HDAC8 Regulates a Stress Response Pathway in Melanoma to Mediate Escape from BRAF Inhibitor Therapy

Use of BRAF inhibitors and BRAF–MEK inhibitor combinations is associated with impressive therapeutic responses and increased overall survival in patients whose melanomas harbor position 600 mutations in BRAF (1). Despite this, most patients ultimately fail therapy, and cures remain rare (1, 2). Although much is now known about the genetic mediators of acquired BRAF and BRAF–MEK inhibitor resistance, there is still an urgent need to better understand the mechanisms underlying treatment failure, particularly at the earliest stages, so that new therapeutic strategies and drug combinations can be developed (2–5). The process of early adaptation to therapy remains poorly defined but appears to involve the adoption of a slow-growing “persister” state that is marked by dedifferentiation, phenotypic plasticity and some recovery of MAPK signaling (6). This early rebound in MAPK signaling is frequently mediated through increased receptor tyrosine kinase (RTK) signaling, with a number of studies now implicating roles for IGF1R, EGFR, Axl, c-MET, PDGFR, and EphA2 (7–10).

In our previous studies, we used comprehensive mass spectrometry–based phosphoproteomics to identify a ligand-independent EphA2-driven signaling network as a driver of an aggressive, epithelial–mesenchymal transition (EMT)-like phenotype in melanoma cells with acquired BRAF inhibitor resistance (11). This S897-EphA2–driven signaling network was dependent upon continuous MAPK pathway inhibition and was reversed following drug withdrawal for >3 weeks (11). The plasticity of this drug-induced phenotype suggested these changes could be epigenetically mediated (11). In the present study, we asked whether a common transcriptional state that emerged when melanoma cells were subjected to stress allowed melanoma cells to survive diverse insults. Our work identified a novel role for HDAC8 as a mediator of phenotype switching and the therapeutic adaptation of melanoma cells to BRAF inhibition. Unexpectedly, we found that HDAC8 regulates BRAF inhibitor sensitivity and acquired drug resistance through direct effects upon c-Jun acetylation, leading to transcriptional rewiring and increased RTK and MAPK signaling. Together, these results point to a new role for the histone deacetylases in regulating the cell signaling networks at the protein acetylation level that mediates therapeutic escape.

The 1205Lu, WM164, and SKMEL-28 cell lines were a generous gift from Dr. Meenhard Herlyn (The Wistar Institute, Philadelphia, PA). The dual BRAF and MEK inhibitor–resistant (RR) lines 1205LuRR, SKMEL28RR, and WM164RR were established as previously described (12). Panobinostat, PCI-34051, and erlotinib were from Selleckchem. Hypoxia was achieved via an oxygen control glove box (Coy Labs) for 24 hours in conditions containing 94% N₂, 1% O₂, and 5% CO₂. All cells were tested for Mycoplasma contamination every 3 months using the Plasmotest-Mycoplasma Detection Test (Invivogen). Last test date was March 18, 2019. Each cell line was authenticated using the Human STR human cell line authentication service (ATCC), and frozen stocks of cells were discarded after 10 passages.

Lysates were acquired and processed for Western blot and immunoprecipitation as previously described (11). The anti-HDAC3 and anti-HDAC8 antibodies were described in refs. 13 and 14. The antibodies against HDAC1 (2062), HDAC2 (2540), BIM (2933), Mcl-1 (4572), phospho-ERK (9101), ERK (9102), phospho-CRAF (56A6, 9427), CRAF (D4B3J, 53745), phospho-EphA2 (D9A1, 6347), EphA2 (D4A2, 6997), phospho-AKT (D9E, 4060), AKT (9272), phospho-c-Jun (54B3, 2361), c-Jun (60A8, 9165), and acetyl (9441) were purchased from Cell Signaling Technology. Anti-HDAC6 (H-300, sc-11420) was purchased from Santa Cruz Biotechnologies. Anti-HDAC11 (ab47036) was purchased from Abcam. Anti-Vinculin (G8796) and anti-GAPDH (V9131) were purchased from Millipore Sigma. Ac-SMC3 was a kind gift from Forma Therapeutics. Phospho-RTKs were measured with the Human Phospho-Receptor Tyrosine Kinase Array Kit (R&D Biosystems). Activated and total Ras were measured with the Active Ras Pull-Down and Detection Kit (ThermoFisher). For each experiment, all antibodies were probed on the same blot. In cases where bands were similar, the blots were washed with Restore Western Blot Stripping Buffer for 10 minutes before a new antibody was used.

