NRF2 transcriptionally activates the heat shock factor 1 promoter under oxidative stress and affects survival and migration potential of MCF7 cells

Functional up-regulation of heat shock factor 1 (HSF1) activity through different posttranslational modifications has been implicated in the survival and proliferation of various cancers. It is increasingly recognized that the HSF1 gene is also up-regulated at the transcriptional level, a phenomenon correlated with poor prognosis for patients with different cancers, including breast cancer. Here, we analyzed the transcriptional up-regulation of HSF1 in human cells upon arsenite- or peroxide-induced oxidative stress. Sequential promoter truncation coupled with bioinformatics analysis revealed that this activation is mediated by two antioxidant response elements (AREs) located between 1707 and 1530 bp upstream of the transcription start site of the HSF1 gene. Using shRNA-mediated down-regulation, ChIP of NRF2, site-directed mutagenesis of the AREs, and DNase I footprinting of the HSF1 promoter, we confirmed that nuclear factor erythroid-derived 2-like 2 (NRF2, also known as NFE2L2) interacts with these AREs and up-regulates HSF1 expression. We also found that BRM/SWI2-related gene 1 (BRG1), a catalytic subunit of SWI2/SNF2-like chromatin remodeler, is involved in this process. We further show that NRF2-dependent HSF1 gene regulation plays a crucial role in cancer cell biology, as interference with NRF2-mediated HSF1 activation compromised survival, migration potential, and the epithelial-to-mesenchymal transition and autophagy in MCF7 breast cancer cells exposed to oxidative stress. Taken together, our findings unravel the mechanistic basis of HSF1 gene regulation in cancer cells and provide molecular evidence supporting a direct interaction between HSF1 and NRF2, critical regulators of two cytoprotective mechanisms exploited by cancer cells.

Healthy cells up-regulate heat shock factor 1 (HSF1) activity when challenged with a stress such as oxidative stress to protect its normal protein structure and function (1). HSF1 drives expression of a group of inducible heat shock protein chaperones (HSPs) such as HSP70 and HSP90 to refold the affected proteins to their native conformation or drive the removal of aggregates (1)(2)(3). HSF1 also controls the up-regulation of many genes associated with different cellular processes other than the HSPs (4 -6).
HSF1 exists in unstressed cells in an inactive state within an HSP90-immunophilin-p23 complex (7,8). Per a current model, a misfolded protein generated under stress titrates away the chaperones from the inactive complex, thereby freeing HSF1 to form a homotrimer. The trimeric complex binds to its recognition sequence, heat shock elements consisting of inverted 5Ј-nGAAn-3Ј motifs on the promoters of its target genes (9). Furthermore, various posttranslational modifications, such as phosphorylation, acetylation, and sumoylation, regulate the activity of HSF1. A noncoding RNA was shown to play an important role in regulating HSF1 function (10,11). Studies also suggested that HSF1 engages a distinct set of coactivators or corepressors to regulate genes in a stress-specific manner (12).
Normally, cellular HSF1 is inactivated as the stress is ameliorated. In contrast, cancer cells maintain HSF1 in a constitutively activated state to sustain their proliferative potential (13). This event is driven by the relatively higher level of oxidative stress in cancer cells caused by higher NADPH oxidase activity and higher mitochondrial rate of respiration associated with enhanced aerobic glycolysis and a defective antioxidant response (14). Many cancer cells, in particular, maintain an elevated level of the HSPs such as HSP90, HSP70, HSP40, and HSP27 because of overexpression of various oncoproteins. In fact, mice devoid of the HSF1 gene (hsf1Ϫ/Ϫ) are much less susceptible to chemical-induced transformation (15).
An up-regulated state of the HSF1 gene in different types of human cancers has been reported (16). HSF1 mRNA level was significantly up-regulated in oral squamous cell carcinoma, correlating the tumor size with a relatively higher level of HSF1 in the nuclear compartment (17). HSF1 was found to be overexpressed in the malignant prostate and in pancreatic cancer cells as well (13). Furthermore, HSF1 gene amplification and higher HSF1 levels were reported in hepatocellular carcinoma (18). Elevated HSF1 mRNA levels were correlated with higher grades of breast tumors (19). Although all these studies supported transcriptional up-regulation of HSF1 with cancer growth and proliferation, little is known about the underlying mechanism of this cancer cell-specific HSF1 regulatory mechanism.
Nuclear factor erythroid-derived 2-like 2 (also named NRF2) controls cellular redox potential in response to oxidative stress (20 -22). NRF2 in a healthy cell is constantly degraded through ubiquitination and proteasomal degradation by its association with KEAP1, a substrate adapter protein for the E3 ubiquitin ligase complex. Disruption of its binding with KEAP1 leads to stabilization and nuclear translocation of NRF2. NRF2 binds to the antioxidant response element (ARE) 3 on the promoter of its target antioxidant responsive genes as a heterodimer with a small Maf protein (20,(23)(24)(25).
Both NRF2 and HSF1 were reported to control certain genes such as heme oxygenase (26,27) and HSP70 (28) associated with cytoprotection. Increasing body of evidence suggested that these two factors also engage in mutual compensatory activities (20). However, it remains to be understood how these activities are precisely regulated.
We investigated HSF1 promoter regulation under oxidative stress in human cells induced by arsenite treatment as a model (condition) to obtain insight into the phenomenon observed in cancer cells. We identified two AREs in the promoter of the HSF1 gene responsible for its sensitivity to oxidative stress. We have further shown that the AREs engage in a multiprotein activation complex with NRF2 and the chromatin modifier BRG1 to mediate HSF1 transcription under arsenic stress. Furthermore, MCF7 cells appear to depend on this NRF2-dependent HSF1 transcription for their growth and proliferation.

