1. Introduction
Uveal melanoma (UM) is an ocular malignancy that arises from melanocytes residing in the uveal tract, which consists of the iris, ciliary body and choroid. It is the second most common type of melanoma and the most common primary intraocular malignancy in adults, affecting approximately 5.1 individuals per million per year; it is most frequent in Caucasians [
1,
2], as a fair skin and light eye color have been identified as host susceptibility factors [
3,
4].
In general, local tumor control is excellent, with large primary ocular melanoma being treated by enucleation, and small- to medium-sized tumors by application of a radioactive plaque, stereotactic irradiation or proton beam therapy [
5,
6,
7,
8,
9,
10]. Despite excellent regional tumor control, UM is still often lethal: up to 50% of patients will develop metastatic disease, for which no effective treatment exists [
11]. The liver is involved in approximately 90% of cases with metastasized disease [
12]. Metastatic disease may develop at any time from the initial diagnosis of the primary tumor to several years after diagnosis [
13].
Several pathological characteristics of the primary tumor are known to be associated with an infaust prognosis. These include a large size, ciliary body involvement, epithelioid cell type, extrascleral invasion and the presence of extravascular matrix loops [
14,
15,
16,
17,
18]. Furthermore, specific genetic features, such as monosomy 3, amplification of chromosome 8q, and loss of chromosome 1p, correlate with poor survival [
19,
20,
21,
22,
23]. On the contrary, an additional copy of chromosome 6p is associated with a favorable prognosis [
24,
25]. Microarray gene expression analyses have resulted in the identification of two classes of UMs: class 1 tumors have low metastatic risk, while class 2 tumors are associated with a high rate of metastatic death [
26,
27,
28].
Recently, mutations in specific genes such as
BAP1 (BRCA1-associated protein-1),
SF3B1 (splicing factor 3b subunit 1), and
EIF1AX (eukaryotic translation initiation factor 1A, X-linked) have been reported to have prognostic value [
29,
30,
31]. Aberrant DNA repair during the evolution of many malignancies and, accordingly, genomic instability is considered a hallmark of cancer cells [
32]. Recent research in UM has focused on genetics, with the aim of unraveling UM biology and identifying specific aberrations that underlie the development of UM and may indicate potential targets of therapy [
29,
30,
31,
33]. The BAP1 protein, the loss of which correlates to a poor prognosis in UM [
29,
34,
35], has been shown to promote DNA double-strand break repair [
36]. Yet, the role of DNA repair in tumor development and progression remains poorly studied. Although counterintuitive, DNA-repair proteins in compensating pathways may be targets for cancer therapeutics [
37,
38]. Since tumor cells that have lost a repair pathway may (over)rely on them (principle of synthetic lethality), one may try to block DNA-repair proteins to decrease the ability of UM cells to repair DNA damage. This may subsequently help to sensitize tumors to traditional anti-cancer treatment by chemotherapy or radiotherapy [
39].
However, it is not yet known whether and how the DNA-repair pathways are involved in the initiation and progression of UM. We, therefore, set out to analyze the expression of genes involved in DNA repair in UM and looked for genes that were associated with prognosis in UM.
To test our hypothesis that genes involved in DNA repair are differentially expressed between tumors with a favorable and unfavorable prognosis, we determined the expression of such genes in 64 UMs and made a comparison between tumors with and without loss of chromosome 3. Additionally, the relation with survival was evaluated for differentially expressed genes. Interesting associations were validated in two other sets of UM and a potential druggable target was explored further.
