Report

Histone demethylase KDM6A directly senses oxygen to control chromatin and cell fate

See allHide authors and affiliations

Science  15 Mar 2019:
Vol. 363, Issue 6432, pp. 1217-1222
DOI: 10.1126/science.aaw1026

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 sensing is central to metazoan biology and has implications for human disease. Mammalian cells express multiple oxygen-dependent enzymes called 2-oxoglutarate (OG)-dependent dioxygenases (2-OGDDs), but they vary in their oxygen affinities and hence their ability to sense oxygen. The 2-OGDD histone demethylases control histone methylation. Hypoxia increases histone methylation, but whether this reflects direct effects on histone demethylases or indirect effects caused by the hypoxic induction of the HIF (hypoxia-inducible factor) transcription factor or the 2-OG antagonist 2-hydroxyglutarate (2-HG) is unclear. Here, we report that hypoxia promotes histone methylation in a HIF- and 2-HG–independent manner. We found that the H3K27 histone demethylase KDM6A/UTX, but not its paralog KDM6B, is oxygen sensitive. KDM6A loss, like hypoxia, prevented H3K27 demethylation and blocked cellular differentiation. Restoring H3K27 methylation homeostasis in hypoxic cells reversed these effects. Thus, oxygen directly affects chromatin regulators to control cell fate.

Oxygen’s appearance in Earth’s atmosphere was a watershed that created the evolutionary selection pressure for a conserved pathway used by metazoans to sense and respond to changes in ambient oxygen that converges on the heterodimeric HIF (hypoxia-inducible factor) transcription factor. Under normoxic conditions, the HIFα subunit is prolyl hydroxylated by the EglN (egg-laying defective nine) isoenzymes of the 2-OGDD dioxygenase family. Hydroxylated HIFα is marked for degradation by the von Hippel-Lindau (VHL) ubiquitin ligase complex. Hypoxia inactivates the EglNs and thereby stabilizes HIFα, which then associates with HIF1β [also called ARNT (aryl hydrocarbon nuclear translocator)] and transcriptionally activates genes that promote adaptation to inadequate oxygen (1).

The 2-OG-dependent dioxygenase family also includes the collagen prolyl hydroxylases, the JmjC (Jumonji C) domain histone lysine demethylases (KDMs), the ten-eleven translocation (TET) DNA hydroxylases, and ~50 other enzymes that are relatively understudied (2). In contrast to the high-oxygen-affinity (low KM) collagen prolyl hydroxylases, the EglNs exhibit low oxygen affinities (high KM) (1), which enables them to sense physiological changes in oxygen.

Several previous studies have suggested that oxygen regulates histone methylation. Certain KDMs display low oxygen affinities in vitro (3). Moreover, many KDMs are transcriptionally activated by hypoxia and HIF (4), perhaps to compensate for a decrease in their enzymatic specific activity. Finally, hypoxia can induce histone hypermethylation (5). However, in these previous studies, it was unclear whether histone hypermethylation reflected a direct effect of hypoxia on KDMs or was confounded by indirect consequences of hypoxia (and HIF). For example, in some cells hypoxia increases the L-enantiomer of 2-hydroxyglutarate (L-2HG), which is an endogenous inhibitor of 2-OG–dependent dioxygenases (68). Moreover, HIF can potentially affect chromatin in many ways, such as by altering KDM protein levels (see above), by inducing chromatin-modifying enzymes other than KDMs [e.g., TETs (9) and DNA methyltransferases (10)], or by up-regulating transcription factors that enforce an epithelial-mesenchymal transition (EMT) and accompanying epigenetic reprogramming (11).

To rigorously address whether hypoxia has a direct or indirect effect on histone methylation, we lentivirally transduced an Arnt-defective (HIF-inactive) mouse hepatoma cell line (mHepa-1 c4) to express either green fluorescent protein (GFP), wild-type (WT) ARNT, or a functionally inactive ARNT mutant (Δ414) that is missing 414 base pairs from its N terminus, thereby eliminating its basic helix-loop-helix domain and ability to heterodimerize (Fig. 1A and fig. S1, A to C). mHepa-1 c4 did not tolerate prolonged growth in 1% oxygen, which is the oxygen concentration typically used to model hypoxia ex vivo. We therefore used more modest levels of hypoxia (2 to 5%) to study these cells. As expected, canonical HIF-target genes (e.g., Egln3 and Ndrg1) were transcriptionally induced by 5% oxygen in the cells expressing WT ARNT but not in the cells expressing ARNT (Δ414) or GFP (fig. S1D).

