The transcription factor HIF (hypoxia-inducible factor) mediates a highly pleiotrophic response to hypoxia. Many recent studies have focused on defining the extent of this transcriptional response. In the present study we have analysed regulation by hypoxia among transcripts encoding human Fe(II)- and 2-oxoglutarate-dependent oxygenases. Our results show that many of these genes are regulated by hypoxia and define two groups of histone demethylases as new classes of hypoxia-regulated genes. Patterns of induction were consistent across a range of cell lines with JMJD1A (where JMJD is Jumonji-domain containing) and JMJD2B demonstrating robust, and JMJD2C more modest, up-regulation by hypoxia. Functional genetic and chromatin immunoprecipitation studies demonstrated the importance of HIF-1α in mediating these responses. Given the importance of histone methylation status in defining patterns of gene expression under different physiological and pathophysiological conditions, these findings predict a role for the HIF system in epigenetic regulation.
- histone demethylase
- hypoxia-inducible factor 1 (HIF-1)
- Jumonji domain
- transcription factor
Hypoxia is a major component of many human diseases and induces a range of cellular and systemic responses that function to restore oxygen homoeostasis or better adapt cells to the hypoxic environment. At the cellular level these responses include regulation of fundamental physiological processes such as energy metabolism, proliferation, apoptosis, motility and cell-differentiation (for a review see [1,2]).
Important insights into these processes have been gained though the definition of hypoxia-signalling pathways that regulate gene expression in accordance with cellular oxygen availability. HIF (hypoxia-inducible factor) is an α/β heterodimeric transcription factor that binds DNA at HREs (hypoxia-response elements) associated with its transcriptional target genes . HIF itself is regulated by post-translational hydroxylation of specific residues in its α subunits by a set of Fe(II)- and 2OG (2-oxoglutarate)-dependent oxygenases [3,4]. In the presence of oxygen, hydroxylation [catalysed by three closely related prolyl hydroxylase domain enzymes (termed PHD1, 2 and 3)] at two proline residues promotes binding of HIF-α subunits to the pVHL (von Hippel–Lindau protein) tumour suppressor E3 ubiquitin ligase leading to proteasomal degradation (for review see ). In hypoxia, these reactions are suppressed allowing the assembly of an active HIF complex and induction of a range of genes that respond directly or indirectly to HIF.
Previous work has sought to better define the extent of this transcriptional cascade system, with studies demonstrating both an extensive set of direct HIF transcriptional target genes, and indirect effects through the induction of regulatory molecules that themselves influence gene expression [6,7]. Pan-genomic assays of transcript abundance have emphasized the overall importance of the HIF hydroxylase system in defining patterns of hypoxia-inducible gene expression and have revealed many new and important hypoxia-inducible genes [8–10]. In the main, such analyses have focused on defining statistically robust induction of specific genes rather than interrogating patterns of response among specific families encoding functionally or structurally related proteins. Of particular interest in this respect is the 2OG-dependent oxygenase superfamily itself. In addition to the 2OG-dependent oxygenases that function in cellular oxygen sensing, the family encompasses members with diverse functions including post-translational modification of proteins (e.g. collagen modification), fatty acid metabolism and DNA repair [11–13]. Since all 2OG-dependent oxygenases have an obligatory requirement for dioxygen as a co-substrate, the activity of some of these enzymes in cells may be affected by the extracellular oxygen status. Several 2OG-dependent oxygenases, including two of the HIF hydroxylases have been identified as hypoxia-inducible gene products [14–17].
Bioinformatic analyses have predicted that the human genome encodes in excess of sixty 2OG-dependent dioxygenases. Although the functions of many remain unassigned, a number have recently been demonstrated to function in histone modification, particularly histone lysine and arginine demethylation [18,19].
