Failure of BRCA1 Dysfunction to Alter Ovarian Cancer Survival¹

Multiple demographic, surgical, pathological, and molecular parameters have been reported to modify survival of individuals diagnosed with ovarian cancer. Examples include age at diagnosis (1), histology (2, 3), cancer grade (2, 3), FIGO³ stage (4), tumor DNA ploidy (3, 5), residual disease (6, 7), preoperative CA125 (8, 9), and more recently, p53 mutation status (10, 11). Several years ago an exciting observation was made by Rubin et al. (12) that individuals whose cancers resulted from germ-line BRCA1 mutation enjoyed enhanced survival over those individuals whose tumors did not carry BRCA1 mutation. This finding seemed to confirm our earlier observation of prolonged survival associated with familial ovarian cancer (13). However, since these initial reports, conflicting additional studies have suggested that BRCA1 carrier status either may (14, 15) or may not (16, 17) modify the survival for individuals diagnosed with BRCA1-related hereditary ovarian cancer. It is important to be able to provide these individuals and their families with accurate prognostic information.

To a large extent, the published survival studies have suffered from a variety of shortcomings including: (a) potential selection bias introduced by studying a population of long-term survivors (12); (b) the use of populations screened only for BRCA1 founder mutations (14, 16); (c) failure to control for volume of residual disease (12, 15, 16, 17); (d) inclusion of borderline cancers (12); (e) potential differences in chemotherapy between groups (12, 15, 16, 17); and (f) comparisons between groups treated at different institutions (16, 17) or during different eras (12, 16, 17). To date, no study has considered BRCA1 dysfunction attributable to other mechanisms as a confounding variable in the control cohort. This factor should at least be considered, because somatic BRCA1 mutations are found in ∼10% of all ovarian cancer (18, 19, 20, 21). In addition, transcriptional silencing of BRCA1 expression by promoter methylation represents another mechanism whereby the BRCA1 tumor suppressor gene may be down-regulated in a substantial number of ovarian cancers (21, 22). At the molecular level, differences in the p53 mutational spectrum between BRCA1-related ovarian cancers relative to sporadic ovarian cancers without BRCA1 mutation could also directly contribute to survival differences independent of BRCA1 mutation status (23). This hypothesis is based on our previous report that specific p53 mutation types contribute to adverse outcome (10). Thus, it may be critical to select an appropriate control population based both on tumor p53 and theoretical BRCA1 functional status, as well as conventional histopathological and demographic prognostic variables.

To overcome many of the potential limitations just discussed, we have carried out an analysis of a large ovarian cancer cohort diagnosed and/or treated at a single institution, the Holden Cancer Center of the University of Iowa, where ovarian cancers are routinely screened for both p53 mutations and truncating BRCA1 tumor mutations. This allowed matching of tumors with BRCA1 dysfunction because of any of the above three mechanisms with tumors expressing presumed functional BRCA1 while simultaneously controlling for specific p53 mutation type, patient age, FIGO surgical stage, tumor histology and site including grade, residual disease, preoperative CA125, date of diagnosis, and chemotherapy regimen. Survival was determined for individuals with tumors containing BRCA1 dysfunction secondary to germ-line BRCA1 mutation, somatic BRCA1 mutation, or promoter methylation compared with the case-matched controls.

