Transdominant ΔTAp73 Isoforms Are Frequently Up-regulated in Ovarian Cancer. Evidence for Their Role as Epigenetic p53 Inhibitors in Vivo

TP73 is a new family member of TP53, one of the most important tumor suppressor genes in human cancer. Significant structural homology exists between p73 and p53. The human full-length p73, mainly TAp73α and TAp73β (two COOH-terminal splice variants) largely mimic p53 function in experimental systems. p73 activates many of the p53 cellular target genes and is capable of inducing cell cycle arrest and apoptosis (1, 2, 3). Moreover, endogenous TAp73 is able to integrate several p53-dependent death stimuli. For example, DNA-damaging agents, such as cisplatin and γ-ionizing radiation (4, 5, 6) as well as oncogenes like E2F1, cMyc, and E1A (7, 8, 9, 10) have been found to activate endogenous TAp73.

However, the roles of TP53 and TP73 in mammalian tumorigenesis seem to be fundamentally different. In sharp contrast to TP53, the virtual absence of inactivating mutations in over 1,100 human tumors screened to date (11), combined with tumor-associated overexpression of wild-type TP73 in many different types of human cancers as well as the lack of a cancer phenotype in the TP73 knockout mouse (12), are all inconsistent with a suppressor function. Moreover, the classic viral oncoproteins SV40 T antigen, Ad E1B 55 kDa, and human papillomavirus E6, which all target and inactivate p53 during host cell transformation, do not target the TAp73 protein physically or functionally (reviewed in Ref. 13). One possible explanation for the different roles of p53 and p73 in tumorigenesis lies within the different genomic organization of these genes. Whereas p53 encodes one protein, TP73 gives rise to several different NH₂ (and COOH-terminal) isoforms with opposing functions. The discovery of an NH₂-terminally truncated, transactivation-deficient dominant-negative p73 isoform in the mouse shifted the interest toward the p73 NH₂ terminus (14). This so-called ΔNp73 protein plays an essential antiapoptotic role during development of the central and autonomous nervous system by counteracting developmental p53-mediated neuronal death. Aside from the proapoptotic, transactivation-competent TAp73 with a putative suppressor role, four different NH₂-terminally truncated isoforms (ΔNp73, ΔN′p73, Ex2p73, and Ex2/3p73) have now been found in human cancers and/or cancer cell lines (see Fig. 1 A). Each of these isoforms lack all or most of the transactivation domain and therefore are collectively called ΔTAp73. ΔTAp73 products are generated either via alternative exon splicing of the P1 promoter transcript (producing ΔN′p73, Ex2p73, and Ex2/3p73; Refs 15, 16, 17, 18) or via alternative use of the P2 promoter in intron 3 (producing ΔNp73; Ref. 19). The Ex2p73 and Ex2/3p73 isoforms lack either exon 2 or exons 2/3, respectively. Of importance, the transcripts ΔNp73 and ΔN′p73 encode the same protein product. ΔTAp73 isoforms fail to induce cell cycle arrest and apoptosis (19, 20, 21). Their biological importance might lie in the fact that ΔTAp73 proteins retain their DNA-binding and tetramerization competence and thus can act as powerful dominant-negative inhibitors of both wild-type p53 and TAp73 in primary and transformed cells (15, 17, 18, 19, 22).

Evidence is mounting that ΔTAp73 might indeed act as a biologically relevant oncogene in primary human cancer. In cultured human cancer cells, ΔNp73 inhibits the suppressive activity of p53 and TAp73 in colony formation assays (19). ΔTAp73 overexpression results in malignant transformation of immortalized NIH3T3 fibroblasts in vitro that produce tumors in nude mice (18). ΔNp73 promotes immortalization in primary cells and cooperates with oncogenic Ras to drive their transformation in vivo (23). Also, ΔTAp73 could mediate hyperphosphorylation of Rb, resulting in enhanced E2F activity, opening the possibility that ΔTAp73 might also interfere with the Rb tumor suppressor pathway (24). Lastly, in the first clinical study on the impact of ΔNp73 expression in human cancer, ΔNp73 was an independent prognostic marker for reduced progression-free and overall survival in neuroblastoma patients (25).

