Inhibition of Hypoxia-inducible Factor 1 Activation by Carbon Monoxide and Nitric Oxide

It has been proposed that cells sense hypoxia by a heme protein, which transmits a signal that activates the heterodimeric transcription factor hypoxia-inducible factor 1 (HIF-1), thereby inducing a number of physiologically relevant genes such as erythropoietin (Epo). We have investigated the mechanism by which two heme-binding ligands, carbon monoxide and nitric oxide, affect oxygen sensing and signaling. Two concentrations of CO (10 and 80%) suppressed the activation of HIF-1 and induction of Epo mRNA by hypoxia in a dose-dependent manner. In contrast, CO had no effect on the induction of HIF-1 activity and Epo expression by either cobalt chloride or the iron chelator desferrioxamine. The affinity of CO for the putative sensor was much lower than that of oxygen (Haldane coefficient, ∼0.5). Parallel experiments were done with 100 μm sodium nitroprusside, a nitric oxide donor. Both NO and CO inhibited HIF-1 DNA binding by abrogating hypoxia-induced accumulation of HIF-1α protein. Moreover, both NO and CO specifically targeted the internal oxygen-dependent degradation domain of HIF-1α, and also repressed the C-terminal transactivation domain of HIF-1α. Thus, NO and CO act proximally, presumably as heme ligands binding to the oxygen sensor, whereas desferrioxamine and perhaps cobalt appear to act at a site downstream.

It has been proposed that cells sense hypoxia by a heme protein, which transmits a signal that activates the heterodimeric transcription factor hypoxia-inducible factor 1 (HIF-1), thereby inducing a number of physiologically relevant genes such as erythropoietin (Epo). We have investigated the mechanism by which two heme-binding ligands, carbon monoxide and nitric oxide, affect oxygen sensing and signaling. Two concentrations of CO (10 and 80%) suppressed the activation of HIF-1 and induction of Epo mRNA by hypoxia in a dosedependent manner. In contrast, CO had no effect on the induction of HIF-1 activity and Epo expression by either cobalt chloride or the iron chelator desferrioxamine. The affinity of CO for the putative sensor was much lower than that of oxygen (Haldane coefficient, ϳ0.5). Parallel experiments were done with 100 M sodium nitroprusside, a nitric oxide donor. Both NO and CO inhibited HIF-1 DNA binding by abrogating hypoxiainduced accumulation of HIF-1␣ protein. Moreover, both NO and CO specifically targeted the internal oxygendependent degradation domain of HIF-1␣, and also repressed the C-terminal transactivation domain of HIF-1␣. Thus, NO and CO act proximally, presumably as heme ligands binding to the oxygen sensor, whereas desferrioxamine and perhaps cobalt appear to act at a site downstream.
Adaptation to hypoxia is of fundamental importance in developmental, physiological, and pathophysiological processes (1,2). Humans and other mammals respond to low oxygen tension in part by fine tuning the expression of a group of physiologically relevant genes. Erythropoietin (3,4) and tyrosine hydroxylase (5) act in concert to raise blood oxygen levels by enhancing erythropoiesis and ventilation, respectively. Hypoxic induction of genes encoding vascular endothelial growth factor (6) and inducible nitric oxide synthase (7,8) leads to increased angiogenesis and vasodilation. Up-regulation of genes encoding glucose transporters and specific glycolytic isoenzymes (9 -11) maximize ATP production in the setting of reduced oxygen supply. Remarkably, the hypoxic induction of all of these diverse genes appears to depend on a common mode of oxygen sensing and signal transduction, triggering the activation of a critical transcription factor, hypoxia-inducible factor 1 (HIF-1) 1 (for review, see Ref. 1).
The activation of HIF-1 by hypoxia depends on a sensing and signaling process that is poorly understood. There is strong, albeit circumstantial, evidence that the oxygen sensor is a heme protein (1,27). Support for this claim derives largely from experiments demonstrating that heme binding ligands suppress the hypoxic induction of HIF-1 activation and expression of the genes mentioned above. The hypoxic induction of erythropoietin (27), vascular endothelial growth factor (28,29), and other genes (30) was markedly blunted in the presence of carbon monoxide (CO), a molecule whose only established biological function is binding to ferrous atoms on heme groups. Subquently, another heme ligand, nitric oxide (NO), was shown to exert a similar effect (29,31,32). In addition to hypoxia, HIF-1 can be activated by the transition metal Co 2ϩ as well as by iron chelation (33). The mechanisms underlying the activities of these agonists are unknown, although there is circumstantial evidence that they affect the level of reactive oxygen species (34), which may serve as signaling molecules (1). To gain insight into hypoxic sensing and signaling, we have made an in-depth comparison of the downstream effects of CO and NO on HIF-1 activation, HIF-1␣ stability, and the expression of oxygen-responsive endogenous and reporter genes.