Fig. 1 Hypoxia causes HIF-independent histone hypermethylation.

(A to C) Vector schematic (A), histone modification profiling by mass spectrometry (B), and histone immunoblot analysis (C) of Arnt-deficient mouse hepatoma (mHepa-1 c4) cells that were lentivirally transduced to produce the indicated V5-tagged proteins and cultured at the indicated oxygen levels for 4 days. In (B), rows represent two biological replicates of the indicated samples, and the color in each cell represents log2 fold change relative to all samples in the column, normalized for total histone by using an internal control peptide (histone H3: residues 41 to 49). CMV, cytomegalovirus promoter. (D to F) Coomassie blue staining (D) and kinetic analysis of baculovirally expressed and purified JmjC catalytic domains of KDM6A (KDM6A*) (E) and KDM6B (KDM6B*) (F). KM values are mean ± SD (N = 3 independent biochemical measurements). (G) Immunohistochemical (IHC) analysis of the indicated tissues derived from representative male and female age-matched mice.

We next used a multiplexed mass spectrometric assay (12) to quantify changes in histone methylation in response to hypoxia in the isogenic [GFP, ARNT(WT), ARNT(Δ414)] mHepa-1 c4 cells. Unsupervised clustering of histone modification patterns revealed that the isogenic cell lines clustered primarily based on oxygen availability during growth and not HIF status (Fig. 1B). Consistent with prior reports (5), hypoxia promoted the dimethylation and trimethylation (me3) of H3K4 (histone 3, lysine 4), H3K9, and H3K27 (Fig. 1B). Hypermethylation of H3K9 and H3K27 was confirmed by immunoblot analysis (Fig. 1C). We also observed a concomitant decrease in hypomethylated H3K27 (me0 and me1 states) and acetylated H3K27 in response to hypoxia (Fig. 1B), which is consistent with the knowledge that histone methylation and acetylation are reciprocally regulated. The H3K27 hypermethylation was not explained by increased expression of EZH2 (enhancer of zeste homolog 2), which controls bulk H3K27 methylation, or decreased expression of the primary H3K27 histone demethylases KDM6A [also called ubiquitously transcribed TPR protein on the X chromosome (UTX)] and KDM6B (fig. S2).

Hypoxia also promoted histone hypermethylation in VHL−/− RCC4 human renal carcinoma cells (fig. S3). Thus, hypoxia promotes histone hypermethylation both in cells that cannot mount a HIF response (mHepa-1 c4 cells) and in cells that constitutively overproduce HIF (RCC4 cells), arguing that hypoxia’s effects on histone methylation are not caused by changes in HIF activity.

We next explored whether hypoxia’s effects on histone methylation in mHepa-1 c4 cells were caused by metabolic changes that can inhibit KDM activity, such as increased L-2HG or decreased 2-OG (68, 13). In the isogenic mHepa-1 c4 cells, 5% oxygen did not significantly induce either total 2-HG (fig. S4A) or enantiomer-specific 2-HG (fig. S4, B and C) and actually increased 2-OG levels (fig. S4D). 2-HG was modestly induced in parental mHepa-1 c4 cells by more profound levels of hypoxia (0.5 to 2% oxygen) (fig. S4, C and E), albeit as D-2HG rather than L-2HG (fig. S4C). The importance of the latter finding is unclear. Even under these more extreme conditions, the 2-HG levels achieved were ~100-fold below both the intracellular levels in mutant isocitrate dehydrogenase 1 cells (fig. S4F), wherein 2-HG serves as an oncometabolite (14), and the intracellular levels required to promote histone methylation in mHepa-1 c4 cells treated with cell-permeable versions of D-2HG or L-2HG (fig. S4, G and H). The latter observation is consistent with the biochemical D-2HG and L-2HG half-maximal inhibitory concentration values for the KDMs (15).