3. Discussion
Biological cellular responses following DNA damage include DNA damage repair, damage tolerance, cell-cycle checkpoint control and apoptosis. These mechanisms are tightly regulated and which pathway becomes activated depends on the type and severity of the DNA damage. In case of severe damage, the complex signaling pathways may eventually lead to cell cycle arrest (providing the cell more time for repair and tolerance mechanisms) or to apoptosis [
51,
52]. The recognition of expression patterns of the genes involved in DNA repair in UM is the first step in understanding the way these genes might play a role in UM development and may help in identifying new targets for therapy. We evaluated the expression of DNA-repair-related genes in the Leiden cohort of 64 UMs and aimed to identify genes with a variable expression between prognostically favorable and prognostically unfavorable UM. After validation in two other independent cohorts, we identified four genes which were associated with the degree of malignancy in UM: three genes (
BAP1,
WDR48, and
XPC1) showed an association between a low expression and poor survival, while
PRKDC was highly expressed in cases with an unfavorable prognosis. The genes
BAP1,
WDR48, and
XPC1 are all located on chromosome 3p and showed a significantly lower expression in monosomy 3 tumors. A lower expression of the
MLH1 gene, which is also located on chromosome 3p, was significantly related to prognosis in one cohort and showed a near-significant effect in the other cohorts. Since these four genes play a role in DNA repair, we can expect that impaired DNA repair is one of the results of the loss of a copy of chromosome 3. Sustained DNA damage as a result of deficient DNA repair mechanisms may lead to the accumulation of chromosomal abnormalities and gene mutations, which may promote cell growth and proliferation. Chromosome 3 loss does not occur in a single step since small tumors with partial monosomy have been observed [
53], but apparently, loss of the entire chromosome confers a selective advantage that might be mediated by the DNA-repair genes identified here.
BAP1 (BRCA1-associated protein 1) is a gene located on chromosome 3p21.1. The
BAP1 gene encodes a nuclear ubiquitin carboxy-terminal hydrolase, which is a deubiquitinating enzyme [
54]. It has been described to be a tumor suppressor gene in the BRCA-1 control pathway. The BAP1 protein contains binding domains for BRCA1 and BARD1, enzymes that form a heterodimeric complex that functions as a tumor suppressor [
55]. Loss of BAP1 has been shown to be related to a poor clinical outcome in UM [
29]. Similarly, a lower gene expression of
BAP1 in our study corresponded to a poor survival.
Ubiquination and deubiquination regulate essential biological processes such as DNA replication and DNA repair [
42,
55]. In accordance, BAP1 has been shown to play a role in the repair of DNA double-strand breaks by homologous recombination [
36,
56]. It has been suggested that the DNA-repair function of BAP1 may be the molecular basis for its tumor suppressor role in UM [
36].
Another DNA-repair-related gene involved in deubiquitination, which in our study showed a low expression in metastasizing uveal melanoma, is
WDR48. It is also known as
UAF1 and is located in close proximity (on 3p22.2) to
BAP1. UAF1 and USP1, a deubiquitinating enzyme, form the UAF1/USP1 complex, which regulates the Fanconi Anemia DNA-repair pathway [
57]. UAF1 activates USP1, and USP1 regulates the Fanconi Anemia repair pathway by deubiquitinating FANCD2, one of the most important players in this pathway. Fanconi Anemia is an inherited genomic instability disorder that led to the discovery of a novel DNA-repair pathway. The Fanconi Anemia repair pathway plays a role in the repair of DNA cross-links and can be activated after various types of DNA damage, such as ionizing radiation and ultraviolet light [
58,
59]. Accurate deubiquitination of the FANCD2 protein by the USP1/UAF1 complex is essential for an intact Fanconi Anemia pathway and proper DNA damage repair [
60,
61]. Because of this crucial role of the
WDR48 gene, and the association that we found of a low expression of
WDR48 with poor prognosis, a defective Fanconi Anemia repair pathway may play a role in the malignant transformation of UM. Murine fibroblasts deficient in UAF1 have been shown to exhibit profound chromosomal instability [
62].
