Dataset DOI: 10.5061/dryad.z8w9ghxqk

Description of the data and file structure

Files and variables

File: DryadUpload_SupplementaryInformationSData1-4_v14.xlsx

Description: The "Description" tab outlines contents, along with important abbreviations and details for the Supplementary Data. The tabs following the Description tab include Supplementary Data 1-4 (large data tables from the manuscript). Supplementary Data 1 includes a table of baseline clinical annotations, patient outcomes, and genetic features for all study subjects (n = 1140 cases). Supplementary Data 2 includes statistical analysis of uveal melanoma-associated mutations for all study subjects (n = 1140 cases). Supplementary Data 3 includes statistical analysis of small tumors (n = 131 cases) versus larger tumors (n = 1009 cases). Supplementary Data 4 includes statistical analysis of patients with BAP1-mutant tumors (n = 364) versus those with BAP1-wildtype tumors (n = 776).

Variables

• Includes Supplementary Data 1, 3, 6, and 7.

File: DryadUpload_SourceData_v4.xlsx

Description: Source Data file contains tables of raw data used to generate all plots in main and Supplementary Figures. The "Description" tab describes each dataset and includes additional column descriptions that may be helpful for understanding included data. Exact p-values included for all significance testing displayed in figures. Order of tables by figure (panels): Figure 1a, Figure 1b, Figure 1c, Figure 1d, Figure 2a, Figure 2b, Figure 2c, Figure 2d, Figure 3a-f, Figure 4a, Figure 4b, Figure 4c, Figure 4d, Figure 4e, Supplementary Figure 2, Supplementary Figure 3, Supplementary Figure 4, Supplementary Figure 5, Supplementary Figure 6, Supplementary Figure 7

Variables

• Excel file with each spreadsheet containing source data for data presented in manuscript.

File: forcedCallVCFs_chr3p_coog2.2_998pts.zip

Description: The dataset consists of a directory of forced called VCF files from targeted hybrid capture of each patient within the cohort for copy number calls of BAP1 (n=998).

Variables

• ZIP folder of VCF files.

Patient Enrollment

This research complies with all relevant ethical regulations. Federal Wide Assurance (FWA) from the Office of Human Research Protections (OHRP) and Institutional Review Board (IRB) or Ethics Committee approval was obtained in accordance with policies at each participating center, with oversight by the University of Miami IRB. Between January 2017 and April 2020, COOG2 enrolled 1687 subjects with UM involving the choroid, ciliary body and/or iris across 26 ocular oncology centers in the U.S. and Canada and prospectively monitored these subjects for metastatic progression and outcome. Informed written consent was obtained from each patient. Primary treatment was performed according to the standard at each center. Exclusion criteria included patient age less than 18 years, diagnosis of a uveal tumor other than UM (e.g., metastatic cancer), prior radiotherapy, inadequate sample for molecular analysis, and patient withdrawal from the study. Prior photodynamic therapy or transpupillary thermotherapy was allowed if there was evidence of tumor regrowth. No participants were excluded based on sex, ethnicity, or race. Gender was recorded from medical records and used as a proxy for biological sex in this study. No additional data on gender identity was collected. For this analysis, a data lock was performed on March 4, 2024. Subjects were not included for this report if they had a primary iris melanoma (n = 101 cases), lacked adequate residual biopsy material for successful sequencing (n = 212 cases) or had no detectable UMAM (UM-associated mutations) (n = 234 cases).

Description of Data

Data includes prognostic testing results of primary UM from DecisionDx®-UM 15-gene expression profile (15-GEP), which renders a result of Class 1 or Class 2, and DecisionDx®-PRAME, which renders a result of negative or positive. The 15-GEP test employs support vector machine (SVM) to classify each sample as Class 1 or Class 2 and to assign a discriminant score as a measure of confidence in the Class call based on the distance of a given sample to the decision boundary.

