8pter–p23 Deletion Is Associated with Racial Differences in Prostate Cancer Outcome¹

Deletions of chromosome sequences mapping to the short arm of chromosome 8 have been observed frequently in a variety of human cancers. Most of these studies reported deletion of the 8p22–12 region (1, 2, 3, 4, 5, 6, 7). However, deletion of the 8pter–p23 region separate from deletion of other portions of the short arm of chromosome 8 has been reported for laryngeal cancers (8, 9, 10), renal clear cell carcinomas (11), ovarian tumors (12), and prostate tumors (5, 13). Moreover,some of these studies have positively correlated deletion of the 8pter–p23 region with poor prognosis (5, 10). On the basis of these observations, we hypothesized that a minimal common region of deletion within the 8pter–p23 region could be defined for prostate tumors, and that deletion of this region might correlate with disease recurrence using postoperative serum PSA⁵values as surrogate end point markers for survival. We now report the results of an extensive allelotyping of the 8pter–p23 region in 45 human prostate tumors and the association between loss of 8pter–p23 sequences with tumor pathology, patient demographics, and disease outcome.

Prostate tissue was obtained after radical prostatectomy from 45 patients diagnosed with prostate cancer. After an initial pathological evaluation, presumed malignant and normal tissues were snap frozen in liquid nitrogen and stored at −70°C. The tumor pathological stage and grade were defined using standard criteria (14, 15). Tumor specimens comprising areas of at least 70% malignant cells and nontumor specimens comprising normal benign or hyperplastic epithelium were serially sectioned. One section was stained with H&E to define areas of discrete histology. These areas were then microdissected from adjacent nonstained sections, and DNA was extracted as described previously (7). Postoperative serum PSA values were used as surrogate end point markers for disease outcome (16, 17). Patients whose postoperative serum PSA values were undetectable after surgery and remained undetectable thereafter were classified as disease free. In contrast, patients whose postoperative serum PSA values were detectable and escalated either immediately after surgery or at some point thereafter were considered to be in biochemical progression with persistent or recurrent disease. For the purposes of statistical analysis, survival was defined as the number of postoperative disease-free days, i.e., the time until rising PSA values first became detectable (e.g., biochemical progression). PSA follow-up data were available for 44 of the 45 patients in this study, with a median follow-up time of 60 months (5 years). Clinical and pathological information for the specimens and patients included in this study are summarized in Table 1.

PCR amplification assays targeted 14 sequences containing highly polymorphic microsatellite repeats at loci of interest on chromosome 8p23. The linkage order of these markers has been reported as:pter-D8S504-D8S264-D8S1824-D8S1781-D8S262-D8S1798-D8S518-D8S561-D8S277-D8S503-D8S550-D8S265-D8S552D8S261-centromere. Primer sequences and linkage information were obtained from data bases maintained by the Human Genome Database,⁶Center for Genome Research at the Whitehead Institute for Biomedical Research,⁷and the National Center for Biotechnology Information⁸as accessed through the Internet. PCR reactions were accomplished as described previously using 1 μm oligonucleotide primers and 100 ng of oligonucleotide primer end-labeled with[γ-³²P]dATP (7). Aliquots of each reaction were electrophoresed on 6% acrylamide/7 m urea-sequencing gels, and the gels were autoradiographed (5, 7). Allelic loss was scored by visual inspection of two independent observers (J. G. W. and J. A. M.)when the ratio of allelic signal intensities in tumor tissue was ≤50%of that for the same alleles in normal tissue from the same heterozygous patient.

For the purposes of statistical analysis, survival was defined as the number of postoperative disease-free days, i.e., time until rising PSA values first became detectable (e.g.,biochemical progression). Kaplan-Meier estimates for the distributions of outcome, measured as the time to progression, after surgery, were computed for strata defined by 8p loss status. Log-rank tests for homogeneity of disease-free survival time distribution across 8p loss categories were performed both on the entire sample and by race. P ≤ 0.05 was considered to indicate significant differences in survival distribution across 8p loss categories.

