The ABL2 Kinase Regulates an HSF1-Dependent Transcriptional Program Required for Lung Adenocarcinoma Brain Metastasis

The ABL2 kinase regulates an HSF1-dependent transcriptional program required for lung adenocarcinoma brain metastasis

Jacob P. Hoja,1, Benjamin Mayroa,1, and Ann Marie Pendergasta,2

aDepartment of Pharmacology and Cancer Biology, Duke University School of Medicine, Durham, NC 27710

Edited by Joan S. Brugge, Harvard Medical School, Boston, MA, and approved November 6, 2020 (received for review April 24, 2020) Brain metastases are the most common intracranial tumors in independent of the canonical heat shock response (18–25). adults and are associated with increased patient morbidity and Previous work identified a divergent cancer-specific transcrip- mortality. Limited therapeutic options are currently available for tional program driven by HSF1 in highly malignant mammary the treatment of brain metastasis. Here, we report on the discov- epithelial cells and tumor cell lines from the lung, breast, and ery of an actionable signaling pathway utilized by metastatic colon (18). Central to this work was the discovery that genome tumor cells whereby the transcriptional regulator Heat Shock occupancy of HSF1 in the context of tumor malignancy is distinct Factor 1 (HSF1) drives a transcriptional program, divergent from from that of heat shock and drives expression of target genes its canonical role as the master regulator of the heat shock implicated in cell cycle, translation, and DNA repair. More re- response, leading to enhanced expression of a subset of E2F cently, HSF1 expression was also identified in T cell acute lym- transcription factor family gene targets. We find that HSF1 is phoblastic leukemia as critical for tumor cell survival by required for survival and outgrowth by metastatic lung cancer regulating NOTCH1 transcriptional activity (19). Although these cells in the brain parenchyma. Further, we identify the ABL2 studies provide evidence in support of heat shock-independent tyrosine kinase as an upstream regulator of HSF1 protein expres- functions of HSF1, little is known regarding the upstream sion and show that the Src-homology 3 (SH3) domain of ABL2 modulators of HSF1 protein expression as well as the role of directly interacts with HSF1 protein at a noncanonical, proline- HSF1-driven transcriptional programs in the context of tumor independent SH3 interaction motif. Pharmacologic inhibition of metastasis. the ABL2 kinase using small molecule allosteric inhibitors, but In Caenorhabditis elegans, HSF1 was recently shown to func- not ATP-competitive inhibitors, disrupts this interaction. Impor- tion with E2F transcription factors to drive a transcriptional tantly, knockdown as well as pharmacologic inhibition of ABL2 program required during larval stages of development, inde- using allosteric inhibitors impairs expression of HSF1 protein and pendent of its role in the cellular response to heat shock and HSF1-E2F transcriptional gene targets. Collectively, these findings proteotoxic stress (26). In this context, HSF1 binds DNA at reveal a targetable ABL2-HSF1-E2F signaling pathway required for “degenerate” heat shock elements adjacent to GC-rich E2F survival by brain-metastatic tumor cells. binding domains. We now show that HSF1 protein levels are

brain metastasis | ABL2 | HSF1 | lung adenocarcinoma | ABL kinases Significance

rain metastases stemming from primary tumors of the lung Among all cancer types, lung cancer patients exhibit the Bare a major health challenge for patients and often result in highest prevalence of brain metastasis, often associated with devastating neurologic impairments and increased mortality cognitive impairment, seizures, decline in quality of life, and – (1 3). Despite numerous studies focused on dissecting genetic decreased survival. Limited therapeutic options are currently and nongenetic molecular mechanisms employed by metastatic available to treat brain metastasis. A comprehensive under- tumor cells to survive and colonize the unique microenvironment standing of the signaling pathways and transcriptional net- of the brain (4–8), there is a lack of effective therapies to treat works required for survival and growth of brain-metastatic this disease (9). Therefore, there is an urgent need to gain ad- cancer cells is needed to develop effective strategies to treat ditional insights into the mechanisms employed by metastasizing this disease. Here, we report that the Heat Shock Transcription tumor cells for colonization and survival in the brain, which Factor 1 (HSF1) is upregulated in brain-metastatic lung cancer might lead to new rational approaches for pharmacologic cells and is required for brain metastasis in mice. Importantly, intervention. we show that the HSF1-dependent expression of E2F target The transcriptional activator Heat Shock Transcription Factor genes implicated in cell cycle progression and survival is de- 1 (HSF1) is a master regulator of protein homeostasis (10, 11). creased by blood–brain barrier-penetrant ABL allosteric Under normal physiologic conditions, HSF1 monomers are held inhibitors. inactive in the cytoplasm through intramolecular interactions and a regulatory complex consisting of protein chaperones and Author contributions: J.P.H., B.M., and A.M.P. designed research; J.P.H. and B.M. per- – formed research; J.P.H. and B.M. analyzed data; and J.P.H., B.M., and A.M.P. wrote the chaperonin TCP1 ring complex (TRiC) (12 14). In response the paper. to a diverse array of proteotoxic stresses including heat shock, Competing interest statement: A.M.P. is a consultant and advisory board member for The monomeric HSF1 sheds its inhibitory regulatory complex, Pew Charitable Trusts. A.M.P. holds patent no. U.S. 9,931,342 B2, related to this work. translocates to the nucleus, and oligomerizes into transcription- This article is a PNAS Direct Submission. – ally active HSF1 trimers (10, 15 17). Active HSF1 then induces This open access article is distributed under Creative Commons Attribution-NonCommercial- transcription of genes encoding molecular chaperones by binding NoDerivatives License 4.0 (CC BY-NC-ND). to DNA motifs known as heat shock elements (HSEs) (16). 1J.P.H. and B.M. contributed equally to this work. Although the role of HSF1 in the context of protein homeostasis 2To whom correspondence should be addressed. Email: [email protected]. is well established, an emerging body of evidence in recent years This article contains supporting information online at https://www.pnas.org/lookup/suppl/ has shown that HSF1 also functions to drive transcriptional doi:10.1073/pnas.2007991117/-/DCSupplemental. programs implicated in tumor progression and malignancy First published December 14, 2020.

