|Home | About | Journals | Submit | Contact Us | Français|
Hypoxia-inducible factor (HIF), consisting of a labile α subunit and a stable β subunit, is a master regulator of hypoxia-responsive mRNAs. HIFα undergoes oxygen-dependent prolyl hydroxylation, which marks it for polyubiquitination by a complex containing the von Hippel-Lindau protein (pVHL). Among the three Phd family members, Phd2 appears to be the primary HIF prolyl hydroxylase. Phd3 is induced by HIF and, based on findings from in vitro studies, may participate in a HIF-regulatory feedback loop. Here, we report that Phd3 loss exacerbates the HIF activation, hepatic steatosis, dilated cardiomyopathy, and premature mortality observed in mice lacking Phd2 alone and produces a closer phenocopy of the changes seen in mice lacking pVHL than the loss of Phd2 alone. Importantly, the degree to which Phd3 can compensate for Phd2 loss and the degree to which the combined loss of Phd2 and Phd3 resembles pVHL loss appear to differ for different HIF-responsive genes and in different tissues. These findings highlight that the responses of different HIF target genes to changes in prolyl hydroxylase activity differ, quantitatively and qualitatively, in vivo and have implications for the development of paralog-specific prolyl hydroxylase inhibitors as therapeutic agents.
Many human diseases, including anemia, myocardial infarction, and stroke, are closely linked to tissue hypoxia. The transcriptional response to hypoxia in metazoans is mediated primarily by the heterodimeric transcriptional factor hypoxia-inducible factor (HIF), which consists of an α subunit (HIFα) and a β subunit (HIFβ, also called the aryl hydrocarbon receptor nuclear translocator) (21). Under hypoxic conditions, HIFα accumulates, dimerizes with HIFβ, translocates to the nucleus, and transcriptionally activates genes containing hypoxia response elements. Many of these genes regulate processes, such as erythropoiesis, angiogenesis, and energy metabolism, that affect oxygen delivery or oxygen consumption and thereby affect survival in a low-oxygen environment (47).
There are three HIFα family members in humans, called HIF1α, HIF2α, and HIF3α. Both HIF1α (the canonical HIFα family member) and HIF2α contain two transactivation domains, the N-terminal transactivation domain (NTAD) and the C-terminal transactivation domain (CTAD), and can activate transcription when bound to DNA. Whether HIF3α activates transcription is less certain. Multiple HIF3α splice variants have been identified, at least some of which are dominant-interfering with respect to HIFα activity (31, 32).
The mechanism that couples the accumulation of HIFα to oxygen availability has recently come into view. When oxygen is plentiful, HIFα becomes hydroxylated at one (or both) of two conserved prolyl residues (21). HIFα hydroxylated at one prolyl residue is recognized by a ubiquitin ligase complex that contains the von Hippel-Lindau tumor suppressor protein (pVHL), polyubiquitinated, and degraded by the proteasome.
Caenorhabditis elegans and Drosophila species contain a single prolyl hydroxylase, called Egl9, capable of hydroxylating HIFα (8, 9). In higher metazoans, there are three family members (PHD2, PHD1, and PHD3; also called EglN1, EglN2, and EglN3, respectively) that are capable of hydroxylating HIFα (8, 9, 20). These enzymes are 2-oxoglutarate- and iron-dependent dioxygenases that require oxygen in order to function. Moreover, their oxygen Kms are slightly higher than those that would be found in normal cells and tissues (17). Consequently, their activity is highly sensitive to decrements in tissue oxygen availability. In addition to oxygen, these enzymes are sensitive to changes in certain Krebs cycle intermediates and changes in the mitochondrial production of reactive oxygen species (21).
Although all three PHD family members can hydroxylate HIFα in vitro, PHD2 appears to be the primary HIFα prolyl hydroxylase based on findings from cell culture studies in which either PHD1, PHD2, or PHD3 was inactivated with specific small interfering RNAs (siRNAs) (1, 6). In particular, the elimination of PHD2 is sufficient, despite the presence of PHD1 and PHD3, to robustly induce HIFα. The importance of PHD2 as the primary HIF prolyl hydroxylase is also underscored by results from studies of genetically engineered mice. In particular, germ line Phd2 inactivation, like germ line Vhl inactivation, leads to embryonic lethality, whereas embryos lacking Phd1 or Phd3 are viable (12, 29, 35, 53). Somatic Phd2 inactivation, but not the inactivation of Phd1 or Phd3, leads to HIF upregulation and the development of polycythemia due to excessive production of the HIF-responsive gene product erythropoietin (35, 52). Finally, polycythemia has been reported to occur in people with hypomorphic germ line PHD2 mutations (24, 36, 37). Collectively, these findings support the suggestion that PHD2 is the primary enzyme responsible for hydroxylating HIF on proline and thereby earmarking it for destruction.
A second layer of HIFα regulation is provided by factor inhibiting HIF1 (FIH1). FIH1, like the PHD family members, is an iron- and 2-oxoglutarate-dependent dioxygenase. It inhibits HIFα transcriptional activity by hydroxylating a conserved asparaginyl residue located in the CTAD (16, 25). However, the inactivation of Phd2 alone transcriptionally activates a subset of HIF target genes (35, 52), presumably because these genes are driven by the HIFα NTAD or because they are responsive primarily to HIF2α, which is fairly insensitive to FIH1 compared to HIF1α (7, 54).
In stark contrast to mice lacking Phd2, mice lacking Phd1 or Phd3 are, as noted above, viable and grossly normal (53). Moreover, each PHD family member has distinctive patterns of tissue expression (28) and subcellular localization (34), which suggests that PHD family members may have distinct, HIF-independent functions. In this regard, Phd3 appears to be capable of inducing HIF-independent apoptosis in certain settings (26, 41).
