Dual Screen for Efficacy and Toxicity Identifies HDAC Inhibitor with Distinctive Activity Spectrum for BAP1-Mutant Uveal Melanoma Free

The concept behind targeted cancer therapy is to attack specific molecular vulnerabilities in cancer cells while sparing normal cells (1). To date, most such efforts have been focused on the enzymatic inhibition of mutant oncogenes such as KIT and BRAF (2). However, many cancer-driver mutations occur in recessively acting tumor-suppressor genes (3), which are notoriously recalcitrant to therapeutic intervention due to the difficulty in restoring lost function (4). One such tumor suppressor is BAP1, which is mutated in numerous cancer types, including uveal melanoma (UM), renal cell carcinoma, mesothelioma, and cholangiocarcinoma (5–9). BAP1 encodes a ubiquitin C-terminal hydrolase that deubiquitinates numerous substrates, including histone H2A, whereby it regulates transcription of genes involved in development, differentiation, and other processes (8, 10–12). Importantly, BAP1 mutations are often associated with increased metastatic risk and decreased survival (5, 6, 9), suggesting that targeting these mutations may inhibit metastasis. As such, a therapeutic strategy for targeting BAP1 mutations is urgently needed, but to date no such therapy has been identified.

Although synthetic lethality screens are a common strategy for identifying compounds that target tumor suppressors (13, 14), drugs that induce cell death often have a narrow therapeutic window with dose-limiting toxicity (15). These properties would further limit the use of such drugs in the adjuvant setting, where the goal is to treat patients at high risk for metastasis but who are still healthy. Consequently, we took a different approach to identify compounds that counteract the effects of BAP1 mutations (Fig. 1). As a primary high-throughput in vitro screening strategy, we used reversal of transcriptional repression caused by BAP1 loss, rather than cell death as the assay endpoint. Next, we screened the lead compounds using our recently described BAP1-deficient Xenopus embryo phenotype (11) as an in vivo tool for assessing rescue of the phenotype while simultaneously evaluating for toxicity. The final candidate identified by this approach was the HDAC inhibitor quisinostat, which has a distinctive activity spectrum that may explain its efficacy and low toxicity. Finally, we validated the ability of this compound to block the growth of tumors in vivo using a mouse uveal melanoma model, thereby nominating quisinostat as a high-value candidate for adjuvant therapy in patients with high-risk BAP1-mutant uveal melanoma.

Before conducting experiments, all cell lines used in this study were tested for Mycoplasma contamination by direct PCR assay on spent media using forward (GGGAGCAAACAGGATTAGATACCCT) and reverse (TGCACCATCTGTCACTCTGTTAACCTC) primers. PDX-derived BAP1-mutant uveal melanoma cell lines MM28 and MP46, and BAP1-wildtype uveal melanoma cell line MP41 were generously provided by Dr. Sergio Roman-Roman (Curie Institute, Paris) and maintained in 5% pO2 in DMEM/F12 with 10% heat-inactivated FBS, 1% penicillin/streptomycin, 2 mmol/L Glutamax, and 0.5% insulin–transferrin–selenium. Established BAP1-wildtype uveal melanoma cell lines 92.1 and Mel202 were obtained from the ATCC and maintained in 20% pO2 in RPM1 with 10% HI-FBS, 1% penicillin/streptomycin and 2 mmol/L Glutamax (Thermo Fisher Scientific). The 92.1 and Mel202 cells were engineered to inducibly knockdown BAP1 expression by creating shRNA constructs targeting three regions in BAP1 (CGTCCGTGATTGATGATGATA, CCCTGTATATGGATTTATCTT, and CCACAACTACGATGAGTTCAT, collectively known as shBAP1). These were cloned into Tet-pLKO-puro vector (Addgene #21915) and packaged into lentiviral particles by transient transfection into H293T cells, as previously described (16). The lentiviral particles were then transduced into the cells and selected with puromycin (2 μg/mL) after 48 hours. Transduced cells were then clonally selected, and individual clones were plated at <80% confluence and induced with doxycycline hyclate (1 μg/mL) for 48 hours. Whole-cell lysates were analyzed by western blot and probed for BAP1 (H300; Santa Cruz Biotechnology) and beta-actin (#4967S, Cell Signaling Technology). The clones with optimal BAP1 knockdown (>85% depletion) were used for in the study. Uninduced cells were used as control.

