Tissue Microarray Sampling Strategy for Prostate Cancer Biomarker Analysis

High-density tissue microarrays (TMA) are useful for profiling protein expression in a large number of samples but their use for clinical biomarker studies may be limited in heterogeneous tumors like prostate cancer. In this study, the optimization and validation of a tumor sampling strategy for a prostate cancer outcomes TMA is performed. Prostate cancer proliferation determined by Ki-67 immunohistochemistry was tested. Ten replicate measurements of proliferation using digital image analysis (CAS200, Bacus Labs, Lombard, IL, USA) were made on 10 regions of prostate cancer from a standard glass slide. Five matching tissue microarray sample cores (0.6 mm diameter) were sampled from each of the 10 regions in the parallel study. A bootstrap resampling analysis was used to statistically simulate all possible permutations of TMA sample number per region or sample. Statistical analysis compared TMA samples with Ki-67 expression in standard pathology immunohistochemistry slides. The optimal sampling for TMA cores was reached at 3 as fewer TMA samples significantly increased Ki-67 variability and a larger number did not significantly improve accuracy. To validate these results, a prostate cancer outcomes tissue microarray containing 10 replicate tumor samples from 88 cases was constructed. Similar to the initial study, 1 to 10 randomly selected cores were used to evaluate the Ki-67 expression for each case, computing the 90th percentile of the expression from all samples used in each model. Using this value, a Cox proportional hazards analysis was performed to determine predictors of time until prostate-specific antigen (PSA) recurrence after radical prostatectomy for clinically localized prostate cancer. Examination of multiple models demonstrated that 4 cores was optimal. Using a model with 4 cores, a Cox regression model demonstrated that Ki-67 expression, preoperative PSA, and surgical margin status predicted time to PSA recurrence with hazard ratios of 1.49 (95% confidence interval [CI] 1.01–2.20, p = 0.047), 2.36 (95% CI 1.15–4.85, p = 0.020), and 9.04 (95% CI 2.42–33.81, p = 0.001), respectively. Models with 3 cores to determine Ki-67 expression were also found to predict outcome. In summary, 3 cores were required to optimally represent Ki-67 expression with respect to the standard tumor slide. Three to 4 cores gave the optimal predictive value in a prostate cancer outcomes array. Sampling strategies with fewer than 3 cores may not accurately represent tumor protein expression. Conversely, more than 4 cores will not add significant information. This prostate cancer outcomes array should be useful in evaluating other putative prostate cancer biomarkers.