Nonetheless, findings from studies of cells treated with combinations of siRNAs against the individual PHD family members establish that PHD1 and PHD3 can cooperate with PHD2 in cell culture and modulate the HIF response (1, 51). Moreover, mice lacking Phd1 are protected from limb ischemia in a HIF-dependent manner (3), and combined loss of Phd1 and Phd3 in the mouse leads to polycythemia (52). Therefore, Phd1 and Phd3 may, in conjunction with Phd2, affect the HIF response under certain conditions.
Noteworthy in this regard is the fact that PHD3 is itself a HIF target and is markedly induced by hypoxia (1, 2, 33, 51). This suggests that PHD3 may participate in a feedback loop to suppress the HIF response to prolonged hypoxia. Importantly, PHD3 appears to have a lower oxygen Km than PHD2 (51; P. Koivunen, personal communication) and therefore may remain active at intermediate levels of hypoxia that are sufficient to disable PHD2. Moreover, there are quantitative and qualitative differences with respect to HIF activation between mice lacking Phd2 and mice lacking pVHL (35). Among several possibilities, this may reflect residual prolyl hydroxylation of HIFα by Phd1 or Phd3 in cells lacking Phd2.
In order to further analyze HIFα regulation by prolyl hydroxylation in vivo, we generated Phd3−/− mice that are also homozygous for a conditional Phd2 allele. HIFα levels in mice lacking both Phd2 and Phd3 were higher than those in mice lacking either alone. Moreover, concurrent loss of Phd3 led to severe hepatic steatosis, as seen in mice lacking pVHL, and worsened the dilated cardiomyopathy and premature mortality observed previously in mice lacking Phd2. This genetic evidence suggests that Phd3 cooperates with Phd2 to regulate HIFα in vivo. Curiously, the loss of Phd3 did not, however, exacerbate the overproduction of erythropoietin and polycythemia in mice lacking Phd2, indicating that the degree of compensation can differ for different HIF targets and in different tissues.
Vhlflox/flox (VhlF/F) mice were a generous gift from Volker Haase (University of Pennsylvania) (13). Phd3−/− mice were a gift of Regeneron Pharmaceuticals, Inc. (Tarrytown, NY) (45). All animals in this study were backcrossed to C57BL/6 mice at least five times. The Phd2F/F mice and transgenic mice expressing the Cre-estrogen receptor (Cre-ER) transgene under the chicken actin promoter were described previously (13, 15, 19, 35). VhlF/F mice and Phd2F/F mice were crossed with the Cre-ER mice to generate Vhl+/F Cre-ER and Phd2+/F Cre-ER mice, respectively. Vhl+/F Cre-ER and Phd2+/F Cre-ER mice were then crossed with Vhl+/F and Phd2+/F mice, respectively, to generate VhlF/F Cre-ER and Phd2F/F Cre-ER mice, respectively, as well as relevant littermate controls. Phd2F/F Cre-ER mice were mated with Phd3−/− mice to generate Phd2+/F Phd3+/− Cre-ER mice. These mice were mated with Phd2+/F Phd3+/− mice to generate Phd2F/F Phd3+/+ Cre-ER mice, Phd2F/F Phd3−/− Cre-ER mice, and relevant littermate controls. Analogous crosses with mice expressing Cre under the control of the albumin promoter were performed (13).
Where indicated, 3-week-old mice were treated twice by intraperitoneal injections (48 h apart) with 1 mg of tamoxifen (T5648; Sigma-Aldrich, St. Louis, MO) dissolved in 100 μl of corn oil (Sigma). Blood samples were obtained by retro-orbital venous puncture.
Mice or cells were genotyped by PCR using the following primers: Phd2 Fwd1 (for the null allele), 5′-TCCATCCAGTCTGAGTTTCTTTCC-3′; Phd2 Fwd2 (for the wild-type [WT] and conditional [floxed] alleles), 5′-AGATGACCTCCCCAACTCTGCTAC-3′; Phd2 Rev (common primer), 5′-CAGTGTTCTGCCTCCATTTAT-3′; Phd3 Fwd1 (for the WT allele), 5′-GCCGGTAGACCAATGGGAG-3′; Phd3 Rev1 (for the WT allele), 5′-TCGTCAGACAGTCCCTTCAC-3′; Phd3 Fwd2 (for the null allele), 5′-GAGTTTCGAGCAACTTTCCC-3′; and Phd3 Rev2 (for the null allele), 5′-GTCTGTCCTAGCTTCCTCACTG-3′.
Phd2+/+ and Phd2−/− mouse embryonic fibroblasts (MEFs) were described previously (35). Vhl+/+ Cre-ER, VhlF/F Cre-ER, Phd2F/F Phd3+/+ Cre-ER, and Phd2F/F Phd3−/− Cre-ER MEFs were obtained from embryos 12.5 days after fertilization, maintained in Dulbecco's modified Eagle medium supplemented with 10% fetal bovine serum, and immortalized with a retrovirus encoding the K1 variant of simian virus 40 large T antigen (11). To activate Cre, cells were treated with 50 nM 4-hydroxytamoxifen (H7904; Sigma-Aldrich, St. Louis, MO) for 48 to 72 h. Where indicated, medium was supplemented with 1 mM dimethyloxalylglycine (DMOG [D1070; Frontier Scientific, Inc., Logan, UT]).
Cells grown in culture were rinsed with ice-cold phosphate-buffered saline, scraped, and lysed with 1× EBC buffer (50 mM Tris [pH 8.0], 120 mM NaCl, 0.5% NP-40) supplemented with a protease inhibitor cocktail (Complete; Roche Applied Science, Indianapolis, IN). For nuclear extracts, mouse tissue fragments (~100 μl) in 1.5-ml Eppendorf tubes were homogenized in ice-cold buffer containing 10 mM Tris-HCl (pH 7.8), 1.5 mM MgCl2, and 10 mM KCl supplemented with a protease inhibitor cocktail (Roche), 1 mM sodium orthovanadate, 0.5 mM dithiothreitol, and 0.4 mM phenylmethylsulfonyl fluoride (P-7626; Sigma-Aldrich) by using a plastic pestle. The homogenates were centrifuged at 4,500 × g for 5 min at 4°C. The resulting pellets were lysed with 8 M urea buffer containing 40 mM Tris-HCl (pH 7.6). For whole-cell extracts, tissue fragments were directly lysed in 8 M urea buffer.
