HSF2 cooperates with HSF1 to drive a transcriptional program critical for the malignant state

Abstract

Heat shock factor 1 (HSF1) is well known for its role in the heat shock response (HSR), where it drives a transcriptional program comprising heat shock protein (HSP) genes, and in tumorigenesis, where it drives a program comprising HSPs and many noncanonical target genes that support malignancy. Here, we find that HSF2, an HSF1 paralog with no substantial role in the HSR, physically and functionally interacts with HSF1 across diverse types of cancer. HSF1 and HSF2 have notably similar chromatin occupancy and regulate a common set of genes that include both HSPs and noncanonical transcriptional targets with roles critical in supporting malignancy. Loss of either HSF1 or HSF2 results in a dysregulated response to nutrient stresses in vitro and reduced tumor progression in cancer cell line xenografts. Together, these findings establish HSF2 as a critical cofactor of HSF1 in driving a cancer cell transcriptional program to support the anabolic malignant state.

Fig. 1.. HSF2 interacts with HSF1 in diverse cancers.

(A) IP-MS of green fluorescent protein (GFP), HSF1, or HSF2 in four cancer cell lines, indicated. Data for HSF1 and HSF2 are plotted as the number of spectral counts normalized to GFP control. (B) HSF1 LUMIER assay reveals HSF2 as the top HSF1-interacting transcription factor. Plot shows ranked HSF1 LUMIER score (GFP-normalized, log₂-transformed luminescence values). (C) IP for HSF1 (H1) followed by immunoblot for HSF1 and HSF2. WCL, whole-cell lysates as input; 231, MDA-MB-231. (D) EMSA assay using HSPA8 (A8) as bait. Reactions were incubated with HSF1, HSF2, or control IgG antibody as indicated. A6 and A8 represent promoter sequences used for HSPA6 or HSPA8, respectively. A8* indicates a mutated HSE in the competing HSPA8 DNA promoter sequence. An arrow is included to highlight the level of supershift in HSF2 antibody lane. (E) Coessentiality rank (x axis) and correlation coefficient (y axis) for all genes with HSF1 (left) or HSF2 (right). The position of HSF2 or HSF1 in the coessentiality plot is indicated (orange or black dot, respectively).

Fig. 2.. HSF2 shares chromatin occupancy sites with HSF1 in cancer cells.

(A and F) ChIP-seq for HSF2 or HSF1 in the indicated cell line. Dual tracks indicate biological replicates. (B and G) Enriched motifs of ChIP peaks. (C and H) Example ChIP-seq tracks for HSF2 and HSF1 at five loci implicated in HSF cancer programs. (D, E, I, and J) ChIP-seq tracks for HSF1 or HSF2 in HSF knockout cells as indicated. These data merged all replicates for ease of visualization. WT, wild-type, parental cell lines. c1 and c2 represent cell lines expanded from a single wild-type cell clone as control for clonal CRISPR knockouts.

Fig. 3.. HSF2 drives a protumorigenic transcriptional program.

(A) RNA-seq data of 11 cancer cell lines treated with either siRNA targeting HSF1 (siHSF1) or siHSF2 or both. Data are expressed as log₂ fold change (log₂FC) versus the NT control in the respective cell line. The heatmap contains genes called significant for siHSF1, siHSF2, or both by EdgeR. Relative ChIP-seq binding intensity is plotted to the right of the heatmap for both HSF1 and HSF2. GSEA was performed and select enriched gene ontology (GO) terms are displayed along with their FDR-adjusted P value (Q value). (B) Select target genes from (A) are highlighted. PC3Mc, clonally derived, wild-type PC3M; PC3M*, wild-type population with On-Target Plus siRNA as opposed to siGenome siRNA for “PC3M.” (C) Median absolute log₂ fold change signature strength for ChIP-bound genes in (A) for cell lines receiving single and double siRNA treatment. Data are plotted as median ± 95% confidence interval. Additivity was tested (see Materials and Methods) and indicated where P < 0.05; see also fig. S4 (fig. S4, B and D).

Fig. 4.. Long-term loss of HSF2 and HSF1 results in sustained suppression of proteostasis gene expression conserved across cancers.

(A) Knockout of HSF1 or HSF2 was confirmed by immunoblot. (B) Correlation plots of log₂ fold change values for the union of differentially expressed genes for each siRNA-treated sample and each sgRNA knockout sample. (C) Differentially expressed genes in either sgHSF1 or sgHSF2 knockout cells for the indicated cell lines. Relative ChIP-seq binding intensity is plotted to the right of the heatmap for both HSF1 and HSF2. GSEA was performed and select enriched GO terms are displayed along with their FDR-adjusted P value (Q value). (D) Select target genes from (A) are highlighted. (E) Heatmap of single and double-knockout (dKO) MDA-MB-231 cells, plotting genes from (C).

Fig. 5.. HSF2 and HSF1 promote the transcriptional response to cancer-associated stresses.

We treated HSF2 knockout (sgHSF2) or control (sgNT) MDA-MB-231 cells with siHSF1 or NT (siNT). Cells were treated with either 10 mM 2-DG for 24 hours, serum starvation (SS) for 48 hours, or 250 μM cobalt chloride (CoCl₂). (A, C, and E) Differentially expressed genes were determined relative to unstressed sgNT + siNT cells (Ctrl). Enriched GSEA terms are indicated. JNK, c-Jun N-terminal kinase. (B, D, and F) Select genes from (A) to (C) are plotted and labeled as direct or indirect transcriptional targets based on ChIP-seq from Fig. 2.

Fig. 6.. HSF2 is required for tumor progression in cell line xenografts of prostate and breast cancer.

(A) Tumor volume after subcutaneous injection of two independent PC3M cell clones for each of wild type, HSF1 knockout (sgHSF1), or HSF2 knockout (sgHSF2). Data are grouped by genotype, combined across clones. ****P < 0.0001 by two-way analysis of variance (ANOVA) at day 32 after inoculation. (B) Survival analysis of mice bearing PC3M tumors. (C) Tumor volume measurements for MDA-MB-231 xenografts. (D) Survival analysis for of mice bearing MDA-MB-231 tumors. (E) Representative hematoxylin and eosin (H&E) and IHC staining for HSF1 or HSF2 for PC3M cell line xenograft tumor. Scale bars, 100 μm. Immunoblot validation of knockout cells before injection are provided in fig. S2J. (F) RNA-seq performed on tumors from each knockout or control at experiment end point. Data are expressed as log₂ fold change relative to four control tumors. Enriched GO terms are indicated.