Transcription of the HSF1 gene is up-regulated by a proteotoxic stressor
Up-regulation of the HSF1 gene was tested in cells in response to two forms of a proteotoxic stressor, heat shock or arsenite treatment. HT1080 cells were either grown at 37°C (control) or subjected to heat shock (HS) at 42°C for 1 h or treated with 20 M sodium arsenite for 2 h. The treated samples were then analyzed for HSP70, HSF1, and GAPDH levels by real-time PCR (qRT-PCR). These two treatment conditions led to a significant induction of HSP70 both at mRNA and protein levels ( Fig. 1A, right panel). An increase in HSF1 expression notably occurred in arsenite-treated samples only (Fig. 1A). Further experimentation will be required to better understand this HSP70 up-regulation but activation of HSF1 function by differential signaling pathways under these stressors has been reported (12). Notably, HSP70 was shown to be regulated by a few factors other than HSF1 under HS such as NF-Y, CREB, and NF-B (29,30). Up-regulation of HSF1 (and also its target HSP70 gene) both at mRNA and protein levels in different cancer cell lines, HeLa, MCF7, and PC3 but not in immortalized cell lines HEK293 and NKE under those treatment conditions measured by qRT-PCR (Fig. 1B) and immunoblotting (Fig. 1C), respectively, suggested that this is not specific to a particular cancer cell type. ROS, estimated by DCFDA fluorescence (Fig. 1B, right panel) in response to arsenite treatment, is also higher in MCF7 and HT1080 as compared with HEK293 and NKE cells.
To test whether the increase in HSF1 mRNA is because of its de novo RNA synthesis, and not because of an increased stability, we blocked transcription by treating cells with actinomycin D (ACTD), followed by HS or arsenite treatment as described above. Abrogation of arsenite-induced up-regulation of HSF1 and HSP70 mRNA and protein levels upon ACTD treatment suggested that HSF1 was indeed up-regulated at the transcriptional level by arsenite (Fig. 1D).

Identification of sequence elements present in the HSF1 gene promoter responsible for its up-regulation by arsenite treatment
To locate the enhancer element, several truncated versions of HSF1 gene promoter DNA, along with the full-length version, were cloned upstream of the Renilla luciferase reporter in a promoterless pGL4.79 vector ( Fig. 2A, left panel). The full-length promoter construct carries the HSF1 promoter sequence from Ϫ1931 to ϩ78 (ϩ1, transcriptional start site) (pHSF1-FL (Ϫ1931/ϩ78)). A reporter construct pHSP70 (Ϫ450/ϩ91) carrying from Ϫ450 to ϩ91 bp of the HSP70A1A promoter in the same vector was used as a positive control. The empty vector was used as a negative control. As revealed by luciferase mRNA level measured by qRT-PCR, cells transfected with pHSF1-FL (Ϫ1931/ϩ78) construct showed Ͼ7-fold induction with arsenite treatment ( Fig. 2A, right panel). HT1080 cells were chosen for this study because of a relatively low basal level of HSF1 mRNA in them. Assay of other constructs upon an identical treatment condition revealed that the region between Ϫ1720 and Ϫ1360 is responsible for the observed effect ( Fig. 2A, right panel). To identify transcription factors that can bind this region, the nucleotide sequence of HSF1 promoter spanning from Ϫ1931 to Ϫ1360 nucleotides was subjected to transcription factor-binding site analysis by rVista (rVista 2.0, https://rvista.dcode.org) 4 (71). This analysis predicted binding sites for hundreds of transcription factors including three putative binding sites of NRF2, a transcription factor reported to be up-regulated with oxidative stress (22,(31)(32)(33). However, there was no prediction for the presence of any heat shock element motif with confidence (data not shown). NRF2, a basic leucine zipper (bZIP) protein binds to AREs upon oxidative stress in the upstream promoter region of many antioxidative phase II genes like NQO1, HO-1, and TRX (34). The putative NRF2-binding sites ARE-A, ARE-B, and ARE-C on the HSF1 promoter including their predicted scores are shown (Fig. 2B). The nucleotide sequence in capital letters highlights the core-binding regions. Comparison of the consensus ARE sequence with those of HSF1, NQO1, and HO-1 gene promoters using Clustal Omega multiple sequence alignment tool identified distinct types of HSF1 AREs (Fig. 2C) (35)(36)(37)(38)(39).