Cells were treated with drugs (72 hours), harvested, and incubated with Annexin V APC (BD Biosciences). Fluorescence was read on a FACSCalibur (BD Biosciences) and analyzed using Flowjo software. To measure cell death following induction of stress, 300 cells were counted for cell death by trypan exclusion using a 0.4% trypan blue solution (Millipore Sigma).

Cells were treated with drug for 28 days before being stained with a 0.5% Crystal Violet solution. Colonies were quantified using ImageJ software.

Samples from melanoma patients pre- and post-BRAF and BRAF–MEK inhibitor therapy were collected from the University Hospital Essen under a written-informed consent protocol. Formalin-fixed, paraffin-embedded (FFPE) slides were stained for HDAC8 expression using the Ventana Discovery XT automated system and an anti-HDAC8 antibody (Novus Biologicals) at a 1:100 concentration with 60-minute incubation. Staining was detected using the Ventana ChromoMap Red Kit, and slides were counterstained with hematoxylin. For mouse IHC, FFPE slides were stained for phospho-c-Jun (Abcam, ab32385) for 1 hour at a 1:100 concentration, and slides were uploaded into an Aperio AT2 scanner (Leica Biosystems) and visualized using Aperio Imagescape 12.3.3 (Leica Biosystems).

Cells were lysed in an urea lysis buffer (20 mmol/L HEPES, pH 8.0, 9 mol/L urea, 1 mmol/L sodium orthovanadate, 2.5 mmol/L sodium pyrophosphate, and 1 mmol/L β-glycerophosphate), and protein concentration of the lysate was measured by Bradford assay. Extracted proteins (10 mg) were digested by trypsin and enriched for phospho-tyrosine and phospho-serine/threonine as previously described (11). Extracted proteins from each condition [empty vector (EV) or HDAC8] were trypsin digested, and 2 equal aliquots of tryptic peptides (100 μg) were labeled by distinct Tandem Mass Tags (TMT six-plex reagents, ThermoFisher), combined, and subjected to offline high pH Reverse Phase fractionation (15). Each of the fractions was enriched for phosphopeptides using a Phos-SELECT Iron Affinity Gel (Millipore Sigma; ref. 16). Mass spectrometry data were acquired on a QExactive mass spectrometer coupled to a U3000 RSLCnano system (ThermoFisher) as described previously (16). Two technical replicates were performed for the immune-enriched phosphotyrosine samples as well as each of the immobilized metal affinity chromatography (IMAC)-enriched fractions. Label-free quantitation was performed for phosphotyrosine samples, whereas MS2-reporter ion quantitation was performed for IMAC-enriched samples using MaxQuant (1.2.2.5; ref. 17). Data are available in PRIDE (PXD012813 and PXD012812).

Isolated RNA was cleaned using an RNAeasy minicleanup kit (Qiagen) and screened for quality on an Agilent BioAnalyzer. The samples were then processed for RNA sequencing (RNA-seq) using the NuGen Ovation Human FFPE RNA-Seq Multiplex System. The libraries were then sequenced on the Illumina NextSeq 500 sequencer with a 2 × 75-base paired-end run in order to generate 40 to 50 million read pairs per sample. Data are available in GEO (GSE127564).

Combat was used to normalize phosphotyrosine profiles before further analyses (18). Log₂ transformation was applied to all three datasets (RNA-seq, and both phosphorylation experiments). Moderated t statistics were used to compare the RNA expression between baseline (EV) and HDAC8 overexpression (HDAC8) samples in RNA-seq data for each of 18,542 genes using the limma package in R (19). In phosphotyrosine residue data, 172 phosphopeptides were used for assessing differential expression in HDAC8 versus EV samples. In serine/threonine phosphopeptide data, 1,976 phosphopeptides were used in assessing differential expression in HDAC8 versus EV samples. Volcano plots with significant phosphopeptides denoted by fold change > 2 and P value < 0.05 in the contrast between EV and HDAC8 samples were also used for visualization.

Normalized phosphoproteomic data were combined and analyzed using GeneGO software (Metacore, Thomson Reuters). Significant interactions between genes were determined with a cutoff value of P < 0.05. Normalized pY and pS/T proteomic data were uploaded and analyzed by STRING. The most stringent interaction threshold of 0.9 was used to find the most significant interactions upregulated in HDAC8-expressing cells. Significant interactions exported from GeneGO were organized into a global signaling hub using Cytoscape software. RNA-seq data were analyzed by Gene Signature Enrich Analysis (GSEA). The data were analyzed for significant transcription factors using an FDR cutoff of 0.05.

