Report

Hypoxia induces rapid changes to histone methylation and reprograms chromatin

See allHide authors and affiliations

Science  15 Mar 2019:
Vol. 363, Issue 6432, pp. 1222-1226
DOI: 10.1126/science.aau5870

Oxygen sensing revisited

The cellular response to hypoxia (oxygen deficiency) is a contributing factor in many human diseases. Previous studies examining the way in which hypoxia alters gene expression have focused on oxygen-sensing enzymes that regulate the activity of a transcription factor called hypoxia-inducible factor (see the Perspective by Gallipoli and Huntly). Chakraborty et al. and Batie et al. now show that hypoxia can also affect gene expression through direct effects on chromatin regulators. Certain histone demethylases, such as KDM6A and KDM5A, were found to be direct sensors of oxygen. In cell-culture models, hypoxia diminished the activity of these enzymes and caused changes in the expression of genes that govern cell fate.

Science, this issue p. 1217, p. 1222; see also p. 1148

Abstract

Oxygen is essential for the life of most multicellular organisms. Cells possess enzymes called molecular dioxygenases that depend on oxygen for activity. A subclass of molecular dioxygenases is the histone demethylase enzymes, which are characterized by the presence of a Jumanji-C (JmjC) domain. Hypoxia can alter chromatin, but whether this is a direct effect on JmjC-histone demethylases or due to other mechanisms is unknown. Here, we report that hypoxia induces a rapid and hypoxia-inducible factor–independent induction of histone methylation in a range of human cultured cells. Genomic locations of histone-3 lysine-4 trimethylation (H3K4me3) and H3K36me3 after a brief exposure of cultured cells to hypoxia predict the cell’s transcriptional response several hours later. We show that inactivation of one of the JmjC-containing enzymes, lysine demethylase 5A (KDM5A), mimics hypoxia-induced cellular responses. These results demonstrate that oxygen sensing by chromatin occurs via JmjC-histone demethylase inhibition.

Hypoxia is important for physiological and pathological processes in humans (1, 2). Hypoxia activates a specific transcriptional program in cells that allows them to adapt to lower oxygen levels (3, 4). Hypoxia-inducible factors (HIFs), a family of transcription factors, coordinate the majority of these transcriptional changes, with additional input from other transcription factors (3). Hypoxia activation of HIFs is mediated via the inhibition of dioxygenases, most prominently prolyl-hydroxylases (PHDs) and factor-inhibiting HIF (FIH) (5). These enzymes require molecular oxygen as a cofactor, as well as iron and 2-oxoglutarate, for activity. Among such dioxygenases are the Jumonji-C (JmjC) domain–containing histone demethylases (6, 7). The oxygen dependency for only a few of the JmjC enzymes has been measured, and for these, their Km for oxygen is similar to that of PHD enzymes in vitro (8, 9), suggesting that JmjC enzymes can act as molecular oxygen sensors in the cell. A number of studies have shown that histone methylation marks are increased in a variety of mammalian cells exposed to severe and prolonged hypoxia (7, 10, 11).

To investigate if JmjC domain-containing histone demethylases have the potential of being molecular oxygen sensors in the cell, we analyzed histone methylation changes after short periods of hypoxia. We found that hypoxia induced a rapid and robust increase in histone methylation marks on histone-3 (H3) in HeLa cells and the human fibroblast cell line HFF (Fig. 1A and fig. S1A). Many of the changes observed in H3 methylation preceded robust stabilization of HIF-1α, occurring within 30 min of hypoxia exposure (Fig. 1A). This was also observed when cells were treated with a noncleavable analog of 2-oxoglutarate [dimethyloxaloylglycine (DMOG)] or with an iron chelator (desferroxamine) (fig. S1, B and C). Hypoxia-induced histone methylation was also observed using acid extraction of histones (fig. S2, A and B) and quantitative immunofluorescence for H3 lysine-4 trimethylation (H3K4me3) and H3K36me3 (Fig. 1B and fig. S3A). To investigate if these results were HIF dependent, we depleted cells of HIF-1β by using small interfering RNA (siRNA) (Fig. 1C). After 1 hour of hypoxia, there was no reduction in histone methylation, demonstrating that HIF is not required for the changes in hypoxia. Increased levels of histone marks in the absence of HIF could be due to reduced levels of several JmjC histone demethylases, because these enzymes are known to be direct targets of the HIF-1 complex (7, 12). HIF independence was also confirmed by using the selective von Hippel-Lindau (VHL) inhibitor VH298 (13), which activates HIF without altering functions of PHDs or other dioxygenases in cells (Fig. 1D), and in cells constitutively expressing active HIF-1α and/or HIF-2α (fig. S4, A to C). These data indicate that hypoxia and other dioxygenase inhibitors induce rapid changes to several histone methylation marks independently of HIF.

