Abstract

Aim: In an effort to identify patients with uveal melanoma at high risk of metastasis, the authors undertook correlation of gene expression profiles with histopathology data and tumour-related mortality.

Methods: The RNA was isolated from 27 samples of uveal melanoma from patients who had consented to undergo enucleation, and transcripts profiled using a cDNA array comprised of sequence-verified cDNA clones representing approximately 4000 genes implicated in cancer development. Two multivariate data mining techniques—hierarchical cluster analysis and multidimensional scaling—were used to investigate the grouping structure in the gene expression data. Cluster analysis was performed with a subset of 10 000 randomly selected genes and the cumulative contribution of all the genes in making the correct grouping was recorded.

Results: Hierarchical cluster analysis and multidimensional scaling revealed two distinct classes. When correlated with the data on metastasis, the two molecular classes corresponded very well to the survival data for the 27 patients. Thirty two discrete genes (corresponding to 44 probe sets) that correctly defined the molecular classes were selected. A single gene (ectonucleotide pyrophosphatase/phosphodiesterase 2; autotaxin) could classify the molecular types. The expression pattern was confirmed using real-time quantitative PCR.

Conclusions: Gene expression profiling identifies two distinct prognostic classes of uveal melanoma. Underexpression of autotaxin in class 2 uveal melanoma with a poor prognosis needs to be explored further.