Cells were placed in OPTI-MEM media in the presence of the plasmid or siRNA and lipofectamine 2000. Mcl-1 (ON Target SMART pool) siRNA and nontargeting control siRNA were purchased from ThermoFisher. The EV plasmids were purchased from Origene Technologies Inc. For infection of Millipore Sigma shRNA viral particles, infection was performed per the manufacturer's protocol. After 24 hours, the media were removed and replaced with media containing 1 μg/mL puromycin (Millipore Sigma). shRNA against HDAC8 (SHCLNV-NM_018486, TRCN0000004851) was purchased from Millipore Sigma.

EGFR mRNA expression was measured by quantitative RT-PCR. EGFR and GAPDH primers were purchased from Applied Biosystems (AB, Thermo). cDNA was made from isolated RNA with the High Capacity cDNA Reverse Transcriptase Kit (AB), and 100 μg of cDNA was run on a 7900HT Fast Real-Time PCR System for 40 cycles using Taqman master mix (AB). Samples were normalized to control.

To assess ATF2 and c-JUN transcriptional activity, we implemented a dual secreted luciferase assay as previously described (20). At 48 hours after transfection, the cells were treated with specified drugs, and at the indicated times, media samples containing secreted lucifase were harvested and measured for luciferase activity using the Pierce Gaussia Luciferase Glow Assay Kit per the manufacturer's instructions (ThermoFisher).

Binding of c-Jun and c-Jun mutant cells to the consensus JUN DNA sequence was performed using the Mouse/Human/Rat JUN/c-Jun DNA Binding ELISA Kit (LSBio) per the manufacturer's instructions. Samples were read on a plate reader at 450 nm.

The following primers were ordered from Integrated DNA Technologies: 268 mutant: 5′ gcatcgctgc ctccagatgc cgaaaaagga agctggagag aatcg 3′ and 5′ cgatt ctctccagct tcctttttcg gcatctggag gcagcgatgc 3′; 271 mutant: 5′ gcatcgctgc ctccaagtgc cgaagaagga agctggagag aatcg 3′ and 5′ cgatt ctctccagct tccttcttcg gcacttggag gcagcgatgc 3′; and 273 mutant: 5′ gcatcgctgc ctccaagtgc cgaaaaagga gactggagag aatcg 3′ and 5′ cgatt ctctccagtc tcctttttcg gcacttggag gcagcgatgc 3′. Mutant DNA constructs were made by a site-directed mutagenesis kit (ThermoFisher) against a WT c-Jun plasmid (Origene Technologies) per the manufacturer's instructions. Mutant constructs were sequenced (Genewiz) using plasmid DNA and c-Jun primer sequence cgtttggagtcgttgaagttg (IDT). DNA was stably transfected into cells using lipofectamine 2000, and clones were selected for further study. After selection, endogenuous levels of c-Jun were knocked down using a 3′ shRNA for JUN (SHCLNV-NM_002228, TRCN0000039588, Millipore Sigma).

Cells were injected into the hind flank of NOD.CB17-Prkdc^scid/J mice (Taconic) in a solution containing 50% L-15 media (ThermoFisher) with 1 mmol/L HEPES (Millipore Sigma) and 50% Matrigel (BD Biosciences). Ten tumors were used for each group in each experiment. All studies were approved by the University of South Florida's Institutional Animal Care and Use Committee (#IS00004987). PLX 4720 was given using formulated chow (Research Diets), whereas panobinostat and PCI-34051 were given by i.p. injections for the duration of the experiment. Weight and tumor size were measured with calipers and were monitored 3 times weekly.

For all experiments, significance was determined between groups using a one-way ANOVA followed by a post hoc t test. For all in vitro experiments, 3 independent experiments with an n of 3 were used for an overall n of 9 with a representative experiment shown. For in vivo studies, an n of 10 was used for each group.