Fig. 1 Hypoxia increases histone methylation independently of HIF.

(A) Immunoblot analysis of the indicated proteins in HeLa and HFF cells exposed to 1% O2 for the indicated times. (B) Immunofluorescence analysis of H3K4me3 in HeLa cells with or without exposure to 1% O2 for 1 hour. The graph represents individual cell measurements, with means ± SEM (N = 3). au, arbitrary units. ***P < 0.001. (C and D) Immunoblot analysis of the indicated proteins in HeLa cells treated with VH298 for the indicated time points, with or without exposure 1% O2 for 24 hours (C) and with or without exposure to 1% O2 for 1 hour after siRNA depletion of HIF-1β (D).

Prolonged hypoxia has also been associated with changes in reactive oxygen species (ROS) (14) and metabolism (15). We thus determined whether short-term hypoxia induced ROS in our cell system (fig. S5A) and whether ROS was responsible for altering the histone methylation patterns in cells (fig. S5B). Despite a small increase in ROS in HeLa cells exposed to 1 hour of hypoxia (fig. S5A), pretreatment with the ROS scavenger N-acetyl-cysteine did not reduce hypoxia-induced increases in histone methylation (fig. S5B). Cells treated with high levels of hydrogen peroxide, to mimic ROS production, displayed no changes in histone methylation marks after 1 hour but exhibited increases after 24 hours (fig. S5C). These data suggest ROS is not involved in increasing histone methylation marks observed at 1 hour of hypoxia exposure but might contribute to increases in histone methylation after prolonged hypoxia.

Changes in 2-oxoglutarate–related metabolites, such as succinate and fumarate and the oncometabolite 2-hydroxyglutarate (2-HG), have also been shown to impact histone methylation (15, 16). To determine if short-term hypoxia–induced changes in histone methylation can be mimicked by changes in metabolites or oncometabolites, we treated HeLa cells with succinate, fumarate, or cell-permeable 2-HG (fig. S6, A to C). Whereas no changes in histone methylation marks were detected after treatment with succinate (fig. S6A), fumarate could induce some increases in histone methylation (fig. S6B). Treatment with 2-HG (5 mM) increased methylation marks on histone H3 only after prolonged treatment in HeLa cells (fig. S6C). However, this was not seen in HFF cells, where 2-HG treatment only altered H3K4me3 and H3K27me3 after prolonged exposure (fig. S6C). Finally, we measured levels of 2-HG in HeLa and HFF cells exposed to 1 hour or 24 hours of hypoxia; we also used HT1080 cells, which possess an isocitrate dehydrogenase 1 mutation (17), as a positive control for high levels of 2-HG (fig. S6D). Hypoxia did not alter 2-HG levels in HeLa or HFF cells (fig. S6D), suggesting that this oncometabolite is not involved in the mechanism that controls the observed changes to the histone marks.

The analysis of H3 methylation suggests that hypoxia induces global changes to histone methylation marks. We thus performed chromatin immunoprecipitation followed by deep sequencing (ChIP-sequencing) to evaluate histone modifications associated with active gene transcription, namely, H3K4me3 and H3K36me3, in HeLa cells at normoxia or after 1 hour of hypoxia exposure. Analysis of H3K4me3 data (fig. S7, A to C, and dataset S1) identified 12,000 shared peaks between normoxia and hypoxia (Fig. 2A). Our dataset is of the same standard as those produced by the ENCODE consortium (fig. S7, D and E). Hypoxia exposure led to 164 up-regulated and 455 down-regulated peaks (Fig. 2B and dataset S1). Genomic location of H3K4me3 peaks revealed similar distributions in normoxia and hypoxia (fig. S7F). Whereas hypoxia–down-regulated peaks for H3K4me3 were predominantly located at promoters, hypoxia–up-regulated peaks displayed increased occupancy at gene body and intergenic regions (fig. S7F). Intergenic H3K4me3 peaks upregulated in hypoxia mapped mostly to predicted enhancers (fig. S8A). Some of the genes associated with these enhancers were hypoxia inducible in HeLa cells (fig. S8B).

