The multifaceted role of HSF1 in tumorigenesis

Abstract

Heat Shock Factor 1 (HSF1), the master transcriptional regulator of the heat shock response (HSR), was first cloned more than thirty years ago. Most early research interrogating the role that HSF1 plays in biology focused on its cytoprotective functions, as a factor that promotes the survival of organisms by protecting against the proteotoxicity associated with neurodegeneration and other pathological conditions. However, recent studies have revealed a deleterious role of HSF1, as a factor that is co-opted by cancer cells to promote their own survival to the detriment of the organism. In cancer, HSF1 operates in a multifaceted manner to promote oncogenic transformation, proliferation, metastatic dissemination, and anti-cancer drug resistance. Here we review our current understanding of HSF1 activation and function in malignant progression and discuss the potential for HSF1 inhibition as a novel anticancer strategy. Collectively, this ever-growing body of work points to a prominent role of HSF1 in nearly every aspect of carcinogenesis.

Introduction

The heat shock response (HSR) is an adaptive mechanism found in all of cellular life that functions to maintain the health of the proteome in times of elevated temperature and other forms of proteotoxic stress (Lindquist, 1986). The first report of the HSR, nearly 60 years ago, is a story of serendipitous scientific discovery (Ritossa, 1962). A malfunctioning incubator containing drosophila salivary glands overheats and reveals a new pattern of polytene chromosomal puffs– an established marker of active transcription. Within ten years, it became clear that this phenomenon was associated with a rapid and robust induction of specific transcripts. Because this transcriptional response could be controlled temporally by simply changing temperature, the HSR became a powerful and widely utilized system to study gene regulation (Guertin et al., 2010; Teves and Henikoff, 2013). Almost immediately upon proteotoxic insult, the cell redirects its gene expression machinery to transcribe and translate heat shock proteins (HSPs), which function as molecular chaperones that serve to maintain proteins in their folded and functional state. Over the years, the HSR has enabled the discovery of numerous fundamental principles of transcription regulation and continues to be used for this purpose today.

While the power of the HSR as a laboratory tool to study gene regulation has been undeniable, a major question remained: what is the role of the HSR in biology? We now recognize the importance of the HSR and HSPs in nearly all aspects of health and disease. In humans, defects in components of the HSR are associated with neurodegenerative diseases, inflammatory diseases, and numerous rare diseases driven by the misfolding and loss of function of individual mutant proteins (Roth and Balch, 2011; Labbadia and Morimoto, 2015). Tumors, on the other hand, have an exquisite dependence on the HSR. This is most directly evident in the mutated oncoproteins themselves, where the mutations that yield an increase in activity come at the cost of an increased dependence on molecular chaperones (Jolly and Morimoto, 2000; Calderwood et al., 2006; Calderwood and Gong, 2016). Thus, the role of chaperones in cancer has an extensive history and still remains a topic of great interest.

Just over ten years ago, two seminal studies provided the most direct evidence for the dependence of cancers on the HSR by asking the following question (Dai et al., 2007; Min et al., 2007). If targeting individual chaperones holds promise as an anti-cancer therapy, what happens if the entire HSR is targeted? To do this, mouse models deficient in Heat Shock Factor 1 (HSF1), the master regulator of the HSR were employed. These initial studies revealed the critical role of HSF1 in enabling tumor formation. The ten years of research thereafter have been filled with discovery and important mechanistic insights, confirming a fundamental role for this factor in cancer biology. In this chapter, we will discuss our current understanding of the multitude of mechanisms by which HSF1 is activated and enables tumorigenesis by promoting cancer cell survival, proliferation, invasion, and metastasis.

HSF1 structure and function

HSF1 is the canonical member of the highly conserved HSF family of winged alpha helix transcription factors. In mammals, HSF1 contains five domains. The N-terminus contains the DNA-binding domain (DBD). This is followed by the oligomerization domain, which contains leucine zipper repeats 1–3 (LZ1–3; also referred to as heptad repeats HR-A and HR-B), the regulatory domain (RD), and a fourth leucine zipper repeat (LZ4 or HR-C) domain. Lastly, the C-terminus contains the transactivation domain, which interacts with the general transcription machinery to promote the release of promoter-proximal paused RNA Pol II and drive transcription elongation (Anckar and Sistonen, 2011; Neudegger et al., 2016; Vihervaara et al., 2017).

The DBD binds to DNA sequences called heat shock elements (HSE). HSEs are alternating inverted repeats of the sequence [nGAAn]. The C-terminus of the DBD wraps around the DNA and exposes the winged domain of HSF1. Thus, in contrast to other winged helix transcription factors, such as the ETS family of transcription factors with winged domains that directly bind DNA (Buchwalter, Gross and Wasylyk, 2004), the winged domain of HSF1 does not make contact with DNA (Gomez-Pastor, Burchfiel and Thiele, 2018). Rather, the surface of the winged domain remains exposed for protein-protein interactions and post-translational modifications, both of which can affect HSF1 activity. For example, in the absence of stress, EP300 acetylates HSF1 at K80, ablating the positive charge (Westerheide et al., 2009). This in turn reduces the affinity of HSF1 for DNA (Gomez-Pastor, Burchfiel and Thiele, 2018; Jaeger and Whitesell, 2018).