Equal amounts of protein extract, as determined by the Bradford method with an assay kit from Bio-Rad Laboratories (Hercules, CA), were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and blotted onto polyvinylidene difluoride membranes (Millipore, Billerica, MA). Membranes were blocked with Tris-buffered saline containing 5% nonfat dry milk and probed with the following primary antibodies: rabbit polyclonal anti-HIF1α (NB100-479 [Novus, Littleton, CO] or AG10001 [A&G Pharmaceuticals, Columbia, MD]), rabbit polyclonal anti-HIF2α (NB100-122; Novus), rabbit polyclonal anti-Glut1 (NB300-666; Novus), rabbit polyclonal anti-PHD3 (NB100-303; Novus), rabbit polyclonal anti-FIH1 (NB100-428; Novus), mouse monoclonal antitubulin (B-512; Sigma-Aldrich), and mouse monoclonal antivinculin (V9131; Sigma- Aldrich). Bound antibody was detected with horseradish peroxidase-conjugated secondary antibodies (31430/31432; Pierce, Rockford, IL) and Immobilon Western chemiluminescent horseradish peroxidase substrate (Millipore).
mRNA was purified with TRIzol (Invitrogen, Carlsbad, CA) and an RNeasy column (Qiagen, Valencia, CA). A 0.5-μg sample of total RNA was then reverse transcribed with a StrataScript first-strand cDNA synthesis kit (Stratagene, La Jolla, CA). The resulting cDNA was used for real-time PCR with RT2 Profiler PCR arrays (SABiosciences, Frederick, MD) and the Mx 3005 PCR system (Stratagene). The following primers were used: Vhl Fwd, 5′-TTGTGGCTCAACTTCGACGG-3′; Vhl Rev, 5′-GGCAAAAATAGGCTGTCCATCG-3′; Hif1a Fwd, 5′-TGCTCATCAGTTGCCACTTCC-3′; Hif1a Rev, 5′-CCATCTGTGCCTTCATCTCATCTTC-3′; Epas1/HIF2a Fwd, 5′-ACGGAGGTCTTCTATGAGTTGGC-3′; Epas1/HIF2a Rev, 5′-GTTATCCATTTGCTGGTCGGC-3′; Hif3a Fwd, 5′-TATCTGTGAAGCCATCCCCCAC-3′; Hif3a Rev, 5′-TCCTCTCGTCGCAGTATGTGAAC-3′; IPAS Fwd, 5′-TGTTCCTCCTCCCTGATACATAACC-3′; IPAS Rev, 5′-CGTCTTGAAGTTCCTCTTGGTCAC-3′; Hif1an/FIH1 Fwd, 5′-TCAATAAACAGCAGGGGAAACG-3′; Hif1an/FIH1 Rev, 5′-ACAGGGTATGGATAGAGGCACTCG-3′; Phd1 Fwd, 5′-CTGGGCAACTACGTCATCAAT-3′; and Phd1 Rev, 5′-TGCACCTTAACATCCCAGTTC-3′. Sequences of primers for Phd2, Phd3, Epo, Vegf-a, and Pgk1 were published before (35). Commercially available primers (QuantiTect primer assay sets [Qiagen]) were used for Bnip3.
Murine transthoracic echocardiography was conducted with conscious 10-week-old mice using a Vevo 770 high-resolution microultrasound system (Visualsonics, Inc., Toronto, Canada) as described previously (35). Briefly, the heart was imaged in the two-dimensional parasternal short-axis view. An M-mode echocardiogram of the midventricle at the level of the papillary muscle was recorded. Heart rate, posterior wall thickness, and end-diastolic and end-systolic internal dimensions of the left ventricle (LV) were measured from the M-mode image. Fractional shortening of the LV was defined as the end-diastolic dimension minus the end-systolic dimension normalized for the end-diastolic dimension and was used as an index of cardiac contractile function.
All tissues were fixed with buffered 10% formalin solution (SF93-20; Fisher Scientific, Pittsburgh, PA). Heart tissues were fix perfused. For standard hematoxylin and eosin staining or trichrome staining, tissues were embedded in paraffin prior to sectioning. For oil red O staining, fixed tissues were embedded in optimal cutting temperature compound (product no. 4583; Sakura Finetek, Torrance, CA) and then frozen for sectioning. Photomicrographs were obtained with a BX51 microscope (40× objective lens and 10× eyepiece lens; total magnification, ×200), a Q-Color5 digital camera, and Q-Capture suite acquisition software (all from Olympus, Tokyo, Japan).
Hearts were fix perfused with ice-cold buffer containing 2.5% paraformaldehyde, 2.5% glutaraldehyde, and 0.1 M sodium cacodylate (pH 7.4 [catalog no. 15949; Electron Microscopy Sciences, Hatfield, PA]). Heart tissue from the luminal side of the LV wall was embedded for transmission electron microscopy using a Tecnai G2 Spirit BioTWIN microscope (FEI Company, Hillsboro, OR) with a cooled charge-coupled device camera (XR41C; Advanced Microscopy Techniques, Danvers, MA) and Image Caption Engine acquisition software (Advanced Microscopy Techniques).
