repressor protein(s), and possibly regulating DNA-binding affinity, via residues 244–444, and (2) regulating the activation domain through binding of heme to the second regulatory domain (residues 1192–1197).

Despite considerable progress in our understanding of the regulation of Hap1p activity by heme, a number of questions remain. Although recent studies have shown that a synthetic peptide of the HRM (Ala-Lys-Arg-Cys-Pro-Val-Asp-His-Thr-Met) reversibly binds heme with an affinity in the micromolar range (Zhang and Guarente, 1995), it is still not known whether Hap1p is a hemoprotein in vivo. Also unclear is whether Hap1p binds redox-active metals (e.g. Fe). Although current models of Hap1p activation regard heme as a metabolic cofactor, it is possible that heme might function as a redox-sensitive group that either directly or indirectly controls the activity of Hap1p. The binding of heme to the HRM is qualitatively different from that in hemoproteins such as globins and cytochromes (Zhang and Guarente, 1995; Bock et al. 1978; Choma et al. 1994). Whereas the axial iron ligands in cytochromes are typically a histidine/methionine pair or bis-histidine, both of which bind heme tightly and shift the spectral Soret peak to a longer wavelength, the sulfhydryl group of the cysteine residue in the HRM is thought to bind heme iron and shift the Soret peak to a shorter wavelength (Zhang and Guarente, 1995). It has been suggested that this difference allows heme to bind reversibly to the HRM (Zhang and Guarente, 1995), but it is not known whether this difference in binding precludes the possibility that this heme is redox-active. In the absence of heme, Hap1p may actually repress the transcription of genes (e.g. SOD2 and ROX1) that are activated in the presence of heme (Pinkham et al. 1997; Deckert et al. 1995a). It is not yet known how the larger, transcriptionally inactive complex that forms in the absence of heme may repress the transcription of these and possibly other Hap1p-regulated genes. Finally, it is interesting to note that the general transcription factors/mediators Tup1p and Ssn6p (Cyc8p), which are typically involved in repression, play a role in the activation of Hap1p (Zhang and Guarente, 1994); however, the mechanism for this activation is not known (see Rox1p section for a discussion of Tup1p/Ssn6p function). Undoubtedly, as additional cellular components of the Hap1p–DNA complex are identified, we will gain new insight into how the activity of this complex is regulated and how heme functions in the formation and activation of this transcription complex.

The other main transcriptional activator of oxygen-responsive genes in yeast is Hap2/3/4/5p. It is a highly conserved heteromeric complex that binds to CCAAT boxes in UASs containing the consensus sequence TNATTGTT (Forsburg and Guarente, 1988). This heteromer activates the transcription of genes primarily in response to heme/oxygen and/or growth on nonfermentable substrates (e.g. lactate, glycerol). Hap2/3/4/5p consists of four polypeptides: Hap2p, a 265-residue protein that contains a DNA-binding site consisting of 22 essential residues (Olesen et al. 1987; Pinkham et al. 1987; Forsburg and Guarente, 1989a; Olesen and Guarente, 1990; Xing et al. 1993); Hap3p, a 144-residue protein that is also required for DNA-binding (Xing et al. 1993); Hap4p, a 554-residue protein that contains an acidic activation domain (Forsburg and Guarente, 1989a; Olesen and Guarente, 1990); and Hap5p, a 216-residue protein that is required for both assembly and DNA-binding activity of the complex (McNabb et al. 1995). Hap2p and Hap3p are homologous to the mammalian transcription factors CPI-A and CPI-B (human) and CBF-A and CBF-B (rat) (Chodosh et al. 1988; Maity et al. 1990).

The Hap2/3/5p heterotrimer functions in DNA-binding, whereas Hap4p is a regulatory subunit required for the activation of the complex (McNabb et al. 1995). Transcription of both HAP2 and HAP4 is induced by nonfermentable carbon substrates, which may account for the increase in expression of Hap2/3/4/5p-regulated genes in response to these carbon substrates (Forsburg and Guarente, 1989a). There is some evidence that other regulatory proteins may substitute for Hap4p, suggesting that the Hap2/3/5p heterotrimer may be a general transcription factor whose activity is modulated by other regulatory proteins besides Hap4p (Forsburg and Guarente, 1989a; Olesen and Guarente, 1990; reviewed by Zitomer and Lowry, 1992). Moreover, additional cellular factors (e.g. Gcn5p) may be involved in the transcriptional activation of the Hap2/3/4/5p complex (Georgakopoulos and Thireos, 1992).