RNA was isolated with Direct-zol RNA kit (Zymo), and melanin pigment was removed using OneStep PCR Inhibitor Removal Kit (Zymo), according to the manufacturers' instructions. Libraries were prepared using the NEBNext Ultra kit and sequenced on the Illumina NextSeq 500 (single-end 75 base pair sequencing). Sequencing quality was assessed using FastQC (v0.11.3). Reads were trimmed using trim galore, aligned to the human genome build hg38/GRCH38 using STAR (17) and counts were generated using RSEM (18). Differential expression was performed using DESeq2 (19). LogFC value <-0.6 and P_adj value of <0.01 were used as cutoff criteria to identify commonly downregulated genes in class 2 BAP1-mutant versus class 1 BAP1-wildtype tumors from The Cancer Genome Atlas (TCGA), as well as in cell lines with conditional BAP1 knockdown (92.1 and Mel202).

Library of 1280 pharmacologically active compounds (LOPAC¹²⁸⁰ library) was obtained from Sigma-Aldrich. A custom in-house epigenetic library (UM-CTI) was assembled and curated by the University of Miami Center for Therapeutic Innovations (20–23). The compounds were purchased from Selleckchem, Biovision, Tocris, The Structural Genomics Consortium (SGC), Epigenetix, SIGMA, and Cayman.

To calculate the Z-factor value for GLO1 expression, shBAP1-inducible 92.1 cells were plated in 384-well plates at 250 cells/well and induced with doxycycline (1 μg/mL) for 48 hours. Because the chemical compound libraries were reconstituted in 100% DMSO, cells were treated with increasing concentration of the solvent (0%, 0.1%, 0.3%, and 0.9%) for an additional 72 hours to assess its effect on the assay performance. Cells were then washed with PBS and treated with lysis buffer, including DNase I, followed by Stop Solution then placed on ice to begin cDNA synthesis using the TaqMan Reverse Transcription kit (Thermo Fisher Cat. No. N8080234). Reverse transcriptase enzyme was mixed with Reverse Transcription Buffer and added to a PCR reaction plate. Cell lysates were added to the PCR plate and placed in the thermal cycler at 37°C for 60 minutes, 95°C for 5 minutes and 4°C indefinitely. TaqMan gene expression Master Mix protocol was used for qPCR (Thermo Fisher Cat. No. 4369016) and TaqMan primers GLO1 (Assay ID Hs00198702_m1, ThermoFisher) and 18S (Cat. 319413E, Thermo Fisher). The cDNA and Master Mix samples were placed at 50°C for 2 minutes, 95°C for 10 minutes, and 40 cycles of 95°C for 15 seconds and 60°C for 1 minute in the Applied Biosystems QuantStudio 6 Real-Time PCR system. The C_t values obtained from the QuantStudio data analysis software were analyzed by calculating fold-change. Gene expressions were normalized to 18S mRNA expression. Z-factor value for DMSO gradient was calculated as previously described (24). For the cell-based screen, 92.1-Tet-shBAP1 cells were plated in 384-well plates at 250 cells/well and induced with doxycycline (1 μg/mL) for 48 hours. The wells were then treated in duplicate for 72 hours with the LOPAC¹²⁸⁰ library or the UM-CTI library at a final concentration of 10 or 1 μmol/L, respectively. Z-prime value was calculated for all plates. The RQ values were converted to the percentage of positive controls. A cutoff criterion was set at 70% above the mean of positive controls and the hit compounds were identified.

