Targeted Next-Generation Sequencing of 117 Routine Clinical Samples Provides Further Insights into the Molecular Landscape of Uveal Melanoma.

Uveal melanoma (UM) has well-characterised somatic copy number alterations (SCNA) in chromosomes 1, 3, 6 and 8, in addition to mutations in GNAQ, GNA11, CYSLTR2, PLCB4, BAP1, SF3B1 and EIF1AX, most being linked to metastatic-risk. To gain further insight into the molecular landscape of UM, we designed a targeted next-generation sequencing (NGS) panel to detect SCNA and mutations in routine clinical UM samples. We compared hybrid-capture and amplicon-based target enrichment methods and tested a larger cohort of primary UM samples on the best performing panel. UM clinical samples processed either as fresh-frozen, formalin-fixed paraffin embedded (FFPE), small intraocular biopsies or following irradiation were successfully profiled using NGS, with hybrid capture outperforming the PCR-based enrichment methodology. We identified monosomy 3 (M3)-UM that were wild-type for BAP1 but harbored SF3B1 mutations, novel frameshift deletions in SF3B1 and EIF1AX, as well as a PLCB4 mutation outside of the hotspot on exon 20 coinciding with a GNAQ mutation in some UM. We observed samples that harboured mutations in both BAP1 and SF3B1, and SF3B1 and EIF1AX, respectively. Novel mutations were also identified in TTC28, KTN1, CSMD1 and TP53BP1. NGS can simultaneously assess SCNA and mutation data in UM, in a reliable and reproducible way, irrespective of sample type or previous processing. BAP1 and SF3B1 mutations, in addition to 8q copy number, are of added importance when determining UM patient outcome.


Introduction
Uveal melanoma (UM), the most common primary intraocular malignancy in adults, has an incidence of 3-8 individuals per million per year in Caucasians [1,2]. Despite successful treatment of cases were selected, as they were taken post-irradiation with either ruthenium plaque radiotherapy (PRXT) or proton beam radiotherapy (PBR) (Figure 1). All samples had previously undergone routine genetic testing by either multiplex ligation dependent probe amplification (MLPA) or microsatellite analysis (MSA).

Panel Comparison (14 Samples)
Of the initial 14 UM samples analysed for panel comparison, 1/14 (7%) and 3/14 (21%) failed to produce reportable SCNA data with the SureSelect (SureSelect XT HS using SureDesign, Agilent) and TSCA (TruSeq Custom Amplicon using DesignStudio Illumina) panels, respectively. 13/14 (93%) UM samples had available SCNA data from previous MLPA for chr1, 3, 6 and 8; the remaining sample was tested by MSA for chr3 status only. There was 100% agreement for chr3 status between the MLPA/MSA data and that provided by both NGS tests in this initial sample cohort (Table S1-samples marked by an asterisk). There was 100% concordance for GNAQ, GNA11, BAP1, SF3B1 and EIF1AX mutations between both testing platforms. No false positives were detected in any of the samples. Of note, 6/14 UM test samples had been previously submitted by our group to the TCGA-UM study, and there was also 100% concordance for all mutations identified. The SureSelect panel was chosen to test the larger UM cohort, due to its greater success rate in SCNA analysis and better coverage (Table S2).

Cox Regression
Univariate analysis was carried out using a significance level of p < 0.005 after Bonferroni correction. Factors significantly associated with survival were: epithelioid cytomorphology, LBD, UH, ciliary body involvement, BAP1 and chr3 status ( Table 2). These variables were entered into the Cox model and backward selection of covariates was carried out using the likelihood ratio to determine 'goodness of fit' of the model. At the 0.01 significance level, chr3 loss was significantly associated with reduced survival (p ≤ 0.001) with a hazard ratio of 5.949 (Table 3).

Survival
Kaplan-Meier survival curves and tables were examined for all primary UM stratified according to: chr3 status, extra copies of chr8q, and mutations in BAP1 and SF3B1. The following were significantly associated with a reduced survival time: loss of chr3 (Log Rank p < 0.001), BAP1 mutations (Log Rank p < 0.001), M3-UM with more than two copies of 8q (Log Rank p = 0.014) and D3-UM with SF3B1 mutations (Log Rank p = 0.027) ( Figure 2).