The function of heme in regulating the activity of Hap2/3/4/5p is unclear. Whereas the DNA-binding activity of Hap2/3/4/5p appears to be heme-independent, it is not known whether heme affects the overall abundance of the complex or its transcriptional activity. Forsburg and Guarente (1989a) suggested that heme could affect Hap4p post-translationally, but how this could influence the activity of the heteromeric complex is not known. Interestingly, the transcription of some of the Hap2/3/4/5p-regulated genes appears to be heme- and/or carbon-source-independent. These genes, which were not included in Table 1, fall into two classes: those regulated in a heme-independent, carbon-source-dependent manner (including ACO1, CIT1, KGD1, KGD2 and LPD1, which encode tricarboxylic acid cycle enzymes), and those regulated in a heme-independent, carbon-source-independent manner (including HEM1 and HEM3, which encode enzymes in the heme biosynthetic pathway) (reviewed by Pinkham and Keng, 1994). The differences in the regulation of these classes of genes by Hap2/3/4/5p are not understood, and they are mentioned here only for completeness. Finally, there are a number of Hap2/3/4/5p-regulated genes whose expression in response to oxygen, heme and carbon source is not fully known. These genes encode a diverse array of proteins involved in cellular respiration (SDH1, SDH3 and SDH4; Daignan-Fornier et al. 1994), glycolysis (FBP1 and PCK1; Mercado and Gancedo, 1992), glutamate synthesis (GDH1; Dang et al. 1996), sporulation (SPR3; Ozsarac et al. 1995) and vacuolar function (APE1; Bordallo et al. 1995).

Given the diverse array of genes regulated by Hap2/3/4/5p, as well as the number of environmental and physiological factors that apparently modulate Hap2/3/4/5p activity, it is tempting to speculate that this transcription complex is modular and that additional elements (possibly subunits other than Hap4p) may regulate its activity. Further dissection of Hap2/3/4/5p should improve our understanding of the ways in which environmental and physiological factors, including heme and oxygen, regulate the activity of this complex.

The Rox1 protein represses the transcription of nearly all of the hypoxic genes. Its gene, ROX1, was originally identified by the characterization of mutations that resulted in the aerobic derepression of the oxygen- and heme-repressed gene ANB1 (Lowry and Zitomer, 1984). Two classes of mutants were identified. One class, represented by the rox1-a1 allele, is semidominant and pleiotropic, affecting both heme-repressed (ANB1) and heme-induced (CYC1, TIF51A and SOD2) genes. The second class, represented by the rox1-b3 allele, is recessive and affects the expression of only heme-repressed genes (e.g. ANB1). The rox1-b3 mutant strain was later shown to have the same phenotype as a rox1 null strain (Balasubramanian et al. 1993), which suggests that the rox1-a1 strain may contain a second, as yet unidentified, mutation.

The ROX1 gene encodes a 368-amino-acid protein whose N-terminal region shows homology to other high-mobility group (HMG) classes of nonhistone chromatin proteins (Balasubramanian et al. 1993). The region that accounts for the DNA-binding specificity is found within the HMG domain, which lies between residues 9 and 93 (Balasubramanian et al. 1993; Di Flumeri et al. 1996). In vitro, Rox1p synthesized in Escherichia coli or with an in vitro wheat-germ translation system forms oligomers that are dependent upon an intact HMG domain (Di Flumeri et al. 1996; Zitomer et al. 1997). The C-terminal domain is required for transcriptional repression and is presumed to interact with the Ssn6p/Tup1p complex (see below) (Balasubramanian et al. 1993; Zitomer et al. 1997). The function of a third region, containing a run of glutamines (residues 103–123), is unclear (Zitomer et al. 1997).

The DNA sequence to which Rox1p binds has been determined through a series of mutational analyses of the promoter regions of several Rox1p-regulated genes. It is found within the upstream repressor site (URS) and consists of the consensus sequence YYYATTGTTCTC (Y=pyrimidine) (Lowry et al. 1990; Balasubramanian et al. 1993). In vitro binding studies using partially purified Rox1p with synthetic oligonucleotides have verified that Rox1p binds specifically to this consensus sequence (Di Flumeri et al. 1996). This sequence is typically found within 500 base pairs 5´ of the TATA box and has been identified in all known Rox1p-regulated genes (Table 2). Several genes contain multiple copies of the consensus sequence, but one copy is apparently sufficient for repression. From a comparison of these sequences, as well as mutational studies, it appears that substitutions in the first three pyrimidines and the last three nucleotides do not severely inhibit Rox1p binding. There is some evidence that, in addition to the consensus sequence, a flanking T-rich region may also be important for Rox1p binding or activity (Sabova et al. 1993).

Although the precise mechanism of repression by Rox1p is unclear, a heme-dependent cofactor is not required: when transformed with a plasmid carrying the ROX1 gene fused to the GAL1 promoter, hem1 null mutants grown on galactose in the absence of heme repress the transcription of hypoxic genes (Keng, 1992). However, heme is required for the transcription of the ROX1 gene, which is mediated, in part, by Hap1p (Keng, 1992). As stated above, in the absence of heme, Hap1p may repress the transcription of ROX1 (Deckert et al. 1995a). There is also some evidence that Hap1p is not the only transcriptional activator of ROX1 expression. Indeed, ROX1 transcript levels are only moderately depressed in Δhap1 mutants (Deckert et al. 1995a).

