The reprogramming of tumor stroma by HSF1 is a potent enabler of malignancy

Abstract

Stromal cells within the tumor microenvironment are essential for tumor progression and metastasis. Surprisingly little is known about the factors that drive the transcriptional reprogramming of stromal cells within tumors. We report that the transcriptional regulator heat shock factor 1 (HSF1) is frequently activated in cancer-associated fibroblasts (CAFs), where it is a potent enabler of malignancy. HSF1 drives a transcriptional program in CAFs that complements, yet is completely different from, the program it drives in adjacent cancer cells. This CAF program is uniquely structured to support malignancy in a non-cell-autonomous way. Two central stromal signaling molecules-TGF-β and SDF1-play a critical role. In early-stage breast and lung cancer, high stromal HSF1 activation is strongly associated with poor patient outcome. Thus, tumors co-opt the ancient survival functions of HSF1 to orchestrate malignancy in both cell-autonomous and non-cell-autonomous ways, with far-reaching therapeutic implications.

Figure 1. HSF1 activation in cancer-associated fibroblasts within human tumors

(A) Tissue sections of breast resection specimens from 12 patients encompassing both invasive ductal carcinoma and neighboring normal breast lobules (in the same section) were immunostained with anti-HSF1 antibodies (brown, upper panels) or co-stained with anti-HSF1 and anti-SMA (pink) antibodies (lower panels). Representative images are shown. Arrows indicate HSF1-positive CAFs in the left panels, and HSF1-negative normal fibroblasts in the lower right panel. (B) Pie charts depict the distribution of relative nuclear HSF1 staining intensity in the stroma amongst 12 breast resection specimens with matching controls. For each specimen, 4 regions of tumor or normal tissue were evaluated. Statistical significance of the differences between normal and tumor was assessed using repeated-measures ANOVA (p=4e-13), as well as paired t-tests, followed by Bonferroni correction (p<0.01). (C) Representative images of tumor sections from patients with the indicated types of cancer co-stained for HSF1 (brown) and SMA (pink). C and S indicate cancer- or stroma-rich regions, respectively. For normal tissue, E and F indicate regions rich with epithelial cells or fibroblasts, respectively. See also Figure S1.

Figure 2. Stromal Hsf1 status alters tumor progression and histology in human breast xenografts

MCF7 breast cancer cells alone or mixed with WT or Hsf1 null primary MEFs were injected subcutaneously into NOD-scid mice. The experiment was repeated twice, with 4 mice per group in each experiment. (A) The mean tumor volume (total 8 per treatment group) is shown. The distribution of individual measurements is shown in the lower panels, in scatter plots for days 22 and 38 post injection. Error bars, SEM. *p<0.05, **p<0.01. (B) Mice were sacrificed when tumor burden reached size limit and the tumors were excised, fixed and stained with hematoxylin & eosin (H&E, upper panels) or Masson’s trichrome stain (lower panels). All images collected at the same magnification. Scale bar = 50μm. See also Figure S2.

Figure 3. HSF1 in fibroblasts supports cancer cell growth by activating gene expression programs both in cancer cells and in fibroblasts

(A–B) WT or Hsf1 null immortalized MEFs were treated with 10 μg/ml mitomycin C. D2A1 mouse mammary tumor cells stably expressing dsRed (D2A1-dsRed) were seeded on top of the MEFs and allowed to grow for 72h–96h, after which cancer cells were either visualized by fluorescent microscopy (A) or trypsinized and quantitated by flow cytometry (B). The mean of 3 independent experiments is shown. Error bars, SEM ** p<0.005. (C–D) Total RNA was purified from duplicate cultures of D2A1 cancer cells grown with or without WT or Hsf1 null MEFs and sorted as described above. RNA was hybridized to Agilent microarrays, and relative gene expression levels were analyzed using cluster 3.0. For each gene, expression in D2A1 cells grown alone was set to 1, and the relative change in expression upon co-culture with WT or Hsf1 null MEFs was calculated. (C) Overlap of genes differentially expressed in D2A1 cancer cells in the presence of WT or Hsf1 null MEFs. (D) Heat-map depicting fold change in mRNA levels of genes differentially expressed in D2A1 cells grown in co-culture with WT versus Hsf1 null MEFs (in duplicate). Gene ontology (GO) enrichment is shown to the right of the panel. Groups a & c correspond to groups a & c in panel (C). (E) WT or Hsf1 null MEFs were co-cultured with D2A1-dsRed cells as described in (A), but not treated with mitomycin C. After 72–96h, cultures were sorted and mRNA was extracted and hybridized to Agilent microarrays. MEFs cultured without D2A1 cells and processed in the same manner served as controls. Gene expression was analyzed using cluster 3.0 and the differentially expressed genes were clustered into 4 groups. Gene ontology (GO) enrichment is shown to the right of the panel. (F) Gene set enrichment analysis of genes upregulated in WT vs Hsf1 null MEFs co-cultured with cancer cells (groups 1 & 4 in panel E). Enrichment was calculated for the indicated gene sets, and is presented as normalized enrichment score (NES). Statistically significant enrichment (false discovery rate (FDR) q-value<0.05) is shown in red, non-significant enrichment is shown in gray. See also Figure S3 and Tables S1–S5.