Total DNA from hearts was isolated with Gentra Puregene buffer (catalog no. 158906; Qiagen) supplemented with proteinase K (catalog no. 19133; Qiagen). The amounts of mitochondrial DNA (mt-Co1) and nuclear DNA (Rn18s) were compared by real-time PCR using the following primers: mt-Co1 Fwd, 5′-CTGAGCGGGAATAGTGGGTA-3′; mt-Co1 Rev, 5′-TGGGGCTCCGATTATTAGTG-3′; Rn18s Fwd, 5′-CGGCTACCACATCCAAGGAA-3′; and Rn18s Rev, 5′-GCTGGAATTACCGCGGCT-3′.
Genomic DNA from heart, kidney, and liver tissues of 10-week-old mice treated with tamoxifen at 3 weeks of age was extracted with a Puregene kit (Qiagen). One hundred nanograms of DNA was used for real-time PCR with RT2 Profiler PCR arrays (SABiosciences). Primers were designed to amplify only WT or floxed alleles of Phd2 or Vhl (shown schematically in Fig. S1 in the supplemental material) (35).
Conditional Phd2 inactivation in the mouse leads to increased renal erythropoietin production and severe erythrocytosis to a degree similar to that seen after conditional pVHL inactivation (10, 35), suggesting that, at least in the kidney, Phd2 inactivation is sufficient to promote HIFα accumulation and enhanced transcription of the erythropoietin gene. However, some HIF target genes that are induced in mice lacking pVHL are not induced in mice lacking Phd2 (35). Moreover, HIF2α levels in hearts lacking pVHL are significantly higher than those in hearts lacking Phd2 (35). These observations suggest that prolyl hydroxylases in addition to Phd2 regulate HIFα in vivo.
To explore this possibility further, we first studied genetically engineered MEFs in vitro. Although Phd2−/− embryos are not viable, they survive long enough to obtain MEFs (53; also data not shown). In contrast, Vhl−/− embryos do not survive long enough to obtain MEFs (12, 29). To circumvent this problem, we created embryos that are homozygous for a conditional (floxed) Vhl allele (13) and that carry a transgene encoding a Cre-ER fusion protein (15), which can be activated by tamoxifen. We then obtained VhlF/F Cre-ER MEFs and littermate Vhl+/+ Cre-ER MEFs. These cells were treated with tamoxifen to generate MEFs with the floxed Vhl alleles recombined (VhlΔ/Δ MEFs) and Vhl+/+ MEFs, respectively.
As expected, HIF1α protein levels in Phd2−/− MEFs were higher than those in Phd2+/+ littermate control MEFs (Fig. (Fig.1A,1A, compare lanes 3 and 1). Importantly, however, the HIF1α levels in Phd2−/− MEFs were lower than those in VhlΔ/Δ MEFs (Fig. (Fig.1A,1A, compare lanes 7 and 3). Moreover, HIF1α was further induced by the hydroxylase inhibitor DMOG in Phd2−/− MEFs but not in VhlΔ/Δ MEFs (Fig. (Fig.1A,1A, compare lanes 4 and 3 and lanes 8 and 7). The simplest explanation for these data would be compensation of Phd2 loss by Phd1 or Phd3. Note that HIFα frequently appears as a series of fuzzy bands in immunoblot assays, at least partly because of differential phosphorylation.
PHD3 is a HIF target gene (1, 2, 33). Consistent with this property, we documented increased Phd3 mRNA and protein levels in Phd2−/− MEFs compared to Phd2+/+ MEFs (Fig. 1B and C, respectively; in Fig. Fig.1C,1C, compare lanes 3 and 1). The Phd3 protein band detected in Phd2−/− MEF samples (Fig. (Fig.1C,1C, lane 3) is specific because it was not detected in MEF samples derived from Phd3−/− embryos (Fig. (Fig.1C,1C, lanes 7 and 8) and was induced by DMOG (Fig. (Fig.1C,1C, compare lanes 2 and 1 and lanes 4 and 3). Therefore, Phd3 is a strong candidate to compensate for Phd2 loss.
To ask whether Phd3 can compensate for Phd2 loss in vivo, we generated, through appropriate matings, Phd2F/F Phd3+/+ Cre-ER mice and littermate Phd2F/F Phd3−/− Cre-ER mice. The Phd2F/F mice used in these studies were described by us previously (35). VhlF/F Cre-ER mice, harboring a conditional Vhl allele generated by Haase and coworkers (13), were generated in parallel as controls. MEFs obtained from mice with these three genotypes were then expanded and, in some cases, treated with tamoxifen to inactivate the floxed allele. MEFs that were not treated with tamoxifen were studied in parallel as controls. HIF1α was modestly induced after conditional inactivation of Phd2 in Phd3+/+ MEFs (Fig. (Fig.1D,1D, compare lanes 3 and 1). The induction of HIF1α was more pronounced when Phd2 was inactivated in cells that also lacked Phd3 (Fig. (Fig.1D,1D, compare lanes 4 and 3). Importantly, the degrees of Phd2 recombination in the Phd3+/+ and Phd3−/− MEFs were similar (data not shown). In cells lacking Phd2 and Phd3, both HIF1α and the HIF-responsive gene product Glut1 were further induced by DMOG to levels seen in cells lacking pVHL (data not shown). Whether this residual sensitivity to DMOG reflects incomplete recombination of the Phd2 locus or the involvement of a third prolyl hydroxylase, such as Phd1, is not clear.