BAP1 IHC
Seventy of the ninety surgical UM samples (enucleation/local resection) had previously undergone routine immunohistochemistry (IHC) to determine nuclear BAP1 (nBAP1) protein expression; the remaining samples did not have enough material for subsequent IHC analysis. nBAP1 protein was absent in 38/70 cases (54%) of which 31 (82%) UM also had mutations in the BAP1 gene. Of the 7/38 (18%) UM with no BAP1 mutations, four patients had M3-UM and three had died from metastatic disease. Furthermore, 3/32 (9%) UM positively expressed nBAP1 protein but had clear mutations in BAP1, all of which were missense alterations (q.Glu31Lys, q.Cys91Gly and q.Ala142Pro). Cancers 2020, 12, x 7 of 16

BAP1 IHC
Seventy of the ninety surgical UM samples (enucleation/local resection) had previously undergone routine immunohistochemistry (IHC) to determine nuclear BAP1 (nBAP1) protein expression; the remaining samples did not have enough material for subsequent IHC analysis. nBAP1 protein was absent in 38/70 cases (54%) of which 31 (82%) UM also had mutations in the BAP1 gene. Of the 7/38 (18%) UM with no BAP1 mutations, four patients had M3-UM and three had died from metastatic disease. Furthermore, 3/32 (9%) UM positively expressed nBAP1 protein but had clear mutations in BAP1, all of which were missense alterations (q.Glu31Lys, q.Cys91Gly and q.Ala142Pro).

SF3B1 Mutations in M3 UM
SF3B1 mutations have previously been associated with D3-UM with late onset metastasis [22]. In our cohort, 5/25 cases (20%) with SF3B1 mutations died of metastatic UM at the time of study closure. Of these five cases, four tumors were D3-UM and one was a M3-UM with a BAP1 mutation. To investigate the prevalence of SF3B1 mutations in M3-UM that lacked mutations in BAP1, we identified 20 additional cases of M3-UM where DNA was available and previous IHC analysis had demonstrated strong nBAP1 positivity, correlating with wild-type BAP1 [28]. This additional UM cohort consisted of 12 males and 8 females with a mean age of 62 years at primary management (median age 62; range 45-80 years). The mean follow-up period was 48 months (median 61 months; range 6-79 months). Primary management was enucleation 17/20 (85%) and local resection 3/20 (15%). The mean LBD was 14.8 mm (median LBD 14.7; range 9.8-22.7 mm) with a mean UH of 8.0 mm (median UH 8.4; range 1.7-12.4 mm). Full histological assessment is detailed in Table S4. Of these

SF3B1 Mutations in M3 UM
SF3B1 mutations have previously been associated with D3-UM with late onset metastasis [22]. In our cohort, 5/25 cases (20%) with SF3B1 mutations died of metastatic UM at the time of study closure. Of these five cases, four tumors were D3-UM and one was a M3-UM with a BAP1 mutation. To investigate the prevalence of SF3B1 mutations in M3-UM that lacked mutations in BAP1, we identified 20 additional cases of M3-UM where DNA was available and previous IHC analysis had demonstrated strong nBAP1 positivity, correlating with wild-type BAP1 [28]. This additional UM cohort consisted of 12 males and 8 females with a mean age of 62 years at primary management (median age 62; range 45-80 years). The mean follow-up period was 48 months (median 61 months; range 6-79 months). Primary management was enucleation 17/20 (85%) and local resection 3/20 (15%). The mean LBD was 14.8 mm (median LBD 14.7; range 9.8-22.7 mm) with a mean UH of 8.0 mm (median UH 8.4; range 1.7-12.4 mm). Full histological assessment is detailed in Table S4. Of these additional 20 UM, 5 (25%) had mutations in SF3B1; 3/5 (60%) q.Arg625Cys and 2/5 (40%) q.Arg625His. At study closure, all five patients were alive; of interest, one patient developed liver metastases 40 months after primary management but underwent metastasectomy and is still alive 25 months after surgery.