Fig. 2 Hypoxia induces specific changes in H3K4me3 at hypoxic gene signatures.

We performed ChIP-sequencing analysis of H3K4me3 in HeLa cells with or without exposure to 1% O2 for 1 hour (1 h 1% O2 or 21% O2, respectively). (A) Overlap of peaks in each condition. (B) Up-regulated and down-regulated peaks and genes (genes with peaks) in response to 1 hour of 1% O2 exposure. (C) Gene group association analysis showing significant enrichment of gene set signatures (MsigDB) for up-regulated genes in (B). (D) Gene group association analysis of the differentially regulated genes identified in (B), with the indicated datasets. (E) ChIP-qPCR analysis (means ± SEM, N = 3) of H3K4me3 at ChIP-sequencing peaks for the indicated genes in HeLa cells exposed to 1% O2 for the indicated times. IgG, immunoglobulin G. **P < 0.01; ***P < 0.001.

Pathway association analysis using the molecular signatures database (MSigDB) (18, 19) showed that genes with hypoxia–up-regulated H3K4me3 peaks displayed a significant association of hypoxia signaling and epithelial-to-mesenchymal transition signatures (Fig. 2C). Genes with hypoxia–down-regulated H3K4me3 were mostly associated with cell division and oxidative phosphorylation (fig. S9A).

We performed integrative analysis with genes whose expression is differently regulated in response to 16 hours of hypoxia in HeLa cells, as determined by RNA sequencing (RNA-seq), genes that are induced or repressed in response to hypoxia in a conserved manner across multiple cell types and hypoxia exposures (20), and validated HIF target genes (Fig. 2D and dataset S1). Genes with hypoxia–up-regulated H3K4me3 peaks were significantly enriched for HIF targets, cell type-conserved hypoxia-induced genes and the nonredundant combination of hypoxia-induced and HIF target genes (Fig. 2D). Furthermore, H3K4me3 up-regulation at these genes was found almost exclusively at promoters (dataset S1). Although some overlap was found for HeLa hypoxia-induced genes and genes with hypoxia–up-regulated H3K4me3 peaks, there were also HeLa cell hypoxia-inducible genes with H3K4me3 hypoxia–down-regulated peaks (Fig. 2D). Thus, whereas H3K4me3 up-regulation at genes in early hypoxia correlates with hypoxia gene induction, the dynamic nature of gene expression in hypoxia may account for hypoxia–down-regulated H3K4me3 at genes induced after 16 hours of hypoxia in HeLa cells.

Analysis of the H3K36me3 ChIP-sequencing dataset (figs. S10, A to F, and S11, A to E) revealed significant enrichment of genes with H3K36me3 hypoxia–up-regulated peaks for HIF targets (fig. S11E and dataset S1) and significant enrichment of genes with H3K36me3 hypoxia–down-regulated peaks for cell type-conserved hypoxia-repressed genes (fig. S11E and dataset S1). ChIP followed by quantitative polymerase chain reaction (ChIP-qPCR) analysis revealed a marked increase in H3K4me3 at all the predicted genes investigated after 1 hour of hypoxia exposure, with further increases seen at 24 hours of hypoxia (Fig. 2E and fig. S12, A and B). We also observed significant increases in H3K4me3 at ACTB, which was used as a control gene, after 24 hours of hypoxia (fig. S12, A and B). No change was observed in the levels of H3K4me3 at the negative control genes used [BAP1 and Lysine Demethylase 2B (KDM2B)] (Fig. 2E and fig. S12, A and B). The mRNA levels of the investigated genes were unchanged after 1 hour of hypoxia exposure (fig. S13A). These results indicate that elevated H3K4me3 after 1 hour of hypoxia precedes increased gene transcription at the sites studied.

