TY - JOUR
AU - Nikinmaa, Mikko
AB - Abstract Metazoans rely on aerobic energy production, which requires an adequate oxygen supply. During reduced oxygen supply (hypoxia), the most profound changes in gene expression are mediated by transcription factors known as hypoxia-inducible factors (HIFs). HIF alpha proteins are commonly posttranslationally regulated by prolyl-4-hydroxylase (PHD) enzymes, which are direct “sensors” of cellular oxygen levels. We examined the molecular evolution of the metazoan PHD–HIF oxygen-sensing system by constructing complete phylogenies for PHD and HIF alpha genes and used computational tools to characterize the molecular changes underlying the functional divergence of PHD and HIF alpha duplicates. The presence of PHDs in metazoan genomes predates the emergence of HIF alphas. Our analysis revealed an unexpected diversity of PHD genes and HIF alpha sequence characteristics in invertebrates, suggesting that the simple oxygen-sensing systems of Caenorhabditis and Drosophila may not be typical of other invertebrate bilaterians. We studied the early vertebrate evolution of the system by sequencing these genes in early-diverging cartilaginous fishes, elasmobranchs. Cartilaginous fishes appear to have three paralogs of both PHD and HIF alpha. The novel sequences were used as outgroups for a detailed molecular analysis of PHD and HIF alpha duplicates in a major air-breathing vertebrate lineage, the mammals, and a major water-breathing vertebrate lineage, the teleosts. In PHDs, functionally divergent amino acid sites were detected near the HIF alpha-binding channel and beta2beta3 loop that defines its substrate specificity. In HIF alphas, more functional divergence was found in teleosts than in mammals, especially in the HIF-1 alpha PAS domain and HIF-2 alpha oxygen-dependent degradation (ODD) domains, which interact with PHDs. Overall, in the vertebrates, elevated substitution rates in the HIF-2 alpha N-terminal ODD domain, together with a functional divergence associated with the known differences in PHD2 versus PHD1/3 substrate specificity, have contributed to the tighter oxygen-sensitive regulation of HIF-1 alpha than that of HIF-2 alpha. oxygen, hypoxia, transcription factor, enzyme, vertebrates, functional divergence Introduction Oxygen has played a pivotal role in the physiological adaptation of metazoans to a broad range of terrestrial and aquatic environments. Metazoans rely on aerobic energy production, where oxygen is used as the final electron acceptor in the mitochondrion, and so an adequate oxygen supply is necessary for cellular homeostasis. Cellular oxygen levels are sensed via prolyl-4-hydroxylase (PHD) enzymes that are direct “sensors” of cellular oxygen partial pressure (Kaelin and Ratcliffe 2008; Lendahl et al. 2009; Wenger et al. 2009). These enzymes use oxygen as a substrate for hydroxylating proline residues in a transcription factor named hypoxia-inducible factor alpha (HIF alpha). In conditions of reduced oxygen supply (hypoxia), the most profound responses in gene expression are mediated by this transcription factor. Together, these two proteins—the enzyme and the transcription factor—constitute a central physiological regulatory mechanism that is conserved in vertebrates. HIF alphas belong to the bHLH–PAS (basic Helix-Loop-Helix–Per-ARNT-Sim) family of transcription factors, which are involved in the regulation of environmentally induced and developmental gene expression (Kewley et al. 2004). In mammals, HIF controls gene expression involved in a range of processes including angiogenesis, erythropoiesis, glucose and iron transport, glycolysis, and cell-cycle control (Kaelin and Ratcliffe 2008). HIF consists of two subunits, ARNT (aryl hydrocarbon nuclear translocator or HIF beta) and HIF alpha, which confers hypoxia sensitivity to HIF. HIF alpha is tagged for degradation by PHDs in normoxia but is stabilized in hypoxia (Ivan et al. 2001; Jaakkola et al. 2001; Kaelin and Ratcliffe 2008). PHD enzymes belong to the family of iron(II) and 2-oxoglutarate–dependent dioxygenases that covalently modify two proline residues in the oxygen-dependent degradation (ODD) domain of HIF alpha subunits (Myllyharju 2009; Wenger et al. 2009). PHDs share very little sequence identity with the other types of vertebrate prolyl hydroxylases, collagen P4Hs and P4H-TMs (Koivunen et al. 2007). Two invertebrate model organisms, Caenorhabditis elegans and Drosophila melanogaster, are reported to possess only one HIF alpha and PHD, whereas known vertebrate genomes contain three functional duplicates of each: HIF1–3 alpha and PHD1 (EGLN2), PHD2 (EGLN1), and PHD3 (EGLN3) (Kaelin and Ratcliffe 2008). The central role of hypoxia in a number of human diseases has lead to extensive biomedical interest in the PHD–HIF oxygen-sensing system (Myllyharju 2009; Harten et al. 2010). Despite this increasing wealth of medical studies, there are no detailed evolutionary analyses of this regulatory mechanism. In order to place previous experimental results in a phylogenetic context, we set out to characterize the molecular evolution of the PHD–HIF system. First, we studied the emergence of the system by screening the genomes of early-diverging metazoans for HIF alpha and PHD homologs. We then examined the functional divergence of the HIF pathway in teleosts—the major water-breathing lineage of vertebrates—as compared with that of air-breathing mammals. Water-breathing animals face oxygen limitation more often than do air breathers, because the oxygen capacitance of water is only 1/30th that of air. We sequenced HIF alpha and PHD genes in early-branching elasmobranchs and used these sequences as outgroups in the computational analysis of mammalian and teleost gene duplicates. Our analyses clarify the early events in the evolution of the PHD–HIF oxygen-sensing system and identify patterns of molecular change underlying the functional divergence of PHD and HIF alpha genes in vertebrates. Materials and Methods Database Searches and Novel Elasmobranch Sequences In order to investigate the presence of HIF alpha and PHD genes in invertebrate metazoans, we searched for these genes in the available invertebrate genome sequences. For the following species, TBlastN searches were conducted specifically in the separate genome assemblies: yeast (Saccharomyces cerevisiae; http://www.yeastgenome.org/cgi-bin/blast-sgd.pl), choanoflagellate (Monosiga brevicollis) tablet animal (Trichoplax adhaerens), sea anemone (Nematostella vectensis), amphioxus (lancelet; Branchiostoma floridae; http://genome.jgi-psf.org/), hydra (Hydra magnipapillata; http://hydrazome.metazome.net/), purple urchin (Strongylocentrotus purpuratus; http://www.hgsc.bcm.tmc.edu), and the preliminary drafts of two early vertebrates lamprey (Petromyzon marinus; http://pre.ensembl.org/) and a cartilaginous fish, elephant shark (Callorhinchus milii; http://esharkgenome.imcb.a-star.edu.sg/). The hits from the genome searches were followed up in two ways: First, the hits were used in BlastP queries of the National Center for Biotechnology Information (NCBI) nonredundant (nr) protein database. A putative homolog was considered, for example, PHD if the best hit was PHD and was additionally verified with alignments and phylogenies (accession numbers provided in supplementary table. S1, Supplementary Material online). Second, if the expected/hypothesized hits were not found when using BlastP queries of the NCBI nr database or if the acquired protein-coding sequences appeared to be partial ones, we collected the sequences surrounding the primary BlastP hits in the genome contigs and repredicted the gene structure using three independent pieces of software: GENESCAN (http://genes.mit.edu/GENSCAN.html), Augustus (http://augustus.gobics.de/), and FGENESH (http://linux1.softberry.com). Reprediction was done on the BlastP hit regions as well as on at least 20 kb of surrounding regions on the contig, where available. For invertebrate species not mentioned above, protein sequences were directly collected from the NCBI nr protein database using BlastP searches as above. For vertebrates, nucleotide sequences were collected from Ensembl selecting the longest transcripts. This Ensembl collection was supplemented with Genbank sequences using BlastP queries (supplementary table. S1, Supplementary Material online). Epaulette sharks (Hemiscyllium ocellatum) were sampled, and tissues processed as previously described (Rytkönen et al. 2010). The smooth dogfish (Mustelus canis) heart tissues were a kind gift from Dr. John Eriksson (Åbo Akademi, Turku, Finland). Briefly, RNA was extracted from the tissues, cDNA produced, and a collection of degenerate (universal) primers used to obtain primary sequence fragments. Degenerate primers (supplementary table. S2, Supplementary Material online) based on alignments of tetrapod (human, mouse, chicken, and frog) and teleost (zebrafish, medaka, stickleback, and fugu) sequences were designed using netprimer (http://www.premierbiosoft.com/netprimer). “Touch-down” PCRs were conducted as previously described (Rytkönen et al. 2010), and the identities of the sequences were verified with Blast and with alignments. To obtain the complete coding sequences, primers for rapid amplifications of cDNA ends were designed (supplementary table. S2, Supplementary Material online), and PCRs were conducted according to