Effect of NRF2 knockdown on HSF1 expression
To check the dependence of NRF2 for HSF1 induction by arsenite, HT1080 cells were transiently transfected with shRNA 3 4 Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party hosted site.

Control of HSF1 promoter by NRF2
against NRF2 or a scramble (pLKO.1-shNRF2 versus pLKO.1scr) to knock down endogenous NRF2. qRT-PCR analysis suggested that down-regulation of NRF2 prevented induction of HSF1 expression by arsenite (Fig. 3A). Efficient knockdown of NRF2 as determined by immunoblot (Fig. 3B) was reflected in inhibition of HO-1 gene expression (Fig. 3A, lower panel). Inhibition of arsenite-induced HSF1 expression under this condition was also shown by immunoblotting (Fig. 3C). Interestingly, the HSP70 transcript level under the same treatment condition was decreased to a lesser extent, which may reflect compensatory role of HSF1 for NRF2 (see "Discussion").

Functional validation of putative NRF2-binding sites in the HSF1 promoter
To functionally validate the importance of three putative ARE motifs on HSF1 promoter activity, reporter constructs were made in pGL4.79-mCMV vector, where one of these motifs (ARE-A, -B, or -C) was deleted one at a time (Fig. 3D). The NQO1-ARE was cloned in the same vector and used as an NRF2-sensitive positive control. HSP70 promoter cloned in the same vector was also used in the assay. Luciferase expression obtained (by assaying the luciferase mRNA levels) from these ARE-carrying constructs were then compared with a pHSF1-FL construct (Fig. 3D, left panel). As shown (Fig. 3D, middle panel), ARE-B and ARE-C are important for NRF2-dependent HSF1 up-regulation. When compared with the full-length reporter expression, deletion of ARE-B (compare pHSF1-ARE-C with pHSF1-FL; also compare pHSF1-ARE-C with pHSF1-ARE-BC) or ARE-C (compare pHSF1-ARE-AB with pHSF1-FL) reduced luciferase expression most, whereas deletion of ARE-A (compare pHSF1-ARE-BC with pHSF1-FL) did not significantly affect reporter induction by arsenite. Deletion of ; ns, not significant; *, p Ͻ 0.05; ***, p Ͻ 0.001. D, qRT-PCR analysis of HSF1 and HSP70 mRNA in HT1080 cells pre-treated with or without 1 g/ml ACTD for 2 h followed by treatment with HS or arsenite. In all assays, GAPDH mRNA level was used as normalization control. Representative immunoblots show the up-regulation HSF1 and HSP70 as indicated with estimation of their levels shown on the right. In all immunoblot experiments ␤-actin was used as internal reference. Error bar, mean Ϯ S.D. (n ϭ 3); ***, p Ͻ 0.001; ns, not significant.

Control of HSF1 promoter by NRF2
either ARE-B or ARE-C reduced the reporter expression level by ϳ49 and ϳ46%, respectively (Fig. 3D, middle panel). Notably, cells with down-regulated NRF2 showed a significant loss of reporter expression in all of the HSF1-promoter constructs (Fig. 3D, middle and right panels, open arrow) like NQO1-ARE-Luc construct (Fig. 3D, middle and right panels, solid arrow). These results suggest that NRF2 is involved in the induction of HSF1 gene expression by arsenite which acts through the ARE-B and ARE-C motifs.