In previous studies, we identified an S897-EphA2–driven signaling interactome that emerged under continuous BRAF therapy, which was readily reversible following drug withdrawal (11). We reasoned that this network, and therefore BRAF inhibitor resistance, may be in part epigenetically regulated. To explore this mechanism, we treated BRAF–MEK inhibitor-resistant melanoma cell lines (designated RR) with the broad-spectrum HDAC inhibitor panobinostat and found that it decreased both S897-EphA2 and pAKT signaling and restored vemurafenib sensitivity in apoptosis assays (Fig. 1A and B). We next asked whether acquired BRAF inhibitor resistance was associated with an increased expression of specific HDAC genes or proteins by microarray and Western blot analysis, respectively. It was determined that Class I HDACs (HDAC1, HDAC2, HDAC3, and HDAC8), Class IIb HDACs (HDAC6), and Class IV HDACs (HDAC11) were consistently expressed (Supplementary Fig. S1A). Although many HDACs showed alteration following the acquisition of BRAF (designated R) and BRAF–MEK inhibitor resistance, HDAC8 expression was consistently increased (>2-fold) in the 5 of 5 drug-resistant melanoma cell lines (Fig. 1C; Supplementary Fig. S1B). Increased expression of HDAC6 (>2-fold) was also noted in 4 of 5 of the cell lines (M229R, SKMEL-28RR, 1205LuRR, and WM164RR; Fig. 1C; Supplementary Fig. S1B). Expression of HDAC8 and c-JUN was also noted in melanoma cells with intrinsic BRAF inhibitor resistance, whereas those with initial BRAF inhibitor sensitivity expressed little c-JUN and an HDAC8 doublet (Supplementary Fig. S2; ref. 21). A role for HDAC8 in the restoration of drug sensitivity was suggested by the ability of an HDAC8 inhibitor (PCI-34051), but not an HDAC6 inhibitor (tubastatin), to restore the sensitivity of BRAF inhibitor–resistant melanoma cell lines to vemurafenib (Fig. 1D; Supplementary Fig. S3). As increased expression is not always indicative of increased enzymatic activity, we also probed for the validated HDAC8 target, acetylated-SMC3 (22), and noted a decrease in acetylation of SMC3 in the resistant cell lines (Fig. 1E).

To explore whether increased HDAC8 activity was a common response of melanoma cells to stress, we next treated 1205Lu melanoma cells with either UV radiation (254 nm: 3.75 J/m²) or hypoxia (1% O₂ for 24 hours). Exposure to both of these stresses induced HDAC8 expression, with overexpression of HDAC8 leading to reduced cell death following UV irradiation or hypoxia (Fig. 1F and G). The clinical relevance of these findings was investigated through IHC staining of a cohort of matched pre- and post-BRAF inhibitor–treated melanoma patient specimens (Supplementary Table S1). It was found that HDAC8 was either highly expressed at baseline (6/8) and did not change on therapy or showed increased expression posttherapy (2/8 cases; Fig. 1H). Collectively, these data demonstrated that HDAC8 was induced under multiple stress conditions, including BRAF–MEK inhibitor therapy, and that expression of HDAC8 could provide protection to melanoma cells. Continuous drug exposure was required to maintain the HDAC8-driven adapted state, with drug removal for >3 weeks leading to reduced expression of HDAC8 (Supplementary Fig. S4).

We next generated stable HDAC8-expressing clones of drug-naïve WM164 and 1205Lu melanoma cells, that had protein expression levels equivalent to that induced by continuous BRAF inhibitor therapy (Fig. 2A). The introduction of HDAC8 increased the tolerance of melanoma cells to BRAF inhibitor therapy in 4-week colony formation assays (Fig. 2B and C) and led to a significant reduction in vemurafenib-induced apoptosis (Fig. 2D), which was not associated with increased cell proliferation (Supplementary Fig. S5). These effects were also observed following administration with a combination of BRAF–MEK inhibitors (Supplementary Fig. S6A and S6B). Conversely, it was found that the silencing of HDAC8 reversed resistance to vemurafenib in colony formation assays (Fig. 2E–G) and restored apoptosis levels to those of the drug-naïve cell lines (Fig. 2H). We next determined the functional consequences of modulating HDAC8 expression in terms of apoptosis regulation. We focused upon BIM and Mcl-1, as (1) both of these proteins are regulated by mutant BRAF in melanoma cells and are (2) important regulators of the apoptotic response following BRAF inhibition (23, 24). HDAC8 introduction, followed by BRAF inhibitor treatment, was associated with a suppression of proapoptotic BIM expression (Fig. 2I) and the maintenance of Mcl-1 levels (Fig. 2I), while silencing HDAC8 increased BIM expression (Supplementary Fig. S7A–S7C). The critical role of Mcl-1 maintenance in the prosurvival effects of HDAC8 overexpression was demonstrated through the siRNA silencing of Mcl-1, which restored the sensitivity of the HDAC8-overexpressing cells to vemurafenib (Fig. 2J and K).