the manufacturer’s instruction with the SMART kit (BD Biosciences, San Jose, CA). Phylogenies and Shared Synteny Multiple sequence alignments were built with MUSCLE (Edgar 2004) using the default parameters. The alignments were manually curated to identify and remove poorly aligned regions. This approach led us to choose the first 360 amino acids of HIF containing the bHLH–PAS domains for further analysis. For PHD, the N-terminal parts of PHD1 and PHD2 that are not present in PHD3 were excluded, and we analyzed the portion corresponding to the catalytic domain of C-terminal 244 amino acids. ProtTest 1.4 (Abascal et al. 2005) was used to obtain a substitution model that best fits the data (Jones Taylor Thornton 1[JTT] + G), and this model was used in PhyML (Guindon and Gascuel 2003) to obtain the maximum likelihood (ML) phylogeny with bootstrapping (100 replicates). The ML phylogenies were used in subsequent analyses. Sharing of the genomic gene order of the flanking genes and shared synteny were evaluated in human, zebrafish, and tetraodon by collecting the genes flanking HIF and PHD at their genomic loci in Ensembl. For HIF-3 alpha loci, no shared genomic gene order between human and teleosts was directly apparent, and locations of tetraodon HIF-3 alpha flanking genes were searched in the human genome. Rates of Molecular Evolution The average mammalian and teleost amino acid substitution rates were calculated using two different outgroups. Sequences from Amphioxus were used to calculate evolutionary rates between HIF and PHD paralogs. Next, we evaluated the rate shifts between mammals and teleosts within each vertebrate duplication using the epaulette shark sequences as outgroups. Pairwise amino acid distances from the outgroup were calculated in MEGA 3.1 (Kumar et al. 2004), and averages for mammals and teleost were used for calculations. The numbers of amino acid changes (substitutions/aa site/year) were calculated using molecular divergence time estimates: vertebrates–cephalochordates 891 Ma and bony fishes–cartilaginous fishes 520 Ma (Blair and Hedges 2005). For the PHD data set corresponding to the catalytic domain (Homo sapiens PHD2 residues 181–426) and the HIF alpha data set corresponding to the bHLH–PAS domain (1–360), the vertebrate and invertebrate sequences were sufficiently similar to allow calculation using the invertebrate outgroup. For the HIF alpha, the whole CDS, and the C-terminal part of the protein (CDS minus bHLH–PAS domain), only the epaulette shark outgroup was used. The domains mediating the PHD–HIF interactions were chosen for targeted analysis. For PHDs, we selected the beta2beta3 loop (PHD2 237–254) (Chowdhury et al. 2009), which is crucial for specifying the activity of PHDs on the HIF ODDs (Flashman et al. 2008). For HIF, the selection of N-terminal oxygen-dependent degradation (NODD) and C-terminal oxygen-dependent degradation (CODD) regions for targeted analysis was based on the experimentally characterized human ODD region (Koivunen et al. 2006), though excluding regions corresponding to the major indels in vertebrate HIF-1 alpha (Rytkönen et al. 2007) and taking into account the differential distance between prolines in the HIF-1 alpha and HIF-2 alpha primary structures (162 and 126 amino acids, respectively). In both HIF-1 alpha and HIF-2 alpha data sets, we included 63 codon positions in the ODD between the two prolines and 30 codon positions toward the exterior of the ODD prolines (HIF-1 alpha NODD = 373–465, HIF-1 alpha CODD = 502–594, HIF-2 alpha NODD = 376–468, and HIF-2 alpha CODD = 469–561), resulting in 94 aligned codon positions in each analysis. Model Testing of Selection Pressures To test whether the gene paralogs have experienced statistically significant differences in selection pressures, we employed ML estimates of the ratio of nonsynonymous to synonymous substitutions (dN/dS= ω) and nested likelihood ratio tests (LRTs) on a species phylogeny. This was done separately in mammals and teleost lineages. To avoid synonymous substitution saturation, we did not use the duplicate-specific gene phylogeny but a species phylogeny and concatenated sequence data from each set of paralogs (e.g., HIF-1 alpha, HIF-2 alpha, and HIF-3 alpha concatenated). We used the partition data set option (G) (Yang and Swanson 2002) in codeml, which is part of the PAML package (Yang 2007), to test if the gene paralogs have experienced statistically significant differences in selection pressures. The test was performed with the data sets mentioned above. Only the species for which there were data for all three paralogs were included. Teleosts’ PHD data sets were partial: Only three species were included, the catalytic domain consisted only of the C-terminal 177 aa, and the beta2beta3 loop fragment was analyzed only in PHD1 and PHD2. Nested LRTs were performed for the following series of model comparisons: first, mode of selection was equal (ω1 = ω2 = ω3); second, selection pressures were significantly different in one of the three paralogs but equal in two others (ω1 = ω2 ≠ ω3 or ω1 ≠ ω2 = ω3); third, selective pressures were different for each (ω1 ≠ ω2 ≠ ω3). Inside a model, different combinations of free and fixed parameters were tested (Mgene = 0,1,3,4, and Mgene = 3 with fixed κ), including the substitution rate (s), the transition/transversion ratio (κ), codon frequencies (πs) and the dN/dS ratio (ω). HIF-1 alpha and HIF-2 alpha NODD and CODD domains (see above) were analyzed together in a data set of four partitions for both mammals and teleosts. LRTs were done for each model with only substitution rate parameters, and for the best fitting model, more parameters were added and tested. Functional Divergence Analysis We analyzed functional divergence for two periods during the evolutionary history of the HIF and PHD genes: first on the divergence of the vertebrate paralogs following their duplication early in vertebrate evolution (using Amphioxus sequences as outgroups) and second on the divergence within each paralogous group after the split between teleosts and amniotes. We used three computational methods for identifying residues under functional divergence. The Type I method of Gu (1999), implemented in DIVERGE (Gu and Vander Velden 2002), uses an ML procedure to compare evolutionary rates in two predefined clades of sequences. The likelihood of the data given a coefficient of functional divergence θ > 0 is compared with the likelihood given no functional divergence (θ = 0) using an LRT. If the null hypothesis is rejected, posterior probabilities are assigned to amino acid sites, which give the probability of that site being under functional divergence. DIVERGE also implements a second (Type II) method (Gu 2006) that identifies “conserved-but-different” residues between two clades of sequences. The third method, which is conceptually similar to the Type II analysis, uses a simple distance-based approach in which BLOSUM substitution scores are used to quantify the radical or conservative nature of substitutions between two clades of sequences, with the score corrected for alignment column conservation (Toft et al. 2009; Williams et al. 2010). For teleost PHDs, there were data for three species only; therefore, only the paralog analysis is discussed in detail. Results and Discussion Emergence of the PHD/HIF Oxygen-Sensing Pathway in Metazoans We took a comparative genomics approach to identify the point in evolutionary history when the components of the PHD/HIF oxygen-sensing pathway emerged. First, we used protein Blast searches to identify HIF and PHD homologs in the available metazoan genomes and several nonmetazoan outgroups. We found that PHD genes were already present outside metazoans and were recruited to regulate HIFs in the course of early metazoan evolution (fig. 1). The commonly employed model invertebrates, the roundworm C. elegans and the fruit fly D. melanogaster, have only one genomic PHD copy (named Egl-9 and fatiga in these organisms). Interestingly, we found that there are two genomic copies of PHDs present in several invertebrate lineages (figs. 1 and 2). The Chordates (amphioxus B. floridae and purple urchin S. purpuratus) and the Cnidarian sea anemone contained two PHD homologs. The catalytic prolyl hydroxylation domain and surrounding region are conserved in the invertebrate sequences (catalytic domain of human PHD2 has 57% and 38% identities to its amphioxus and 48% and 43% to its sea anemone homologs). It seems feasible that PHDs may have been recruited to HIF regulation from other pre-existing functions. For example, in the amoeba (Dictyostelium discoideum), a PHD1 mediates oxygen signaling during development (West et al. 2007), and in fission yeast (Saccharomyces pombe), Ofd1, a prolyl hydroxylase family member, regulates a transcription factor Sre1, the fission yeast sterol regulatory element binding protein (Hughes and Espenshade 2008). Thus, the recruitment of PHD to interact with a bHLH–PAS transcription factor is a plausible example of pre-existing components evolving to become a novel physiological regulatory system. FIG. 1. View large Download slide Overview of HIF alpha and PHD gene products in selected metazoan lineages. The left column of blocks shows the PHD proteins, and the columns to the right show HIF alpha proteins. In PHD blocks, the symbol M indicates the presence of an MYND-type Zn2+ finger