Hypoxia can incite reactive oxygen species (ROS), which can inhibit 2-OG-dependent dioxygenases (16). Treating mHepa-1 c4 cells with the ROS-inducer tert-Butyl hydroperoxide showed that intracellular ROS levels ~10-fold higher than those observed after exposure to 2% oxygen were required to induce histone methylation (fig. S5). These findings suggested that the HIF-independent effects of hypoxia on KDM activity were not caused by increased L-2HG, decreased 2-OG, or increased ROS but instead were caused by a direct effect of hypoxia on the enzymatic activities of specific KDMs.

In support of this idea, we found that recombinant KDM4B, KDM5A, and KDM6A [also called ubiquitously transcribed TPR protein on the X chromosome (UTX)] have relatively low oxygen affinities that are comparable to the EglN family, whereas recombinant KDM4A, KDM5B, KDM5C, KDM5D, and KDM6B have high oxygen affinities (figs. S6 to S8). Finally, like full-length KDM6A, the isolated KDM6A catalytic domain also had low oxygen affinity compared to its KDM6B counterpart (Fig. 1, D to F, and fig. S8).

We focused our attention on the KDM6 H3K27 demethylases because hypoxia and histone H3K27 methylation have been independently linked to the control of cellular differentiation (1719), because this histone mark can be manipulated with drugs, and because KDM6A has the lowest oxygen affinity of the KDMs tested to date. We first confirmed that hypoxia induced H3K27 methylation in additional human cell lines including 293T embryonic kidney cells, MCF7 breast cells, and SK-N-BE(2) neuroblastoma cells (fig. S9). Moreover, histological analysis showed elevated H3K27 methylation in mouse tissues that are known to be hypoxic, such as the kidney (20), splenic germinal centers (21), and thymus (22), but not in well-oxygenated tissues such as the heart (Fig. 1G). Similarly, H3K27 methylation is increased in hypoxic regions of mouse tumors (2325) (fig. S10). Finally, Gene Set Enrichment Analysis (26) of ~2000 human tumors that were previously annotated as normoxic or hypoxic on the basis of their HIF signature (27) (table S1) revealed that hypoxic tumors had transcriptional signatures indicative of H3K27 hypermethylation (fig. S11 and tables S2 and S3).

To examine the effect of hypoxia on cell differentiation in vitro, we studied C2C12 murine myoblasts. C2C12 differentiate into myosin heavy chain (MyHC)–positive multinucleated myotubes when shifted from serum-rich growth media (GM) to nutrient-poor differentiation media (DM). Hypoxia blocks C2C12 differentiation (fig. S12, A to D) (28, 29). Hypoxic C2C12 cells grown in DM entered a quiescence-like state but more readily proliferated when returned to GM under normoxic conditions, compared with their normoxic counterparts (fig. S12, E to G). Similarly, hypoxia blocked the myogenic differentiation of mouse embryo fibroblasts (MEFs) that conditionally express MyoD (fig. S13).

Eliminating ARNT in C2C12 cells by using CRISPR-Cas9 blocked (rather than accentuated) their ability to differentiate under normoxic conditions (fig. S14, A to C) and did not rescue their ability to differentiate under hypoxia (fig. S14D). Moreover, expression of stabilized versions of HIF1α or HIF2α did not block normoxic C2C12 differentiation (fig. S14, E and F). Thus, the differentiation block exhibited by hypoxia in C2C12 cells is not due to HIF activation.

Total 2-HG was not induced, and L-2HG was induced only about twofold (fig. S15, A to C) in C2C12 cells by 2% oxygen. C2C12 cells that were pharmacologically or genetically manipulated to have intracellular L-2HG levels three- to fivefold higher than levels observed under hypoxia still differentiated in DM (fig. S15, D to G). Therefore, hypoxia’s effects on C2C12 differentiation were not caused by 2-HG.

Similar to hypoxia, treatment of C2C12 cells with the KDM6 family inhibitor GSK-J4 promoted H3K27 hypermethylation and blocked myogenic differentiation (fig. S16). Similar results were obtained with MEFs expressing MyoD (fig. S13). In contrast, KDM-C70, a KDM5 family inhibitor, did not block C2C12 differentiation (fig. S17).

