Early genetic evolution of driver mutations in uveal melanoma

Abstract

Uveal melanoma (UM) is an aggressive eye cancer that frequently results in metastatic death despite successful primary tumor treatment. Subclinical micrometastasis is thought to occur early, when tumors are small and difficult to distinguish from benign nevi. However, the early genetic evolution of UM is poorly understood, and biomarkers for malignant transformation are lacking. Here, we perform integrated genetic profiling of 1140 primary UMs, including 131 small tumors. A clinically available 15-gene expression profile (15-GEP) prospectively validated by our group is more accurate than driver mutations for predicting patient survival. Small tumors are significantly more likely to be in earlier stages of genetic evolution than larger tumors. Further, the 15-GEP support vector machine discriminant score predicts small tumors undergoing transformation from low-risk Class 1 to high-risk Class 2 profile. These results shed light on the early genetic evolution of UM and move us closer to a molecular definition of malignant transformation in this cancer type.

Fig. 1. Genetic landscape of uveal melanomas.

a Oncoprint of 1140 primary uveal melanomas, demonstrating the 7 canonical uveal melanoma associated mutations (UMAMs), along with 15-GEP status, PRAME status, tumor thickness (millimeters), tumor diameter (millimeters), gender, metastatic status (yes or no), and survival status (alive or deceased). Pie charts summarizing variant types for BAP1, SF3B1, and EIF1AX mutations b for all samples with at least one mutation (n = 836 tumors) and c for BAP1 mutations in Class 1 (n = 25 tumors) and Class 2 tumors (n = 339 tumors). Significance was calculated by two-tailed Fisher’s exact test. d Connectivity plot indicating co-occurring mutations, with connector color representing G_q mutation (blue, GNAQ; mauve, GNA11; purple, PLCB4; yellow, CYSLTR2), and connector thickness corresponding to the number of cases. Dashed lines indicate ≤2 cases. Variant types described in the “Methods” section, and relevant data are provided in the Source data file. 15-GEP 15-gene expression profile, PRAME(+) PRAME positive, PRAME (−) PRAME negative, Diam tumor diameter, Thick tumor thickness, CB ciliary body, D-score 15-GEP support vector machine discriminant score.

Fig. 2. Comparison of cancer cell fraction and 15-GEP discriminant score in small versus larger uveal melanomas.

a Scatter plot displaying the distribution of tumor thickness and diameter for 131 small tumors (yellow dots) versus 1009 larger tumors (green dots). b Raincloud plot of TP-corrected VAF for each BSE mutation in small (n = 81 tumors) versus larger tumors (n = 761 tumors). c Raincloud plot of cancer cell fraction (CCF) for each BSE mutation in small (n = 76 tumors) versus larger tumors (n = 696 tumors). d Box plot comparing the 15-GEP discriminant score for small tumors (yellow boxes) versus larger tumors (green boxes), comparing Class 1 (n = 716 tumors), Class 2 (n = 424 tumor) and all tumors (n = 1140 tumor). For box plots in (b–d), the box center line, lower boundary, and upper boundary display the median, 25th percentile, and 75th percentile, respectively. The distance between box boundaries reflects the interquartile range (IQR). Lower whiskers extend to the minima, or the lowest value up to 1.5 times the IQR from the lower box boundary. Upper whiskers extend to the maxima, or the highest value up to 1.5 times the IQR from the upper box boundary. Continuous variables were compared by two-tailed Wilcoxon rank-sum test. Relevant data are available within the Source data file. BSE, mutation in BAP1, SF3B1, or EIF1AX; CCF_BSE, cancer cell fraction for each BSE mutation, TP tumor purity, VAF variant allele frequency, mm millimeter.