domain. In HIF alpha blocks, the number indicates the length of the protein in amino acids. For both proteins, the most ubiquitously expressed human paralogs (hsPHD2 and hsHIF-1 alpha) were chosen as a reference for amino acid identity comparisons that are indicated in percentage values in the blocks (for PHDs the catalytic domain and for HIF alphas whole CDS). In HIF, the P symbol indicates proline hydroxylation motifs; in the case where there is only one symbol, it aligns with the C-terminal ODD domain motif, and when there are two symbols, these indicate the presence of N-terminal and C-terminal ODD domain motifs. The arrows below the P symbols indicate a change in relative importance of the ODD in the regulation. In Chondrichthyes (cartilaginous fishes), “partial s” represent sequence fragments from Smooth dogfish (HIF-3 alpha) and Elephant shark (PHD3). Cnidarians have only one prolyl hydroxylation motif, whereas insect and chordate HIF alphas have two proline hydroxylation sites, corresponding to the vertebrate CODD core and NODD core. Caenorhabditis elegans has only one proline site that is very divergent from both C-terminal and N-terminal motifs. In the Arthropoda lineage, the honeybee (Apis mellifera) HIF homolog has both NODD and CODD motifs present (as has the grass shrimp (Palaemonetes pugio) and the red flour beetle (Tribolium castaneum), but the model organism fruit fly (Drosophila melanogaster) is an exception and has only CODD (supplementary fig. S1A and supplementary fig. S1B, Supplementary Material online). Inspection of other Drosophila species suggests that NODD is present in other insects (bees and beetles) but was specifically lost in the Drosophila genus. FIG. 1. View large Download slide Overview of HIF alpha and PHD gene products in selected metazoan lineages. The left column of blocks shows the PHD proteins, and the columns to the right show HIF alpha proteins. In PHD blocks, the symbol M indicates the presence of an MYND-type Zn2+ finger domain. In HIF alpha blocks, the number indicates the length of the protein in amino acids. For both proteins, the most ubiquitously expressed human paralogs (hsPHD2 and hsHIF-1 alpha) were chosen as a reference for amino acid identity comparisons that are indicated in percentage values in the blocks (for PHDs the catalytic domain and for HIF alphas whole CDS). In HIF, the P symbol indicates proline hydroxylation motifs; in the case where there is only one symbol, it aligns with the C-terminal ODD domain motif, and when there are two symbols, these indicate the presence of N-terminal and C-terminal ODD domain motifs. The arrows below the P symbols indicate a change in relative importance of the ODD in the regulation. In Chondrichthyes (cartilaginous fishes), “partial s” represent sequence fragments from Smooth dogfish (HIF-3 alpha) and Elephant shark (PHD3). Cnidarians have only one prolyl hydroxylation motif, whereas insect and chordate HIF alphas have two proline hydroxylation sites, corresponding to the vertebrate CODD core and NODD core. Caenorhabditis elegans has only one proline site that is very divergent from both C-terminal and N-terminal motifs. In the Arthropoda lineage, the honeybee (Apis mellifera) HIF homolog has both NODD and CODD motifs present (as has the grass shrimp (Palaemonetes pugio) and the red flour beetle (Tribolium castaneum), but the model organism fruit fly (Drosophila melanogaster) is an exception and has only CODD (supplementary fig. S1A and supplementary fig. S1B, Supplementary Material online). Inspection of other Drosophila species suggests that NODD is present in other insects (bees and beetles) but was specifically lost in the Drosophila genus. FIG. 2. View largeDownload slide ML phylogeny of PHD protein sequences using the alignment of the catalytic domain (C-terminal 244 amino acids) with the JTT + G model. Bootstrap values >50% are indicated. Inside each vertebrate paralog, the lineages used for the dN/dS model testing and functional divergence analysis are circled. Notably, all the paralogs have diversified more in teleost than in mammalian radiation. The bootstrap values are displayed only for the branches separating the paralogs and main lineages. The cut in the invertebrate branch indicates that the invertebrate part of the phylogeny was scaled down. FIG. 2. View largeDownload slide ML phylogeny of PHD protein sequences using the alignment of the catalytic domain (C-terminal 244 amino acids) with the JTT + G model. Bootstrap values >50% are indicated. Inside each vertebrate paralog, the lineages used for the dN/dS model testing and functional divergence analysis are circled. Notably, all the paralogs have diversified more in teleost than in mammalian radiation. The bootstrap values are displayed only for the branches separating the paralogs and main lineages. The cut in the invertebrate branch indicates that the invertebrate part of the phylogeny was scaled down. The PAS domain responsible for HIF dimerization has ancient origins, with prokaryote homologs involved in environmental sensing (Taylor and Zhulin 1999). We did not detect HIF alpha-like homologs in the nonmetazoan yeast (S. cerevisiae), choanoflagellate (M. brevicollis), or in the early-diverging metazoan sponge (Amphimedon queenslandica). The major sequence characteristics that distinguish HIF alpha from other bHLH–PAS family members are the regulatory hydroxylation motifs in the ODD domain responsible for ODD of HIF alpha (Koivunen et al. 2006; Kaelin and Ratcliffe 2008). In vertebrates, the ODD domain consists of N-terminal (NODD) and C-terminal (CODD) proline hydroxylation sites (supplementary fig. S1A and supplementary fig. S1B, Supplementary Material online) possessing conserved motifs surrounding a core Leu-X-X-Leu-Ala-Pro signature. The conservation of the core CODD was detected in predicted invertebrate HIF alphas (supplementary fig. S1B, Supplementary Material online), including Cnidarian sea anemone (N. vectensis). A recent experimental study showed that tablet animal (T. adhaerens) has HIF alpha with a slightly divergent ODD core and that this organism has a functional PHD–HIF oxygen-sensing system (Loenarz et al., 2011). Cnidarian sea anemone is the most divergent organism having HIF alpha with both prolyl hydroxylation motif and a C-terminal factor inhibiting HIF (FIH) asparginyl hydroxylation motif (supplementary fig. S2, Supplementary Material online). The bilaterian model species, C. elegans and D. melanogaster, possess simplified HIF genes with only one ODD proline (supplementary fig. S1A and supplementary fig. S1B, Supplementary Material online) and no full FIH hydroxylation motif (supplementary fig. S2, Supplementary Material online). Together with variation in the number of PHD copies among invertebrates, this suggests that work on the HIF oxygen-sensing systems of these model species should be only very carefully projected onto other bilaterian species. We found no correlation between the presence of multiple PHD genes and elaboration of the HIF/PHD regulatory system via multiple ODD cores. For instance, N. vectensis has two PHD genes, but its HIF gene possesses only the CODD motif, whereas arthropods (with the exception of D. melanogaster among the species so far studied) have one PHD but have both the NODD and the CODD in their HIF alpha. This apparent uncoupling may suggest that the presence of multiple PHDs was not a prerequisite for the emergence of two regulatory ODD domains in HIF alpha. Chordates appear to have two PHD copies and one HIF alpha with both proline hydroxylation motifs. In vertebrates there are three PHDs and three HIF alphas, of which HIF-1 alpha and HIF-2 alpha have both proline motifs. The NODD and CODD motifs are present in early-branching chordate and most arthropod sequences, suggesting that the fine tuning of the PHD–HIF interaction by two separate hydroxylation motifs was established before the evolution of vertebrates. This existing reservoir of PHD–HIF alpha interactions was then further elaborated after early vertebrate genome duplications. Regulation via Myeloid, Nervy, and DEAF-1 Domain Is an Ancient Character of PHDs The phylogeny (fig. 2) of metazoan PHD sequences suggests a scenario where the three jawed vertebrate copies have emerged from one invertebrate gene copy. Of the invertebrate PHDs amphioxus (A), purple urchin (A) and insect homologs most closely resemble this putative ancestral vertebrate gene, whereas other invertebrate sequences have more affinity to the homolog of the early-diverging tablet animal. PHD2 has the lowest distance to the invertebrate root of the three vertebrate duplicates. This is in agreement with a previous report that, based on only two invertebrate sequences, identified vertebrate PHD2 as the ancestral form of the vertebrate gene family (Taylor 2001). The major difference between the three vertebrate PHD coding sequences is the composition of the N-terminal parts of the protein. PHD1 and PHD3 lack recognizable interaction domains, whereas PHD2 possesses a MYND (myeloid, Nervy, and DEAF-1) type zinc finger domain. It was recently discovered that this domain anchors PHD2 via the protein FKBP38 to the mitochondrial or endoplasmic reticulum membranes (Barth et al. 2009). This study suggested that membrane bound PHD2 is degraded, whereas cytosolic PHD2 is stable. Thus, MYND-mediated interaction may be crucial