The differential oxygen affinities of KDM6A and KDM6B suggested that the ability of hypoxia to promote H3K27 methylation and block differentiation is caused specifically by a loss of KDM6A activity. Indeed, down-regulating KDM6A, but not KDM6B, with different short hairpin RNAs (shRNAs) phenocopied the effects of hypoxia on differentiation (Fig. 2, A and B, and fig. S18, A to D) (30). Moreover, eliminating KDM6A in C2C12 cells with CRISPR-Cas9 blocked their ability to differentiate unless they were rescued with a single guide RNA (sgRNA)–resistant KDM6A cDNA (Fig. 2, C to E). In contrast, eliminating KDM6B had no effect (fig. S18, E and F), and eliminating KDM5A promoted differentiation (fig. S19). Bulk H3K27me3 was oxygen-insensitive in the Kdm6a-deficient C2C12 cells, consistent with KDM6A being the primary oxygen sensor amongst the KDM6 paralogs (Fig. 2F).

Fig. 2 Regulation of myogenic differentiation by the KDM6A H3K27 demethylase.

(A to B) Immunofluorescence analysis of C2C12 cells that were lentivirally transduced to express the shRNAs targeting Kdm6A (A) or Kdm6B (B) and then cultured in DM for 4 days. DAPI, 4′,6-diamidino-2-phenylindole. (C to E) Immunoblot analysis of C2C12 cells lentivirally transduced to express the indicated sgRNAs and cultured for 4 days either in GM or DM, as indicated. (E) Immunoblot analysis of C2C12 cells expressing, where indicated, Kdm6a sg2 [described in (C) and (D)] that were lentivirally transduced to produce either GFP (control) or wild-type human KDM6A [6A(WT)] and then cultured in DM at the indicated oxygen concentrations for 4 days. The mouse Kdm6A sg2 target sequence is not conserved in human KDM6A. Con sg, Nontargeting control sgRNA. (F) Immunoblot analysis of histones from C2C12 cells expressing the indicated sgRNA that were cultured at the indicated oxygen concentrations for 3 days. (G to J) Heatmap representing mRNA levels determined by RNA sequencing (RNA-seq) (G) and H3K27me3 levels determined by chromatin immunoprecipitation sequencing (ChIP-seq) analysis at the Actc1 (H), Myl1 (I), and the Myog (J) loci from two biological replicates (A and B) of C2C12 cells cultured in the indicated media for 4 days at the indicated oxygen concentration.

Previous work showed that KDM6A is directly recruited to myogenic targets during differentiation (30). We therefore investigated whether differentiation programs driven by KDM6A activity involve transcriptional changes that depend on H3K27me3 elimination. Comparison of transcriptional signatures of normoxic C2C12 cells grown in GM versus DM revealed profound differences in transcriptional output, particularly of muscle-specific target genes (Fig. 2G, fig. S20, and table S5). Hypoxia (and the consequent differentiation block), however, blunted the transcriptional differences between these two conditions (fig. S20A), which was associated with a failure of these cells to induce muscle-specific markers in DM (Fig. 2G and fig. S20B). H3K27me3 status typically represses transcription. The inability of C2C12 cells grown in DM to activate late myogenic genes (e.g., Actc, Myl1, and Myog) under hypoxia correlated with a failure to erase H3K27me3 at those loci (Fig. 2, H to J, and fig. S21), presumably because of inactivation of KDM6A. This was specific because H3K4me3 decreased at late myogenic genes under hypoxia (fig. S21).

Loss of cellular differentiation is a hallmark of cancer, and KDM6A is a human tumor suppressor gene that is inactivated in a variety of cancers, including leukemia, kidney cancer, and bladder cancer (31). Gene Set Enrichment Analysis showed that a myogenic differentiation gene set, which presumably also contains genes linked to differentiation in other contexts, is more highly expressed in KDM6A WT bladder cancers compared with KDM6A mutant tumors (fig. S22 and table S6).