NRF2 binds to predicted AREs of the HSF1 promoter in vitro
Electrophoretic mobility shift assay (EMSA) was carried out to examine the binding of NRF2 to ARE-A, ARE-B, and ARE-C elements of the HSF1 promoter. These motifs, along with their mutant versions were synthesized individually as 26-bp doublestranded oligonucleotides and labeled at their 5Ј-ends with [␥-32 P]ATP (used as probes). Nuclear extract (Nuc) isolated from HT1080 cells, untreated or pretreated with arsenite, were NRF2 binds an ARE as a heterodimer with a small Maf protein to elicit its transactivation function (31,33). Binding to an Figure 2. Analysis of HSF1 gene promoter elements responsible for its sensitivity to arsenite treatment. A, luciferase reporter assay was performed in HT1080 cells using full-length and different truncated versions of the HSF1 promoter as indicated by schematic drawing. Numbers indicate nucleotide positions with respect to ϩ1 as the transcriptional start site. qRT-PCR analysis of promoter activity (-fold increase in luciferase mRNA level) in arsenite-treated versus untreated samples expressed as relative luciferase mRNA levels represented by scatter plots. Error bar, mean Ϯ S.D. (n ϭ 3); ***, p Ͻ 0.001; ns, not significant. Relative luciferase mRNA level was normalized by measuring GFP mRNA level from a vector cotransfected with the reporter vector as a transfection control. B, putative AREs with corresponding scores, locations, and core sequences on the HSF1 promoter as revealed by rVISTA analysis (left panel). Schematic shows putative AREs with their relative positions on the HSF1 promoter (right panel). C, Clustal Omega alignment of nucleotide sequences of HSF1 promoter AREs with those of NQO1 and HO-1 promoters. The nucleotides matching the logo generated by the WebLogo program are marked by asterisks.

Control of HSF1 promoter by NRF2
ARE sequence was tested with a purified His 6 -NRF2 and His 6 -MafG protein mixed together in varying molar ratios. NRF2 with MafG in the molar ratio of 1:2 resulted in a visible complex with all the AREs (Fig. 4C, lanes 4, 7, and 10, arrowhead). The complexes formed on NQO1-ARE (using this ratio of NRF2 and MafG) were challenged with increasing concentrations of unlabeled NQO1-ARE, ARE-B, and ARE-C. As revealed, complex assembled on the ARE-C was less stable than that on the NQO1 and ARE-B (Fig. 4D). Direct measurement of the equilibrium dissociation constant (K D ) suggested the complex formed on ARE-B is of higher stability compared with that on ARE-C with K D of 0.983 M and 1.847 M, respectively, whereas the K D of the complex formed on the NQO1-ARE was 0.672 M (Fig. 4E).
A direct association of NRF2-MafG with the AREs is also reflected in the DNase I footprinting assay. As revealed, protection by the complex to DNase I digestion extended to nucleotides beyond the core motifs of all three AREs including the NQO1 (Fig. 4F). Interestingly, binding of NRF2-MafG complex induced DNase I hypersensitivity only at the ARE-B (asterisk).

NRF2 interacts with HSF1 promoter AREs in cells
Redox imbalance activates NRF2 in a cell which should in principle be neutralized by a ROS quencher, N-acetyl cysteine (NAC). This was essentially found when internal ROS levels were measured (Fig. 5A).

Control of HSF1 promoter by NRF2
To test if NRF2 is recruited to the AREs on the HSF1 promoter, chromatin immunoprecipitation (ChIP) analysis was performed in cells pretreated or untreated with arsenite in the presence or absence of NAC. For convenience, cells overexpressing FLAG-NRF2 were used. In the assay, ChIP-enriched DNA carried ARE-B and ARE-C regions only and NQO1-ARE (positive control), but not that of ARE-A. This was greatly reduced by NAC pre-treatment. This indicated that NRF2 asso-ciates only with ARE-B and ARE-C upon arsenite treatment. Results are summarized by bar graphs where enrichment of the untreated control samples was defined as 1 (Fig. 5B).
Increase in arsenite-induced HSF1 mRNA expression depended on ROS generation. Expression of HO-1, an NRF2 target gene, was monitored as a positive control in the experiment (Fig. 5C). Tertiary butylhydroquinone (TBHQ), an inducer of mitochondrial oxidative stress and NRF2 activity (40), effec-

Control of HSF1 promoter by NRF2
tively induced HSF1 promoter activity as tested by assaying the sensitivity of the HSF1 promoter-directed reporter and endogenous HSF1 mRNA levels (Fig. 5D). Azadiradione, an inducer of HSF1 activity (41), had little effect on HSF1 mRNA and HO-1 levels, although it significantly induces HSP70 mRNA (Fig. 5D). HSF1 mRNA induction is, therefore, a direct consequence of NRF2 activity caused by oxidative stress.