Two general transcription factors/mediators, Ssn6p (Cyc8p) and Tup1p, are required for Rox1p-mediated repression. When either protein is absent, Rox1p-regulated genes are expressed constitutively (Balasubramanian et al. 1993; Deckert et al. 1995a). Both Tup1p and Ssn6p are required for the activity of other DNA-binding transcriptional repressors that function to regulate a diverse array of cellular activities, including cell-type recognition and catabolite repression (Keleher et al. 1992; Schultz and Carlson, 1987; Trumbly, 1988; Williams and Trumbly, 1990). Although these factors do not bind DNA directly, they are thought to be recruited to form a complex with the DNA-binding repressors (Keleher et al. 1992; Varanasi et al. 1996; Tzamarias and Struhl, 1994, 1995). It has been proposed that Ssn6 provides the link to a pathway-specific DNA-binding protein (e.g. Rox1p), while Tup1p mediates repression (Tzamarias and Struhl, 1994, 1995). Indeed, Zitomer et al. (1997) recently collected evidence for a direct interaction between Ssn6 and Rox1p, and they suggest that Ssn6 may be involved in stabilizing the protein–DNA complex. The precise role of Tup1 in the formation of this complex and in repression is unclear.

The transcriptional regulation of genes by Rox1p can be summarized by the following model. In the presence of heme (e.g. aerobiosis), the ROX1 gene is transcribed, mediated in part by Hap1p activation, and translated. Cellular levels of Rox1p are regulated via autorepression, in that Rox1p represses the expression of its own gene (Deckert et al. 1995a). Once translated, Rox1p enters the nucleus, binds to consensus sequences in the URS(s) of hypoxic genes, and forms a complex with Ssn6p/Tup1p, resulting in the transcriptional repression of these genes. There is some evidence that Rox1p may bind to URSs with different affinities, which may, in part, account for differences in the transcript levels of a number of hypoxic genes under identical conditions of heme and oxygen availability (see discussions in Zitomer et al. 1997). In the absence of heme (e.g. anaerobiosis), the transcription of ROX1 is repressed, caused in part by Hap1p-mediated repression. Rox1p levels subsequently decline, repression is released, and the hypoxic genes are transcribed. A recent report suggests that Rox1p may be ‘rapidly’ degraded in the absence of heme/oxygen, although the mechanism responsible for this is not known (Zitomer et al. 1997).

In addition to ROX1, REO1 was identified as an aerobic repressor of the heme-mediated, hypoxic genes COX5b and ANB1 (Trueblood and Poyton, 1988). However, complementation studies between rox1 and reo1 strains have yielded conflicting results concerning whether REO1 and ROX1 are the same or different genes (Trueblood and Poyton, 1988; Lambert et al. 1994). To resolve this confusion, we recently characterized phenotypic differences between reo1 and rox1 mutants, performed additional complementation studies, and sequenced the ROX1 gene in reo1 strains (Kwast et al. 1997). These studies revealed a similar degree of aerobic derepression for all of the hypoxic genes examined in Δrox1, reo1 and the resulting diploid strain. Sequence analyses of ROX1 in reo1 strains revealed a frame-shift mutation in the 5´-end of the ROX1-coding region. This mutation results in a nonsense codon in the sixth position of the coding region (Kwast et al. 1997). Thus, it appears that reo1 is, in fact, an allele of ROX1.

In addition to ROX1 (REO1), there are several other genes whose products are thought to repress the transcription of hypoxic genes. These include ROX3, ROX5 and ROX6 (Rosenblum-Vos, 1988; Rosenblum-Vos et al. 1991). These genes were identified in mutant hunts for factors regulating the expression of CYC1, CYC7 or ANB1. Although mutations in rox3 result in the aerobic derepression of CYC7, ROX3 transcript levels increase during anaerobiosis in a heme-independent manner, and deletion of its product is lethal. These results suggest that Rox3p is probably a general transcription factor (Rosenblum-Vos et al. 1991). Indeed, recent studies have shown that Rox3p (synonymous with Ssn7p and Rmr1p) is a mediator and component of the RNA polymerase II holoenzyme (Gustafsson et al. 1997). Thus, through its involvement in the transcriptional regulation of CYC7, Rox3p contributes to the global stress response in S. cerevisiae (Gustafsson et al. 1997). The functional roles that ROX5 and ROX6 play in the repression of hypoxic genes are unclear and neither gene has been cloned.

Finally, recent studies have identified several genes, including DAN1, GPD2 and SRP1, that are repressed by heme or oxygen in an apparently Rox1p-independent manner (Sertil et al. 1997; Ansell et al. 1997; Donzeau et al. 1996). The expression of DAN1 is similar to that of other hypoxic genes in the following respects: its transcription is induced by anoxia, blocked by the addition of heme to anoxic cells and constitutive in heme mutants (Sertil et al. 1997). Because Rox1p does not influence its expression, these results suggest the existence of a parallel heme-dependent regulatory system. Similarly, the expression of GPD2, encoding an isoenzyme of NAD⁺-dependent glycerol-3-phosphate dehydrogenase, is induced under anoxia in a Rox1p-independent manner (Ansell et al. 1997). However, its transcription appears to be modulated by the redox state of the cell, suggesting a regulatory pathway that is different from that of other hypoxically expressed genes (Ansell et al. 1997).