in the regulation of PHD2. Most invertebrate Bilaterian PHDs have the MYND domain (fig. 1 and supplementary fig. S3, Supplementary Material online). In all of these species, the MYND domain is characterized by the presence of the typically conserved cysteines (6) and histidines (2). In the case of the second purple urchin PHD copy (which is more distant from vertebrate PHDs), the predicted CDS includes the N-terminus without the MYND domain. In nonbilaterian metazoa, we could not detect an MYND domain in the predicted gene structures. In some cases, this is explained by only partial gene prediction, leading to a partial sequence of the N-terminal part of the gene. Support for the ancestral presence of the MYND domain comes from an Alveolata marine protist (Perkinsus olseni) PHD that has an intact MYND domain (Leite et al. 2008). The most parsimonious interpretation of these data is that the regulation of PHD activity at the mitochondrial or ER membrane is an ancient feature of oxygen sensing that may have been lost in some lineages (fig. 1 and supplementary fig. S3, Supplementary Material online). This conclusion will remain tentative until more data on the distribution of the MYND domain in non and early-branching metazoans become available. Vertebrate Diversification of the PHD–HIF System It is now well established that two rounds of genome duplication took place during the early evolution of vertebrates (Kuraku et al. 2009; Ravi et al. 2009). The exact timing of the second whole-genome duplication is still unclear, with most data suggesting that it took place before the divergence of the lamprey from gnathostomes (jawed vertebrates; Kuraku et al. 2009) or somewhat later (Putnam et al. 2008). In the lamprey genome, we detected fragments of two separate HIF homologs, whereas the PHD fragments were too short to be informative. A complete HIF alpha CDS was found in contig2889 (supplementary table. S1, Supplementary Material online). In our phylogeny based on the bHLH–PAS domain (fig. 3), it grouped closer to the vertebrate HIF-2 alpha than to HIF-1 alpha or HIF-3 alpha. It also contained the HIF-2 alpha-specific exon 15. Although a second HIF homolog (contig26464) was only partial (302 aa), a phylogeny based on the 190 residues (supplementary fig. S4, Supplementary Material online) suggested that both lamprey HIF alphas result from a lamprey-specific duplication of HIF-2 alpha. Tentatively, because the lamprey HIF alphas fall within the vertebrate HIF1/2/3 radiation, they represent a further data point in favor of the second whole-genome duplication occurring before the split between the lamprey and jawed vertebrates. FIG. 3. View largeDownload slide ML phylogeny of HIF alpha protein sequences using the first 360 predicted amino acids of the alignment (bHLH–PAS domain) with the JTT + G-model. Bootstrap values >50% are indicated. Inside each vertebrate paralog, the lineages used for the dN/dS model testing and functional divergence analysis are circled. Notably, all the paralogs have diversified more in teleost than in mammalian radiation. The bootstrap values are displayed only for the branches separating the paralogs and main lineages. The cut in the invertebrate branch indicates that the invertebrate part of the phylogeny was scaled down. FIG. 3. View largeDownload slide ML phylogeny of HIF alpha protein sequences using the first 360 predicted amino acids of the alignment (bHLH–PAS domain) with the JTT + G-model. Bootstrap values >50% are indicated. Inside each vertebrate paralog, the lineages used for the dN/dS model testing and functional divergence analysis are circled. Notably, all the paralogs have diversified more in teleost than in mammalian radiation. The bootstrap values are displayed only for the branches separating the paralogs and main lineages. The cut in the invertebrate branch indicates that the invertebrate part of the phylogeny was scaled down. Cartilaginous fishes, which include the elasmobranchs, are an early-branching gnathostome lineage that diverged from the lineage leading to tetrapods and teleosts approximately 450 Ma (Sansom et al. 1996). We provide evidence that cartilaginous fishes possess three copies of both PHD and HIF alpha that correspond to those found in teleosts and tetrapods. We sequenced PHD2, PHD1, HIF-1 alpha and HIF-2 alpha genes from epaulette shark (H. ocellatum). These genes share a high identity with their mammalian and teleost counterparts; for example, shark PHD2 has 78% identity to human PHD2 and shark HIF-1 and HIF-2 alphas have highly conserved ODD domains (see supplementary fig. S1A and S1B, Supplementary Material online). PHD3 and HIF-3 alpha did not amplify from the cDNA collection from 10 different epaulette shark tissues even after extensive universal primer and PCR optimizations. In the elephant shark (C. milii) draft genome, we identified orthologs of amniote and teleost PHD1, 2, and 3 (supplementary fig. S5, Supplementary Material online), but we only detected very short sequence fragments for HIF-3 alpha. However, we were able to amplify HIF-3 alpha from the heart cDNA of another elasmobranch species, the hypoxia-sensitive smooth dogfish (Mustelus canis). The absence or very low transcription of PHD3 and HIF-3 alpha in epaulette shark transcriptome may reflect the biology of the hypoxia-tolerant animal. PHDs are negative regulators of HIF alphas, and in mammals, HIF-3 alpha variants negatively regulate HIF-1 alpha (Makino et al. 2002; Jang et al. 2005) and HIF-2 alpha (Maynard et al. 2007). Epaulette sharks are constantly exposed to hypoxic stress in their natural environment. This may cause the hypoxia response to be constantly primed and reduced inhibition of the HIF response to hypoxia via HIF-3 alpha or PHDs could be beneficial. In contrast, smooth dogfish that inhabit generally well-oxygenated areas may benefit from tight regulation of the HIF response. Molecular Evolution of Vertebrate PHD Duplicates A phylogeny of invertebrate and vertebrate PHDs suggests that the vertebrate duplicates arose from a single invertebrate gene (fig. 2), with the first duplication leading to a split between PHD2 and PHD1/3, followed by the divergence of the PHD1 and 3 genes. Genomic-shared synteny analysis (supplementary table. S3, Supplementary Material online) confirms that human and teleost PHD1, PHD2, and PHD3 are true one-to-one orthologs. The C-terminal catalytic domain (Chowdhury et al. 2009) is conserved across all three vertebrate PHD duplicates, but PHD1 and 3 have lost the N-terminal MYND domain (supplementary fig. S3, Supplementary Material online). Our analysis demonstrates that the PHD duplicates arising from the teleost-specific whole-genome duplication were not universally conserved, with only zebrafish (Danio rerio) having a detectable whole CDS from two PHD2 duplicates. Species from Tetraodontidae (Fugu rubripes and Tetraodon nigrividis) have one predicted whole CDS PHD1 homolog and another only partial PHD1 homolog. An analysis of the catalytic domain of PHDs indicated that all three duplicates have been generally under strong selective constraint (table 1). Despite this, both the PHD2 lineage versus PHD1/3 lineage (P < 0.05) and PHD1 versus PHD3 lineages (P < 0.01) were subject to significant functional divergence (supplementary table. S4, Supplementary Material online). With our distance-based analysis of functional divergence (fig. 4 and supplementary table. S5, Supplementary Material online), we detected positions of functional divergence in regions that are crucial in the catalytic activities and the substrate selectivity of the enzyme. Four of the detected PHD3 FD sites (E260G, D278R, R281L, and G294E/A; all P < 0.01) are located close to the channel binding HIF-1 alpha CODD peptide (fig. 4). We measured the minimum Euclidean distances between these four positions in PHD2 and the CODD residues (supplementary table. S6, Supplementary Material online). All four residues are within 6 Å of the HIF CODD (with the closest site being 294 at <3 Å) well within the usual range over which protein–protein interactions can occur (Gloor et al. 2005) and thus may be involved in catalytic interactions. Three of the FD positions with the greatest statistical significance (PHD1/3vs2 D246V/I, PHD1vs3 S247P, and PHD3vs1 S248K; all P < 0.01) are concentrated in a beta2beta3 loop (fig. 4) that has been experimentally characterized (Chowdhury et al. 2009). In PHD2, this loop displays considerable conformational changes upon ligand binding via Arg-252 and Asp-254 (Chowdhury et al. 2009) and is reported to determine the substrate specificity of the enzyme toward HIF CODD or NODD (Villar et al. 2007; Flashman et al. 2008). This pattern of having the most significant FD sites of different PHD lineages closely localized could indicate differential compensatory mutations that have been important for the emergence for the substrate selectivity. Lastly, three of the detected PHD2 functional divergence sites (A224D/E, D232T, and R236T; all P < 0.01) locate together in the region separating beta2beta3 loop from the channel where the enzyme binds CODD peptide (fig. 4). Table 1. Summary of the Average Amino Acid Substitution Rates and ML Hypothesis Testing of Selection Pressures for PHDs. Data Set  Lineage (n)    Subs. Ratea   dN/dS  Best Fitting Model after LRTsb    Invert.  