FBXW7 modulates cellular stress response and metastatic potential through ​HSF1 post-translational modification.

​Heat-shock factor 1 (​HSF1) orchestrates the heat-shock response in eukaryotes. Although this pathway has evolved to help cells adapt in the presence of challenging conditions, it is co-opted in cancer to support malignancy. However, the mechanisms that regulate ​HSF1 and thus cellular stress response are poorly understood. Here we show that the ubiquitin ligase ​FBXW7α interacts with ​HSF1 through a conserved motif phosphorylated by ​GSK3β and ​ERK1. ​FBXW7α ubiquitylates ​HSF1 and loss of ​FBXW7α results in impaired degradation of nuclear ​HSF1 and defective heat-shock response attenuation. ​FBXW7α is either mutated or transcriptionally downregulated in melanoma and ​HSF1 nuclear stabilization correlates with increased metastatic potential and disease progression. ​FBXW7α deficiency and subsequent ​HSF1 accumulation activates an invasion-supportive transcriptional program and enhances the metastatic potential of human melanoma cells. These findings identify a post-translational mechanism of regulation of the ​HSF1 transcriptional program both in the presence of exogenous stress and in cancer.

Organisms respond to stressors by activating adaptive mechanisms to restore homeostasis. Environmental and intrinsic factors elicit the highly conserved heat-shock response, orchestrated by the transcription factor HSF1. Upon stress, HSF1 induces gene expression of heat-shock proteins (HSPs), which act as molecular chaperones and restore protein homeostasis[1-3]. It has long been noted that cancer cells bolster their chaperone system to cope with stress caused by increased protein production due to aneuploidy, increased protein folding requirements and proteasome overwhelming[4]. HSF1 deficiency protects against tumorigenesis driven by different oncogenic stimuli[5-7]. In addition, depletion of HSF1, which itself is not a bona fide oncogene, decreases the viability of multiple cancer cell lines, a phenomenon coined as “non-oncogene addiction”[6-13]. Apart from its classic role as a major activator of chaperone-encoding genes, HSF1 also regulates a malignant-specific transcriptional program, critical for cancer cells and tumor microenvironment[14-16]. However, the signaling pathways modulating the HSF1 cancer-specific activity remain unknown.

Heat-shock response activation-attenuation is an intricate process as the HSF1 protein undergoes extensive post-translational modifications[17-19]. Protein stability controlled by the ubiquitinproteasome pathway is an emerging theme in human cancer. FBXW7, a substrate-targeting subunit of the SCF (Skp1-Cul1-F box) ubiquitin ligase complex[20] targets several key regulators of proliferation, growth and apoptosis for proteasomal degradation[21-29]. FBXW7 is mutated in a significant portion of diverse human cancers[30]. We investigate here the mode of post-translational regulation of HSF1 and demonstrate an interaction between FBXW7 and HSF1. We show that FBXW7α controls the stability of nuclear HSF1 and modulates the attenuation phase of the heat-shock response. Moreover, FBXW7 deficiency enhances the metastatic ability of melanoma via HSF1 stabilization and alteration of the HSF1 malignant transcriptional program. Altogether, our data suggest that a tumor suppressor, FBXW7, regulates heat-shock response and cancer cell stress response and metastatic potential via modification of HSF1.

HSF1 is a substrate of the FBXW7α ligase

To identify substrates of the ubiquitin ligase FBXW7α, we performed tandem affinity purification of FBXW7α and identified its interacting proteins by 2D LC-MS/MS (Fig. 1a; Supplementary Table 1). Interestingly, HSF1, similar to MYC, was detected in FBXW7α immunoprecipitates (Fig. 1b). However, the HSF1 interaction with a WD40 domain mutant FBXW7α, that lacks the ability to bind protein substrates but binds the Cullin 1 complex, was significantly reduced (Fig. 1b). In addition, endogenous FBXW7 and HSF1 were found to interact (Supplementary Fig. 1a). Analysis of HSF1 protein sequence revealed the presence of two conserved amino-acid sequences resembling the canonical FBXW7 degradation motif (degron) S/TPPXS/T[20], one of which (SPPQS), contains evolutionary conserved phosphoamino acids (Fig. 1c).

Figure 1

HSF1 is a substrate of the FBXW7α ubiquitin ligase

(a) Network of FBXW7α-interacting partners. Serial immunoprecipitation experiments from HEK293 cells coupled to mass-spectrometry based analysis revealed a large number of known substrates (NFKB2, MYC, MED13L, MED13), already characterized members of the Cullin 1 complex (SKP1, CUL1) and putative interactors (MED1, HSF1). The FBXW7 degrons on various substrates are indicated. (b) FBXW7α binds to HSF1 through specific residues in the WD40 domain. HEK293T cells were transfected with constructs encoding FLAG tagged HSF1, and FLAG-HA tagged empty vector (EV), or FLAG-HA tagged FBXW7α or FLAG-HA tagged FBXW7α (WD40), a substrate binding mutant, in which three residues within one of the seven WD40 repeats of FBXW7α have been mutated[57]. HA-tagged FBXW7α was immunoprecipitated (IP) from cell extracts with anti-HA resin, followed by immunoblotting as indicated. The left panel shows inputs. (c) Alignment of the human HSF1 protein region containing the putative degron with HSF1 from various organisms. Conserved phospho-amino acids (amino acids 303 and 307 in human sequence) are highlighted. Uncropped blots are shown in Supplementary Fig. 8.

We next mapped the FBXW7α -binding motif of human HSF1. To investigate which amino acids participate in the interaction of HSF1 with FBXW7α, we mutated the two putative FBXW7α degrons (amino acid positions 303-307 and 363-367). We found that an HSF1 mutant containing alanine substitutions at both Ser303 and Ser307 [HSF1(Ser303/307Ala)] failed to bind FBXW7α (Fig. 2a). Further mutational analysis revealed both Ser303 and Ser307 as necessary residues contributing to its interaction with FBXW7α (Fig. 2b). Notably, phosphorylation of HSF1 on Ser303/307 by GSK3β and ERK1 respectively, has been suggested to play a role in down-modulating HSF1's activity during recovery from stress[31-36]. Treatment of cells with a GSK3 inhibitor (GSK3i IX and XVI) or a MEK inhibitor (U0126) markedly decreased the affinity of FBXW7α for HSF1 (Fig. 2c, d; Supplementary Fig. 1b). This is consistent with the notion of a priming role for the Ser307 phosphorylation for subsequent phosphorylation on Ser303[37]. Moreover, depletion of FBXW7 increased the half-life of nuclear HSF1 (Fig. 2e).

