Skip to content

Transcriptomic analysis of prostate cancer progression

Student thesis: Doctoral Thesis

Prostate cancer has a highly variable clinical course, which cannot be fully explained by clinicopathological variables. New biomarkers are needed to better inform treatment decisions at diagnosis to prevent the overtreatment of indolent disease, whilst allowing more aggressive disease to be appropriately treated with interventional and adjunctive treatments.

This discovery study used TaqMan arrays to measure the expression of a panel of 91 putative markers of prostate cancer progression in archival biopsy samples from 67 prostate cancer patients treated with radical prostatectomy and/or radiotherapy. The gene expression data were correlated with clinical outcome and binary logistic regression analysis was used to identify predictive models of 5-year biochemical/clinical recurrence.

Multivariate logistic regression models were focused to a final set of 28 two-gene and 9 three-gene models of recurrence which showed the highest overall performance in terms of calibration (Likelihood Ratio Test statistic (χ2)) and discriminatory accuracy (Area Under the Curve), whilst retaining parsimony (smallest number of predictor variables).
The best performing models according to χ2 were AMACR.EFNA5.SDHA (χ2(3) = 22.483, p<0.0001) and AMACR.HPRT1.SDHA (χ2(3) = 22.265, p<0.0001). In addition to demonstrating a highly significant improvement in fit to the data in comparison to the null model, these models had high discriminatory accuracy (Area Under the Curve of 0.812 and 0.829 respectively). By selecting an optimal threshold for categorisation of the patients into the recurrent or non-recurrent group, AMACR.EFNA5.SDHA could correctly categorise 76.1% of patients with a sensitivity and specificity of 81% and 73.9% respectively. AMACR.HPRT1.SDHA could correctly categorise 85.1% of patients with a sensitivity and specificity of 81% and 87% respectively.
Conclusion The study identified a selection of two- and three-gene models with potential to predict 5-year recurrence in this set of prostate cancer patients. These models may provide novel prognostic information beyond standard prognostic variables at the time of treatment selection but would need to be externally validated in another set of prostate cancer patients to confirm their generalizability as prostate cancer biomarkers. The findings contributed to knowledge regarding disease progression in prostate cancer. Several genes of interest were identified, some of which have previously been described in relation to prostate cancer progression and others that showed a novel association. A large proportion of the identified genes were key genes in metabolic pathways, underlining the importance of metabolic reprogramming in prostate cancer progression. 
Original languageEnglish
Awarding Institution
Award dateJun 2018


Relations Get citation (various referencing formats)

ID: 13518760