Identification of novel chemotherapeutic strategies for metastatic uveal melanoma

Melanoma of the uveal tract accounts for approximately 5% of all melanomas and represents the most common primary intraocular malignancy. Despite improvements in diagnosis and more effective local therapies for primary cancer, the rate of metastatic death has not changed in the past forty years. In the present study, we made use of bioinformatics to analyze the data obtained from three public available microarray datasets on uveal melanoma in an attempt to identify novel putative chemotherapeutic options for the liver metastatic disease. We have first carried out a meta-analysis of publicly available whole-genome datasets, that included data from 132 patients, comparing metastatic vs. non metastatic uveal melanomas, in order to identify the most relevant genes characterizing the spreading of tumor to the liver. Subsequently, the L1000CDS2 web-based utility was used to predict small molecules and drugs targeting the metastatic uveal melanoma gene signature. The most promising drugs were found to be Cinnarizine, an anti-histaminic drug used for motion sickness, Digitoxigenin, a precursor of cardiac glycosides, and Clofazimine, a fat-soluble iminophenazine used in leprosy. In vitro and in vivo validation studies will be needed to confirm the efficacy of these molecules for the prevention and treatment of metastatic uveal melanoma.

In the present study, we made use of bioinformatics to analyze the data obtained from three public available microarray datasets on uveal melanoma in attempt to identify novel putative chemotherapeutic options for the metastatic disease.

Material and Methods
Microarray dataset selection and meta-analysis. The NCBI Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/) 7 was used to identify suitable microarray datasets comparing localized vs. metastatic uveal melanomas. Three datasets were included in the study GSE22138 8 -GSE73652 9 , and GSE44295, that included whole-genome transcriptomic data from primary melanoma cells obtained upon eye enucleation ( Table 1). The following information were extracted from each of the studies that were selected: GEO accession; sample type; platform; numbers of patients and controls; and gene expression data. Briefly, the GSE22138 dataset included 28 samples of non-metastatic tumor and 35 samples from enucleated patients having distant metastasis. Patients were included in the group of localized tumors if no metastasis were observed in 36 months of follow-up. The GSE73652 dataset included 5 samples of non-metastatic tumor and 8 samples from enucleated patients having distant metastasis. Metastasis-free cases were included only if there was > 6 years follow-up. No detailed clinical data are available for GSE44292, that included 32 samples from patients with localized disease and 24 sample from patients with metastasis. Illumina platforms were pre-processed using Beadarray prior to quantile normalization (Dunning et al., 2007), while Affymetrix platforms were pre-processed and quantile normalized using the robust multiarray average (RMA). The datasets were uploaded to INMEX (http:// www.inmex.ca/INMEX) 10 , and the data annotated by converting probe ID to Entrez IDs. For each probe-set, intensity values were auto-scaled and a data integrity check was performed prior to the meta-analysis stage. Batch effects were corrected using the "ComBat" function. A random effects model of effect size (ES) measure was used to integrate gene expression patterns from the three datasets. The random effects model presumes that different studies present substantial diversity, and evaluates between-study variance as well as within-study sampling error. Genes with FDR < 0.01 were identified as Differentially Expressed (DE) and selected for further analysis.
Gene Ontology (GO) and Drug Prediction Analysis. Functional relationships among DE gene were obtained from GeneMania (http://genemania.org/ http://genemania.org/) 11 . GeneMANIA searches publicly available genomics and proteomics data, including data from gene and protein expression profiling studies and primary and curated molecular interaction networks and pathways, to find related genes. The network weighting method is 'Gene-Ontology (GO) based weighting, Biological Process based' . This weighting method assumes the input gene list is related in terms of biological processes (as defined by GO).
The PANTHER (protein annotation through evolutionary relationship) classification system 12 (http:// www.pantherdb.org/) was used to classify input genes according to their function (PANTHER protein class). PANTHER is a comprehensive system that combines gene function, ontology, pathways and statistical tools that enable to analyze large-scale, genome-wide data from sequencing, proteomics or gene expression experiments.