Phd3 is induced by Phd2 inactivation and may participate in a negative feedback loop to dampen the HIF response. To model this process in vitro, we again treated Phd2F/F Phd3+/+ Cre-ER MEFs and littermate Phd2F/F Phd3−/− Cre-ER MEFs with tamoxifen in vitro to generate Phd2Δ/Δ Phd3+/+ and Phd2Δ/Δ Phd3−/− MEFs. At various time points following Phd2 inactivation, the cells were subjected to immunoblot analysis (Fig. (Fig.1E).1E). As expected, HIF1α, Glut1, and Phd3 protein levels increased after acute Phd2 inactivation in Phd3+/+ cells (Fig. (Fig.1E,1E, compare lanes 2 and 1). HIF1α and Glut1 levels peaked 6 days after Phd2 inactivation in Phd3+/+ cells (Fig. (Fig.1E,1E, lane 3) and diminished thereafter such that they approached baseline values by day 12 (Fig. (Fig.1E,1E, compare lanes 5 and 1). In contrast, the induction of both HIF1α and Glut1 after Phd2 loss in cells lacking Phd3 was more pronounced and sustained, with protein levels remaining elevated even 12 days after Phd2 inactivation (Fig. (Fig.1E,1E, compare lanes 7 to 10 and lane 6). The recombination efficiencies of Phd2 in the Phd3+/+ and Phd3−/− MEFs were comparable (data not shown). These results suggest that Phd3 regulates the intensity and duration of the HIF response to Phd2 loss.
We and others showed before that conditional Phd2 inactivation in the mouse causes the development of polycythemia, cardiac failure, and premature mortality (35, 52). We next generated littermate Phd2F/F Phd3+/+, Phd2F/F Phd3+/+ Cre-ER, and Phd2F/F Phd3−/− Cre-ER mice. These mice were treated with tamoxifen at 3 weeks of age to generate Phd2F/F Phd3+/+, Phd2Δ/Δ Phd3+/+, and Phd2Δ/Δ Phd3−/− mice, respectively. In parallel, VhlF/F Cre-ER mice were treated with tamoxifen at 3 weeks of age to generate VhlΔ/Δ mice. The efficiencies of Cre-mediated recombination of the conditional Phd2 and VHL alleles in the three organs examined (heart, liver, and kidney) were comparable (see Fig. S1 in the supplemental material).
Consistent with the findings of earlier studies, Phd2 inactivation led to a time-dependent increase in red blood cell production (Fig. (Fig.2A)2A) associated with an increase in circulating erythropoietin (Fig. (Fig.2B).2B). Surprisingly, the development of polycythemia in this setting was not substantially different, in either severity or rapidity of onset, from that observed in mice lacking both Phd2 and Phd3 or in mice lacking pVHL (Fig. (Fig.2A).2A). These results further underscore the importance of Phd2 with respect to the control of red blood cell production. Serum erythropoietin levels in mice lacking both Phd2 and Phd3 were likewise not demonstrably different from those in mice lacking Phd2 alone (Fig. (Fig.2B).2B). In contrast, serum erythropoietin levels in mice lacking pVHL were significantly higher than those in mice lacking Phd2 (with or without Phd3) (Fig. (Fig.2B).2B). This finding suggests that the levels of erythropoietin achieved in mice lacking Phd2 are sufficient to attain nearly maximal red blood cell production, although it remains possible that the trend toward more severe polycythemia in mice lacking pVHL would become statistically significant with the inclusion of more mice. Notably, the erythropoietin gene is one of many HIF-responsive genes dedicated to red blood cell production, and the biological effects of erythropoietin in these mouse models are likely to reflect cooperative effects between the erythropoietin gene and other such genes. In particular, these models may not mirror the administration of recombinant erythropoietin as a single agent.
Consistent with the results of an earlier study (35), circulating vascular endothelial growth factor (VEGF), in contrast to erythropoietin, was not increased in mice lacking Phd2 but was increased in mice lacking pVHL (Fig. (Fig.2B).2B). Intriguingly, circulating VEGF was also not increased in mice lacking both Phd2 and Phd3, despite robust HIF accumulation and measurably increased VEGF mRNA levels in a variety of tissues (see below). The lack of a systemic increase in circulating VEGF may reflect, at least in part, differences between VEGF and erythropoietin with respect to stability and binding to extracellular matrix proteins (5).
Despite the above-mentioned similarities in red blood cell production, mice lacking both Phd2 and Phd3 or lacking pVHL died substantially earlier than mice lacking Phd2 alone (Fig. (Fig.2C),2C), implying strong genetic interaction between Phd2 and Phd3. It should be noted that the median survival time of Phd2Δ/Δ mice was approximately 40 weeks in the present study and less than 15 weeks in our earlier work (35). In the earlier study, however, the first dose of tamoxifen was given in utero, whereas in the present study the first dose was given at 3 weeks of age.
Premature mortality in mice lacking Phd2 appears to be due, at least in part, to cardiac failure (35). Cardiac failure in this setting is likely caused by chronic volume overload, hyperviscosity, and an cell-intrinsic role of Phd2 in cardiomyocytes (J. Moslehi, Y. A. Minamishima, and W. G. Kaelin, Jr., unpublished data). The hearts from 10-week-old Phd2F/F Cre-ER mice treated with tamoxifen at 3 weeks of age were only minimally enlarged and retained normal systolic function in vivo (Fig. (Fig.3A3A and and4),4), in contrast to previous results for Phd2F/F Cre-ER mice treated in utero (35). The hearts from 10-week-old Phd3−/− mice appeared to be normal (data not shown). In contrast, the hearts from 10-week-old Phd2F/F Phd3−/− Cre-ER mice treated with tamoxifen at 3 weeks of age (Phd2Δ/Δ Phd3−/− mice) were markedly enlarged and exhibited profound systolic dysfunction in vivo (Fig. (Fig.3B3B and and4)4) compared to those from tamoxifen-treated littermate controls that lacked the Cre-ER transgene (Phd2F/F Phd3−/− mice) or tamoxifen-treated Phd2F/F Phd3+/+ Cre-ER (Phd2Δ/Δ Phd3+/+) mice. Moreover, the sizes and functions of hearts lacking Phd2 and Phd3 resembled those of hearts lacking pVHL (Fig. (Fig.33 and and44).