RNA Preparation and Analysis-Total RNA was extracted by the TRIzol method (Life Technologies, Inc.) according to the manufacturer's protocol and dissolved in 100% formamide. A plasmid, containing a portion of exon 5 of the erythropoietin (Epo) gene (bp 2650 -2890) inserted into pBluescript, was linearized with XhoI and in vitro transcribed in the presence of [ 32 P]CTP (DuPont NEN) with T7 polymerase (Promega). RNase protection assays were performed with 40 -100 g of RNA hybridized at 55°C overnight with the Epo riboprobe. For quantification of gels, a PhosphorImager was used with ImageQuant software (Molecular Dynamics, Sunnyvale, CA). Results are presented as the mean Ϯ S.E. from independent experiments.
Transient Transfection-Human embryonic kidney 293 cells and human hepatoma Hep3B cells were transfected in suspension using the calcium phosphate method as described elsewhere (19). In general, 293 cells were transfected with 2 g of each expression vector plus 0.25 g of pCMV-␤gal, whereas Hep3B cells were transfected with 2 g of pCMV-␤gal and 2 g of reporter plasmid. Cells were treated with 100 M sodium nitroprusside as indicated. After 24 h of incubation, whole cell extracts were prepared as described previously (19) for electrophoretic mobility shift assays and Western blotting. For reporter gene assays, cells were harvested 48 h after transfection and assayed for luciferase activity as described elsewhere (19).
Calculation of the Haldane Coefficient for Oxygen versus CO Binding-To calculate the fractional saturation of the oxygen sensor with ligand (O 2 versus CO), we used a comprehensive and accurate set of earlier experimental data from our laboratory (37) on the relationship between pO 2 and level of Epo mRNA. These data, in combination with experimental quantitation of Epo mRNA, enabled the estimation of pO s values, the pO 2 simulated by O 2 ϩ CO ligation. Because the relevant experiments were carried out in 1% O 2 (ϭ7 torr), the Haldane coefficient M was calculated from the equation: M ϭ (pO s Ϫ 7)/CO torr .

Effect of Carbon Monoxide and Nitric Oxide on the Induction of Epo mRNA-Of all the genes up-regulated by hypoxia, Epo
shows the most robust response. Therefore we began by examining the effect of CO and NO on the induction of Epo mRNA in Hep3B cells. Epo mRNA, protected by a 240-bp riboprobe during ribonuclease protection assays, displayed a doublet banding pattern (Fig. 1, A and B). As expected, hypoxia, cobalt, and desferrioxamine markedly induced Epo mRNA expression. CO suppressed hypoxic induction of Epo mRNA, almost to the level of normoxic controls (Fig. 1A, lanes 3 and 4). In determining the Haldane coefficient for the oxygen sensor, we assumed that CO and O 2 bind to the same site, similar to other heme proteins. At 10 and 80% CO, the calculated Haldane coefficients were 0.8 and 0.2, respectively. Interpretation of these values may be confounded by the possibility that, like cytochrome c oxidase, CO is likely to bind reversibly to the heme group of the oxygen sensor, whereas oxygen binding may be irreversible if it undergoes reduction to superoxide (1,38). However, the coefficients that we obtained were reproducible and probably valid, owing to the high flow of gas used in the incubations and the likelihood that the pO 2 was maintained at a constant level. Cobalt

FIG. 1. Effect of CO and NO on induction of Epo mRNA expression in Hep 3B cells.