Discussion
This is the largest study to date to profile UM using bespoke targeted NGS panels. It identified chr3 as the most significant factor associated with metastatic death and demonstrated for the first time that irradiated UM samples can be successfully profiled using NGS with no observable differences in quality when compared to non-irradiated UM samples. We identified a subset of M3-UM-patients without nBAP1 loss that demonstrated mutations in SF3B1 and also describe concurrent disruptive frameshift deletions in SF3B1 and EIF1AX. This is consistent with the observation in one case sequenced in TCGA that harboured both an EIF1AX and an atypical SF3B1 (T663P) mutation [11]. We also observed co-occurring mutations in BAP1 and SF3B1 and EIF1AX and SF3B1. Novel mutations were also identified in TTC28, KTN1, CSMD1 and TP53BP1. Of interest, we identified a mutation in PLCB4 that does not fall within the hotspot on exon 20 and coincides with a GNAQ mutation. Furthermore, chr3 results obtained using the NGS panel were comparable to previous MLPA and MSA analyses. We recommend that this bespoke NGS panel ultimately replaces MLPA/MSA testing in routine labs, with the possibility of incorporating molecular data into prognostic tools-e.g., the LUMPO (Liverpool Uveal Melanoma Prognosticator Online), which was recently externally validated in a multicentre study [29].

Enrichment Comparison
Hybrid capture and PCR-based enrichment methods in NGS vary in how targeted regions are enriched [30]. Hybrid capture methodologies like the SureSelect XT HS used in this study, involve shearing gDNA into smaller fragments, library preparation and hybridisation with targeted biotinylated RNA baits. Using magnetic streptavidin beads, these baits can be separated, and the hybridised library amplified; whilst PCR-based methods hybridise a custom oligo pool flanking-targeted regions on unfragmented gDNA. These are then extended and ligated, and PCR is performed to integrate indexes and sequencing primers. The PCR-based method has the advantages of requiring lower DNA inputs with shorter preparation times. In our study, hybrid capture outperformed the PCR-based enrichment in terms of a larger percentage of reads mapped and a greater mean depth of coverage. Although there were no differences in the ability to call single nucleotide variants (SNV), there was an increased SCNA analysis failure rate for the PCR-based method. Similar comparison investigations in other cancer types found limited sensitivity of PCR-based sequencing, with several variants being missed due to regions of high guanine-cytosine content and suboptimal PCR conditions, yielding a minimal coverage not found when using hybrid capture [31][32][33]. An increased incidence of false positives and missed variants in PCR-based enrichment was also reported when evaluating hybrid capture versus PCR-based methods for whole-exome sequencing [34]. In contrast to our comparison, neither study found differences between the success rates of SCNA analysis.

Comparison with Previous MLPA
In the current study, we were able to successfully examine both SCNA and SNV using a single NGS assay in fresh, FFPE and also irradiated tissues. Only one sample failed to produce a clear genotype, but this was expected because of a low yield of library post-capture. Furthermore, 10/116 (9%) UM samples were discordant with the original MLPA/MSA analyses for chr3: 2 were isodisomy 3, which had been classified as D3 by MLPA due to its limitations in detecting acquired homozygosity; two were shown to have regions of deletion not identified in previous MLPA, most likely due to an increased number of probes covering chr3 on the NGS panel. Of the remaining six discordant samples, four had been classified as M3 by MLPA but as D3 by NGS; two of these cases had SF3B1 mutations but all patients were alive at the study closure. Two had been classified as D3 by MLPA but M3 by NGS; one had a BAP1 mutation and both patients had died from metastatic disease. For chr1, 6 and 8, the discordance between the MLPA and the NGS SCNA was greater at 17-26% of UM cases, which is likely a result of the low probe coverage for these chromosomes on the MLPA panel. Whilst the median 8q copy number was the same in D3-UM and M3-UM, the 8q copy number burden was generally higher in M3-UM. This was reflected by a reduced survival in M3-UM with an 8q copy number of 4 or more consistent with previous reports that 8q dosage is an important predictor of outcome in UM [11,35].

Irradiated Samples
This is the first study to examine irradiated UM samples using a NGS panel. No diminished quality or ability to genotype these tumors was observed amongst these samples. This is consistent with our findings using MSA/MLPA to genotype irradiated UM [36][37][38].

