Differential expression of CD10 in prostate cancer and its clinical implication

This study was carried out under approval by the University of Washington IRB for the research use of excess tissue from surgeries. Frozen blocks of tumor tissue in OCT were retrieved from the tumor bank in the Department of Urology, and multiple 5-μm serial sections were cut from each specimen and fixed in cold acetone for CD10 immunohistochemistry. Details of this procedure have been previously described [1]. Each specimen was assigned a numeric code with a letter code indicating the site origin of the block (right apex, left mid, etc.). Cancer glands were scored for the presence of CD10 staining, and overall percentage of glands staining positive in the different Gleason component patterns was scored by a single pathologist (LDT). Since benign glands stain positive for CD10, this provided a reliable gauge for positive cancer staining. An isotype control was also done as described [1]. Eighty-seven patients, 53 with and 34 without pathologically organ-confined prostate cancer at the time of RP were identified. The CD10 results of these specimens were reported in ref. 1, and a formal statistical analysis relating CD10 staining and clinicopathological characteristics is presented here. In scoring, each tumor in the particular section examined was characterized as containing what percentage (0 to 100%) of Gleason pattern 3, pattern 4 or pattern 5; then within each pattern, the percent positive for CD10 staining. Fourteen patients with lymph node metastasis found at the time of RP were identified and included in this study. Clinical characteristics were obtained from the medical records.

To further test our methods for estimating percent CD10 staining by specific Gleason pattern for correlation with clinical parameters we elected to perform an identical analysis with a separate CD antigen. Another cohort of 66 patients was scored for CD13 immunohistochemistry and subjected to the same statistical analysis described below. The cancer glands were similarly characterized, and correlation between staining and the same disease parameters was investigated. CD13 was chosen, because it, like CD10, is a cell surface peptide-processing enzyme (ANPEP, aminopeptidase N) expressed by luminal cells, and a majority of cancer glands were also found not to express CD13. Like CD10, a smaller percentage of cancer glands were positive for CD13. The CD13 cohort was not identical to the CD10 cohort, but there was some overlap of subjects. Because of the limited number of lymph node metastases in our collection, the same node specimens were serially stained for CD10 and CD13. Monoclonal CD10 (clone HI10a) and CD13 (clone WM15) antibodies were obtained from BD PharMingen (San Diego, CA). In our hands, these antibodies worked well for frozen sections [1]. Antibodies were used at a concentration of 4 ng/μl. Immunolocalization was done by an indirect avidin-biotin-peroxidase method as described previously [1]. In both these investigations, tissue microarray was not used because of inherent sampling problems with such arrays as only a small portion of any tumor is represented. This would make correlation with clinical outcome difficult as we previously pointed out [1]. While the sections used here were more representative, cancer foci in areas not captured by the sections were nevertheless missed.

Standard statistical summaries and procedures were used for visualization and significance testing of associations with CD10, and with CD13, including scatterplots with Pearson product-moment and Spearman nonparametric correlation measures and tests, cumulative distribution functions with Kolmogorov-Smirnov and Wilcoxon nonparametric tests, and 1-way ANOVA. Nonparametric methods were included due to non-normal distributions and the presence of ordinal variables. For calculation of odds ratios for lymph node metastasis and for estimating PSA free survival, strong staining was defined as > 20% positivity based on the observed maximum separation of the cumulative distributions. PSA free survival was calculated using the Kaplan-Meier method and survival curves were compared using the log-rank test. Only patients with pathologically localized (pT2-T3N0) disease with an undetectable PSA after RP were included in the survival analysis. Patients with lymph node positive disease and patients who received adjuvant therapy were excluded. PSA recurrence was defined as any value ≥ 0.2 ng/ml after an undetectable PSA after RP.

DNA array analysis of cancer cell lines [14] and sorted prostate cell populations [15] was done with Affymetrix HG-U133 GeneChips. CD26 was used to sort luminal cells, CD104 to sort basal cells, and CD49a to sort stromal fibromuscular cells. In cell sorting, prostatectomy tissue samples free of cancer were digested with collagenase and the resultant cells were partitioned on Percoll density gradients into epithelial and stromal fractions. Magnetic cell sorting (MACS) was used to select the targeted cell populations after labeling with CD antibodies conjugated to R-phycoerythrin (PE). The array datasets were comprised of 2 replicates for the cell lines and 5 replicates for the sorted cells. In addition to the cell types mentioned, datasets were obtained for benign tissue (NP), tumor tissue (CaP), sorted cancer cells from a primary tumor (CD26+ CaP), sorted endothelial cells (CD31) [15], putative epithelial stem cells (CDw338/antibody clone 5D3). A description of the sorted cancer cells and 5D3 cells will be reported elsewhere. Data analysis was done with GeneSpring software.

Both CD10 and CD13 are expressed by luminal cells of the prostatic epithelium but their expression in cancer is heterogeneous [1]. Examples of all four possible CD10/CD13 cancer cell types – CD10^-/CD13^-, CD10^-/CD13⁺, CD10⁺/CD13^-, CD10⁺/CD13⁺ – were detected in our cohort. Normal epithelium present in the specimens showed the characteristic strong CD10 and CD13 expression at the luminal plasma membrane. Fig. 1 shows frozen section examples of tumors stained serially for CD10 and CD13. In all three cases, the cancer glands were mostly negative for CD10. In two cases, the cancer glands were positive for CD13, and a small subpopulation of CD13⁺ tumor cells was present in one case. These examples showed that CD13⁺ cancer cells could be found in tumors with either a glandular (Gleason 3) or non-glandular (Gleason 4) morphology. Overall, the predominant cancer cell phenotype was CD10^-/CD13^-, and this was validated by DNA array analysis data shown in Fig. 2. As can be seen, the expression pattern was similar for CD13 and CD10 (except in the cancer cell lines) with both CD10 and CD13 expression down-regulated in cancer specimens. For the cell lines, the expression profiles of these two genes were in agreement with the reported flow cytometry and immunocytochemistry data [11, 16]. The data also showed that expression of any particular gene was variable in cancer; one cancer cell type may express it while another one may not. Their differential expression in cell lines may reflect the situation in tumors, where either positive or negative cancer cell types are found.