A, cells were exposed to 20% O 2 (N) or 1% O 2 (H) and incubated in either control medium (lanes 1-4) or in media containing 100 M cobalt (lanes 5-8) or 130 M desferrioxamine (lanes 9 -11). Aliquots of cells were exposed to 10% carbon monoxide (lanes 3, 7, and 11-13), and 80% CO (lanes 4 and 8). Ribonuclease protection assays were performed using radiolabeled wild-type Epo riboprobe (250 bp of the fifth exon, which includes the 3Ј untranslated region). The bar graph below shows densitometric quantitation of replicate experiments: means Ϯ S.E. of with the number of experiments indicated above each bar. B, experiments similar to A tested the effect of 100 M SNP on induction of Epo mRNA by hypoxia and cobalt. and hypoxia had an additive effect on Epo mRNA levels, whereas desferrioxamine and hypoxia did not. Combined cobalt and hypoxia induction was decreased by 10 and 80% CO treatment, to the levels of normoxic cobalt induction. Cobalt induction of Epo mRNA under normoxic conditions was not suppressed by 10% CO (Fig. 1A, lane 14). Likewise, CO had no effect on desferrioxamine-induced Epo mRNA levels (Fig. 1A,  lane 11). These results indicate that CO does not suppress Epo mRNA expression in a nonspecific toxic manner.
Parallel experiments were done with 100 M SNP as a source of NO. As shown in Fig. 1B, lane 3, this concentration of SNP mimicked CO in suppressing the induction of Epo mRNA by hypoxia. However, in contrast to CO, SNP suppressed induction of Epo mRNA by cobalt (Fig. 2, lanes 5, 7, and 9). These results with SNP are consistent with those of Sogawa et al. (32), who tested a hypoxia-inducible reporter gene with SNP and two other NO donors.
Effect of CO and NO on the Activation of HIF-1-To delineate the mechanism by which CO and NO suppressed Epo mRNA expression, we examined how these agents affected the activation of HIF-1, the transcription factor responsible for the regulation of Epo and a number of other genes induced by hypoxia. Extracts prepared from hypoxic cells, when incubated with radiolabeled wild-type HIF-1 binding sequence, displayed doublet bands in gel shift assays. In contrast, single bands were observed in extracts prepared from cells treated with cobalt (Fig. 2, lane 2) or desferrioxamine. Similar results have been reported by others (12,33,39). Hypoxia-induced HIF-1 binding was attenuated by treatment with CO in a dose-dependent manner (Fig. 2, lanes 5 and 6). In contrast, cobalt and desferrioxamine induction of HIF-1 binding was not affected by CO treatment (Fig. 2, lanes 7 and 8).
Similar to what was observed with CO, 100 M SNP significantly suppressed hypoxia-induced HIF-1 DNA binding in Hep3B cells, whereas neither constitutive nor nonspecific binding was affected (Fig. 2, lane 14). To investigate the mechanism underlying the inhibitory effects of CO and SNP, 293 cells were transfected with the CMV-driven vectors overexpressing HIF-1 (HA-tagged HIF-1␣ and ARNT, respectively). Both CO and NO abrogated DNA binding of overexpressed HIF-1 (Fig. 2, lanes  11 and 17). It is noteworthy that under normoxia, despite exogenous overexpression of HIF-1, no HIF-1 binding activity was detected (Fig. 2, lanes 9 and 15). Together these results indicate that both endogenous and exogenous HIF-1 are regulated by the same signaling pathway.
Because HIF-1 activity is primarily determined by the abun-  3 and 4; C, lanes 4 -6), or with pGal4-ODD (C, lanes 7-9). Twenty-four hours after transfection, the cells were then subjected for 4 h to normoxia (N) or hypoxia (H) in the absence or presence of 80% CO (B) or 100 M SNP (C). Cells extracts were then prepared and analyzed with EMSAs using 32 P-labeled oligonucleotides containing an HIF-1 binding site (B, lanes 1-4; C, lanes 1-6) or Gal4 binding site (C, lanes 7-9). The endogenous HIF-1 complexes are marked with an arrowhead, whereas those composed of HIF-1␣(401⌬603) and ARNT are marked with asterisks, and Gal4 binding is marked with an arrow. PAS, periodic aryl hydrocarbon receptor-simultaneous. dance of HIF-1␣ protein (19), we examined by Western blot whether CO and NO affect HIF-1␣ accumulation in hypoxic cells. As shown in Fig. 2, bottom, exposure of hypoxic cells to 10 and 80% CO resulted in a dose-dependent reduction in the abundance of HIF-1␣ protein (Fig. 2, lanes 5 and 6), comparable with the decrease in HIF-1 DNA binding on EMSA. Importantly, CO had no effect on the abundance of HIF-1␣ induced by cobalt (Fig. 2, lanes 7 and 8), again indicating that CO was not exerting a nonspecific toxic effect. In a like manner, we examined whether NO affects HIF-1␣ accumulation in hypoxic cells. Consistent with the results obtained with CO, SNP blocked hypoxia-induced accumulation of endogenous HIF-1␣ (Fig. 2,  lane 14). Moreover, both CO and NO suppressed accumulation of transfected HIF-1␣ (Fig. 2, lanes 11 and 17). In all these results, the level of HIF-1␣ closely paralleled that of HIF-1 binding activity.