To investigate patterns of induction by hypoxia we compiled a refined list of human 2OG-dependent oxygenases, reanalysed microarray-based transcriptome profiling of the MCF7 breast cancer cell line in normoxic and hypoxic conditions , and tested the role of the HIF system in such responses using a series of genetic interventions on the HIF pathway. The results define genes encoding the JmjC-domain-containing histone demethylases as a new class of gene regulated by the HIF system and demonstrate consistent patterns of induction by HIF-1α among different family members.
Human osteosarcoma (U2OS), human breast cancer (MCF7) and human cervical carcinoma (HeLa) cells were cultured in DMEM (Dulbecco's modified Eagle's medium; Invitrogen) supplemented with 10% (v/v) FCS (fetal calf serum; Sigma). Human neuroblastoma cells (IMR 32) were cultured in Eagle's minimal essential medium (Invitrogen) with 10% (v/v) FCS, 2 mM glutamine (Invitrogen) and non-essential amino acids. Human promyelocytic leukemia (HL60) cells were grown in RPMI (Invitrogen) with 15% (v/v) FCS. All cells were maintained in a humidified atmosphere of 5% CO2 and 21% O2. Both R38 [human renal carcinoma (RCC4) cells that express a tetracycline-inducible HA (haemagglutinin)-tagged VHL transgene)] and C29 (U2OS cells expressing tetracycline-inducible HIF-1α) have been previously described [20,21] and were cultured in DMEM with 10% (v/v) tetracycline-free FCS (BD Bioscience) supplemented with 1 μg/ml G418 and 5 μg/ml blasticidin (Sigma). In the experimental work, cells were cultured under normoxic conditions (5% CO2 and 21% O2) or in hypoxic conditions (16 h at 5% CO2 and 0.5% O2). Where specified, cells were treated with DMOG (dimethyloxalylglycine; 1 mM for 16 h) or DFO (desferrioxamine; 100 μM for 16 h).
siRNA (small interfering RNA) treatment of cells
MCF7 and U2OS cells were seeded at 30% confluency and grown in normoxic conditions. Cells were transfected with siRNA (20 nM) using Oligofectamine (Invitrogen) according to the manufacturer's protocol. After 24 h, the cells were transfected again and exposed to hypoxia for 16 h after which they were washed, lysed, and RNA and protein was extracted. The siRNA oligonucleotide sequences used were as previously described .
RNA extraction and cDNA synthesis
Total RNA was extracted using the Sigma Total RNA kit (Sigma) according to the manufacturer's protocol. First strand cDNA synthesis was generated from 5 μg of total RNA (GE Healthcare).
Protein extraction and immunoblot analysis
Preparation of cell extracts and immunoblot analyses were performed as described . Primary antibodies used were mouse anti-(HIF-1α) (BD Transduction Laboratories), rabbit anti-(HIF-2α) (Novus Biologicals), rabbit anti-JMJD1A (Abcam), mouse HRP (horseradish peroxidase)-conjugated anti-HA (Dako) and mouse HRP-conjugated anti-actin (Abcam).
ChIP (chromatin immunoprecipitation) assays
ChIP assays were performed using a modified version of the Upstate protocol (Millipore). In brief, MCF7 cells were incubated with 2 mM DMOG for 16 h to increase HIF-α levels. DNA-binding proteins were cross-linked to DNA using formaldehyde at a final concentration of 1% (w/v) for 10 min at 25 °C, followed by treatment with glycine (125 mM) for a further 5 min. Cells were washed in PBS, lysed in SDS lysis buffer, and sonicated (Sonics & Materials, VCX 500). The supernatant was collected by centrifugation (15000 g for 10min at 4 °C) and pre-cleared with protein A–agarose beads (Millipore). Chromatin was then incubated overnight with rabbit polyclonal anti-sera to HIF-1α (PM14) and HIF-2α (PM9)  before adding protein A–agarose for a further 1 h. Pre-immune serum was used as a negative control. The beads were washed, and immunoprecipitated complexes were eluted into 1% (w/v) SDS and 0.1 M sodium bicarbonate elution buffer. Cross-linking was reversed by overnight incubation at 65 °C and followed by protein digestion with proteinase K. DNA was recovered by phenol/chloroform extraction and ethanol precipitation.
