Purchase this article with an account.
Wojtek Drabarek, Serdar Yavuzyigitoglu, Askar Obulkasim, Job van Riet, Kyra N. Smit, Natasha M. van Poppelen, Jolanda Vaarwater, Tom Brands, Bert Eussen, Robert M. Verdijk, Nicole C. Naus, Hanneke W. Mensink, Dion Paridaens, Eric Boersma, Harmen J. G. van de Werken, Emine Kilic, Annelies de Klein, for the Rotterdam Ocular Melanoma Study Group; Multi-Modality Analysis Improves Survival Prediction in Enucleated Uveal Melanoma Patients. Invest. Ophthalmol. Vis. Sci. 2019;60(10):3595-3605. doi: https://doi.org/10.1167/iovs.18-24818.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Uveal melanoma (UM) is characterized by multiple chromosomal rearrangements and recurrent mutated genes. The aim of this study was to investigate if copy number variations (CNV) alone and in combination with other genetic and clinico-histopathological variables can be used to stratify for disease-free survival (DFS) in enucleated patients with UM.
We analyzed single nucleotide polymorphisms (SNP) array data of primary tumors and other clinical variables of 214 UM patients from the Rotterdam Ocular Melanoma Study (ROMS) cohort. Nonweighted hierarchical clustering of SNP array data was used to identify molecular subclasses with distinct CNV patterns. The subclasses associate with mutational status of BAP1, SF3B1, or EIF1AX. Cox proportional hazard models were then used to study the predictive performance of SNP array cluster-, mutation-, and clinico-histopathological data, and their combination for study endpoint risk.
Five clusters with distinct CNV patterns and concomitant mutations in BAP1, SF3B1, or EIF1AX were identified. The sample's cluster allocation contributed significantly to mutational status of samples in predicting the incidence of metastasis during a median of 45.6 (interquartile range [IQR]: 24.7–81.8) months of follow-up (P < 0.05) and vice versa. Furthermore, incorporating all data sources in one model yielded a 0.797 C-score during 100 months of follow-up.
UM has distinct CNV patterns that correspond to different mutated driver genes. Incorporating clinico-histopathological, cluster and mutation data in the analysis results in good performance for UM-related DFS prediction.
This PDF is available to Subscribers Only