Uveal melanoma, the most common eye malignancy, causes severe visual morbidity and is fatal in approximately 50% of patients. Primary uveal melanoma can be cured by surgery or radiotherapy, but the metastatic disease is treatment refractory. To understand comprehensively uveal melanoma genetics, we conducted single-nucleotide polymorphism arrays and whole-genome sequencing on 12 primary uveal melanomas. We observed only approximately 2,000 predicted somatic single-nucleotide variants per tumor and low levels of aneuploidy. We did not observe an ultraviolet radiation DNA damage signature, but identified SF3B1 mutations in three samples and a further 15 mutations in an extension cohort of 105 samples. SF3B1 mutations were associated with good prognosis and were rarely coincident with BAP1 mutations. SF3B1 encodes a component of the spliceosome, and RNA sequencing revealed that SF3B1 mutations were associated with differential alternative splicing of protein coding genes, including ABCC5 and UQCC, and of the long noncoding RNA CRNDE.
Significance: Our data show that despite its dismal prognosis, uveal melanoma is a relatively simple genetic disease characterized by recurrent chromosomal losses and gains and a low mutational burden. We show that SF3B1 is recurrently mutated in uveal melanoma, and the mutations are associated with aberrant alternative splicing. Cancer Discov; 3(10); 1122–9. ©2013 AACR.
This article is highlighted in the In This Issue feature, p. 1083
Uveal melanoma arises in the iris, ciliary body, and choroid. Light skin complexion, fair hair, blue eyes, and the presence of cutaneous nevi are risk factors in uveal melanoma (1), but the ultraviolet radiation (UVR)-associated increase in cutaneous melanoma that has occurred in countries such as Australia over the past 4 decades has not been accompanied by parallel increases in uveal melanoma (2–4). Thus, the role of UVR in uveal melanoma etiology is unclear.
Class I uveal melanomas present a low risk of metastasis, whereas class II tumors are highly metastatic and are characterized by monosomy of chromosome 3 and gain of 8q. Mutually exclusive mutations in GNAQ or GNA11, the principal driver oncogenes in uveal melanoma, occur in approximately 85% of cases (5, 6), and inactivating mutations in the tumor suppressor BAP1 occur in approximately 85% of metastatic tumors and are associated with disease dissemination (7). Recently, exome sequencing of uveal melanomas has identified recurrent mutations in EIF1AX and SF3B1 (8, 9), predominantly in low-grade tumors. Somatic mutations in SF3B1, which encodes a component of the spliceosome, also occur in hematologic, breast, and pancreatic cancers (10–13). Mutant SF3B1 is associated with differential gene splicing in chronic lymphocytic leukemia (14), but aberrant splicing in SF3B1-mutant uveal melanoma has not been reported (9).
To gain insight into uveal melanoma genetics, we conducted single-nucleotide polymorphism (SNP) array analysis, whole-genome sequencing (WGS), and RNA sequencing (RNA-seq) on 12 frozen primary uveal melanoma samples. Despite its dismal prognosis, we find that uveal melanoma has a remarkably low mutation burden, and we did not observe a UVR DNA damage signature. However, we found recurrent mutations in SF3B1 that were associated with differential alternative splicing of both coding and noncoding genes that may play a role in the etiology of this disease.
Our discovery cohort comprised 12 primary uveal melanoma T3–T4 tumors that represented different histologic types (one epithelioid cell, three spindle cell, and eight mixed cases) and were treated by primary enucleation (Supplementary Table S1). One case was metastatic at diagnosis and six patients subsequently developed metastases. SNP array analysis was conducted using Illumina HumanOmni2.5 SNP arrays, and whole-genome sequencing was conducted on the Illumina HiSeq 2000 platform. The whole-genome sequence coverage was >30× (Supplementary Table S2), and the data were aligned to the reference genome and duplicate reads excluded. The whole genomes were compared with their matched normal DNA to identify chromosomal translocations, short insertions/deletions (indels), and somatic single-nucleotide variants (SNV).