H3K4me3 is predicted to be erased by several JmjC enzymes, namely, KDM2B and the KDM5A-D family. Immunoblotting and mRNA analysis after siRNA-mediated depletion of these enzymes in HeLa cells revealed that KDM5A had the strongest effect, increasing mRNA levels of BNIP3L and KLF10 (Fig. 3A and figs. S14, A to C, and S15, A and B). STAG2, LOX, and ENO1 mRNA levels only increased significantly when treated with one siRNA against KDM5A (fig. S16, A and B). These changes were mirrored in the levels of H3K4me3 present at promoters of each gene tested, with significant increases in KLF10, BNIP3L, STAG2, and LOX when HeLa cells were siRNA depleted of KDM5A, but not in the negative controls, BAP1 and KDM2B (Fig. 3B and fig. S15C). These results indicate that KDM5A is directly involved in the regulation of H3K4me3 at the genes we identified as having early hypoxia induction of this histone modification. Functional cellular analysis of KDM5A depletion in HeLa cells resulted in reduced proliferation and a sub-G1 cell population under normoxia and hypoxia (Fig. 3C and fig. S16A). KDM5A depletion also increased the levels of autophagy markers and apoptosis BCL-2 homology (BH3) domain-containing proteins (fig. S16B).

Fig. 3 KDM5A regulates promoter H3K4me3 methylation and gene expression for a subset of hypoxia-inducible genes and controls cellular responses.

(A) Results of qPCR analysis (means ± SEM, N = 3) showing mRNA expression levels for the indicated genes in HeLa cells siRNA depleted of KDM2B, KDM5A, KDM5B, or KDM5C. (B) ChIP-qPCR analysis results (means ± SEM, N = 3) of H3K4me3 at ChIP-sequencing peaks for the indicated genes in HeLa cells siRNA depleted of KDM5A. (C) Cell proliferation analysis results (means ± SEM, N = 3) in HeLa cells siRNA depleted of KDM5A, with or without exposure to 1% O2 for 1 hour (1 H). **P < 0.01; ***P < 0.001.

Our results indicate that chromatin can sense oxygen at the cellular level, via JmjC domain-containing enzymes, in a similar manner to HIF-1/2α, via PHD enzymes. We thus investigated the dynamics of histone methylation marks in hypoxia. A feature of oxygen sensing by PHDs is the rapid reinstatement of basal HIF-1α levels upon reoxygenation (21). Therefore, we analyzed how rapidly histone methylation marks returned to normoxic levels upon restoration of normal oxygen tensions in HeLa cells (Fig. 4A). As expected, HIF-1α levels quickly returned to basal levels when cells were exposed to 21% oxygen after 24 hours of exposure to 1% oxygen (Fig. 4A). Similar results were also observed for all histone marks analyzed (Fig. 4A). To further delineate the oxygen sensitivity of histone methylation marks, we repeated our analysis in cells exposed to different levels of oxygen and compared these marker levels to the HIF-1α stabilization pattern (Fig. 4B). The histone methylation marks increased at the same oxygen concentration as HIF-1α, indicating that at least some of the JmjC-containing enzymes have similar oxygen requirements as the PHDs (Fig. 4B). This is in agreement with the reported Km values for the JmjC enzymes KDM4A and KDM4E (8, 9).

Fig. 4 Oxygen sensitivity of cellular histone methylation marks.

(A and B) Immunoblot analysis of the indicated proteins in HeLa cells with or without exposure to 1% O2 for 24 hours followed by 1, 2, or 4 hours of reoxygenation at 21% O2 (A) or with or without exposure to 15%, 10%, 5%, or 1% O2 for 1 hour (21% O2 is control) (B). (C) HEK293 cells overexpressing KDM5A wild type or the indicated mutants, with or without exposure to 1% O2 for 24 hours. (D) Quantitative proteomic analysis for KDM5 family members in HeLa cells. (E) Immunoblot analysis of the indicated proteins in HeLa cells siRNA depleted of KDM5A and exposed to 15% or 10% O2 for 1 hour.