These data suggest that KDM6A inactivation by hypoxia promotes the persistence of H3K27me3 and prevents the transcriptional reprogramming required for differentiation. If true, the effects of hypoxia on differentiation should be redressed by inhibiting H3K27 methyltransferase activity (Fig. 3A). Like mHepa-1 c4 cells, hypoxia did not alter the protein levels of the EZH H3K27 methyltransferases in C2C12 cells (fig. S23A). Inhibiting the H3K27 methyltransferase EZH2 with CRISPR-Cas9 (Fig. 3B and fig. S23B) or with the drug GSK126 (fig. S23, C and D) reduced H3K27me3 levels and partially rescued the ability of C2C12 cells to differentiate under hypoxic conditions. By contrast, rescue with the G9a and GLP methyltransferase inhibitor UNC638 was ineffective (fig. S23, C and D). Finally, GSK126 rescued the hypoxia-induced differentiation block in human primary myoblasts and in MEFs conditionally expressing MyoD (fig. S23, E and F).

Fig. 3 Restoring the balance of H3K27 methyltransferase activity to H3K27 demethylase activity rescues myogenic differentiation under hypoxic conditions.

(A) Model for control of H3K27 methylation by the indicated opposing demethylases (“erasers”) and methyltransferases (“writers”). (B) Immunoblot analysis of C2C12 cells lentivirally transduced to express the indicated sgRNAs and cultured under the indicated conditions. Vinc, Vinculin. (C) Structural models of the KDM6A (pink) and KDM6B (cyan) catalytic pockets. The nonconserved M1190 (KDM6A) → T1434 (KDM6B) and the E1335 (KDM6A) → D1579 (KDM6B) are highlighted. Peptidic H3K27me3 substrate (yellow), Fe+2 (orange), 2-OG (green), and Zn+2 (dark purple) are shown. (D and E) Michaelis-Menten plots (inset Lineweaver-Burk plot) and measured oxygen KM and Vmax values (mean ± SD, N = 3) of recombinant KDM6A wild type and the MT/ED mutant. (F and G) Real-time quantitative polymerase chain reaction analysis (mean ± SD, N = 3) of the indicated genes (F) and immunofluorescence analysis (G) of C2C12 cells transduced to produce WT human KDM6A or the KDM6A MT/ED variant and then cultured in DM at the indicated oxygen concentrations for 4 days. *P < 0.05 ns, not significant; Rel., relative.

Hypoxia can also, in a HIF-independent manner, alter the differentiation of human mammary epithelial cells (25), causing an EMT (fig. S24, A and B) and up-regulation of the cancer stem-like marker CD44 (fig. S24C). Similar to our findings with C2C12 cells, these hypoxia-associated changes were phenocopied by pharmacologic (GSK-J4) (fig. S24, A to C) or genetic (CRISPR-Cas9) disruption of KDM6A (fig. S24, D to F) and rescued by the EZH inhibitor GSK126 (fig. S24, A to C).

Finally, we tried to directly increase KDM6A’s oxygen affinity. We reasoned that certain nonconserved residues in the catalytic domains of KDM6A and KDM6B caused their vastly different oxygen affinities. We overlaid published models of the catalytic JmjC domains of KDM6A with KDM6B (32, 33) and noted two nonconserved residues, M1190 (KDM6A) → T1434 (KDM6B) and E1335 (KDM6A) → D1579 (KDM6B), lining the 2-OG– and Fe+2-binding pocket (Fig. 3C). As predicted, a KDM6A variant that harbored these two KDM6B-like changes (MT/ED: M1190 → T and E1335→ D) displayed a twofold increased affinity for oxygen in vitro, albeit at the cost of a decreased Vmax (Fig. 3, D and E). WT KDM6A and the MT/ED variant were comparably insensitive to L-2HG and to ROS (fig. S25, A to D), which was induced less than twofold by 2% oxygen in C2C12 cells (fig. S25E). Both wild-type KDM6A and the MT/ED rescued the ability of KDM6A-deficient C2C12 cells to differentiate under normoxia (fig. S26A). The double mutant, however, was superior to WT KDM6A at rescuing differentiation under hypoxic conditions in both parental and KDM6A-deficient C2C12 cells, presumably because of its enhanced oxygen affinity (Fig. 3, F and G, and fig. S26, B to D).

Independent lines of research have shown that oxygen and H3K27 methylation each regulate embryological development, cellular differentiation, stemness, and malignant transformation (1719). We propose that these observations are linked. Specifically, we argue that oxygen has both direct and indirect effects on chromatin and that the former involves enzymes such as KDM6A, which couple changes in oxygen availability to changes in H3K27 methylation and the transcriptional control of cell fate.