The SNP arrays revealed low levels of aneuploidy in 11 of the tumors and tetraploidy in tumor #7 (Supplementary Fig. S1; Supplementary Table S3). The diploid tumors presented recurrent chromosome 3 monosomy (nine tumors), losses of 1p (five tumors), 6q (three tumors), and 8p (five tumors), and gains in 6p (four tumors) and 8q (seven tumors; Supplementary Figs. S1 and S2). These aberrations are characteristic of uveal melanoma, but we also observed loss of 16q in three tumors and gain of chromosome 11 in two (Supplementary Figs. S1 and S2). The presence of two identical copies of chromosome 3 in tumor #7 suggested that chromosome 3 monosomy preceded acquisition of the tetraploid state. Tumor #7 also displayed trisomy for 1q and monosomy for 8p. Commensurate with the SNP array data, whole-genome sequencing also revealed a low level of chromosomal aberrations. We predicted two to 59 interchromosomal translocations, 0 to seven intrachromosomal translocations, no inversions, two to 25 large deletions, and 0 to five large insertions (Supplementary Fig. S2; Supplementary Table S4). Thus, the frequency of structural variations in uveal melanoma is only approximately 40% of that reported in cutaneous melanoma and only approximately 20% of that reported in acral melanoma (Fig. 1A; refs. 15, 16). It is also only approximately 10% of that we recently found in mucosal melanoma (Fig. 1A; ref. 17).
The whole-genome sequencing also revealed a very low number of SNVs. We predicted only 1629–2604 (median 2112) somatic SNVs and 47–178 (median 67) somatic short indels (Table 1). The SNV mutation rate of <1 per Mb (Table 1) is markedly lower than is seen in most other types of cancer (Fig. 1B) and significantly lower than is seen in cutaneous (∼30,000 mutations/genome), mucosal (∼8,000 mutations/genome), or acral (∼5,000 mutations/genome) melanoma (15–17).
UVR-induced DNA damage is characterized by C>T transitions at the 3′ end of pyrimidine dinucleotides (18), a “signature” that accounts for 80% to 90% of mutations in cutaneous melanoma, and up to 60% of the mutations in acral melanoma (15, 16). Although C>T (G to A) transitions were the most common mutation in uveal melanoma, they accounted for only approximately 35% of the lesions (Fig. 1C) and were not enriched at the 3′ position of pyrimidine dimers (Fig. 1D). Thus, UVR-induced DNA damage does not seem to play a role in uveal melanomagenesis.
In line with the generally low level of mutations, we observed very few coding region mutations, predicting only 4–19 nonsynonymous SNVs per tumor; of which 92% of those tested were validated by Sanger sequencing (Supplementary Table S5). We also predicted only 0–2 coding region indels per tumor (Table 1). The only recurrent mutations were A>T, p.Q209L mutations in GNA11 (seven tumors) and T>A/T>G, p.Q209L/P mutations in GNAQ (three tumors; Supplementary Table S6). We did, however, observe a small number of nonrecurrent mutations in individual genes with possible functional significance, including C>T p.P107L in GNA15 that was coincident with A>T, p.Q209L GNA11 in tumor #10, C>T p.G8R in EIF1AX in tumor #8, and BAP1 mutations in seven tumors (Supplementary Table S6).
Critically, we also observed nonrecurrent mutations in SF3B1 in three tumors (T>G, p.K666T: tumor #6; T>C, p.K700E: tumor #11; C>T, p.R625H: tumor #12), and although there were no chromosome losses or gains in the region containing SF3B1 (2q33.1), these data suggested a role for SF3B1 in uveal melanoma. We screened SF3B1 in 105 additional consecutive archival primary uveal melanomas and detected 15 additional mutations (eight p.R625H, four p.R625C, one p.R625P, one p.R625L, one p.K666T; Supplementary Table S7). Our overall mutation rate of 15% (18/119) is similar to the rate (18.6%) recently reported for SF3B1 mutations in uveal melanoma by whole-exome sequencing (9), but note that in addition to the R625 codon mutations reported therein, we also observed K666 and K700 codon mutations (Supplementary Fig. S3). The SF3B1 mutations are inversely associated with chromosome 3 monosomy, and, notably, they are associated with improved progression-free and cancer survival (Supplementary Table S7; Supplementary Fig. S4).