Arsenite-induced HSF1 up-regulation requires chromatin remodeler BRG1
NRF2, a member of the cap 'n' collar family of transcription factors, recruits CREB binding protein coactivator to its target gene promoters through cooperative interaction of its two domains Neh4 and Neh5 in a stress-independent manner (42,43). The Neh3 domain of NRF2 was shown to recruit CHD6 to the NQO1 gene both under basal and tert-butylhydroquinone induction (44). A BRG1-containing coactivator was shown to be required for NRF2-dependent activation of the HO-1, but not NQO1 gene (45). A possible involvement of a chromatinremodeling complex on the HSF1 promoter was tested by measuring the effect of down-regulation of subunits of different chromatin-remodeling complexes such as SSRP1 of the FACT complex, BRG1 (SMARCA4) of the SWI2/SNF2 complex, and SMARCA3 of the SWI/SNF complex (HLTF) using corresponding siRNAs or shRNAs (Fig. 6A). Among these, only BRG1 down-regulation affected arsenite-mediated HSF1 expression to a significant level (Fig. 6B). Reduced recruitment of RNApolymerase II specifically to the promoter region of HSF1 (HSF1pr) upon BRG1 inhibition, is indicative of NRF2 dependence on BRG1 for its function in response to arsenite stress (Fig. 6C).

Effect of NRF2-dependent HSF1 expression on proliferation and epithelial to mesenchymal transition of MCF7 cells induced by oxidative stress
Up-regulation of HSF1 gene expression at the levels of protein and mRNA was reported in breast cancer patients by a cohort study (19,46). Therefore, MCF7 cells were selected to validate the importance of NRF2-dependent HSF1 expression in a cellular function. HSF1 and NRF2 functions were downregulated by shRNA or all-trans retinoic acid (RA), respectively, alone or in combination, to assess their effects on cell morphology, proliferation, viability, as well as epithelial-mesenchymal transition (EMT) of MCF7 cells in response to arsenite treatment. NRF2 in the presence of RA is defective in ARE binding because it preferably forms complex with retinoic acid receptor ␣ when RA is present (47). The optimum concentra-

Control of HSF1 promoter by NRF2
tion of RA on NRF2 function inhibition was determined by its dose-dependent activity, i.e. inhibition versus toxicity study (Fig. S1). Dependence of MCF7 cells to NRF2-dependent HSF1 transcription demonstrated a drastic change in their morphology; MCF7 developed apoptotic vesicles when exposed to arsenite as well as when treated with HSF1 shRNA and RA (Fig.  S3, arrows). Inhibition of the factors NRF2 and HSF1, alone or in combination, also resulted in a marked alteration in the proliferation (Fig. 7A) and the viability (Fig. 7B) of the cells (Fig. 7C,  Fig. S3). Similar effect on HO-1 expression by arsenite treatment in cells pre-treated with RA or NRF2 shRNA suggested that these two treatments are functionally comparable (Fig.  7C). EMT, a crucial developmental event in cells, has been shown to correlate with the aggressiveness of cancer (48). To further assess the effect of HSF1 and NRF2 inhibition on tumorigenicity, the expression levels of two important EMT markers, E-cadherin and N-cadherin, were determined by immunoblotting (49,50). After 4 days of arsenite exposure, E-cadherin, an epithelial marker, decreased in the arsenite-treated set, but not so when both HSF1 and NRF2 were inhibited (Fig. 7D). The arsenite-induced up-regulation of mesenchymal marker N-cadherin, in contrast, was abrogated following inhibition of HSF1 and NRF2 (Fig. 7D).
Autophagy was earlier shown to help cancer cell survival and migration through overcoming intracellular and extracellular stress (51). To examine if autophagy has any role in this case, immunoblotting was carried out to estimate the level of LC3II, a marker of autophagy activation (52). The level of ATG7, a HSF1 target (53) and an E1 ubiquitin ligase involved in the autophagosome formation (54), was also monitored. Inhibition of HSF1 and NRF2 alone or in combination appeared to affect the levels of LC3II (Fig. 7D). Next, role of ATG7 and HSF1 alone and in combination was analyzed. The results suggested that down-regulation of these factors which inhibit autophagy also impairs EMT (Fig. 7E). Involvement of NRF2 and HSF1 in cell migration in response to arsenite treatment was monitored by wound healing assay. Significant block in migration/wound healing of cells was observed when both the factors were inhibited (Fig. 7F). Induction of HSF1 by peroxide treatment was found to be similarly dependent on NRF2, as RA treatment reduced HSF1 mRNA transcription similar to that in the HO-1 gene (Fig. 7G, S1B). The role of HSF1 and NRF2 on cell viability was also measured in response to hydrogen peroxide treatment yielding results similar to that of arsenite (Fig. 7H).