The residual material was analyzed with the custom UMAM NGS (next-generation sequencing) panel DecisionDx®-UMSeq. Genetic variants were included that were classified as tier I, II, or III according to the guidelines of the College of American Pathologists (CAP), American Society of Clinical Oncology (ASCO), and Association for Molecular Pathology (AMP). This testing was performed in a CAP-accredited, CLIA-certified clinical laboratory (Castle Biosciences, Inc., Friendswood, TX, USA). Mutations were classified as nonsense (introduction of a premature stop codon), stop-loss or start-loss (loss of stop or start codon preventing translation), frameshift insertion or deletion (shift of codon reading frame via addition or subtraction of a non-triplet set of nucleotides), non-frameshift insertion or deletion (addition or removal of a codon without shifting the reading frame), block substitution (alteration of multiple sequential codons), splice site alteration (alteration of splice donor or acceptor site), missense (substitution of one amino acid).The following BAP1 variants were called pathogenic: nonsense, stop-loss, start-loss, frameshift and non-frameshift insertions and deletions, block substitutions, and splice site alterations. The remaining (predominantly missense) variants were called pathogenic if they: (1) were previously reported as pathogenic in the ClinVar Database 32, (2) exhibited a SIFT score less than or equal to 0.05, or (3) exhibited a PolyPhen2 score greater than or equal to 0.5.

Tumor purity (TP), the percentage of cells in a sample that are tumor cells, was inferred from the variant allele frequency (VAF) of the Gq mutation, assuming that the Gq mutation is the founder mutation, is a heterozygous alteration, and is therefore present at 50% VAF in tumor cells. In rare cases with more than one Gq mutation, the mutation with the highest frequency (and presumably the earlier initiating mutation) was used. As such, TP=min([VAF-Gq-mutant x 2],100%). The variant allele fraction (VAF) for BAP1, SF3B1, and EIF1AX mutations was corrected for TP using the following equation: TP-corrected VAF-BSE = VAF-BSE ÷ TP. Samples without a detectable Gq mutation could not be corrected for VAF and, thus, were not included in analyses requiring TP-corrected VAF-BSE. Next, we estimated the cancer cell fraction (CCF) for each BSE mutation, representing the proportion of UM cells that harbor a given mutation, which requires a correction for allelic copy number. SF3B1 is located on chromosome 2, which is not frequently altered in UM20,22. Thus, SF3B1 mutations were assumed to be heterozygous and CCF-SF3B1 = min (TP-corrected VAF-SF3B1 x 2, 100%). EIF1AX is located on the X chromosome, which is also rarely lost in UM. Thus, gender was used to calculate mutant CCF-EIF1AX, where females were assumed to have an EIF1AX mutation at 50% and males at 100% of TP-corrected VAF. Thus, the CCF-EIF1AX for females was calculated as CCF-EIF1AX=min(TP-corrected VAF-EIF1AX x 2, 100%), whereas the CCF-EIF1AX for males was assumed to be equal to TP-corrected VAF-EIF1AX. BAP1 is located at chromosome 3p2133, which frequently undergoes copy number loss in UM. To detect loss of heterozygosity (LOH) and calculate CCF for BAP1, we developed a custom targeted CNV sequencing panel containing 74 loci across chromosome 3p that was performed on the same sample used for the 15-GEP/PRAME classifier and UMAM NGS panel (description below). For BAP1-mutant tumors with retention of heterozygosity for chromosome 3p, the CCF-BAP1 was calculated as CCF-BAP1=min(TP-corrected VAF-BAP1 x 2, 100%). For tumors demonstrating LOH for chromosome 3p (LOH3p), CCF-BAP1 was assumed to be equal to TP-corrected VAF-BAP1.

For the custom CNV sequencing panel, B-allele frequencies and log fold-change (lfc) read depths across chromosome 3p were compared to a reference DNA panel of normals (PON), comprising peripheral blood mononuclear cell (PBMC) samples from 64 patients. Variant call format (VCF) files were analyzed using Wheeljack (https://github.com/covingto/KRCGTK/releases/tag/v0.1). Copy-number loss for chromosome 3p was detected by consistent b-allele frequencies at 100% and a decreased lfc read depth of less than 0. Isodisomy for chromosome 3p was identified by consistent b-allele frequencies at 100% and a lfc read depth of approximately 0. For downstream analyses, samples demonstrating either copy number loss or isodisomy for chromosome 3p were called as LOH3p, whereas samples without these aberrations were called as retention of heterozygosity for 3p. Calls were made by hand and adjudicated by 3 authors. Variability across b-allele and read depth plots was used to assign confidence scores with 0, 1, 2, and 3 corresponding to very low, low, medium, and high confidence, respectively. A confidence score of 2 or 3 was required for use in downstream analyses.