Figure 2

HSF1 interacts with FBXW7α through a conserved degron sequence phosphorylated by GSK3β and ERK1

(a) HSF1 binds FBXW7α through a conserved degron. HEK293T cells were transfected with FLAG-HA tagged FBXW7α and constructs encoding FLAG tagged HSF1 or HSF1(Ser303/307Ala) or HSF1(Ser363/367Ala). HA-tagged FBXW7α was immunoprecipitated (IP) from cell extracts with anti-HA resin, followed by immunoblotting as indicated. The left panel shows inputs. (b) Both Ser303 and Ser307 in HSF1 are required for the interaction with FBXW7α. HEK293T cells were transfected with FLAG-HA tagged FBXW7α and constructs encoding FLAG tagged HSF1 or HSF1(Ser303/307Ala) or HSF1(Ser303Ala) or HSF1(Ser307Ala). HA-tagged FBXW7α was immunoprecipitated (IP) from cell extracts with anti-HA resin, followed by immunoblotting as indicated. (c) Interaction between HSF1 and FBXW7α depends on GSK3β activity. HEK293T cells were transfected with constructs encoding FLAG tagged HSF1 and FLAG-HA tagged FBXW7α. Cells were treated with GSK3i IX (10 μM for 10 h) or DMSO. HA-tagged FBXW7α was immunoprecipitated (IP) from cell extracts with anti-HA resin, followed by immunoblotting as indicated. (d) Interaction between HSF1 and FBXW7α depends on ERK1 activity. HEK293T cells were transfected with constructs encoding FLAG tagged HSF1 and FLAG-HA tagged FBXW7α. Cells were treated with MEK1 inhibitor U0126 (10 μM for 2 h) or DMSO. HA-tagged FBXW7α was immunoprecipitated (IP) from cell extracts with anti-HA resin, followed by immunoblotting as indicated. (e) FBXW7α controls the half-life of nuclear HSF1. HEK293T cells were infected with the indicated shRNA-encoding lentiviruses. Cells were treated with 2 μg/ml cycloheximide for the indicated length of time. Nuclear fractions were analyzed by immunoblotting as indicated. TATA-binding protein (TBP) was used as loading control. (f) FBXW7α depletion abolishes HSF1 ubiquitylation in vivo. HEK293T cells were transfected with constructs encoding FLAG tagged HSF1 and Histidine-Myc tagged ubiquitin and infected with the indicated shRNA-encoding lentiviruses. Cells were heat shocked at 42°C for 1 h to induce ubiquitylation. Histidine tagged proteins were immunoprecipitated from whole cell extracts with nickel (Ni)-NTA beads, followed by immunoblotting for HSF1. Uncropped blots are shown in Supplementary Fig. 8.

Finally, we tested the ubiquitylation status of HSF1. Upon heat stress, HSF1 poly-ubiquitylation was markedly increased, suggesting a tagging mechanism for rapid proteasomal degradation upon stress removal (Fig. 2f; Supplementary Fig. 1c). In sharp contrast, induced HSF1 polyubiquitylation was significantly reduced upon FBXW7 silencing (Fig. 2f; Supplementary Fig. 1c), suggesting that FBXW7 binds and directly controls HSF1 ubiquitylation.

Loss of FBXW7 leads to defective nuclear HSF1 clearance and prolonged heat-shock response

To determine whether FBXW7 affects HSF1 protein turnover, we utilized the human colorectal cancer cell line HCT116, an isogenic cell line containing a homozygous deletion (KO) of the FBXW7 gene and its wild type (WT), FBXW7-expressing counterpart[38]. HSF1 shuttles between the cytoplasm and the nucleus, but upon stress it accumulates in the nucleus[19, 39]. Subcellular fractionation analysis of HSF1 revealed identical cytoplasmic HSF1 protein levels in WT and FBXW7 KO cells under basal conditions (37°C; Supplementary Fig. 2a). Strikingly, increased baseline HSF1 nuclear levels were observed in FBXW7 KO cells, compared to WT cells (Fig. 3a), in agreement with the nuclear localization of FBXW7α protein. Next, we exposed both cell lines to heat shock and noticed a marked increase of nuclear HSF1 levels. We also observed a rapid reduction of nuclear HSF1 during the recovery period, in WT cells (Fig. 3a). In contrast, we noted a prolonged stabilization of nuclear HSF1 in FBXW7 KO cells (Fig. 3a). HSF1 mRNA levels did not change during heat shock or recovery (Supplementary Fig. 2b). Subsequently, we asked whether loss of FBXW7 affects the clearance of nuclear HSF1, accumulated upon exposure to MG132, a known proteotoxic stressor. We observed reduced degradation of HSF1 in FBXW7 KO cells, compared to WT cells, upon MG132 removal (Fig. 3b). However, we could not detect any significant difference in the cytoplasmic fraction (Supplementary Fig. 2c). Thus, FBXW7 deficiency results in defective clearance of nuclear HSF1 upon exogenous stress removal.

Figure 3

FBXW7 deficiency results in nuclear HSF1 accumulation and prolonged heat-shock response upon exposure to exogenous stress

(a) Loss of FBXW7 results in accumulation of nuclear HSF1 during recovery from heat shock. HCT116 WT and FBXW7 KO cells were heat shocked (42°C for 1 h) following recovery for the indicated time. Nuclear fractions were analyzed by immunoblotting as indicated. (b) Loss of FBXW7 results in accumulation of nuclear HSF1 during recovery from proteotoxic stress. HCT116 WT and FBXW7 KO cells were treated with MG132 (1 μM for 10 h) following recovery for 3 h. Nuclear fractions were analyzed by immunoblotting as indicated. (c) FBXW7 KO cells show defective attenuation of the heat-shock response pathway. Heat maps showing fold changes in expression of HSF1 targets comparing recovery (37°C for 2 h) to heat shock (42°C for 1 h) as determined by high-throughput RNA-Seq. The common targets of HSF1 in HCT116 WT and FBXW7 KO cells after heat shock, as revealed by ChIP-Seq analysis, are displayed on the heat map. Well-characterized genes, positively regulated by HSF1, are indicated. Each column represents a biological replicate. (d) Overlap of genes bound by HSF1 in HCT116 WT and FBXW7 KO cells, under basal conditions. (e) Representative ChIP-Seq tracks for common gene loci between WT and KO (HSPD1/E1, HSP90AB1) and unique for KO gene loci (EIF4A2, CCT6A). The scale corresponds to RPM. (f) Read density profile around TSS's of common HSF1 targets in WT and FBXW7 KO cells. (g) HSPD1 and HSP90AB1 mRNA expression in HCT116 WT and FBXW7 KO cells, under basal conditions (P<0.001 for WT versus KO; unpaired t-test). Error bars indicate mean ± SD, and n=3 independent experiments. Uncropped blots are shown in Supplementary Fig. 8.