BAP1 Mutations
The frequency of BAP1 mutations in the present study was 43% in total, occurring in 82% of M3-UM; these data are consistent with the findings of others [11,14,19,25]. The presence of a BAP1 mutation in UM was associated with a worse survival. We have previously reported that nBAP1 + M3-UM have a better prognosis as compared with nBAP1 − M3-UM [21]; however, interestingly in this current study, M3-UM that were wild-type for BAP1 (10/57; 18%) did not correlate with an increased survival time as compared with M3-UM with BAP1 mutations. This may be due to either the observation that BAP1 mutations do not always correlate with loss of nBAP1 protein expression, or to the smaller cohort of patients in the present study [28,39].

SF3B1 Mutations
The frequency of SF3B1 mutations in UM ranges in the literature from 11-34% [14,25], and in this study SF3B1 mutations occurred in 21% of cases. SF3B1 mutations are reported to occur mainly in D3-UM associated with late onset metastasis and decreased survival (22). This is consistent with our study in which 20/25 (80%) SF3B1 mutations occurred in D3-UM with a significantly reduced survival time as compared with D3/SF3B1wt UM (p = 0.027).
A novel disruptive frameshift deletion in SF3B1 of 15 nucleotides was observed in p.Lys653_Ser657del on heat domain 4, outside the hotspot region of codon 625; the significance of this is unclear. Of particular interest in our study are five M3-UM or UM with PL of chromosome 3 with SF3B1 mutations. Two of these UM harboured BAP1 mutations, previously described in one other study (11); one patient succumbed to metastatic disease 12 months after primary management, and the second patient died of other causes 99 months (8.25 years) later. Three SF3B1 mutations were recorded in M3-BAP1wt UM, a phenomenon only observed in one other study to date [11]. To examine this further, we tested an additional 20 cases of M3-UM with nBAP1 positivity and identified five cases with SF3B1 mutations; at the time of study closure, all five patients were alive. Additional cases and longer follow-up are required to fully understand the clinical relevance of SF3B1 mutations in M3-UM.

EIF1AX Mutations
EIF1AX mutations were detected in the present study in 19% of UM, which is consistent with that reported by other groups [11,14,18,25]. Interestingly, two UM demonstrated mutations in both EIF1AX and SF3B1 despite previous reports describing that these occur in a mutually exclusive manner [11,25]. Of note, both patients died from metastatic disease at 34 and 58 months, respectively, after primary treatment. EIF1AX mutations are typically associated with D3-UM; however, we identified two M3-UM that displayed mutations in this gene. A novel disruptive frameshift deletion of 6 nucleotides from the coding sequence was also identified in p.Arg14_Gly15del of EIF1AX.

Initiating Mutations
Mutations in GNAQ and GNA11 occurred in 89% of UM in a mutually exclusive manner (53% and 39%, respectively), consistent with the literature [11,14,25]. Mutations predominantly occurred in exon 5 for GNAQ and GNA11, and two UM had mutations in exon 4. One sample contained two unusual mutations in exon 4 of GNA11 p.R214K and p.R214S. These regions do not lie within any of the known functional domains of GNA11 and have not been previously described; their effect on GNA11 protein function is unknown. Mutations in CYSLTR2 were found in two UM in the hot spot region p.L129Q in exon 1 and occurred in a mutually exclusive manner to mutations in GNAQ and GNA11, as previously reported [17]. Consistent with our general understanding of the function of these mutations, there were no differences in survival outcome based on the mutational status of the driver mutations GNAQ, GNA11 and CYSLTR2.
Disruptive frameshift deletions in p.M549_G556delinsI and M561_G568delinsI mutations were observed in PLCB4 in a single UM sample. These cases also showed a p.R183Q mutation in GNAQ. Previous studies identified recurrent mutations in PLCB4 in a hot-spot region p.D630Y and p.D630N on exon 20 [18]. The mutation identified in our study occurred in exon 18 and is the first mutation in this region to be described in UM. Though it was initially thought that PLCB4 mutations occurred in a mutually exclusive manner to GNAQ, GNA11 and CYSLTR2, our study and that of Robertson et al. [11] demonstrate PLCB4 mutations concurrent to GNAQ and GNA11 mutations.