Discussion
Results presented here have unraveled a molecular basis of transcriptional up-regulation of cellular HSF1 gene by NRF2 upon oxidative stress induced by sodium arsenite or hydrogen peroxide ( Figs. 1 and 7, G and H). Two AREs located between 1707 and 1517 bp upstream of the HSF1 promoter were identified to be responsible for mediating the NRF2 function (Figs. 2  and 3). We also showed that this event plays an important role in survival, proliferation, and migration of breast cancer cells (Fig. 7). Constitutive activation of NRF2 has been implicated by earlier studies in cancer cell survival and proliferation (32,55).
An important finding of this study is the identification of two unique AREs by comparison with an up-to-date consensus ARE-sequences. Although the sequence analysis software identified ARE-A, ARE-B, and ARE-C, functional assays validated only ARE-B and ARE-C as the functional AREs. These two AREs differed from that of the NQO1 or HO-1 gene (Fig. 2C). For example, the ARE-B element matched the upstream TGA present in both NQO1 and HO-1 AREs but not downstream.

Control of HSF1 promoter by NRF2
The ARE-C element matched the downstream AGC present in both NQO1 and HO-1 AREs but it carries TTA instead of TGA upstream. Assembly of the complex with nuclear extracts on the ARE-B and ARE-C, with mobility similar to that of NQO1, suggests the formation of a complex involving similar players (Fig. 4C). The K D value of ARE-C, however, was measured to be two and three times of that of ARE-B and NQO1, respectively (Fig. 4E). The K D of NRF2-MafG heterodimer binding to ARE-B with a higher affinity than that of ARE-C suggests that the requirement of GC at 9, 10 position of an ARE may not be essential in all ARE sequence context (36). Although it remains to be experimentally verified, the presence of T at position 3 of ARE-C (which is less conserved in the consensus ARE) may be responsible for its lower affinity to bind the NRF2-MafG heterodimer (Fig. 4E). Mismatch in positions 4 and 10 may have rendered putative ARE-A inactive (although r-VISTA scored it higher than ARE-C. (Fig. 2, B, left panel, and C).

Control of HSF1 promoter by NRF2
The sensitivity of both NRF2-and HSF1-guided pathways to a proteotoxic stressor is known; HSF1 and NRF2 were shown to activate a common set of genes with overlapping functions in heat shock and antioxidant response pathways (56). NRF2 and HSF1 were reported to compensate for each other's activity (20). NRF2 increase HSP70 expression during methionine deprivation in a process independent of HSF1 function, suggesting thatinhibitionofHSF1activitycanbeinpartbypassedbyNRF2-dependent antioxidant response (57). NRF2 knockout mice showed a higher level of unfolded protein response, thus a higher level of HSF1 activation (58). This compensatory function may be responsible for failure to detect any significant change of HSP70 transcription induction in cells with downregulated NRF2 (Fig. 3A). HSP70 was also regulated by factors other than HSF1 (29). Although this study convincingly showed that cellular HSF1 is a target of NRF2, especially under oxidative stress, it remains to be determined why earlier ChIP sequencing studies did not detect NRF2 presence on the HSF1 promoter (35,36,38). Studies here suggest that HSF1 transcription was differentially regulated by oxidative stress in different cells. For example, it was induced 8-fold more in HT1080 compared with the NKE cells (Fig. 1B). Additional investigation is necessary to conclude that the extent of dependence on NRF2 is determined in a cell by its endogenous ROS level. This study shows that HSF1 gene is up-regulated in cells by oxidative stress in proportion to their endogenous ROS levels (Fig. 1B). Efficient neutralization of the effect of arsenite by N-acetyl cysteine suggested that a major mediator of arsenite action was through increasing the level of cellular ROS (Fig. 5C).
We also found that NRF2 requires BRG1 to regulate HSF1 promoter activity (Fig. 6). Down-regulation of FACT and helicase-like transcription factor (HLTF) did not affect arseniteinduced HSF1 transcription. An association of P-TEFb, Mediator, and DNA-bending protein is anticipated in the process. MED16 was shown to be required in NRF2 target gene expression (59). Future studies will show how NRF2, recruited at the identified ARE-B and -C, communicates with RNA polymerase II at the core promoter (60).
This study highlighted an important role of NRF2 and HSF1 collaboration in cancer cells survival and progression. Our results suggested that both NRF2 and HSF1 alone or in combination play a positive role in arsenite-induced EMT and migration in MCF7 cells (Fig. 7, D-H) (61). These two factors may support these processes by enhancing autophagy (Fig. 7, D and  E), an important process many cancer cells utilize for their benefit (20,51,53,62).
Many cancer cells have an elevated level of oxidative stress (63). An activated ROS signaling supports prosurvival pathways through facilitating several cellular processes such as inactivating tumor suppressor genes, enhancing glucose consumption, the rate of oncogenic mutation, and anaerobic respiration (64). NRF2 plays a crucial role in maintaining higher ROS in cells. Oncogenes can direct increased NRF2 expression (65). This gain of function in NRF2-controlled interactome has been reported in lung, breast, skin, esophageal, gallbladder, renal, pancreatic, and endometrial cancer (66,67). NRF2 exerts checks and balances on the cellular ROS level by up-regulating expression of its target genes such as GSH peroxidase and per-oxiredoxin (20,64). At the same time, cancer cells exhibit an elevated level of HSF1 activity, i.e. up-regulated levels of HSPs to support their prosurvival property (15,16). The role of various posttranslational modifications modulating HSF1 activity has been under extensive investigation (5). Altogether, the results presented here shed light on the molecular basis of transcriptional up-regulation of the HSF1 gene for adaptation to oxidative stress and its correlation with aggressiveness of cancer.