We then hypothesized that stabilization of nuclear HSF1 in FBXW7 deficient cells might affect HSF1 target gene expression upon stress removal. We heat shocked the HCT116 WT and FBXW7 KO cells (42°C for 1 hour) and monitored the expression of the heat-shock-inducible HSPA6 gene after 2 hours of recovery at 37°C. In WT cells, HSPA6 expression was diminished during the recovery period. Strikingly, in the FBXW7 KO cells, HSPA6 expression was higher after recovery, compared to heat shock (Supplementary Fig. 3a). Identical results were obtained in HEK293T cells upon FBXW7 silencing (Supplementary Fig. 3b). To assess the change of HSF1 targets expression after heat shock and recovery, in a high throughput approach, RNA sequencing (RNA-Seq) was performed in HCT116 WT and FBXW7 KO cells. To identify direct HSF1 targets, we first performed chromatin immunoprecipitation coupled to high-throughput sequencing (ChIP-Seq) for HSF1 after heat-shock treatment of HCT116 (WT and KO) cells. We focused on high-confidence HSF1 targets and monitored their expression during heat shock and recovery in the two cell lines. In contrast to WT cells, the expression of positively regulated HSF1 targets, such as HSPH1, HSPB1 and HSPE1, continued to increase in FBXW7 KO cells during recovery from heat shock (Fig. 3c). Notably, the majority of common HSF1 heat-shock response targets (both positively and negatively regulated) in HCT116 WT and FBXW7 KO cells displayed opposite expression pattern during recovery from heat-shock treatment (Fig. 3c). However, HSF1 was expressed at similar basal levels in the two cell lines (Supplementary Fig. 3c). Thus, FBXW7 loss results in defective attenuation of the heat-shock response.

Loss of FBXW7 leads to aberrant HSF1 genome occupancy and protects cells from proteotoxic stress

To determine whether altered HSF1 stability can control genome occupancy by this transcription factor and subsequently alter HSF1 regulated transcriptional profiles, we performed ChIP-Seq analysis in WT and FBXW7 KO HCT116 cells under basal conditions. A small (52) number of gene loci were bound by HSF1 in HCT116 WT cells. In contrast, in the FBXW7 KO cells we identified a much larger number (485) of genes bound by HSF1 (Fig. 3d), suggesting that HSF1 protein increased stability leads to new binding sites on the genome (Fig. 3e, EIF4A2, CCT6A). The vast majority of genes bound by HSF1 in WT cells were also bound in FBXW7 KO cells. Notably, however, for many of the genes bound in both cell lines, promoter HSF1 binding (as defined by the peak density) was significantly stronger in the FBXW7 KO cells compared to WT cells (Fig. 3e, HSPD1/E1, HSP90AB1; 3f). Stronger binding of HSF1 in FBXW7 KO cells translated to higher expression of common with WT target genes (Fig. 3g).

To assess the functional significance of the altered HSF1 transcriptional program in the FBXW7 KO cells, we asked whether these cells have a survival advantage in the presence of proteotoxic stress. We monitored cell survival over 3 days in the presence of sublethal concentrations of the proteasome inhibitor MG132 and the HSP90 inhibitor Radicicol. FBXW7 KO cells were significantly more resistant to both proteotoxic stressors, compared to WT cells (Supplementary Fig. 4a). To test the FBXW7-HSF1 interplay in a non-cancer context, we examined mouse embryonic fibroblasts (MEFs) isolated from Fbxw7 conditional knockout mice[40]. We observed nuclear accumulation in Fbxw7 MEFs compared to Fbxw7 and WT counterparts (Supplementary Fig. 4b). We then assessed MEF survival upon exposure to proteotoxic drugs. We found that Fbxw7 MEFs displayed increased survival compared to Fbxw7 and WT MEFs (Supplementary Fig. 4c). Notably, the survival advantage was abolished upon downregulation of HSF1, indicating a stringent HSF1 requirement for protection against proteotoxicity (Supplementary Fig. 4c).

HSF1 protein levels and gene target expression are associated with metastasis in melanoma

To further investigate the role of FBXW7-HSF1 interplay in human cancer progression and metastasis, we focused on melanoma, a solid tumor in which FBXW7 is frequently mutated and inactivated[23, 30, 41]. We tested HSF1 protein expression using immunohistochemistry in human specimens. We found that HSF1 protein levels were significantly increased as melanoma progressed to metastatic disease (P<0.001 for metastatic versus primary or nevi) (Fig. 4a). To assess the functional significance of HSF1 accumulation in melanoma, we asked whether HSF1 gene targets were upregulated during disease progression. Previously characterized HSF1 targets, including HSPD1, HSPE1, HSPH1 and CKS2 were expressed in significantly higher levels in metastatic melanoma compared to primary melanoma and normal skin (Fig. 4b). In contrast, FBXW7 expression was significantly reduced in metastatic melanoma, while HSF1 mRNA expression did not change significantly (Fig. 4b), suggesting post-translational regulation. In agreement with this model, we found that FBXW7 protein expression shows opposite pattern to HSF1 expression during disease progression (Figure 4c). To further investigate the hypothesis that HSF1 activation is associated with poor disease outcome, we expanded the primary human melanoma specimens cohort. These tumors were scored for levels of HSF1 and survival outcomes were investigated. Melanoma patients whose primary tumor expressed high levels of nuclear HSF1 had significantly decreased recurrence-free survival relative to the patients with low HSF1 levels (P=0.01, median follow up 5 years; Fig. 4d). These studies connect the FBXW7:HSF1 interaction to melanoma metastasis and disease progression.

Figure 4

HSF1 protein levels and HSF1 targets expression are associated with metastasis and disease progression in melanoma

(a) IHC staining with anti-HSF1 antibody of the indicated tissue types (dysplastic nevi n=48, primary n=39 and metastatic n=45; P<0.001 for metastatic versus primary or nevi; unpaired t-test; scale bar, 200 μm). (b) Box plots showing expression of HSF1 targets (HSPD1, HSPE1, HSPH1, CKS2), FBXW7 and HSF1, derived from microarray analysis of normal skin (n=4), primary (n=42) and metastatic melanoma (n=40) samples[58]. Whiskers represent the upper and the lower limits of the range. Boxes represent the first and third quartile, and the line represents the median (unpaired t-test). (c) IHC staining with anti-FBXW7 and anti-HSF1 antibodies of the indicated tissue types (dysplastic nevi n=59, primary n=53 and metastatic n=53; P<0.001 for nevi FBXW7 versus HSF1, P<0.01 for primary FBXW7 versus HSF1, P<0.05 for metastatic FBXW7 versus HSF1; unpaired t-test; scale bar, 200 μm). (d) Kaplan-Meier survival curves of patients with tumors expressing high (2D, n = 76) or low levels of nuclear HSF1 (including 0, 1F and 1D, n = 101; P= 0.01; Log rank test). Staining was scored according to the intensity (0-2) and distribution (Focal <50%, Diffuse ≥50%). (e) FBXW7 expression inversely correlates with HSF1 mRNA signature. FBXW7α (NM_033632) expression levels were measured by RNA-Seq in 325 Skin Cutaneous Melanoma (SKCM) samples from TCGA. HSF1 mRNA signature was defined as 1,864 mRNA expressions that were significantly correlated with HSF1 expression (NM_005526) with Spearman Correlation Coefficient > 0.3 and FDR < 5% in TCGA samples. Among them, 830 mRNAs with positive correlations were defined as the positive signature, while 1,034 mRNAs with negative correlations were defined as the negative signature. The first Principal Component (PC1) of the HSF1 positive signature was negatively correlated with FBXW7α expression (Spearman Correlation Coefficient = -0.4; P< 0.001); while the PC1 of the HSF1 negative signature was positively correlated with FBXW7α expression (Spearman Correlation Coefficient = 0.42; P< 0.001).