A multivariable Cox proportional hazards regression model was fitted to the entire data to estimate the effects of pathological tumor stage,age, race, preoperative serum PSA value, and 8p loss pattern on survival time, and the significance of each was tested, controlling for all other variables in the model. A step-down variable selection procedure, as described by Mantel (18), was adopted to arrive at a final model. Because our primary goal was to assess the effect of 8p loss patterns on outcome, terms related to such effects were not considered for deletion from the model at any step. Thus, the final model consisted of predictors that had a significant effect on disease-free survival time in the presence of 8p loss terms in the model.

Two separate approaches were followed to assess the effect of 8p loss patterns on outcome in a multivariable analysis. In the first approach, the effect of each 8p loss category, i.e., loss in the 8pter–p23 region, loss in other parts of 8p, and loss of all informative 8p loci, compared with no 8p loss was assessed. In the second approach, the outcome of those with losses in the 8p terminal arm was compared with the outcome of all other patients in the sample.

Associations among predictors were tested prior to fitting the multivariable models using suitable t tests andχ ² tests. Multivariable analyses following both approaches, and with 8p loss terms and all other predictors, except race, that were significant or approached significance in the fit to the entire set of patients was performed for each race (19).

Forty-five paired normal and tumor samples from radical prostatectomy specimens were examined for allelic dosage at 13 loci spanning 0–29 cM within 8pter–p23 and two loci mapping more proximally, D8S549 at 31 cM and D8S261 at 39 cM. Twenty-eight of the 45 tumors examined (62%) demonstrated loss of at least one locus mapping to 8p. Although the deletion patterns demonstrated by these 28 tumors were complex, three allelic loss groups, as well as a minimal common region of deletion, could be discerned. Twelve of the 28 tumors (43%) demonstrated loss of all or part of 8pter–p23 spatially separate from more proximal loci (designated as 8pter–p23 in Table 2). Five of these 12 tumors also demonstrated more distal deletions, but these were not contiguous with the more extensive proximal region of deletion. Six of the 28 tumors (21%) demonstrated loss of all 8p loci examined in this study (all 8p; Table 2), and 10(36%) demonstrated loss of isolated loci (cases 80, 111, 500, and 210)or two or more loci spanning both proximal and distal sequences (other 8p; Table 2). Within the 12 tumors demonstrating loss of all or part of 8pter–p23 spatially separate from more proximal loci, three–504, 520,and 410–defined the distal boundary of loss at D8S264, and two–9 and 125–defined the proximal boundary of loss at D8S1798. Thus, the minimal common region of deletion within 8pter–p23 may be defined by a 5-cM interval spanning loci D8S264-D8S1824-D8S1781-D8S262-D8S1798. An autoradiographic example of deletions at the D8S1824 locus is shown in Fig. 1.

For the purposes of statistical analysis, survival was defined as the number of postoperative disease-free days and ended when rising postoperative serum PSA values first became detectable(e.g., biochemical progression). Survival was assessed for the four patterns of 8p loss observed in the prostate tumors examined:8pter–p23 loss, loss of other 8p sequences, loss of the entire 8p arm,and no loss. Log-rank tests indicated a significant difference in survival among the four groups, with the 8pter–p23 loss group demonstrating the most rapid rate of biochemical failure over time(log-rank P = 0.0426; Fig. 2). This significant difference was maintained when survival of the 8pter loss group was compared against the other three allelic loss groups combined (log-rank P = 0.0068; Fig. 2). Moreover, 50%of patients whose tumors demonstrated 8pter–p23 loss were in biochemical failure by 340 days after surgery compared with 1340 days for all other patients (Fig. 2). These findings show that loss of the 8pter–p23 chromosomal region is specifically and significantly associated with poor disease outcome in prostate cancer patients.