In the CD10 cohort, a trend was seen that in tumors with predominantly pattern 3 cancer (Gleason score 3+3 = 6) the percentage of positive CD10 staining was less than 5–10%. Higher percentages were found in tumors with patterns 4 or 5, although some pattern 3 tumors were also positive [1]. Correlations between CD10 percentage and Gleason score, clinical stage and pre-operative serum PSA level are summarized in Table 1 and are as follows. The first scatter plot (Fig. 3) shows the relationship between percent positive and Gleason score with jitter added to allow separation of multiple observations with identical data values for viewing purposes. The result was pooled based on relative similarity at advanced grades, combining all cases more advanced than 3+3 into a single group (n = 69) and compared to that of 3+3 (n = 18). Clinically, there is a clear behavioral difference between Gleason pattern 3 and pattern 4/5 tumors. The result, shown as cumulative distributions, was very highly statistically significant (Wilcoxon Z = 3.645, P = 0.0003; Kolmogorov-Smirnov D = 0.551, P = 0.0003) with higher percent positive associated with advanced Gleason scores. The mean increased from 4.1% to 32%, while the median increased from 2% to 20% for Gleason score 6 and Gleason score 7, 8 and 9 tumors, respectively. The maximum separation of the cumulative distributions, 100%-44.9%, occurred at 15% positive. The result was significant across all Gleason scores (1-way ANOVA F = 4.214 with 3 and 83 degrees of freedom, P = 0.00796). The Spearman correlation was ρ = 0.336, P = 0.00148 with n = 87. The second scatter plot (fig. 4) shows the relationship between percent positive and stage. The result was pooled based on the relative sparseness at advanced stages (combining stages T3 and T4), and was separated to compare organ confined (T2N-, n = 49) to non-organ confined stages (T2N+, T3, T4, n = 38) as shown in the cumulative distributions. The differences were statistically significant (Wilcoxon Z = 2.316, P = 0.0206; Kolmogorov-Smirnov D = 0.320, P = 0.0253; Spearman ρ = 0.234, P = 0.0293 with n = 87), with more advanced stages having significantly higher percent positive staining. The mean increased from 20.6% to 33.5%, while the median increased from 5% to 23.4% for the organ confined and non-organ confined stages, respectively. The maximum separation of the cumulative distributions, 71.4%-39.5%, occurred at 17% positive. The third scatter plot (Fig. 5) shows the relationship between percent positive and pre-operative PSA level (logarithmic scale, n = 84 observations after 3 were omitted due to missing values). There was a statistically significant relationship, with a correlation of 0.262 (P = 0.012), with higher percent positive associated with higher PSA levels. The relationship and its significance were robust to the 3 possible outliers with the highest PSA levels (correlation of 0.241, P = 0.0299, when these were omitted). The Spearman correlations were ρ = 0.300, P = 0.00555 with n = 84, and ρ = 0.279, P = 0.0118 with n = 81 minus the 3 outliers.

The second analysis showed no statistically significant correlations between CD13 and the same pathologic and clinical parameters. A scatter plot (Fig. 6) shows the relationship between percent positive and Gleason score. The result was analyzed by combining all cases that were more advanced than 3+3 into a single group (n = 54) and compared to that of 3+3 (n = 12). The resulting cumulative distributions showed that the relationship was not statistically significant (Wilcoxon Z = 1.439, P = 0.075; Kolmogorov-Smirnov D = 0.296, P = 0.355; Spearman ρ = 0.0700, P = 0.576 with n = 66). The next scatter plot shows the relationship between percent positive and stage. When the result was separated to compare T2N- (n = 39) to the more advanced stages (n = 27), as shown in the cumulative distributions, the differences were again not statistically significant (Wilcoxon Z = 0.548, P = 0.584; Kolmogorov-Smirnov D = 0.214, P = 0.460; Spearman ρ = -0.111, P = 0.375 with n = 66).

Nearly all lymph node metastases available in our collection (8/9) displayed strong staining for CD10 (Fig. 7). Seventy percent of RP specimens from patients with node metastasis showed strong staining (> 20%) for CD10, compared to 30% in the entire cohort (OR = 3.4, 95% CI: 1.08–10.75, P = 0.019). Thus, patients with tumors staining heavily for CD10 were more likely to harbor lymph node metastasis at the time of RP. Tumor cell lines derived from node metastases, LNCaP, C4-2 (Fig. 2) and xenograft LuCaP 35 [16], are CD10⁺.

Thirteen patients in our cohort with a mean follow-up of 29 months experienced PSA recurrence after treatment while 32 remained disease free. Figure 8 depicts the Kaplan-Meier PSA free survival stratified by percent positive staining for CD10. Patients with > 20% staining for CD10 showed significantly worse PSA free survival after RP (P = 0.03).