33486–33495 | PNAS | December 29, 2020 | vol. 117 | no. 52 www.pnas.org/cgi/doi/10.1073/pnas.2007991117 Downloaded by guest on September 30, 2021 up-regulated in brain-metastatic lung cancer cells and that HSF1 metastatic colonization and outgrowth of lung cancer cells in an regulates expression of E2F gene targets in this setting. Genetic in vivo model of brain metastasis. To examine this hypothesis, we inhibition of HSF1 ablates E2F target gene expression and performed intracardiac injections of PC9-BrM3 cells transduced dramatically impairs survival of lung cancer brain metastases both with dox-inducible tet-shNTC or tet-shHSF1 shRNAs and in vitro and in mouse models. coexpressing a luciferase-Tomato (pFuLT) lentiviral reporter. Because therapeutic targeting of transcription factors such as Mice were injected with brain metastatic PC9-BrM3 cells in the HSF1 and E2F is challenging, we sought to identify actionable left cardiac ventricle and, after assessing colonization of the upstream regulators of this HSF1-E2F transcription network. brain at day 10 postinjection with bioluminescent imaging (BLI), We recently characterized a TAZ-AXL-ABL2 feed-forward mice were administered 3 mg/mL dox water to induce shRNA signaling axis that is activated in brain-metastatic lung cancer expression. The 10-d time point was selected as previous studies cell lines and which is required for successful colonization of have shown that metastasizing tumor cells in circulation com- these cells in the brain (27). We now show that ABL2 modulates plete extravasation into the brain parenchyma within 7 d post- expression of the HSF1 and E2F transcription factors and that injection (7, 28). We found that the growth rate of metastases in knockdown or allosteric inhibition of ABL2 impairs expression mice injected with PC9-BrM3 cells expressing shRNA against of HSF1-dependent and E2F-dependent transcription programs. HSF1 was markedly impaired compared to mice injected with We report that HSF1 protein is up-regulated in brain-metastatic nontarget shRNA control cells (Fig. 1F). Metastatic disease cancer cells downstream of the ABL2 tyrosine kinase and is re- burden within the brain was significantly decreased in the HSF1 quired for their survival both in vitro and in vivo. Importantly, we knockdown group as measured 1 mo after induction with dox show that HSF1-E2F target gene expression is pharmacologically water (Fig. 1G). The rate of whole-body metastatic burden was targetable with ABL kinase allosteric inhibitors (27). These data similarly delayed in the HSF1 knockdown cells (Fig. 1 H and I). support a critical role for an actionable ABL2-HSF1-E2F path- Together, these results show that HSF1 is necessary for the way in promoting brain metastasis. colonization and outgrowth of metastatic tumor cells in an in vivo model of lung cancer brain metastasis. Notably, analysis Results of human lung adenocarcinoma patient microarray datasets Expression of HSF1 Is Up-Regulated in Brain-Metastatic Lung correlating messenger RNA (mRNA) expression with patient Adenocarcinoma Cells. HSF1 expression was previously shown to survival showed that high expression of HSF1 correlates with be up-regulated during progression of normal epithelial cells poor overall survival (Fig. 1J). toward a malignant, tumorigenic state (18). Thus, we questioned whether HSF1 might be differentially expressed in brain- HSF1 Regulates Transcriptional Expression of E2F Gene Targets in MEDICAL SCIENCES metastatic cells compared to the parental tumor cell lines from Brain-Metastatic Lung Cancer Cells in a Heat Shock-Independent which they were derived. Immunoblot analysis of HSF1 protein Manner. To investigate the mechanism by which HSF1 pro- expression in EGFR mutant (PC9 and HCC4006) and KRAS motes survival of metastatic lung cancer cells, we employed un- mutant (H2030) human lung adenocarcinoma cells compared to biased transcriptional profiling of these cells with or without their respective brain-metastatic variants PC9-BrM3, HCC4006- HSF1 knockdown in order to gain insights into the transcrip- BrM, and H2030-BrM3, revealed elevated levels of HSF1 pro- tional target gene signatures altered by loss of HSF1. We per- tein in the brain-metastatic derivatives (Fig. 1A). Further, im- formed RNA-sequencing (RNA-seq) on PC9-BrM3 cells munofluorescence staining of PC9-BrM3 cells revealed high transduced with dox-inducible nontarget control shRNA or levels of HSF1 in the nucleus, indicative of transcriptionally shHSF1 constructs and treated with dox for 48 h, a time point at active HSF1 (Fig. 1B). which HSF1 protein is depleted but prior to loss in cell viability HSF1 was previously shown to be essential for T cell acute observed at longer time points (Fig. 1 C and D). Gene set en- lymphoblastic leukemia cell survival (19). Therefore, we sought richment analysis (GSEA) was used on an expanded list of gene to investigate whether the elevated expression of HSF1 in brain- signatures from the mSigDB database from The Broad Institute metastatic cells was required for their survival and growth. To to identify pathways affected by HSF1 loss (29). The signatures examine this possibility, we transduced cells with doxycycline most depleted upon HSF1 knockdown corresponded to E2F (dox)-inducible lentiviral short hairpin RNAs (shRNAs) against family gene targets and G2/M cell cycle checkpoint targets, either nontargeting control (shNTC) or HSF1 and measured cell consistent with the observed increase in G2/M cell cycle arrest viability across multiple time points after induction of shRNA (SI Appendix, Figs. S1A and S2 A–C and Fig. 2 A–C) and loss of expression. Immunofluorescence and immunoblot analysis Cyclin B1 expression upon HSF1 knockdown (Fig. 1C). Inter- demonstrated the effectiveness of HSF1 knockdown (Fig. 1 B estingly, signatures related to the unfolded protein response, and C). Knockdown of HSF1 with multiple shRNAs resulted in a while slightly depleted under knockdown conditions, scored robust decrease in cell viability starting 48 h post-shRNA in- relatively poorly despite the known role for HSF1 in this process duction, which correlated with the loss of HSF1 protein (Fig. 2A). Similarly, expression of gene targets representative of (Fig. 1 C–E and SI Appendix, Fig. S1 A–E). This loss corre- a canonical heat shock response remained relatively stable de- sponded with an increase in G2/M cell cycle arrest as revealed by spite HSF1 depletion in these cells (Fig. 2D). A list of conserved FACS analysis (SI Appendix, Fig. S1A) and decreases in Cyclin E2F target genes from a number of E2F gene signatures iden- B1 protein expression (Fig. 1C). Knockdown of HSF1 also tified by GSEA, which we confirmed by RT-qPCR to be regu- resulted in increased levels of cleaved PARP, indicative of ap- lated by E2F family members, was used to generate an HSF1- optotic induction (Fig. 1C). Parental lung cancer cells exhibited dependent E2F gene panel (Fig. 2D, upper bracket and SI Ap- decreased survival upon HSF1 knockdown, but to a lesser extent pendix, Fig. S2 D and E). than the brain-metastatic variants (Fig. 1 D and E and SI Ap- We next reasoned that if HSF1 promoted transcription of a pendix, Fig. S1 B–E). These data show that brain-metastatic lung subset of E2F target genes, then expression of these target genes cancer cells are highly dependent on HSF1 for growth might also be increased in brain-metastatic cells expressing high and survival. HSF1 levels relative to parental cells. RNA-seq analysis and GSEA comparing PC9-BrM3 cells with PC9 parental cells HSF1 Knockdown Impairs Metastatic Outgrowth and Tumor Cell revealed enrichment for E2F target gene expression as well as Survival In Vivo. Given that loss of HSF1 robustly impairs sur- G2/M checkpoint targets in the PC9-BrM3 cells (Fig. 3A and SI vival of brain-metastatic lung cancer cells in vitro, we next Appendix, Fig. S3 A–C). RT-qPCR analysis was used to validate evaluated whether HSF1 expression might be required during increased expression of the HSF1-E2F gene panel in PC9-BrM3

Hoj et al. PNAS | December 29, 2020 | vol. 117 | no. 52 | 33487 Downloaded by guest on September 30, 2021 A PC9 PC9-BrM3HCC40064006-BrMH2030 2030-BrM3 B Dapi HSF1 Merge HSF1 1.00 9.11 1.00 96.30 1.00 2.75

Actin LUAC C Tet-shHSF1 7480 Tet-shHSF1 7481 shNTC Time + Dox (h) 0 24 48 72 0 24 48 72 HSF1

1.00 0.360.08 0.0011.00 0.73 0.20 0.05 Cl. PARP

1.00 1.11 0.73 3.36 1.00 0.60 0.46 3.04 Cyclin B1

1.00 0.95 0.29 0.04 1.00 0.85 0.24 0.32 shHSF1 7480 Actin

D PC9 Parental vs BrM3 1.5 PC9 shNTC F PC9-BrM3 G PC9 shHSF1 7480 PC9-BrM3 shNTC PC9-BrM3 1.0 PC9BrM3 shHSF1 7480 24 * **** 0.5 **** 22 * 107 Relative Cell Viability 0.0 Tet-shNTC 0123456 20

6 PC9 Parental vs BrM3 10 E Brain Flux 18 1.5 PC9 shNTC [p/s] Flux Total 5 PC9 shHSF1 7481 10 Log2(Flux (photons/sec)) 16 1.0 PC9-BrM3 shNTC ns PC9-BrM3 shHSF1 7481 Tet Tet 0.5 **** ** shNTC shHSF1 Tet-shHSF1 7480 Tet-shHSF1

Relative Cell Viability 0.0 0123456 Time (Days + Dox)

H PC9-BrM3 I PC9-BrM3 J 1.8 1.8 Tet-shNTC Tet-shHSF1 7480 9 100 HSF1 n shNTC High 1.6 1.6 shHSF1 7480 Mid 8 Low 1.4 1.4 7 50 1.2 1.2 ** % OS 6 1.0 1.0 Log(photons/sec) Log2(Flux (photons/sec)) Whole Body Met. Burden p-value < 0.0001 Overall metastatic burde 0 0.8 0.8 5 0102030405040 102030 0 0 1020304050 0 50 100 150 200 250 Days +Dox Days +Dox Days +Dox Time (months)

