HSP70 Inhibition Limits FAK-Dependent Invasion and Enhances the Response to Melanoma Treatment with BRAF Inhibitors

Metastatic melanoma is an aggressive and usually fatal malignancy. Advanced melanomas have multiple oncogenic drivers, and despite advances with BRAF and immune checkpoint inhibitors, resistance to therapy is a significant problem (1). Given the incomplete efficacy and durability of current treatment options, and the enormous propensity of melanomas for resistance, there is an urgent need to identify new molecular targets for melanoma. In this work, we test the hypothesis that the major stress-inducible heat shock protein 70 (also called HSP72 or HSPA1A but hereafter referred to as HSP70) might provide a novel therapeutic avenue, both alone and in combination with current therapies.

The HSP70 family has over 10 members, many of which show tissue-specific expression and/or organelle-specific localization (2, 3). In contrast with constitutively expressed forms like HSPA5 (Bip, GRP78), HSPA8 (HSC70), and HSPA9 (GRP75), all of which are required for cell viability, the major stress-induced form, HSP70, is expressed at extremely low levels in unstressed cells and is not required for life (4). Consequently, HSP70 inhibitors show tumor-selective cytotoxicity, with limited toxicity in normal, nontransformed cells (5). We have developed and characterized a series of HSP70 inhibitors, beginning with the “first-in-class” compound, PES (2-Phenylethynesulfonamide, empirical formula C₈H₇NO₂S, molecular weight 181.2). PES binds specifically to the lesser-conserved substrate-binding domain (SBD) of HSP70, and it does not bind to family members like BiP and GRP75, or to the chaperone HSP90 (6). We found that PES is broadly cytotoxic to cancer but not normal cells, and it markedly protects Eμ-myc mice from lymphoma, without any evidence of organ or general toxicity (6). Subsequently, we have developed and characterized two derivatives of PES, called PES-Cl and PET-16, which show superior cytotoxicity and enhanced efficacy as anticancer agents, both in vitro and in vivo (7–9). The chemical structures of PES, PES-Cl, and PET-16 are depicted (Supplementary Fig. S1). Notably, silencing or deletion of HSP70 (HSPA1A/A1B) in tumor cells reduces the cytotoxicity of these compounds, supporting the premise that they are specific for HSP70 and are not other family members (6, 9). Mechanistically, these compounds are cytotoxic by virtue of their ability to inhibit proteasome function, alter the solubility and proper folding of HSP70 client proteins, and via their ability to inhibit autophagy, a cancer-critical survival pathway (6–8, 10). We elucidated the mechanism of action of PES, PES-Cl, and PET-16 using a combination of isothermal calorimetry and by solving the crystal structure of PET-16 bound to the SBD of the closely related bacterial orthologue of HSP70, DnaK. These analyses revealed that PET-16 interacts with loop alpha-beta of the SBD and functions as an allosteric regulator to prevent allosteric cycling of HSP70 (9). The specificity of PES derivatives for HSP70, and their efficacy on tumor lines in vitro, provided us a unique opportunity to study the role of this chaperone in melanoma.

In this work, we conducted the first tumor microarray of melanoma for HSP70, using an antibody specific for the major stress-induced form that does not cross-react with other isoforms (11). We show that HSP70 is highly expressed in a large percentage of melanomas, and that expression of this protein increases with advanced stage. We use the technique of reverse phase protein array (RPPA) in melanomas to identify HSP70 client proteins, and identify pFAK and BRAF as two melanoma-relevant and specific client proteins of HSP70. We show that inhibition of pFAK using the HSP70 inhibitor PET-16 leads to decreased ability of melanomas to migrate, invade, and metastasize in vivo. In addition, we show that the V600E mutant form of BRAF, present in up to 50% of melanoma tumors, is a client of HSP70. Along these lines, we show that PET-16 synergizes with BRAF inhibitors, retains cytotoxicity in melanomas resistant to BRAF inhibitors, and extends the durability of treatment with BRAF inhibitors. These combined data augur well for the eventual use of HSP70 inhibitors for melanoma therapy.