We reasoned that the introduction of HDAC8 increased melanoma cell survival under stress by rewiring the signaling network. To explore this, we utilized mass spectrometry–based phosphoproteomics to map the entire signaling network. The introduction of HDAC8 into drug-naïve BRAF-mutant melanoma cell lines led to significant increases in the tyrosine phosphorylation of 5 peptides and the serine/threonine phosphorylation of 113 peptides (Fig. 3A). These data demonstrated HDAC8 overexpression enriched for networks associated with the adoption of an EMT, as well as MAPK and AP-1 transcription factor signaling (Fig. 3B). These findings with HDAC8 mirrored those reported previously by our group on melanomas with acquired BRAF inhibitor resistance (11). Grouping of the proteomic data into cellular processes using STRING analysis demonstrated HDAC8 to be involved in ribosomal function, RNA binding, cell-cycle regulation, ERK signaling, and organization of the cytoskeleton (Fig. 3C). Analysis of individual phosphopeptides identified the emergence of a signaling interactome that was dependent upon MAPK1 and c-Jun (Fig. 3D). Other members of the HDAC8-driven signaling network included MAPK pathway members (p38 MAPKα and p38MAPKγ), cytoskeleton regulators (FAK, paxillin, stathmin, LIMA1, PTRF, and MARCKS), cell cycle/spindle regulators (CDK1, ASPM, and TPX2), transcriptional initiation (EIF6 and EEF1D), and PKC signaling (PRKCD).

Our phosphoproteomic studies identified MAPK1 as a major HDAC8-regulated signaling hub. We next used two isogenic cell line pairs transduced with either EV or HDAC8 to evaluate its role in MAPK signaling. HDAC8 introduction increased baseline phospho-ERK levels in both cell lines, and MAPK signaling was maintained in the presence of a BRAF inhibitor, i.e., the drug never inhibited the pathway by >50% (Fig. 4A and B). These effects were also seen following administration of a combination of BRAF–MEK inhibitors (Supplementary Fig. S6A and S6B). Conversely, shRNA knockdown of HDAC8 reduced MAPK signaling in the presence of a BRAF inhibitor (Supplementary Fig. S7A–S7C). The increased MAPK signaling we observed occurred upstream of ERK, with a more pronounced and rapid induction of phospho-CRAF signaling being noted in the HDAC8-expressing cells compared with the EV controls (Fig. 4C). Ras-GTP pulldown experiments demonstrated that HDAC8 overexpression increased the level of Ras-GTP loading, indicating the reactivation of signaling upstream of RAF (Fig. 4D).

We next turned our attention to RTKs and used RTK arrays to demonstrate that HDAC8 introduction altered the basal phosphorylation of multiple RTKs including EGFR, c-MET, and FGFR3 (Fig. 4E and F; Supplementary Fig. S8A–S8D). Among the RTKs identified, EGFR appeared critical for the increased MAPK signaling associated with HDAC8, with studies showing that erlotinib resensitized HDAC8-expressing melanoma cells to BRAF inhibitor–mediated apoptosis (Fig. 4G). Use of the c-MET inhibitor, crizotinib, or the FGFR inhibitor, BGJ398, also resensitized HDAC8-expressing melanoma cells to BRAF inhibition (Supplementary Fig. S9A and S9B). Together, these data indicate that increased HDAC8 activity contributes to stress tolerance through maintenance of survival signaling.

We next performed RNA-seq analyses on our isogenic (EV and HDAC8 introduced) cell lines (Fig. 5A) and used GSEA to identify transcriptional programs associated with HDAC8 expression. One of the top hits was an AP-1 gene signature, indicative of c-Jun transcriptional activity (Fig. 5B). Unbiased kinome array analysis showed HDCA8 introduction to be associated with increased c-JUN, p53, AKT, and HSP60 phosphorylation (Supplementary Fig. S10A and S10B). Functional studies showed HDAC8 introduction led to increased c-Jun phosphorylation following BRAF inhibitor treatment (Fig. 5C) and enhanced c-Jun transcriptional activity both immediately following and at 4 hours after BRAF inhibitor treatment (Fig. 5D). A role for increased c-Jun expression/activity in BRAF inhibitor tolerance was indicated by the observation that c-Jun silencing restored vemurafenib sensitivity to HDAC8-expressing melanoma cells (Supplementary Fig. S11A and S11B).

