Abstract 1990: Exploratory analysis of the usefulness of the top-scoring pairs (TSP) of genes for prediction of prostate cancer progression

Abstract

Abstract Predicting prostate cancer progression after radical prostatectomy is most challenging. Gene expression profiling is widely used to identify genes associated with cancer progression. Usually candidate genes are identified in gene-by-gene comparisons of expression, but recent reports suggest that relative expression of a gene pair more efficiently predicts cancer progression. The rank-based top-scoring pair (TSP) algorithm classifies phenotypes according to relative expression of a gene pair [1]. In many cases TSP provides robust, efficient classification of phenotypes, but a large number of tests can lead to a large number of false-positive results. We modified the standard TSP algorithm by controlling the false-discovery rate and computing sensitivity and specificity of the test and applied the modified approach to Gene Expression Omnibus dataset GSE10645 [2]. The cited study analyzed the association between gene expression and outcome after initial therapy by comparing expression of ∼ 1000 candidate genes in 213 patients with no evidence of disease progression during 7 years after radical retropubic prostatectomy (RRP) with that in 213 patients whose disease progressed to the metastatic form. We used TSP to predict which patients would experience systemic tumor progression and which would stay disease free. Relative expression of TPD52L2/SQLE, BCS1L/SQLE, and CEACAM1/BRCA1 gene pairs predicted patients who will develop metastases after RRP with > 99% specificity but ∼10% sensitivity. The same gene pairs were validated in 3 independent prostate cancer datasets from GEO. Combining 2 pairs of genes (TPD52L2/SQLE and CEACAM/BRCA1) improved sensitivity without compromising specificity: 21.5% sensitivity and 99.5% specificity. Functional annotation of the TSP genes using the Ingenuity approach showed that they cluster by a limited number of biologic functions and pathways, including the molecular mechanisms of cancer, insulin receptor signaling, integrin signaling, and regulation of actin-based motility by Rho. In validation analysis with independent datasets, lower expression of genes from arginine and proline metabolism relative to that of genes from the insulin receptor-signaling pathway may be used to classify primary tumors vs distant metastases in cancer progression. In conclusion, comparative analysis of the expression of 2 genes and resulting pathways may be a simple, effective classifier for predicting prostate cancer progression.