Since 1991 all of the individuals diagnosed with ovarian cancer treated at the University of Iowa Hospitals and Clinics have participated in a prospective study approved by the University of Iowa Committee for the Protection of Human Subjects to analyze a variety of prognostic markers in the outcome of their ovarian cancers. For the purposes of this study, we have included ovarian, fallopian tube, and primary peritoneal carcinomas. Collectively we refer to these as ovarian cancers. No borderline cancers were included in this analysis. p53 mutations have been characterized based initially on SSCP screening with confirmatory sequencing (10, 23, 24, 25) and subsequently with complete sequencing of the p53 gene in nearly 350 individuals. Using snap-frozen fresh tumor samples, we have developed recently a screening strategy for BRCA1 dysfunction based on the identification of both germ-line and somatic BRCA1 null mutations in a PTT assay that covers the entire BRCA1 open reading frame (21, 26). Whereas the PTT technique will not detect missense mutations, it does facilitate identification of cancers that are BRCA1 null secondary to BRCA1 promoter methylation (21) as well as atypical splice variants (26). We have also shown this screening strategy to be superior to the more conventional SSCP screening of BRCA1 (26). Briefly, mRNA and DNA are prepared from tumor tissue (24, 26). The mRNA is reverse transcribed into cDNA template using random hexamers (24). Five overlapping PCR amplifiers, containing T7 promoters, are generated for in vitro translation (TNT Quick Coupled Transcription/Translation System; Promega, Madison, WI). Three fragments are DNA-based, covering exon 11, whereas two fragments are cDNA-based covering the rest of the open reading frame, each overlapping of a portion of exon 11. Subsequent sequencing validates that abnormal PTT profiles are caused by either germ-line or somatic mutations depending on whether or not sequence abnormalities are also present in peripheral blood DNA. If a full-length protein product is identified in the coupled translation reaction, we have assumed this to represent wild-type BRCA1 sequence. Samples were consecutive based on the availability of both snap-frozen tissue and germ-line DNA plus a minimum of 1 year follow-up data. A subset was also screened by SSCP (18), and the 4 cancers with missense mutations that do not result in PTT have also been included in the present study. Thus far 40 cancers containing a BRCA1 mutation have been identified among 250 unselected ovarian cancers. Of these, 24 resulted from germ-line BRCA1 mutations, whereas 16 represent somatic BRCA1 ovarian cancer tumor mutations (18, 22). There were 21 protein truncating and 3 missense mutations identified in the germ-line cohort, whereas the somatic BRCA1 cohort contained 1 missense and 15 truncating mutations. An additional 19 ovarian cancers were found to contain methylated CpG sequences in the region of nucleotides 1536–1622 of the BRCA1 promoter (GenBank accession no. U37574) by methylation-specific PCR (21). For these cancers, we were unable to detect BRCA1 mRNA, whereas expression of the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase mRNA and other gene mRNAs were readily detected.

All cancers containing a BRCA1 mutation or BRCA1 promoter hypermethylation were entered into the present study. Data collected on individuals diagnosed with these cancers included FIGO stage, histological type, tumor grade, p53 mutation status, preoperative CA125, residual disease, date of diagnosis, date of death or last follow-up, site of disease, and initial chemotherapy regimen. For each case with BRCA1 dysfunction, a control was selected on the basis of optimizing the match between: (a) specific p53 mutations; (b) age; (c) FIGO surgical stage; (d) residual disease; (e) histological type; (f) tumor grade; (g) preoperative CA125; and (h) site of disease. Chemotherapy regimens and date of diagnosis was also recorded.

Evidence from breast (27, 28), colon (29), and lung (30) cancer indicates that p53 mutations within certain regions of the gene have a greater influence on protein structure and may markedly modify clinical outcome. Accordingly, each BRCA1-associated p53 mutation was characterized as affecting the zinc-binding domains, which include the L2 and L3 loops, tetramerization domain, or DNA-binding domains including the loop-sheet-helix portion, S2-S21 portion, COOH terminus of the S10 β-strand, the H2 helix, the L3 loop, or DNA contact residues themselves (31). To select a control, an exact match was considered to have been made in the absence of a mutation, confirmed by sequencing the entire open reading frame when both the case and the control were immunonegative or the case and the control were immunopositive. All of the protein truncating p53 mutations were considered equivalent. An identical mutation at an identical amino acid residue was also considered to be an exact match. A near match was recorded if a different amino acid appeared as a mutation at the same residue where mutation was documented in the case or alternatively if mutations at two different residues involved in the same function were paired. For example, all of the missense mutations at the DNA contact residues 241 and 248 were considered to be near matches. Similarly, a mutation at residue 247 within the L3 loop was a near match to a different amino acid change at residue 257, also within the L3 loop of the conserved region of the p53 protein. A poor match was said to have occurred if the case and the control matched a functional with a nonfunctional amino acid change. This was generally only done if it was necessary to maximize the matches of patient age, residual disease, and FIGO stage. For example, we considered it better to match a stage IV mucinous tumor in a 60-year-old with 2 cm of residual disease with a stage IV primary peritoneal serous carcinoma in a 63-year-old with 3 cm of disease containing a different amino acid mutation than to obtain an exact match by pairing the case with a stage I mucinous cancer in a 37-year-old with no residual disease. In other words, attention was directed toward all of the major match parameters rather than just matching on the basis of a p53 mutation.