All human melanoma cell lines were obtained from the Wistar Institute collection and were confirmed by genotyping. These cells were maintained in MCDB153 (Sigma)/Leibovitz's L-15 (Cellgro) medium (4:1 ratio) supplemented with 2% FCS and 2 mmol/L CaCl₂. H1299 cells, B16-F10, and fibroblasts were obtained from the ATCC within 6 months of their use, and were maintained in DMEM (Invitrogen). The FS5 cell line, kindly provided by Ashani Weeraratna (Wistar Institute, Philadelphia, PA), was maintained in RPMI 1640, and the Yumm1.7 cell line (kindly provided by Marcus Bosenburg, Yale University, New Haven, CT) was maintained in DMEM/F12 (Invitrogen). All the above media were supplemented with 10% FBS (Invitrogen) and 100 U/mL penicillin and streptomycin. Cell stocks were fingerprinted using AmpFLSTR Identifiler PCR Amplification Kit from Life Technologies at The Wistar Institute Genomics Facility. PET-16 (Triphenyl(phenylethynyl)phosphonium bromide; Sigma; catalog #S16773) and vemurafenib/PLX4032 (S1267; Selleckchem) were prepared as 50 mmol/L stock solutions in DMSO and stored at -80°C.

Western blotting was performed as described (7). Primary antibodies used in this study include HSP70 (4873S; Cell Signaling Technology), anti-HA (3724; Cell Signaling Technology), total FAK (EMD Millipore; 05-537), FAK Tyr-397-P (44-625G; Invitrogen), BRAF (sc-5284; Santa Cruz Biotechnology), and GAPDH (14C10; Cell Signaling Technology; 2118). Secondary antibodies conjugated to horseradish peroxidase were used at a dilution of 1:10,000 (Jackson Immunochemicals). ECL (Amersham; RPN2232) was applied to blots, and protein levels were detected using autoradiography. Densitometry quantification of protein signals was performed using ImageJ software (NIH). For IP westerns, 1,000 μg of lysate was immunoprecipitated as described (12) with 0.5 μg of antisera to HSP70, followed by SDS-PAGE, transfer, and Western analysis using antisera to total FAK, FAK Tyr-397-P, and BRAF. The 1205Lu cell line with shRNA knockdown of PTK2/FAK was generated by infection with the lentiviral vector pLKO.1-puro carrying either of two shRNA sequences against human PTK2: shRNA (CCGGTCGAATGATAAGGTGTA; TRCN0000001620 and CAACAGGTGAAGAGCGATTAT; TRCN0000194984). Stable cells were selected using puromycin (1 μg/mL), and PTK2/FAK knockdown was confirmed by Western blot.

Human melanoma cells were plated at 2,000 cells per well in 96-well plates. After 24 hours, cells were treated with serial dilutions of individual drugs or combinations of two drugs at a constant molar ratio. After 72 hours, cell viability was measured with Alamar blue (Life Technologies; DAL1025) using a SynergyHT plate reader (BioTek). Combination index (CI) values, established by the Chou–Talalay method (13), were calculated using the CompuSyn software package (CompuSyn). BRAF inhibitor–resistant cells were generated by growing parental cells in the presence of 1 μmol/L, 5 μmol/L, or 10 μmol/L PLX4720 for at least four passages. Resistance was confirmed by IC₅₀ analysis.

Immunofluorescence of fixed paraffin-embedded tissue sections was performed as previously described (14). The following primary antibodies were used: FAK Tyr-397-P (44-625G; Invitrogen), FAK Tyr-397-P (700255; Life Technologies), SQSTM1/p62 (ab56416; Abcam), and SQSTM1/p62 (sc-28359; Santa Cruz Biotechnology). Tissue microarrays (TMA) containing 0.9-mm cores of 77 nevi, 8 in-situ, 50 invasive, and 103 metastatic malignant melanomas constructed at the Fox Chase Cancer Center (Philadelphia, PA); full details about these samples are previously described (15). Paired melanoma tissue specimens before and after vemurafenib treatment were obtained at the Abramson Cancer Center, Philadelphia, PA. All samples were procured under approved Institutional Review Board protocols at the relevant institution. The HSP70 monoclonal antibody (1:100; Cell Signaling Technology; 4873S) is specific for the major stress-induced form of HSP70, and does not cross-react with other family members (11). This was incubated with the tissue after heat-induced epitope retrieval with citrate buffer (pH 6.0). The samples were incubated with primary antibody overnight followed by incubation with a biotinylated secondary antibody (Jackson Immuno Research; 1:200) for 30 minutes. Detection was performed using Nova Red for 15 minutes, and the slides were counterstained with hematoxylin (Fisher Scientific). Immunohistochemical stain for HSP70 was interpreted semiquantitatively by assessing the intensity and extent of staining on the tissue sections present on the slides according using the H-Score method as previously described (16). HSP70 expression was scored in blinded manner by a board-certified dermatopathologist (Xiaowei Xu, University of Pennsylvania, Philadelphia, PA).