Fig. 1. HSF1 protein is up-regulated in brain-metastatic lung adenocarcinoma cells and is required for cell survival in vitro and colonization of the brain in vivo. (A) Immunoblot analysis of HSF1 protein expression in parental vs. brain-metastatic lung adenocarcinoma (LUAC) cell lines. Protein quantification data for brain-metastatic lines were normalized to actin loading control and corresponding parental cell line expression. (B) Immunofluorescence staining and imaging of HSF1 protein (red) in PC9-BrM3 cells transduced with dox-inducible lentiviral shRNAs against nontarget control (shNTC) or HSF1 (shHSF1 7480). Dapi, nuclear stain (blue). (C) Immunoblot analysis of HSF1 protein expression in PC9-BrM3 cells transduced with distinct dox-inducible shRNA clones targeting HSF1 (7480 or 7481) and treated ± 500 ng/mL dox for the indicated timepoints (H). Protein quantification data were normalized to actin loading control and to corresponding 0 h (untreated) timepoint. (D and E) Cell-Titer Glo assay measuring cell viability in PC9 parental vs. PC9-BrM3 cells transduced with shRNAs against nontarget control or HSF1 clone 7480 (D) or clone 7481 (E) and treated with dox for the indicated timepoints. For all experiments, n = 3 biological replicates per condition. Statistical analysis performed by two-way ANOVA followed by Fisher’s multiple comparison post hoc testing. *P < 0.05; **P < 0.01; ****P < 0.001; ns, not significant. Representative BLI (F) and quantification of brain metastasis burden (G) in mice injected with PC9-BrM3 cells transduced with dox-inducible shRNAs against nontarget control (NTC) or HSF1 (clone 7480). Mice were treated with 3 mg/mL dox water for shRNA induction starting on day 10 postintracardiac injection. After 28 d on dox water, mice were subjected to BLI analysis. Statistical analysis performed by unpaired two-tailed t test. *P < 0.05. (H and I) Individual spider plots (H) and quantification (I) of whole-body metastatic burden in mice as described in Fig. 1 F and G. Statistical analysis was performed using a mixed-effects model in GraphPad Prism 8 software to account for missing values due to premature animal death. **P < 0.01. (J) Kaplan–Meier survival analysis correlating overall survival with mRNA expression of HSF1 in human lung adenocarcinoma patients. Survival groups were separated by tertile based on mRNA expression (n = 720 total patients), and statistical analysis was performed using log-rank Mantel–Cox test.

33488 | www.pnas.org/cgi/doi/10.1073/pnas.2007991117 Hoj et al. Downloaded by guest on September 30, 2021 ABHALLMARK GSEA D PC9-BrM3 shNTC vs shHSF1 HSF1 Knockdown E2F Targets G2M Checkpoint E2F8 Spermatogenesis MCM7 MYC Targets V1 MYC Targets V2 DNMT1 3 Protein Secretion l RANBP1 Unfolded Protein Response Mitotic Spindle DCTPP1 Interferon Alpha Response NES = -2.40 MCM2 DNA Repair p-value = <0.001 Estrogen Response Late PRKDC Glycolysis MCM4 NOTCH Signaling MCM6 Angiogenesis arget Pane MTORC1 Signaling T ATAD2 Tet-shNTC Tet-shHSF1 2 IL2 STAT5 Signaling SMC1A HEME Metabolism SMC3 Fatty Acid Metabolism E2F IL6 JAK STAT3 Signaling NASP Xenobiotic Metabolism UV Response UP MCM3 Adipogenesis CDC6 KRAS Signaling DN Peroxisome B4GALNT3 Oxidative Phosphorylation ESR1 Bile Acid Metabolism 1 Hedgehog Signaling C INSR Interferon Gamma Response DUSP1 Pancreas Beta Cells WNT Beta Catenin Signaling AP4B1 Reactive Oxygen Species Pathway DCLRE1B Apical Surface [Log2(Fold Change)] PI3K AKT MTOR Signaling TRAK1 k Myogenesis LTBP1 Expression Relative mRNA Androgen Response Estrogen Response Early IL7R 0 Complement NES = -2.07 SOX7 Coagulation p-value = <0.001 FMNL2 TGF Beta Signaling KRAS Signaling Up PTPRE Apical Junction ZBTB7A Apoptosis Heat Shoc

P53 Pathway (Mendillo et al 2012) VPS37D Allograft Rejection Tet-shNTC Tet-shHSF1 ZC3HAV1 Epithelial Mesenchymal Transition Inflammatory Response AFAP1 Hypoxia FGD4 -1

TNFA Signaling via NFKB MEDICAL SCIENCES UV Response DN SRGAP3 Cholesterol Homeostasis SORBS1 2 0 FBXL14 -2 NES

Fig. 2. Depletion of HSF1 in brain-metastatic lung cancer cells impairs expression of E2F target gene signatures independent of the canonical heat shock response. (A) Waterfall plot of Hallmark GSEA signatures from RNA-seq data ranked by normalized enrichment score (NES) for PC9-BrM3 cells with inducible expression of shNTC or shHSF1. Negative NES, depleted signature; positive NES, enriched signature. Dotted line represents P value cutoff <0.05. (B and C)

GSEA plots showing E2F targets (B)orG2/M checkpoint signatures (C) depleted in PC9-BrM3 Tet-shHSF1 knockdown cells. (D) Heatmap depicting relative expression of the E2F target gene panel and representative heat shock gene panel.

and HCC4006-BrM cells relative to parental PC9 and HCC4006 consisting of TTTnnCGC (where “n” is either C or G) (SI Ap- cells, respectively (Fig. 3 B and C). Notably, RT-qPCR analysis pendix, Fig. S4A) (30). Given that loss of HSF1 impairs mRNA also revealed that E2F family members themselves were differ- expression of E2F family target genes in brain metastatic cells, entially expressed in brain-metastatic cells with increased ex- we hypothesized direct HSF1 binding may occur at degenerate pression of E2F1, E2F2, E2F7, and E2F8, whereas expression of HSEs located near promoter regions of known E2F transcrip- other E2F family members remained unchanged (SI Appendix, tional targets. To investigate this possibility, we analyzed the Fig. S3D). Further, immunoblot analysis of brain-metastatic cells promoter regions of various E2F targets from the E2F target revealed enrichment of E2F1 and E2F8 relative to the varied gene panel and found TTTnnCGC motifs upstream of their re- expression of other E2F family members, implying a potential spective transcriptional start sites (SI Appendix, Fig. S4 B–D). role for E2F1 and E2F8 in regulating expression of E2F target We then searched for partial permutations of the canonical HSE genes in this system (Fig. 3D and SI Appendix, Fig. S2 D and E). sequence containing varying segments of this DNA motif (SI Of note, analysis of human lung adenocarcinoma patient Appendix, Fig. S4A). We detected a significant number of pre- microarray data revealed that high expression of E2F1 and E2F8 dicted degenerate HSEs, which were primarily found immedi- was each individually correlated with poor overall survival (SI ately downstream of the E2F binding motif and upstream of the Appendix, Fig. S3 E and F). Taken together, these data show that transcriptional start site, consistent with the report of E2F-HSF1 loss of HSF1 in brain-metastatic lung cancer cells results in de- coregulated DNA binding in C. elegans (SI Appendix, Fig. creased expression of select E2F target genes with minimal im- – pact on expression of canonical heat shock gene targets. S4 B D). In contrast, analysis of the promoter regions of HSF1 was recently shown to function with E2F transcription HSP90AB1 and HSPA6, both known target genes of HSF1 dur- factors in C. elegans to drive a developmental transcriptional ing the canonical heat shock response, showed the presence of program divergent from the canonical heat shock response (26). expected canonical HSEs upstream of the TSS (SI Appendix, Fig. Central to this work was the observation that in the context of S4 E and F), but no E2F DNA binding motifs were found near development, HSF1 occupies genomic loci containing a “de- the transcriptional start sites of HSP90AB1 and HSPA6. Im- generate” or partial HSE nearby E2F DNA binding motifs. The portantly, chromatin immunoprecipitation (ChIP)-qPCR analy- canonical DNA element recognized by trimeric HSF1 in re- sis in PC9-BrM3 and HCC4006-BrM cells revealed empirically sponse to heat shock consists of triple tandem inverted nGAAn that HSF1 occupancy is enriched at predicted degenerate HSEs repeats (e.g., nTTCnnGAAnnTTCn or derivations thereof) (11). upstream of E2F target genes (Fig. 3 E and F). Collectively, E2F family members bind DNA sequences at GC-rich motifs these findings reveal the presence of degenerate HSF1-binding

Hoj et al. PNAS | December 29, 2020 | vol. 117 | no. 52 | 33489 Downloaded by guest on September 30, 2021 A B E2F Target Gene Signature C E2F Target Gene Signature HALLMARK GSEA 2.0 E2F8 MCM6 PC9 Parental vs PC9-BrM3 MCM2 MCM4 Apical Surface ATAD2 MCM3 n MCM3 n DCTPP1 G2M Checkpoint 2.0 1.5 Hedgehog Signaling DCTPP1 PRKDC MCM4 essio KRAS Signaling DN MCM7 pressio E2F Targets MCM7 E2F8 Ex Interferon Alpha Response CDC6 MCM2 1.0 Mitotic Spindle RANBP1 DNMT1 Estrogen Response Early SMC3 1.0 SMC1A e mRNA e mRNA Expr e mRNA v Myogenesis NASP v ATAD2 NOTCH Signaling DNMT1 CDC6 0.5 Relati