For the statistical analyses, the two groups of subjects were considered to be independent rather than dependent. Although we sought to individually match cases and controls, the primary purpose of the matching was to produce two independent groups that were similar with respect to multiple factors known to influence disease outcome. We do not believe that the matching procedure resulted in a close pairing of individual subjects that should be retained in the analysis. The distributions of categorical variables in the two groups were compared using the Pearson χ² test. Because CA125 levels were not normally distributed, the two groups were compared using the Mann-Whitney U test. Distributions of other continuous variables in the two groups were compared using the two-sample t test. Time-to-event distributions were estimated using the Kaplan-Meier method, and the two groups were compared using the log-rank test. Deaths in the absence of cancer were treated as censored so that our end point was death because of disease. Deaths from disease have occurred in 38 (64%) of the individuals in both the study and control populations. No patient died of any other cancer. A Cox proportional hazards model was developed to determine the relative impact of all of the study variables on overall survival in a multivariate analysis. Both forward and reverse stepwise modeling was carried out. Confidence intervals were calculated with 95% limits. A two-sided P < 0.05 was required for significance. Calculations were carried out on an IBM compatible computer using the SPSS 10.0 package (SPSS Inc., Chicago, IL)

Tumors were initially stratified on the basis of expected BRCA1 dysfunction versus case-matched controls, which expressed nontruncating BRCA1 sequences as determined by the PTT assay. BRCA1 dysfunction was additionally characterized as secondary to germ-line or somatic BRCA1 mutation, or because of transcriptional silencing of the BRCA1 gene. For each case, a matched control was selected as described in “Patients and Methods.” Table 1 compares the match parameters for the study populations. The overall p53 mutation rate was nearly 80% including a 29% p53 null mutation rate. There were 34 identical matches, 19 near matches, and 6 poor matches on the basis of the p53 mutation match criteria outlined in “Patients and Methods.” The average age at diagnosis was 58.5 (95% confidence limit, 55.3–61.8) years for individuals with BRCA1 dysfunction versus 60.2 (95% CL, 57.3–63.2) years for those without BRCA1 dysfunction (P > 0.05). Eighty-six percent of the cancers and controls were advanced stage cancers (stage III or IV). There was no difference in FIGO stage distribution between subgroups. The cancers were uniformly of high grade, FIGO grade 2 or 3, as only a single grade 1 cancer was found in the entire study population. The predominant histology was serous or adenocarcinoma not otherwise specified. Residual disease averaged 1.98 cm (95% CL, 1.24–2.72) versus 1.81 cm (95% CL, 1.16–2.47) in the study cohorts (P > 0.05). It was noted in a subset analysis that cancers with BRCA1 somatic mutations were cytoreduced to a larger volume of residual disease (3.71 cm; 95% CL, 1.56–5.87); however, this was also true for their matched controls (4.07 cm; 95% CL, 1.94–5.62). In contrast the residual disease volumes for the germ-line BRCA1 or promoter-methylated cancers and their controls varied between 0.95 and 1.51 cm (P < 0.01). Preoperative CA125 levels were highly variable among and between cohorts. However, there were no statistically significant differences in preoperative CA125 levels between cohorts (Mann-Whitney U test). There were no differences in the distribution by site of origin between or among cohorts. Nearly 85% of the study cancers originated in the ovary.

Two other important parameters, which might determine clinical outcome but which were not part of our match criteria, were also evaluated between cohorts: (a) the date of diagnosis; and (b) subsequent chemotherapy regimens delivered. The mean date of diagnosis for the study cohorts varied between November 14, 1994 and December 19, 1994: a difference of only 35 days. As summarized in Table 1, review of the chemotherapy regimens delivered to these patients confirms that all of the study groups were managed identically after cytoreductive surgery, providing reassurances that potential survival differences between cohorts could not be attributed to the year of diagnosis and changing practice patterns. Nearly all of the patients received a platinum containing chemotherapy along with either cyclophosphamide or a taxane. To date, there have been 10 deaths in individuals with somatic BRCA1 tumor mutations and 12 deaths in matched controls. Fourteen deaths have occurred in individuals with germ-line BRCA1 mutations and 15 deaths in matched controls, whereas there were 14 deaths recorded in individuals with BRCA1 promoter methylation and 13 deaths in matched cohorts. There were 4 postoperative deaths, 2 in the BRCA1 dysfunction cohort and 2 in case-matched controls. These events were treated as censored. Therefore, the event frequency and number of censored cases was identical between groups consistent with achievement of an important goal of closely matched study and control populations.