XPC (Xeroderma Pigmentosum, complementation group C) is the third gene located on chromosome 3p. Its low expression was associated with poor survival in our study. The
XPC gene, located in the region 3p25.1, encodes a protein that helps to form the XPC repair complex and is involved in the early steps of the DNA Nucleotide Excision Repair (NER) pathway. Mutations in
XPC that impair the production of the XPC protein are related to Xeroderma Pigmentosum (XP), a rare recessive disorder, which makes patients extremely sensitive to ultraviolet light. This results in the frequent development of skin tumors, mainly in areas of the body exposed to the sun. The XPC protein acts a sensor detecting DNA damage [
63,
64,
65,
66]. The association of the low expression of
XPC with poor survival in UM is interesting, since evidence for the association of ultraviolet light exposure and UM development is inconclusive. However, XPC may play a role that is independent of its direct function related to UV-damage, as evidenced by the association of epigenetic silencing of
XPC with shorter survival in bladder cancer [
67]. The XPC repair complex contains the CETN2 protein, which shows a significantly lower expression in metastasizing UMs in the two validation cohorts of our study (
Table 4) [
68]. Xeroderma Pigmentosum is associated with a higher risk for ocular malignancies [
69].
In contrast to the genes discussed above, the
PRKDC gene that is located on chromosome 8q11.21 was found to be associated with worse survival when highly expressed [
70]. A heatmap showing the patients that developed metastases makes it clear that a low BAP1 expression (blue) is associated with a high PRKDC expression (
Figure 7).
PRKDC encodes the catalytic subunit of DNA-dependent serine/threonine protein kinase (DNA-PKcs). DNA-PK is involved in the repair of double-strand breaks (DSBs) by non-homologous end-joining (NHEJ) [
71,
72,
73]. DSBs can develop due to the effects of reactive oxygen intermediates or by exogenous agents such as ionizing radiation and anticancer chemotherapeutic drugs [
74].
High expression of DNA-repair proteins such as DNA-PKcs may increase the ability of tumor cells to withstand damage caused by chemotherapy or irradiation. Accordingly, increased DNA-PKcs activity was related to glioma resistance to cisplatin chemotherapy [
76]. Moreover, upregulation of DNA-PKcs was detected after irradiation of oral squamous cell carcinoma (OSCC) cells that were resistant to radiotherapy. Targeting DNA-PKcs has been suggested as a novel sensitization therapy of OSCC, and it has been shown to increase anticancer drug sensitivity in osteosarcoma cell lines [
77,
78]. Since the majority of primary UMs is treated by radiotherapy and certain chemotherapeutic targets are being tested for their effectiveness in killing UM metastases, elucidating the role of DNA-PKcs in UM may pave the way for sensitization therapy in UM by inhibiting DNA-PKcs. While some preliminary results indicate that inhibition of DNA-PKcs by NU7026 sensitizes UM cell lines for the topoisomerase I inhibitor, Topotecan, studies on cervical and breast cancer cells, as well as on lung cancer cells, have shown that this treatment sensitizes tumor cells to radiation treatment [
79,
80]. As far as we know, this combination has not been tried on UM cells. Van Oorschot et al. showed that the combination of hyperthermia and treatment with NU7441 led to an even better sensitization [
79].
We demonstrate that gain of chromosome 8q is related to a higher expression of
PRKDC in our cases, as well as in the TCGA cohort and UM cell lines. It is known that amplification of chromosome 8q is associated with an adverse clinical outcome in UM [
25,
81]. Although the exact mechanisms by which gain of chromosome 8q confers its malignant effect has not yet been elucidated, overexpression of
DDEF1 has been suggested as one potential mechanism [
82]. A recent study in prostate cancer has shown that the DNA-PKcs protein encoded by
PRKDC modulates cell invasion and migration and acts as a strong driver of tumor progression and metastasis [
45]. In addition, activated DNA-PKcs has been correlated with increased proliferation, decreased apoptosis and poor survival in hepatocellular carcinoma [
46]. In accordance, DNA-PKcs has been shown to be involved in normal cell cycle progression by controlling proper chromosome segregation and cytokinesis [
83].