The animal protocols used in this work were evaluated and approved by the Institutional Animal Care and Use Committee (IACUC) of the Universityof Miami (#18–031). All activities were performed in compliance with federal state, and institutional regulations. Xenopus laevis embryos were prepared as previously described (11). When embryos reached the 1-cell stage, they were injected with 15 ng of control morpholino CtrlMO (AGCTTTCGTTCATGTAACCTCCTCA) or bap1-directed morpholino Bap1MO (AGCCTTTATTCATGTTGCCTCCTCC) using a Nanoject II Auto-Nanoliter Injector (Drummond Scientific, Inc.). After injections, the embryos were maintained in 4% ficoll/MMR solution at 18°C until they reached stage 6 and then transferred to 0.1 MMR buffer containing varying concentrations of each compound tested. Embryos were then incubated for additional 48 hours at 21°C and scored for morphological abnormalities at early neurula stage and/or in the early tailbud stages. For each drug concentration, embryos were evaluated for phenotype expression using our previously established grading system (11). A total of n = 300 embryos from 3 different females were evaluated.

The animal protocols used in this work were evaluated and approved by the Institutional Animal Care and Use Committee (IACUC) of the Universityof Miami (#15–197). All activities were performed in compliance with federal state, and institutional regulations. Mice were randomized with respect to sex using the block randomization technique. For each mouse, 2 million MM28, MP46 or MP41 uveal melanoma cells were resuspended in ice-cold Matrigel (Corning 354248) and injected subcutaneously into the intrascapular region above the fat pad in NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ JAX immunodeficient (NSG) mice. Drug or vehicle treatments were initiated 3 weeks after the graft date for MP46 and after 2 weeks for MP41 and MM28 xenografts, due to differences in growth rates. Quisinostat was resuspended in 100% DMSO (50 mg/mL) and then reconstituted in 20% hydroxypropyl-β-cyclodextrin saline solution (final pH 8.7) at a final concentration of 0.17% (w/v) or 3.57 mmol/L quisinostat. Quisinostat or vehicle was administered via intraperitoneal injection on Mondays and Thursdays at a dose of 10 mg/kg body weight. Xenografts were imaged every 2 weeks by ultrasound (Vevo3100; Fujifilm Visualsonics) to assess tumor growth over time.

Uveal melanoma cells were plated in sextuplicate wells (1.0 × 10³ cells per well) and treated with increasing concentrations of MEKi (trametinib) and/or quisinostat for 72 hours. Cell viability was determined using MTT assay as previously described (25) with the addition of 100 μL of DMSO added to each well before the isopropanol solution.

Unless otherwise specified, statistical significance was assessed using the χ² test for ordinal variables and Mann–Whitney test for continuous variables using MedCalc version 19.1.

On the basis of the hypothesis that BAP1 loss promotes tumor progression and metastasis by altering the transcription of key genes, we sought to identify a representative BAP1-regulated gene that is silenced by BAP1 loss in uveal melanoma to screen for compounds that restore its transcription. First, we analyzed RNA-seq data from the 80 uveal melanoma samples from TCGA to identify genes that are differentially expressed between class 1 and 2 tumors (Fig. 2A; Supplementary Table S1), as previously defined (26). Then, we performed a similar differential expression analysis of RNA-seq data obtained from BAP1-wildtype 92.1 and Mel202 uveal melanoma cells engineered to inducibly express shRNA directed against BAP1 (Fig. 2A; Supplementary Table S1 and Supplementary Fig. S1A). Only 4 genes were differentially expressed in all 3 systems: GLO1, EDARADD, DAPL1, and BAP1 itself (Fig. 2B and C). The largest change in expression was observed for GLO1 (Fig. 2C), which was chosen for subsequent high-throughput screens.