In addition to DAN1 and GPD2, SRP1 (Donzeau et al. 1996) and other genes, such as SUT1 (Bourot and Karst, 1995) and TIP1 (Donzeau et al. 1996), are hypoxic genes that are expressed under anaerobic conditions, but the trans-acting factor(s) that mediates their expression is not known. Finally, the URSs in a number of hypoxic genes also contain the core sequence AAACGA (Sabova et al. 1993; Turi and Loper, 1992), but a transcriptional repressor that may interact with this sequence has not been identified. Thus, at present, Rox1p is the only known transcriptional repressor of hypoxic genes. However, as suggested by these studies and others discussed below, additional trans-acting factors and regulatory pathways are probably involved in modulating the expression of some of the hypoxic genes.

From the preceding discussion, it is clear that heme works together with a variety of trans-acting factors to regulate the transcription of oxygen-responsive genes in yeast. Further evidence for the central role of heme in the transcriptional regulation of these genes is provided by the following experimental observations. First, heme-deficient mutants derepress the transcription of hypoxic genes and repress the transcription of aerobic genes, irrespective of oxygen concentration (Hodge et al. 1989; Lowry and Lieber, 1986). Second, the addition of hemin to anoxic cells increases transcription of aerobic genes and decreases transcription of hypoxic genes (Hodge et al. 1989; Lowry and Lieber, 1986). Thus, heme can act either in a positive manner, activating transcription of primarily the aerobic genes, or in a negative manner, repressing the transcription of primarily the hypoxic genes. As a product of the mitochondrion and as a prosthetic group of respiratory cytochromes, heme may be ideally suited to coordinate the expression of the mitochondrial and nuclear genes involved in the biogenesis of the mitochondrial respiratory chain (Forsburg and Guarente, 1989b; Padmanaban et al. 1989). Indeed, heme is intimately entwined with energy production via its link as a prosthetic group in the mitochondrial cytochromes and other proteins directly involved in oxygen use and redox reactions. In the following sections, we provide an overview of the regulation of cellular levels of heme and discuss the functional roles of heme in transcription and oxygen-sensing pathways.

The heme biosynthetic pathway in yeast has been fairly well characterized and reviewed elsewhere (Labbe-Bois and Labbe, 1990; Pinkham and Keng, 1994). Two steps in its biosynthesis require molecular oxygen as an electron acceptor: the formation of protoporphyrinogen, catalyzed by coproporphyrinogen III oxidase, and the formation of protoporphyrin by protoporphyrinogen IX oxidase. Because of the requirement for oxygen in heme synthesis, it has been argued that cellular concentrations of heme reflect oxygen concentration (reviewed by Zitomer and Lowry, 1992). For this to be true, the rate-limiting step in heme synthesis must be oxygen-dependent. Except at very low oxygen concentrations, this does not appear to be the case. In aerobically growing cells, δ-aminolevulinate (ALA) accumulates as a result of low levels of ALA dehydratase (porphobilinogen synthase) and its low substrate affinity, suggesting that ALA dehydratase is rate-limiting under aerobic conditions (reviewed by Labbe-Bois and Labbe, 1990; Pinkham and Keng, 1994). However, under ‘near-anoxic’ (trace oxygen) conditions, coproporphyrinogen III oxidase, an oxygen-utilizing enzyme, is probably rate-limiting (reviewed by Labbe-Bois and Labbe, 1990; Pinkham and Keng, 1994). Moreover, the activity of coproporphyrinogen III oxidase increases in response to near-anoxic conditions (Miyake and Sugimura, 1968) and in strains with a defect in heme biosynthesis, regardless of the position of the block in the pathway (Labbe-Bois et al. 1980; Urban-Grimal and Labbe-Bois, 1981; Rytka et al. 1984). In addition, whereas other genes encoding enzymes in the heme biosynthetic pathway are thought to be expressed constitutively with respect to oxygen, the transcription of HEM13, which encodes coproporphyrinogen III oxidase, is repressed by heme/oxygen (mediated by Hap1p and Rox1p; Verdiere et al. 1991; Keng, 1992). This oxidase has a high affinity for oxygen, with an estimated K_m of below 0.1 μmol l⁻¹ O₂ (see discussion by Labbe-Bois and Labbe, 1990), suggesting that heme levels would not reflect oxygen concentration until near-anoxic conditions. Supporting this view is the lack of evidence that cellular heme concentrations vary appreciably at higher oxygen concentrations (Labbe-Bois and Labbe, 1990). Therefore, if cellular concentrations of heme are controlling the expression of oxygen-responsive genes, heme probably acts as an on–off switch only at extremely low oxygen levels. Because of the difficulty in measuring free heme levels in cells, this hypothesis has not yet been tested.

