HIF1A employs CDK8-mediator to stimulate RNAPII elongation in response to hypoxia

Abstract

The transcription factor HIF1A is a key mediator of the cellular response to hypoxia. Despite the importance of HIF1A in homeostasis and various pathologies, little is known about how it regulates RNA polymerase II (RNAPII). We report here that HIF1A employs a specific variant of the Mediator complex to stimulate RNAPII elongation. The Mediator-associated kinase CDK8, but not the paralog CDK19, is required for induction of many HIF1A target genes. HIF1A induces binding of CDK8-Mediator and the super elongation complex (SEC), containing AFF4 and CDK9, to alleviate RNAPII pausing. CDK8 is dispensable for HIF1A chromatin binding and histone acetylation, but it is essential for binding of SEC and RNAPII elongation. Global analysis of active RNAPII reveals that hypoxia-inducible genes are paused and active prior to their induction. Our results provide a mechanistic link between HIF1A and CDK8, two potent oncogenes, in the cellular response to hypoxia.

Figure 1. CDK8 and CDK19 regulate distinct gene expression programs

(A) Western blots showing mutually exclusive interaction of CDK8 and CDK19 with the CDK-module of Mediator. CDK-module subunits (CDK8, CDK19, Cyclin C, MED12, MED13), MED26 and CDK9 were detected in pulldowns from protein extracts of HCT116 cells expressing HaloTag (HT) alone, HT-CDK8 or HT-CDK19. Increasing twofold dilutions are shown to indicate relative abundance of interacting subunits. (B) Microarray analysis showing the effects of CDK8 (shCDK8) or CDK19 (shCDK19) knockdown on the gene expression profiles of HCT116 cells under conditions leading to exponential cell proliferation, sorted by effect of shCDK8 or shCDK19 (>1.5 fold change from shControl, p<0.05). (C–E) Microarray analysis showing the effects of CDK8 or CDK19 knockdown on genes induced by (C) genotoxic stress upon 5-fluorouracil treatment (5FU, 12 h), (D) by glucose deprivation (24 h) or (E) by hypoxia (1% O₂, 24 h). In all cases induction is defined as >1.5 fold, p<0.05. Heatmaps are color-coded by log2 fold-change, treated vs. untreated shControl cells for induction, shCDK8 or shCDK19 treated vs. shControl treated for knockdown effect. Venn diagrams compare the numbers of genes induced by each stimuli to those decreased by shCDK8 and shCDK19 under each condition (≥1.1 fold change, p<0.05). See also Figure S1 and Table S1.

Figure 2. CDK8 is a coactivator of many hypoxia-inducible genes

(A) Relative expression of hypoxia-inducible genes as measured by Q-RT-PCR for control and CDK8 knockdown HCT116 cells in normoxia or after 24 h hypoxia (1% O₂). Expression values were normalized to 18S rRNA and are expressed relative to the control normoxia value. Error bars represent SEM from three independent biological replicates. (B) Relative expression of hypoxia-inducible genes as measured by Q-RT-PCR for HIF1A+/+ and HIF1A−/− HCT116 cells in normoxia or after 24 h hypoxia (1% O₂). (C) Venn diagrams displaying the proportion of hypoxia-inducible genes with CDK8 peaks and those negatively affected by CDK8 knockdown in normoxia and hypoxia. (D) CDK8 binding profiles around transcription start sites in normoxia and hypoxia as determined by ChIP-seq. Shown are 10 kb regions centered on all unique RefSeq TSS (left panel) and the TSS of hypoxia-inducible genes (right panel). Heatmaps are ranked by the CDK8 normoxia signal and the color scale represents tags per 50bp. (E) Metagenes centered on TSS showing CDK8 binding profiles for RefSeq genes (top) and hypoxia-inducible genes (bottom) in normoxia (blue) and hypoxia (red). (F) Dot plot comparing CDK8 enrichment in normoxia (x-axis) and hypoxia (y-axis) at all CDK8 peak regions (blue) and at peaks associated with hypoxia-inducible genes (red). The grey line represents a 1:1 relationship (no change) between normoxia and hypoxia. (G) Genome browser views of CDK8 binding at HIF1A target loci under both normoxic (grey) and hypoxic conditions (black). See also Figure S2 and Table S2.