Human skin reconstructs were generated as previously described (17). Briefly, dermal reconstructs consisted of rat tail type I collagen contracted for 4 days with embedded dermal fibroblasts (7.5 × 10⁴ per dish) before seeding 1205Lu cells (0.83 × 10⁵) together with keratinocytes (4.17 × 10⁵) on top of the dermal reconstructs over a 4-day period in media before air exposure until day 18 to allow epidermal differentiation. During the last 7 days, skin reconstructs were treated with DMSO or PET-16 that were added to the medium. At day 18, skin reconstructs were harvested and fixed in 10% neutral buffered formalin for 2 to 3 hours, embedded into paraffin, sectioned (5 μm), and then stained with hematoxylin and eosin.

Triplicate samples of WM793 and 1205Lu cells were treated with vehicle, 1 μmol/L, 3 μmol/L, 5 μmol/L, and 10 μmol/L PET-16 for 24 hours. Following standard protocols of the RPPA Core Facility at MD Anderson Cancer Center (Houston, TX), cells were lysed on ice, and lysates were cleared by centrifugation and denatured in SDS sample buffer, and then submitted for analysis as described (18, 19). The pair-wise group comparisons were done using two-sample t test, and the Benjamini–Hochberg method was used for correction for multiple testing. Proteins that for a PET-16 concentration passed FDR < 5% in both cell lines, and fold change > 1.2 threshold in at least one cell line were considered significantly differentially expressed. Data were visualized as expression heatmaps using Microsoft Excel.

For wound-healing assays, melanoma cells were seeded in 24-well plates and grown to 90% to 95% confluency. Subsequently, a scratch was placed in the middle of the well with a sterile 200 μL pipette tip. After washing once with serum-free media, the respective treatments were performed. Media with DMSO were added to control wells, and media with different drugs were added to respective wells before starting the time-lapse imaging. The concentrations of drug used were shown not to induce cell death using live/dead staining (Calcein/PI; Sigma Aldrich). Images were captured every 15 minutes for 72 hours. Each reading was performed in triplicate. For Transwell migration assays, 6-well Transwell assay dishes (3414; Corning Life Sciences) were used to examine the migration of melanoma cells. 1205Lu cells were pretreated with PET-16 for 24 hours, and then seeded onto the filters in serum-free media at 20 × 10⁴ cells/well. Media containing 20% FCS were placed in the lower well to act as a chemoattractant. Following a 24-hour incubation period at 37°C, the inserts were removed and washed three times in PBS. Cells remaining on the upper surface of the insert (nonmigrated cells) were removed gently using a cotton swab, and cells on the lower surface were stained with the Differential Quick Stain Kit (24606-250; Polysciences, Inc.). After staining, inserts were rinsed three times by dipping into PBS and left to air-dry. Colored cells that had migrated through the pores of the polycarbonate membrane were visualized using a Nikon Eclipse E600 microscope.

