Background
We identify and validate accurate diagnostic biomarkers for prostate cancer through a systematic evaluation of DNA methylation alterations.
Materials and methods
We assembled three early prostate cancer cohorts (total patientsâ=â699) from which we collected and processed over 1300 prostatectomy tissue samples for DNA extraction. Using realâtime methylationâspecific PCR, we measured normalized methylation levels at 15 frequently methylated loci. After partitioning sample sets into independent training and validation cohorts, classifiers were developed using logistic regression, analyzed, and validated.
Results
In the training dataset, DNA methylation levels at 7 of 15 genomic loci (glutathione Sâtransferase Pi 1 [GSTP1], CCDC181, hyaluronan, and proteoglycan link protein 3 [HAPLN3], GSTM2, growth arrestâspecific 6 [GAS6], RASSF1, and APC) showed large differences between cancer and benign samples. The best binary classifier was the GAS6/GSTP1/HAPLN3 logistic regression model, with an area under these curves of 0.97, which showed a sensitivity of 94%, and a specificity of 93% after external validation.
Conclusion
We created and validated a multigene model for the classification of benign and malignant prostate tissue. With false positive and negative rates below 7%, this threeâgene biomarker represents a promising basis for more accurate prostate cancer diagnosis.