Reagents and antibodies
All the reagents were of analytical grade and were purchased from SRL India, Merck, or Sigma-Aldrich. All primers and oligonucleotides used were procured from Sigma-Aldrich. Restriction enzymes were bought from New England Biolabs.

Heat shock and arsenite treatments
The cells were plated the day before for treatment. Heat shock was conducted by submerging the plate of cells (grown to ϳ70% confluency) in water bath pre-warmed at 42°C for 1 h. The temperature in the medium reached 42°C within 5 min. The growth medium was not replaced either before or after heat shock. Treatment with sodium arsenite (20 M) was carried out for 2 h unless otherwise indicated.

Preparation of whole cell extracts and Western blot analysis
The cells were harvested from culture plates with a cell scraper on ice in ice cold PBS and rinsed twice with PBS. Whole cell extracts were prepared essentially using the procedure described earlier (29). Cell extracts containing 20 -50 g total protein or as appropriate were electrophoresed in 10%

Control of HSF1 promoter by NRF2
(acrylamide:bisacrylamide::29:1) denaturing PAGE and the resolved proteins were transferred to PVDF membrane (EMD Millipore). The membrane was blocked with 3% BSA or 5% fat-free milk in PBST (137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO4, 1.8 mM KH 2 PO4, pH 7.4, supplemented with 0.1% Tween 20) for 30 min at room temperature and incubated with a desired primary antibody at an optimized dilution for overnight at 4°C. The next day, the membrane was washed three to four times with PBST followed by incubation with secondary antibody for 1 h at room temperature. The membrane was then washed as before and developed by chemiluminescent detection (Clarity Bio-Rad) according to manufacturer's instructions. For quantification of the immunoblots, band intensities were measured using ImageJ software considering ␤-actin as normalization control and values were plotted using GraphPad Prism 6.0.

RNA isolation, semiquantitative RT-PCR, and qRT-PCR
Total RNA was isolated from cultured cells using TRIzol reagent as per manufacturer's protocol (Life Technologies). Isolated RNA was dissolved in ultrapure water and quantified using BioSpectrometer (Eppendorf). For cDNA preparation, 1 g total RNA was reverse transcribed using iScript cDNA Synthesis Kit (Bio-Rad) in 10 l reaction volume. For semiquantitative RT-PCR, a 10-fold (or as appropriate) diluted cDNA was probed with gene-specific primers using Taq polymerase. Expression of GAPDH mRNA in the same sample under analysis was considered as endogenous control. qRT-PCR was performed using SYBR Green (SYBR Green Hot JumpStart) according to the manufacturer's protocol (Sigma-Aldrich). Results were expressed as the ratio of treated samples to untreated samples normalized to their GAPDH levels. Measured numbers of threshold were converted to relative -fold change using 2 Ϫ⌬⌬CT method (68). The primer sets used for estimation of transcript levels for different genes are described in the supporting information (Table S1).