All animal experiments were approved by the Institutional Animal Care and Use Committee and were performed in an Association for the Assessment and Accreditation of Laboratory Animal Care–accredited facility. B16-F10 cells were pretreated with DMSO, 3 μmol/L PES-CL, or 3 μmol/L PET-16 for 24 hours and were injected (2.5 × 10⁵ cells/100 μL PBS) into the tail vein of C57BL/6 mice (The Jackson Laboratory). At day 7, weekly intraperitoneal injections with DMSO vehicle, 20 mg/kg PES-CL, or 5mg/kg PET-16 were initiated in randomly assigned groups of 10 mice for each treatment arm (n = 10). After 3 weeks, the mice were euthanized, the lungs were harvested, photographed, and embedded in paraffin for further immunohistochemical analysis. For the allograft model, 1 × 10⁶ of pooled, transfected Yumm1.7 mouse melanoma cells engineered to express high level of HSP70 versus a parental cell line were subcutaneously injected into the right flank of C57Bl/6 mice. For the next 20 days, tumor burden was monitored twice a week by volume calculation using caliper measurements. For the xenograft model, 5 × 10⁵ of 1205Lu human melanoma cells were subcutaneously injected into the right flank of female NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice at a Matrigel/culture medium ratio of 1:1. When tumors reached an average tumor volume of approximately 120 to 200 mm³, animals were randomized into treatment groups. Mice were treated with (i) PET-16 (5 mg/kg per i.p. injection twice a week, n = 10), (ii) vemurafenib/PLX4720 (200 mg/kg of PLX-containing chow, n = 10), (iii) vemurafenib/PLX4720 (200 mg/kg of PLX-containing chow) plus 5 mg/kg PET-16 (n = 10), and (iv) with PET-16 (10 mg/kg per i.p. injection twice a week, n = 5). Buffered saline (n = 10) and normal chow served as vehicle controls. Tumor growth was measured every 3 to 4 days using calipers; tumor volumes were calculated according to the formula V = (L x W x W)/2 (mm³), where V is tumor volume, L is tumor length, and W is tumor width. At day 27, the study was completed, and the tumors were harvested and embedded in paraffin for further immunohistochemical analysis.

For in vitro studies, the Student t test was performed using at least three independent experiments. For animal experiments, tumor weight was compared using t test between two groups. The effect of treatment on the change of tumor volume was examined using mixed model analysis. For TMA scores from human tissues, the Wilcoxon rank-sum test was used to compare TMA scores between melanoma and nevi. Cuzik's trend test was used to examine the trend of TMA scores from the tissues without melanoma to tissues with different stage of melanoma. Paired t test was used to compare TMA scores between before and after therapy. A P value < 0.05 was considered significant.

There are some reports that show that HSP70 (HSPA1A/B) is overexpressed in melanoma, and may be associated with drug-resistant melanoma (20–22). However to date, no studies have performed a comprehensive staining for the major, heat shock inducible form of HSP70 protein in melanoma tumors versus benign nevi. Toward this end, we used an HSP70 monoclonal antibody specific for the cytosolic stress-induced form of this protein, and not cross-react with other family members, in order to stain a TMA composed of 77 nevi, 8 melanoma in situ, 50 invasive primary melanomas, and 103 metastatic melanomas. There was a statistically significant difference in HSP70 staining in melanomas compared with nevi (mean ± SD score melanoma vs. nevi P = 0.0003; Fig. 1A–C). In addition, there was a significant correlation between HSP70 expression and increasing stage of cancer, and the highest median scores for HSP70 were in metastatic melanoma (Cuzik's trend test, P < 0.0001; Fig. 1D).

We next determined whether HSP70 mRNA levels correlated with patient survival using the program PrognoScan, which analyzes expression data and survival information from The Cancer Genome Atlas database (23). This analysis revealed that in 38 melanoma samples, the increased level of HSP70 mRNA was significantly associated with poorer survival (P = 0.001; Supplementary Fig. S2). This association did not hold true for other family members (not shown). These data support the assignment of HSP70 as a potential marker of melanoma aggressiveness, but they do not reveal whether HSP70 overexpression is a driver, or a consequence, of melanoma progression. To address this issue, we generated matching melanoma cell lines that express low and high levels of HSP70, and compared their tumorigenic properties. Specifically, we obtained the Yumm1.7 murine melanoma cell line generated from the BRAF-V600E/PTEN conditional knockout mouse (24). This cell line expresses modest levels of HSP70, and it was used to create pooled subclones that express high levels of HA-tagged HSP70 (Yumm1.7-HSP70; Fig. 1E, inset). Parental Yumm1.7 and HSP70-overexpressing cell lines were injected into C57Bl/6 mice as xenografts, and tumor volume was assessed over time. Notably, there was a marked increase in melanoma growth rate, tumor volume, and tumor weight in two independent lines that overexpress HSP70 compared with vector-alone controls (Fig. 1E and F). We also noted increased Ki67 staining, indicative of proliferating cells, as well as decreased staining for cleaved caspase-3, a marker of apoptotic cells, in tumors that overexpress HSP70, relative to parental controls (Fig. 1G). These data support HSP70 as a significant driver of melanoma growth and progression.