Previous studies have demonstrated that c-Jun is acetylated at Lys268, Lys271, and Lys273 (25). We performed immunoprecipitation studies and demonstrated that the introduction of HDAC8 led to the deacetylation of c-Jun (Fig. 5E). A structural analysis revealed that the three potential acetylation sites (268, 271, and 273) are located within the DNA-binding domain of c-Jun (Fig. 5F). A series of acetylation-deficient K→R c-Jun mutants were generated at each of the three individual lysines (K268R, K271R, and K273R; Supplementary Fig. S12), along with the silencing of the endogenous protein through a 3′-UTR–directed shRNA. Mutating lysine 273 led to a reduction of BRAF inhibitor sensitivity by both apoptosis (Fig. 5G) and colony formation assays (Supplementary Fig. S13A and S13B). Introduction of K273R c-JUN also limited the proapoptotic effects of combined HDAC8 and BRAF inhibition in apoptosis assays (Supplementary Fig. S13C). Functionally, these effects were associated with increased levels of ERK phosphorylation in addition to decreased levels of BIM expression following BRAF inhibition (Fig. 5H). Mutating lysine 273 also increased the binding of c-Jun to the consensus JUN/c-Jun DNA sequence as determined by ELISA (Fig. 5I) and significantly increased levels of EGFR mRNA as measured by qRT-PCR (Fig. 5J). These results were supported by kinome and RTK arrays that demonstrated K273R introduction to be also associated with increased EGFR phosphorylation and enhanced p53, AKT, STAT3, WNK1, and HSP60 signaling (Supplementary Figs. S14A, S14B, S15A, and S15B).

As the final step, we asked whether HDAC8 inhibition improved BRAF inhibitor responses in vivo. For the initial studies, we injected isogenic WM164 and 1205Lu melanoma cells that expressed EV, had HDAC8 expression (HDAC8), or were stably knocked down (shRNA) for HDAC8 into the flanks of NSG mice. When tumor volumes reached 25 to 40 mm³, treatment with the BRAF inhibitor PLX4720 was initiated. In these xenografts, melanoma cells expressing HDAC8 showed resistance to BRAF inhibitor treatment, whereas the melanoma with HDAC8 stably knocked down “crashed” following initiation of BRAF inhibitor treatment (Fig. 6A and B). At the completion of the experiments, the HDAC8 shRNA knockdown tumors were significantly smaller than both the HDAC8-expressing and EV control cells. Western blot studies confirmed the increased expression of HDAC8 and showed this to be associated with a suppression of BIM expression under BRAF inhibitor therapy (Fig. 6C). Increased nuclear accumulation of phospho-c-JUN was also seen in the tumors with HDAC8 expression (Fig. 6D). It was not possible to analyze the HDAC8 shRNA tumors by Western blot due to very low tumor volumes after BRAF inhibitor therapy. We then determined whether similar results could be achieved with small-molecule HDAC inhibitors. Here, we used two HDAC inhibitors (the broad-spectrum HDAC inhibitor panobinostat and the HDAC8-specific inhibitor PCI-34051) in combination with the BRAF inhibitor PLX4720. For these studies, the animals received a lead-in dose of each of the HDAC inhibitors (to mimic the effects of having the HDACs silenced prior to initiating BRAF inhibitor therapy) before continuing treatment with the combination of HDAC and BRAF inhibitors. Cotreatment with both drugs significantly reduced tumor growth compared with either agent alone and was associated with durable responses in these model systems (Fig. 6E and F).

Adaptation to therapy is a major factor that limits the long-term responses of BRAF-mutant melanoma patients to BRAF inhibitor monotherapy and BRAF–MEK inhibitor combination therapy (3, 6, 26). Despite this, relatively little is known about the early events that permit limited numbers of cells to evade the effects of BRAF inhibition. Work from our group and others has demonstrated that diverse therapeutic interventions, including targeted therapy and immune therapy, induce a dedifferentiated state that is reminiscent of an EMT (11, 27–31). Melanoma cells that have undergone this transition typically exhibit increased invasion and resistance to most therapies (11, 27–31). Previous studies from our lab showed this phenotype to be reversible upon drug withdrawal and possibly epigenetically mediated (11). Given the postulated links between stress, phenotype switching, and drug resistance in melanoma, we asked whether there was a unifying cellular program that regulated the response of melanoma cells to multiple stresses.

We began by demonstrating that the drug-adapted, EMT-like state (here marked by increased S897-EphA2 signaling) could be reversed following treatment with HDAC inhibitors such as panobinostat. HDACs are enzymes that catalyze the hydrolysis of acetyl groups from acetylated proteins. The HDACs have many targets, both nuclear and cytoplasmic, with the best characterized of these being the N-terminal tails of histones (32). Our studies revealed that HDAC8 was frequently upregulated in melanoma cells with acquired BRAF and BRAF–MEK inhibitor resistance and that introduction of exogenous HDAC8 conveyed resistance to MAPK-targeted therapies. HDAC8 is a poorly characterized Class I HDAC found in both the nucleus and cytoplasm (14, 33). As well as its nuclear activity as a histone deacetylase, HDAC8 also has a number of nonhistone substrates including p53, cortactin, and SMC3 (22, 34, 35). HDAC8 was not the only HDAC whose expression was altered upon chronic BRAF inhibitor treatment, with increased HDAC6 expression being observed in some of the resistant cultures. Despite this, inhibition of HDAC6 did not restore sensitivity to BRAF inhibition, suggesting that this HDAC played a minor role in the escape from BRAF inhibitor therapy.