The L1000CDS 2 was used to identify potential chemotherapeutics for metastatic uveal melanoma. L1000CDS 2 enables to find L1000 small molecule signatures that match input gene signatures. The L1000 mRNA gene-expression dataset is generated as part of the Library of Integrated Network-based Cellular Signatures (LINCS) project, a NIH Common Fund program. LINCS aims to profile the molecular effects of small molecules on human cells 13 . When gene lists are submitted to L1000CDS 2 , the search engine compares the input gene lists to the DE genes computed from the LINCS L1000 data and returns the top 50 matched signatures. The result score is the overlap between the input DE genes and the signature DE genes divided by the effective input. The effective input is the length of the intersection between the input genes and the L1000 genes. L1000CDS 2 currently covers chemically perturbed gene expression profiles from 62 cell-lines and 3,924 small molecules. Also, L1000CDS 2 allows to predict effective drug combinations by comparing every possible pair among the top 50 signatures and computing the potential synergy for each pair.

Results
Meta-analysis of gene expression in metastatic uveal melanoma. Three GEO data sets were identified for the following analysis (Table 1). These datasets consisted of primary uveal melanoma data, and included a total of 67 metastatic tumors and 65 localized primary tumors. Details of the individual studies are presented in Table 1. Figure 1A shows a Principal Component Analysis of the three separate microarray datasets included in the meta-analysis.
A total of 64 genes were identified, which were consistently modulated in metastatic tumors. Among these 64 DE genes, 29 were upregulated and 35 were downregulated. A list of the significant upregulated and downregulated genes is shown in Tables 2 and 3, respectively. An heatmap showing the 64 DE genes is presented in Fig. 1B The functional relationships among the up-and down-regulated genes, respectively, are presented in Fig. 2A,B. Among the upregulated genes, the top three represented protein classes were: Transferase, which included phosphatases (CDC25B and IMPA1) and proteases (IDE and TYSND1); Enzyme Modulator, including G protein modulators (SGSM2, SRGAP2 and SIPA1L2) and the protease inhibitor, HPN; Nucleic Acid Binding, i.e. the DNA helicase, CHD7, and the DNA ligase, LIG1 (Fig. 2C). Among the downregulated genes, the top three protein classes represented were: Nucleic Acid Binding, that included DNA binding proteins (H2AFY2, NDN and MBNL1) and RNA binding proteins (ANG, EIF4A2 and EIF4A3); Hydrolase, including the proteases, ABHD6 and LTA4H, and the lipase, PLCD1; Enzyme Modulator, including the G protein modulator, PLCD1, the kinase modulator, HOOK1, and the protease inhibitor, VWA5A (Fig. 2D).  Prediction of novel chemotherapeutics for metastatic uveal melanoma. The L1000CDS 2 web-based utility was used to predict small molecules and drugs targeting the metastatic uveal melanoma gene signature. Figure 3 shows a clustergram with the top ranked L1000 perturbations (e.g. those with most anti-similar signatures). The complete list of predicted chemotherapeutics is presented as Table 4. Several of these drugs are FDA-approved and already used in the clinic or tested in clinical trials, as indicated in Table 4. Among the predicted chemotherapeutics, the most represented classes were: histone deacetylase inhibitors (that included HDAC6 inhibitor ISOX, BRD-K13810148, Trichostatin A and Vorinostat), and anti-infectious/parasitic drugs (i.e., Clofazimine, Erythromycin ethylsuccinate, Demeclocycline and Quinacrine hydrochloride). The top identified drugs were the following: BRD-K07220430 (Cinnarizine), an anti-histaminic drug used for motion sickness, was predicted for its ability to downregulate CHAC1 and to upregulate MBNL1, LPAR6, PLSCR4, NDN, ABHD6, ZSCAN18 and ZBTB20; Digitoxigenin, for its ability to downregulate CDC25B, IDE, INTS8 and MTDH, while upregulating F11R, ID2, and RAB11FIP1; clofazimine, a fat-soluble iminophenazine used in leprosy, which is able to downregulate CDC25B, CHAC1, and SHC1, and to upregulate ABHD6, PLOD2, PLSCR4, ZBTB20, ZSCAN18. Accordingly, the top three most promising drug combination found were: BRD-K07220430 and Digitoxigenin; Digitoxigenin and OSSK_645683; BRD-K07220430 and HDAC6 inhibitor ISOX (for the complete list, see Table 5).