The Oxygen-dependent Degradation Domain Is Responsible for Inhibition by CO and NO-We recently identified an ODD domain within HIF-1␣ that plays a regulatory role for hypoxiainduced stabilization of HIF-1␣. Internal removal of this domain rendered HIF-1␣ stable irrespective of oxygen tension (25). The inhibitory effect of CO and SNP on the accumulation of HIF-1␣ protein prompted us to examine whether the ODD domain is targeted by treatment with these agents. As in the experiment shown in Fig. 2, lanes 9 -11 and 15-17, 293 cells were co-transfected with a vector expressing ARNT and another vector expressing wild-type HIF-1␣ or the ODD-deleted HIF-1␣ (401⌬603; Fig. 3A). Consistent with previous results (25), normoxic cell extracts prepared from cells transfected with HIF-1␣(401⌬603) gave rise to strong constitutive HIF-1 binding (Fig. 3C, lane 4). In contrast to wild-type HIF-1␣, HIF-1␣(401⌬603) was resistant to treatment with either CO (Fig. 3B, lane 4) or SNP (Fig. 3C, lane 6). These results indicate that the ODD domain is required for the suppressive effects of CO and SNP.
Gal4 is normally a highly stable protein. However, when fused to the ODD domain, it undergoes oxygen-dependent degradation (25). To confirm that abrogation of HIF-1␣ abundance by SNP is specifically mediated by the ODD domain, a plasmid expressing Gal4-ODD was transfected into 293 cells and assayed for Gal4 DNA binding activity. Like wild-type HIF-1, Gal4-ODD exhibited very weak binding activity under normoxia (Fig. 3C, lane 7) and robust induction with hypoxia (Fig.  3C, lane 8). Of note, hypoxia-induced binding was markedly reduced by SNP treatment (Fig. 3C, lane 9).
In an effort to elucidate further the mechanism underlying the effect of nitric oxide on the ODD domain, we tested the possibility that NO could be forming an SNO adduct with a cysteine sulfhydryl group on the ODD. Accordingly, Cys-520, the only cysteine residue present in the ODD domain, was mutated to serine. As in experiments depicted in Figs. 2 and 3, HIF-1 binding activity was analyzed after cells were transfected with ARNT and either wild-type HIF-1␣ or the mutant HIF-1␣(C520S). Interestingly, the mutant HIF-1␣ was still sensitive to SNP treatment (Fig. 4A, lane 3), which is in agreement with the loss of HIF-1␣ protein accumulation demonstrated by Western blotting (Fig. 4B, top panel). As shown in Fig. 4B, bottom panel, probing of the same blot with anti-ARNT antibodies demonstrated that ARNT protein levels remained unchanged.
CO and NO Inactivate HIF-1␣ Transactivating Activity-Previously Sogawa et al. (32) showed that SNP specifically inhibited hypoxia-induced activity of a luciferase reporter containing four copies of the HIF-1 binding site. Because we found that expression of HIF-1␣(401⌬603) gave rise to stable HIF-1 DNA binding activity that was insensitive to CO and SNP treatment (Fig. 3, B and C), we wondered whether this stable HIF-1 activity could transactivate an HIF-1 target reporter gene in the presence of SNP. A luciferase reporter EpoE-luc was used to test the effect of SNP in Hep3B cells co-transfected with wild-type HIF-1 (HIF-1␣ and ARNT) or the ODD domaindeleted HIF-1 (HIF-1␣(401⌬603) and ARNT) as mentioned above. The reporter EpoE-luc contains a native HIF-1 binding site within a 50-bp region of the human erythropoietin 3Ј enhancer (19). In agreement with the results of Sogawa et al. (32), SNP drastically inhibited hypoxia-induced luciferase activity when wild-type HIF-1 was transfected (Fig. 5, left). Surprisingly, a similar result was obtained when HIF-1␣(401⌬603) was tested (Fig. 5, right). It is noteworthy that HIF-1␣ contains two transactivation domains; one is within the ODD domain, and the other is at the C terminus (23,24), in contrast to HIF-1␣(401⌬603), which has only the latter. Under normoxic conditions the transactivating activity of HIF-1␣(401⌬603) was substantially higher than that of the wild type, as observed previously (25). In addition, hypoxia further stimulated the reporter activity.
