A tissue biomarker panel predicting systemic progression after PSA recurrence post-definitive prostate cancer therapy

Abstract

Background: Many men develop a rising PSA after initial therapy for prostate cancer. While some of these men will develop a local or metastatic recurrence that warrants further therapy, others will have no evidence of disease progression. We hypothesized that an expression biomarker panel can predict which men with a rising PSA would benefit from further therapy.

Methodology/principal findings: A case-control design was used to test the association of gene expression with outcome. Systemic (SYS) progression cases were men post-prostatectomy who developed systemic progression within 5 years after PSA recurrence. PSA progression controls were matched men post-prostatectomy with PSA recurrence but no evidence of clinical progression within 5 years. Using expression arrays optimized for paraffin-embedded tissue RNA, 1021 cancer-related genes were evaluated-including 570 genes implicated in prostate cancer progression. Genes from 8 previously reported marker panels were included. A systemic progression model containing 17 genes was developed. This model generated an AUC of 0.88 (95% CI: 0.84-0.92). Similar AUCs were generated using 3 previously reported panels. In secondary analyses, the model predicted the endpoints of prostate cancer death (in SYS cases) and systemic progression beyond 5 years (in PSA controls) with hazard ratios 2.5 and 4.7, respectively (log-rank p-values of 0.0007 and 0.0005). Genes mapped to 8q24 were significantly enriched in the model.

Conclusions/significance: Specific gene expression patterns are significantly associated with systemic progression after PSA recurrence. The measurement of gene expression pattern may be useful for determining which men may benefit from additional therapy after PSA recurrence.

Figure 1. Nine genes with significantly different expression in cases with systemic disease progression (SYS) versus controls with PSA recurrence (PSA).

P-values (t-test) for the SYS case/PSA control comparison are shown. Controls with no evidence of disease recurrence (NED) are also included.

Figure 2. Areas under the curve (AUCs) for three clinical models, the final 17 gene/probe model and the combined clinical probe models.

A. The training set AUCs for three clinical models, the final 17 gene/probe model and the combined clinical/17 gene/probe model. B. The validation set AUCs for three clinical models, the final 17 gene/probe model and the combined clinical/17 gene/probe model. C. The training set AUCs of 4 previously reported gene expression models of prostate cancer aggressiveness compared with the clinical model C alone and with the 17 gene/probe model. D. The validation set AUCs of 4 previously reported gene expression models of prostate cancer aggressiveness compared with the clinical model C alone and with the 17 gene/probe model. For an explanation of the clinical models see Table 6. (E and F) A comparison of the training and validation set AUCs for each of the model. E. AUCs of the each of the gene/probe models alone. F. AUCs of each of the gene/probe models with the inclusion of clinical model C.

Figure 3. Systemic progression-free and overall prostate cancer-specific survival in the PSA control and SYS case groups.

A) Systemic progression-free survival for the patients classified in the poor outcome category and for those in the good outcome category in the PSA control group–17 gene/probe model. B) Prostate cancer-specific overall survival for the patients classified in the poor outcome category and for those in the good outcome category in the SYS case group–17 gene/probe model. C) Prostate cancer-specific overall survival for patients classified in the poor outcome category and for those in the good outcome category in the SYS case group-Lapointe et al. 2004 recurrence model.

Figure 4. Expression results for ERG, ETV1 and ETV4 among the men with no evidence of disease progression (NED), PSA recurrence (PSA) and systemic progression (SYS).

(A) Each overlapping set of three bars (blue, red and green) represent a different case or control. Thresholds for overexpression are ERG>3200, ETV1>6000 and ETV4>1400. (B) The numbers of cases showing overexpression of one or more of ERG, ETV1 and ETV4 are shown.