The LZ1–3 domain forms intermolecular hydrophobic interactions to mediate HSF1 oligomerization. The LZ1–3 domain can also form an intramolecular interaction with the LZ4 domain via hydrophobic and ionic contacts (Rabindran et al., 1993; Zuo et al., 1994). This intramolecular interaction inhibits HSF1 oligomerization by sequestering the LZ1–3 domain, preventing it from forming intermolecular interactions with the LZ1–3 domain of other HSF molecules. In mammalian cells, the oligomerization state of HSF1 is clearly important for its activity. Monomeric HSF1 does not bind DNA and is thus inactive as a transcription factor (Gomez-Pastor, Burchfiel and Thiele, 2018). The molecular mechanisms that govern these intrinsic conformational changes in response to stress are not well understood.

While a number of models have been proposed to explain HSF1 activation, a consensus is emerging around the chaperone titration model (Abravaya et al., 1992; Shi, Mosser and Morimoto, 1998; Zheng et al., 2016a). In this model, HSF1 is normally bound by chaperones in the cytoplasm. Upon proteotoxic stress, an increased number of misfolded protein substrates compete with HSF1 for chaperone binding, unleashing active HSF1 to drive gene expression. Active HSF1 promotes the transcription of chaperone genes to restore protein homeostasis. Once proteostasis is restored, chaperones are free to inactivate HSF1, completing the negative feedback loop (Shi, Mosser and Morimoto, 1998; Gomez-Pastor, Burchfiel and Thiele, 2018; Kijima et al., 2018).

HSF1 regulation by post translational modification:

The RD of HSF1 has long been known to undergo global hyperphosphorylation upon thermal stress that involves the simultaneous phosphorylation of at least 15 serine and threonine residues. While heat-induced global phosphorylation has been used as a marker for HSF1 activation, a series of recent studies demonstrated that this event is largely uncoupled from its transcriptional activity (Budzyński et al., 2015; Zheng et al., 2016b). Specifically, Budzynski et al. generated an HSF1 variant in which 15 S/T phosphorylation sites were simultaneously mutated to alanine within the RD. Surprisingly, this HSF1 mutant was still able to localize to the nucleus, bind HSEs, and increase HSP gene expression in response to acute proteotoxic stress (Budzyński et al., 2015). A subsequent study in yeast went a step further, in which 152 of all 153 serine/threonine residues of yHSF1 were simultaneously mutated to either alanine or aspartate (Zheng et al., 2016b). Remarkably, both of these variants were still functional and capable of driving gene expression during the HSR. These studies did reveal some differences between yeast and human HSF1– the phosphorylation-deficient mutant of human HSF1 moderately increases heat shock-induced transcriptional activity while the phosphorylation-deficient mutant of yeast HSF1 moderately reduces heat shock-induced transcriptional activity. Regardless, these studies collectively demonstrate that global phosphorylation is not necessary for HSF1 function in response to acute proteotoxic stress, but rather acts to fine-tune the transcriptional activity of HSF1 (Budzyński et al., 2015; Zheng et al., 2016b).

While clearly not required for heat shock induction, it is possible that phosphorylation of HSF1 enables it to sense and respond to the physiological stresses that accompany anabolic metabolism, biomass expansion, cellular proliferation, and other cell state fluctuations that occur in normal physiology and disease. These phosphorylation events, driven by diverse signaling pathways, can both promote and inhibit HSF1 activation. The phosphorylation events that positively regulate HSF1 include those on S230 and S320 mediated by calcium/calmodulin-dependent protein kinase II (CaMKII) and protein kinase A (PKA), respectively (Holmberg et al., 2001; Zhang et al., 2011). They also include phosphorylation of S326, which is often used as a marker of HSF1 activation even though it is not required for its activity (Guettouche et al., 2005; Chou et al., 2012). On the other hand, phosphorylation events that negatively regulate HSF1 activity include S121 by proinflammatory protein kinase MAPKAP kinase 2 (MK2) (Wang et al., 2006), S363 by c-Jun NH2-terminal kinase (JNK) (Dai et al., 2000), as well as S303 and S307 (Chu et al., 1998; Wang et al., 2003).

Other post-translational modifications (PTMs) in the regulatory domain, such as acetylation and SUMOylation, have also been identified that regulate HSF1. Moreover, there is evidence of crosstalk between individual PTMs that ultimately impact HSF1 activation. The most prominent example of this is an acetylation-sumoylation switch at K298 that is sensitive to the phosphorylation status of the neighboring S303. Acetylation of K298 prevents proteasome-dependent degradation and therefore increases HSF1 stability (Raychaudhuri et al., 2014). On the other hand, sumoylation of K298, which requires phosphorylation of S303, inhibits HSF1 activity (Hietakangas et al., 2003). These residues are contained within a bipartite ψKxExxSP motif, named PDSM (phosphorylation-dependent sumoylation motif), comprising a SUMO consensus site and a proline-directed phosphorylation site (Hietakangas et al., 2006). PDSM is highly conserved and found in numerous proteins, most notably other transcription regulators, including GATA-1, MEF2, and PPARγ (Hietakangas et al., 2006; Yang and Grégoire, 2006). Taken together, there does not seem to be any individual PTM that is sufficient to dramatically alter HSF1 activity– at least in response to thermal stress. Rather, the collective effect of these modifications is to tune HSF1 activity (Anckar and Sistonen, 2011). There is still much left to discover with regard to the nature of this PTM combinatorial code, and the precise effects on HSF1 activation, especially in the context of disease-relevant states.