In addition to its effects on transcription, heme affects a large number of other cellular processes, including protein translation, transport, assembly and degradation (reviewed by Padmanaban et al. 1989). Because of the multitude of cellular activities that are influenced by heme, its synthesis, degradation and distribution are thought to be tightly regulated (Padmanaban et al. 1989). Although its synthesis is subject to feedback regulation that ensures adequate production, virtually nothing is known about the fate of heme in yeast once it is made in the mitochondrion (Labbe-Bois and Labbe, 1990). Yeast apparently lack heme oxygenase (Labbe-Bois and Labbe, 1990), which catalyzes the first step in heme degradation in higher eukaryotes. Moreover, heme degradation products (e.g. biliverdin, bilirubin) have not been found in yeast. Thus, there is no evidence that heme is degraded in yeast cells; under anoxic conditions, heme may simply be diluted as cell mass increases during anaerobiosis.

Similarly, little is known about the intracellular trafficking of heme. Once it is synthesized in the mitochondrion, heme must be distributed to other cellular compartments, including microsomes, peroxisomes, the nucleus and the cytosol. Given the high affinity of heme for proteins and lipids, it is likely that this distribution is carrier-mediated. There is some evidence in support of this view, at least in higher eukaryotes (Meier et al. 1978). Thus, it is unlikely that pools of free heme accumulate in these different cellular compartments. All of the steps in the synthesis, distribution and degradation or dilution of heme could be important regulatory checkpoints. A thorough understanding of the role of heme in regulating cellular function, including transcription, awaits further characterization of these processes.

There are at least three feasible pathways in which heme could be involved in the transcriptional regulation of genes; one involves control by the concentration of heme, and the other two invoke control by the redox state of hemoproteins (reviewed by Poyton and Burke, 1992; Bunn and Poyton, 1996). These different regulatory pathways are not mutually exclusive.

In the first type of pathway, heme would serve simply as a metabolic cofactor or ligand that binds to transcriptional components and regulates their activity, as has been proposed for the heme-dependent activation of Hap1p (Fytlovich et al. 1993; Zhang and Guarente, 1995). In this type of pathway, the concentration of heme, and not its redox state, would modulate the activity of transcriptional component(s). At very low oxygen concentrations, i.e. near the K_m of oxygen for coproporphyrinogen III oxidase, cellular heme concentrations could provide an effective on–off switch for the transcription of both the hypoxic and aerobic genes.

In the second type of pathway, heme would function as a redox-sensitive component of either a transcription factor or effector element that regulates the activity of transcriptional components (reviewed by Poyton and Burke, 1992; Bunn and Poyton, 1996). For example, the redox state of heme bound to Hap1p or other heme-mediated transcription factors could conceivably control their activity. Although Hap1p has been shown to bind heme in vitro, it is not known whether this heme is redox-active. Measurements of Hap1p activity in different redox environments would help clarify the role of heme in Hap1p activation.

In the third type of pathway, the redox (or spin) state of the iron in a hemoprotein oxygen sensor would modulate the activity of transcriptional components either directly or indirectly. For example, for mammalian cells, considerable evidence suggests that the redox state of a hemoprotein oxygen sensor controls the expression of a large number of hypoxic genes, acting through a transcriptional activator, hypoxia-inducible factor 1 (HIF-1) (reviewed by Bunn and Poyton, 1996; Ratcliffe, 1998; Huang et al. 1998). Perhaps the best-characterized hemoprotein oxygen sensor is FixL in nitrogen-fixing bacteria, Rhizobium sp.; the spin state of Fe in the heme moiety of FixL regulates its kinase activity, which controls the activity of a transcriptional component, FixJ (Gilles-Gonzalez et al. 1991, 1994; Gilles-Gonzalez and Gonzalez, 1993; reviewed by Bunn and Poyton, 1996). It is not known whether any of the oxygen-responsive transcription factors in yeast are differentially phosphorylated in response to oxygen or whether hemoproteins per se are involved in oxygen-sensing pathways in yeast.

Recently, we completed a number of experiments addressing the role of heme and hemoproteins in regulating the transcription of oxygen-responsive genes in yeast. Previous experiments examining the transcription of these genes have been performed primarily using cells grown either aerobically or anaerobically, or with hem1 mutants in the presence or absence of δ-aminolevulinate, which bypasses the hem1 defect. While these approaches have identified genes that are oxygen- and/or heme-sensitive and have helped to define many of the trans-acting factors and cis-sites that control the expression of these genes, they do not address how oxygen is sensed or the functional role of heme in regulating their transcription. For example, one question raised by these studies is whether these genes respond in a graded fashion to oxygen concentration or in an all-or-none fashion to the presence or absence of oxygen. Moreover, measurement of their transcript levels as a function of oxygen concentration could provide insight into the functional role of heme in controlling their expression. For example, if cellular concentrations of heme were controlling the expression of both the aerobic and hypoxic genes, we would predict that their transcript levels would not vary with declining oxygen level until its concentration approached the K_m of coproporphyrinogen III oxidase. Below this oxygen concentration, transcript levels of the aerobic genes would decline, while those of the hypoxic genes would increase.