These results raised the possibility that, in addition to impacting on the protein stability of HIF-1␣, CO and NO may also affect its C-terminal activation domain. To address this question directly, we took advantage of a Gal4 fusion system in which the stable C-terminal transactivation domain was fused to the Gal4 DNA binding domain (Gal4-CAD, Fig. 6A) so that transactivating activity of the fusion protein could be assessed by a luciferase reporter containing five copies of the Gal4 binding site. Transactivating activity of the C-terminal portion of HIF-1␣ in a Gal4 fusion system can be enhanced by hypoxia without altering the protein abundance (23,24). To that end, Hep3B cells were co-transfected with a reporter (Gal4-luc) and a vector expressing Gal4 alone (Gal4) or Gal4 fusion (Gal4-CAD) and subjected to normoxic or hypoxic treatment in the presence or absence of either CO or SNP. As expected, Gal4 alone failed to stimulate reporter activity when challenged by hypoxia (Fig. 7, B and C). In contrast, the reporter activity was markedly increased under hypoxia when the Gal4 fusion protein was expressed, but this induction was suppressed by both CO (Fig. 7B) and SNP (Fig. 7C). Taken together, these data indicate that CO and NO exerted their inhibitory effects by targeting the C-terminal transactivating domain of HIF-1␣ as well as its protein stability.
NO Suppression Is Not Mediated by Guanylate Cyclase-Because the predominant physiological action of nitric oxide is mediated through an increase in cGMP by activation of soluble guanylate cyclase (40), we also examined whether cell membrane-permeable 8-bromo-cGMP, an analog of cGMP, simulates the effect of SNP. No loss of HIF-1 binding was detected with up to 1 mM 8-bromo-cGMP (Fig. 7, lane 3). This result makes it very unlikely that the inhibitory effect of SNP is mediated through a pathway involving soluble guanylate cyclase. DISCUSSION We have compared the effects of carbon monoxide and nitric oxide on the sensing of oxygen and the signaling pathway leading to HIF-1 activation. Although CO and NO both bind to ferrous iron in heme proteins, the two ligands differ in major respects. CO is chemically inert and undergoes no known chemical modifications in biological systems other than perhaps oxidation to CO 2 coupled with reduction of cytochrome c oxidase (41) and other heme proteins (42). In contrast to CO, the biochemistry of NO is much more complex. NO binds to and activates its primary biological target, soluble guanylate cyclase, much more readily than does CO (43). Unlike CO, NO can also bind to ferric heme. Moreover, unlike CO, NO is capable of reacting with oxygen and superoxide (44), possibly affecting the signaling pathway for HIF-1 activation. Finally, unlike CO, NO can form S-nitroso derivatives of a broad repertoire of membrane, cytosolic, and nuclear proteins (44).
Our experiments show that CO and NO have similar effects on oxygen sensing and signaling: they both suppress expression of target genes through inhibition of HIF-1 activation, and both trigger destabilization of HIF-1␣ via the ODD domain and inhibition of the C-terminal activation domain. The most parsimonious mechanism is binding of these ligands to the heme group of the oxygen sensor. Our experiments with CO, along with those of Liu et al. (29), are at odds with a recent report by Srinivas et al. (45), who concluded that CO was not effective in suppressing expression of a hypoxia-inducible reporter gene. They tested only a single concentration of CO, 6%, and in fact observed ϳ17% suppression, which was statistically significant. We show in this report that the oxygen sensor has a low affinity for CO. Thus, it is likely that Srinivas et al. (45) would have obtained a more robust effect had they used a higher concentration of CO. Our finding that CO abrogates hypoxic stabilization of HIF-1␣ protein, in a dose-dependent manner (Fig. 2D), is in conflict with a recent report by Liu et al. (29), who found that 5% CO suppressed HIF-1 activation but did not affect the level of HIF-1␣ protein. In contrast to their experiments, we used the same cell extract for both EMSA and HIF-1␣ Western blot analysis. Alternatively, the discrepancy between our results and those of Liu et al. (29) may be explained in part by their use of only a single low concentration of CO.