Other Mutations
We observed low frequency (3%) somatic mutations in genes originally identified by Royer-Bertrand et al. (6%), namely in TTC28, CSMD1, KTN1 and TP53BP1 [14]. Most of these genes are involved in various cellular processes, e.g., cell cycle regulation [40], cell migration and proliferation [41,42], kinesin binding [43] and DNA double-strand break repair [44]. Our NGS panel was custom-designed to have full coverage of the TTC28, CSMD1, KTN1 and TP53BP1 genes, and because of its targeted nature had greater coverage in comparison to whole-exome sequencing methodologies. Due to their low frequency in this study, no association could be made between the mutations in TTC28, CSMD1, KTN1 or TP53BP1 and UM with particular clinical or morphological features. It is worth noting that previously described mutations in SRSF2, DLK2 or FBXW7 were not detected in this large study [14,20,45].

Patients
In this retrospective cohort study, primary UM samples were collected from 117 patients who were treated at the Liverpool Ocular Oncology Centre (LOOC), Liverpool University Hospitals NHS Foundation Trust, between January 2008 and May 2015. This time period was chosen to allow sufficient follow-up (median, 65 months). The follow-up period was calculated from the date of primary management to either study end (23 September 2019) or to death from metastatic disease or other causes. Patients were treated either by radiotherapy or surgical resection, and their UM was genotyped using either MLPA or MSA, as described below.

Specimen Characteristics
Specimens consisted of DNA (stored at −80 • C) previously extracted from fresh biopsies all preserved in CytoLyt (Cytyc Corp) and stored at 4 • C, fresh-tumor tissue snap-frozen in liquid nitrogen and stored at −80 • C, and FFPE UM samples stored at room temperature. Twenty-six of the DNA samples analysed were post-irradiation specimens.

Study Design
The clinical endpoint examined in this study was death from metastatic disease. Patients who died from causes other than those relating to UM were included in the study, and data for these records were treated as right-censored cases for evaluation purposes. This study conformed to the principles of the Declaration of Helsinki and Good Clinical Practice guidelines. Approval for the study was

Morphological/Histological Studies
All samples underwent routine histopathological and cytological workup assessing cell type, mitotic count, and presence of PAS+ connective tissue loops where possible (28). Furthermore, 90/117 enucleation and local resection specimens had a full histological workup, whilst 27/117 biopsies and endoresection specimens underwent cytological examination only. Additionally, IHC analysis of nBAP1 expression was undertaken in 70/117 cases, as described previously [21].

DNA Extraction and Quantification
Methods for DNA extraction from FFPE and frozen UM have been published elsewhere [46]. DNA integrity of FFPE samples was qualified by performing a qPCR using the Agilent NGS FFPE QC Kit. (Agilent Technologies Inc., Cheadle, UK).

Chromosomal SCNA Analysis
MLPA (MRC Holland, The Netherlands) and MSA were used to assess SCNA, and subsequent comparison with NGS data were undertaken during routine genetic testing of patient samples, as previously described [47,48]. Cases yielding >100 ng of DNA were tested using MLPA, whilst MSA was undertaken for UM samples with lower DNA yields.

Next-Generation Sequencing
Two custom NGS panels were designed: SureSelect XT HS using SureDesign (Agilent) and TruSeq Custom Amplicon (TSCA) using DesignStudio (Illumina). Both panels were designed to cover mutations in GNAQ (exons 4 & 5), GNA11 (exons 4 & 5), SF3B1 (exons 12 & 14), EIF1AX (exons 1 and 2), and all exons of BAP1, FBXW7, DLK2, CSMD1, CYSLTR2, KTN1, TP53BP1, SRSF2, PLCB4, TTC28 and BRAF (negative control). Both enrichment methods included the incorporation of unique molecular identifiers or barcodes to reduce errors and quantitative bias introduced by the amplification process. For the SureSelect XT HS, additional probes were included to examine SCNA in chr1: 1541 probes; chr3: 1287 probes; chr6: 1094 probes; chr8: 933. The TSCA panel included additional probes to examine SCNA in chr3: 83 amplicons; chr6: 76 amplicons and chr8: 67 amplicons. Chr1 was not included in the TSCA NGS panel due to tiling limitations. As the panels were worked up on larger resection samples, the DNA input was 50 ng for both panels. Libraries were constructed using either the SureSelect XT HS Reagent and Capture Library Kit (Agilent Technologies Inc., Cheadle, UK) or TruSeq Custom Amplicon Low Input Kit (Illumina Inc., United Kingdom), according to the manufacturer's instructions. The two panels were tested and compared using 14 frozen UM samples, 8 of which had been previously profiled by The Cancer Genome Atlas (TCGA) UM study [11], and 6 had available data from previous genotyping plus an additional two reference samples (Genome In A Bottle, HDx).
The SureSelect XT HS was subsequently selected to test a larger cohort of 95 fresh and 13 FFPE UM samples with reference samples included in each sequencing run. The DNA input varied (5 ng-25 ng) depending upon the sample type.