To better understand how HDAC8 regulates signaling in melanoma cells, we performed phosphoproteomic analyses of isogenic melanoma cell line pairs and noted the emergence of a signaling network dependent upon MAPK1 and c-Jun following the introduction of HDAC8. These findings closely mirrored our previous proteomic studies that identified an EGFR, c-JUN signaling network being associated with acquired BRAF inhibitor resistance (11). In functional terms, introduction of HDAC8 was associated with increased baseline levels of phospho-ERK and the maintenance of MAPK signaling following BRAF inhibitor treatment. It is likely that the shallower level of ERK inhibition associated with HDAC8 introduction reduces drug efficacy. In the clinical setting, >90% ERK inhibition is required for responses in melanoma patients (36). The HDAC8-mediated adaptation occurred upstream, at the level of RTK signaling, with increases noted in Ras-GTP loading and phosphorylation of CRAF at S338. Ultimately, the increased level MAPK signaling throughput prevented the melanoma cells from being primed for cell death through a mechanism including reduced levels of BIM expression and maintenance of prosurvival Mcl-1 levels (24, 37, 38). Both BIM and Mcl-1 are known to be regulated through the MAPK pathway, with BIM in particular being rapidly targeted for degradation following its phosphorylation by MAPK at Ser69 (23). Mcl-1 exerts its antiapoptotic activity by binding to and blocking the function of BIM-EL and through inhibition of proapoptotic Bak/Bax. In melanoma, Mcl-1 conveys resistance to anoikis, and its downregulation is required for the cytotoxic activity of the HSP90 inhibitor XL888 (24, 39).

As both our proteomics and RNA-seq analyses suggested that HDAC8 expression was associated with c-Jun signaling and Jun/AP-1–driven transcription, we next asked whether HDAC8 mediated its effects via direct c-Jun regulation. c-Jun is a key transcriptional regulator of melanoma cells that has been implicated in melanoma progression, phenotype switching, and therapy resistance (40–42). The expression and activation of c-Jun is subject to complex regulation at both the transcriptional and the posttranslational levels. In BRAF- and NRAS-mutant melanoma cells, c-Jun activation occurs as a result of a complex signaling loop dependent upon ERK-mediated GSK3 and CREB phosphorylation (40). Other recent studies have tied the activation of c-Jun to decreased expression of the ERK target gene SPRY-4, following BRAF inhibition (43). Work in other systems has suggested that Jun transcriptional activity can be regulated through acetylation at Lys268 (25). Through immunoprecipitation and site-directed mutagenesis studies, we here demonstrated that HDAC8 was required for deacetylation of c-Jun at Lys273 and that the introduction of K273R mutant of c-Jun mimicked the effects of HDAC8. Mechanistically, it was found that the introduction of the K273R acetyl mutant of c-Jun led to increased transcription of EGFR, the maintenance of ERK signaling, and the escape of the melanoma cells from BRAF inhibitor therapy. There is already evidence that c-Jun transcriptional activity can be induced in response to stresses such as UV (44). Our work provides the first evidence that HDAC8 activity is increased in responses to multiple, diverse cellular stresses and that this turn initiates a transcriptional program that is associated with increased melanoma cell survival (29, 43, 45).