Histologic examination revealed only occasional myocytes showing degenerative changes, manifested as mild hypereosinophilia, in hearts from Phd2Δ/Δ Phd3+/+ mice (Fig. 5B and F) compared to control hearts (Fig. 5A and E). In contrast, severe cardiomyopathic changes such as myocyte dropout, increased interstitial fibrosis, myofibrillar degeneration (Fig. (Fig.5C),5C), and focal inflammatory cell infiltration (Fig. 5C and G) were observed in Phd2Δ/Δ Phd3−/− hearts. These changes were similar to those observed in VhlΔ/Δ hearts (Fig. 5D and H).
Transmission electron microscopy revealed mild mitochondrial swelling in Phd2Δ/Δ Phd3+/+ hearts (Fig. (Fig.5J)5J) compared to control hearts (Fig. (Fig.5I).5I). Strikingly, the ultrastructural changes, including mitochondrial swelling and degeneration, myofibrillar degeneration, and intracellular edema, were more pronounced in the Phd2Δ/Δ Phd3−/− hearts (Fig. (Fig.5K).5K). These ultrastructural changes were similar to those observed in VhlΔ/Δ hearts (Fig. (Fig.5L5L).
To clarify whether mitochondria are actually lost in Phd2Δ/Δ Phd3−/− hearts, we measured the amount of mitochondrial DNA by real-time PCR. The copy number of the mitochondrial gene mt-Co1, normalized to the amount of nuclear genomic DNA (Rn18s), was significantly decreased in Phd2Δ/Δ hearts compared to controls (Fig. (Fig.5M).5M). The decrease was even more pronounced, however, in Phd2Δ/Δ Phd3−/− hearts and in VhlΔ/Δ hearts (Fig. (Fig.5M5M).
We next asked whether the more pronounced cardiomyopathy in hearts lacking both Phd2 and Phd3 was associated with increased HIF activation relative to that in hearts lacking Phd2 alone. Protein and mRNA were isolated from the hearts of 10-week-old mice that were treated with tamoxifen at 3 weeks of age as described above. In hearts lacking Phd2 alone, we observed a modest increase in Phd3 mRNA and protein levels (Fig. (Fig.6A;6A; see also Fig. S2 in the supplemental material), consistent with the findings of our MEF studies. In addition, hearts lacking Phd2 alone exhibited a modest increase in HIF2α and a minimal increase in HIF1α (Fig. (Fig.6A,6A, lane 3), consistent with the data in our earlier report (35). HIFα levels were not increased in hearts lacking Phd3 alone (Fig. (Fig.6A,6A, lane 2). However, the accumulation of both HIF1α and HIF2α, as well as the activation of HIF-responsive mRNAs such as those for Bnip3, VEGF-A, and Pgk1, was significantly increased in hearts lacking both Phd2 and Phd3 compared to that in hearts lacking Phd2 alone (Fig. (Fig.6A),6A), consistent with the enhanced phenotype of the former compared to the latter. Interestingly, HIF2α and specific HIF-responsive mRNAs in hearts lacking pVHL still accumulated to significantly higher levels than those in hearts lacking both Phd2 and Phd3, suggesting that Phd1 may remain active in the latter setting to regulate HIF2α in this context (Fig. (Fig.6A6A).
pVHL inactivation, but not Phd2 inactivation, in the mouse liver causes hepatomegaly and hepatic steatosis (13, 35). We therefore next asked if Phd3 also plays a compensatory role in the context of Phd2 loss in the liver. Livers from 10-week-old Phd2Δ/Δ Phd3+/+ mice showed evidence of mild congestion, consistent with polycythemia, but were otherwise histologically normal (Fig. (Fig.7B),7B), consistent with findings in earlier studies. Mild steatosis was detectable, however, after oil red O staining (Fig. (Fig.7F).7F). In contrast, severe steatosis, as indicated by cytoplasmic vacuolization and oil red O accumulation, was observed in Phd2Δ/Δ Phd3−/− mice and in VhlΔ/Δ mice (Fig. 7C to D and G to H). Similar results were observed in a transgenic mouse strain expressing Cre recombinase under the control of the albumin promoter, in which the conditional alleles were inactivated specifically in the liver (Fig. 7I to P). This observation, coupled with the lack of polycythemia in the mice with liver-specific Phd2Δ/Δ Phd3+/+ and Phd2Δ/Δ Phd3−/− genotypes (data not shown), argues that the observed changes reflect a cell-autonomous effect of Phd loss on the liver.
Like the hearts, livers lacking Phd2 accumulated increased Phd3 mRNA and protein (Fig. (Fig.6B;6B; see also Fig. S2 in the supplemental material). Interestingly, the loss of Phd3 in Phd2Δ/Δ livers caused increased accumulation of HIF2α compared to that in livers lacking Phd2 alone but did not accentuate the accumulation of HIF1α (Fig. (Fig.6B).6B). Like those in hearts lacking both Phd2 and Phd3, the levels of accumulation of HIF2α, but not HIF1α, in livers lacking both Phd2 and Phd3 were still significantly lower than those in hearts lacking pVHL. In contrast to the rate of electrophoretic mobility of HIF1α from hearts lacking Phd2, however, that of HIF1α from livers lacking Phd2 (with or without Phd3) was noticeably higher than that of HIF1α from hearts lacking pVHL. Among several possibilities, this finding may reflect differential phosphorylation or differences related to prolyl hydroxylation status.
Surprisingly, the inactivation of Phd2, with or without simultaneous inactivation of Phd3, caused minimal or no increase in hepatic production of erythropoietin mRNA, in stark contrast to the inactivation of pVHL, despite the accumulation of both HIF1α and HIF2α. In contrast, other HIF-responsive mRNAs, including VEGF mRNA and Pgk1 mRNA, were induced after Phd2 inactivation in the liver, and the accumulation of such mRNAs increased further when Phd3 was concurrently inactivated (Fig. (Fig.6B).6B). This discrepancy between erythropoietin mRNA and the other HIF-responsive mRNAs may reflect a threshold phenomenon with respect to HIF activation that is target gene specific (with the erythropoietin gene having a higher threshold than, for example, the VEGF or Pgk1 gene) and/or the influence of other non-HIF transcription factors.