To determine the underlying mechanism whereby increased HSP70 is associated with melanoma tumorigenesis, we sought to identify melanoma-relevant ‘client’ proteins for HSP70. In this case, we define a “client” protein as one whose solubility/folding requires HSP70, and that interacts with HSP70. To identify these, we first used the technique of reverse RPPA (19, 25) in order to identify proteins whose signal was decreased following 24-hour treatment with different concentrations of PET-16. This analysis revealed 34 proteins that showed significantly decreased expression in two different cell lines at several different doses of PET-16 after 24 hours, representing candidate HSP70 client proteins (Supplementary Fig. S3). Expressions for six of these proteins that had a FDR <5% in both cell lines and a fold change > 1.5 are depicted (Fig. 2A). Of these six, we selected two for further analysis due to their significance to melanoma and metastasis: these were phospho-FAK (p-FAK/PTK2) and BRAF. Western analysis to confirm RPPA results revealed a marked decrease in the level of p-FAK (Y397) and BRAF following 24 hours of PET-16 treatment in three different melanoma lines with mutant (1205Lu and WM793) and wild-type (WM852) BRAF (26; Fig. 2B). To confirm this finding, we performed immunofluorescence for p-FAK in PET-16–treated cells. This analysis revealed a dramatic loss of p-FAK staining in WM793 and WM852 cell lines after PET-16 treatment (Fig. 2C); notably, this loss of p-FAK staining was evident after as little as 2-hour exposure to PET-16 (A. Budina-Kolomets, unpublished data). Immunoprecipitation–Western blot analysis confirmed that both p-FAK and BRAF were detectable in a complex with HSP70 (Fig. 2D and E; Supplementary Fig. S4A). These analyses confirm p-FAK and BRAF as HSP70 client proteins.

Given the role of FAK (particularly the autophosphorylated, activated p-Y397 form) in tumor-invasive phenotypes, we next sought to assess the impact of PET-16 on migration and invasion. We first performed conventional “scratch-wound” assays wherein the ability of tumor cells to migrate and fill in an area is assessed by time-lapse microscopy. For these studies, we were careful to use concentrations, and timepoints of PET-16 treatment were not cytotoxic, and that did not inhibit cell proliferation (Supplementary Fig. S4B and S4C). These scratch-wound assays revealed that after PET-16 treatment, the ability of 1205Lu cells to migrate and fill in a wound was markedly impaired (Fig. 3A). By 48 hours, the DMSO control sample was completely filled in, whereas the PET-16 sample was less than 50% completed (Fig. 3B). Time-lapse microscopy confirmed that PET-16–treated cells continued to proliferate, while ceasing to migrate (see Supplementary Movie). To extend this finding, we performed Transwell migration assays. Again, at doses that are noncytotoxic or cytostatic, we noted significantly impaired migration of 1205Lu cells following treatment with PET-16 (Fig. 3C).

We next assessed the impact of low doses of PET-16 on the ability of metastatic melanoma to invade in three dimensional (3D) skin reconstructions. The skin reconstructions recapitulate human skin by including keratinocytes, fibroblasts, and melanoma cells, embedded in a collagen matrix (17). Highly invasive melanoma tumors then invade into the fibroblast/collagen matrix, reminiscent of the actual scenario in the skin. Skin reconstructs made with the 1205Lu melanoma and treated with DMSO showed extensive ability to invade into the fibroblast layer (Fig. 3D, left). In contrast, 1205Lu cells treated with 0.5 μmol/L PET-16 failed to invade (right). The depth of invasiveness in multiple skin reconstructs was measured, and the difference in invasiveness between DMSO and PET-16–treated cells was highly significant (0.08 mm vs. 0.01 mm; P < 0.05; Fig. 3E). Similar findings were made in the primary melanoma line FS5 (Supplementary Fig. S4D). Boyden chamber assays were then used to measure the ability of melanoma cells to migrate and invade through a collagen matrix. These assays confirmed that melanoma invasiveness is markedly impeded by PET-16 (Fig. 3F and G). Time course analyses indicated that invasion of 1205Lu cells was impaired after as little as 2 hours of incubation with PET-16 (Supplementary Fig. S4E and S4F). We next sought to determine whether the inhibition of invasiveness by PET-16 was due to the specific inhibition of p-FAK. To address this, we silenced FAK in 1205Lu cells using either of two different short hairpins, and assessed the ability of PET-16 to inhibit invasion. As expected, silencing of FAK with either of two short hairpins (Supplementary Fig. S4G) markedly diminished the ability of cells to invade in Boyden chamber assays (Fig. 4A and B). Notably, however, silencing FAK diminished the ability of PET-16 to further inhibit invasion, and the percent inhibition of invasion by PET-16 was markedly greater in shControl cells, compared with shFAK (P = 0.005; Fig. 4B and C). These data point to FAK as a significant target for the ability of PET-16 to inhibit melanoma invasion and migration.