In this study, we show that inhibition of DNA-PKcs results in decreased proliferation of UM cells. A recent study by Kotula et al. in the cutaneous melanoma cell line SK28 demonstrated that DNA-PKcs has pro-metastatic activity by modulating the tumor microenvironment through controlling the secretion of, e.g., matrix metalloproteinases (MMPs) and tissue inhibitors of matrix metalloproteinases (TIMPs) [
84]. We found a low and variable expression of MMPs and TIMPs in the majority of UM cell lines we analyzed and we did not observe an evident regulatory effect following DNA-PKcs inhibition. Since DNA-PKcs is postulated to be a driver of invasion and metastasis, we analyzed the effect of DNA-PKcs inhibition on the expression of an epithelial-to-mesenchymal transformation (EMT)—associated factors that have been shown to play a role in the invasiveness of UM cells (
ZEB1,
TWIST1,
SNAIL1) [
47]. Although the basal expression of these factors was low in the UM cell lines, we observed a decrease in the expression of the pro-metastatic
SNAIL1 upon DNA-PKcs inhibition. The inhibition of the protein interaction between DNA-PKCs and Snail1 has been suggested to be an effective strategy for inhibiting tumor migration [
85].
Considering the suggested pro-metastatic functions of DNA-PKcs, it is conceivable that an increased expression of
PRKDC, as a result of amplification of 8q, may contribute to the malignant progression in UM. This would imply that DNA-PKcs could be a potential target for therapy in UM. Furthermore, the use of inhibitors of DNA-repair proteins is a promising option for treating metastases, since cancer cells only retain some DNA-repair modules and are dependent on these for survival [
86].
4. Materials and Methods
4.1. Study Population
Our ‘training set’ contained 64 UMs obtained by primary enucleation at the Leiden University Medical Center (LUMC), Leiden, The Netherlands, between 1999 and 2008. Patient and tumor characteristics are shown in
Table 1. Sufficient frozen material of these tumors was available and DNA of adequate quality could be retrieved. Survival data was retrieved from the patients’ charts and from the Netherlands Comprehensive Cancer Organisation (
https://iknl.nl/over-iknl/about-iknl), and updated in March 2017. In The Netherlands, general physicians report every cancer patient to the Netherlands Comprehensive Cancer Organisation, which collects and registers information on the survival status by contacting the general physicians yearly. The follow-up in The Netherlands is not intensive because of a lack of effective treatments for UM metastases and patients are often referred back to their general physician after treatment of the primary tumor. The median follow-up time was 62 months and no patient was lost to follow-up.
Validation of the data was performed using two independent cohorts of post enucleation surgery patients: microarray datasets from Genoa and Paris, and RNAseq data of The Cancer Genome Atlas (TCGA) project [
42,
75]. Sixty-three untreated uveal melanoma provided by the Biological Resource Centre of Institut Curie (GSE2213840) [
86] and 48 UM samples from the Genoa cohort (GSE2783141 and GSE5188042) [
41,
87] were obtained from the Gene Expression Omnibus (
www.ncbi.nlm.nih.gov/geo/). The datasets were combined and normalized as described.
The study followed the tenets of the Declaration of Helsinki (World Medical Association of Declaration 1964; ethical principles for medical research involving human subjects) and the Medical Ethics Committee of the LUMC, Leiden, The Netherlands, had no objection regarding this research (G16.076/NV/gk).
4.2. Histologic Examination
After opening the enucleated bulbus, a part of the tumor was retrieved and snap frozen at −80 °C. The remaining tumor tissue was formalin fixed (4% neutral-buffered) and embedded in paraffin. A conventional histologic evaluation by an ophthalmic pathologist for confirmation of diagnosis and determination of characteristics was done. Parameters such as largest basal diameter (LBD, in millimeters), thickness (in millimeters), mitotic count (per 2mm
2 at 40× magnification, 8 high-power fields), tumor location, cell type (assessed according to the Armed Forces Institute of Pathology atlas) [
88] were evaluated on 4 μm-thick hematoxylin and eosin-stained sections. The 8th edition of the AJCC Cancer Staging Manual [
89] was used to stage tumors according to the TNM classification system.