Figure 3. CDK8 regulates RNAPII elongation without affecting HIF1A binding or histone acetylation

Quantitative ChIP analysis of (A) CDK8, (B) total RNAPII, (C) serine-5 and (D) serine-2-phosphorylated RNAPII CTD (S5P, S2P), (E) HIF1A, (F) H3K9ac and (G) H4ac histone acetylation, (H) the CDK9 subunit of P-TEFb, (I) the bromodomain protein BRD4, (J) the SEC subunit AFF4, the core mediator subunits (K) MED1 and (L) MED26, (M) the DSIF subunit Spt5 and (N) the TFIIH subunit CDK7 at the ANKRD37 locus in control and CDK8 knockdown HCT116 cells in normoxia or after 24 h hypoxia (1% O₂). Values are plotted as a percentage of the maximum signal for that locus. Error bars represent SEM from three independent biological replicates. Grey shading indicates the transcribed region. See also Figure S3.

Figure 4. HIF1A is required for recruitment of CDK8, Mediator and P-TEFb to chromatin

(A) Quantitative ChIP analysis of total RNAPII, serine-5 and serine-2-phosphorylated RNAPII CTD (S5P, S2P), histone acetylation (H3K9ac and H4ac), CDK8, MED1 and CDK9 at the ANKRD37 locus in HIF1A+/+ and HIF1A−/− cells. Values are plotted as a percentage of the maximum signal. Error bars represent SEM from three independent biological replicates. Grey shading indicates the transcribed region. (B) Schematic of HIF1A protein domains cloned as GST-fusions for use as affinity bait proteins. (C) Western blot analysis of proteins purified by their affinity for GST-HIF1A or GST-VP16 transactivation domains. GST-alone served as a negative control. VP16-interacting proteins were diluted ten-fold lower to avoid overloading. See also Figure S4.

Figure 5. Hypoxia-inducible genes are active and paused in normoxia

(A) Gene profiles showing levels of transcriptionally engaged RNA polymerases (GRO-seq, sense strand) and RNAPII occupancy (ChIP-seq) in comparison to CDK8 occupancy (ChIP-seq) in normoxia for all RefSeq genes. Heatmap color scales represent the read density for 160 bins across the transcribed region of each gene with 1 kb upstream and downstream flanking regions (20 bins of 50 bp each). Heatmaps are ranked by gene activity as determined from GRO-seq. (B) Relative proportions of genes in each transcription class for RefSeq genes (top) and hypoxia-inducible genes (bottom). (C) Metagenes showing average GRO-seq (sense strand) and CDK8 ChIP-seq signals across RefSeq genes (blue) in comparison to the average signals across hypoxia-inducible genes (red). Units are mean tags per bin for 160 bins across the transcribed region of each gene with 1 kb upstream and downstream flanking regions (20 bins of 50 bp each) (D) Cumulative distribution plots of pausing index for active RefSeq genes (blue) and active hypoxia-inducible genes (red). Distributions are significantly different (Kolmogorov-Smirnov test, P = 0.0347). (E) Cumulative distribution plots of gene activity for active RefSeq genes (blue) and active hypoxia-inducible genes (red). Distributions are significantly different (Kolmogorov-Smirnov test, P < 0.0001). (F) Genome browser views showing GRO-seq reads in the sense- (blue) and anti-sense (red) direction relative to each gene, and RNAPII ChIP-seq signal (grey) at HIF1A target loci under normoxic conditions. See also Table S3.

Figure 6. A new model of HIF1A transactivation

See Discussion for details.