An evaluation of blood mRNA expression array signatures derived from unsupervised analyses in the identification of prostate cancers with poor outcome.

Abstract

4577 Background: Prostate cancer (PCa) is the commonest male cancer in the Western world. Its biology and prognosis are heterogeneous and improved tools are needed to predict outcome more reliably. Here we studied the feasibility and clinical utility of whole blood mRNA expression array studies for this purpose. Methods: Whole blood samples were collected into PaxGene tubes between 08/2007 and 04/2008 from 100 PCa patients (pts): 31 good prognosis pts selected for active surveillance (AS) and 69 advanced castration resistant prostate cancer (CRPC) pts. RNA was extracted, amplified and biotinylated; unsupervised genome-wide expression profiles were analysed with HGU133plus2 (Affymetrix) microarrays and using Bayesian Latent Process Decomposition (LPD) analyses (Carrivick et al, 2006) in 94 patients; associations with outcome were studied. Results: LPD analysis of the whole blood mRNA expression data divided the pts into 4 separate groups (LPD1 to 4, with 10 unclassifiable pts). LPD1 (n=14) and LPD2 (n=18) consisted almost entirely of CRPC pts (14/14; 17/18); the single active surveillance patient in LPD2 had a rapidly rising PSA and required a prostatectomy. All the 4 groups included CRPC pts (LPD3:15/31; LDP4: 12/21); LPD1 CRPC pts had poorer overall survival (median 10.7 months, CI-95% 4.2-17.2) than CRPC pts in LPD2 to 4 (median 26.5 months, CI-95% 18.1-34.9, p=0.00007). Other baseline characteristics associated with LPD1 were increased number of treatment lines, worse performance status, and a higher proportion of detectable TMPRSS2/ERG transcripts. Group 1 membership remained the strongest prognostic factor in a multivariate analysis (HR 5.0, CI-95% 2.1-11.9, p = 0.0002). Gene signatures in the poor prognosis LPD group 1 were associated with increased CD71+ early erythroid cells and a decreased B-cell and T-cell immune response. A 9-gene signature (TERF2IP, TMCC2, GABARAPL2, SNCA, RIOK3, TFDP1, SLC4A1, HMBS, STOM) classified samples as group LPD 1 with a very low misclassification rate (0.012). Conclusions: PCa patient outcome can be predicted using gene expression signatures from peripheral blood.