Interferon Gamma Response Relati SMC3 Spermatogenesis MCM6 Fold Change (Normalized to 18S) Estrogen Response Late SMC1A Fold Change (Normalized to 18S) NASP PRKDC RANBP1 Androgen Response 0 0 Adipogenesis PC9 PC9 HCC4006 HCC4006 Heme Metabolism Parental -BrM3 Parental -BrM Reactive Oxygen Species Pathway RM ANOVA: p-value < 0.0001 RM ANOVA: p-value < 0.0001 Protein Secretion DNA Repair Pancreas Beta Cells **** PC9-BrM3 UV Response UP D E ChIP: HSF1 TGF Beta Signaling 0.06 PI3K AKT MTOR Signaling PC9 BrM3 HSF1 **** Apical Junction IgG Peroxisome E2F1 0.04 **** WNT Beta Catenin Signaling 1.00 6.61 ** Allograft Rejection IL6 JAK STAT3 Signaling E2F2 0.02 Percent Input (%) ns Bile Acid Metabolism 1.00 3.62 Fatty Acid Metabolism Epithelial Mesenchymal Transition 0.00 Apoptosis E2F3 Inflammatory Response AD2 E2F8 HBB SMC1A AT Complement 1.00 0.24 RANBP1 TNFA Signaling via NFKB Coagulation E2F4 HCC4006-BrM Hypoxia 1.00 0.81 ChIP: HSF1 Cholesterol Homeostasis F **** UV Response DN E2F6 0.15 HSF1 KRAS Signaling Up **** P53 Pathway 1.00 1.12 IgG **** Glycolysis 0.10 MYC Targets V2 E2F8 IL2 STAT5 Signaling ** 1.00 5.15 Angiogenesis 0.05 Xenobiotic Metabolism ns MYC Targets V1 HSP90 Percent Input (%) MTORC1 Signaling 1.00 1.01 0.00 Oxidative Phosphorylation Unfolded Protein Response Actin E2F8 HBB SMC1A ATAD2 RANBP1 -2 -1 0 1 2 NES

Fig. 3. Brain-metastatic lung cancer cells exhibit increased expression of E2F target gene signatures and exhibit HSF1 binding at E2F target genes. (A) Waterfall plot of Hallmark GSEA signatures from RNA-seq data ranked by NES for PC9 parental vs. PC9-BrM3 cells. Negative NES, depleted signature; positive NES, enriched signature. Dotted line represents P value cutoff <0.05. (B and C) Heatmaps of mRNA expression (RT-qPCR) for E2F target genes in PC9 parental vs. PC9-BrM3 cells (B) or HCC4006 parental vs. HCC4006-BrM cells (C). (D) Immunoblot analysis of the indicated proteins expressed in PC9 parental vs. PC9-BrM3 cells. Quantification of protein expression was normalized to actin loading control and parental expression. (E and F) ChIP-qPCR analysis of HSF1 occupancy at degenerate HSEs nearby the indicated E2F target genes in PC9-BrM3 (E) or HCC4006-BrM cells. HBB (Hemoglobin) gene included as negative control. n = 3 biological replicates. For all experiments, *P < 0.05; **P < 0.01; ***P < 0.005; ****P < 0.001; ns, not significant.

sequences, occupied by HSF1 and distinct from the canonical by ABL2 knockdown (Fig. 4A). Interestingly, reciprocal coimmu- HSE motif, near a subset of E2F transcriptional targets. noprecipitation of endogenous HSF1 and ABL2 in PC9-BrM3 cells showed a strong interaction between these two molecules (Fig. 4 B ABL2-Dependent Regulation of HSF1 Protein Expression in and C). In contrast to HSF1, which shuttles between the cytoplasm Brain-Metastatic Lung Cancer Cells. Our data reveal that HSF1 is and the nucleus, ABL2 is a cytoplasmic protein and lacks a nuclear required for survival of brain-metastatic cancer cells and suggest localization sequence. Coimmunoprecipitation and immunoblot that targeting of HSF1 could be exploited as a potential thera- analysis between nuclear and cytoplasmic fractions in cells peutic strategy to treat this disease. The structure of HSF1 is not cotransfected with ABL2-GFP and Flag-tagged HSF1 (Flag-HSF1) readily amenable to inhibition by small molecules; however, revealed that the interaction between these two molecules is re- HSF1 expression and activity might be impaired through inhi- stricted to the cytoplasm (SI Appendix,Fig.S5A). bition of upstream HSF1 regulators. As our previous work (27) To characterize the interaction between ABL2 and HSF1, we showed that loss of ABL2 in lung cancer cells impaired brain employed WT, kinase-inactive (K317M), SH2-inactive (R198K), metastasis outgrowth, similar to the phenotype induced by loss of or SH3-inactive (P158L) ABL2-GFP–tagged proteins (Fig. 4D). HSF1 in vivo, we evaluated whether ABL2 might also regulate Cotransfection of GFP-tagged ABL2 WT or ABL2 mutant HSF1 expression required for survival and colonization of brain- proteins with His-tagged HSF1 followed by coimmunoprecipi- metastatic cells. Notably, knockdown of ABL2 in PC9-BrM3 tation assays revealed that expression of the ABL2 P158L mu- cells in vitro resulted in a near complete loss of measurable tant resulted in a complete loss of the interaction with HSF1 HSF1, E2F1, and E2F8 proteins (Fig. 4A). Consistent with a heat (Fig. 4E). Further, GST in vitro pulldown assays of purified shock-independent role for HSF1 in brain metastasis, protein His-HSF1 and GST-ABL2 SH2 or SH3-SH2 domains revealed expression of HSP90 was not altered by the loss of HSF1 induced a direct interaction between HSF1 and ABL2 that is

33490 | www.pnas.org/cgi/doi/10.1073/pnas.2007991117 Hoj et al. Downloaded by guest on September 30, 2021 ABIgG IP EF shNTC shABL2 ABL2-eGFP - WT R198K P158L K317M ABL2 Blot: ABL2 Flag-HSF1 + ++ ++ IP Flag-HSF1 - WT del187-195del188-193 1.00 0.11 ABL2 ABL2-eGFP + +++ IP: Flag ABL1 IP: HSF1 (GFP) ABL2 1.00 0.64 (GFP) ABL2 WCL HSF1 HSF1 HSF1 (Flag) 1.00 0.01 HSF1 Actin (Flag) ABL2 E2F1 (GFP) 1.00 0.02 HSF1 ABL2

E2F8 C IgG IP (Flag) WCL (GFP) 1.00 0.04 pCRKL

Blot: HSF1 WCL IP (Y207) HSF1 HSP90 (Flag) 1.00 1.29 p-CRKL IP: ABL2 CRKL (Y207) Actin 1.00 0.21 ABL2 Actin CRKL HSF1 Actin Actin G His-HSF1 WT WT del187-195del188-193 GST-ABL2-SH3 - +++ GST + -- - WT del187-195del188-193 D ABL2 His 384 16 120 130 203 409 1% Input HSF1 GST

K188LIQFLIS195

Fig. 4. ABL2 kinase regulates HSF1 and E2F protein expression. (A) Immunoblots of the indicated proteins in PC9-BrM3 cells transduced with lentiviral shRNAs against nontarget control or ABL2. Quantification was normalized to actin loading control and shNTC, with the exception of phospho-CRKL, which was normalized to total CRKL. (B) Coimmunoprecipitation pulldown assay for endogenous HSF1 protein and immunoblots of indicated proteins in PC9-BrM3 cells. MEDICAL SCIENCES WCL, whole cell lysate. (C) Coimmunoprecipitation pulldown assay for endogenous ABL2 protein and immunoblots of indicated proteins in PC9-BrM3 cells. (D) Linear protein structures for ABL2 and HSF1. (E) Coimmunoprecipitation pulldown assay of HSF1 (Flag) in 293T cells transiently transfected with His-tagged HSF1 and GFP-tagged ABL2 WT, ABL2 R198K, ABL2 K317M, or empty vector control. (F) Coimmunoprecipitation pulldown assay of HSF1 (Flag) in 293T cells transiently transfected with Flag-tagged HSF1 WT or indicated deletion mutants with ABL2-eGFP. (G) GST in vitro pulldown assay of purified HSF1 (His) WT or deletion mutants incubated with purified ABL2 SH3 domain (GST).