HSF1 in carcinogenesis

Two landmark publications provided the first compelling evidence for a direct role of HSF1 in malignancy. In the first study, Hsf1 loss selectively suppressed the formation of lymphomas in a p53-deficient mouse model (Min et al., 2007). In the second study, Hsf1 loss dramatically reduced the susceptibility to tumor formation driven by oncogenic Ras in a classical chemical skin carcinogenesis mouse model, and by a tumor suppressor p53 hotspot mutation (Dai et al., 2007). In this study, Hsf1 knockout mice had reduced numbers and volumes of tumors, and an increase in tumor-free survival (Dai et al., 2007). These initial studies highlight HSF1 as a prominent example of non-oncogene addiction and provide a rationale for targeting HSF1 as an anti-cancer strategy–exploiting the “addiction” of tumors to this evolutionarily conserved survival mechanism.

The role of HSF1 as a critical pro-tumorigenic factor has been corroborated in a number of additional murine models of cancer. These include a malignant peripheral neural sheath tumor (MPNST) model driven by p53 and Nf1 loss (Dai et al., 2012), a mammary tumorigenesis model driven by Her2/Neu overexpression (Xi et al., 2012), a Hepatocellular Carcinoma (HCC) mouse model driven by chemical carcinogenesis (Jin, Moskophidis and Mivechi, 2011), and a T-cell Acute Leukemia (T-ALL) model driven by oncogenic Notch1 (Kourtis et al., 2018). In all cases, Hsf1 deletion resulted in a profound reduction in tumor burden, and a corresponding increase in survival of the host.

In humans, elevated nuclear expression of HSF1 is common across diverse types of cancer. These include carcinomas of the breast, cervix, colon, liver, lung, pancreas, and prostate (Dudeja et al., 2011; Santagata et al., 2011; Fang, Chang and Yang, 2012; Mendillo et al., 2012). In addition, these also include mesenchymal tumors such as meningioma (Mendillo et al., 2012), and hematopoietic malignancies such as multiple myeloma and T-ALL (Heimberger et al., 2013; Kourtis et al., 2018).

While HSF1 is expressed in diverse types of cancers, its expression within tumors of any individual type of cancer is heterogeneous. For example, nearly half of breast tumor samples from 1,841 women who participated in the Nurses’ Health Study have elevated and uniform levels of nuclear HSF1 protein expression (Santagata et al., 2011). Around 30% of the samples have either weak or heterogenous nuclear HSF1 expression. The remaining 20% of the samples are negative for nuclear HSF1 expression. In this study, HSF1 overexpression and nuclear localization were strongly associated with reduced survival. These general trends– heterogeneity of HSF1 expression in tumors within a subtype, with high expressing tumors exhibiting more aggressive clinical behavior– are found in many other types of tumors (Santagata et al., 2011; Fang, Chang and Yang, 2012; Mendillo et al., 2012; Engerud et al., 2014; Liao et al., 2015; Liang et al., 2017; Zhou et al., 2017; Wan et al., 2018). Taken together, these studies using mouse models and human tumor specimens demonstrate the extraordinary breadth and importance of HSF1 activation across a diverse spectrum of cancers.

How is HSF1 activated in cancer?

The mechanisms by which HSF1 is activated in cancers appear to be as diverse as the tumor types in which it operates. HSF1, long known for its ability to respond to diverse proteotoxic insults, is clearly well situated to respond to many of the stresses that arise during tumorigenesis. These include cell-autonomous proteotoxic stresses, such as those that accompany increased rates of protein synthesis (Santagata et al., 2013), aneuploidy (Oromendia, Dodgson and Amon, 2012; Santagata et al., 2013), and misfolded proteins arising from genetic mutations. These also include non-cell-autonomous stresses, such as those that accompany hypoxia, altered nutrient availability, and inflammation (Luo, Solimini and Elledge, 2009). In these scenarios, the mechanistic basis of HSF1 activation is similar to that which occurs during the heat shock response, i.e. HSF1 is released from normal sequestration by chaperones either by an increase in unfolded substrates or by an increase in protein synthesis and its requisite HSP70-dependent co-translational processing.

Recent work has revealed a multitude of mechanisms by which tumors can seemingly circumvent the requirement of elevated proteotoxic stress and activate HSF1 directly by increasing HSF1 expression, nuclear localization, or stability. Several studies demonstrate that oncogenic signaling pathways activated by either oncogene activation or tumor suppressor loss can directly modulate HSF1 activity, lending support for a proactive mode of activation rather than the canonical stress-sensing mode of activation. HER2 (ERBB2), the oncogenic driver of the eponymous subtype of breast cancers (Slamon et al., 1989), promotes HSF1 activation by at least two mechanisms. HER2 drives PI3K/AKT signaling, which in turn promotes the phosphorylation and inactivation of glycogen synthase kinase-3 beta (GSK3B), which ordinarily phosphorylates HSF1 on S303/307 to inhibit its activity (Chu et al., 1996, 1998; Khaleque et al., 2005). Moreover, PI3K/AKT signaling drives mTOR activation, which in turn phosphorylates HSF1 at S326 and enhances HSF1 transcriptional activity (Chou et al., 2012; Schulz et al., 2014). HSF1 activation driven by this HER2-initiated signaling cascade is of critical importance. In HER2-driven breast cancer cell lines, HSF1 promotes proliferation, migration, mammosphere formation, and xenograft tumor formation, while reducing senescence and apoptosis (Meng, Gabai and Sherman, 2010; Xi et al., 2012; Carpenter et al., 2015). Likewise, in a HER2+ mouse model, HSF1 enables tumor formation, vascularization, and lung metastasis (Gabai et al., 2012; Xi et al., 2012).