Shark  CDS  Mammals (9)  PHD2  0.40  0.31  0.06  ω(PHD2) ≠ ω(PHD1) = ω(PHD3)      PHD1  0.43  0.54  0.05  Different rs. κ. ω      PHD3  0.47  na  0.05      Teleosts (3)  PHD2  0.40  0.34  0.06  ω(PHD2) = ω(PHD1) ≠* ω(PHD3)      PHD1  0.45  0.61  0.06  Different rs      PHD3  0.53  na  0.06    b2b3 Loop  Mammals (9)  PHD2  0.44  0.32  0.01  ω(PHD2) = ω(PHD1) = ω(PHD3)      PHD1  0.37  0.74  0.01  Homogenous      PHD3  0.57  na  0.01      Teleosts (3)  PHD2  0.44  0.39  0.00  ω(PHD2) = ω(PHD1)      PHD1  0.46  0.95  0.00  Homogenous      PHD3  0.47  na  na    Data Set  Lineage (n)    Subs. Ratea   dN/dS  Best Fitting Model after LRTsb    Invert.  Shark  CDS  Mammals (9)  PHD2  0.40  0.31  0.06  ω(PHD2) ≠ ω(PHD1) = ω(PHD3)      PHD1  0.43  0.54  0.05  Different rs. κ. ω      PHD3  0.47  na  0.05      Teleosts (3)  PHD2  0.40  0.34  0.06  ω(PHD2) = ω(PHD1) ≠* ω(PHD3)      PHD1  0.45  0.61  0.06  Different rs      PHD3  0.53  na  0.06    b2b3 Loop  Mammals (9)  PHD2  0.44  0.32  0.01  ω(PHD2) = ω(PHD1) = ω(PHD3)      PHD1  0.37  0.74  0.01  Homogenous      PHD3  0.57  na  0.01      Teleosts (3)  PHD2  0.44  0.39  0.00  ω(PHD2) = ω(PHD1)      PHD1  0.46  0.95  0.00  Homogenous      PHD3  0.47  na  na    NOTE.—Average amino acid substitution rates (substitutions/site year × 10−9) for mammals and teleost were calculated using one invertebrate outgroup for all paralogs and shark outgroups inside a given paralog. Selection pressures between the paralogs were tested in inside lineage partitioned data sets and dN/dS from the best-fitting model; the model and parameters are shown. a Amino acid substitution rate (substitutions/site year × 10−9) using molecular estimate of the divergence time, for Invertebrate Amphioxus 891 Ma and for sharks 525 Ma. b Parameters: substitution rate (s), transition/transversion ratio (κ), codon frequencies (πs) and dN/dS ratio (ω). In the case of ω(1) = ω(2) ≠ ω(3) versus ω(1) ≠ ω(2) = ω(3), the model with better fitting likelihood value was chosen. The log likelihood values and LRTs of the models are available from the authors on request. View Large Table 2. Summary of the Average Amino Acid Substitution Rates and ML Hypothesis Testing of Selection Pressures for HIF alphas. Data Set  Lineage    Subs. Ratea  dN/dS  Best Fitting Model after LRTsb  CDS  Mammals (10)  HIF1  0.67  0.10  ω(HIF1) ≠ ω(HIF2) ≠ ω(HIF3)      HIF2  0.69  0.12  Different rs. κ. πs. ω      HIF3  na  0.18      Teleosts (6)  HIF1  0.72  0.17  ω(HIF1) = ω(HIF2) ≠ ω(HIF3)      HIF2  0.81  0.17  Different rs. κ. ω      HIF3  na  0.21    HIF1 and 2 ODDs  Mammals (10)  1 NODD  0.80  0.05  ω(1N) ≠ ω(2N) ≠ ω(1C) ≠ ω(2C)      2 NODD  0.89  0.18  Different rs. κ. πs. ω      1 CODD  0.63  0.07  Separate setsc      2 CODD  0.59  0.08      Teleosts (6)  1 NODD  0.86  0.20  ω(1N) ≠ ω(2N) ≠ ω(1C) = ω(2C)      2 NODD  1.03  0.31  Different rs. κ. πs. ω      1 CODD  0.83  0.18        2 CODD  0.77  0.18    Data Set  Lineage    Subs. Ratea  dN/dS  Best Fitting Model after LRTsb  CDS  Mammals (10)  HIF1  0.67  0.10  ω(HIF1) ≠ ω(HIF2) ≠ ω(HIF3)      HIF2  0.69  0.12  Different rs. κ. πs. ω      HIF3  na  0.18      Teleosts (6)  HIF1  0.72  0.17  ω(HIF1) = ω(HIF2) ≠ ω(HIF3)      HIF2  0.81  0.17  Different rs. κ. ω      HIF3  na  0.21    HIF1 and 2 ODDs  Mammals (10)  1 NODD  0.80  0.05  ω(1N) ≠ ω(2N) ≠ ω(1C) ≠ ω(2C)      2 NODD  0.89  0.18  Different rs. κ. πs. ω      1 CODD  0.63  0.07  Separate setsc      2 CODD  0.59  0.08      Teleosts (6)  1 NODD  0.86  0.20  ω(1N) ≠ ω(2N) ≠ ω(1C) = ω(2C)      2 NODD  1.03  0.31  Different rs. κ. πs. ω      1 CODD  0.83  0.18        2 CODD  0.77  0.18    NOTE.—Average amino acid substitution rates (substitutions/site year × 10−9) for mammals and teleost were calculated using elasmobranch outgroup inside a given paralog. Selection pressures (dN/dS) between the paralogs were tested in inside lineage partitioned data sets and dN/dS from the best fitting model; the model and parameters are shown. a Amino acid substitution rate using epaulette shark sequences as outgroup (substitutions/site year × 10−9) and using molecular estimate of the divergence time for sharks, 525 Ma. b Parameters: substitution rate (s), transition/transversion ratio (κ), codon frequencies (πs), and dN/dS ratio (ω). In the case of ω(1) = ω(2) ≠ ω(3) versus ω(1) ≠ ω(2) = ω(3), the model with better fitting likelihood value was chosen. The log likelihood values and LRTs of the models are available from the authors on request. c In this case, a separate analysis of the four sets is shown, as a more complex model was nearly significant (LRT df 2 = 5,26 [P(0,05) = 5,99]) when using partition data set option (G). View Large FIG. 4. View largeDownload slide Residues under functional divergence in vertebrate PHDs. Colors denote the part of the phylogeny in which those sites experienced functional divergence (P < 0.01, distance-based method): PHD1/3 after split from PHD2, blue; PHD2 lineage, red; PHD1 lineage, green; PHD3 lineage, yellow. An HIF fragment corresponding to the CODD domain is visible in the bottom left of the image. The Homo sapiens PHD2 (hsPHD2) structure (Chowdhury et al. 2009) was visualized in iMol (http://www.pirx.com/iMol/) and numbering is according to hsPHD2. FIG. 4. View largeDownload slide Residues under functional divergence in vertebrate PHDs. Colors denote the part of the phylogeny in which those sites experienced functional divergence (P < 0.01, distance-based method): PHD1/3 after split from PHD2, blue; PHD2 lineage, red; PHD1 lineage, green; PHD3 lineage, yellow. An HIF fragment corresponding to the CODD domain is visible in the bottom left of the image. The Homo sapiens PHD2 (hsPHD2) structure (Chowdhury et al. 2009) was visualized in iMol (http://www.pirx.com/iMol/) and numbering is according to hsPHD2. The molecular evolution of PHD2 and PHD1 was more closely examined in teleosts and mammalian lineages using the novel elasmobranch sequences as an outgroup. Amino acid substitution rates in mammalian and teleost PHD2 genes were moderate when using either amphioxus or our epaulette shark sequences as the outgroup (table 1). However, the use of the epaulette shark outgroup when calculating the substitution rates for PHD1 revealed an unexpected rate shift. In particular, the rate of evolution in PHD1 beta2beta3 loop is more than twice faster in both mammals and teleosts using the shark outgroup compared with that using the amphioxus outgroup. In agreement with these rate shifts, the Type I method (Gu 1999) indicated that PHD1, but not PHD2, was subject to functional divergence between mammals and teleosts (supplementary table. S4, Supplementary Material online). This is reflected in the Gu type II functional analysis where 14 functionally divergent positions are detected in PHD1 versus only one in PHD2 (posterior probability >0.75, supplementary table. S5, Supplementary Material online). These results support a scenario where the PHD2 catalytic domain and beta2beta3 loop did not experience rate shifts at the protein level and conserved the interactions characteristic of the ancestral PHD gene. In contrast, the functional diversification of PHD1 was most likely prevalent in the Osteichthyes (bony fishes) lineage after its separation from cartilageous fishes but not after the divergence of the Actinopterygii (teleosts) and Sarcopterygii (including mammals) lineages. These changes in the PHD1 lineage have resulted in less effective hydroxylation of the HIF core NODD compared with PHD2 as has been demonstrated in vitro for mammalian components (Hirsila et al. 2003). Molecular Evolution of the Vertebrate HIF Alpha Duplicates The phylogeny (fig. 3) of the elasmobranch HIF alpha sequences, together with all other available HIF sequences, confirms that the three HIF isoforms present in extant vertebrate genomes are products of two successive gene duplication events that occurred prior to the divergence of cartilaginous fishes (elasmobranch and chimera) from the lineage leading to tetrapods and teleosts. The phylogeny suggests a scenario in which the ancestral HIF alpha first duplicated to preHIF1/2 and preHIF-3 alpha. In a second round of duplication, preHIF1/2 alpha duplicated to HIF-1 alpha and HIF-2 alpha. A shared genomic gene order of the flanking genes and shared synteny (supplementary table. S7, Supplementary Material online) confirms that human HIF-1 alpha and HIF-2 alpha are orthologous to teleost HIF-1 alpha and HIF-2 alpha. Our results are more equivocal about the relationships among the HIF-3 alpha genes in cartilaginous fishes, teleosts, and mammals, which is currently a matter of some debate (Law et al. 2006; Richards 2009). Because there is one HIF-3 alpha copy in each lineage, the most parsimonious interpretation is that they are orthologs. No shared gene order was detectable between the human and teleost HIF-3 alpha genomic regions, but tetraodon HIF-3 alpha flanking genes were present on same chromosome as human HIF-3 alpha (supplementary table. S7, Supplementary Material online). The smooth dogfish (M. canis) HIF-3 alpha that we sequenced grouped unexpectedly within the teleost HIF-3 alpha genes, rather than as an outgroup to teleost and amniote sequences (fig. 3). This tentatively supports a scenario in which modern HIF-3 alpha genes derive from loss of different ancestral HIF-3 alpha duplicates in cartilaginous fishes/teleosts on the one hand and amniotes on the other. The analysis of teleost model genomes