RT-Q-PCR (real-time quantitative PCR)
RT-Q-PCR for mRNA quantification employed Taqman gene expression assays on a StepOne thermocycler (Applied Biosystems). Normalization was to β-actin mRNA and relative gene expression was calculated using the ΔΔCT method. Per reaction, 50 ng of cDNA template was used and three biological replicates, each in triplicate, were performed for each experiment. For quantification of HIF-binding sites, ChIP DNA (5 ng) from each of input, pre-immune sera, PM14 and PM9 was subjected to RT-Q-PCR using oligonucleotides designed to amplify putative HRE consensus sequences (See Supplementary Table S1 at http://www.BiochemJ.org/bj/416/bj4160387add.htm). The fold-enrichment of each HRE in the DMOG-treated cells was determined using the ΔΔCT method.
Statistical comparisons were performed using a Student's unpaired t test.
The human genome encodes multiple sub-groups of 2OG-dependent oxygenases that are regulated by hypoxia
Using structurally informed sequence analysis coupled to functional knowledge, we grouped known or predicted human 2OG-dependent oxygenases into subfamilies, and defined the existence or otherwise of regulation by hypoxia by reference to comparative whole-genome expression arrays  conducted on normoxic and hypoxic MCF7 cells (Table 1). The data indicated that some, but not all, transcripts are substantially up-regulated in hypoxic cells. Reported patterns of induction by hypoxia were confirmed for several subfamilies of enzymes, including the HIF PHDs, some of the pro-collagen PHDs and the PLODs (pro-collagen lysyl hydroxylases) [4,14–17]. In addition, several other classes of 2OG-dependent oxygenases contained members that manifest significant up-regulation by hypoxia. Particularly striking was the up-regulation of transcripts encoding the Jumonji-domain-containing proteins JMJD1A and JMJD2B. These proteins belong to the JMJD1 (otherwise known as JHDM2 and containing JmjC and modified zinc-finger domains) and JMJD2 (otherwise known as JHDM3 and containing JmjC, JmjN, plant homology domains and Tudor domains) subfamilies of 2OG-dependent oxygenases that have been demonstrated to possess HDM (histone lysine demethylase) activity [18,25,26]. To better define this new class of hypoxia-regulated gene we performed further experiments on seven members (JMJD1A–C and JMJD2A–D) of the 2OG oxygenase superfamily that have been definitively assigned as possessing HDM activity.
First, to verify the results of the microarray analysis and to determine whether regulation by hypoxia was manifested in different cell types, expression of all seven transcripts was analysed by RT-Q-PCR in a panel of cell lines. MCF7, U2OS, HeLa, IMR32 and HL60 cells were exposed to hypoxia (0.5% O2 for 16 h). Consistent patterns of induction by hypoxia were observed across all five cell lines (Figure 1A). The most striking induction by hypoxia was observed for JMJD1A (ranging from 2.6–4.7-fold) and JMJD2B (ranging from 2.5–4.6-fold) mRNA (Figure 1A). Other family members did not manifest induction by hypoxia, with the exception of JMJD2C which manifested modest (up to 2-fold) induction in some cell types.
Regulation of genes encoding Jumonji-domain-containing proteins JMJD1 and JMJD2 by the HIF hydroxylase pathway
Inspection of the microarray data indicated that 2OG oxygenase genes manifesting induction by hypoxia were also commonly induced by DMOG, a cell-penetrating competitive inhibitor of many 2OG oxygenases including the hydroxylases that regulate HIF . To pursue the potential role of HIF hydroxylase pathways in the regulation of genes encoding JMJD1 and JMJD2 family members, we extended this analysis to other cell types, and tested responses to both DMOG (1 mM for 16 h) and DFO (100 μM for 16 h), agents that up-regulate HIF by inhibition of 2OG oxygenases, including the HIF hydroxylases . Results are shown for U2OS cells (Figures 1B and 1C). As expected we observed robust up-regulation of HIF-1α by immunoblot analysis that equalled or exceeded that achieved by hypoxic exposure (Figure 1B). RT-Q-PCR analysis of JMJD1 and JMJD2 transcripts demonstrated substantial up-regulation, the pattern being closely similar to that observed for hypoxia (Figure 1C). JMJD1A, JMJD2B and JMJD2C were again the most strongly induced genes, with responses to DMOG exceeding 20-fold induction for JMJD2B (Figure 1C). Assays of the effects on endogenous JMJD protein levels are limited by the availability of appropriate antibodies. Nevertheless we were able to confirm that JMJD1A protein levels were increased significantly both under hypoxia and with the addition of DMOG (Figure 1D).