We next sought to assess the ability of HSP70 inhibitors to inhibit melanoma metastasis. To do this, we utilized the B16-F10 melanoma cell line, which is commonly used in metastasis assays because it is pigmented and easy to track as lung metastases after tail vein injection. We first established the IC₅₀ for PET-16 and PES-Cl in the B16-F10 cell line. We found that this cell line shows an IC₅₀ of 15.5 μmol/L and 17.5 μmol/L for PES-Cl and PET-16, respectively (Supplementary Fig. S5A and S5B). We pretreated B16-F10 cells for 24 hours with dilution vehicle alone, or 3 μmol/L of either PES-Cl or PET-16; this low dose caused no loss of cell viability of B16-F10 cells, even when cells were treated every other day for 10 days (Supplementary Fig. S5C). Cells were treated for 24 hours and then injected into the tail veins of C57Bl/6 mice; after 1 week, mice were treated once per week with intraperitoneal doses of dilution vehicle, PES-Cl (20 mg/kg) or PET-16 (5 mg/kg; Fig. 5A). After three treatments, we analyzed the lungs of these mice for the number and size of metastases. Treatment with DMSO resulted in lungs that were full of melanoma tumors after 3 weeks (Fig. 5B). However, treatment with either PES-Cl or to a greater extent PET-16 resulted in a profound reduction in the number and volume of metastases in mice. This was evident visually by melanin pigment (Fig. 5B), and by hematoxylin and eosin staining of sectioned lungs (Fig. 5B, bottom plots). Quantification of metastases in PES-Cl and PET-16–treated mice revealed a marked decrease in metastases in these mice, compared with control (Fig. 5C). Immunofluorescent staining of the melanoma tumors in treated mice revealed that phospho-FAK staining was markedly reduced in the tumors of treated mice, confirming that the drugs reached these tumors (Fig. 5D). We confirmed that the autophagy scaffold protein p62^SQSTM1, which accumulates and aggregates following PET-16 treatment, showed accumulation in the tumors of treated mice, especially in mice treated with PET-16 (Supplementary Fig. S5D and S5E).

In addition to p-FAK, our data indicated that BRAF is a client of HSP70. To explore this further, we obtained five matched pairs of melanoma biopsy samples from patients, prior to treatment with the BRAF inhibitor vemurafenib (before therapy), and after resistance and relapse from therapy (after therapy). These matched samples were stained with HSP70 antibody and scored in a blinded manner. Notably, in all five matched samples, the level of staining of HSP70 (red stain) increased significantly following the emergence of resistance (Fig. 6A), in a manner that was statistically significant (P < 0.05; Fig. 6B). To extend this finding, we next assessed the IC₅₀ for the BRAF inhibitor Vemurafenib (PLX4032) in the presence and absence of PET-16. At a constant molar ratio of 10:1 (PET-16:PLX4032), the IC₅₀ for PLX4032 decreased 4-fold and 8-fold, respectively, in 1205Lu and WM35 cells, suggesting that these two drugs cooperate (Fig. 6C and D). We next assessed synergy using the CI analysis in five different melanoma cell lines with mutant BRAF, as per ref.13. Notably, there was evidence for synergy of PET-16 and PLX4032 in all five cell lines, and there was strong synergy (cI<0.5) in two lines, WM35 and WM164 (Fig. 6E). Synergy was evident at several different ratios of each drug (Supplementary Table S1), and was consistent using an independent assay for cell viability (Trypan Blue; Supplementary Fig. S7C). Recent studies indicate that other autophagy inhibitors can synergize with BRAF inhibitors, mediated by the induction of ER stress (27); however, we found limited evidence that PET-16 causes significant ER stress in melanoma (Supplementary Fig. S6).