Figure 4. TGFβ and SDF1 mediate the support of cancer cell growth by stromal HSF1

(A) The relative expression of Sdf1, Tgfβ1 and Tgfβ2 in WT or Hsf1 null immortalized MEFs was measured by qPCR, and normalized to Gapdh. The mean of 3 experiments is shown. Error bars, SEM. (B–C) WT or Hsf1 null immortalized MEFs were co-cultured with D2A1-dsRed cells as explained in Figure 3A, in the presence or absence of 10 ng/ml TGFβ1 and 100 ng/ml SDF1. After 96h, cells were either visualized by fluorescent microscopy (B) or quantitated by flow cytometry (C). The percentage of cancer cells in co-culture is presented. The experiment was repeated 3 times, in triplicate. Representative results of one experiment are shown as mean +/− SEM. (D) Immortalized WT or Hsf1 null mitomycin-treated MEFs were pretreated, or not, with LY2109761 for 30 minutes before co-culture with D2A1-dsRed cells. Cultures were continued for 72h, with daily supplementation of LY2109761 (or not, as control), and then analyzed as in (C). The experiment was repeated 3 times, in triplicate. Results are expressed as the mean relative number of cancer cells, normalized to non-drug-treated co-cultures with WT MEFs. Error bars, SEM. (E) Immortalized WT MEFs stably expressing shRNA hairpins targeting Smad2 (shSmad2) or GFP (shGFP) were co-cultured with D2A1 cells, treated and analyzed as in (C). The percentage of cancer cells in the co-culture is presented. (F) Chromatin immunoprecipitation (IP) was performed with anti-HSF1 antibodies using material prepared from MCF7 tumor xenografts. Normal rat-IgG served as a negative IP control. IPs were analyzed by qPCR with primers targeting potential heat shock elements in mouse Sdf1 and Tgfβ2. Primers targeting an intergenic region in the mouse DNA, not expected to be amplified, were used as a negative control. The experiment was repeated twice, tumors from 3 mice were used for each experiment. Representative results from one experiment are shown as mean +/− SEM, *p<0.05, **p<0.01. See also Figure S4.

Figure 5. Increased HSF1 activation in the stroma is associated with decreased survival in breast cancer patients

(A–C) Analysis of Hsf1 mRNA expression levels in the stroma of 53 breast cancer patients from (Finak et al., 2008). (A) The association between Hsf1 expression and tumor grade is presented in a box & whiskers plot. (B) Kaplan-Meier (KM) analysis of patients stratified by Hsf1 expression. (C) The correlation between Hsf1 expression and HER2 status is presented in a box & whiskers plot. (D) Breast cancer resections from 46 early-stage patients were stained with anti-HSF1 antibodies and scored for HSF1 protein activation (relative nuclear staining intensity) in the stroma by immunohistochemistry. Association of stromal HSF1 activation with disease-free survival was assessed by KM analysis. *p<0.05. See also Figure S5 and Table S6.

Figure 6. Increased HSF1 activation in the stroma is associated with decreased survival in lung cancer patients

(A) Lung cancer resections from 5 patients were stained with anti-HSF1 (brown), anti-SMA (brown) or a combination of both antibodies (HSF1 in brown; SMA in red). Representative images are shown. Scale bar = 20μm (B–C) Lung cancer resections from 72 patients with Stage I disease were stained with anti-HSF1 antibodies and scored for HSF1 activation in the stromal cells and in the cancer cells. (B) HSF1 stromal scores are correlated with disease-free survival by KM analysis. (C) KM analysis of disease-free survival for patients with concordant high or low HSF1 scores in both stromal cells and cancer cells. (D) Stromal HSF1 levels in KRAS mutant tumors (n=18) from the lung cancer cohort correlate with disease-free survival by KM analysis. See also Figure S6 and Table S7.