Fig. 1,A compares survival of individuals whose cancers contained what would be expected to be nonfunctional BRCA1 on the basis of mutation, including both germ-line and somatic mutations or promoter silencing with the case-matched, sporadic, wild-type BRCA1 cancers. The median survival was ∼3.5 years (95% CL, 2.5–4.6) versus 4.1 years (95% CL, 2.9–5.3) for the BRCA1 dysfunction and case matched controls, respectively. Fig. 1,B compares survival between the individuals with BRCA1 dysfunction stratified on the basis of the mechanism of BRCA1 dysfunction. Median survival was 2.3 years (95% CL, 1.8–2.8), 2.8 years (95% CL, 2.0–3.6), and 4.5 years (95% CL, 3.8–5.2) depending on whether the origin of the nonfunctional BRCA1 was secondary to promoter-methylation, somatic, or germ-line BRCA1 mutation, respectively. Fig. 1,C compares survival between individuals with germ-line BRCA1 mutations, (median = 4.5 years; 95% CL, 3.8–5.2) and their case-matched controls (median = 4.6 years; 95% CL, 1.6–7.7). Table 2 summarizes the median survival data for all of the BRCA1 dysfunction subsets. In each case, there was no detectable difference in median survival of individuals with BRCA1-dysfunctional ovarian cancers when compared with a carefully matched population of control cancers with detectable wild-type BRCA1 protein by PTT assay. The 10-year survival of ∼25% was virtually the same between groups. Finally, a Cox proportional hazard model was constructed. We entered residual disease, FIGO stage, tumor grade, histology, preoperative CA125, type of p53 mutation, and type of BRCA1 dysfunction into the model. For this multivariate analysis, only residual disease (P = 0.0001), age (P = 0.01), and FIGO stage (P = 0.011) were significant predictors of overall survival. Neither overall BRCA1 dysfunction nor any individual type of BRCA1 dysfunction entered the overall model as a prognostic factor for clinical outcome.

Ovarian cancer is a very heterogeneous disease. Thus, comparative survival studies between cohorts must be carefully controlled to avoid misinterpretations. The present study is the first to control for inactivation of BRCA1 by mechanisms other than germ-line carrier status. It is also the first to control for p53 mutations. Mutation of the p53 tumor suppressor gene represents a molecular prognostic factor, which may influence tumor response to treatment. Altered p53 expression and mutation clearly compromise breast cancer survival (27, 28). Our group (10) and others (11) have shown recently that p53 mutation can influence ovarian cancer survival as well. Furthermore, we have demonstrated that the spectrum of p53 mutation associated with both somatic and germ-line BRCA1 cancers varies from the spectrum seen with sporadic cancers (23). Given these significant molecular prognostic findings, a cleaner answer to the BRCA1 dysfunction ovarian cancer survival question was anticipated based on simultaneously controlling for BRCA1 and p53 mutation status along with the traditional histopathological and demographic variables long known to influence ovarian cancer outcome. To this end, we selected eight variables, each known to independently modify ovarian cancer outcome to construct case-control matches. No other study has attempted to control for more than four variables.

At least two hypotheses, which are not mutually exclusive, could reasonably be invoked to explain a survival advantage for individuals carrying germ-line mutation of the BRCA1 gene. Given a potentially important role for BRCA1 in the DNA repair process (32), some have speculated that traditional cornerstone chemotherapeutics, such as the platinum agents, could demonstrate selectively enhanced cancer cell killing (14). In addition, or alternatively, germ-line BRCA1 mutation in association with loss of the wild-type allele, may result in ovarian cancer by either accelerating or by-passing one or more additional carcinogenic events such as loss of other tumor suppressor genes or up-regulation of various oncogenes. Such a mechanism would explain the tendency for hereditary BRCA1-related ovarian cancers to occur in younger-aged women. These observations suggest that a comparative survival analysis of germ-line BRCA1-related ovarian cancers could be enhanced by considering outcomes in women with BRCA1 inactivation occurring through mechanisms other than inheritance. Such mechanisms include both somatic tumor BRCA1 mutation and gene silencing by promoter hypermethylation. Unknowingly including differing proportions of cancers with BRCA1 inactivation through these alternative mechanisms in control cohorts of sporadic cancers may explain the conflicting BRCA1 survival data in the literature on the basis of happenstance.