To test the involvement of HIF in these responses we next analysed cells bearing different genetic manipulations of the HIF system. First, we analysed changes in expression in response to normoxic overexpression of HIF-1α using the C29 cell line, a stable U2OS transfectant bearing a doxycycline-inducible HIF-1α transgene . Exposure of normoxic C29 cells to doxycycline resulted in sustained induction of HIF-1α (Figure 2Ai) which was associated with sustained, although more moderate, induction of JMJD1A, JMJD2B and JMJD2C over the 48 h period of doxycycline exposure (Figure 2Aii).
We then analysed responses in R38 cells, a VHL-defective RCC (RCC4) transfectant that re-expresses a doxycycline-inducible HA-tagged wild-type VHL transgene . In keeping with baseline VHL-defective status, R38 cells expressed increased levels of HIF-1α subunits in normoxia, whereas treatment with doxycycline over a 48 h period resulted in sustained expression of HA–VHL and reduction of HIF-1α (Figure 2Bi). Consistent with a role for HIF in the regulation of the JMJD1 and JMJD2 gene families, down-regulation of HIF-1α was associated with down-regulation of JMJD1 and JMJD2 transcripts in a pattern that was the inverse of that observed for hypoxia and induction of HIF-1α in that JMJD1A, JMJD2B and JMJD2C showed the greatest reduction in expression (Figure 2Bii).
Finally we tested the role of HIF-1α and HIF-2α isoforms using siRNA-mediated knockdown in hypoxic MCF7 cells. As indicated in Figure 2(Ci), MCF7 cells expressed both HIF-1α and HIF-2α; siRNAs directed against each HIF-α isoform, but not control siRNA, resulted in specific knockdown of the target. Analysis of JMJD transcripts from hypoxic cells indicated that siRNA directed against HIF-1α substantially reduced expression of JMJD1A, JMJD2B and JMJD2C, whereas siRNA directed against HIF-2α had essentially no effect (Figure 2Cii).
Binding of HIF-α subunits to the promoters of JMJD1A and JMJD2B
Taken together, these results indicate that induction of JMJD1A, JMJD2B and JMJD2C by hypoxia is mediated directly or indirectly by HIF-1α. Direct transcriptional activation at HIF target gene loci is mediated by binding of the HIF-α/β complex to HREs containing the core motif RCGTG . To identify such sites at loci encoding JMJD1 and JMJD2 family members, we searched the March 2006 Human Genome assembly (http://genome.ucsc.edu/) for HIF-binding sites that were highly conserved among mammalian species, within the regions (−5 to +0.5 kB) of the assigned transcriptional start site. Three potential HIF-binding sites within these regions were identified in the three members that were demonstrated to be induced in hypoxia by HIF-1α (JMJD1A, JMJD2B and JMJD2C) and no such sites in the remaining four members (JMJD1B, JMJD1C, JMJD2A and JMJD2D), suggesting that the three hypoxia-inducible genes might be direct target genes.