In vivo models were then used to demonstrate the requirement for HDAC8 in the adaptation of melanomas to BRAF inhibitor therapy. Treatment of HDAC8-silenced melanoma xenografts with the BRAF inhibitor PLX4720 showed them to be unable to adapt to therapy. In contrast, introduction of HDAC8 into the same melanoma cells made them BRAF inhibitor tolerant, and the tumors grew rapidly in the presence of drug. To determine if these effects could be recapitulated by small-molecule HDAC inhibitors, we performed two experiments in which drug-naïve melanoma cells were cotreated with either a broad spectrum HDAC inhibitor (panobinostat) and PLX4720 or an HDAC8-specific inhibitor (PCI-03451) and PLX4720. In both cases, the combination of the BRAF inhibitor and the HDAC inhibitor out performed either single agent, with particularly striking effects being seen for the broad-spectrum HDAC inhibitor/BRAF inhibitor combination. Although significantly improved responses were seen for the HDAC8 inhibitor plus the BRAF inhibitor, these were not as impressive as with panobinostat or HDAC8 silencing. Possible explanations for this difference include the potential minor contribution of other HDACs to the process of therapeutic escape, or the failure of PCI-03451 to inhibit HDAC8 to the same extent as the HDAC8 shRNA silencing in vivo. Nevertheless, these findings provide a strong rationale to pursue the development of more selective and potent HDAC8 inhibitors for future evaluation as drugs that can limit phenotype switching and therapeutic escape in melanoma. In support of this goal, there is already evidence that melanomas with acquired BRAF–MEK inhibitor resistance exhibit sensitivity to the broad-spectrum HDAC inhibitor vorinostat (46), and that HDAC inhibitors can restore expression of BIM and BMF in melanomas with acquired BRAF inhibitor resistance (47).

In summary, we have shown that HDAC8 is a critical driver of a cellular program that allows melanoma cells to rapidly adapt to multiple cellular stresses, including BRAF inhibitor therapy. The mechanism of this adaptation is complex and involves the deacetylation of c-Jun leading to a transcriptional program that allows melanoma cells to rewire their signaling to maintain MAPK pathway activity. To date, attempts to therapeutically target c-Jun and, indeed, phenotype switching in melanoma have proven to be difficult. The development of potent HDAC8 inhibitors is a promising strategy to limit this plasticity in melanoma cells allowing therapeutic responses to be improved.

J.C. Becker is advisor at 4SC, eTheRNA, and CureVac; reports receiving commercial research grant from Alcedis and IQVIA; reports receiving honoraria from the speakers' bureau of Amgen, Sanofi, Pfizer, and Merck; and is a consultant/advisory board member for Sanofi, Amgen, Merck, and Pfizer. D. Schadendorf reports receiving honoraria from the speakers' bureau of Novartis, BMS, MSD, Sanofi, Roche, and Pierre-Fabre and is a consultant/advisory board member for Novartis, BMS, Merck-EMD, Immunocore, Array, MSD, Roche, Pierre-Fabre, Sanofi, Philogen, Inflarx, SunPharma, and Hexal. E. Seto reports receiving honoraria from the speakers' bureau of Otsuka Pharmaceutical and is a consultant/advisory board member for Otsuka Pharmaceutical, Institute of Biological Chemistry, and EMD Millipore. V.K. Sondak is a consultant/advisory board member for Array, Bristol Myers Squibb, Regeneron, Polynoma, Pfizer, Genentech Roche, Novartis, and Merck. K.S.M. Smalley reports receiving commercial research grant from Forma Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.F. Emmons, J.C. Becker, V.K. Sondak, J.D. Licht, K.S.M. Smalley

Development of methodology: M.F. Emmons, J.C. Becker, K.S.M. Smalley

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.F. Emmons, F. Faião-Flores, R. Sharma, J.L. Messina, J.C. Becker, D. Schadendorf, K.S.M. Smalley

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.F. Emmons, F. Faião-Flores, R. Thapa, J.L. Messina, J.C. Becker, V.K. Sondak, Y.A. Chen, J.D. Licht

Writing, review, and/or revision of the manuscript: M.F. Emmons, F. Faião-Flores, R. Sharma, R. Thapa, J.L. Messina, J.C. Becker, D. Schadendorf, V.K. Sondak, J.M. Koomen, Y.A. Chen, E.K. Lau, K.S.M. Smalley

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.F. Emmons, E. Seto, V.K. Sondak, J.M. Koomen, L. Wan

Study supervision: M.F. Emmons, V.K. Sondak, K.S.M. Smalley

Other (experimental/technical guidance): E.K. Lau

We would like to thank Bin Fang, Ph.D., for his assistance with the phosphoproteomic experiments and Divya Bhat for her assistance with the in vivo experiments.

This study was supported by SPORE grant P50 CA168536 (to K.S.M. Smalley, V.K. Sondak, J.L. Messina, and Y.A. Chen), NCI R21 CA216756 (to K.S.M. Smalley), Florida Department of Health 8BC03 (to K.S.M. Smalley and J.D. Licht), and Forma Therapeutics (to K.S.M. Smalley). This work has been supported in part by the Proteomics and Metabolomics Core, the Biostatistics and Bioinformatics Core, the Tissue Core, and Flow Cytometry Core Facility at the Moffitt Cancer, an NCI-designated Comprehensive Cancer Center (P30-CA076292).

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.