A multivariable Cox proportional hazards model was fitted to the data to estimate the effects of pathological tumor stage, patient’s age at surgery, patient’s race, preoperative serum PSA values, and 8p loss patterns on survival, and the significance of each was tested,controlling for all other variables in the model. Because pathological tumor stage is strongly correlated with combined Gleason score(Mantel-Haenszel P = 0.002) in our sample, it was decided not to include both predictors simultaneously in the analysis. Separate analyses for each indicated that tumor stage was a better predictor of outcome and hence was preferred over combined Gleason score.

Following our first approach, a step-down variable selection algorithm eliminated three predictors in successive steps (model 1; Table 3A). Of all of the variables in the initial model, patient race was the least significant predictor of outcome and was not considered further. Patient age at surgery and preoperative serum PSA, in that order, were dropped in subsequent fits. As described earlier, the 8p loss pattern terms were never considered for deletion from the model at any step. Our final model consisted of pathological tumor stage(P = 0.014) and 8p loss terms as important predictors of outcome. Although pathological stage was significant at every step of the model selection process, the effect of 8pter–p23 loss on outcome compared with patients with no 8p loss was seen to be significant only in the absence of preoperative PSA in the model(P = 0.022). This analysis also estimated that loss of the 8pter–p23 region, apart from loss of other 8p sequences or loss of the entire 8p arm, was associated with an estimated 3.36-fold increase in risk (95% confidence interval, 1.19–9.49) for poor disease outcome compared with patients with no 8p loss. Similar results were observed following our second approach (model 2; Table 3B).

These results are consistent with results obtained from univariate log-rank tests and Kaplan-Meier survival function estimates and indicate that loss of the 8pter–p23 chromosomal region may be specifically associated with poor disease outcome in prostate cancer patients (Fig. 2). Suitable t tests indicated no significant difference in mean preoperative serum PSA levels between the 8pter–p23 loss group and other patients. Therefore, we attribute the lack of significance of the effect of 8pter–p23 loss on survival in the presence of preoperative serum PSA to the differential effect these variables individually have on outcome for each race.

Although equivalent numbers of ACM (14 of 24; 42%) and AAM (12 of 20; 40%) patients experienced biochemical failure, log-rank tests indicate that time to failure was significantly associated with 8pter–p23 loss only for ACM patients. Separate multivariable analyses for ACM and AAM prostate cancer patients were performed to estimate the effects of pathological stage, preoperative PSA, and 8p loss categories on survival. Of all variables in the initial Cox model fitted to the entire data, only these predictors were either significant or approached significance (Table 3C). Loss of the 8pter–p23 region was seen to be significantly associated with poor survival, compared with patients with no 8p loss, for ACM (P = 0.006) but not AAM (P = 0.707) patients (Table 3C). On the other hand,preoperative serum PSA was an important predictor for AAM(P = 0.042) but not ACM (P = 0.897)patients (Table 3D). This analysis estimated that, for ACM patients,loss of the 8pter–p23 region conferred a 10.85-fold higher risk (95%confidence interval, 2.57–45.83) of poor outcome, compared with patients with no 8p loss. These data suggest that loss of the 8pter–p23 chromosomal region may be specifically and significantly associated with poor disease-free survival in ACM but not AAM prostate cancer patients.

The predictive accuracy of our multivariable model selection procedure suffers because of our small sample size and the not-so-small set of known and important prognostic factors to be controlled for in a meaningful assessment of the effect of 8p loss patterns on outcome. The results of our statistical analyses will require external validation on other similar patient populations. Further study with a larger sample of patients, and hence, greater statistical power and more precise estimates of risk ratios with narrower confidence limits are sought to confirm our findings.