Heme proteins vary markedly in their relative affinities for CO versus O 2 . Human hemoglobin binds CO ϳ210 times more tightly than O 2 . In contrast, the yeast flavohemoglobin binds CO and O 2 with approximately equal affinity. 2 Likewise, b-type cytochromes, which have been proposed to function in oxygen sensing (1,38), are likely to have very low affinity for CO. We estimate that the oxygen sensor in Hep3B cells has a Haldane coefficient of ϳ0.5, a value significantly different from the coefficient of ϳ10 that Warburg (46) obtained for yeast cytochrome oxidase, which pertains to a wide range of higher organisms. This difference is difficult to reconcile with a recently proposed mitochondrial model of oxygen sensing (47,48). Despite the inherent inaccuracies in our measurement of a "functional" Haldane coefficient for oxygen sensing, the value we have obtained may prove useful in screening candidate genes suspected to encode the oxygen sensor. An agreement between the "functional" value that we have obtained and a coefficient obtained from spectra of overexpressed protein would lend support to a candidate, whereas a significant discrepancy would rule it out.
The low affinity of the oxygen sensor for CO may have adaptive significance. The primary toxicity of CO in higher organisms is attributable to its high affinity binding to hemoglobin, locking the tetramer in the "oxy" conformation and thereby increasing oxygen affinity and decreasing oxygen unloading, resulting in tissue hypoxia. The organism needs intact oxygen sensors to adapt to this hypoxic stress. These sensors would be unresponsive to CO-induced hypoxia if they had high affinity for CO.
Our conclusion that CO and NO are acting at the same site is supported by experiments that rule out other known effects of NO. The fact that a mutation of the single cysteine in the ODD domain did not affect the function of HIF-1 (Fig. 4) implies that the NO effect is not attributable to S-nitrosylation of HIF-1␣. Moreover, the fact that the C520S mutation has no effect on the activation of HIF-1 and its suppression by NO argues that SH-dependent redox chemistry is not involved in oxidant-dependent degradation. This result contrasts strikingly with an analogous phenomenon, the iron-dependent degradation of iron regulatory protein 2, the protein that regulates either the translation or the stability of mRNAs encoding proteins critical in iron homeostasis. Iwai et al. (49) have recently shown that a cluster of conserved cysteine residues is required for the iron-dependent oxidative attack on the protein.
An alternate explanation for the suppressive effect of NO is via binding to its most prominant target, guanylate cyclase. In contrast to findings of Liu et al. (29) but similar to those reported by Sogawa et al. (32), we found no effect of 8-bromo-cGMP on HIF-1 activation (Fig. 5), making it unlikely that suppression of HIF-1 activation depends on up-regulation of guanylate cyclase.
It is possible that NO and CO act independently of one another with NO generating reactive oxygen species (ROS) directly, whereas CO could generate ROS indirectly through redox cycling of cytochrome c (46) and other electron acceptors 2 H. Zhu, J. Olsen, and A. Riggs, personal communicacation. (42), including the oxygen sensor. Cytochrome c participates in the oxidation of CO to CO 2 by reduction of its heme iron. Oxidation of cytochrome c back to the ferric state would reduce oxygen to superoxide, enabling the generation of ROS via the iron-catalyzed Fenton reaction.
Our work supports and extends the paradigm that HIF-1 activation correlates with the abundance of HIF-1␣ protein. In addition, both CO and NO suppressed hypoxic induction of the C-terminal activation domain, which is independent of HIF-1␣ stability. Our demonstration of the inhibitory effect of CO and NO on HIF1␣ stability and transactivation supports the notion of two distinct pathways that lead to inhibition of HIF-1 activity. The conjoint destabilization of HIF-1␣ and inhibition of C-terminal transactivation of HIF-1␣ are both distal events in the signaling pathway, triggered by CO and NO, which act appear to act proximally, very likely at the level of the oxygen sensor. The only difference between the two ligands that we could find was that the induction of Epo mRNA by cobalt was suppressed by NO but not by CO (Fig. 1). This result is consistent with our hypothesis that this transition metal can substitute for iron in the oxygen sensor and mimic deoxy heme (27). Carbon monoxide cannot bind to cobalt-substituted heme, whereas NO binds to cobalt heme with high affinity (50). However, the action of cobalt may be independent of the oxygen sensor, as others have suggested (48,51). In contrast to cobalt, the activation of HIF-1 by the iron chelator desferrioxamine is unlikely to directly affect oxygen sensing, consistent with the inability of CO to suppress induction by this agent. The most plausible explanation of the effect of desferrioxamine is that free iron functions as a Fenton reagent, catalytically increasing levels of ROS. Therefore iron chelation will result in a lower level of ROS, which appears to be critical in signaling the activation of HIF-1 (1).