Finally, to investigate the direct link between HSF1-regulated gene signature and FBXW7α expression, we analyzed human melanoma cases from the Cancer Genome Atlas (TCGA) and generated an HSF1 “gene signature” comprised of mRNAs positively or negatively correlated with HSF1 levels. HSF1 transcription factor exerts both positive and negative role on the regulation of its targets and mRNAs positively correlated with HSF1 expression comprised the “positive signature” while mRNAs negatively correlated with HSF1 expression comprised the “negative signature”. In complete agreement with our proposed model, the HSF1 “positive signature” is negatively correlated with FBXW7α expression (P<0.001), while the “negative signature” is positively associated with FBXW7α expression (P<0.001; Fig. 4e).

FBXW7 controls HSF1 protein stability in human melanoma

We next monitored a panel of melanoma cell lines for nuclear HSF1 levels. Notably, melanoma cells deficient for FBXW7 (WM39, WM3862, WC00125) had significantly higher levels of nuclear HSF1, compared to WT cell lines (SKMEL24, SKMEL5, SKMEL28, COLO829; Fig. 5a). Furthermore, FBXW7 siRNA-mediated knockdown resulted in nuclear HSF1 accumulation in 501mel melanoma cell line (Fig. 5b). Interestingly, FBXW7 knockdown did not have any effect on cytoplasmic HSF1 levels (Supplementary Fig. 5a). We reasoned that the increased HSF1 nuclear accumulation upon FBXW7 knockdown might translate to elevated expression of HSF1 gene targets. Indeed, we were able to show increased expression of classical-HSF1 targets upon FBXW7 knockdown (Fig. 5c; Supplementary Fig. 5b).

Figure 5

FBXW7 regulates nuclear HSF1 levels and invasion ability in human melanoma

(a) FBXW7 deficiency stabilizes nuclear HSF1 in melanoma. Nuclear fractions from wild type (COLO829, SKMEL28, SKMEL5, SKMEL24) and deficient for FBXW7 (WC00125, WM3862, WM39) melanoma cell lines were analyzed by immunoblotting as indicated. (b) FBXW7 knockdown results in nuclear HSF1 accumulation. 501mel cells were treated with non-coding (NC) siRNA or siRNA against FBXW7. Nuclear fractions were analyzed by immunoblotting as indicated. (c) FBXW7 knockdown results in increased HSF1 targets expression in melanoma. DNAJB1 and HSP90AB1 mRNA expression in 501mel cells treated with non-coding (NC) siRNA or siRNA against FBXW7 (P<0.001 for NC siRNA versus FBXW7 siRNA; unpaired t-test, and n=3 independent experiments) (d) 451Lu cells were treated with the BRAF inhibitor Vemurafenib (2 μM, 9 h) and MEK inhibitor Trametinib (50 nM, 9 h). Nuclear fractions were analyzed by immunoblotting as indicated. (e) FBXW7 depletion results in increased invasion in melanoma. 451Lu, SKMEL239 and A375 cells were infected with the indicated shRNA-encoding lentiviruses. One week after transduction, trans-well invasion assay was performed (P<0.05 for 451Lu shLUC versus shFBXW7, P<0.001 for SKMEL239 scrambled versus shFBXW7, P>0.05 for A375 shLUC versus shFBXW7; n= 10 fields per biological replicate; 4 biological replicates; unpaired t-test). Nuclear fractions were analyzed by immunoblotting as indicated. (f) Representative microphotographs of 451Lu and A375 cells after 8 days of infection with the indicated shRNA-encoding lentiviruses (scale bar, 200 μm). Error bars indicate mean ± SD.. Uncropped blots are shown in Supplementary Fig. 8.

As we demonstrated earlier, the FBXW7:HSF1 interaction and substrate degradation requires phosphorylation of HSF1 by GSK3β and ERK1 (Fig. 2c, d). The RAS/RAF/MEK/ERK signaling pathway is hyperactivated in a large fraction of human melanoma[42]. To test whether the BRAF/MEK inhibition results in alteration of HSF1 protein levels, we treated the BRAF mutant cell line 451Lu with the BRAF inhibitor Vemurafenib and the MEK inhibitor Trametinib. We observed stabilization of nuclear HSF1 upon treatment with either inhibitor (Fig. 5d). In contrast, cytoplasmic HSF1 levels remained unaffected upon treatment (Supplementary Fig. 5c).

To investigate whether FBXW7 deficient melanoma cells show aberrant nuclear HSF1 clearance upon exogenous stress removal, we heat shocked WT (SKMEL28) and FBXW7 deficient (WM39) cells and monitored nuclear HSF1 levels. Both cell lines mounted an efficient heat-shock response as indicated by the accumulation of nuclear HSF1 (Supplementary Fig. 5d). During recovery, in SKMEL28 cells, nuclear HSF1 levels returned to untreated levels. In sharp contrast, in FBXW7-deficient WM39 cells, nuclear HSF1 levels remained stable during the recovery period (Supplementary Fig. 5d). Overexpression of FBXW7α partially restored HSF1 degradation during recovery (Supplementary Fig. 5d).

Accumulation of nuclear HSF1 upon FBXW7 loss enhances the metastatic potential in human melanoma

In a recent comparative genomic screen for the identification of oncogenic metastasis drivers[43] HSF1 was suggested to be pro-invasive. We hypothesized that downregulation of FBXW7, and consequently stabilization of HSF1, may impact metastasis in human melanoma. To test this hypothesis, we assessed in vitro the invasive capacity of melanoma cells upon FBXW7 knockdown. Notably, downregulation of FBXW7 resulted in significantly increased invasion in the melanoma cell line 451Lu, compared to control cells (Fig. 5e). In agreement with this finding, FBXW7 downregulation transformed the non-invasive melanoma cell line SKMEL239 into highly invasive (Fig. 5e). As a control, FBXW7 downregulation had no effect on the invasive capacity of A375 cells that express negligible levels of nuclear HSF1 (Fig. 5e). Cytoplasmic HSF1 levels were comparable in all cell lines (Supplementary Fig. 5e). Finally, to test whether HSF1 function is critical for the survival of melanoma, we modulated HSF1 expression using a previously validated shRNA[7]. The viability of nuclear HSF1-expressing 451Lu melanoma cells was strongly affected by the HSF1 silencing (Fig. 5f; Supplementary Fig. 5f, g). In sharp contrast, A375 cells were largely unaffected by HSF1 depletion (Fig. 5f; Supplementary Fig. 5f, g). Collectively, our data suggest that HSF1 is a critical substrate of FBXW7 and its accumulation upon FBXW7 downregulation modulates the invasive capacity of melanoma cells that are “addicted” to HSF1 for survival.

To investigate the effect of FBXW7 downregulation and subsequent nuclear HSF1 accumulation on metastatic capacity of melanoma in vivo, 451Lu cells transduced with shFBXW7 or shLUC were injected into immune-compromised recipient mice. Although FBXW7 knockdown did not affect primary tumor growth (Fig. 6a, b; Supplementary Fig. 6a), it led to dramatic increase in the formation of metastatic foci in the lungs (Fig. 6c, d). Primary tumors consisting of melanoma cells carrying shFBXW7 exhibited prominent nuclear HSF1 staining compared to their control counterparts. Ki-67 staining did not reveal any difference in proliferation rates between the two treatments (Fig. 6e). Also, FBXW7 downregulation did not affect differentiation of melanoma cells (Supplementary Fig. 6b).