dependent on the presence of the ABL2 SH3 domain (SI interaction between a noncanonical, proline-independent motif Appendix,Fig.S5B). of HSF1 with the ABL2 SH3 domain. Reciprocally, to identify the region of HSF1 responsible for its interaction with ABL2, we generated HSF1 truncation mutants Allosteric Inhibition of the ABL Kinases Impairs HSF1-E2F Expression. by site-directed mutagenesis (SI Appendix, Fig. S5C). Coimmu- We previously showed that the BBB-penetrable allosteric in- noprecipitation of the HSF1-truncated proteins in cells hibitor ABL001 (Asciminib) is effective in mouse models of cotransfected with ABL2-GFP revealed a loss in the interaction lung cancer brain metastasis (27). Thus, we examined if the between HSF1 and ABL2 upon removal of the LZ1-LZ3 do- loss of HSF1 protein observed in ABL2 knockdown cells mains (amino acids 130–206) in the HSF1 1–129 truncation was also induced by pharmacologic inhibition of the ABL2 mutant (SI Appendix, Fig. S5D). Curiously, this region of HSF1 kinase with ABL001. Treatment of HCC4006-BrM cells with lacks proline-rich sites that normally typify canonical SH3- ABL001 resulted in a dose-dependent decrease in the ex- binding proteins, suggesting that a noncanonical SH3-binding pression of HSF1, E2F1, and E2F8 proteins, while expression A motif might mediate the interaction between HSF1 and ABL2. of HSP90 remained unchanged (Fig. 5 ). Notably, inhibition of the ABL kinases with ABL001 via oral gavage in mice To test this possibility, we generated 10-amino acid truncation harboring established lung cancer brain metastases resulted mutants of HSF1 from residues 166–206 to more precisely in near complete loss of HSF1 protein expression and identify the residues responsible for binding the SH3 domain of – inhibited phosphorylation of the ABL kinase substrate CRKL ABL2. We found that the loss of residues spanning 187 195 of (Fig. 5B). ABL2-dependent regulation of HSF1 protein ex- HSF1 completely ablated the interaction between these two pression occurs posttranscriptionally, as RT-qPCR analysis of molecules (SI Appendix, Fig. S5E). As recent work revealed that PC9-BrM3 cells treated with ABL001 showed no measurable SH3 domains can bind to proline-independent hydrophobic change in HSF1 mRNA expression (SI Appendix,Fig.S6A) motifs (31), we hypothesized that a hydrophobic six amino acid and it is independent of the proteasome, as proteasomal in- – KLIQFL motif spanning residues 188 193 of HSF1 (Fig. 4D) hibition with MG132 in cells treated with an ABL allosteric might be responsible for the SH3-dependent interaction with inhibitor did not rescue HSF1 protein expression (SI Ap- ABL2. Deletion of this domain in HSF1 del188-193 and pendix,Fig.S6B). HSF1 del187-195 mutants ablated the interaction between HSF1 ABL kinase inhibitors can be classified into two broad cate- and ABL2 in cells (Fig. 4F). Furthermore, GST in vitro pulldown gories: 1) selective allosteric inhibitors that target the unique assays with purified His-HSF1 WT, His-HSF1 del187-195, or myristate-binding pocket located within the ABL kinase (SH1) His-HSF1 del188-193 mutants in the presence of purified GST- domain and 2) ATP-competitive inhibitors that bind to the cat- ABL2 SH3 protein revealed a complete loss in the interaction alytic site within the kinase domain and target not only ABL but between the ABL2 SH3 domain and the HSF1 deletion mutants several other tyrosine kinases (32). Notably, comparison of ABL (Fig. 4G). Together these findings uncover an intermolecular allosteric inhibitors (GNF5, ABL001) with ATP-competitive

Hoj et al. PNAS | December 29, 2020 | vol. 117 | no. 52 | 33491 Downloaded by guest on September 30, 2021 ABCHCC4006-BrM D DMSOGNF5ABL001NilotinibImatinibDasatinib ABL001 (uM) 01020 E2F1 GNF5 Nilotinib HSF1 Brain Mets in vivo 1.00 0.10 0.12 1.43 0.77 1.82 Time (h)0244872 0 24 48 72 1.00 0.57 0.11 E2F8 Vehicle ABL001 HSF1 E2F1 #1 #2 #1 #2 1.00 0.10 0.09 1.33 0.70 0.40 1.00 0.87 0.26 0.38 1.51 1.88 1.38 2.57 CDC6 1.00 0.36 0.12 HSF1 ABL2 1.00 0.04 0.02 1.08 0.55 0.22 1.00 1.30 0.49 0.41 0.30 0.39 0.81 2.05 E2F8 1.00 0.75 0.05 0.07 p-CRKL HSF1 Cyclin B1 1.00 0.20 0.21 (Y207) 1.00 0.24 0.25 1.05 0.67 0.44 1.00 0.84 0.22 0.07 1.00 0.15 0.01 0.01 0.97 1.42 1.44 2.06 HSP90 HSP90 CRKL Cl. Parp 1.00 0.93 0.91 1.00 0.70 0.75 0.97 0.98 0.88 p-CRKL Actin ABL2 0.00 0.01 0.39 1.00 0.00 0.00 0.00 0.00 (Y207) 1.00 0.55 0.34 0.98 0.98 0.48 p-CRKL 1.00 0.03 0.02 p-CRKL (Y207) 1.00 0.99 0.48 0.07 0.43 0.01 0.01 0.03 (Y207) CRKL 1.00 0.35 0.55 0.27 0.05 0.02 CRKL CRKL Actin Actin Actin

EFDMSO DMSO GNF5 Nilotinib GPC9-BrM3 H HCC4006-BrM ABL2-eGFP + + ++ ATAD2 1.00 FLAG-HSF1 ATAD2 1.00 - +++ CDC6 CDC6 ABL2 DCTPP1 (GFP) DCTPP1 DNMT1 DNMT1

0.75 ion E2F8 0.75 E2F8 ss

HSF1 FLAG IP: MCM2 (FLAG) MCM2 MCM3 MCM3 ABL2 MCM4 (GFP) 0.50 MCM4 0.50 HSF1 MCM6 MCM6

(pS326) MCM7 MCM7 Expre mRNA e ive mRNA Expression ive mRNA iv HSF1 NASP NASP t (FLAG) PRKDC 0.25 PRKDC 0.25 Relat WCL Rela pCRKL Fold Change (Normalized to 18S) RANBP1 RANBP1 Fold Chagne (Normalized to 18S) (Y207) SMC1A SMC1A CRKL SMC3 SMC3 0 0 Actin 02448 02448 ABL001 (h) ABL001 (h)

Fig. 5. ABL allosteric inhibitors impair expression of HSF1 and E2F proteins and downstream transcriptional targets. (A) Immunoblots of indicated proteins in HCC4006-BrM cells treated with indicated concentrations of ABL001 for 72 h. Quantification normalized to actin loading control and DMSO control (left lane). (B) Immunoblots of indicated proteins in established in vivo brain metastases from mice injected with H1975 lung cancer cells and treated with vehicle (n = 2) or 100 mg/kg ABL001 (n = 2) via oral gavage. Mice were treated at 3, 12, and 24 h prior to harvesting brain metastases. Protein quantification data normalized to actin loading control and the leftmost lane (Vehicle #1). (C) Immunoblots of PC9-BrM3 cells treated with DMSO or the ABL allosteric inhibitors GNF5 (10 μM) or ABL001 (10 μM), and ATP-competitive Nilotinib (1 μM), Imatinib (2 μM), or Dasatinib (150 nM) for 48 h. Indicated equipotent doses for each drug were

selected after determining corresponding IC50 values. Phospho-CRKL Y207 blot used as surrogate marker for ABL kinase inhibition. (D) Immunoblots of PC9- BrM3 cells treated with 10 μM GNF5 or 1 μM Nilotinib for the indicated timepoints. Protein quantification data normalized to actin loading control and the leftmost lane (DMSO). For all immunoblots of phospho-CRKL, quantification data were normalized to total CRKL levels. (E) Coimmunoprecipitation pulldown assay for Flag-HSF1 in 293T cells cotransfected with ABL2-GFP and treated with the indicated inhibitors for 24 h. (F) Proposed diagram of the SH3-dependent interaction between ABL2 and HSF1 in the absence (Upper) and presence (Lower) of ABL allosteric versus ATP-competitive pharmacologic inhibitors. (G and H) RT-qPCR analysis of E2F target gene mRNA expression in PC9-BrM3 (G) or HCC4006-BrM (H) cells treated with 10 μM ABL001 for indicated timepoints.