Fig. 2 illustrates the effect of oxygen concentration on transcript levels of an aerobic gene (Fig. 2A), an aerobic and hypoxic gene pair (Fig. 2B), and a hypoxic gene (Fig. 2C). Fig. 2A shows the relative mRNA levels of the aerobic gene COX4 in cells grown at different oxygen concentrations (Burke et al. 1997). COX4 transcript levels decline gradually between 200 and 1 μmol l⁻¹ O₂, and then decline sharply below this concentration. Transcripts of a number of other subunits of cytochrome c oxidase, including COX6, COX7, COX8 and COX9, show a similar trend with respect to oxygen concentration (Burke et al. 1997). Under nominally oxygen-free conditions (anoxia), mRNAs of all of these genes are detectable, varying from 8 to 40% of their normoxic levels (Burke et al. 1997).

Fig. 2B shows the dose–response curve of transcripts of the aerobic and hypoxic gene pair COX5a/COX5b as a function of oxygen concentration. Unlike the other aerobic COX genes examined, COX5a transcript levels (solid line) vary little as a function of oxygen concentration between 200 and 10 μmol l⁻¹ O₂, but decline sharply below approximately 5–1 μmol l⁻¹ O₂ (Burke et al. 1997; K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). In contrast, transcript levels of the hypoxic gene COX5b (dashed line) are undetectable, or nearly so, between 200 and 0.25 μmol l⁻¹ O₂, and then they increase sharply below this concentration. Similar oxygen-dependent transcript profiles were found for the aerobic and hypoxic gene pairs CYC1/CYC7, TIF51A/ANB1 and AAC2/AAC3 (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). In all cases, a sharp break in their transcript levels occurs between approximately 1 and 0.5 μmol l⁻¹ O₂, with the aerobic isoform being optimally expressed above this concentration and the hypoxic isoform expressed below it.

Fig. 2C shows the dose–response curve of the hypoxic gene CPR1 (NCP1). Unlike the hypoxic genes that encode isoforms, CPR1 transcripts are detectable between 200 and 0.5 μmol l⁻¹ O₂ (at approximately 30% of their anoxic levels), but their levels also increase sharply below 0.5 μmol l⁻¹ O₂. Similar transcript profiles were found for the hypoxic genes HEM13, HMG1, HMG2, ERG11 and OLE1; transcript levels of all of these genes also show sharp changes, typically below 1 μmol l⁻¹ O₂ (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation).

Another way to investigate how oxygen affects the transcription of these genes is to shift cells from anoxic to aerobic conditions and vice versa. We used such shift experiments to examine the kinetics of induction of aerobic genes and decline in transcript levels of hypoxic genes (Fig. 3) as well as the kinetics of induction of hypoxic genes and decline in transcript levels of aerobic genes (Fig. 4). The large change in oxygen concentration that occurs during the first 10 min after the shift is shown in the insets in Figs 3A and 4A. Under the experimental conditions used, the mass-doubling time of S. cerevisiae strain JM43 was approximately 4 h under anoxic conditions and approximately 2.4 h under aerobic conditions.

Fig. 3 shows the effect of shifting cells from steady-state anoxic conditions to aerobic conditions on transcript levels of COX4 (Fig. 3A), COX5a and COX5b (Fig. 3B) and the hypoxic genes CPR1, CYC7 and HEM13 (Fig. 3C) (Burke et al. 1997; K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). In general, transcript levels of aerobic genes, as illustrated by COX4, respond rapidly to reoxygenation and then increase slowly, reaching half of their aerobic steady-state levels (t_1/2) typically in less than 1 h (Burke et al. 1997). Transcript levels of aerobic isoforms, such as COX5a (Fig. 3B, solid line), typically increase more rapidly, with t_1/2 values ranging from 5 min (TIF51A) to 45 min (AAC2) (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). Finally, transcript levels of most hypoxic genes, as illustrated by COX5b (dashed line, Fig. 3B) and HEM13 (dashed line) and CPR1 (solid line) in Fig. 3C, decline rapidly, with t_1/2 decay values ranging from 5 min (HEM13, COX5b, AAC3) to 70 min (OLE1) (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). Currently, it is not known whether the decline in transcript levels of these hypoxic genes is due to the effect of oxygen on transcript synthesis, transcript stability or both. Surprisingly, upon reoxygenation, CYC7 transcript levels (dash-dotted line, Fig. 3C) increase, reaching nearly five times their levels under anoxia, before declining (Burke et al. 1997; K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation).

Fig. 4 shows the effect of shifting cells from steady-state aerobic conditions to anoxic conditions on transcript levels of these same genes. In general, transcript levels of aerobic genes (e.g. COX4, Fig. 4A), as well as the aerobic isoforms (e.g. COX5a, solid line in Fig. 4B), decline rapidly, with t_1/2 decay values ranging from 5 to 60 min (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation). In contrast to the comparatively rapid induction of aerobic genes (Fig. 3), there is a remarkable variation in both the time course for induction of the hypoxic genes as well as their transcript profiles (Fig. 4B,C). Whereas some hypoxic genes are induced fairly rapidly (e.g. HEM13, dashed line in Fig. 4C), others are induced much more slowly, not reaching fully anoxic levels until nearly 20 h (e.g. CYC7, dashed-dotted line in Fig. 4C). Moreover, many of the hypoxic genes exhibit large overshoots (e.g. HEM13 and CPR1, solid line in Fig. 4C) above their steady-state anoxic levels, while others simply plateau (e.g. COX5b, dashed line in Fig. 4B, and CYC7) (K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation).