Figure 6

Nuclear HSF1 accumulation upon FBXW7 depletion results in increased metastasis in vivo

In vivo metastasis assay with 451Lu cells transduced with control (shLUC) or FBXW7 shRNA injected subcutaneously in NOG/SCID mice. (a) FBXW7 knockdown does not affect tumor growth or (b) tumor mass (shLUC: n=8; shFBXW7: n=7; n corresponds to number of mice per condition; ns: non significant; unpaired t-test) Error bars indicate mean ± SD. (c) Macroscopic pictures of mouse lungs and H&E-stained sections of lung metastases at termination of the experiment. Black circles mark metastatic foci (scale bar, 100 μm). (d) Whisker plots show the number of metastases per lung. Mean ± SEM is depicted shLUC: n=8; shFBXW7: n=7; n corresponds to number of mice per condition).(****P<0.0001; unpaired t-test). (e) HSF1, Ki67 and H&E-stained sections of subcutaneous tumors resected at termination of the experiment show that FBXW7 knockdown induces HSF1 nuclear accumulation but not proliferation (scale bar, 100 μm). (f) HSF1 is necessary for the increased metastatic potential of melanoma cells upon FBXW7 depletion. 451Lu cells were transduced with control (shLUC) or FBXW7 shRNA. Melanoma cells that acquired increased metastatic potential were subsequently transduced with the indicated shRNA lentiviruses. Transwell invasion assay was performed 3 days after transduction (P<0.001 for shScr versus shHSF1; n= 10 fields per biological replicate; 4 biological replicates; unpaired t-test). Error bars indicate mean ± SD.

To verify that accumulation of other reported FBXW7 substrates does not impact the increased metastatic potential of melanoma cells upon FBXW7 depletion, we examined a panel of known substrates. Of the substrates tested, downregulation of FBXW7 resulted in significant stabilization of only HSF1 and Cyclin E in 451Lu cells (Supplementary Fig. 7a). Similar stabilization of Cyclin E upon FBXW7 downregulation was observed in the “non-HSF1 addicted” cell line A375, suggesting that nuclear accumulation of Cyclin E is not sufficient to alter the metastatic potential of melanoma cells (Supplementary Fig. 7b; Fig. 5e). In addition, overexpression of Cyclin E did not increase the invasive capacity of melanoma cells (Supplementary Fig. 7c), suggesting that HSF1 is the key FBXW7 substrate controlling melanoma metastasis.

We hypothesized that depletion of HSF1 may abolish the increased metastatic potential of melanoma cells with reduced FBXW7 expression. To test this hypothesis, we utilized the in vitro invasion system and used hairpins that only partially silence HSF1 to avoid the acute effects on cell survival. We were able to show that HSF1 deficiency completely suppressed the increased invasive potential of melanoma cells with reduced FBXW7 expression (Fig. 6f; Supplementary Fig. 7d, e). In contrast, Cyclin E (CCNE1) downregulation had no significant effect on invasion (Fig. 6f; Supplementary Fig. 7d, e). These studies connect the FBXW7:HSF1 module to human cancer metastasis and invasion.

HSF1 orchestrates a metastasis-promoting transcriptional program in melanoma

To gain insights into the molecular basis of HSF1 metastasis-supportive transcriptional program, we performed HSF1 ChIP-Seq analysis using the 451Lu melanoma cells. Initial analysis suggested a preference for promoter binding (Fig. 7a). Gene enrichment analysis (GSEA) of HSF1 targets using publicly available data[44] revealed significant enrichment for gene sets containing genes significantly altered in metastatic relative to primary melanomas (Fig. 7a, b). To better focus on bona fide melanoma metastasis regulators, we exploited recent studies that described a number of pro-invasive and oncogenic genes in melanoma[43]. Strikingly, our analysis revealed that several of these invasion/metastasis drivers are directly bound by HSF1 (Fig. 7b; MTHFD2, HMGB1 and ITGB3BP). We hypothesized that nuclear HSF1 accumulation upon FBXW7 knockdown may result in upregulation of the melanoma transcriptional program. To test this hypothesis, we monitored the expression of selected metastasis-related HSF1 targets, upon FBXW7 knockdown. FBXW7 silencing resulted in significant upregulation of such HSF1 targets (Fig. 7c). Based on these findings, we suggest that HSF1 orchestrates a metastasis-supportive transcriptional program that is bolstered upon nuclear accumulation due to FBXW7 deficiency/mutation.

Figure 7

HSF1 drives a metastatic-supportive transcriptional program that is affected by FBXW7 expression levels

(a) Genomic distribution of the regions of HSF1 occupancy (promoter, intragenic or intergenic). GSEA enrichment plot showing significant enrichment of the top 600 HSF1 targets for genes linked to metastasis in melanoma (Normalized Enrichment Score, NES=1.72; P=0.006). (b) Representative ChIP-Seq tracks of HSF1 metastasis-related targets. The scale corresponds to Reads Per Million (RPM). (c) mRNA expression analysis of HSF1 metastasis-related targets upon depletion of FBXW7 (P<0.01 for shLUC versus shFBXW7 for ADAM17 or ADAM22 or MTHFD2 and P<0.001 for shLUC versus shFBXW7 for HMGB1 or ITGB3BP; unpaired t-test). Error bars indicate mean ± SD, and n=3 independent experiments.

DISCUSSION

The present study sheds light on the regulation of the heat-shock response pathway attenuation phase through modulation of HSF1 stability by the ligase FBXW7. Moreover, it illustrates that FBXW7 modulates the malignant transcriptome via stabilization of HSF1. Since HSF1 provides critical stress relief and metastatic advantage in cancer, it is tempting to speculate that HSF1 stabilization is an important arm of the FBXW7 tumor suppressor activity in all tumors where FBXW7 mutations and genomic loses are reported. Also, given that FBXW7 function is frequently impaired in cancer by mechanisms other than mutations and allelic loss[41, 45, 46], we propose that the impact of HSF1 stabilization on cancer progression is broad.

Despite decades of study, HSF1 regulation during classic heat-shock response is not fully understood. The extensive array of post-translational modifications of HSF1 during the activation/attenuation cycle adds complexity. In the light of current knowledge, attenuation of the heat-shock response pathway involves weak interactions of HSF1 with HSPs and SIRT1-dependent control of HSF1 DNA binding activity[47-50]. Recently, a genome-wide RNAi screen suggested that the nuclear proteasome has a pivotal role in the regulation of the heat-shock response[51]. However, no specific ubiquitin complex was previously connected to HSF1 ubiquitylation. We were able to show here that FBXW7 links the activated HSF1 to the degradation machinery and FBXW7 activity affects both the classical role of HSF1 in the heat-shock response and its role in the orchestration of the cancer-supportive transcriptome (Supplementary Fig. 7f). It has been previously shown that HSF1 amino acids at positions 303 and 307 are important for the down-modulation of the heat-shock response[31-36]. Our study shows that these amino acids mediate the interaction of HSF1 with the CUL1:FBXW7 ubiquitylation complex. Importantly, we show that this interaction is controlled by molecular pathways frequently deregulated in cancer (GSK3β, ERK1). Since HSF1 is a powerful modifier of tumorigenesis[7, 52], it remains to be seen whether altered function of these kinases in different cancers converges on HSF1 stability.