4.3. Genetic Analyses
DNA and RNA were isolated from fresh-frozen tissue. DNA for single nucleotide polymorphism (SNP) analysis was extracted with the QIAmp DNA Mini kit and RNA for gene-expression profiling with the RNeasy Mini Kit (both from Qiagen, Venlo, The Netherlands). SNP array analysis to determine the chromosome copy number was performed with the Affymetrix 250K_NSP microarray chip (Affymetrix, Santa Clara, CA, USA) on all 64 UMs and with the Affymetrix Cytoscan HD chip (Affymetrix) on the cell lines. The Chromosome Analysis Suite (ChAS, version 2.0225) from Affymetrix was used to determine chromosome copy numbers. Gene-expression profiling at the transcriptional level was carried out on RNA of 64 UMs using 35,244 probes from the Illumina HT-12v4 chip (Illumina, San Diego, CA, USA).
RNA for real-time PCR analysis in cell lines was isolated using the SV total RNA isolation kit (Promega, Madison, WI, USA), then cDNA was synthesized using the reverse transcriptase reaction mixture, as indicated by Promega. qPCR was performed using SYBR green mix (Roche Diagnostics, IN, USA) in a C1000 touch Thermal Cycler (Bio-Rad laboratories, Hercules, CA, USA). Relative expression of PRKDC and SNAIL1 was determined compared to housekeeping genes CAPNS1 and SRPR. The untreated samples average was set at 1.
RNAseq analysis in the cell lines was conducted at Institut Curie (Paris, France) after isolation of total RNA using a NucleoSpin Kit (Macherey-Nagel, Düren, Germany). cDNA synthesis was conducted with MuLV Reverse Transcriptase in accordance with the manufacturers’ instructions (Invitrogen, Carlsbad, CA, USA), with quality assessments conducted on an Agilent (Santa Clara, CA, USA) 2100 Bioanalyzer. Libraries were constructed using the TruSeq Stranded mRNA Sample Preparation Kit (Illumina) and sequenced on an Illumina HiSeq 2500 platform using a 100 bp paired-end sequencing strategy. TopHat (v2.0.6) was used to align the reads against the human reference genome Hg19 RefSeq (RNA sequences, GRCh37) downloaded from the UCSC Genome Browser (
http://genome.ucsc.edu). Gene expression was determined by featureCounts and normalized using DESeq2.
Heatmaps and hierarchical clustering were performed in R (heatmap.plus package) using Euclidean distance and average linkage. Gene expression data were also correlated with Uveal Melanoma subtypes according to Robertson et al. [
42].
4.4. Gene Selection Procedure
We identified 121 genes encoding proteins involved in DNA repair mechanisms, based on a literature review on DNA repair, using the platforms Gene, Online Inheritance in Man (OMIM), Kyoto Encyclopedia of Genes and Genomes (KEGG) and PubMed. As our goal was to identify genes with a variable expression level, we determined the standard deviations of the expression levels of the DNA repair gene probes on the Illumina chip (
n = 178) (Appendix
Table A1). Certain genes were analyzed multiple times because they are encoded by different Illumina probes (in that case, the distinction between probes is made by placing letters in alphabetic order at the end of the gene name), while 18 genes were not analyzed since they were not on the Illumina chip. A selection of genes was made based on a cut-off value of the standard deviation of the expression (
Figure 1). A cut-off value of >0.5 would result in 6 genes, of >0.4 in 15 genes, and a cut-off value of >0.3 would lead to a total of 44 genes (encoded by 49 probes). A cut-off value of >0.3 was chosen to have a reasonably-sized group of genes with an acceptable level of variation in expression. The median expression of the probes of these 44 genes was compared between disomy 3 (D3) and monosomy 3 (M3) tumors and corrected for multiple testing using the Bonferroni method. A total of 13 Genes which were significantly differentially expressed after Bonferroni correction were selected for further analysis.