The observed H3K4 hypermethylation and H3K9 hypermethylation in hypoxic Arnt-deficient mHepa-1 c4 cells, together with our biochemical studies, suggest that at least one H3K4 and one H3K9 histone demethylase also act as oxygen sensors. For example, our biochemical data, together with the data in the accompanying manuscript (34), argue that KDM5A plays such a role. Profound hypoxia can also inhibit other 2-OG–dependent enzymes, including TETs, leading to DNA hypermethylation (27).

Mammalian embryological development occurs in a hypoxic environment and mammalian stem cells are maintained in hypoxic niches. It is well established that hypoxia can affect stemness and cellular differentiation by activating HIF and HIF-target genes such as Oct4 (19). Such effects are not mutually exclusive with direct effects of oxygen on histone methylation and might serve to reinforce one another. Hypoxia promotes stemness in both metazoans and plants, but the HIF pathway is only present in the former (35). It is possible that oxygen sensing by histone demethylases evolutionarily preceded the emergence of oxygen-sensitive transcription factors.

Supplementary Materials

www.sciencemag.org/content/363/6432/1217/suppl/DC1

Materials and Methods

Figs. S1 to S26

Tables S1 to S6

References (3655)

References and Notes

Acknowledgments: We thank H. Zhao, D. Lambrechts, and P. Carmeliet for sharing their TCGA tumor annotations. We thank R. Looper for synthesizing and sharing the esterified 2-HG, M. K. Koski and R. Wierenga for help with structural modeling, and T. Aatsinki and E. Lehtimaki for technical assistance. We thank the Broad GDAC group for help with the GDAC-GSEA source code. We thank M. Oser for sharing mouse KDM5A sgRNAs and V. Koduri for assistance in acquiring microscopic images. Funding: W.G.K. was supported by grants from the NIH (R01CA068490, P50CA101942, and R35CA210068). A.A.C. was supported by grants from the Friends of Dana-Farber and the NIH (Cancer Biology Training grant: T32CA009361 and the DF/HCC Kidney SPORE CEP and DRP award: P50CA101942). P.K. was supported by Academy of Finland grants (266719 and 308009), the S. Juselius Foundation, the Jane and Aatos Erkko Foundation, and the Finnish Cancer Organizations. T.L. was supported by the Finnish Medical Foundation and, with P.K., the Emil Aaltonen Foundation. S.K.M. is supported by an American Cancer Society postdoctoral fellowship (PF-14-144-01-TBE) and by a Career Enhancement Project award from the Dana-Farber/Harvard Cancer Center Brain SPORE. W.G.K. is an HHMI Investigator. Author contributions: A.A.C., P.K., and W.G.K. conceived experiments and wrote the manuscript; A.A.C., T.L., M.M., A.E.R., A.L.C., and R.B.J. performed experiments. A.A.C., T.L., M.M, A.E.R., M.A.B., M.Y.T., Y.J.M., S.R.M., S.K.M., B.A.O., Z.T.H., J.D.J., S.S., M.C.H., R.B., P.K., and W.G.K. analyzed data. Competing interests: W.G.K. has financial interests related to Lilly Pharmaceuticals (Board of Directors), Agios Pharmaceuticals [scientific advisory board (SAB)], Cedilla Therapeutics (founder), Fibrogen (SAB), Nextech Invest (SAB), Peloton Therapeutics (SAB), Tango Therapeutics (founder), and Tracon Pharmaceuticals (SAB). W.G.K. is a coinventor on a patent (US6855510B2) covering pharmaceuticals and methods for treating hypoxia, which has been licensed to Fibrogen. S.S. has a consulting or advisory role in AstraZeneca/MedImmune and Merck; patents, royalties, and other intellectual property from Biogenex Laboratories; and research funding from AstraZeneca, Exelixis, and Bristol-Myers Squibb. R.B. receives research funding from Novartis. Data and materials availability: All data are available in the manuscript or the supplementary material. RNA-seq and ChIP-seq data have been uploaded to the GEO (GSE114086).
View Abstract

Stay Connected to Science

Navigate This Article