Discussion
Melanoma of the uveal tract accounts for 5% of all melanomas and, with an incidence of about 6 cases per million person-years, represents the most common primary intraocular malignancy 14,15 . Despite improvements in diagnosis and more effective local therapies for primary cancer, the rate of metastatic death has not changed in the past forty years 15 .  Several genes and pathways have been identified to be involved in the progression and metastasis of uveal melanoma, such as c-Met 16 , Hepatocyte Growth Factor (HGF) 16,17 , Insulin-like Growth Factor-1 Receptor (IGF-1R) 18 , the CXCL12-CXCR4 pathway 17,19 , VEGF 20 , Mda-9/syntenin 21 and the PTP4A3 phosphatase 8 . However, despite     increasing knowledge in the biology of uveal melanoma, no therapeutic strategies has been found to be effective. Metastatic uveal melanoma is resistant to current systemic chemotherapy and, to date, no clear role has been established for chemotherapy in several clinical trials, that reported objective response rates < 20% 4 . Single-agent chemotherapies (e.g., dacarbazina, fotemustine, paclitaxel, temozolomide, camptothecin, treosulfan) or combination chemotherapies (e.g. gemcitabine/treosulfan, cisplatin/gemcitabine/treosulfan, carboplatin/paclitaxel/ sorafenib) have shown poor response rates and immunotherapy with ipilimumab, antiangiogenetic treatment strategies using bevacizumab combined with interferon-α 2b, temozolomide, or aflibercept have not proven to be superior to chemotherapy 22 . In order to prevent metastatic disease, several adjuvant studies are also being conducted using ipilimumab, sunitinib, valproic acid, and crizotinib for high-risk patients 4 . Adjuvant treatment with bacillus Calmette-Guerin and interferon-α , as well as, intra-arterial hepatic infusion of fotemustine, have been previously studied in an effort to reduce the occurrence of liver metastasis, but none of these studies has demonstrated significant improvement in metastasis free survival or OS 4 .
The limited efficacy of current chemotherapies proves the medical need for more effective treatment strategies in metastatic uveal melanoma. In the present study, a meta-analysis of three whole-genome microarray datasets on primary tumor samples from the enucleated eyes of patients with localized or metastatic disease, has been performed in order to investigate the genes associated to the metastatic properties of uveal melanoma and to identify potential chemotherapeutic strategies to be used in metastatic disease.
It was recently developed a 15-gene qPCR-based assay that discriminate between primary uveal melanomas that have a low metastatic risk and a high metastatic risk 23 . This platform is currently used in a College of American Pathologists (CAP)-accredited Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory on fine needle aspiration samples and on formalin-fixed specimens 23 . A few of the genes screened in this assay correspond to those identified by the current analysis, namely ID2 and LTA4H, among the downregulated genes. Some of the other DE genes, share similar function to those included in the assay, i.e. EIF4A2 and EIF4E2 (to EIF1B).
As regards the putative drugs here identified, some of them are already in clinical use, such as BRD-K07220430 (Cinnarizine) 24 , clofazimine 25 , mesoridazine besylate 26 , erythromycin ethylsuccinate 27 ; other are in preclinical development (e.g. HDAC6 inhibitor ISOX) 28 or do not have any known pharmacological target, e.g. OSSK_645683. Interestingly, some of the identified drugs are histone deacetylase inhibitors (such as, HDAC6 inhibitor ISOX, BRD-K13810148, Trichostatin A and Vorinostat), that have already been shown efficacy on uveal melanoma preclinical models [29][30][31][32][33][34][35][36] 30 . In addition, the Class III-specific HDAC inhibitor, Tenovin-6, was shown to be able to eradicate cancer stem cells in 92.1 and Mel 270 cells 29 . Venza in 2014 31 , reported that the Class-I histone deacetylase inhibitor, MS-275, due to its ability to reduce c-FLIP expression, is able to increase TRAIL-induced cell death in uveal melanoma cell lines. MS-275, is currently tested in a Phase 1 study for the treatment of patients with Refractory Solid Tumors, including intraocular melanoma, or Lymphomas (NCT00020579). Vorinostat, a small molecule inhibitor of class I and II histone deacetylases, is currently tested in two Phase 2 Study on Ocular Melanoma With Extraocular Extension, Recurrent Uveal Melanoma and Grade IV Metastatic Uveal Melanoma (NCT00121225, NCT 01587352).
