Serum GADD45a methylation is a useful biomarker to distinguish benign vs malignant prostate disease

Abstract

Background: Prostate-specific antigen (PSA) screening for prostate cancer results in a large number of unnecessary prostate biopsies. There is a need for specific molecular markers that can be used in combination with PSA to improve the specificity of PSA screening. We examined GADD45a methylation in blood DNA as a molecular marker for prostate cancer diagnosis.

Methods: The study included 82 men, with PSA levels >4 ng ml(-1) and/or abnormal digital rectal exam, who underwent prostate biopsy. We compared GADD45a methylation in DNA from serum and buffy coat in 44 patients (22 prostate cancer and 22 benign). GADD45a methylation in serum DNA was examined in 82 patients (34 cancer and 48 benign).

Results: There was no significant difference in buffy coat GADD45a methylation between cancer and benign patients. Serum GADD45a methylation was significantly higher in cancer than in benign patients. Classification and regression tree predictive model for prostate cancer including risk groups defined by PSA, free circulating DNA (fcDNA) level and GADD45a methylation yielded specificity of 87.5%, sensitivity of 94.1% and receiver operator characteristic curve area of 0.937.

Conclusions: Serum GADD45a methylation in combination with PSA and fcDNA level was useful in distinguishing benign from prostate cancer patients.

Figure 1

Methylation in GADD45a CpG 1 in matched serum and buffy coat DNA samples. Box plots of percent GADD45a methylation in serum (A and D) and buffy coat (B) by biopsy result. Low correlation was seen between GADD45a methylation levels in serum and buffy coat by Ms-SNuPE analysis (C). GADD45a serum methylation was higher in cancer as compared with benign patients measured by Ms-SNuPE (A) and pyrosequencing (Pyro) (D). High correlation was observed between serum GADD45a methylation analysed by Ms-SNuPE and pyrosequencing (E).

Figure 2

Percent GADD45a methylation in serum (pyrosequencing) by biopsy result. Jittered box plots of original data (A), and of log base 2-transformed data (B).

Figure 3

(A) Classification tree based on cut points for PSA, fcDNA and sum of GADD45a four-CpG methylation (CpGs_sum). Age, race and ethnicity were also included in each model statement under RPART but none defined partition. Below each prediction node label (corresponding to majority vote) is shown the number of observations from benign and cancer groups, respectively. The corresponding misclassification rates were 8/82 (9.8%) overall, 6/48 (12.5%) in benign and 2/34 (5.9%) in cancer group. These implied specificity 42/48 (87.5%) and sensitivity 32/34 (94.1%). In addition, the leave-one-out cross-validation misclassification error was 9.9%. (B) Receiver-operating curves for comparison of the classification tree models including PSA, fcDNA and GADD45a methylation (solid and dashed lines) to other fitted models.