We created a high-throughput platform to screen for compounds that rescue GLO1 expression in 92.1 uveal melanoma cells. The optimized platform demonstrated a high-throughput compatibility Z-factor value of 0.518 (Supplementary Fig. S1B). An initial screen of the LOPAC¹²⁸⁰ library of pharmacologically active compounds resulted in 11 hits, based on ≥70% rescue of GLO1 expression compared with DMSO control (Fig. 2D). We then screened a custom in-house epigenetic library of 180 compounds, which yielded 8 hits, all HDAC inhibitors (Fig. 2D). Dose–response assays were performed for all preliminary hits to eliminate false positives and to determine IC₅₀ values. IC₅₀ values were then used to compare potency of each candidate compound to restore GLO1 expression. Only one of LOPAC¹²⁸⁰ library compounds, ouabain, passed the dose-response validation test, for an overall hit rate for this library of 0.08% (Fig. 2E). Ouabain is a cardiac glycoside and Na/K-ATPase inhibitor that was not deemed a viable candidate due to potential cardiac side effects. Remarkably, all 8 hits from the in-house epigenetic library passed the dose–response validation test, for an overall hit rate of 4.4% for the epigenetics library (Fig. 2D and E; Supplementary Table S2).

Next, using IC₅₀ values generated in the primary assay, we used our recently described BAP1-deficient Xenopus embryo developmental model to screen the remaining lead compounds for both efficacy and toxicity (11). Trapoxin A was eliminated, as this was an irreversible HDAC inhibitor. Pracinostat and AR-42 (IC₅₀ for GLO1 rescue 3.73 and 2.65 μmol/L, respectively), demonstrated no toxicity but also no rescue (Supplementary Fig. S2). Therefore, these and other compounds with IC₅₀ > 1 μmol/L (trichostatin A, givinostat, and LAQ824) were eliminated. CUDC-907 (GLO1 rescue IC₅₀ = 0.81 μmol/L) caused severe toxicity in control and BAP1-deficient embryos. In contrast, quisinostat (GLO1 rescue IC₅₀ = 0.46 μmol/L) effectively rescued the BAP1-deficient phenotype in the nanomolar range with no detectable toxicity in control embryos even at micromolar concentrations (Fig. 3A and C). Because MEK inhibitors have shown some activity in uveal melanoma (27), we tested the effects of combining quisinostat with the MEK inhibitor trametinib, but this resulted only in a modest additive effect (Supplementary Fig. S3).

To test the potential clinical value of quisinostat in the adjuvant setting for high-risk BAP1-mutant uveal melanoma, we evaluated whether this compound could repress the growth of small uveal melanoma xenografts using an established mouse intrascapular fat pad model (28). We tested two BAP1-mutant (MM28 and MP46) and one BAP1-wildtype (MP41) PDX-derived cell lines (28). Quisinostat markedly reduced tumor growth over a 2-month period in both BAP1-mutant lines (MM28 and MP46), whereas it did not affect the growth of the BAP1-wildtype line (MP41; Fig. 4A–C).

Few drugs have been identified that target mutations in tumor suppressors, despite the pervasiveness of these aberrations in human cancer. In this study, we identify quisinostat as a potential targeted therapy in BAP1-mutant uveal melanoma, and we demonstrate the potential value of a novel multi-step drug-screening strategy for identifying such compounds. Capitalizing on the known role of BAP1 as an epigenetic regulator of gene expression, we used an initial high-throughput drug screen in which transcriptional restoration of a reporter gene, rather than cell death, was used as the endpoint. This resulted in no viable candidates from a library of pharmacologically active compounds, but multiple hits from an epigenetic compound library, all of which were HDAC inhibitors, lending credibility to the specificity of this approach. The secondary screen involved our distinctive in vivo BAP1-deficient embryonic developmental model to assess for efficacy and toxicity. This screen eliminated several candidates due to toxicity and/or lack of efficacy, leaving quisinostat as the lead candidate. Finally, we validated that BAP1 loss sensitized uveal melanoma tumors to quisinostat in vivo.