It next became logical to test the efficacy of combination therapy using these two compounds. Toward this end, we analyzed subcutaneous tumors from the 1205Lu cell line in immunocompromised mice; this cell line was chosen because it shows only moderate sensitivity to BRAF inhibition (28). Forty mice were injected subcutaneously with 1205Lu cells, and tumors were allowed to engraft for 8 days, after which mice were randomly separated into four groups. Each group of 10 mice was treated twice weekly with intraperitoneal doses of dilution vehicle alone, PET-16 (5 mg/kg), PLX4720 in chow (PLX4032 is the clinical analogue of PLX4720), or the combination of PET-16 (5 mg/kg) plus PLX4720 in chow (Fig. 6F). There were linear increases in tumor volume in DMSO-treated mice, whereas PLX4720 led to an initial decrease in tumor volume, followed by waning antitumor efficacy with time (blue line; Fig. 6G). Addition of low-dose (5 mg/kg) PET-16 greatly enhanced the efficacy of PLX4720 [P < 0.0001 comparing PLX4720 vs. PET-16 (5 mg/kg) plus PLX4720; Fig. 6G, red line]. Immunofluorescence and IHC analyses confirmed markedly decreased p-FAK and BRAF staining in PET-16–treated tumors (Supplementary Fig. S7A and S7B). We did not observe any weight loss or signs of toxicity for any of these treatment arms (A. Budina-Kolomets; unpublished data). As a final confirmation of the usefulness of PET-16 for melanoma treatment, we analyzed the efficacy of this compound in tumors with acquired resistance to BRAF inhibitors. We found no difference in the cytotoxicity of PET-16 in tumor cells resistant to PLX4032, compared with parental tumor lines (Fig. 6H). These data support the use of HSP70 inhibitors for melanoma therapy.

It is well established that one of the best markers for the detection of tumor versus normal cells is the stress-induced heat shock protein HSP70 (5). In this study, we chose to focus on HSP70 and melanoma, due to the enormous propensity in this tumor type for aneuploidy and mutation, both of which lead to considerable proteotoxic stress (29). Currently, the role for HSP70 in melanoma is not clear. Two studies showed that HSP70 overexpression occurs in melanoma and is associated with BRAF mutation (20) and drug resistance (21). In contrast, another study found high levels of HSP70 associated with improved prognosis, especially in patients with tumors greater than 1.5 mm in size (22). In this study, we provide evidence that HSP70 supports melanoma progression. We show that HSP70 overexpression causes increased tumor proliferation and decreased apoptosis, and that this protein shows the highest expression in metastatic melanoma. The reasons for the discrepancies regarding the role of HSP70 in melanoma are unclear, but may reflect the fact that this protein can localize to different compartments in the cell. When HSP70 localizes to cytosolic, nuclear, and lysosomal compartments, it promotes cancer progression; alternatively, this protein can be extracellular and promote antitumor immunity (30).

One consistent finding among multiple groups is that silencing or inhibiting HSP70 in melanoma cell lines and other tumors leads to senescence and apoptosis (7, 31–34). Both a peptide aptamer that binds and inhibits this chaperone (35) and a small-molecule inhibitor of HSP70 (36) have shown anticancer efficacy in an allograft model of mouse melanoma. HSP70 is an obligate co-chaperone for HSP90, and targeting HSP90 has shown significant promise in the treatment of melanoma (37, 38). Therefore, HSP70 is a compelling new therapeutic target for this tumor. The group of Garrido showed previously that a peptide–aptamer inhibitor of HSP70 was efficacious against the murine melanoma allograft model B16-F10, grown subcutaneously. Interestingly, in the tumors of treated mice, this group noted that regression of the tumors was accompanied by significant immune cell infiltration (35). To explore this same possibility, we analyzed PET-16–treated melanoma tumors and found that these treated tumors also showed marked accumulation of leukocytes infiltrating the tumors, compared with the vehicle controls (Supplementary Fig. S5F, arrows). These data indicate that HSP70 inhibitors may affect immune cell recruitment, and they support the testing in preclinical models of combinations of HSP70 inhibitors with immune checkpoint antibodies. We are currently pursuing this avenue.