Sanger Sequencing
Exon 14 of SF3B1 was sequenced using PCR-based capillary Sanger sequencing in an additional twenty M3-UM with unusual nBAP1+ protein expression [21]. Oligonucleotides were constructed by Eurofins Genomics; forward 5'-GGCCGAGAGATCATTTCT-3, reverse 5'-AAGAAG GGCAATAAAGAAGGA-3', product size 289bp. PCR was performed in a reaction volume of 50 µL containing 100 ng of genomic DNA, 0.25 µL of Thermo-Start Taq DNA Polymerase (Thermo Scientific), 5 µL of HP Buffer, 4 µL of 25 mM MgCl 2 , 2 µL of dNTP (2 mM each), 31.25 µL Nuclease Free water and 1 µL of each of the primers. The thermal cycling profile was as follows: initial denaturation at 95 • C for 15 min and 35 rounds of amplification at 95 • C for 15 s, 55 • C for 30 s and 72 • C for 1 min. A final extension step at 72 • C for 5 min was added. PCR products were purified using the QIAquick PCR purification kit (Qiagen, United Kingdom) according to the manufacturer's protocol. Sequencing of PCR products was carried out by GATC at Eurofins Genomics in accordance with ISO 17025. Sequencing data were analysed using Chromas Lite (2.1.1., Technelysium Pty Ltd.).

NGS Data Analysis
NGS libraries were sequenced on the Illumina MiSeq platform (2 × 250 bp paired-end) by the Centre for Genomic Research (www.cgr.liv.ac.uk), University of Liverpool, UK. Base-calling and de-multiplexing of indexed reads were performed by CASAVA version 1.8.2 (Illumina) to produce the raw sequence data in FASTQ format. The raw FASTQ reads were trimmed to remove Illumina adapter sequences using Cutadapt version 1.2, and low-quality bases using Sickle version 1.200.

Statistical Analysis Methods
Survival time (months) was calculated from the date of primary management until death from metastases or study closure on 23 September 2019. Median survival time was estimated using the Kaplan-Meier product limit method. Univariate associations between survival time, clinical, histological and genetic features were examined using Cox proportional hazards regression models. Analyses were undertaken using SPSS Statistics v.24 (IBM), Microsoft R 3.5.1 and the packages rms, cmprsk and mstate. Cut-offs for SCNA used established values based on previous clustering analysis carried out at our centre: log rank < 0.85 loss, > 1.15 amplification [49]. The allelic frequency threshold to call a mutation was 10%.

Conclusions
Our bespoke UM NGS panel enables detailed SCNA and mutational information to be obtained from small UM biopsies, FFPE material and previously irradiated UM. This is in distinct contrast to some current methodologies, which, when applied to biopsies, can only determine chr3 status due to the low DNA yield. Moreover, consistent with other reports, BAP1 and SF3B1 mutations in addition to 8q copy number are of added importance when determining patient outcome and moves UM stratification away from a binary genetic classification based on chr3 copy number only. Identifying metastatic risk groups with greater precision than is currently possible with SCNA assessment alone will have implications on the frequency at which patients are followed up for subsequent liver imaging, and the imaging techniques applied, as well as on patient selection for clinical trials. Although at present, mutations in UM are not therapeutically actionable, it is hoped that continued advances in our understanding of this disease will result in the use of these biomarkers to predict response to emerging therapies.