Abstract
Abstract: :
Purpose: Uveal melanomas exhibit a continuous spectrum of clinical and pathologic changes with no discrete prognostic stages. This study was performed to determine whether uveal melanomas may segregate into distinct prognostic groups based on gene expression. Methods: Gene expression profiling and/or real time PCR was performed on 33 primary uveal melanoma samples using Affymetrix GeneChips®. Data were analyzed by hierarchical clustering, primary component analysis, self–organizing maps, signal–to–noise ratio and weighted voting algorithms. Survival analysis was performed by life–table analysis. Results: Uveal melanomas surprisingly clustered into two discrete classes based on gene expression. Class 1 tumors tended to contain more spindle cells, and no patients in this class have died to date. Class 2 tumors tended to contain more epithelioid cells, and all deaths (which were all due to metastasis) occurred in this group. The 5–year survival probability of class 1 was 100% and class 2 was 41% (P = 0.007). Comparison of predictive value revealed that this molecular classification outperformed traditional pathologic grading, tumor diameter, thickness, location, and local invasion. Many of the genes that predicted class membership are involved in developmental processes. Conclusions: Despite a lack of discrete clinical or pathologic prognostic stages, uveal melanomas naturally group into two classes based on gene expression. Since this classification strongly predicts metastatic risk, this finding could potentially be used to develop a clinical predictive test for uveal melanoma patients.
Keywords: gene microarray • tumors • melanoma