In this work, we show that pFAK is a novel HSP70 client protein and that inhibition of p-FAK by PET-16 explains part of the mechanism for reduced invasiveness following HSP70 inhibition. We should note however that two other proteins implicated in melanoma metastasis were identified in our RPPA assay as downregulated following PET-16 treatment: ERBB3 and FOXM1 (Supplementary Fig. S3; refs. 39, 40). Therefore, it remains formally possibly that the inactivation of these proteins also contributes to the ability of PET-16 to inhibit melanoma metastasis. We also find that mutant B-RAF is an HSP70 client protein, and that treatment of tumor cell lines in vitro and in vivo leads to markedly decreased expression of mutant B-RAF. Further, we find that PET-16 synergizes with PLX4032 in five different melanoma cell lines, and that PET-16 improves the durability of response to BRAF inhibition. There are two possible explanations for this synergy. The first is that PET-16 and BRAF inhibit BRAF in parallel. The second is that common mechanisms for resistance to BRAF inhibitors include overexpression of AKT, mTOR, C-RAF, and RB (41); these were all identified as HSP70 client proteins in our RPPA analysis (Supplementary Fig. S3). The emergence of resistance to BRAF inhibitors is a significant clinical problem, so the identification of compounds like PET-16 is an important advance.

G.B. Mills reports receiving commercial research grant from Adelson Medical Research Foundation, AstraZeneca, Critical Outcome Technologies, Komen Research Foundation, Nanostring; has received honoraria from the speakers bureau of AstraZeneca, ISIS Pharmaceuticals, Nuevolution, and Symphogren; has ownership interest (including patents) in Catena Pharmaceuticals, Myriad Genetics, PTV Ventures, and Spindletop Ventures; and is a consultant/advisory board member for Adventist Health, AstraZeneca, Blend, Catena Pharmaceuticals, Critical Outcome Technologies, HanAl Bio Korea, ImmunoMET, Millenium Pharmaceuticals, Nuevolution, Precision Medicine, Provista Diagnostics, Signalchem Lifesciences, and Symphogen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Budina-Kolomets, M.R. Webster, J. I-Ju Leu, A. Guerrini, D.L. George, M.E. Murphy

Development of methodology: A. Budina-Kolomets, M.R. Webster, J. I-Ju Leu, A. Guerrini, R.K. Amaravadi, X. Xu, M.E. Murphy

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Budina-Kolomets, M.R. Webster, J. I-Ju Leu, M. Jennis, C. Krepler, W. Xu, G. Karakousis, L. Schuchter, H. Wu, G.B. Mills, X. Xu, A.T. Weeraratna

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Budina-Kolomets, M.R. Webster, J. I-Ju Leu, C. Krepler, A.V. Kossenkov, R.K. Amaravadi, X. Yin, Q. Liu, G.B. Mills, X. Xu, D.L. George, A.T. Weeraratna, M.E. Murphy

Writing, review, and/or revision of the manuscript: A. Budina-Kolomets, M.R. Webster, J. I-Ju Leu, M. Jennis, A.V. Kossenkov, W. Xu, G. Karakousis, L. Schuchter, R.K. Amaravadi, H. Wu, G.B. Mills, X. Xu, D.L. George, A.T. Weeraratna, M.E. Murphy

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Budina-Kolomets, M.R. Webster, X. Xu, M.E. Murphy

Study supervision: W. Xu, M.E. Murphy

Other (performed RPPA analysis and provided the RPPA data set): Y. Lu

The authors thank Gao Zhang and Meenhard Herlyn for help with the RPPA analysis, and members of the Murphy laboratory (Pat Kung and Subhasree Basu) for help and advice. This work utilized the following Core facilities at The Wistar Institute: the Laboratory Animal Facility, the Imaging Facility, the Histochemistry Facility, and the Bio-informatics Facility.

This work was supported by NIH R01 139319 (M.E. Murphy), P01 114046-07 (M.E. Murphy, D.L. George, and A.T. Weeraratna), and R01 CA174746-01 (A.T. Weeraratna), as well as T32 CA009171-36 (M.R. Webster). The IHC and IF studies were supported by the Molecular Pathology & Imaging Core, Perelman School of Medicine, University of Pennsylvania (NIH P30 DK050306). The RPPA analysis was funded by the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation to Meenhard Herlyn (The Wistar Institute). Support for Core Facilities utilized in this study was provided by Cancer Center Support Grant (CCSG) CA010815 to The Wistar Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.