So does ovarian cancer BRCA1 mutation alter survival? Table 3 places findings of the current study into perspective with five studies published previously that have also attempted to address this question of potential clinical significance. The answer appears to depend both on the population studied and the selection process used to identify the control cohorts.

Other than the report by Boyd et al. (14), the present study is the only one to correlate ovarian cancer BRCA1-related survival by evaluating a population treated and tested at a single institution. Four investigations have a potential selection bias in that they use familial registry cases (12, 15, 16, 17) or have done limited BRCA1 screening in a single ethnic group only for common founder mutations (14). The current study is the only one based on consecutive cases wherein all of the cancers were screened for BRCA1 null mutation throughout the BRCA1 open reading frame. Consequently, we are able to report outcomes of germ-line BRCA1 ovarian cancers for a study population, which included 2 African-Americans, 2 Ashkenazi Jews, and 20 cancers in individuals of Central and Northern European extraction. A total of 30 different BRCA1 mutations including 19 different germ-line BRCA1 mutations have been sequenced in this group (18, 21, 26). This is a larger variety of mutations than reported in any other survival analysis. However, it is likely that the study by Pharoah et al. (17) contained at least this many.

Caution must be observed comparing between studies because of population differences. The reported median survival for BRCA1 mutant ovarian cancers varies from a low of 1.7 years (17) to a remarkably high 9.6 years (15). The latter is nearly three times that reported by Boyd et al. (14) for stage III and IV cancers in Ashkenazi Jews carrying one of two founder mutations. It is one and a half times that reported by Rubin et al. (12) in a cohort that also included borderline ovarian cancers and contained a 20% incidence of early stage disease. Using our carefully controlled BRCA1 dysfunction population and case-matched controls, BRCA1 dysfunction did not alter ovarian cancer outcome. Median survival was 3.5–4.1 years for both groups, essentially identical to the Boyd finding for Ashkenazi Jews with founder BRCA1 mutations (14). However, Boyd et al. (14) did not control for the incidence and type of p53 mutations, nor did they rule out other cancers with BRCA1 dysfunction in their sporadic ovarian cancer cohort. Sample size in the present study is relatively small and does not provide the statistical power to detect small survival differences. Application of the method of Freedman (33) to our BRCA1 dysfunction cohort with a median survival of 3.5 years allows us to estimate the magnitude of the differences that could be detected with 80% power based on a two-sided test at the 5% level of significance. The median survival of the control cohort would have to be no longer than 2.2 years. Similarly, based on the observed median survival of the control cohort of 4.1 years, the median survival of the BRCA1 dysfunction cohort would need to be increased to at least 8.7 years. These results assume that survival is exponentially distributed and that the two groups are compared after 6 years of follow-up. Therefore, we were adequately powered to detect the magnitude of survival advantage reported in two (12, 15) of the three positive studies (12, 14, 15).

We found nearly twice the median survival for individuals with germ-line BRCA1 mutations compared with individuals with ovarian cancers with BRCA1 dysfunction secondary to somatic mutation or promoter hypermethylation. However, there was no survival difference between these individual cohorts and their case-matched controls. The apparent survival difference between cohorts may have been secondary to a greater volume of residual disease after surgical cytoreduction in the somatic BRCA1 mutant group (P < 0.01) and a higher frequency of p53 null mutations (P = 0.14) in the group with BRCA1 promoter hypermethylation. These comparisons clearly stress the importance of comparing similar populations. Taken together, these data suggest that whereas BRCA1 dysfunction may influence overall survival in selected populations, like so many other factors, it is not the single ovarian cancer prognostic driving force behind favorable or unfavorable actuarial survival for ovarian cancer individuals with hereditary ovarian cancer.

We thank Sara McClain, Jenny Rathe, Amara Lucke, Matthew Buller, and Lisa Blake for technical assistance. Rebecca Sandersfeld and Linda Sanders provided assistance with manuscript preparation and patient follow up respectively. Drs. Barrie Anderson, Joel Sorosky, and Anil Sood provided tissue samples from their patients for study.