Sequence and structural analyses of KDM5A, KDM5B, and KDM5C active sites revealed high levels of similarity among all three enzymes (fig. S17, A and B). The only visible difference is that serine-464 in KDM5A and KDM5C (position 495) is changed to a cysteine in KDM5B (position 480), creating a bigger onward interface in KDM5A than with the other KDM5 family members (fig. S17, A and B). However, there is no discernible change to the activity of KDM5A when serine-464 is changed to a cysteine, suggesting that this site is not responsible for the specificity we are proposing in response to hypoxia (fig. S18, A and B). Additional mutations across key domains, where residues were different between KDM5 family members, demonstrated that the JmjN (T30 and S34 to A) and Plant Homeodomain 1 (PHD1) (M297L) domains control demethylase activity even at 1% oxygen in HEK293 cells (Fig. 4C and fig. S18C). Given that these domains do not coordinate coactivators such as oxygen or 2-oxoglutarate, it is likely that they increase affinity for the histone tail (PHD1) and increase activity by potentiating dimerization (JmjN domain), as was recently discovered for the KDM4 family (22).

The mutational analysis of KDM5A still did not answer how the specificity of oxygen-dependent responses is achieved in cells. We thus hypothesized that this could be due to levels of the respective enzymes in cells. As such, we examined available quantitative proteomic datasets and collected the copy number information for several members JmjC domain-containing enzymes in HeLa cells (23) (Fig. 4D and fig. S19, A to E). KDM5A is the most abundant KDM5 family member, followed closely by KDM5C, whereas KDM5B is much less abundant (Fig. 4D). Our KDM5A overexpression analysis results (Fig. 4C and fig. S18B) support this hypothesis. Furthermore, siRNA depletion of KDM5A resulted in similar H3K4me3 levels in HeLa cells exposed to 15% or 10% oxygen, thereby setting a new sensitivity threshold for oxygen in these cells (Fig. 4E).

Although the oxygen Km for KDM5A is currently unknown, our data suggest that this property alone is insufficient to explain how different tissues respond to different oxygen tensions. Rather, and based on the data presented here, levels of JmjC histone demethylases in combination with oxygen affinity contribute to the overall sensitivity to oxygen in a given cell, promoting specificity across different tissues. Up-regulation of histone modifications associated with active gene transcription, through JmjC enzyme inhibition in early hypoxia, may coordinate the subsequent transcriptional changes. Reduction in active histone methylation marks might be compensated by up-regulation of repressive marks at specific genes. Based on our data and the accompanying report (24), JmjC domain-containing enzymes thus play an important role in the control of a variety of cellular responses in an oxygen-sensitive manner. These enzymes, including KDM5A, are evolutionary conserved from yeast to humans (2527), suggesting that oxygen sensing by chromatin via JmJC enzymes contributes to a conserved response to hypoxia across species. Furthermore, it suggests that oxygen sensing by chromatin is, in phylogenetic evolutionary terms, older than that of oxygen-regulated transcription factors such as HIFs.

Supplementary Materials

www.sciencemag.org/content/363/6432/1222/suppl/DC1

Materials and Methods

Figs. S1 to S19

Dataset S1

References (2850)

References and Notes

Acknowledgments: We thank S. Cottrill and A. Clark for technical assistance and A. Hermann for help with fluorometric measurements. We also thank the Centre for Cell Imaging and the accounts team in IIB at the University of Liverpool for help. Funding: S.R. was supported by grants from Cancer Research UK (C99667/A12918), Wellcome Trust (097945/B/11/Z; 206293/Z/17/Z), an MRC DTP training grant (to M.B.), a Tenovus Scotland small grant, and the University of Liverpool. Author contributions: M.B. and S.R. conceived the experiments and wrote the manuscript. M.B., J.F., M.F., J.W.W., and S.R. performed experiments. M.B., J.F., M.F., P.S., and S.R. analyzed the data. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the main manuscript and supplementary material. ChIP-sequencing and RNA-seq data have been uploaded to the NCBI GEO database (accession numbers GSE120339 and GSE120675).
View Abstract

Stay Connected to Science

Navigate This Article