inhibitors (Nilotinib, Imatinib, and Dasatinib) revealed that cells these findings, our work suggests that, in contrast to ATP- treated with the ABL allosteric inhibitors exhibited a profound competitive inhibition, allosteric inhibition of the ABL2 kinase loss in HSF1, E2F1, and E2F8 proteins relative to ATP-competitive results in an inactive state wherein the ABL2 SH3 domain is inhibitor treatment (Fig. 5C). Expression of CDC6, an HSF1-E2F sterically inaccessible to binding with SH3-interacting proteins target gene, was preferentially decreased by the ABL allosteric in- such as HSF1 (Fig. 5F). Collectively, our results show that in hibitors, whereas expression of HSP90 was not affected by any of contrast to the ATP-competitive inhibitors, the ABL allosteric the inhibitors tested (Fig. 5C). Similar to HSF1 knockdown, treat- inhibitors phenocopy the effects of ABL2 knockdown and ment with the ABL allosteric inhibitor GNF5 resulted in loss of mightbeusedtodisruptHSF1-E2Ftranscriptioninbrain- Cyclin B1 expression as well as a robust increase in cleaved PARP metastatic lung cancer cells. In this regard, PC9-BrM3 and levels that were not observed by treatment with Nilotinib (Fig. 5D). HCC4006-BrM cells treated with ABL001 exhibited profound These findings are consistent with published data showing that decreases in mRNA expression of the HSF1-E2F target gene in contrast to ABL allosteric inhibitors, the ATP-competitive panel (Fig. 5 G and H). Together, these results demonstrate inhibitors elicit ERK activation in lung cancer cells and other that HSF1-E2F transcription networks can be pharmacologi- tumors (33). Notably, coimmunoprecipitation and immuno- cally targeted with ABL kinase allosteric inhibitors. blot analysis of Flag-HSF1 cotransfected with ABL2-GFP in cells treated with either an allosteric or ATP-competitive in- An HSF1-E2F Transcriptional Signature is Up-Regulated in Lung hibitor revealed a complete loss in the interaction between Adenocarcinoma Patients and Is Predictive of Poor Survival. To HSF1 and ABL2 only under allosteric inhibition (Fig. 5E). evaluate whether the HSF1-E2F target gene signature is clini- Structural studies revealed striking differences in ABL kinase cally relevant, we analyzed patient datasets to determine if conformation and activity states under ATP-competitive– HSF1-E2F target gene expression is predictive of patient survival inhibited vs. allosteric-inhibited conditions (34). In line with outcomes. Analysis of microarray-based mRNA expression

33492 | www.pnas.org/cgi/doi/10.1073/pnas.2007991117 Hoj et al. Downloaded by guest on September 30, 2021 correlated to overall patient survival data revealed that high signature co-occurred in the same patients, and mutual exclu- expression of the HSF1-E2F–coregulated target genes ATAD2, sivity analysis showed high co-occurrence for almost all gene CDC6, PRKDC, RANBP1, MCM2, MCM3, MCM4, and MCM7 pairs tested from the HSF1-E2F gene panel (Fig. 6 I and J). each individually were predictive of significantly poorer overall Collectively, these data provide supporting evidence for an survival in lung adenocarcinoma patients (Fig. 6 A–H). In ad- HSF1-E2F–coregulated transcriptional gene signature that may dition, analysis of TCGA datasets for lung adenocarcinoma be clinically relevant as a prognostic marker for patient survival showed that high expression of the HSF1-E2F target gene in lung adenocarcinoma (Fig. 6K).

AB CD 100 ATAD2 100 CDC6 PRKDC RANBP1 High 100 100 High Mid High High Mid Low Mid Mid Low Low Low 50 50 50

% OS 50 % OS % OS % OS

p-value < 0.0001 p-value < 0.0001 p-value < 0.0001 p-value < 0.0001 0 0 0 0 0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250 Time (months) Time (months) Time (months) Time (months) EF GH MCM2 MCM3 MCM4 MCM7 100 100 High 100 100 High Mid High High Mid Low Mid Mid Low Low Low 50 50 50 50 % OS % OS % OS % OS

p-value = 0.004 p-value < 0.0001 p-value < 0.0001 p-value < 0.0001

0 0 0 0 MEDICAL SCIENCES 0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250 Time (months) Time (months) Time (months) Time (months)

TCGA Provisional - LUAC - mRNA I DNMT1 K NASP ABL2-HSF1-E2F Signaling in LCBM RANBP1 MCM4 MCM6 Cell Survival MCM2 Cell Proliferation CDC6 ATAD2 P-Y- ON ABL2 MCM7 HSF1 E2F8 G2M Checkpoint Activation MCM3 PRKDC SMC3 SMC1A Cytoplasm DCTPP1 Nucleus Pathway: Expression Heatmap -3 3 n = 515 patients E2Fs HSF1 HSF1-E2F Target Gene Signature TCGA Provisional - LUAC Degenerate HSE Mutual Exclusivity Analysis 3 J ATAD2 Allosteric Inhibition CDC6 (ABL001, GNF5) DCTPP1 Genetic Depletion (shRNA) DNMT1 2 E2F8

MCM2 Co-occurence P-Y- ABL2 MCM3 OFF HSF1 Cyclin B1 G2M Checkpoint MCM4 1 Inhibition MCM6 Cleaved Apoptosis MCM7 PARP NASP Cytoplasm PRKDC 0 Nucleus RANBP1 Pathway: SMC1A E2Fs SMC3 HSF1 HSF1-E2F Target Gene Signature -1 Mutually-Exclusive Log2 Degenerate HSE E2F8 CDC6 MCM2MCM3MCM4MCM6MCM7NASP SMC3 ATAD2 DNMT1 PRKDC SMC1A Odds Ratio DCTPP1 RANBP1

Fig. 6. Expression of HSF1-E2F target genes is predictive of poor survival outcomes and co-occurs in lung adenocarcinoma patients. (A–H) Overall survival (OS) curves derived from human lung adenocarcinoma patient microarray datasets with high versus low expression of HSF1-E2F target genes ATAD2 (A), CDC6 (B), PRKDC (C), RANBP1 (D) MCM2, (E), MCM3 (F), MCM4 (G), and MCM7 (H). Curves were generated using KMPlot analysis tool and patient groups were stratified by tertile (n = 720 total patients). For all survival curves, statistical analysis performed by log-rank Mantel–Cox test. (I) Clustered heatmap depicting mRNA expression for HSF1-E2F target gene panel in human lung adenocarcinoma provisional patient dataset available through TCGA. Data were analyzed usingthe cBioPortal online analysis software. (J) Mutual exclusivity analysis of HSF1-E2F target gene expression in the same TCGA lung adenocarcinoma dataset to

demonstrate high co-occurrence of target gene expression. Heat map scaling based on log2 odds ratio as calculated using cBioPortal analysis tool. (K) Model diagram illustrating ABL2-HSF1-E2F signaling in lung adenocarcinoma brain metastasis.