suggested that teleost-specific HIF alpha duplicates were generally not retained—only zebrafish has detectable full CDS from two duplicates of each of the HIFs (HIF-1 alpha a/b, HIF-2 alpha a/b, and HIF-3 alpha a/b). Taken together with the presence of partial euteleostei (Gasterosteus aculeatus and T. nigrividis) HIF-2 alpha b sequences, tentative support is given to a scenario where the HIF alpha duplicates observed in zebrafish would have originated in the ancient teleost whole-genome duplication rather than in ostariphysii/cyprinid-specific duplications. All vertebrate HIF-3 alpha duplicates underwent a major functional shift after separation from the HIF1/2 lineage. The major exon losses in the HIF-3 alpha C-terminus restricted our alignment and analysis to the HIF-3 alpha bHLH–PAS domain. These deletions also included the C-terminal activation domain that interacts with transcriptional activators, such as p300 (Lando et al. 2002; Pasanen et al. 2010) so that HIF-3 alphas are predicted to negatively regulate other HIF alphas. Our analysis of functional divergence highlights molecular changes in HIF-3 alpha bHLH–PAS domains. Even though amino acid substitution rates and dN/dS ratios in HIF-3 alpha are only slightly higher compared with those in HIF-1 or 2 alphas (table 2 and supplementary table. S8, Supplementary Material online), we detected 19 amino acid positions that were subject to significant functional divergence in the branch leading to HIF-3 alpha compared with only four amino acid positions in the branch leading to preHIF1/2 (fig. 5). Five of the 19 detected HIF-3 alpha sites (152–173) align in a part of PAS A domain that is included in the inhibitory splice variants human HIF-3alpha4 that negatively regulates both HIF-1 alpha (Jang et al. 2005) and HIF-2 alpha (Maynard et al. 2007) and mouse inhibitory PAS domain protein that negatively regulates HIF-1 alpha (Makino et al. 2002). These sites may contribute to the dimerization properties of HIF-3 alpha variants that enable them to function as dominant repressors. FIG. 5. View largeDownload slide Idealized diagram showing functional divergence analysis of selected HIF alpha duplicates. Only results from HIF1/2 versus HIF-3 alpha paralogs, mammalian HIF-1 alpha versus teleost HIF-1 alpha orthologs and mammalian HIF-2 alpha versus teleost HIF-2 alpha orthologs are shown. Distance-based method was used (with P < 0.01). For analysis of paralogs, we used an invertebrate protein as an outgroup, whereas for analysis of mammalian and teleost lineages, we used elasmobranch HIF-1 alpha and HIF-2 alpha proteins as outgroups. The significant functional divergence positions are shown as lines in their corresponding positions on the primary structure. Amino acid positions discussed in the text are shown, first amino acid is the outgroup and the latter one is lineage specific. The color code shows the domain positions in a schematic overview of the HIF1/2 (below). Basic Helix–Loop–Helix (bHLH) and Per-ARNT-Sim (PASA, PASB/C) domains are involved in DNA binding and dimerization of the protein and are marked with red. NODD and CODD domains are involved in PHD dependent degradation of HIF and are marked with blue and green, respectively. Transactivation domains (NAD, CAD) confer regulation via protein–protein interactions to other transcriptional activators. The proportion of functionally divergent sites of the whole coding sequence and of a composite of both ODD domains is shown (% values above the diagrams), and the positions of crucial prolyl hydroxylation sites are depicted (P). FIG. 5. View largeDownload slide Idealized diagram showing functional divergence analysis of selected HIF alpha duplicates. Only results from HIF1/2 versus HIF-3 alpha paralogs, mammalian HIF-1 alpha versus teleost HIF-1 alpha orthologs and mammalian HIF-2 alpha versus teleost HIF-2 alpha orthologs are shown. Distance-based method was used (with P < 0.01). For analysis of paralogs, we used an invertebrate protein as an outgroup, whereas for analysis of mammalian and teleost lineages, we used elasmobranch HIF-1 alpha and HIF-2 alpha proteins as outgroups. The significant functional divergence positions are shown as lines in their corresponding positions on the primary structure. Amino acid positions discussed in the text are shown, first amino acid is the outgroup and the latter one is lineage specific. The color code shows the domain positions in a schematic overview of the HIF1/2 (below). Basic Helix–Loop–Helix (bHLH) and Per-ARNT-Sim (PASA, PASB/C) domains are involved in DNA binding and dimerization of the protein and are marked with red. NODD and CODD domains are involved in PHD dependent degradation of HIF and are marked with blue and green, respectively. Transactivation domains (NAD, CAD) confer regulation via protein–protein interactions to other transcriptional activators. The proportion of functionally divergent sites of the whole coding sequence and of a composite of both ODD domains is shown (% values above the diagrams), and the positions of crucial prolyl hydroxylation sites are depicted (P). Molecular Evolution in HIF-2 Alpha NODD Domain May Be Associated with the Divergence in Oxygen Sensitivity To address the interactions between HIF alpha and PHD crucial for oxygen sensing, we first investigated the evolution of the HIF alpha NODD and CODD domains in detail. We concentrated on the activatory HIF-1 alpha and HIF-2 alpha duplicates that have both a NODD and CODD domain present in their coding sequences and analyzed air-breathing mammals and water-breathing teleost in separate data sets. Based on comparisons of amino acid substitution rates and dN/dS ratios (table 2), we found that the CODD is generally subject to slightly more stringent negative selection in vertebrates than the NODD. In both lineages, but particularly in mammals, the selection pressure on the HIF-2 alpha NODD appears to be relaxed. Analysis of functional divergence with the distance-based method identified more conserved, radical changes in the teleost ODDs (HIF-1 3.8% and HIF-2 6.5% of the ODD amino acid sites) compared with mammals (both HIF-1 and HIF-2: 1.6%; fig. 5). This suggests that adaptive evolution may underlie the elevated dN/dS values in teleosts, whereas in mammals, relaxed constraint is a more likely explanation. In the absence of complete structural information for HIF alpha, the significance of the detected functional divergence sites in teleosts remains unclear. Our results agree with previous in vitro studies on these mammalian genes: PHDs hydroxylate the HIF-1 alpha CODD and the HIF-2 alpha CODD with a similar efficiency (Appelhoff et al. 2004), but the ubiquitous PHD2 hydroxylates the NODD of HIF-1 alpha more efficiently than that of HIF-2 alpha. Also, it has been reported that hydroxylation of NODD is mainly PHD2 dependent and is relatively more important in HIF-1 alpha regulation in hypoxic than in normoxic conditions (Appelhoff et al. 2004; Chan et al. 2005). Our results suggest that the oxygen-dependent regulation via NODD has become relatively less important for HIF-2 than HIF-1 alpha in vertebrates. As this effect was clearer in mammals, the relaxation of negative selection on the HIF-2 alpha NODD may be directly correlated with the reduced demand of air breathers to sustain HIF-2 alpha sensitivity to lowered oxygen tensions. This result is consistent with reports that HIF-2 alpha plays a more important role in chronic hypoxia than HIF-1 alpha does (Holmquist-Mengelbier et al. 2006). More HIF Alpha Duplicate Functional Divergence in Water-Breathing Teleosts Than in Air-Breathing Mammals When considering the whole HIF alpha gene-coding sequences, the evolution of vertebrate HIF-1 alpha and HIF-2 alpha has been rather similar: They have similar exon structures, share the major regulatory domains, and have experienced similar substitution rates across mammalian and teleost species (table 2 and supplementary table. S8, Supplementary Material online). The slightly higher substitution rate in teleost HIF-2 alpha compared with its mammalian counterpart is attributable to the C-terminal regions near the HIF-2 alpha-specific exon. This exon, which encodes 57 residues, is located 27 codons upstream of the asparginyl hydroxylation motif of HIF-2 alpha (which centers at Asn-847) and therefore may be important in transcriptional regulation involving p300 (Lando et al. 2002). The dN/dS ratios after model testing were slightly higher in teleosts than in mammals in both HIF-1 alpha and HIF-2 alpha (table 2 and supplementary table. S8, Supplementary Material online). This suggests either relaxed selective constraint or adaptive evolution due to aquatic hypoxia in the teleost lineage but may also partly reflect evolution of redundant pairs generated by the teleost whole-genome duplication (Brunet et al. 2006; Ravi and Venkatesh 2008). Our functional divergence analysis provides some evidence for adaptive evolution in water-breathing teleosts. With the distance-based method, we detected more functionally divergent amino acid sites in both teleost HIF-1 alpha (3.5%) and teleost HIF-2 alpha (2.4%) than their mammalian counterparts (1.8% and 1.6%; fig. 5 and supplementary table. S9, Supplementary Material online). In the highly conserved bHLH–PAS domain, we found 10 significant (P < 0.01) sites in teleost HIF-1 alpha compared with only 2 significant