Quisinostat has a distinctive activity spectrum (29), including the highest potency of any compound in the Probes and Drugs portal (https://www.probes-drugs.org/home/) for HDAC4. Interestingly, most clinically available HDAC inhibitors have very low potency for HDAC4, which was recently identified as a key downstream target of BAP1 (11). We previously showed that earlier generation HDAC inhibitors such as valproic acid and vorinostat could induce differentiation and transcriptomic rewiring of uveal melanoma cells in vitro, but only at higher concentrations (30). The low potency of these compounds, especially against HDAC4, may explain their lack of efficacy in BAP1-mutant uveal melanoma (31) and their dose-limiting toxicity (32). Indeed, although quisinostat and CUDC-907 have comparable nanomolar affinity for HDAC1, HDAC2, HDAC11 and HDAC10 (Supplementary Table S2), CUDC-907 has low affinity for HDAC4 and failed to pass our secondary in vivo screen (Supplementary Fig. S2). Similarly, Pracinostat and AR-42 with low affinity for HDAC4 also failed the secondary in vivo screen (Supplementary Table S2 and Supplementary Fig. S2). Moreover, the lower potency of quisinostat in a BAP1-wildtype uveal melanoma cells (33) is consistent with preferential sensitivity of BAP1-mutant uveal melanoma to this compound.

Our screening approach differs conceptually from synthetic lethal screens, which have suggested that BAP1 mutations may increase vulnerability to PARP inhibitors (34), which are in clinical trials in the metastatic setting (ClinicalTrials.gov Identifier: NCT03207347). In contrast, the epigenetic mechanism of action of an epigenetic compound such as quisinostat may be more appropriate in the adjuvant setting to prevent the progression of micrometastatic disease to overt metastasis in patients with high-risk uveal melanoma and other BAP1-mutant cancers. Our study provides a proof of principle and framework for future screens of additional compound libraries to identify novel therapeutics targeting loss of BAP1 and other tumor suppressors.

J.N. Kuznetsoff reports grants from Department of Defense during the conduct of the study. D.A. Rodriguez reports grants from Department of Defense during the conduct of the study. N.T. Chee reports grants from Department of Defense during the conduct of the study. E.R. Roberts reports grants from Department of Defense during the conduct of the study. C.-H. Volmar reports grants from Department of Defense during the conduct of the study. J.W. Harbour reports grants from Department of Defense during the conduct of the study, as well as other from Castle Biosciences outside the submitted work, and a patent for US 9,809,856 and US 9,441,277 with royalties paid from Castle Biosciences. No disclosures reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

J.N. Kuznetsoff: Conceptualization, investigation, methodology, writing–original draft, writing–review and editing. D.A. Owens: Formal analysis, investigation, methodology, writing–review and editing. A. Lopez: Investigation, methodology. D.A. Rodriguez: Formal analysis. N.T Chee: Investigation. S. Kurtenbach: Formal analysis, visualization. D. Bilbao: Investigation. E.R. Roberts: Investigation. C.-H. Volmar: Investigation. C. Wahlestedt: Conceptualization. S.P. Brothers: Conceptualization, formal analysis, writing–review and editing. J.W. Harbour: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.

This work was supported by DOD W81XWH-15–1-0578, Alcon Research Institute, Research to Prevent Blindness, Inc. Senior Scientific Investigator Award, and a generous gift from Dr. Mark J. Daily (to J.W. Harbour), NCI P30CA240139 (Sylvester Comprehensive Cancer Center), NEI P30EY014801 (Bascom Palmer Eye Institute), and Research to Prevent Blindness Unrestricted Grant (Bascom Palmer Eye Institute). We acknowledge the assistance of the Molecular Therapeutics Shared Resource (MTSR), Biostatistics and Bioinformatics Shared Resource (BBSR), Oncogenomics Shared Resources (OGSR), and Cancer Modeling Shared Resource (CMSR) of the Sylvester Comprehensive Cancer Center of the University of Miami Miller School of Medicine.

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.