Hoj et al. PNAS | December 29, 2020 | vol. 117 | no. 52 | 33493 Downloaded by guest on September 30, 2021 Discussion Cell Lines and Cell Culture. Human non-small cell lung cancer (NSCLC) cell lines While HSF1 has been shown to support initiation and progres- HCC4006 and H1975 were purchased from American Type Culture Collection (ATCC). PC9 parental, H2030 parental, and H2030-BrM3 cells were a gift sion of distinct types of primary tumors, little is known regarding from Joan Massagué (Memorial Sloan Kettering Cancer Center, New York, the role of HSF1 during metastasis (18, 20, 22). We now show NY). PC9-BrM3 and HCC4006-BrM cell lines were derived in the A.M.P. lab- that brain-colonizing lung cancer cells up-regulate expression of oratory by serial intracardiac injection as described in the following section. HSF1 and that loss of HSF1 in these cells results in dramatic Parental and derivative cell line pairs were subjected to short tandem repeat decreases in cell survival in vitro and metastatic burden in vivo. profiling through the Duke University DNA Analysis Facility Human cell line Additionally, our data show that the transcriptional response authentication service to confirm their authenticity. NSCLC lines were driven by HSF1 is strikingly divergent from its canonical role as maintained in RPMI 1640 (Life Technologies) supplemented with 10% the master regulator of the heat shock response. Unbiased tetracycline-screened fetal bovine serum (FBS, HyClone), 10 mM Hepes, transcriptomic profiling comparing brain-metastatic and parental 1 mM sodium pyruvate, and 0.2% glucose. H293T cells used for transfection and virus production were purchased from ATCC and were maintained in cell lines reveals significant enrichment of E2F transcription DMEM (Life Technologies) with 10% FBS (Corning). All cultures were main- family gene signatures in brain-metastatic cells whose expression tained at 37 °C in humidified air containing 5% CO2. In studies employing is dependent on HSF1. Previous work in C. elegans revealed an the use of dox-inducible constructs, the dose of dox (Millipore Sigma) was HSF1-E2F transcriptional program is required during develop- determined empirically. ment, whereby HSF1 binds DNA at promoter regions of E2F For experiments assessing effects of pharmacologic inhibitors in vitro target genes in a heat shock-independent manner (26). Similarly, (GNF5, ABL001, Nilotinib, Dasatinib, Imatinib), drugs were dissolved in DMSO analysis of promoter regions for predicted HSF1-E2F–coregulated with the final concentration of DMSO in culture media not exceeding 0.1% genes in human cells revealed the presence of “degenerate” HSEs vol/vol. The proteasomal inhibitor MG132 (Sigma) was dissolved in DMSO for downstream of E2F DNA-binding motifs which were confirmed use in cell-based in vitro assays. Inhibitors were synthesized by the Duke University Small Molecule Synthesis Facility and were validated by liquid empirically to be occupied by HSF1, thus supporting the hypothesis 1 chromatography–mass spectrometry and H-NMR techniques in addition to that HSF1 is a cofactor required for expression of these E2F gene cell-based assays confirming effects on both cell viability (Cell-Titer Glo) and targets in metastatic lung cancer cells. Importantly, analysis of hu- kinase target inhibition (Western blot). man patient data revealed that high expression of an HSF1-E2F transcriptional signature co-occurs in patient populations with poor Intracardiac Injection. All animal studies were performed in accordance with survival outcomes, suggesting that an HSF1-E2F target gene panel protocols approved by the Duke University Division of Laboratory Animal may hold prognostic value in patients with aggressive metastatic Resources Institutional Animal Care and Use Committee. PC9-BrM3 cells disease. Future studies comparing samples of paired patient primary transduced with a stable pFU-luciferase-Tomato (pFuLT) lentiviral vector in tumors and corresponding brain metastases should be undertaken addition to shRNA constructs as described in figure legends were injected to evaluate whether HSF1 protein expression and an HSF1- intracardially, and mice were subsequently monitored in vivo using an IVIS – XR bioluminescent imager to confirm both proper anatomical injection as E2F coregulated transcription network may be relevant to pro- well as to monitor for metastatic disease progression. Eight- to 12-wk-old gression in this disease setting. Furthermore, the importance of age-matched female athymic nu/nu mice were used for all studies (Jackson HSF1 activation and E2F-coregulated transcription in extracranial Laboratory). Mice were anesthetized with 5% isoflurane prior to injections. metastasis merits further investigation. For all studies, 4 × 105 lung cancer cells suspended in 100 μL of PBS were The ABL kinases, ABL1 and ABL2, function downstream of injected into the left cardiac ventricle with a 30-gauge needle and animals known HSF1 activators including EGFR, HER2, and RAS (20, were monitored until full recovery from anesthesia. 33, 35–38). Additionally, the ABL kinases regulate expression ABL001 (Asciminib) was used to pharmacologically inhibit the ABL kinases and activity of multiple factors that function as HSF1 tran- in tumor-bearing mice in vivo and was prepared as a suspension in sterile scriptional coregulators, including E2F1 (39–41). Recent work by 0.5% methylcellulose/0.5% Tween-80 as described previously (42). To eval- our laboratory uncovered a critical role for the ABL2 kinase in uate the pharmacologic effects of ABL001 on ABL kinase activity and downstream signaling in brain metastases in vivo, mice were injected with the regulation of a TAZ-dependent transcriptional network re- H1975 pFulT lung cancer cells and metastatic burden was allowed to prog- quired for the early colonization of lung cancer brain metastases ress over the span of 1 mo. Mice with established brain metastases were (27). We now show that the SH3 domain of ABL2 forms a administered either vehicle or 100 mg/kg ABL001 via oral gavage at 3, 12, complex with a hydrophobic KLIQFL motif located within the and 24 h prior to euthanasia and tissue harvest. Immediately following eu- LZ1-3 domains of HSF1, and knockdown of ABL2 results in thanasia, mouse brains were freshly dissected and tissue sections with visible impaired expression of HSF1, E2F1, and E2F8 proteins in brain- brain metastases were excised and lysed in RIPA buffer containing protease- metastatic lung cancer cell lines. Notably, we found that phar- phosphatase inhibitor mixture (Cell Signaling). Harvested samples were macologic inhibition of the ABL kinases using selective ABL digested for 15 min on a rotator at 4 °C followed by centrifugation for allosteric inhibitors, but not ATP-competitive inhibitors, ablates 15 min at 4 °C to clear the lysate of myelin and cell debris. Protein lysates were temporarily stored at −20 °C prior to immunoblotting. the physical interaction between ABL2 and HSF1, results in markedly decreased expression of HSF1, E2F1, and E2F8 pro- Derivation of Brain-Metastatic Lung Cancer Cell Lines. Derivation of brain- teins in brain-metastatic lung cancer cells, and results in deple- metastatic cell lines was performed as described previously (5). Eight- to tion of HSF1-E2F transcriptional targets. These findings 12-week-old athymic nude mice were injected intracardially with either PC9, highlight potential differences affecting intramolecular and in- HCC4006, or H1975 parental cell lines (4 × 105 cells per injection) and were termolecular protein–protein interactions induced by allosteric monitored for metastatic progression using an IVIS XR bioluminescent im- versus ATP-competitive kinase inhibitors that might have imager. Approximately 30 d postintracardiac injection, subsets of mice pre- portant therapeutic implications. Importantly, the targetable senting with advanced brain metastases were euthanized and whole brains nature of the ABL2-HSF1-E2F signaling network identifies ABL were collected, minced, and digested in RPMI culture medium supplemented allosteric inhibitors as a potentially effective therapy for the with 0.125% collagenase III and 0.1% hyaluronidase. Cells suspended in di- gestion media were then placed on a rotator for 4–5 h at room temperature. treatment of metastatic lung cancers characterized by high ex- Following incubation, cells were centrifuged and resuspended in 0.25% pression of HSF1. trypsin for 10 min at 37 °C, followed by resuspension in culture media containing 1× Anti-Anti (Thermo Fisher). Cell lines were expanded to near- Materials and Methods confluence in culture prior to sorting for Tomato-positive cells for future Additional experimental details and methods are provided in SI Appendix, experimental use. including procedures for RNA-seq analysis, real-time PCR analysis, immunoblotting, immunofluorescence, and patient survival analysis. RNA sequenc- Statistical Analysis. Statistical analyses were performed using GraphPad Prism ing raw data have been deposited in the National Center for Biotechnology 8 software. Mouse numbers per group were determined through statistical Information Gene Expression Omnibus (GEO) under accession no. GSE149246. power calculations (α = 0.05) where 10 mice per group allows for 90% power

33494 | www.pnas.org/cgi/doi/10.1073/pnas.2007991117 Hoj et al. Downloaded by guest on September 30, 2021 to detect intergroup differences of 50% and assuming intragroup variability study are available upon completion of a Materials Transfer Agreement. of 25%. For Kaplan–Meier survival analysis, P values were calculated using RNA-seq data have been deposited in GEO (accession no. GSE149246). log-rank (Mantel–Cox) testing. Statistical comparisons of two groups were conducted using Student’s t tests (unpaired, two-tailed). For comparisons ACKNOWLEDGMENTS. We thank Dr. Joan Massagué for providing the PC9 involving more than two groups, data were evaluated by ANOVA followed parental, H2030 parental, and H2030-BrM3 cell lines, as well as Dr. Anthony by post hoc testing as described in figure legends. Post hoc testing was Koleske (Yale University, New Haven, CT) for providing DNA plasmids. We performed only when statistical significance was achieved from the ANOVA. thank the Duke University genome sequencing facility for providing assis- For comparisons between groups of unequal size, the mean value was used tance with RNA-seq data acquisition, the Duke Flow Cytometry Shared Re- to allow for statistical analysis by ANOVA. For all tests, we considered a P source for assistance with cell sorting, and the Duke Light Microscopy Core value less than 0.05 as statistically significant. Data shown represent aver- Facility. We acknowledge Courtney McKernan and Ashley Colemon for tech- nical assistance. This work was supported by NIH National Cancer Institute ages ± SEM unless otherwise indicated in figure legends. Grants R01CA195549 (to A.M.P.), F31CA22496001 (to J.P.H.), F99CA245732- 01 (to J.P.H.), F31CA243293-01A1 (to B.M.), and 5T32GM007105-44 (to J.P.H. Data Availability. Requests for resources and reagents found in this study and B.M.), the Lung Cancer Research Foundation Free to Breathe Metastasis should be directed to and will be fulfilled by the corresponding author, A.M.P. Research Grant (to A.M.P.), the Emerson Collective, and Duke SPORE in Brain ([email protected]). All unique and stable reagents generated in this Cancer Grant P50CA190991.