The absence of hepatic erythropoietin mRNA accumulation after Phd2 inactivation, together with the observed increase in circulating erythropoietin (Fig. (Fig.2),2), led us to examine the kidneys of mice lacking Phd2. Importantly, Phd2 inactivation in the kidney, in contrast to that in the liver, led to a significant increase in erythropoietin mRNA production (Fig. (Fig.6C),6C), presumably accounting for the increased circulating erythropoietin. Like that in the heart and liver, the loss of Phd2 in the kidney led to the upregulation of Phd3 protein (Fig. (Fig.6C;6C; see also Fig. S2 in the supplemental material) and concurrent inactivation of Phd3 enhanced the accumulation of HIF1α and HIF2α in kidneys lacking Phd2, although the levels of HIF2α remained near the detection limit (Fig. (Fig.6C).6C). Unexpectedly, however, the induction of some HIF-responsive mRNAs, including the erythropoietin and Pgk1 mRNAs, was not increased in kidneys lacking both Phd2 and Phd3 compared to that in kidneys lacking Phd2 alone (Fig. (Fig.6C).6C). The well-studied HIF target adrenomedullin also behaved like erythropoietin insofar as it was induced in the kidney but not the liver after Phd2 inactivation (with little additional increase upon Phd3 loss) but was induced in both organs after Vhl loss (data not shown). The situation was reversed for HIF targets such as hexokinase 2, which was sensitive to Phd inactivation in the liver but required Vhl inactivation for induction in the kidney (data not shown). Therefore, there is not a stereotypical response to Phd inactivation among HIF target genes and across tissues.
Erythropoietin mRNA accumulated in kidneys lacking Phd2 alone, despite the fact that we observed little or no accumulation of HIF1α and HIF2α. This finding suggests that increases in HIFα accumulation that are at or near the level of detection of our Western blot assay can still translate into a biological effect. On the other hand, we did not see the induction of erythropoietin in the livers lacking Phd2 (with or without Phd3), despite increases in HIF1α and HIF2α that were readily detectable by Western blot analysis. This result suggests that the amount of HIFα required to induce erythropoietin in the kidney is smaller than the amount required in the liver. A caveat, however, is that our Western blot analyses of crude kidney extracts may not be indicative of the degree of HIF stabilization in the small minority of renal cells that are capable of producing erythropoietin. Regardless of the explanation, our data clearly indicate that Phd2 inactivation is sufficient to induce renal erythropoietin production but is not sufficient, even when combined with Phd3 loss, to induce significant hepatic erythropoietin production. The latter can be achieved, however, by pVHL inactivation (Fig. (Fig.6B)6B) or by treatment with small molecules that inhibit all three PHD family members (44).
As expected, the accumulation of HIF1α and HIF2α proteins in tissues lacking Phd activity or pVHL was associated with relatively minor changes in HIF1α and HIF2α mRNA accumulation (see Fig. S3 in the supplemental material). Multiple HIF3α mRNA splice variants, including some which can act as dominant-negatives with respect to the activation of HIF target genes, have been described previously. In tissues lacking pVHL, we detected an increase in HIF3α mRNAs encoding proteins predicted to activate transcription as well as mRNAs encoding dominant-negative HIF3α (inhibitory PAS domain protein) variants. An increase in the former was also detected in some tissues lacking Phd activity (see Fig. S3 in the supplemental material). Although the significance of these findings is not yet clear, they suggest that induction of HIF3α further modulates the response to Phd or pVHL loss in vivo.
We found that the concurrent loss of Phd3 in cells and tissues lacking Phd2 leads to enhanced HIF accumulation and increased activation of HIF target genes. Moreover, concurrent loss of Phd3 exacerbates a number of the phenotypes observed in mice lacking Phd2, including hepatic steatosis, dilated cardiomyopathy, and premature mortality. Therefore, Phd3 and Phd2 are partially redundant in vivo, with Phd3 partially compensating for Phd2 loss. The loss of Phd3 alone does not measurably affect HIF accumulation in the tissues we examined, nor does it lead to the phenotypes described above, consistent with Phd2's being the primary HIF prolyl hydroxylase under normal conditions. However, Phd3 appears to be poised to modulate the HIF response when Phd2 function is compromised, such as would occur during prolonged hypoxia.
We have not yet formally proven that the pathology observed in mice lacking Phd2 and Phd3 is HIF dependent. However, the hepatopathy and cardiomyopathy observed in such mice are remarkably similar to those observed after VHL inactivation in the heart and liver. Through appropriate genetic crosses, it has been established that HIF deregulation is both necessary and sufficient for these VHL-dependent liver and cardiac alterations (23, 27, 39, 40; J. Moslehi, Y. A. Minamishima, and W. G. Kaelin, Jr., unpublished data). This fact and the knowledge that HIF is activated in mice lacking Phd2 and Phd3 strongly argue that the deregulation of HIF contributes to the pathological changes observed upon Phd inactivation in vivo.
In a previous study, it was reported that mice lacking both Phd1 and Phd3, but not mice lacking the individual family members, developed polycythemia due to increased hepatic erythropoietin production, implying cooperation between Phd1 and Phd3 (52). We have not observed increased red blood cell production, however, in Phd1−/− Phd3−/− mice with a mixed genetic background (S. Schlisio and W. G. Kaelin, Jr., unpublished data), possibly due to strain differences.
We consistently observed higher HIFα levels in cells lacking pVHL than in cells lacking Phd2 and Phd3, although the magnitude of the difference appears to be tissue and HIFα family member dependent. The simplest explanation would be that Phd1 hydroxylates HIFα in cells lacking Phd2 and Phd3. Tissue dependence presumably reflects, in part, differential expression of the three Phd family members in different cell types, while differences between HIF1α and HIF2α may reflect differences in their abilities to serve as substrates for the different Phd family members in vivo.