The Mitogen Activated Protein Kinase (MAPK) pathway is commonly altered in multiple types of cancers, with activating mutations in RAF or RAS occurring most frequently. Mutations in RAS activate downstream effectors that include MEK, which interacts with and phosphorylates HSF1 at S326 (Figure 1) (Tang et al., 2015). In addition, loss of the tumor suppressor neurofibromatosis type 1 (NF1) can also activate Ras leading to increased HSF1 phosphorylation, trimerization, nuclear localization, and transcriptional activity (Figure 1) (Dai et al., 2012).

Figure 1. HSF1 activation by multiple mechanisms in cancer.

1) Chaperone Sequestration: Under basal conditions, HSF1 remains suppressed by heat shock proteins such as HSP70. In response to proteotoxic stress or increased levels of protein synthesis, HSF1 is titrated away from HSP70. Subsequently, HSF1 is phosphorylated, trimerizes and translocates to the nucleus. In the nucleus, HSF1 binds to HSEs of target genes and induces gene expression. 2) Oncogenic Signaling: Oncogenic signaling pathways activated by either oncogene activation or tumor suppressor loss can regulate HSF1 activity. HER2 drives PI3K/AKT signaling, which in turn promotes the phosphorylation and inactivation of GSK3β. GSK3β phosphorylates HSF1 on S303/307 to inhibit its activity. Activating mutations in RAS activate downstream effectors that include MEK, which interacts with and phosphorylates HSF1 on S326 leading to its activation. Loss of the tumor suppressor neurofibromatosis type 1 (NF1) can also activate RAS leading to increased levels of HSF1 phosphorylation, trimerization, nuclear localization, and transcriptional activation. Loss of tumor suppressor kinase LKB1 inhibits AMPK, which normally inhibits HSF1 via S121 phosphorylation. 3) DNA copy number: An increase in HSF1 gene copy number can increase HSF1 mRNA and protein levels. 4) mRNA expression levels: NOTCH1 binds directly to the promoter of the HSF1 gene leading to an increase in HSF1 mRNA and protein levels. 5) Protein stability: The F-box/WD repeat-containing protein 7 (FBXW7) is an E3 ubiquitin ligase that targets HSF1 for ubiquitylation and proteasomal degradation. Loss of FBXW7 in many cancers leads to an increase in HSF1 stability.

The tumor suppressor kinase LKB1 is inactivated in diverse human cancers that include lung cancer, cervical cancer, and melanoma (Zhou, Zhang and Marcus, 2014). LKB1 normally activates the metabolic stress sensor AMPK, which suppresses HSF1 activity through S121 phosphorylation, preventing its nuclear translocation, DNA-binding, and transcriptional activity. Loss of AMPK results in increased HSF1 activation (Dai et al., 2015),which in pancreatic ductal adenocarcinoma, promotes invasion and migration (Chen et al., 2017). Taken together, mutations in oncogenes and tumor suppressors converge to induce oncogenic signaling networks, which activate HSF1 to enable malignant progression (Figure 1).

Perhaps the most common mechanisms by which HSF1 is activated in cancer are those which simply increase HSF1 expression levels. One mechanism through which this increase occurs is due to an increase in the copy number of the HSF1 gene itself. (Figure 1). The HSF1 locus resides on chromosomal segment 8q24.3 of the human genome, which is among the most frequently amplified regions across all human cancers (Zhang et al., 2017). HSF1 amplification is likely the mechanistic basis for the increased HSF1 mRNA and protein levels found in ovarian, breast, and prostate cancers, among others (Santagata et al., 2011; Powell et al., 2016). In other cancers, HSF1 can be upregulated directly at the level of transcription. A recent study demonstrated that the oncogene NOTCH1, which is hyperactivated in T cell acute lymphoblastic leukemia (T-ALL), binds directly to the promoter of HSF1 to drive its transcription (Figure 1). This leads to increased HSF1 protein levels and consequently, increased HSF1-dependent transcription of HSPs. In addition, NOTCH1 also binds directly to many of these same HSP genes to independently drive their expression. This NOTCH1-HSF1-HSP feedforward loop is essential for T-ALL pathogenesis. Loss of HSF1 eradicates leukemia in mouse models of T-ALL, while sparing normal hematopoiesis. Moreover, disruption of this feedforward loop at any node, by depletion of NOTCH1, HSF1 or any of the downstream HSP targets, suppresses the growth of human T-ALL (Kourtis et al., 2018).

