Background rationale
The accurate estimation of prostate cancer progression risk early in the course of the disease is essential for optimal management. Risk stratification is predominantly based upon systematic prostate sampling at biopsy, but undergrading of tumours is frequent, occurring in up to 50% of Victorian men with significant implications for their care. It has previously been reported that the presence of prostate cancer affects gene transcription in associated benign prostate acini, however it is unknown if there are detectable differences in this field effect between men who do not require treatment (low risk disease) and those that do (high risk disease). We were therefore interested to determine if a transcriptional signature exists in benign tissue that can risk stratify men with localised prostate cancer at the time of diagnosis.
Methods
Here, employing machine learning techniques, we compare and contrast the ability of transcriptional markers derived from benign prostatic tissue (proximal to the tumor foci) and adipose tissue (periprostatic and subcutaneous) to predict the tumor grade from a total of 194 patients.
Results
Surprisingly, we show that selected transcripts (n<5) from both subcutaneous and periprostatic adipose tissue have a significative predictive power, using support vector machine alone, with an area under the curve greater than 0.90. On the contrary, transcripts from the more proximal prostate tissue show no significative predictive power.
Conclusions
This study provides the foundation for a potential prognostic test for prostate cancer and offers new insight into the biology of high risk disease.