Fig. 3. Insights into early genetic evolution from 15-GEP discriminant score and BSE cancer cell fraction.

a Typical Class 1 tumors with high discriminant scores: Case #003-025, near-clonal SF3B1 mutation; Case #007-071, sub-clonal EIF1AX mutation. b Discordant Class 1 tumors with high discriminant score: Case #027-106, near-clonal EIF1AX mutation and subclonal BAP1 mutation. c Discordant Class 1 tumors with low discriminant scores: Case #026-504, clonal BAP1 and SF3B1 mutations and partial LOH3p involving a limited region around the BAP1 locus; Case #017-119, clonal SF3B1 mutation and LOH3p; Case #028-183, near-clonal BAP1 mutation, LOH3p and very low discriminant score (0.02). d Class 2 tumors with high discriminant scores and bi-allelic BAP1 loss: Case #021-013, sub-clonal BAP1 mutation; Case 023-059, near-clonal EIF1AX mutation and subclonal BAP1 mutation. e Class 2 tumors with intermediate discriminant scores: Case #026-435, near-clonal SF3B1 mutation and LOH3p but no detectable BAP1 mutation; Case #019-019, sub-clonal BAP1 mutation and no detectable LOH3p. f Class 2 tumors with low discriminant scores: Case #017-035, near-clonal EIF1AX and BAP1 mutations with LOH3p; Case #021-116, near-clonal BAP1 mutation with no detectable LOH3p. The length of the connector between the ancestor melanoma cell (gray circle) and the melanoma (blue or red circle for Class 1 or Class 2, respectively) is proportional to tumor diameter (in millimeters) at the time of tumor sampling. The length of the extension beyond the melanoma is proportional to time to death or last follow-up (in months) with final status indicated. Survival status for living and dead patients is indicated by gray or black bar, respectively. Mutation nomenclature is described in the “Methods” section. Relevant data are available in the Source data file. 15-GEP 15-gene expression profile, ANM alive no metastasis, DOM dead of metastasis, D-score support vector machine discriminant score, LOH3p loss of heterozygosity of chromosome 3p; (−) PRAME negative, (+) PRAME positive.

Fig. 4. Features of BAP1 mutations by 15-GEP Class status.

a Raincloud plot depicting tumor diameter in relation to 15-GEP Class and BAP1 mutation status (n = 1140 tumors). b Raincloud plot depicting tumor thickness in relation to 15-GEP Class and BAP1 mutation status (n = 1140 tumors). c Box plot comparing raw discriminant scores by 15-GEP Class and BAP1 allelic dosage reflected in BAP1 mutation and LOH3p status (n = 905 tumors). Box center line, lower boundary, and upper boundary for box plots in (a–c) represent the median, first quartile, and third quartile, respectively, while box range reflect the interquartile range (IQR). Lower whiskers extend to the minima, or the lowest value up to 1.5 times the IQR from the first quartile. Upper whiskers extend to the maxima, or the highest value up to 1.5 times the IQR from the third quartile. Survival analysis plots displaying the d metastasis-free survival and e overall survival probabilities for Class 1 (n = 715 tumors) and Class 2 (n = 418 tumors) UM according to absolute discriminant score at specified time points including 12 (red curves), 24 (light blue curves), 36 (green curves), 48 (dark blue curves), and 60 (orange curves) months. Significance for continuous variables was determined by two-tailed Wilcoxon rank-sum test. Significance for survival analysis was calculated by Cox proportional hazard analysis by Wald test. All data are available in the Source data file. Exact p values for thickness and diameter in Class 2/BAP1^mut versus Class 1/BAP1^wt were 4.8 × 10⁻²⁹ and 1.8 × 10⁻¹⁶, respectively, and in Class 2/BAP1^mut versus Class 1/BAP1^mut were 2.6 × 10⁻¹⁰ and 3.8 × 10⁻⁷, respectively. 15-GEP 15-gene expression profile, MFS metastasis-free survival, OS overall survival, BAP1^wt BAP1 wildtype, BAP1^mut BAP1 mutant.

Fig. 5. Hypothesis for relationship between BAP1 dosage, tumor immune microenvironment, and discriminant score.

BAP1 dosage decreases as BAP1-deficient tumor cells outcompete BAP1-wildtype tumor cells, leading to altered composition of infiltrating immune cells in the tumor immune microenvironment (TIM). Since the 15-GEP includes genes expressed in tumor cells, immune cells or both, inversion of the SVM discriminant score from the Class 1 side to the Class 2 side of decision boundary occurs progressively as the transcriptional effects of BAP1 loss accrue in both tumor and immune cells. This would explain why there is not a strict association between the fraction of cancer cells harboring mutant BAP1 (CCF_BAP1) and the discriminant score, as the rate at which the TIM changes following BAP1 loss may differ between individuals. This would also explain why transitional tumors with low discriminant score tend to be small, whereas larger tumors, which have had longer for these transcriptional changes to occur, tend to have high discriminant scores.