SF3B1 encodes subunit 1 of splicing factor 3b, a component of the spliceosome, so to evaluate the effects of SF3B1 mutations on uveal melanoma transcripts, we hybridized three SF3B1-mutated tumors and three SF3B1–wild-type tumors to Affymetrix Human Transcriptome Arrays (HTA2), which contain both exon and exon–exon junction probes. Three hundred twenty-five genes were predicted to be differentially expressed, with 46 genes upregulated and 279 genes downregulated in the SF3B1-mutant compared to the SF3B1–wild-type tumors (Supplementary Table S8). Gene Ontology (GO) and pathway analysis of the differentially expressed genes did not predict any significant GO term(s), and did not predict KEGG or REACTOME pathway enrichment. However, splicing level analysis predicted 130 genes that contained at least one differentially regulated exon and/or splicing pattern (Supplementary Table S9). Manual inspection of the predicted events by the GenoSplice EASANA visualization interface revealed eight high-confidence or very high-confidence events, including alternative terminal exons (four events), alternative 3′ acceptor splice sites (two events), alternative cassette exons (one event), and intron retention (one event; Table 2). Critically, six of these events (GUSBP11, UQCC, ANKHD1, ADAM12, CRNDE, and ABCC5) were also identified when we analyzed the RNA-seq data from Harbour and colleagues (Table 2).
Next, we compared the RNA-seq data from our three SF3B1-mutant tumors to our nine SF3B1–wild-type tumors. For this, we used DEXSeq, a Bioconductor package that uses generalized linear models to detect differential exon usage (19), and also MATS, which uses a Bayesian statistical framework to identify alternative splicing (20). Forty-seven genes were predicted to be differentially spliced in these two populations by at least one algorithm (Supplementary Tables S10 and S11). Strikingly, when we compared our analysis of our HTA2 and the Harbour and colleagues data with our RNA-seq analyses, three alternative splicing events involving CRNDE, ABCC5, and UQCC, were identified by all three analyses (Table 2). These data suggest that CRNDE, ABCC5, and UQCC are strong candidates for alternative splicing in SF3B1-mutant tumors; hence, we examined the sequencing profiles for the three genes in our RNA-seq data. We normalized the number of mapped bases for these genes and compared nucleotide coverage at each base in SF3B1-mutant and wild-type tumors (Fig. 2).
The profiles for UQCC revealed clear evidence of alternative terminal exon use in the SF3B1–wild-type and SF3B1-mutant tumors (Fig. 2A). For CRNDE, we observed near-uniform representation of all bases of exon 4 in the SF3B1–wild-type tumors, but an enrichment of the reads at the 3′ end of this exon in the SF3B1-mutant tumors (Fig. 2B). Finally, in ABCC5, we observed clear evidence of differential splicing of intron 8 (Fig. 2C). Critically, we detected all three of these splicing events when we analyzed the RNA-seq data from Harbour and colleagues (Supplementary Fig. S5; ref. 9), providing independent confirmation of these alternative splicing events.
Finally, to further validate our findings, we assessed alternative splicing of GUSBP11, UQCC, ANKHD1, GAS8, F8, ADAM12, CRNDE, and ABCC5, the eight genes that provided the strongest evidence of splicing, by quantitative real-time PCR in 74 independent uveal melanomas, comprising 58 SF3B1–wild-type tumors and 16 SF3B1-mutant tumors. This analysis confirmed that all eight genes were alternatively spliced in SF3B1-mutant tumors compared with SF3B1–wild-type tumors (Fig. 2D; Supplementary Fig. S6).
We describe here the first whole-genome sequencing of uveal melanoma, and our data reveal that this is a comparatively simple genetic disease characterized by recurrent chromosomal gains and losses and a relatively low number of SNVs and structural variants. The tumor genomes display a homogenous SNV burden, both in terms of number (1,629–2,604) and mutation class, and, notably, they do not display a canonical UVR-induced DNA damage signature at pyrimidine dinucleotides. The absence of this signature negates an obvious influence of UVR in the etiology of this disease.