Previous studies have suggested that HSF1 activation is associated with poor outcome in breast cancer[39]. Interestingly, reduced FBXW7 expression or inactivating mutations is significantly correlated with poor patient prognosis in multiple cancers [41, 53-56]. Our data unify these observations and prove that HSF1 accumulation due to FBXW7 altered expression provides an advantage in cancer cells during disease progression. Indeed, we show a direct correlation between FBXW7 and HSF1 protein expression to increased metastatic potential and decreased recurrence-free survival of melanoma patients. These data suggest that HSF1 could be a biomarker and a therapeutic target in this disease.

METHODS

Protein Immunoprecipitation

HEK293 were treated for 6 h with 10 μM MG132 to allow the accumulation of FBXW7 interacting proteins. Cells were then lysed in lysis buffer (50 mM Tris-HCl, ph 7.5, 150 mM NaCl and 2 mM EDTA) containing invitrosol (Invitrogen), PMSF, NaF, NaVO3 and proteinase inhibitors (Sigma). Tandem purification was performed. First immunoprecipitation was carried out overnight with anti-FLAG agarose beads. Next day, beads were washed and FLAG-HA-FBXW7 was eluted with 3xFLAG peptide. The eluate was used for a second immunoprecipitation with anti-HA agarose beads. Beads were thoroughly washed and subsequently digested for MS analysis.

Trypsin Digestion for MS Analysis

50 μl of 50 mM Ammonium Bicarbonate solution were added to the beads for trypsin digestion. 2 μg of trypsin was added to each sample along with 2 mM CaCl2 and incubated at 37°C for 16 hours. Samples were centrifuged subsequently for 30 min at 14,000 rpm, and the supernatants were transferred to fresh tubes to be acidified with 90% formic acid (2% final) to stop proteolysis. The soluble peptide mixtures were collected for LC-MS/MS analysis.

Multidimensional Protein Identification Technology (MudPIT) and Tandem Mass Spectrometry

Peptide mixtures were pressure-loaded onto a 250 μm inner diameter (i.d.) fused-silica capillary packed first with 3 cm of 5 μm strong cation exchange material (Partisphere SCX, Whatman), followed by 3 cm of 10 μm C18 reverse phase (RP) particles (Magic C18 AQ 3 μm, Michrom. Bioresources, Auburn, CA). Loaded and washed microcapillaries were connected via a 2 μm filtered union (UpChurch Scientific) to a 100 μm i.d. column, which had been pulled to a 5 μm i.d. tip using a P-2000 CO2 laser puller (Sutter Instruments), then packed with 13 cm of 3 μm C18 reverse phase (RP) particles (Magic C18 AQ 3 μm, Michrom. Bioresources, Auburn, CA) and equilibrated in 5% acetonitrile, 0.1 % formic acid (Buffer A). This split-column was then installed inline with a NanoLC Eskigent HPLC pump. The flow rate of channel 2 was set at 300 nl/min for the organic gradient. The flow rate of channel 1 was set to 0.5 μl/min for the salt pulse. Fully automated 11-step chromatography runs were carried out. Three different elution buffers were used: 5% acetonitrile, 0.1% formic acid (Buffer A); 98% acetonitrile, 0.1% formic acid (Buffer B); and 0.5 M ammonium acetate, 5% acetonitrile, 0.1% formic acid (Buffer C). In such sequences of chromatographic events, peptides are sequentially eluted from the SCX resin to the RP resin by increasing salt steps (increase in Buffer C concentration), followed by organic gradients (increase in Buffer B concentration). The last chromatography step consists in a high salt wash with 100% Buffer C followed by acetonitrile gradient. The application of a 2.5 kV distal voltage electrosprayed the eluting peptides directly into a LTQ-Orbitrap XL mass spectrometer equipped with a nano-LC electrospray ionization source (ThermoFinnigan). Full MS spectra were recorded on the peptides over a 400 to 2000 m/z range by the Orbitrap, followed by five tandem mass (MS/MS) events sequentially generated by LTQ in a data-dependent manner on the first, second, third, and fourth most intense ions selected from the full MS spectrum (at 35% collision energy). Mass spectrometer scan functions and HPLC solvent gradients were controlled by the Xcalibur data system (ThermoFinnigan, San Jose, CA).

Database Search and Interpretation of MS/MS Datasets

Tandem mass spectra were extracted from raw files, and a binary classifier - previously trained on a manually validated data set - was used to remove the low quality MS/MS spectra. The remaining spectra were searched against a human protein database downloaded as FASTA-formatted sequences from IPI protein database (database released on January, 2010)[59]. To calculate confidence levels and false positive rates, we used a decoy database containing the reverse sequences of annotated protein sequences appended to the target database, and the SEQUEST algorithm[60] to find the best matching sequences from the combined database.

SEQUEST searches were done using the Integrated Proteomics Pipeline (IP2, Integrated Proteomics Inc., CA) on Intel Xeon X5450 X/3.0 PROC processor clusters running under the Linux operating system. The peptide mass search tolerance was set to 50 ppm. No differential modifications were considered. At least half tryptic status was imposed on the database search, so the search space included all candidate peptides whose theoretical mass fell within the 50 ppm mass tolerance window. The validity of peptide/spectrum matches was assessed in DTASelect2[61] using SEQUEST-defined parameters, the cross-correlation score (XCorr) and normalized difference in cross-correlation scores (DeltaCN). The search results were grouped by charge state (+1, +2, and +3) and tryptic status (fully tryptic, half-tryptic, and non-tryptic), resulting in 9 distinct sub-groups. In each one of the sub-groups, the distribution of XCorr and DeltaCN values for (a) direct and (b) decoy database hits was obtained, and the two subsets were separated by quadratic discriminant analysis. Outlier points in the two distributions (for example, matches with very low Xcorr but very high DeltaCN) were discarded. Full separation of the direct and decoy subsets is not generally possible; therefore, the discriminant score was set such that a false positive rate of 1% was determined based on the number of accepted decoy database peptides. This procedure was independently performed on each data subset, resulting in a false positive rate independent of tryptic status or charge state. In addition, a minimum sequence length of 7 amino acid residues was required, and each protein on the final list was supported by at least two independent peptide identifications unless otherwise specified. These additional requirements – especially the latter - resulted in the elimination of most decoy database and false positive hits, as these tended to be overwhelmingly present as proteins identified by single peptide matches. After this last filtering step, the false identification rate was reduced to below 1%.