To test whether these loci did indeed bind HIF-α subunits, we performed ChIP assays using anti-(HIF-1α) antibodies (PM14) or anti-(HIF-2α) antibodies (PM9) or control pre-immune sera, and carried out RT-Q-PCR analysis of precipitated DNA using primers designed to amplify each of the predicted HREs. To provide further controls, primers were also designed to a region of the JMJD1A promoter that does not contain an HRE, and to the JMJD2A promoter that does not contain an HRE and did not exhibit regulation by HIF or hypoxia in the transcript analyses. Significant and reproducible enrichment of DNA containing HREs at the JMJD1A and JMJD2B loci was observed with both anti-(HIF-1α) and anti-(HIF-2α) antibodies (Table 2). More moderate enrichment was observed at the JMJD2C locus with anti-(HIF-1α) but not anti-(HIF-2α). No enrichment of the non-HRE-containing amplicons was detected with either antibody.
The present study has defined genes encoding Jumonji-domain-containing 2OG-dependent oxygenases, with histone demethylase activity, as a new class of HIF-responsive hypoxia-inducible genes. We focused our analysis on two groups of closely related genes; those encoding JMJD1A–C proteins that contain only the catalytic JmjC domain, and JMJD2A–D that contain both JmjN and JmjC domains (as well as Tudor and plant homology domains in the case of JMJD2A–C) . Among these families, we observed consistent patterns of regulation with genes encoding JMJD1A, JMJD2B and JMJD2C, but not other members, being inducible by hypoxia. Although JMJD1A and JMJD2B showed particularly striking regulation by hypoxia, prompting us to focus on the JMJD groups, a range of other genes encoding human 2OG-dependent oxygenases showed hypoxia-inducible behaviour in the MCF7 cell gene expression array (Table 1). Consistent with previous reports, regulation by hypoxia was observed for the HIF hydroxylases PHD2 and PHD3 [4,16,17], and for some members of the procollagen PHD and PLOD groups [14,15]. However, and also consistent with previously reported data [4,20], PHD1 and FIH (factor inhibiting HIF), an asparaginyl hydroxylase, which like PHD2 and PHD3 are proposed to act as sensing enzymes in the HIF system, are not, or much less significantly, regulated by hypoxia. Modest regulation by hypoxia was also noted in the MCF7 cell gene expression array for other classes of Jumonji-domain-containing proteins including members of the JARID (Jumonji AT-rich interactive domain) and UTX/Y (ubiquitously transcribed tetratricopeptide repeat containing, X/Y chromosome) groups. Several of these proteins also have histone demethylase activity so that it is possible that regulation by hypoxia extends more widely across this group of enzymes.
Dynamic control of histone methylation status is proposed to regulate chromatin assembly and gene expression [18,26]. Histone lysine methylation is currently known to occur on H3 (histone 3) at Lys4, Lys9, Lys27, Lys36 and Lys79, and on H4 at Lys20; each site potentially existing as mono-(me1), di-(me2), or tri-(me3) methylated forms associated with distinct biological functions [18,26–28]. Although the patterns of substrate specificity are as yet incompletely defined, some HDMs are reported to be highly site-specific. For instance JMJD1A has so far been shown to exhibit demethylation activity on H3K9me1 and H3K9me2, but not H3K9me3 ; JMJD2B exhibits poor enzyme activity in vitro, but has been demonstrated to specifically reduce H3K9me3 when overexpressed in cells ; JMJD2C exhibits demethylation activity on both H3K9me3 and H3K36me3 . Interestingly, specific HDMs have been associated with the regulation of particular transcriptional pathways. For instance JMJD1A is associated with the AR (androgen receptor) and has been shown to positively regulate certain AR target genes [29,32]. In embryonic stem cells, JMJD1A and JMJD2C enhance the expression of genes associated with self-renewal and are targets of the transcription factor Oct-4 . This suggests that expression levels of specific histone demethylases may themselves be regulated as components of physiological control systems directing patterns of gene expression. Interestingly, Oct-4 has itself been identified as a gene responding specifically to HIF-2α in embryonic stem cells .