The utility of the present data may reside primarily in the potential identification of effective adjuvant treatments. In light of the fact that there is no therapy for metastatic disease, the most promising strategy to improve survival is to treat patients in an adjuvant setting. Although there might be non-metastasizing melanomas, which never metastasize, even if not treated, however the intra-tumoral genetic heterogeneity suggests an ongoing evolutionary tumoral process 37 . Adjuvant interventions could settle up the uncertainty about the effect of ocular treatment on survival, since some patients might unnecessarily sacrifice their visus in the hope of a longer life-expectancy; while other patients with small melanomas might succumb due to metastasis because treatment has been delayed. Presently, there is no adjuvant approach that improves outcome, but the impact of ocular treatment, in terms of therapeutic benefit, has ethical implications.
The present study has several advantages but also limitations. It represents, to date, the largest meta-analysis comparing metastatic vs. non metastatic uveal melanomas, encompassing whole-genomic data from a total of 132 patients. A strict and rigorous statistics has been applied to data in order to sort out the most relevant genes characterizing the spreading of tumor to the liver. Among the upregulated genes, the highest effect size was detected for JPH1, encoding for Junctophilin-1, a component of junctional complexes between the plasma membrane and endoplasmic/sarcoplasmic reticulum, providing functional cross-talk between the cell surface and intracellular calcium release channels (http://www.genecards.org/cgi-bin/carddisp.pl?gene= JPH1). Among the significant downregulated genes, the gene with the highest effect size was PLSCR4, encoding for the Phospholipid Scramblase 4, which has been found to be strongly expressed in the neuropil of malignant gliomas and in cytoplasm of liver cancers, colorectal cancers and malignant melanomas (http://www.proteinatlas. org/ENSG00000114698-PLSCR4/cancer). Comparing to the original analysis, the PTP4A3 phosphatases and the cancer-testis antigen, PRAME, were dropped out because of too little statistical significance. In addition, although Scientific RepoRts | 7:44564 | DOI: 10.1038/srep44564 the role of the PKC, MAPK and PI3K/AKT/mTOR signaling cascades has been extensively examined in uveal melanoma [38][39][40] , however, whole genome transcriptomic analysis may fail to detect their modulation, because of the primary contribution related to post-transcriptional modifications. Also, our analysis was very statistical stringent (FDR < 0.01), and this may account for the differences from previously published work 1 . Furthermore, the discrepancies with previous analysis may be due to the enlargement of the number of samples which could influence DE genes detection as respect to smaller sample groups.
Investigating the genes found to be significantly modulated in the present analysis may help to identify the molecular mechanisms involved in the liver metastasis of uveal melanoma, and may allow the identification of novel pharmacological targets for the prevention and treatment of liver involvement. Moreover, our study was aimed at predicting putative pharmacological schemes to apply as adjuvant or curative therapy, making use of the Library of Integrated Network-based Cellular Signatures (LINCS) project database. The repurposing of drugs currently approved for use in the clinical setting may expedite the design of phase II-III clinical trials, being cost-effective and reducing the risks associated with early stages of drug development 41 . The highest ranking scores were found for Cinnarizine, Digitoxigenin and Clofazimine. Cinnarizine has already been shown to be effective in vitro against lymphoma and multiple myeloma 42 and to inhibit melanogenesis in mouse B16 melanoma cells 43 , and Clofazimine was found effective in preclinical models of pancreatic ductal adenocarcinoma and triple-negative breast cancer 44,45 . However, this study has also limitations. Despite the practicality of this bioinformatic approach, there are drawbacks to point out. First, the efficacy of a drug is more complex than the simple match of expression profiles. Drugs have to reach tissue-specific concentrations to exert an effect, and the route and timing of administration is essential for the drug to be effective and to limit side effects. A preliminary in vitro testing on both primary and established cancer cell lines will be required before running pilot phase II clinical trials. On the other hand, for successful result of the clinical trials, the appropriate selection and recruitment of patients will be also key to evaluate the potential activity of the compounds in the clinical setting.