We confirm that GNAQ and GNA11 are the most commonly mutated driver oncogenes and that BAP1 is the most commonly mutated tumor suppressor. In addition, we confirm SF3B1 as recurrently mutated in 15% of cases, with a mutation hotspot at codon R625. SF3B1 mutations have been reported in hematologic, breast, and pancreatic cancers (10–13). Intriguingly, in those cancers, codon K700 mutations predominate, whereas in uveal melanoma, R625 codon mutations predominate. This suggests either that the gene mutations have distinct etiology, so different hotspots are targeted in each disease, or that the diverse biology of the diseases favors selection of discrete mutations. Notably, in common with previous studies (8, 9), we confirm that in uveal melanoma SF3B1 mutations are associated with better prognosis. Thus, in uveal melanoma and myelodysplastic syndrome, SF3B1 mutations are associated with improved outcome, whereas in chronic lymphocytic leukemia (CLL), SF3B1 mutations are associated with poorer prognosis (21).
SF3B1 encodes subunit 1 of splicing factor 3b, a component of the spliceosome, a large intracellular machine that processes precursor mRNA into mature transcripts. Specifically, splicing factor 3b anchors precursor mRNA onto the spliceosome to define the splicing site. Previous studies show that spliceosomal component mutations can alter splicing within a gene, can cause intron retention, or can cause aberrant alternative splicing, affecting protein isoform balance and thereby cell proliferation and differentiation (22, 23). In CLL, SF3B1 mutations are associated with alternative splicing at the 3′ ends of genes to generate truncated variants of the vitamin C transporter SLC23A2, the T-cell regulator TC1RG1, and the forkhead transcription factor FOXP1. We show that in uveal melanoma, SF3B1 mutations are also associated with alternative splicing (Fig. 2D). We show that, in common with CLL, SF3B1 mutations in uveal melanoma are associated with alternative splicing of the 3′ end of transcripts, for example in UQCC, which encodes the ubiquinol-cytochrome c reductase complex chaperone, a protein implicated in bone development and stature. We also see differential splicing of ABCC5, a multidrug resistance-associated protein that is implicated in breast cancer metastasis (24) and colorectal cancer (25). Interestingly, in uveal melanoma, we observe evidence of intron retention in ABCC5 in the SF3B1–wild-type samples compared with the mutant samples, suggesting that this gene is more efficiently spliced in the SF3B1-mutant than wild-type tumors.
More intriguing, we show that SF3B1 mutations are associated with cryptic alternative splicing of exon 4 of CRNDE transcripts (NR_034105 and NR_034106). This long noncoding RNA exists in several alternatively spliced forms and is upregulated in both solid tumors and leukemias (26, 27). CRNDE in general, and exon 4 in particular, is alternatively spliced in colorectal cancer and the alternative forms are thought to regulate gene expression by regulating chromatin-modifying enzymes (27, 28). We show that in uveal melanoma, mutations in SF3B1 are associated with cryptic alternative splicing within exon 4 of CRNDE, and, considering the comparatively simple genetics of uveal melanoma, it will be important to determine how alternative splicing of this noncoding gene affects cellular function.
In conclusion, we show that despite its appalling prognosis, uveal melanoma is a relatively simple genetic disease characterized by recurrent chromosomal losses and gains, and a low mutational burden. We confirm that GNAQ/GNA11 is the most commonly mutated oncogene and BAP1 the most commonly mutated tumor suppressor. We identify SF3B1 mutations in approximately 15% of cases and show these are associated with better prognosis, which will guide clinical management of this disease. Intriguingly, we show that SF3B1 mutations are associated with diverse alternative splicing events, including alternative terminal exon usage, intron retention, and cryptic splicing within exons of both protein coding and noncoding genes. Future studies will focus on how these events affect uveal melanoma biology.
Twelve patients with uveal melanoma were included in the WGS study (Supplementary Table S1) from whom tumor and matched blood samples were obtained. This study was approved by the ethics committee of Institut Curie (Paris, France), and informed consent was obtained from all subjects.