Cell culture and drug treatment

Cell lines were grown at 37°C in an atmosphere of 5% CO2. Cell lines were obtained from ATCC, unless otherwise stated. Cell lines were identified as negative for mycoplasma contamination. Heat shock was performed at 42°C for 1 h. HEK293T cells were maintained in DMEM medium (Invitrogen) containing 10% (v/v) FBS and 1% (v/v) penicillin/streptomycin. HCT116 cells were cultured in McCoy's 5A (Gibco) containing 10% (v/v) FBS and 1% (v/v) penicillin/streptomycin. A375 cells were cultured in DMEM medium (Invitrogen) containing 10% (v/v) FBS and 1% (v/v) penicillin/streptomycin. 451Lu cells were cultured in TU2% medium containing 80% (v/v) MCDB153, 20% (v/v) Leibovitz's L-15 (Invitrogen), 2% (v/v) FBS, 5 μg/ml bovine insulin, 1.68 mM CaCl2 and 1% (v/v) penicillin/streptomycin. SKMEL239 cells were cultured in RPMI medium containing 10% (v/v) FBS and 1% (v/v) penicillin/streptomycin. 501mel, MM485 and Colo829 were maintained in RPMI 1640 (Gibco) containing 10% (v/v) FBS; SKMEL5 and SKMEL28 in EMEM (Lonza) containing 10% (v/v) FBS; SKMEL24 in MEM (Lonza) containing 10% (v/v) FBS; WC00125, WM3862 and WM39 in tumor media (MCDB153 containing 20% Leibovitz's L-15, 2% fetal bovine serum, 5 μg/ml insulin, 1.68 mM CaCl2). The following drugs were used: MG132 (Peptide Institute), Cycloheximide (Sigma), BIO ((2′Z,3′E)-6-bromoindirubin- 3′-oxime; GSK3 inhibitor IX; Calbiochem), U0126 (1,4-diamino-2,3-dicyano-1,4-bis[2-aminophenylthio] butadiene; MEK1/2 inhibitor; Cell Signaling), Vemurafenib (PLX4032; BRAFV600E inhibitor; Selleckchem), Trametinib (GSK1120212;MEK1/2 inhibitor; Selleckchem).

Biochemical Methods

For immunoprecipitation experiments, HEK293T were transfected with 10 μg of each vector per 10 cm plate. Two days later, cells were treated with proteasome inhibitor (MG132, 10 μM) for 3 h and cells were lysed in 150 mM NaCl, 50 mM Tris (pH 8.0), 1% NP-40 supplemented with complete protease and phosphatase inhibitors (Roche). HA-tagged proteins were affinity purified from cleared lysates with anti-HA affinity gel (40 ul beads per confluent 10 cm plate; Sigma), washed thoroughly and boiled in SDS loading buffer with 1% β-mercaptoethanol for PAGE separation, followed by western blot analysis. Detection of protein ubiquitylation was carried out as described previously[62]. To quantify protein levels in subcellular compartments, fractionation was performed using subcellular protein fractionation kit for cultured cells (Thermo Scientific). 30 μg of whole cell lysate was loaded per lane, separated on a 4%–12% NuPage Bis-Tris polyacrylamide gel (Invitrogen) and subjected to western blot analysis. The following antibodies were used: FLAG M2 (1:5000, Sigma, F3165), Hsf1 (1:1000, Cell Signaling, 4356), c-Myc (1:1000, Cell Signaling, 9402), Skp1 (1:1000, BD, 610530), Myc-Tag (1:3000, Cell Signaling, 2272), a-Tubulin (1:5000, Santa Cruz, 8035), Lamin B (1:1000, Santa Cruz, 6217), Mcl-1 (1:1000, Santa Cruz, 819), mTOR (1:1000, Cell Signaling, 2972), SREBP-1 (1:1000, Santa Cruz, 8984), Cyclin E (1:1000, Santa Cruz, 247) and Notch1 (1:1000, Santa Cruz, 6014). Representative blots (Figs 1b, 2, 3a,b, 5a,b,d,e) are shown from three experiments.

Quantitative Real-Time PCR

For mRNA quantification, total RNA was isolated from one million cells for each condition and replicate using the RNeasy Plus Mini Kit (QIAGEN). RNA was quantified by absorbance at A260 nm and 1 μg of total RNA used for cDNA synthesis using Superscript III first strand synthesis kit (Invitrogen). Real time PCR reactions were carried out using SYBR Green Master Mix (Roche) and run with a Lightcycler 480 II (Roche). The following primer sequences were used for cDNA quantification: HSF1 for CCGTGTCCTGTGGTTTGGTT, rev CTGTCTTGTCCGTCCATCCA; HSPA6 for AGCAGTTGTGGCACTCAAG, rev TCACAGCTGACTTATCACGAAG; HSPD1 for ACGGCTTGCAAAACTTTCAG, rev TTAAGGGCATCTGTAACTCTGTC; HSP90AB1 for AGCAGCAGTAGTGGGACCAT, rev TAGAAGTGGCAGCAATCACG; DNAJB1 for GAGATCTTCGACCGCTACG, rev CCATGGAATGTGTAGCTGAAAG; HSPE1 for ACACTAGAGCAGAGTACGAGTC, rev CAGCACTCCTTTCAACCAATAC; HSPH1 for CACCAGAAAACCCAGACACT, rev GGGAGACTGTGAGGTTTGTT; CCNE1 for TCTTTGTCAGGTGTGGGGA, rev GAAATGGCCAAAATCGACAG; ADAM17 for TCTGAGAGCAAAGAATCAAGC, rev TCCTATTCCTGACCAGCGTG; ADAM22 for CAGTCTTGCCTCCATGTTCA, rev GCCTTTGGAACGTCATTCAT; HMGB1 for AGGATCTCCTTTGCCCATGT, rev TGAGCTCCATAGAGACAGCG; ITGB3BP for TTCTCAACTTTTGATAGCAACATCA, rev GACTCCCGACTTGACACGAT; MTHFD2 for CACTCTTCTACCTCCTGCCG, rev TTCGCCCTTTCCACCTC.

Immunohistochemistry

Immunohistochemistry was performed on Formalin-Fixed, Paraffin Embedded (FFPE) slides using the Avidin/Biotinylated enzyme system (VECTASTAIN ABC, Vector Laboratories). Heat induced epitope retrieval was performed using a pressure cooker with Tris-EDTA Buffer. Samples were provided by the Biospecimen Core of the NYU IMCG. The number of melanoma samples was selected based on the homogeneity of HSF1 staining in nevi, primary and metastatic samples. No statistical method was used to predetermine sample size. No samples were excluded or randomized. The study protocol was approved by the NYU Institutional Review Committee (IRB). All NYU patients signed informed consent. The anti-HSF1 antibody (1:10000, Thermo Scientific, RT-629-PABX). The anti-FBXW7 antibody (1:2000, Abcam, 109617). The slides were reviewed and scored by IMCG pathologist (FD) according to the intensity (0-2) of the staining as well as distribution (focal <50%, diffuse ≥50%). The IMCG pathologist was blinded while scoring the samples. In addition, rabbit anti-human Ki67, clone Sp6 (1:400; Thermoscientific; MA1-90584), mouse anti-human Melan-A, clone A103 (1:75; Dako; M7196) and mouse anti-human Tyrosinase, clone T311 (1:100; Biocare Medical; 155) were used.