Plasmid constructs
The human HSP70 promoter (pHSP70 (Ϫ450/ϩ91)) and 5Ј-deletion fragments of the human HSF1 promoter (Fig. 2) were generated by PCR from human genomic DNA using primers containing restriction sites. An XhoI and HindIII recognition site was added to the PCR product through 5Ј-ends of forward and reverse primers, respectively. Amplified fragments were subsequently cloned into the pGL4.79 [hRluc/neo] reporter vector (Promega) pre-digested with the same restriction enzymes. Cloning of the ARE site from NQO1 gene promoter (5Ј-TCCGCAGTCACAGTGACTCAGCAGAATCTG-3Ј) and HSF1 promoter versions was done in pGL4.79-mCMV between Acc65I and BglII sites. The pGL4.79-mCMV vector is a modified pGL4.79[hRluc/neo] vector with a minimal CMV (mCMV) enhancer element (5Ј-TAGAGGGTATATAATG-GAAGCTCGACTTCCAG-3Ј) cloned between BglII and Hin-dIII sites. Nucleotide sequences of PCR primers used are listed in supporting information (Table S2).

Luciferase reporter assay
HT1080 cells grown in 12-well plates to ϳ70% confluence were transfected with different reporter constructs carrying the HSF1 promoter variants using Lipofectamine 2000 (29). Next day, cells were distributed into two wells and treated with or without 20 M arsenite for 2 h, followed by isolating the RNAs by TRIzol reagent. Expression levels of different genes were measured by semiquantitative RT-PCR and qRT-PCR from the cDNA prepared from the total RNA pretreated with DNase I. Relative -fold increase in Renilla-luciferase (Rluc) mRNA levels in the presence of arsenite was calculated with respect to the untreated control as well as carrier internal vector control. The levels of Rluc mRNA in samples transfected with promoterless pGL4.79 [hRluc/neo] vector was considered as 1. Unless otherwise stated, reporter gene assays were carried out in three parallel experiments or repeated at least three times.

Control of HSF1 promoter by NRF2
Healthcare). DNase I footprinting was performed as described (70).

ChIP assay
HT1080 cells transfected with a FLAG-tagged NRF2 overexpression plasmid (NC16 pCDNA3.1 FLAG, Addgene plasmid no. 36971) were treated with 20 M sodium arsenite or mock for 2 h. Treatment with N-acetyl cysteine (5 mM) was carried out for 3 h prior to treatment with arsenite. Preparation of sheared chromatin, incubation with anti-FLAG antibody (Sigma-Aldrich, F7425) for collecting the associated DNA, and PCR were carried out (31). The DNA was quantified by quantitative PCR using the primers listed in supporting information (Table S3). ChIP against RNA Polymerase II was performed with anti-RNA Polymerase II antibody (Abcam, ab817).

Cell proliferation analysis
Cell proliferation was measured by viability determination using 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay or by direct cell counting. Cells were seeded at 0.5 ϫ 10 5 cells per well in 12-well plates and incubated under standard conditions. After the experiment, cells were incubated with 0.5 mg/ml MTT solution (Himedia, RM1131) at 37°C, 5% CO 2 for 4 h in the dark. After 4 h incubation the MTT solution was removed and DMSO was added to the plates. A 570 value was measured with BioSpectrometer (Eppendorf). For counting, cells were trypsinized after washing with PBS. After making thorough suspension, samples were counted using hemocytometer.

Wound healing assay
Cells (2 ϫ 10 5 ) were grown in complete medium in 6-well plates until cells were 80 -90% confluent. The undersurface of each well was marked with three parallel straight lines. A scratch was made across the surface of the plate using sterilized disposable 200 l pipette tip to remove the complete layer of cells within the scratch area at 90°to the orientation lines. Each well was then washed twice with PBS, replaced with growth medium supplemented with 5% serum, and treated as mentioned. The width of the scratch was imaged using a microscope with camera (Leica) immediately (0 h) or after 24 or 48 h of treatment at the same magnification settings. Scratch-gap area was measured using ImageJ software and values were plotted using GraphPad Prism 6.0.

ROS quantification by DCFDA assay
Cells pre-treated as appropriate were incubated with 5 M DCFDA for 30 min. After removing the DCFDA by washing twice with ice-cold PBS, the labeled cells were trypsinized, rinsed, and finally resuspended in PBS. Oxidation of DCFDA to the highly fluorescent 2Ј,7Ј-dichlorofluorescein proportional to ROS generation was analyzed by measuring 10,000 events/sample using FACS (BD FACSVerse, BD Biosciences).

Statistical analysis
All experiments were performed at least three times. Plotting of graphs and statistical analysis was performed by one-way analysis of variance (ANOVA) with appropriate post hoc tests, using GraphPad Prism 6.0 software. Scatter plots represent individual data points combined with mean Ϯ S.D. Values were considered to be statistically significant when p Ͻ 0.05.