HSF1 nuclear protein levels can also be elevated by reducing the rate of HSF1 degradation. The F-box/WD repeat-containing protein 7 (FBXW7) is a component of the multi-subunit ubiquitin ligase (SCF), which functions in the ubiquitin-dependent proteasome degradation pathway. FBXW7 is a well-characterized tumor suppressor associated with multiple cancers, such as carcinomas of the breast, prostate, and pancreas, among others (Akhoondi et al., 2007). A recent study demonstrated that elevated nuclear HSF1 protein levels correlate with loss of FBXW7 in melanoma where HSF1 promotes metastatic and invasive properties. Mechanistically, FBXW7 is an E3 ubiquitin ligase that targets HSF1 for ubiquitylation and proteosomal degradation (Figure 1) (Kourtis et al., 2015). Given that FBXW7 is mutated in many cancers, it is likely this mechanism of increasing the levels of active HSF1 is relevant in other types of cancers. More broadly, FBXW7-dependent degradation of HSF1 may also be a contributing factor in neurodegenerative diseases such as Huntington Disease (Gomez-Pastor et al., 2017).

Importantly, the mechanisms described above are not inconsistent with the canonical chaperone-sequestration model of HSF1 activation. Rather, an increase in HSF1 levels simply reduces the degree to which chaperone sequestration is required to achieve the same level of HSF1 activity. In sum, HSF1 is not only activated in response to oncogenic stresses but is also activated by a number of other mechanisms, such as those mediated by oncogenic signaling and those that simply increase HSF1 expression levels, which collectively explain the breadth of HSF1 activation and function in tumorigenesis.

How does HSF1 support the malignant state in cancer?

HSF1 regulates a transcriptional network of classical HSPs and a wide array of other genes directly involved in many of the hallmark processes of cancer, including cancer cell proliferation, invasion, and energy metabolism (Hanahan and Weinberg, 2011). In contrast to HSF1-dependent transcription during heat shock (Mendillo et al., 2012) or viral infection (Filone et al., 2014), where the net effect of HSF1 activity is a profound induction of chaperones (e.g. HSF1-dependent steady-state mRNA levels of some HSP70s are increased hundreds of fold upon heat shock), HSF1-dependent transcription in cancer results in a more nuanced effect on transcription. In this scenario, HSF1 tunes the expression of targets that support a diverse array of biological processes. It is worth noting that the HSP and non-HSP target genes most sensitive to HSF1 activity have a moderate reduction (~2-fold) in steady-state mRNA levels after HSF1 depletion in whole population experiments. It is likely that there are subpopulations of cells where HSF1 has a more profound effect on transcription, and these cells might be particularly important in promoting aggressive cancer phenotypes and drug resistance. Lastly, a subset of genes that are bound by HSF1 directly have increased expression upon HSF1 depletion in cancer cells (Mendillo et al., 2012). One possible explanation is that HSF1 can suppress the expression of a subset of its targets through mechanisms that have yet to be defined. However, another possibility is that HSF1 still drives the gene expression of these targets, but HSF1 loss enables a more potent transcription factor to bind and drive gene expression to even higher levels. In sum, HSF1 rewires the cancer cell transcriptome, with implications that are discussed in more detail below.

HSF1 regulation of cancer cell proteostasis:

The HSP genes regulated by HSF1 include HSP70, HSP90 and other co-chaperones that are often expressed at elevated levels in cancers. While these chaperones will be covered in more detail in other chapters of this volume, we will briefly discuss how these genes may contribute to the HSF1 cancer program.

HSP70 has a well-established role in promoting survival through its regulation of apoptosis, senescence, and autophagy (Murphy, 2013). For example, HSP70 associates with the caspase recruitment domain (CARD) of Apaf1 and inhibits the formation of the apoptosome, which is normally required for activation of pro-caspase 9 (Alnemri et al., 2000; Green et al., 2000). In addition, HSP70 depletion leads to the release of cytochrome c and a decrease in lysosome integrity– the lysosome becomes permeable and releases cathepsin B, a protease that may activate caspases directly (Dudeja et al., 2009). In breast and pancreatic cancer, HSP70 depletion decreases cancer cell growth by significantly inducing cell death (Nylandsted et al., 2002; Phillips et al., 2007). Thus, it is conceivable that HSF1 suppresses cancer cell apoptosis by promoting HSP70 transcription. In support of this idea, Jacobs et al. reveals an HSF1-HSP70/BAG3 axis required to prevent apoptosis in colon cancers. Here, HSP70 and its nucleotide exchange factor BAG3 (Bcl-2 associated athanogene domain 3) interact with the anti-apoptotic mediator BCL2 to prevent its degradation resulting in reduced levels of apoptosis (Jacobs and Marnett, 2009). In addition, HSF1, by promoting HSP70 expression, inhibits the phosphorylation and activation of the pro-apoptotic, c-Jun N-terminal kinase (JNK), which is known to induce apoptosis (Jacobs and Marnett, 2007). Likewise, in chronic lymphocytic leukemia (CLL), inhibition of HSF1 and HSP70 induces apoptosis in vitro (Åkerfelt, Morimoto and Sistonen, 2010; Frezzato et al., 2019). Collectively, these studies demonstrate that HSF1 promotes cancer cell survival at least in part through its regulation of HSP70 (Figure 2).

Figure 2. Schematic of HSF1 targets and their role in malignancy.