In addition to quantitative differences, we observed qualitative differences between the accumulation of HIF1α in cells lacking pVHL and that in cells lacking both Phd2 and Phd3. The electrophoretic mobility of HIF1α from pVHL-defective livers and kidneys was retarded compared to that of HIF1α from livers lacking both Phd2 and Phd3. Although the molecular basis for this differential remains to be elucidated, it further underscores that the loss of pVHL function may not be synonymous with the loss of Phd function with respect to HIFα activation, even were comparable levels of HIF1α to accumulate in these two settings. At a minimum, one would predict that pVHL-deficient cells would accumulate hydroxylated HIFα and that Phd-deficient cells would accumulate nonhydroxylated HIFα. The HIFα prolyl hydroxylation sites reside within one of HIFα's two transactivation domains (the NTAD) and may conceivably affect HIFα transactivation function, quantitatively or qualitatively, irrespective of changes in stability. Furthermore, the presence of pVHL may, through direct binding, alter the ability of FIH1 to inhibit the remaining HIFα transactivation domain (the CTAD) (30).
Some of these considerations likely also bear on our finding that the degree to which Phd3 compensates for Phd2 loss appears to differ among different cell types. For example, in tissues lacking Phd2, the combined loss of Phd3 further increased both HIF1α and HIF2α levels in the heart, increased only HIF2α in the liver, and increased primarily HIF1α in the kidney. Phd3 levels are particularly high in the heart (28, 43), and the results of previous siRNA experiments have suggested that Phd3 has more influence on HIF2α than HIF1α (1). The effects of losing Phd3 also differ, in a cell type-dependent manner, among different HIF target genes. For example, the combined loss of Phd2 and Phd3, in contrast to pVHL inactivation, did not activate erythropoietin in the liver, whereas Phd2 loss is apparently sufficient to activate erythropoietin in the kidney. For other targets, such as Bnip3 and VEGF, Phd2 loss causes partial activation that is increased further by concurrent Phd3 loss. Our findings suggest that different HIF target genes have distinctly different dose-response curves with respect to their transcriptional induction in response to increases in HIFα protein levels and that these dose-response curves are also profoundly influenced by cell type. The shapes of these dose-response curves are probably determined by a number of factors, including the effects of HIFα asparaginyl (and perhaps prolyl) hydroxylation on HIF transactivation function, the requirement for HIF1α compared to HIF2α, and the effects of cis-acting non-HIF transcription factors.
Concurrent Phd3 loss exacerbated the hepatic and cardiac phenotypes but not the hematological phenotype in mice lacking Phd2. In particular, this finding argues that the cardiac phenotype is not merely a secondary consequence of chronic volume overload and hyperviscosity. Morever, we observed the loss of mitochondria in Phd-deficient and pVHL-deficient hearts, as well as increased transcription of Bnip3, which has been implicated in mitochondrial autophagy (14, 46, 48, 55). These data are consistent with Phd loss causing cardiomyopathy because of a cardiomyocyte-specific role for Phd in the control of mitochondrial autophagy. Our preliminary data obtained using cardiac cell-specific Phd inactivation are consistent with this idea (J. Moslehi, Y. A. Minamishima, and W. G. Kaelin, Jr., unpublished data), as is the recent finding that cardiac cell-specific pVHL inactivation causes HIF-dependent dilated cardiomyopathy (27).
Small organic molecules that inhibit HIFα prolyl hydroxylases are being developed for the treatment of anemia (18, 44) and ischemic diseases (42, 49). However, to what extent these drugs inhibit the three individual PHD family members, as well as perhaps other 2-oxoglutarate- and iron-dependent dioxygenases such as FIH1, is not completely known. Our studies suggest that the inhibition of Phd2 would suffice to promote red blood cell production. It remains to be determined whether other or additional Phd family members, perhaps in conjunction with FIH1, will need to be inhibited to achieve maximal benefits for other indications such as ischemic tissue preservation. Notably, circulating VEGF levels were not increased in mice lacking Phd2, even when the lack of Phd2 was combined with the loss of Phd3. The differential behavior of erythropoietin and VEGF genes in Phd-deficient mice suggests that it will be possible to pharmacologically activate HIF and subsets of HIF target genes without necessarily causing a marked increase in VEGF, which would theoretically promote the development of hypotension, tissue edema, and increased angiogenesis.
The results of our studies further suggest that chronic treatment with prolyl hydroxylase inhibitors may lead to hepatic steatosis and cardiomyopathy. A caveat here is that these inhibitors, when used in the clinic, will almost certainly lead to partial, intermittent loss of Phd function whereas our genetic model achieves continuous, complete loss of Phd function in cells that have undergone successful recombination of the floxed Phd2 allele. It remains to be determined whether the threshold of Phd2 inhibition required for therapeutic effects will substantially differ from the threshold required for the development of hepatic or cardiac pathology. Moreover, preclinical models suggest that acute HIF stabilization would be beneficial in the setting of acute ischemic insults, such as occur with myocardial infarction (22, 38) or stroke (4, 50), even if chronic HIF stabilization proves to be deleterious. Mouse models such as those described here should be powerful tools to study the potential advantages and disadvantages of pharmacologically manipulating HIF activity in vivo.
We thank members of the Kaelin laboratory for helpful discussions, Regeneron Pharmaceuticals for Phd3−/− mice, and Volker Haase for VHLF/F mice.
J.M. was supported by an NIH Clinical Research Training Program (T32) for cardiovascular medicine. W.G.K. was supported by NIH, is a Howard Hughes Medical Institute (HHMI) Investigator, and is a Doris Duke Distinguished Clinical Scientist. W.G.K. has a financial interest in Fibrogen, Inc., which is developing prolyl hydroxylase inhibitors as drugs.
Published ahead of print on 31 August 2009.
†Supplemental material for this article may be found at http://mcb.asm.org/.