Taken together, these findings provide insight concerning a number of aspects of the transcriptional regulation of these genes. First, transcript levels of both the aerobic and hypoxic genes are determined by the concentration of oxygen, not merely by its presence or absence. Second, compared with the aerobic genes, the hypoxic genes are regulated much more tightly in terms of the oxygen concentrations over which they are optimally expressed. Third, the induction of the aerobic genes is much more rapid than that of the hypoxic genes. Fourth, there is a marked variation in both the time course for induction and transcript profiles of the hypoxic genes. And fifth, for nearly all of these oxygen-responsive genes, a sharp break occurs in their transcript levels at approximately 1 μmol l⁻¹ O₂, with transcript levels of the aerobic genes declining sharply and those of most hypoxic genes increasing sharply below this concentration. It is interesting that this sharp change occurs near the estimated K_m for O₂ of coproporphyrinogen III oxidase (approximately 0.1 μmol l⁻¹ O₂). Although we have not directly measured levels of free heme at these different oxygen concentrations, these data are consistent with models that invoke regulation by cellular concentrations of heme. However, the gradual decline in mRNA levels of a number of aerobic genes – and the increase in mRNA levels of some of the hypoxic genes (see K. E. Kwast, P. V. Burke and R. O. Poyton, in preparation) – over a range of oxygen concentration (1–200 μmol l⁻¹) in which heme levels are thought not to vary (Labbe-Bois and Labbe, 1990) is difficult to reconcile with such a model. Instead, these data suggest that other regulatory mechanisms (e.g. the redox state of heme, hemoproteins or other metalloproteins) are controlling the expression of these genes at higher oxygen concentrations.

One way to test the hypothesis that the redox state of hemoprotein(s) is involved in regulating the transcription of oxygen-responsive genes is to clamp the redox state under different conditions of oxygen availability. An experimental approach that has been widely used to study the involvement of hemoproteins in oxygen-sensing pathways in higher eukaryotes is to use carbon monoxide (CO). CO is an inert gas that has a remarkably high specificity for ferrous (FeII) heme groups in hemoglobin, myoglobin, certain cytochromes and other hemoproteins that bind oxygen reversibly. Under hypoxic conditions, CO markedly reduces the transcription of a number of hypoxically induced genes in mammalian cells (Goldberg et al. 1988; Kietzmann et al. 1993; Goldberg and Schneider, 1994; reviewed by Bunn and Poyton, 1996) and inhibits the expression of several proteins in turtle hepatocytes (reviewed by Hochachka et al. 1996). Although the precise target of CO in these cells has not been determined, these results are consistent with CO binding to a ferrous hemoprotein oxygen sensor and preventing a redox change in the heme moiety that is required for the induction of these genes at low oxygen tension (Goldberg et al. 1988; reviewed by Bunn and Poyton, 1996). In mammalian cells, this hemoprotein oxygen sensor is proposed to function as an oxidase that reduces oxygen to superoxide, which acts as a second messenger.

To determine whether the redox state of hemoprotein(s) may also control the expression of oxygen-responsive genes in yeast, we recently examined the effects of CO on the induction of a number of hypoxic genes (COX5b, CYC7, ANB1, AAC3, HEM13, HMG1, HMG2, ERG11, CPR1 and OLE1) and on the decline in transcription of several aerobic genes (COX5a, COX6, CYC1, TIF51A, AAC2 and ROX1) after shifting cells from aerobic to anoxic conditions (K. E. Kwast, P. V. Burke, B. Staahl, S. Fontaine and R. O. Poyton, in preparation). This set includes genes whose transcription is regulated by Hap1p, Hap2/3/4/5p and/or Rox1p. Interestingly, CO affected only a small subset of the hypoxic genes: CO totally blocked the anoxic induction of CYC7 and OLE1 and partially blocked the induction of COX5b (K. E. Kwast, P. V. Burke, B. Staahl, S. Fontaine and R. O. Poyton, in preparation). The transcription of the other hypoxic genes and of all the aerobic genes examined was not affected appreciably by CO. These results define two classes of hypoxic genes: CO-sensitive and CO-insensitive. They also suggest that CO is not acting through any of the known heme-activated transcription factors or through Rox1p. Rather, these findings point to additional pathways (and probably trans-acting factors) that mediate the expression of these hypoxic genes.