Our studies, in a range of different cell lines, have identified a different and more direct connection between histone demethylases and the HIF system. Pharmacological and genetic analyses demonstrated functional control of particular JMJD-encoding genes by HIF, specifically by HIF-1α, whereas ChIP assays defined HIF-α-binding sites in the promoters of these hypoxia-inducible genes. Interestingly, although ChIP assays indicated that both HIF-1α and HIF-2α bound to promoter sequences, siRNA-mediated suppression suggested that HIF-1α, but not HIF-2α, was responsible for induction of JMJD1A and JMJD2B in hypoxia, even though both HIF-α isoforms were induced under the experimental conditions. Such behaviour has been noted at certain other HIF target gene loci [22,35], and is consistent with recent data indicating that HIF-α isoform transcriptional selectivity is mediated by post-DNA binding mechanisms conveyed by sequences distal to the DNA binding and dimerization domains [24,35]. The nature of these mechanisms, together with the reasons why endogenous HIF-2α appears transcriptionally inactive at many loci, is incompletely understood, although evidence has been provided for the existence of a titratable repressor that limits HIF-2α activity in many cell types .
Although hypoxia has long been recognized to have major effects on genome integrity, gene expression and cell differentiation, which probably involve epigenetic controls, detailed analyses of hypoxic effects on epigenetic modification of DNA and histones have so far been limited. However, hypoxia has been noted to alter levels of both histone acetylation and histone methylation at certain sites [36,37]. A recent study demonstrated that changes in both histone methyltransferase and histone demethylase activity contributed to changes in global H3K9 methylation status in response to hypoxic stress and focused on the up-regulation of the histone methyltransferase G9a in hypoxia by both HIF-dependent and HIF-independent mechanisms .
The current evidence for direct regulation of genes encoding specific histone demethylases by HIF-1α and that in a recently published report describing up-regulation of JMJD1A by HIF-1α  suggests another means by which activation of the HIF pathway may contribute to the modulation of histone marks and gene expression profiles in hypoxic cells. Given the existence of several different mechanisms by which HIF and hypoxia itself may alter histone methylation, the difficulty of separating these effects accurately, and uncertainties regarding the extent to which transcriptional up-regulation of JMJDs is reflected in biological activity, it is currently difficult to predict effects of the present findings on overall cellular demethylase activity. Since the JMJD1 and JMJD2 enzymes are 2OG-dependent oxygenases that have an absolute requirement for dioxygen as a co-substrate it could be argued that any increases in enzyme abundance in hypoxia simply compensate for reduced activity. However, given the specificity of the HIF-mediated hypoxia-inducible response for particular HDMs with different methyl-lysine substrate preferences, and different biological effects at particular transcriptional loci, this appears unlikely. Rather, we propose that this new class of HIF transcriptional target adds another level of control that shapes the overall response to hypoxia. The present findings raise interesting questions as to whether the hypoxia-inducible histone demethylases are specifically associated with the HIF transcriptional complex or with specific classes of hypoxia-inducible target gene loci and, if so, how they function to modulate patterns of gene expression under conditions of varying oxygen availability.
We are grateful to Cancer Research U.K., the Biotechnology and Biological Sciences Research Council and the Wellcome Trust for the support of this research. C. L. is in receipt of a Rhodes Scholarship. We thank Matthew Cockman, Ya-Min Tian, Mathew Coleman, Nicolas Granatino and Chris Pugh for their contributions to the work.
Abbreviations: ChIP, chromatin immunoprecipitation; DFO, desferrioxamine; DMEM, Dulbecco's modified Eagle's medium; DMOG, dimethyloxalylglycine; FCS, fetal calf serum; FIH, factor inhibiting HIF; HA, haemagglutinin; HDM, histone lysine demethylase; HIF, hypoxia-inducible factor; HRE, hypoxia-response element; HRP, horseradish peroxidase; JMJD, Jumonji-domain containing; 2OG, 2-oxoglutarate; PHD, prolyl hydroxylase domain; PLOD, pro-collagen lysyl hydroxylase; RT-Q-PCR, real-time quantitative PCR; siRNA, small interfering RNA; (p)VHL, von Hippel–Lindau protein
- © The Authors Journal compilation © 2008 Biochemical Society