sites in the mammalian counterpart. These constitute two clusters of functionally divergent sites in the teleosts HIF-1 alpha bHLH–PAS domain: S149V, Q152T, L155S, and G158A/T in the PAS A domain and N205T/E/A, H209P/D, C210N/S, and C219Y between the PAS A and B domains. The latter region may be involved in the spatial orientation of the coordinated dimerization of PAS A and PAS B (Card et al. 2005; Konietzny et al. 2009). Previous evidence for positive selection was found in this region at position 203 (Rytkönen et al. 2008). We hypothesize that the detected four radical substitutions may have led to a change in the dimerization properties of teleost HIF1 alpha. This, taken with our data on HIF-3 alpha bHLH–PAS domain, and also the report on HIF-2 alpha-specific PAS B ligand binding (Scheuermann et al. 2009), emphasizes that subfunctionalization of HIF duplicates may also have occurred within the largely conserved PAS domains of HIF alpha protein, not only in the C-terminal domains. Conclusions This study has comprehensively examined the molecular evolution of the metazoan PHD–HIF oxygen-sensing system, by placing existing experimental work in a phylogenetic context and using modern computational tools to characterize the molecular changes underlying the functional divergence of HIF alpha and PHD duplicates. The PHD–HIF oxygen-sensing system provides an excellent example of how pre-existing components can be recruited into a new system during evolution. The PAS domain responsible for HIF dimerization has ancient origins with prokaryote homologs involved in environmental sensing (Taylor and Zhulin 1999). Oxygen-sensing homologs of PHDs were already present in nonmetazoan eukaryotes (West et al. 2007; Hughes and Espenshade 2008) and during metazoan evolution were recruited into a bHLH–PAS transcription factor circuitry. Early metazoan gene duplications of bHLH–PAS transcription factors resulted in the appearance of HIF-like bHLH–PAS transcription factors before the divergence of the cnidarian and bilaterian lineages. The PHD–HIF interaction was further refined in Bilaterians by the introduction of two separate proline motifs that permit more precise coordination of oxygen-dependent HIF regulation. The prechordate origins of this innovation could not be deduced from studies on Drosophila or Caenorhabditis, as these model organisms have only one predicted proline in the HIF gene product. Therefore, these animals may have atypical PHD–HIF systems, emphasizing the need for caution when extrapolating results from these to other bilaterian organisms. The early vertebrate genome duplications led to further refinement in oxygen sensing; we have shown that cartilaginous fishes contain three duplicates of both PHD and HIF alpha in their genomes. Our analysis of functional divergence identified sites that may underlie the differential preference of PHDs for specific HIF alpha duplicates. We found that PHD1, but not PHD2, experienced a period of accelerated evolution early in vertebrate evolution associated with functional divergence in the beta2beta3 loop, which mediates PHD substrate specificity (Villar et al. 2007; Flashman et al. 2008; Chowdhury et al. 2009). In this loop, we detected functional divergence sites in three PHD lineages that may indicate differential compensatory mutations. Second, for PHD3, we found four FD sites in the vicinity of the HIF binding channel that may be involved in the PHD3-specific features of ODD binding. In addition to the divergence of PHDs, molecular evolution in HIF alpha ODD domains has contributed to PHD–HIF interactions. In particular, the HIF-2 alpha NODD domain has been subject to relaxed selective constraint in vertebrates. Our analysis suggests that divergence in both PHD1/3 and HIF-2 alpha ODD domains has resulted in less intensive oxygen-dependent regulation of HIF-2 alpha than HIF-1 alpha. More functional divergence was detected in water-breathing teleosts compared with air-breathing mammals, especially in HIF-2 alpha ODD and HIF-1 alpha PAS domains. These amino acid changes in HIF alphas may have been beneficial for teleosts in the highly variable oxygen tensions of aquatic environments. Overall, with the increasing medical interest in the regulation of the HIF system by specific inhibitors of PHD activity (Myllyharju 2009; Harten et al. 2010), we anticipate that evolutionary insights into PHD–HIF interactions will be useful from both the ecological and medical perspectives. We thank Mario Fares and Erica Leder for useful discussions. We thank Maria Zwart for help in making the graphics. The University of Turku, the Academy of Finland, and the Irish Research Council for Science, Engineering and Technology are acknowledged for funding. References Abascal F,  Zardoya R,  Posada D.  ProtTest: selection of best-fit models of protein evolution,  Bioinformatics ,  2005, vol.  21 (pg.  2104- 2105) Google Scholar CrossRef Search ADS PubMed  Appelhoff RJ,  Tian YM,  Raval RR,  Turley H,  Harris AL,  Pugh CW,  Ratcliffe PJ,  Gleadle JM.  Differential function of the prolyl hydroxylases PHD1, PHD2, and PHD3 in the regulation of hypoxia-inducible factor,  J Biol Chem ,  2004, vol.  279 (pg.  38458- 38465) Google Scholar CrossRef Search ADS PubMed  Barth S,  Edlich F,  Berchner-Pfannschmidt U,  Gneuss S,  Jahreis G,  Hasgall PA,  Fandrey J,  Wenger RH,  Camenisch G.  Hypoxia-inducible factor prolyl-4-hydroxylase PHD2 protein abundance depends on integral membrane anchoring of FKBP38,  J Biol Chem ,  2009, vol.  284 (pg.  23046- 23058) Google Scholar CrossRef Search ADS PubMed  Blair JE,  Hedges SB.  Molecular phylogeny and divergence times of deuterostome animals,  Mol Biol Evol ,  2005, vol.  22 (pg.  2275- 2284) Google Scholar CrossRef Search ADS PubMed  Brunet FG,  Crollius HR,  Paris M,  Aury JM,  Gibert P,  Jaillon O,  Laudet V,  Robinson-Rechavi M.  Gene loss and evolutionary rates following whole-genome duplication in teleost fishes,  Mol Biol Evol ,  2006, vol.  23 (pg.  1808- 1816) Google Scholar CrossRef Search ADS PubMed  Card PB,  Erbel PJA,  Gardner KH.  Structural basis of ARNT PAS-B dimerization: use of a common beta-sheet interface for hetero- and homodimerization,  J Mol Biol ,  2005, vol.  353 (pg.  664- 677) Google Scholar CrossRef Search ADS PubMed  Chan DA,  Sutphin PD,  Yen SE,  Giaccia AJ.  Coordinate regulation of the oxygen-dependent degradation domains of hypoxia-inducible factor 1 alpha,  Mol Cel Biol ,  2005, vol.  25 (pg.  6415- 6426) Google Scholar CrossRef Search ADS   Chowdhury R,  McDonough MA,  Mecinovic J,  Loenarz C,  Flashman E,  Hewitson KS,  Domene C,  Schofield CJ.  Structural basis for binding of hypoxia-inducible factor to the oxygen-sensing prolyl hydroxylases,  Structure ,  2009, vol.  17 (pg.  981- 989) Google Scholar CrossRef Search ADS PubMed  Edgar RC.  MUSCLE: a multiple sequence alignment method with reduced time and space complexity,  BMC Bioinformatics ,  2004, vol.  5 (pg.  1- 19) Google Scholar CrossRef Search ADS PubMed  Flashman E,  Bagg EAL,  Chowdhury R,  Mecinovic J,  Loenarz C,  McDonough MA,  Hewitson KS,  Schofield CJ.  Kinetic rationale for selectivity toward N- and C-terminal oxygen-dependent degradation domain substrates mediated by a loop region of hypoxia-inducible factor prolyl hydroxylases,  J Biol Chem ,  2008, vol.  283 (pg.  3808- 3815) Google Scholar CrossRef Search ADS PubMed  Gloor GB,  Martin LC,  Wahl LM,  Dunn SD.  Mutual information in protein multiple sequence alignments reveals two classes of coevolving positions,  Biochemistry ,  2005, vol.  44 (pg.  7156- 7165) Google Scholar CrossRef Search ADS PubMed  Gu X.  Statistical methods for testing functional divergence after gene duplication,  Mol Biol Evol ,  1999, vol.  16 (pg.  1664- 1674) Google Scholar CrossRef Search ADS PubMed  Gu X.  A simple statistical method for estimating type-II (cluster-specific) functional divergence of protein sequences,  Mol Biol Evol ,  2006, vol.  23 (pg.  1937- 1945) Google Scholar CrossRef Search ADS PubMed  Gu X,  Vander Velden K.  DIVERGE: phylogeny-based analysis for functional-structural divergence of a protein family,  Bioinformatics ,  2002, vol.  18 (pg.  500- 501) Google Scholar CrossRef Search ADS PubMed  Guindon S,  Gascuel O.  A simple, fast, and accurate algorithm to estimate large phylogenies by maximum likelihood,  Syst Biol ,  2003, vol.  52 (pg.  696- 704) Google Scholar CrossRef Search ADS PubMed  Harten SK,  Ashcroft M,  Maxwell PH.  Prolyl hydroxylase domain inhibitors: a route to HIF activation and neuroprotection,  Antiox Redox Signaling ,  2010, vol.  12 (pg.  459- 480) Google Scholar CrossRef Search ADS   Hirsila M,  Koivunen P,  Gunzler V,  Kivirikko KI,  Myllyharju J.  Characterization of the human prolyl 4-hydroxylases that modify the hypoxia-inducible factor,  J Biol Chem ,  2003, vol.  278 (pg.  30772- 30780) Google Scholar CrossRef Search ADS PubMed  Holmquist-Mengelbier L,  Fredlund E,  Lofstedt T, et al.  ,  (12 co-authors).  Recruitment of HIF-1 alpha and HIF-2 alpha to common target genes is differentially regulated in neuroblastoma: HIF-2 alpha promotes an aggressive phenotype,  Cancer Cell ,  2006, vol.  10 (pg.  413- 423) Google Scholar CrossRef Search ADS PubMed  Hughes BT,  Espenshade PJ.  Oxygen-regulated degradation of fission yeast SREBP by Ofd1, a prolyl hydroxylase family member,  EMBO J ,  2008, vol.  27 (pg.  1491- 1501) Google Scholar PubMed  Ivan M,  Kondo K,  Yang HF,  Kim W,  Valiando J,  Ohh M,  Salic A,  Asara JM,  Lane WS,  Kaelin WG.  