HSF1 promotes carcinogenesis by activating the canonical heat shock proteins (HSPs) and non-canonical genes. HSF1 activates canonical HSPs such as HSP70 and HSP90. HSPs contribute to cancer programs by different mechanisms. HSP70 inhibits apoptosis; while HSP90 drives heterogeneity, which in turn can leads to tumor transformation and drug resistance. DNAJB8 can differentiate non-cancer stem cells into cancer-stem cells (CSC) by inducing expression of SOX2. HSF1 also promotes tumorigenesis by enhancing transcription of genes that encode non-HSP factors. HSF1 promotes the expression of HuR, which is involved in translation and/or mRNA stability. HSF1 also activates the expression of adhesion proteins such as vinculin, which promotes cell adhesion and spreading.

HSP90 is another chaperone protein with a well-characterized role in malignancy (Whitesell and Lindquist, 2005). HSP90 associates with substrates or “client proteins” involved in diverse cellular processes that include signal transduction, immune response, development and DNA repair. Instead of recognizing a specific sequence within a protein, HSP90 is thought to recognize structurally unstable conformations of client proteins (Taipale et al., 2012, 2014; Schopf, Biebl and Buchner, 2017). In addition, mutations in these proteins can increase conformational instability that render them more dependent on HSP90 and other chaperones to maintain their proper folding and activity (Sahni et al., 2015). In support of this idea, there is an extensive body of research that demonstrates the importance of HSP90 in chaperoning mutated oncoproteins critical in carcinogenesis (Whitesell and Lindquist, 2005; Jaeger and Whitesell, 2018). In fact, Geldanamycin and other members of the benzoquinone ansamycin class of HSP90 inhibitors were originally thought to directly inhibit the V-Src oncogene, which reflects the extraordinary dependence of V-Src and other oncogenic tyrosine kinases on HSP90 (Whitesell et al., 1994). Beyond stabilizing essential oncogenic clients, HSP90 may also impact tumorigenesis by promoting the evolution of heritable new traits. A large body of work in model organisms has established a role for HSP90 in promoting phenotypic robustness by masking the deleterious effects of destabilizing mutations and regulating the folding of a diverse spectrum of signaling proteins (Rutherford and Lindquist, 1998; Queitsch, Sangstert and Lindquist, 2002; Jarosz and Lindquist, 2010). Because of this, it has been described as an “evolutionary capacitor” due to its ability to store phenotypic variance, which can be released upon cellular stress (Jarosz, Taipale and Lindquist, 2010). This HSP90-mediated link between cellular stress and phenotypic diversification is likely to have important implications in human malignancies (Whitesell and Lindquist, 2005; Jarosz, 2016). In human models of breast cancer, modest HSP90 inhibition, which does not possess anticancer activity on its own, strongly impaired the emergence of resistance to hormone antagonists in cell culture and in mice (Whitesell et al., 2014). In another example, modest HSP90 inhibition exacerbated the chemosensitivity of cells that encode mutant Fanconi anemia pathway proteins (Karras et al., 2017). Thus, by promoting the transcription of HSP90 and other chaperones, HSF1 will likely affect the oncogenic signaling circuitry, heterogeneity and evolvability of human cancers (Figure 2) (Jaeger and Whitesell, 2018).

HSF1 regulation of mRNA processing and protein synthesis:

The direct targets regulated by HSF1 in malignancy include genes that are involved in mRNA processing and protein synthesis, processes that are often aberrantly regulated to support the increased levels of cell proliferation associated with malignant progression (Truitt and Ruggero, 2016). As one example, HSF1 directly regulates the splicing factor TRA2B (also known as also SFRS10) (Mendillo et al., 2012; Kajita et al., 2013). Mutation analysis of the TRA2B promoter revealed that two of three HSEs are particularly important for the induction of TRA2B in response to oxidative stress (Kajita et al., 2013). In breast, cervical, ovarian, and colon cancer, TRA2B upregulation has been suggested to play a role in metastasis by affecting the splicing of genes involved in proliferation and cell survival (Best et al., 2013). Beyond simply promoting proliferation, TRA2B has been reported to modulate other processes relevant to tumorigenesis, including lipid metabolism (Pihlajamäki et al., 2011) and developmental gene regulation (Figure 2) (Dichmann, Walentek and Harland, 2015).

The RNA-binding protein HuR (ELAVL1) provides another example of an important effector of the HSF1 cancer program that has been particularly well studied (Figure 2). HSF1 binds the promoter of HuR to directly promote its expression (Gabai et al., 2012; Mendillo et al., 2012; Chou et al., 2015; Holmes et al., 2018). In turn, HuR binds an AU-rich consensus motif in the 3’-untranslated region of RNA targets to enhance their translation and/or stability (Srikantan and Gorospe, 2012). HuR targets include hypoxia-inducible factor 1 (HIF-1), β-catenin, and Rictor which promote angiogenesis, invasion, stem cell renewal (Gabai et al., 2012; Chou et al., 2015; Holmes et al., 2018). Another HuR target is SIRT1 (Abdelmohsen et al., 2007; Gabai et al., 2012; Chou et al., 2015; Holmes et al., 2018), which can deacetylate HSF1 to increase its activity (Westerheide et al., 2009). This HSF1-HuR-SIRT1 circuit has been shown to promote HSF1 activity in response to DNA-damage-mediated senescence (Kim et al., 2012) and serves as one of several examples of a link between HSF1 and the response to DNA damage, further detailed below.