Consecutive patients diagnosed at the Institut Curie between January 2006 and December 2008 who underwent primary enucleation and with sufficient material at the Biobank were included in the validation series with exclusion of patients with metastasis at diagnosis. Patient characteristics are reported in Supplementary Table S6. The follow-up for this analysis ended in December 2012, with a median time of 38 months. During this period, 57 patients (53%) developed metastatic disease and 50 patients (47%) died. Metastatic melanoma was the cause of the death in 43 patients (40%). DNA was extracted from frozen materials or formalin-fixed paraffin-embedded (FFPE) sections. SF3B1, GNAQ, GNA11, and BAP1 were sequenced by Sanger methods. Oligonucleotide primer sequences are available upon request.
DNA and RNA Extraction and Sequence Analysis
Tumor DNA and RNA were provided by the Biological Resource Center of the Institut Curie. The DNA was extracted from frozen tumor or FFPE samples using a standard phenol/chloroform procedure. The total RNA was isolated from frozen tumor samples using TRIzol reagent, and cDNA synthesis was conducted with MuLV Reverse Transcriptase in accordance with the manufacturers' instructions (Invitrogen), with quality assessments conducted on an Agilent 2100 bioanalyzer. For Sanger sequencing, genomic DNA was amplified by PCR and the products were sequenced using dye-terminator chemistry as previously described (16). Primer sequences are available upon request. Sequences were visualized using Sequencher software.
Whole-Genome Sequencing and Analysis
Extracted DNA samples were sequenced on three lanes of Illumina Hiseq2000 sequencers to produce paired-end reads of 100 bp. FASTQ files from each lane were aligned to the human reference genome (GRCh37). Data for each sample were merged and duplicate reads were marked using Picard (http://picard.sourceforge.net/). Somatic variants were identified by comparing matched tumor and normal genomes (Supplementary Methods).
SNP Array Analysis
Genome-wide genotyping for tumor and blood DNA samples was conducted on Illumina HumanOmni2.5 SNP arrays. Raw data files were processed using GenomeStudio and somatic alterations were identified (Supplementary Methods).
RNA Extraction and Array Hybridization
Total RNA was isolated from frozen biopsy using a miRNeasy Mini Kit (Qiagen) and quality assessment was conducted using RNA 6000 Nano labchip (Bioanalyzer, Agilent) and by a Nanodrop spectrophotometer (Thermo). Total RNA integrity number values were between 7.7 and 9 (average: 8.65). Affymetrix Human Transcriptome Array 2.0 ST arrays were hybridized according to Affymetrix recommendations using the Ambion WT protocol (Life Technologies) and Affymetrix labeling and hybridization kits. One hundred nanograms of total RNA were processed in parallel with an external MAQC A RNA to control robustness of data. Labeled DNA mean yield was 7.19 μg (min: 6.27 μg; max: 7.57 μg). Affymetrix GeneChip Human Transcriptome 2.0 ST microarrays (HTA2) were hybridized with 4.7 μg of labeled DNA. Raw data, transcript data, and exon data were controlled with Expression console (Affymetrix) at the Institut Curie microarray core facility. The benefit of this array is to highlight spliced RNA isoforms using both exon and exon–exon junction probes that can measure excluded or included exons/regions.
Microarray Data Analysis
Affymetrix HTA2 dataset analysis was conducted by GenoSplice technology (www.genosplice.com and Supplementary Methods). We conducted an unpaired Student t test to compare gene intensities between SF3B1–wild-type and SF3B1-mutated tumors. Genes were considered significantly differentially expressed when fold change was ≥1.5 and P ≤ 0.05 (unadjusted P). Analysis at the splicing level was first conducted taking into account only exon probes (“EXON analysis”; see Supplementary Methods). Results were considered statistically significant for unadjusted P values ≤0.05 and fold changes ≥1.5 for SPLICING PATTERN analysis and unadjusted P values ≤0.05 and fold changes ≥ 2.0 for EXON analysis. After bioinformatics analysis of microarray data, a manual inspection using the GenoSplice EASANA interface was conducted to select high-confident events.