HIF alpha targeted for VHL-mediated destruction by proline hydroxylation: implications for O-2 sensing,  Science ,  2001, vol.  292 (pg.  464- 468) Google Scholar CrossRef Search ADS PubMed  Jaakkola P,  Mole DR,  Tian YM, et al.  ,  (13 co-authors).  Targeting of HIF-alpha to the von Hippel–Lindau ubiquitylation complex by O-2-regulated prolyl hydroxylation,  Science ,  2001, vol.  292 (pg.  468- 472) Google Scholar CrossRef Search ADS PubMed  Jang MS,  Park JE,  Lee JA,  Park SG,  Myung PK,  Lee DH,  Park BC,  Cho S.  Binding and regulation of hypoxia-inducible factor-1 by the inhibitory PAS proteins,  Biochem Biophys Res Commun ,  2005, vol.  337 (pg.  209- 215) Google Scholar CrossRef Search ADS PubMed  Kaelin WG,  Ratcliffe PJ.  Oxygen sensing by metazoans: the central role of the HIF hydroxylase pathway,  Molecular Cell ,  2008, vol.  30 (pg.  393- 402) Google Scholar CrossRef Search ADS PubMed  Kewley RJ,  Whitelaw ML,  Chapman-Smith A.  The mammalian basic helix-loop-helix/PAS family of transcriptional regulators,  Int J Biochem Cell Biol ,  2004, vol.  36 (pg.  189- 204) Google Scholar CrossRef Search ADS PubMed  Koivunen P,  Hirsila M,  Kivirikko KI,  Myllyharju J.  The length of peptide substrates has a marked effect on hydroxylation by the hypoxia-inducible factor prolyl 4-hydroxylases,  J Biol Chem ,  2006, vol.  281 (pg.  28712- 28720) Google Scholar CrossRef Search ADS PubMed  Koivunen P,  Tiainen P,  Hyvarinen J,  Williams KE,  Sormunen R,  Klaus SJ,  Kivirikko KI,  Myllyharju J.  An endoplasmic reticulum transmembrane prolyl 4-hydroxylase is induced by hypoxia and acts on hypoxia-inducible factor alpha,  J Biol Chem ,  2007, vol.  282 (pg.  30544- 30552) Google Scholar CrossRef Search ADS PubMed  Konietzny R,  Konig A,  Wotzlaw C,  Bernadini A,  Berchner-Pfannschmidt U,  Fandrey J.  Peers C.  Haddad GG,  Chandel NS.  Molecular imaging into in vivo interaction of HIF-1 alpha and HIF-2 alpha with ARNT,  Hypoxia and consequences from molecule to malady ,  2009(pg.  74- 81) Kumar S,  Tamura K,  Nei M.  MEGA3: integrated software for molecular evolutionary genetics analysis and sequence alignment,  Brief Bioinf ,  2004, vol.  5 (pg.  150- 163) Google Scholar CrossRef Search ADS   Kuraku S,  Meyer A,  Kuratani S.  Timing of genome duplications relative to the origin of the vertebrates: did cyclostomes diverge before or after?,  Mol Biol Evol ,  2009, vol.  26 (pg.  47- 59) Google Scholar CrossRef Search ADS PubMed  Lando D,  Peet DJ,  Gorman JJ,  Whelan DA,  Whitelaw ML,  Bruick RK.  FIH-1 is an asparaginyl hydroxylase enzyme that regulates the transcriptional activity of hypoxia-inducible factor,  Genes Dev ,  2002, vol.  16 (pg.  1466- 1471) Google Scholar CrossRef Search ADS PubMed  Law SHW,  Wu RSS,  Ng PKS,  Yu RMK,  Kong RYC.  Cloning and expression analysis of two distinct HIF-alpha isoforms—gcHIF-1alpha and gcHIF-4alpha—from the hypoxia-tolerant grass carp, Ctenopharyngodon idellus.,  BMC Mol Biol ,  2006, vol.  7 pg.  15  Google Scholar CrossRef Search ADS PubMed  Leite RB,  Brito AB,  Cancela ML.  An oxygen molecular sensor, the HIF prolyl 4-hydroxylase, in the marine protist Perkinsus olseni,  Protist ,  2008, vol.  159 (pg.  355- 368) Google Scholar CrossRef Search ADS PubMed  Lendahl U,  Lee KL,  Yang H,  Poellinger L.  Generating specificity and diversity in the transcriptional response to hypoxia,  Nat Rev Genet ,  2009, vol.  10 (pg.  821- 832) Google Scholar CrossRef Search ADS PubMed  Loenarz C,  Coleman ML,  Boleininger A,  Schierwater B,  Holland PW,  Ratcliffe PJ,  Schofield CJ.  The hypoxia-inducible transcription factor pathway regulates oxygen sensing in the simplest animal, Trichoplax adhaerens,  EMBO Rep ,  2011, vol.  12 (pg.  63- 70) Google Scholar CrossRef Search ADS PubMed  Makino Y,  Kanopka A,  Wilson WJ,  Tanaka H,  Poellinger L.  Inhibitory PAS domain protein (IPAS) is a hypoxia-inducible splicing variant of the hypoxia-inducible factor-3 alpha locus,  J Biol Chem ,  2002, vol.  277 (pg.  32405- 32408) Google Scholar CrossRef Search ADS PubMed  Maynard MA,  Evans AJ,  Shi W,  Kim WY,  Liu FF,  Ohh M.  Dominant-negative HIF-3 alpha 4 suppresses VHL-null renal cell carcinoma progression,  Cell Cycle ,  2007, vol.  6 (pg.  2810- 2816) Google Scholar CrossRef Search ADS PubMed  Myllyharju J.  HIF prolyl 4-hydroxylases and their potential as drug targets,  Curr Pharm Des ,  2009, vol.  15 (pg.  3878- 3885) Google Scholar CrossRef Search ADS PubMed  Pasanen A,  Heikkila M,  Rautavuoma K,  Hirsila M,  Kivirikko KI,  Myllyharju J.  Hypoxia-inducible factor (HIF)-3a is subject to extensive alternative splicing in human tissues and cancer cells and is regulated by HIF-1 but not HIF-2,  Int J Biochem Cell Biol ,  2010, vol.  42 (pg.  1189- 1200) Google Scholar CrossRef Search ADS PubMed  Putnam NH,  Butts T,  Ferrier DEK, et al.  ,  (37 co-authors).  The amphioxus genome and the evolution of the chordate karyotype,  Nature ,  2008, vol.  453 (pg.  1064- 1071) Google Scholar CrossRef Search ADS PubMed  Ravi V,  Lam K,  Tay BH,  Tay A,  Brenner S,  Venkatesh B.  Elephant shark (Callorhinchus milii) provides insights into the evolution of Hox gene clusters in gnathostomes,  Proc Natl Acad Sci U S A ,  2009, vol.  106 (pg.  16327- 16332) Google Scholar CrossRef Search ADS PubMed  Ravi V,  Venkatesh B.  Rapidly evolving fish genomes and teleost diversity,  Curr Opin Genet Dev ,  2008, vol.  18 (pg.  544- 550) Google Scholar CrossRef Search ADS PubMed  Richards JG.  Richards JG,  Farrell AP,  Brauner CJ.  Metabolic and molecular responses of fish to hypoxia,  Hypoxia in fishes ,  2009 San Diego (CA) Elsevier(pg.  443- 485) Rytkönen KT,  Renshaw GMC,  Ashton KJ,  Williams-Pritchard G,  Leder EH,  Nikinmaa M.  Elasmobranch qPCR reference genes: a case study of hypoxia preconditioned epaulette sharks,  BMC Mol Biol ,  2010, vol.  11 pg.  27  Google Scholar CrossRef Search ADS PubMed  Rytkönen KT,  Ryynänen HJ,  Nikinmaa M,  Primmer CR.  Variable patterns in the molecular evolution of the hypoxia-inducible factor-1 alpha (HIF-1 alpha) gene in teleost fishes and mammals,  Gene ,  2008, vol.  420 (pg.  1- 10) Google Scholar CrossRef Search ADS PubMed  Rytkönen KT,  Vuori KAM,  Primmer CR,  Nikinmaa M.  Comparison of hypoxia-inducible factor-1 alpha in hypoxia-sensitive and hypoxia-tolerant fish species,  Comp Biochem Physiol D ,  2007, vol.  2 (pg.  177- 186) Sansom IJ,  Smith MM,  Smith MP.  Scales of thelodont and shark-like fishes from the Ordovician of Colorado,  Nature ,  1996, vol.  379 (pg.  628- 630) Google Scholar CrossRef Search ADS   Scheuermann TH,  Tomchick DR,  Machius M,  Guo Y,  Bruick RK,  Gardner KH.  Artificial ligand binding within the HIF2 alpha PAS-B domain of the HIF2 transcription factor,  Proc Natl Acad Sci U S A ,  2009, vol.  106 (pg.  450- 455) Google Scholar CrossRef Search ADS PubMed  Taylor MS.  Characterization and comparative analysis of the EGLN gene family,  Gene ,  2001, vol.  275 (pg.  125- 132) Google Scholar CrossRef Search ADS PubMed  Taylor BL,  Zhulin IB.  PAS domains: internal sensors of oxygen, redox potential, and light,  Microbiol Mol Biol Rev ,  1999, vol.  63 pg.  479  Google Scholar PubMed  Toft C,  Williams TA,  Fares MA.  Genome-wide functional divergence after the symbiosis of proteobacteria with insects unraveled through a novel computational approach.,  PLoS Comp Biol ,  2009, vol.  5 pg.  e1000344  Google Scholar CrossRef Search ADS   Villar D,  Vara-Vega A,  Landazuri MO,  Del Peso L.  Identification of a region on hypoxia-inducible-factor prolyl 4-hydroxylases that determines their specificity for the oxygen degradation domains,  Biochem J ,  2007, vol.  408 (pg.  231- 240) Google Scholar CrossRef Search ADS PubMed  Wenger RH,  Camenisch G,  Stiehl DP,  Katschinski DM.  HIF prolyl-4-hydroxylase interacting proteins: consequences for drug targeting,  Curr Pharm Des ,  2009, vol.  15 (pg.  3886- 3894) Google Scholar CrossRef Search ADS PubMed  West CM,  van der Wel H,  Wang ZA.  Prolyl 4-hydroxylase-1 mediates O-2 signaling during development of Dictyostelium10.1242/dev.000893,  Development ,  2007, vol.  134 (pg.  3349- 3358) Google Scholar CrossRef Search ADS PubMed  Williams TA,  Codoner FM,  Toft C,  Fares MA.  Two chaperonin systems in bacterial genomes with distinct ecological roles,  Trends Genet ,  2010, vol.  26 (pg.  47- 51) Google Scholar CrossRef Search ADS PubMed  Yang ZH.  PAML 4: phylogenetic analysis by maximum likelihood,  Mol Biol Evol ,  2007, vol.  24 (pg.  1586- 1591) Google Scholar CrossRef Search ADS PubMed  Yang ZH,  Swanson WJ.  Codon-substitution models to detect adaptive evolution that account for heterogeneous selective pressures among site classes,  Mol Biol Evol ,  2002, vol.  19 (pg.  49- 57) Google Scholar CrossRef Search ADS PubMed  © The Author 2011. Published by Oxford University Press on behalf of the Society for Molecular Biology and Evolution. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
TI - Molecular Evolution of the Metazoan PHD–HIF Oxygen-Sensing System
JF - Molecular Biology and Evolution
DO - 10.1093/molbev/msr012
DA - 2011-01-12
UR - https://www.deepdyve.com/lp/oxford-university-press/molecular-evolution-of-the-metazoan-phd-hif-oxygen-sensing-system-7RMn8DtqRL
SP - 1913
EP - 1926
VL - 28
IS - 6
DP - DeepDyve
ER -