HSF1 regulation of DNA repair:

Recent work has revealed other roles for HSF1 in managing DNA repair, most prominently in response to genotoxic stress that arises from anticancer therapies. Fujimoto et al. reveal an HSF1-PARP13-PARP1 complex that is required for tumorigenesis. Here, HSF1 recruits PARP1 to chromatin as a ternary complex with PARP13. In response to DNA damage, PARP1 is released from the complex and is redirected to sites of DNA damage to promote DNA repair (Fujimoto et al., 2017). The loss of either HSF1 or PARP13 reduces PARP1 chromatin occupancy and the efficiency of homologous recombination repair (HRR). As an interesting extension of this work, the group went on to show that genotoxic stress that disrupts the HSF1-PARP13-PARP1 complex reduces HSP70 expression during the HSR suggesting crosstalk between genotoxic and proteotoxic stresses (Fujimoto et al., 2018). Ordinarily, the HSF1-PARP13-PARP1 complex binds to the HSP70 promoter to promote its expression during the HSR. The results of this study show that, in response to thermal stress, PARP1 is redistributed throughout the HSP70 locus resulting in HSP70 PARylation, which is required for HSF1 binding to the HSP70 promoter for optimal HSP70 induction. Further support for a role of HSF1 in maintaining genome integrity is the HSF1-BCL6-TOX axis in germinal center (GC) B cells. HSF1-dependent activation of BCL6 represses the expression of BLC6 targets that include TOX, a DNA binding protein of the HMG-box family that is involved in chromatin assembly, transcription, and replication. As a result of TOX repression, DNA repair mechanisms are enhanced in cancers with high levels of stresses leading to chemoresistance (Fernando et al., 2019). These studies reveal the multifaceted mechanisms by which HSF1 is involved in the maintenance of genome integrity in tumorigenesis.

HSF1 regulation of energy metabolism:

HSF1 has a role in regulating cellular metabolism, and thus may play a role in the aberrant metabolism that has long been recognized as a hallmark of malignancy. As one example, HSF1 directly increases the transcription of lactate dehydrogenase A (LDH-A), induces lactate production, and consequently promotes glycolysis (Dai et al., 2007; Zhao et al., 2009). This HSF1-mediated addiction to glucose can be exploited with the natural product englerin A (EA). EA induces insulin resistance, which deprives tumor cells access to glucose, and simultaneously activates protein kinase C-θ, which activates HSF1, leading to a lethal scenario in highly glycolytic tumors (Sourbier et al., 2013).

In mouse hepatocytes, HSF1 loss reduces the levels of NAD⁺, ATP, and glucose resulting in an increase in AMPK activation, and a reduction in mitochondrial respiration and lipid synthesis (Qiao et al., 2017). Mechanistically, this occurs at least in part because HSF1 directly promotes the transcription of NAMPT, which maintains intracellular NAD+ levels through the NAD⁺ salvage pathway. A very recent study has also revealed that HSF1 can inhibit AMPK activity independent of its transcriptional activity through a physical interaction that reduces the affinity of AMPK to AMP (Su et al., 2019). Because AMPK normally limits the activity of the lipogenic transcription factor SREBP1, HSF1 can promote lipogenesis through the expression of SREBP1 target genes that include fatty acid synthase (FASN) and low-density liporprotein receptor (LDLR). Interestingly, HSF1 can also directly regulate the transcription of FASN and LDLR (Mendillo et al., 2012) demonstrating that HSF1 uses both transcriptional and non-transcriptional mechanisms in a feed forward-like manner to regulate lipogenesis.

HSF1 regulation of cell motility, migration and adhesion:

HSF1 directly regulates the expression of a number of genes involved in cell motility, migration and adhesion, which may be particularly important in promoting cancer cell invasion and metastasis. In one study, overexpression of a constitutively active variant of HSF1 promoted the anchorage-independent growth, migration and metastatic dissemination of melanoma cells by directly suppressing the transcription of Vinculin (Toma-Jonik et al., 2015), a focal adhesion gene previously observed to suppress invasion and metastasis (Goldmann et al., 2013). A different study of the pro-metastatic function of HSF1 in melanoma highlighted ITGB3BP as another example of an HSF1 target gene involved in cell migration. However, this study suggested that HSF1 target genes involved in other processes, such as the proliferation gene CKS2, the metabolic enzyme MTHFD2, and canonical HSPs, among others, are also important (Figure 2). Taken together, these studies provide a plausible mechanistic basis for the identification of HSF1 as one of six metastasis promoting genes in a genome-wide screen for drivers of melanoma invasion as well as the observation that HSF1 is correlated with poor clinical outcomes in malignant melanoma (Scott et al., 2011). More broadly, because HSF1 directly regulates many of these pro-metastatic targets described above in other types of cancer, this same aspect of HSF1 regulation may explain the correlations observed between HSF1 activation and aggressive phenotypes in multiple myeloma, and cancers of the pancreas, prostate, lung and breast (Dudeja et al., 2011; Santagata et al., 2011; Fang, Chang and Yang, 2012; Mendillo et al., 2012; Nakamura et al., 2014; Toma-Jonik et al., 2015).