1. W. Schuette, Treatment of brain metastases from lung cancer: Chemotherapy. Lung 23. C. Xi, Y. Hu, P. Buckhaults, D. Moskophidis, N. F. Mivechi, Heat shock factor Hsf1 Cancer 45, S253–S257 (2004). cooperates with ErbB2 (Her2/Neu) protein to promote mammary tumorigenesis and 2. K. J. Stelzer, Epidemiology and prognosis of brain metastases. Surg. Neurol. Int. 4, metastasis. J. Biol. Chem. 287, 35646–35657 (2012). S192–S202 (2013). 24. X. Jin, D. Moskophidis, N. F. Mivechi, Heat shock transcription factor 1 is a key de- 3. B. D. Fox, V. J. Cheung, A. J. Patel, D. Suki, G. Rao, Epidemiology of metastatic brain terminant of HCC development by regulating hepatic steatosis and metabolic syn- tumors. Neurosurg. Clin. N. Am. 22,1–6 (2011). drome. Cell Metab. 14,91–103 (2011). 4. Q. Zeng et al., Synaptic proximity enables NMDAR signalling to promote brain me- 25. K. H. Su, S. Dai, Z. Tang, M. Xu, C. Dai, Heat shock factor 1 is a direct antagonist of tastasis. Nature 573, 526–531 (2019). AMP-activated protein kinase. Mol. Cell 76, 546–561.e8 (2019). 5. M. Valiente et al., Serpins promote cancer cell survival and vascular co-option in brain 26. J. Li, L. Chauve, G. Phelps, R. M. Brielmann, R. I. Morimoto, E2F coregulates an es- metastasis. Cell 156, 1002–1016 (2014). sential HSF developmental program that is distinct from the heat-shock response. 6. Q. Chen et al., Carcinoma-astrocyte gap junctions promote brain metastasis by cGAMP Genes Dev. 30, 2062–2075 (2016). transfer. Nature 533, 493–498 (2016). 27. J. P. Hoj, B. Mayro, A. M. Pendergast, A TAZ-AXL-ABL2 feed-forward signaling Axis – 7. E. E. Er et al., Pericyte-like spreading by disseminated cancer cells activates YAP and promotes lung adenocarcinoma brain metastasis. Cell Rep. 29, 3421 3434.e8 (2019). MRTF for metastatic colonization. Nat. Cell Biol. 20, 966–978 (2018). 28. Y. Kienast et al., Real-time imaging reveals the single steps of brain metastasis for- – 8. D. J. H. Shih et al., Genomic characterization of human brain metastases identifies mation. Nat. Med. 16, 116 122 (2010). drivers of metastatic lung adenocarcinoma. Nat. Genet. 52, 371–377 (2020). 29. A. Subramanian et al., Gene set enrichment analysis: A knowledge-based approach 9. P. E. Fecci et al., The evolving modern management of brain metastasis. Clin. Cancer for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U.S.A. 102, MEDICAL SCIENCES – Res. 25, 6570–6580 (2019). 15545 15550 (2005). 10. J. Anckar, L. Sistonen, Regulation of HSF1 function in the heat stress response: Im- 30. E. Morgunova et al., Structural insights into the DNA-binding specificity of E2F family transcription factors. Nat. Commun. 6, 10050 (2015). plications in aging and disease. Annu. Rev. Biochem. 80, 1089–1115 (2011). 31. A. Gkourtsa, J. van den Burg, T. Avula, F. Hochstenbach, B. Distel, Binding of a proline- 11. R. Gomez-Pastor, E. T. Burchfiel, D. J. Thiele, Regulation of heat shock transcription independent hydrophobic motif by the Candida albicans Rvs167-3 SH3 domain. Mi- factors and their roles in physiology and disease. Nat. Rev. Mol. Cell Biol. 19,4–19 crobiol. Res. 190,27–36 (2016). (2018). 32. E. K. Greuber, P. Smith-Pearson, J. Wang, A. M. Pendergast, Role of ABL family kinases 12. D. W. Neef et al., A direct regulatory interaction between chaperonin TRiC and stress- in cancer: From leukaemia to solid tumours. Nat. Rev. Cancer 13, 559–571 (2013). responsive transcription factor HSF1. Cell Rep. 9, 955–966 (2014). 33. J. J. Gu et al., Inactivation of ABL kinases suppresses non-small cell lung cancer me- 13. Y. Shi, D. D. Mosser, R. I. Morimoto, Molecular chaperones as HSF1-specific tran- tastasis. JCI Insight 1, e89647 (2016). scriptional repressors. Genes Dev. 12, 654–666 (1998). 34. L. Skora, J. Mestan, D. Fabbro, W. Jahnke, S. Grzesiek, NMR reveals the allosteric 14. J. Zou, Y. Guo, T. Guettouche, D. F. Smith, R. Voellmy, Repression of heat shock opening and closing of Abelson tyrosine kinase by ATP-site and myristoyl pocket transcription factor HSF1 activation by HSP90 (HSP90 complex) that forms a stress- inhibitors. Proc. Natl. Acad. Sci. U.S.A. 110, E4437–E4445 (2013). sensitive complex with HSF1. Cell 94, 471–480 (1998). 35. D. Srinivasan, R. Plattner, Activation of Abl tyrosine kinases promotes invasion of 15. M. Akerfelt, R. I. Morimoto, L. Sistonen, Heat shock factors: Integrators of cell stress, aggressive breast cancer cells. Cancer Res. 66, 5648–5655 (2006). – development and lifespan. Nat. Rev. Mol. Cell Biol. 11, 545 555 (2010). 36. C. C. Mader et al., An EGFR-Src-Arg-cortactin pathway mediates functional matura- 16. T. Neudegger, J. Verghese, M. Hayer-Hartl, F. U. Hartl, A. Bracher, Structure of human tion of invadopodia and breast cancer cell invasion. Cancer Res. 71, 1730–1741 (2011). heat-shock transcription factor 1 in complex with DNA. Nat. Struct. Mol. Biol. 23, 37. D. Li, A. Yallowitz, L. Ozog, N. Marchenko, A gain-of-function mutant p53-HSF1 feed – 140 146 (2016). forward circuit governs adaptation of cancer cells to proteotoxic stress. Cell Death Dis. 17. P. A. Mercier, N. A. Winegarden, J. T. Westwood, Human heat shock factor 1 is pre- 5, e1194 (2014). – dominantly a nuclear protein before and after heat stress. J. Cell Sci. 112, 2765 2774 38. L. Meng, V. L. Gabai, M. Y. Sherman, Heat-shock transcription factor HSF1 has a (1999). critical role in human epidermal growth factor receptor-2-induced cellular transfor- 18. M. L. Mendillo et al., HSF1 drives a transcriptional program distinct from heat shock to mation and tumorigenesis. Oncogene 29, 5204–5213 (2010). support highly malignant human cancers. Cell 150, 549–562 (2012). 39. M. C. Birchenall-Roberts et al., The p120-v-Abl protein interacts with E2F-1 and reg- 19. N. Kourtis et al., Oncogenic hijacking of the stress response machinery in T cell acute ulates E2F-1 transcriptional activity. J. Biol. Chem. 272, 8905–8911 (1997). lymphoblastic leukemia. Nat. Med. 24, 1157–1166 (2018). 40. K. K. Wong et al., v-Abl activates c-myc transcription through the E2F site. Mol. Cell. 20. C. Dai, L. Whitesell, A. B. Rogers, S. Lindquist, Heat shock factor 1 is a powerful Biol. 15, 6535–6544 (1995). multifaceted modifier of carcinogenesis. Cell 130, 1005–1018 (2007). 41. Y. Parada et al., BCR-ABL and interleukin 3 promote haematopoietic cell proliferation 21. N. Kourtis et al., FBXW7 modulates cellular stress response and metastatic potential and survival through modulation of cyclin D2 and p27Kip1 expression. J. Biol. Chem. through HSF1 post-translational modification. Nat. Cell Biol. 17, 322–332 (2015). 276, 23572–23580 (2001). 22. C. Dai et al., Loss of tumor suppressor NF1 activates HSF1 to promote carcinogenesis. 42. A. A. Wylie et al., The allosteric inhibitor ABL001 enables dual targeting of BCR-ABL1. J. Clin. Invest. 122, 3742–3754 (2012). Nature 543, 733–737 (2017).

Hoj et al. PNAS | December 29, 2020 | vol. 117 | no. 52 | 33495 Downloaded by guest on September 30, 2021