Another experimental approach that has been widely used to address the involvement of hemoproteins in oxygen-sensing pathways is to examine whether transition metals induce the transcription of hypoxic genes under normoxic conditions. When added to cells, transition metals (e.g. Co, Ni) are thought to be incorporated into hemoproteins in place of Fe. Indeed, it has been shown that these metals can serve as substrates for ferrochelatase, and Co has been shown to be incorporated into heme both in vivo (Sinclair et al. 1979) and in cultured cells (Sinclair et al. 1982). Unlike Fe, these metals either cannot bind O₂ (Ni) or have an exceedingly low affinity for O₂ (Co). In mammalian cells, these metals typically induce the same set of hypoxic genes whose expression is blocked by CO under hypoxic conditions (reviewed by Bunn and Poyton, 1996).

In yeast, we find that Co and Ni induce the expression specifically of the CO-sensitive genes OLE1 and, in some experimental conditions, CYC7 under normoxic conditions in a concentration-dependent manner (K. E. Kwast, P. V. Burke, B. Staahl, S. Fontaine and R. O. Poyton, in preparation). Our findings with both CO and transition metals suggest that these hypoxic genes are induced via a pathway involving an oxidoreduction state change in a hemoprotein oxygen sensor. Although we have not definitively identified the hemoprotein involved in this oxygen-sensing pathway, a large body of evidence suggests that the mitochondrial respiratory chain, probably cytochrome c oxidase, is involved in this signaling pathway (K. E. Kwast, P. V. Burke, B. Staahl, S. Fontaine and R. O. Poyton, in preparation); the anoxic induction of OLE1 and CYC7 is specifically blocked by inhibiting electron transport either with metabolic poisons (e.g. CN⁻, antimycin A) or with mutations in the respiratory chain (cox⁻ or cob⁻ strains). The nature of the cross-talk between the mitochondrion and nucleus that is required for the transduction of this signal is being studied.

Overall, our experiments examining oxygen thresholds, the induction/decline in transcript levels following a shift in oxygen concentration and the effects of CO and transition metals on the transcription of oxygen-responsive genes suggest that there are multiple pathways and mechanisms involved in regulating the oxygen-dependent expression of genes in yeast. With regard to the functional roles of heme, we have collected indirect evidence that cellular concentrations of heme regulate the transcription of these genes, which is probably mediated by the Hap proteins. However, the oxygen-dependent change in transcript levels of a subset of both the aerobic and hypoxic genes over a range of oxygen concentration in which heme concentration is thought not to vary suggests that other regulatory mechanisms are involved, possibly ones involving redox changes. Finally, of all of the oxygen-responsive genes examined, only a small subset of the Rox1p-dependent, hypoxic genes was affected by CO and transition metals. This suggests that these effects are exerted through pathways that are independent of Rox1p and Hap1p regulation. Moreover, it is interesting that the effect of these treatments apparently overrides the regulation by Rox1p on the transcription of these genes, given that ROX1 transcript levels are not affected by these treatments (K. E. Kwast, P. V. Burke, B. Staahl, S. Fontaine and R. O. Poyton, in preparation). Taken together, these findings indicate that there are multiple regulatory and oxygen-sensing pathways involved in modulating the expression of oxygen-responsive genes in yeast.

During the past decade, considerable progress has been made in our understanding of oxygen-sensing pathways in both prokaryotes and eukaryotes. These pathways share intriguing similarities, with hemoproteins playing a central role as proximal oxygen sensors. In yeast, the role of heme in oxygen-sensing pathways has been proposed to be somewhat different from that in higher eukaryotes; rather than being a redox-sensitive prosthetic group of a sensor, heme has been viewed primarily as a redox-insensitive, metabolic cofactor required for the function of transcription factors. However, our studies suggest that heme may act in both capacities in yeast.

This review provides testimony to the pivotal role that heme plays in the oxygen-dependent transcription of genes. Given that cellular levels of heme appear to control the transcription of a large set of both aerobic and hypoxic genes, it is possible that coproporphyrinogen III oxidase acts, in many respects, as an oxygen sensor. Our studies of the transcription of these genes as a function of oxygen concentration lend support to the view that heme may act as a redox-insensitive cofactor for transcription factors but, at the same time, the results suggest that additional regulatory mechanisms – possibly involving redox-sensitive sensors – are involved at higher oxygen concentrations. Furthermore, work with CO and transition metals indicates that multiple pathways are involved in the regulation of a subset of the hypoxic genes and points to the involvement of redox-sensitive hemoprotein(s). Indeed, these studies suggest that oxygen-sensing mechanisms involving hemoproteins may be universal in prokaryotes and both lower and higher eukaryotes. Further studies with yeast should define these pathways and identify oxygen sensor(s) in this simple eukaryotic organism and may provide insight into an area of research that has presented a considerable challenge to investigators working with higher eukaryotic cells.

We thank the Company of Biologists for the opportunity to present our work and for bringing together a diverse array of researchers studying many aspects of oxygen metabolism, as shown by the breadth of work in this volume. In addition, we thank the community of yeast researchers for providing preprints and reprints of recent work. The preparation of this article was supported, in part, by an American Heart Association Fellowship to K.E.K., and National Institutes of Health Grants GM-30228 and GM-39324 and Tobacco Research Council Grant 4557 to R.O.P.