HSF1 regulation of cell state:

Finally, HSF1 may promote malignant progression by altering cancer stem cell (CSC) - like characteristics. There is a wealth of data that implicates CSCs in tumor formation, metastatic dissemination, and drug resistance (Scheel and Weinberg, 2012). A series of recent studies have reported elevated HSF1 expression in CSC-like cancer cells coincident with an increase in CSC markers that include CD44, SOX2, and Nanog (Wang et al., 2015; Yasuda et al., 2017; Kusumoto et al., 2018). Functionally, HSF1 promotes tumor sphere formation independent of cell proliferation, suggesting that these changes are due to changes in CSC-like characteristics (Wang et al., 2015). Mechanistically, it is possible that HSF1 mediates these effects through canonical regulation of HSPs. In support of this possibility, cellular stresses that include thermal and oxidative stress have been shown to differentiate non-CSC into CSC-like cancer cells. In this mechanism, HSF1 drives the expression of DNAJB8, a member of the HSP 40 family, which in turn is critical for SOX2 upregulation (Kusumoto et al., 2018). However, a non-mutually exclusive possibility is that HSF1 mediates these effects through the direct regulation of either developmental genes such as JARID2 or genes involved in other processes, such as the splicing factor TRA2B that has been reported to regulate the expression of the CSC marker CD44 (Figure 2). Future studies will be required to better understand the molecular mechanisms by which the complex network of HSF1 target genes contribute to promote tumorigenesis. Because most HSF1 target genes are also regulated by other factors, experiments that attempt to phenocopy the effects of HSF1 loss by knocking out its target genes are not optimal. Ideally, strategies should be employed that disrupt the regulatory circuitry of HSF1 at subsets of its targets (e.g. targeting the HSE of HSP genes directly) that leave other mechanisms of their regulation intact. While challenging, advances in genome engineering make these types of experiments an exciting possibility.

Outlook

The association of the HSF1 cancer program with anabolic cellular processes, metastases and death suggests an evolutionary origin distinct from cancer itself (Mendillo et al., 2012; Jaeger and Whitesell, 2018). Moreover, the broad range of cancer types in which HSF1 is activated lends further support that this program originated to support basic biological processes (Santagata et al., 2011; Mendillo et al., 2012). Recent work has revealed that the sole heat-shock factor in C. elegans drives an essential transcriptional program during development that is distinct from the heat shock response. In this context, the binding of E2F recruits HSF1 to genes containing a distinct consensus sequence comprising a GC-rich motif coupled to a degenerate HSE to drive a transcriptional program essential for C. elegans larval development. Thus, certain aspects of HSF1 function in cancer may derive from an ancient role in development that is conserved across species (Li et al., 2016). Related to this, HSF2, one of several HSF1 paralogs, has been reported to function as a stress sensor during development (Akerfelt, Morimoto and Sistonen, 2010). In addition, several recent studies demonstrate that HSF2 also has a role in tumorigenesis (Björk et al., 2016; Zhong et al., 2016). However, in contrast to the pro-tumorigenic HSF1, there is data to suggest that it can both suppress (Björk et al., 2016) and promote tumorigenesis (Zhong et al., 2016), depending on the context. Much remains to be understood regarding the role of HSF2 and potentially other HSFs in cancer, including whether there is an interplay in cancer cells between these HSFs, which can form hetero-oligomers (Alastalo et al., 2003; He et al., 2003; Jaeger et al., 2016).

The multitude of mechanisms by which HSF1 operates to support tumorigenesis make it an attractive target for cancer therapy. However, targeting transcription factors with small molecules is notoriously challenging. Because of this, defining the critical upstream regulatory nodes that feed into HSF1 provides an indirect strategy to identify pharmacologically tractable targets to disrupt the HSF1 cancer program. In fact, most reported inhibitors of HSF1 are the result of phenotypic screens that act through either critical co-factors or nodes upstream of HSF1. In one example, an inhibitor that targets eIF4A (Iwasaki, Floor and Ingolia, 2016), which is involved in cap recognition during translation initiation, was identified that leads to the inactivation of HSF1 (Santagata et al., 2013). In another example, an inhibitor that targets CDK9, which is involved in transcription elongation, was identified that leads to the inhibition of HSF1 activity during the HSR (Rye et al., 2016).

Several emerging strategies provide hope that we will be able to develop potent and specific inhibitors that directly target HSF1. In one recent example, molecular modeling was used to predict molecules that would bind the HSF1 DNA binding domain. A subsequent screen of candidates led to the development of HSFI115, which was shown to bind the DNA binding domain and inhibit HSF1 transcriptional activity (Vilaboa et al., 2017). In another recent pair of examples, dominant negative peptide screens resulted in the identification of peptides that target the DNA-binding domain, the trimerization domain (Dorrity, Queitsch and Fields, 2019), and the LZ4 domain of HSF1 (Ran et al., 2018), all of which led to the inhibition of HSF1 activity. Another emerging strategy is PROTAC (Proteolysis-Targeting Chimera), which harnesses the power of the ubiquitin-proteasome to selectively target and degrade proteins. A PROTAC is a bivalent molecule comprising one domain that selectively binds a target and another domain that binds an E3 ligase, resulting in the ubiquitination and degradation of the target protein (Pettersson and Crews, 2019). PROTAC could theoretically be coupled to molecules based on the molecular modeling and peptide screens described above to develop even more potent molecules that not only inactivate, but also degrade HSF1. This is all the more important considering that HSF1 is now appreciated to function in both the tumor cell and the tumor ecosystem (Scherz-Shouval et al., 2014); (discussed in an accompanying chapter) to impact nearly all aspects of tumorigenesis.