4.5. Cell Lines, DNA-PKcs Inhibition, and Proliferation Assay
Cell lines OMM2.5 (originally called OMM1.5 derived from a liver metastasis) and Mel270, which are derived from the same patient, were obtained from Dr. Bruce Ksander [
90] and maintained in RPMI supplemented with 10% FBS (fetal bovine serum) and antibiotics. MM28 was obtained from Dr. Sergio Roman-Roman [
91] and grown in IMDM supplemented with 20% FBS and antibiotics. The OMM1 cell line, maintained in RPMI supplemented with 10% FBS and antibiotics was established by Dr. Gré Luyten [
92]. Cell line 92.1 was developed in Leiden by Dr. Martine Jager [
93]. MM28 cells lack BAP1 expression, whereas Mel270, OMM2.5, and OMM1 cells are BAP1-positive.
To evaluate the effect of DNA-PKcs inhibition on the expression of pro-metastatic factors, the expression of these factors was evaluated in a primary UM cell line (Mel270) and in a metastatic UM cell line (MM28) before and after treating the cells with 10 µM NU7026 (#13308, Cayman Chemical, Ann Arbor, MI, USA, stock concentration 20 mM in DMSO) for 5 days. In order to analyze the effect of the DNA-PKcs inhibitor on the growth of these UM cell lines, the cells were seeded in triplicate in 96-well plates. Treatment with NU7026 was started the next day. Cells were replenished with fresh medium with or without drugs after three days. Relative survival was determined after five days with the use of the CellTitre-Blue cell viability assay (Promega) according to the manufacturer’s protocol.
4.6. Statistical Analysis
For data analysis, we used the statistical programming language R version 3.0.1 (R: A Language and Environment for Statistical Computing, R Core Team, R foundation for Statistical Computing, Vienna, Austria, 2014,
http://www.R-project.org) supplemented with specialized packages for SNP and RNA analysis. The main package used for SNP analysis was aroma.affymetrix, supported by ‘DNAcopy’ (Venkatraman E. Seshan and Adam Olshen, DNAcopy: DNA copy number data analysis. R package version 1.34.0), ‘sfit’ (Henrik Bengtsson and Pratyaksha Wirapati (2013), sfit: Multidimensional simplex fitting. R package version 0.3.0/r185,
http://R-Forge.R-project.org/projects/matrixstats/), and ‘R.utils’ (Henrik Bengtsson (2014), R.utils: Various programming utilities, R package version 1.29.8,
http://CRAN.R-project.org/package=R.utils). The ‘Aroma.Affymetrix’ package made it possible to use the information from the SNP microarrays to determine copy number values [
94,
95].
The packages used for RNA microarray analysis were ‘limma’ version 3.16.8, and the specific packages for Illumina microarrays: ‘lumi’ version 2.12.0, ‘annotate’ (R. Gentleman, annotate: Annotation for microarrays, R package version 1.38.0), and the database package ‘IlluminaHumanv4.db’ (Mark Dunning, Andy Lynch and Matthew Eldridge, IlluminaHumanv4.db: Illumina HumanHT12v4 annotation data (chip IlluminaHumanv4), R package version 1.18.0).
The statistical software package SPSS v.20.0.0 (IBM SPSS Statistics for Windows, IBM Corp., Armonk, NY, USA) was used for data analysis. Population characteristics were described using medians and percentages. The Mann–Whitney U test was performed to analyze numerical variables between two groups, and the Kruskal–Wallis test in case more than two groups were compared. Kaplan–Meier survival curves were made and the log rank test was used to analyze significance. Differences were considered to be significant if p < 0.05 after correction for multiple testing.