In this study, 62% of 45 tumors examined demonstrated loss of at least one locus mapping on 8p, and almost half of those tumors(12 of 28) demonstrated independent loss of the 8pter–p23 region. The highest contiguous frequencies of loss were observed within an extremely p-terminal 5-cM interval including loci D8S264-D8S1824-D8S1781-D8S262-D8S1798 which, together,defined a minimal common region of loss that was deleted independently of other 8p sequences. This region is almost identical to that reported by Sunwoo et al. (8) as the minimal common region of deletion on 8pter in squamous cell carcinomas of the larynx and oral cavity spanning loci D8S264-D8S1806-D8S1824-D8S1781-D8S1788. Using physical mapping information provided by the Center for Genome Research at the Whitehead Institute for Biomedical Research⁷ and reported by Sun et al.(9), this interval may be described contiguously as D8S264-D8S1806-D8S1824-D8S1781-D8S1788-D8S262-D8S1798,with the markers used by both the current study and that of Sunwoo et al. in bold italic (8). This interval includes an 8pter region described as frequently deleted in oropharyngeal squamous cell carcinomas by Wu et al.(20) and also overlaps with that reported deleted in squamous cell carcinomas of the head and neck by Ishwad et al. (21). Moreover, homozygous deletions within this interval have been reported by Ishwad et al.(21) spanning D8S1781-D8S262 in three oral squamous carcinoma cell lines, and by Sun et al. (9)spanning D8S1824-D8S1788 in four head and neck squamous carcinoma cell lines. Taken together, these studies suggest that the interval between D8S264 and D8S1798 likely contains at least one tumor suppressor gene whose inactivation is critical for tumorigenesis in the prostate, larynx, and oral cavity.

Previous studies have shown that biochemical progression,or the development of detectable serum PSA values, within 2 years after radical prostatectomy predicts distant metastases in 90–93% of cases (16). Moreover, the median actuarial time to death after the development of metastatic disease is 5 years (17). Therefore, the detection of rising postoperative serum PSA values serves as a suitable surrogate end point marker for prostate cancer recurrence and death from metastatic disease. Our laboratory had reported previously a positive correlation between the deletion of 8pter sequences and biochemical progression (5). That observation was confirmed by the current study, which found that separate deletion of 8pter–p23 was significantly associated with decreased survival measured as biochemical progression (log-rank P = 0.0068). Moreover, 50% of patients whose tumors demonstrated 8pter–p23 loss were in biochemical progression by 340 days after surgery compared with 1340 days for all other patients. This conclusion was supported by a Cox proportional hazards analysis of the data showing that tumor pathological stage (P =0.013) and 8pter–p23 loss (P = 0.0984) were clearly associated with poor survival, and that loss of the 8pter–p23 region, apart from loss of other 8p sequences or loss of the entire 8p arm, was associated with a 2.7-fold increase in risk for poor disease outcome. Scholnick et al. (10) have shown that loss of the 8pter–p23 region, especially of the D8S264locus, in squamous cell carcinomas of the head and neck is significantly associated with both a shorter disease-free interval and reduced disease-specific survival. Our data now show that loss of the 8pter–p23 region, inclusive of the D8S264 locus, in human prostate tumors is also associated with a shorter disease-free interval for prostate cancer patients. These studies suggest that loss of the 8pter–p23 region serves as a marker for poor prognosis and may inactivate tumor suppressor genes critical for tumorigenesis in at least two types of human tumors.

A unique feature of the current study is the finding that,although equivalent numbers of ACM and AAM patients experienced biochemical progression, loss of the 8pter–p23 region was significantly associated with poor survival for ACM, but not AAM,patients by both Kaplan-Meier (log-rank P = 0.0024) and Cox proportional hazards model (P = 0.006) analysis. Moreover, this analysis showed that loss of the 8pter–p23 region conferred a 8.67-fold higher risk for poor survival for ACM patients. These data suggest that loss of the 8pter–p23 chromosomal region is specifically and significantly associated with poor disease outcome in ACM, but not AAM, prostate cancer patients. Recent studies have suggested that racial differences in the morbidity and mortality associated with prostate cancer, especially for non-organ-confined disease, may be based upon genetic and epigenetic differences (22). Our study shows that specific loss of the 8pter–p23 region increases the risk for poor survival for ACM patients, but that other genetic events likely contribute to poor survival for AAM prostate cancer patients. Further studies should confirm the prognostic value of 8pter–p23 loss as a marker for prostate cancer recurrence and progression and should facilitate the isolation of putative tumor suppressor genes mapping to this region.