FBXW7 Transient Knockdown

For transient knockdown of FBXW7, human melanoma cell lines were transfected with scrambled siRNA (D-001810-10, ON-TARGETplus Non-targeting Pool) or FBXW7-specific siRNA (L-004264-00) (ON-TARGETplus SMARTpool, Dharmacon, Lafayette, CO, USA) using JetPrime (Polyplus transfection, France). Seventy-two hours after siRNA transfection, cells were collected and total RNA was extracted for further analyses.

In vitro Invasion Assay

Cell migration was measured using transwell inserts (8 μm pores, BD Falcon). 451Lu cells (30.000 cells per insert) were suspended in serum-free medium over a Matrigel coating (Becton Dickinson), and medium supplemented with serum was used as chemoattractant. Cells that migrated after 12 hours were fixed in glutaraldehyde 0.1% solution, stained with 0.5% crystal violet, and counted in ten different fields using an inverted microscope. For each independent experiment, four replicates per condition were run. The average of cell counts from four inserts per condition was used for plotting results.

In vivo Metastasis Assay

451Lu cells transduced with shLuc or shFBXW7 lentiviruses were resuspended in growth media at a concentration of 2x 106 cells/150 uL, aliquoted into Eppendorf tubes (150 uL) and maintained on ice until injection. Immediately prior to injection, cell aliquots were mixed with 150 uL Matrigel (Becton Dickinson). Cell/Matrigel (1:1) suspensions were injected subcutaneously in the right flank of NOD/Shi-scid/IL-2Rγnull (NOG) 8 week old female mice (shLUC n=8; shFBXW7 n=7). No statistic method was used to predetermine sample size. The experiments were not randomized. No samples were excluded from the analysis. In vivo experiments were performed in compliance with a referenced protocol (120405-03) approved by the Institutional Animal Care and Use Committee (IACUC). When tumors were palpable (15 days), length and width measurements were made with calipers 3 times weekly until the animals were sacrificed. Tumor volume was calculated by the following formula: a2*b/2, where a is the width and b is the length. 40 days after subcutaneous injection, all animals were sacrificed to assess tumor mass and quantify lung metastasis. Tumors were extracted, weighed, formalin-fixed and paraffin-embedded. Lungs, liver, spleen, and kidney were removed for analysis of metastasis. Ventral and dorsal macroscopic images of metastasis-bearing lungs were taken with a fluorescent dissecting microscope equipped with a black and white camera. Pictures of the same lung were stitched and macroscopic metastases were quantified by counting lesions in both sides of lung per mouse. Representative images from tumors from 5 mice per condition are included. Data were plotted using GraphPad PRISM and significance determined by unpaired t test.

Detection of Cell Death

Cells were transduced with scrambled and shRNA targeting HSF1 as indicated, collected and directly stained by addition of Pacific Blue™-Annexin V (Biolegend) and 1 μg/mL propidium iodide (PI) 10 mM HEPES/NaOH (pH 7.4), 140 mM NaCl and 2.5 mM CaCl2 for 20 by flow cytometry (LSRII; Becton Dickinson) and the population with Annexin-V positive and PI negative staining was considered apoptotic. Representative images (5f) are shown from three experiments.

Survival Assays

Relative cell growth and survival was measured in 96-well microplate using the resazurin cell viability kit (Cell Signaling). Cells were plated at low density (~3000 cells/well) and exposed for 3 days to the indicated concentrations of proteotoxic drugs.

Statistics

Data are presented as the mean ± s.d. and were plotted using GraphPad PRISM. Significance was determined using an unpaired Student's t-test. For multiple comparisons, we used two-way ANOVA, corrected by the post-hoc Bonferroni test.

Chromatin Immunoprecipitation

For ChIP-Seq, 5×106 HCT116 cells were used for each immunoprecipitation. 5 μg of HSF1 antibody (Cell Signaling, 4356) was used per immunoprecipitation. ChIP and ChIP-Seq experiments were performed as described previously[63].

RNA-Sequencing Library Preparation

Total RNA was isolated from 1 million HCT116 cells using RNeasy Plus kit (Qiagen). Ribosomal RNA was depleted using RiboZero magnetic kit (Epicenter) according to manufacturer's protocol. cDNA preparation and strand-specific library construction was performed using the dUTP method as described previously[64]. Libraries were sequenced on the Illumina HiSeq 2000 using 50bp single end reads.

Data Sources and Computational Tools

Human assembly version hg19/GRCh37 and ENSEMBL annotations release 69 were used for the RNA-sequencing, ChIP-sequencing and data integration analyses. Bowtie[65] version 0.12.7 was used for alignment of sequenced reads. RNA-sequencing data analysis was performed using the DEGseq[66]. MACS[67] version 1.4.2 was used for ChIP-sequencing peak discovery. GenomicTools[68] version 2.7.2 was used for performing genomic interval mathematical operations, genomic interval annotations and ChIP-seq profile construction.

HSF1 Peak Identification

HSF1 ChIP-sequenced reads were aligned using bowtie (with default parameters, except for –m 1 so as to report only unique alignments) on human assembly version hg19. Peak discovery was performed with MACS using default parameters except for using a fragment size of 270bp estimated by Agilent 2100 Bioanalyzer. Sonicated input was used as a control for peak discovery. HSF1 peaks were characterized according to their genome-wide distribution into: (a) 1kb TSS-flanking regions of transcript isoforms, (b) gene body regions (excluding any overlapping regions with (a)), and (c) upstream regions of minimum 10kb and up to a maximum of 100kb (excluding any overlapping regions with (a) or (b)).

RNA-Sequencing Analysis

Differential gene expression analysis was performed for each matched recovery versus heat-shock pairs, separately in each biological replicate and cell line (WT or KO). Analysis was performed using DEGseq[66]. Input to DEGseq (i.e. gene RNA-seq counts) were computed using GenomicTools[68]. P-value cutoff (adjusted for multiple hypothesis testing) for differential expression was set at 1e-5, and genes presented in the heatmap have absolute log2 fold-change greater than 0.5 and FPKM value greater than 2.0 (in heat shock, recovery or both). To enforce more stringent criteria for significant differential expression, we ran DEGseq on each matched heat shock-recovery pair of biological replicates, and filtered out genes with inconsistent changes between replicates. The genes shown in the heatmap are the common HSF1 heat-shock response targets identified by ChIP-Seq analysis in HCT116 WT and FBXW7 KO cells and are significantly differentially expressed in recovery versus heat shock in at least one of the cell lines.

HSF1 mRNA signature and correlation analysis with FBXW7 expression

mRNA expression (level 3 normalized RNAseq results) of 325 Skin Cutaneous Melanoma (SKCM) cases were obtained from the Cancer Genome Atlas (TCGA) Data Portal (http://tcga-data.nci.nih.gov). HSF1 mRNA signature was defined as mRNA expressions that were significantly correlated with HSF1 expression (NM_005526) with Spearman Correlation Coefficient > 0.3 and FDR < 5% in TCGA samples. Among them, mRNAs with positive correlations were defined as the positive signature, while mRNAs with negative correlations were defined as the negative signature. Spearman correlation coefficient was used to examine the correlation between FBXW7 (NM_033632) expression and first Principal Component (PC1) of HSF1 positive signature and PC1 of HSF1 negative signature respectively.