RNA from the 12 tumor samples was sequenced (Supplementary Methods). Reads were aligned using Tophat (29). Differential splicing analysis between the mutant (n = 3) and wild-type SF3B1 (n = 9) samples was conducted using DEXseq (19), and events with a false discovery rate (FDR) of <0.1 were regarded as significant. Differential splicing analysis was also conducted by MATS (20), using the mapped read bamfiles as input, and events with an FDR <0.1 were regarded as significant. The RNA-seq data from the study conducted by Harbour and colleagues (9) were downloaded from the NCBI Sequence Read Archive (http://www.ncbi.nlm.nih.gov/sra?term=SRA062359) and analyzed using the same methodology, with the exception that read lengths were trimmed to 99 bp. One of the samples described as wild-type SF3B1 was predicted to have an SF3B1 R625C mutation (Supplementary Fig. S7) and was designated SF3B1 mutant for the differential splicing analysis.
Splice Variant Analysis in the Validation Series
The validation set was used to measure the predicted splice variant of eight genes using specific probes (Supplementary Table S12). A total of 1.3 ng of cDNA were analyzed in duplicate to quantify spliced and unspliced forms by real-time PCR. Forty-five cycles of quantitative PCR were conducted in 384-well plates using QuantiTect SYBR Green reagents (Qiagen) on the ABI 9700HT device. To perform the splice variant analysis, three steps were performed per gene of interest. First, Ct values were averaged per sample, then a ratio of spliced form was calculated per sample using the formula 2−(Ct form1 mRNA − Ct form2 mRNA). Finally, for each splicing event, a Mann–Whitney U test was applied between SF3B1-mutated and wild-type cases.
Cancer-specific survival was calculated from the date of diagnosis to death from uveal melanoma or last follow-up. Event-free survival (EFS) was calculated from the surgical resection to development of metastasis or last follow-up. Survival curves were constructed using the Kaplan–Meier method and the difference between groups was compared with the log-rank test. Chi-square and Fisher exact tests were used to determine association between variables. P values less than 0.05 (two-sided) were considered statistically significant.
Whole-genome, RNA-seq, and SNP array data have been submitted to the European Genome-phenome Archive under study accession number EGAS00001000472.
Disclosure of Potential Conflicts of Interest
R. Marais has received honoraria from the speakers' bureau of Roche, has ownership interest (including patents) in The Institute of Cancer Research, and is a consultant/advisory board member of Novartis, Servier, and GlaxoSmithKline. No potential conflicts of interest were disclosed by the other authors.
Conception and design: S.J. Furney, M. Pedersen, S. Turajlic, S. Roman-Roman, M.-H. Stern, R. Marais
Development of methodology: S.J. Furney, M. Pedersen, D. Gentien
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Pedersen, D. Gentien, A.G. Dumont, A. Rapinat, L. Desjardins, S. Piperno-Neumann, S. Roman-Roman, R. Marais
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.J. Furney, M. Pedersen, D. Gentien, A.G. Dumont, S. Piperno-Neumann, P. de la Grange, S. Roman-Roman, R. Marais
Writing, review, and/or revision of the manuscript: S.J. Furney, M. Pedersen, A.G. Dumont, L. Desjardins, S. Piperno-Neumann, S. Roman-Roman, M.-H. Stern, R. Marais
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Gentien, A.G. Dumont, L. Desjardins, R. Marais
Study supervision: M.-H. Stern, R. Marais
This work was supported by Cancer Research UK (ref: C107/A10433; C5759/A12328), the Wenner-Gren Foundations, Stockholm, Teggerstiftelsen; The Harry J Lloyd Charitable Trust; and The French National Cancer Institute (INCa). A.G. Dumont is supported by the INCa grant “MeluGene.”
The authors thank Dorine Bellanger and Aurore Rampanou for technical assistance, Aurélie Kamoun for initial splicing analysis, and Odette Mariani for managing tumor samples.
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).
- Received July 1, 2013.
- Revision received July 12, 2013.
- Accepted July 12, 2013.
- ©2013 American Association for Cancer Research.