Mutations in the p53 gene are extremely common in epithelial ovarian cancers, particularly those with serous histology. Recent analyses of prophylactic salpingo-oophorectomy specimens from BRCA1 carriers show that the acquisition of these mutations may be a very early event in the development of many cases [3, 21, 22]. Given the prevalence and potentially pivotal nature of these mutations in the etiology of the disease, we sought to define transcriptional patterns associated with p53 mutational status in a series of early and advanced serous ovarian cancers. Since all cancers are of the serous histology, this should serve to reduce false associations that might arise in a study of mixed histologic types where the prevalence of p53 mutations varies, i.e., patterns of expression that are related to histology rather than the disruption of p53 activity.
In a total of 89 cancers, we sequenced the entire p53 coding region and performed expression analysis using Affymetrix U133A arrays. We categorized the cancers by stage (40 stage I/II and 49 stage III/IV cancers). TP53 mutational status was considered as a binary (mutations that alter the coding sequence versus wild type) and ternary (premature chain terminating mutation, missense, and wild type) variable to include possible differences in biology between the types of mutation. The overall mutation rate of 66% in this series is comparable to previous studies that have sequenced the full coding sequence of p53 in ovarian cancers [1, 2, 8, 23–25]. Analyzed by stage, mutations were more prevalent in the early stage cancers (77% versus 63%) and the early stage cancers had a significantly higher incidence of null mutations (38% versus 8%). This is consistent with the report by Leitao et al. that focused on early stage ovarian cancers . In 21 early stage serous cancers, 14 (67%) harbored mutations and of these, 9 were chain terminating. Therefore, while the mutation prevalence between early and late stage serous cancers is comparable, the rate of null mutations is significantly different (p < 0.001 for the combined data from Leitao et al. and our current study). Since mutations are not likely to be lost during disease progression, the high frequency of null mutations in the early stage cancers may be an indication of differences in molecular pathogenesis relative to advanced cancers. Whether this is also related to the more favorable clinical outcome of early stage cancers is unknown.
We broadly categorized p53 mutations as either premature chain terminating (null) or missense in our gene expression analysis. Beyond the incidence of these mutations in early and advanced ovarian cancers, there is a significant amount of evidence indicating that increased stability of missense p53 mutant proteins and subsequent accumulation in the nucleus leads to various transcriptional and biological consequences over and above what is seen when the protein is absent (as is the case with most chain terminating mutations) [26–28].
Several studies correlated the type of p53 mutation with clinical variables of epithelial ovarian cancers. Sood et al. examined the presence of distant metastases (parenchyma of the liver or spleen or extra-abdominal) in relation to p53 mutations . The most significant finding was that distant metastasis was 8-fold more common in patients with cancers that carried null mutations compared to those with either missense mutations or wild type p53. In addition, Shahin et al. found that ovarian cancers with null mutations and functionally null tumors (based upon lack of p53 immunostaining) had the worst prognosis . Therefore, two independent studies found that null mutations, at least in advanced stage cancers, seem to confer a more aggressive biology related to metastasis and outcome. While our current study was not specifically designed to test disease outcome, the advanced stage cancers were derived from women who survived less than 3 years after initial diagnosis (n = 26) and those who survived greater than 7 years (n = 23) . Comparing mutation status to outcome in our advanced stage cases, the overall prevalence of mutations was significantly higher in the short term survivors (81% versus 30%) and both the long and short term survivor groups contained 2 cancers with null mutations. Therefore, while mutation status does correlate with survival in our series of advanced stage cancers, the primary difference is in the rate of missense mutations. A consistent and reproducible association between p53 mutations and ovarian cancer outcome remains elusive.
The primary goal of the current study was to determine whether a gene expression signature related to p53 status exists in ovarian cancer and could provide molecular insight into the disease. To our knowledge, this is the first exercise of its type applied to epithelial ovarian cancer but similar studies have been published for other cancers. Further, there is a large literature on transcriptional targets of p53 activity. Therefore, our study can be placed judiciously into this broader context. In the present study, using standard analytic approaches, gene lists of varying significance were derived between groups of tumors based on p53 status. A comparison of TP53 wild type versus mutant ovarian cancer of all stages yielded a set of 9 differentially expressed genes (at the p < 0.05 level using a relatively stringent false discovery criterion). Within sample predictions using these 9 genes yields an accuracy of 86.5%. Of these 9 genes, only DDB2 has other experimental evidence indicating that it is a bona fide p53 target. Transcription of DDB2 is activated by human p53 protein, the promoter element interacts with p53, it is part of the xeroderma pigmentosum complementation group E (XPE), and is involved in DNA damage recognition and repair [17, 29–32]. Further, this is one of the 52 genes described by Troester et al. that discriminates breast cancers with p53 mutations. This gene appears to be a highly plausible candidate for membership in a cancer associated p53 gene expression signature.
With no appropriate validation set yet available, the strength of this overall predictor of TP53 mutation status cannot be evaluated directly. However, we do present a series of other predictors that can be tested within our data set and related to other published p53 signatures. In particular, the most robust approach within the constraints of the current study is to compare signatures derived in early versus late stage disease. Since all cancers were of serous histology, with relatively equal distribution by grade, the biologic consequences of a p53 mutation should be similar. Therefore, a signature derived on early or late stage cancers and then applied to the remaining cohort for "validation" should provide insight into the strength of the predictor. We observed minimal overlap in the gene sets that constitute these individual predictors (Table 5) presaging a failure to validate. Indeed, for three out of four of these predictors, accuracy in the validation sets hovered around 50%. However, testing the gene set derived from early stage cancers that were categorized by binary p53 status on the set of advanced cancers resulted in predictions with 86% accuracy. Examination of the genes constituting this predictor provides further support that this is a biologically meaningful result (Table 6). Of the 42 genes in this list, 6 (DDB2, CDC2, TRIP13, ACTR2, PCR1, and SAFB) have either been described in a p53 related breast cancer signature or implicated as p53 target genes by other means. For comparison, there is a single gene (MYBL2) in common between the two published p53 breast cancer signatures and no genes in common between either of the breast signatures and a similarly derived colon cancer signature [9, 10, 33].
The primary confounder in discerning a clear expression signature of TP53 mutation may be the fundamental importance of this pathway in the development of serous ovarian cancers. In the current series, two thirds of the cancers harbored p53 mutations. The question is whether the remaining third have other alterations that result in inactivation of downstream p53 functions. Numerous mechanisms have been described that may impact on the activity of p53 and there is no widely accepted metric for evaluating this parameter, particularly from a frozen tissue sample. Gene expression patterns indicative of mutational status in ovarian cancers are discernible, but they are not highly predictive in out of sample validation with the exception of a signature developed specifically on early stage cancers. However, a number of genes within this signature have a high degree of biologic plausibility. The confirmation of several of these genes in a p53 related breast cancer signature suggests that a core set of expression markers could be developed to assess p53 activity in multiple cancer types.