Given these findings plus the overexpression of p73 in human tumors, the possibility arises that in fact the dominant-negative ΔTAp73 isoforms (rather than TAp73) might be the physiologically relevant components of p73 overexpression in tumors, functionally overriding an accompanying increase in TAp73 expression. The existence of this inhibitory family network could readily explain the paucity of p73 mutations in human tumors. Of note, many of the previous p73 overexpression studies in human cancers did not use primer pairs or antibodies specific for TA/ΔTAp73 isoforms and therefore determined the composite sum of p73 levels derived from all variants. Therefore, up-regulation of ΔTAp73 forms might have contributed to the elevated “p73” levels found previously in human cancers. Thus far, only a few studies, encompassing a very limited number of tumors, focused on the specific expression of ΔTAp73. In the largest such study that had taken place previously, we reported that ΔNp73 transcripts are overexpressed in 73% of 37 malignant gynecological tumors, compared with the patients’ matched normal tissues of origin. Moreover, 31% of 52 breast cancers overexpressed ΔNp73 compared with normal breast tissue (19).

Therefore, a comprehensive and detailed survey of expression profiles is urgently needed to clarify if tumor-specific up-regulation of ΔTAp73 isoforms is a clinically important phenomenon in human cancers. This knowledge is critical for our understanding of the true role of TP73 in tumorigenesis. We therefore undertook this study, the largest and most detailed to date, in primary tumors and determined expression levels of all ΔTAp73 and TAp73 products. Using isoform-specific quantitative real-time reverse transcription (RT)-PCR, we analyzed 100 primary ovarian carcinomas and compared their various isoform mRNA levels to a group of 48 normal tissues. The results greatly substantiate the significance of deregulation of these dominant-negative inhibitors in human cancer and provide further support for an oncogenic role of TP73 in tumor formation.

Ovarian cancer tissues from 100 patients were collected during surgery at the Department of Obstetrics and Gynecology, University of Innsbruck, Austria, in compliance with and approved by the Institutional Review Board. Their histological subtypes are given in Table 1. Tissues were immediately snap frozen in liquid nitrogen, pulverized in the frozen status, and stored at −80°C until usage. Normal ovarian control tissues from 16 patients were obtained from the Southern Division of the Cooperative Human Tissue Network (Birmingham, Alabama). An additional 32 normal tissues were collected during surgery at Stony Brook University in compliance with and approved by the Institutional Review Board.

RNA was extracted from tissue using the RNAeasy Mini Kit (Qiagen Sciences, Germantown, MD). High RNA quality was confirmed by Tris-acetate-EDTA agarose gel electrophoresis (Fig. 1 B). Total RNA (5 μg) was reverse transcribed using the Thermo Script RT-PCR system (Invitrogen) and random primers.

Primer pairs for TAp73, Ex2p73, and Ex2/3p73 were designed using Gene Runner version 3.05 and Primer Picking Software (http://www.genome.wi.mit.edu7cgi-bin/primer/primer3_www.cgi). Several different primer pairs for each of these isoforms were designed and tested before the optimal ones were chosen for all additional experiments. Primers for ΔNp73 and ΔN′p73 were used as described previously (18).

All primer pairs detected a unique specific cDNA (see Fig. 1,A). For TAp73, Ex2p73, and Ex2/3p73, primers spanning splice junctions were used. For ΔNp73, the upstream primer detected a sequence unique for this transcript (5′ untranslated region of exon 3′). Specificity for ΔN′p73 was achieved by the unique combination of the upstream and downstream primer. Specificity of each primer pair was confirmed by detecting single bands of amplicons on agarose gels (Fig. 1,C) and by sequencing the PCR products. Also, all primers were queried against the nonredundant Human Genome Database (National Center for Biotechnology Information). In no case were unrelated gene matches found. Although two nonpaired primers matched to the related TP63 gene, their respective return primer did not. Table 2 lists all primer pairs for the five examined p73 isoforms and for the internal normalizer 28S rRNA.

Absolute copy numbers for all isoforms were determined by real-time RT-PCR using the SYBR Green Dye method (Qiagen Sciences). 28S rRNA was used to normalize samples (26). PCR reaction was performed in a 20-μl reaction mixture containing 1/32 of the reverse transcription reaction, 10 μl of QuantiTect SYBR Green PCR Master Mix, and 6.25 pmol of each primer. The Master Mix is a premixed solution containing HotStartTaq DNA Polymerase, QuantiTect SYBR Green PCR buffer, deoxynucleoside triphosphates, SYBR Green I, and ROX (a passive reference dye). The PCR conditions were as follows: initial melting at 95°C for 15 min followed by 39 cycles of 94°C denaturation for 30 s; annealing for 30 s; elongation at 72°C for 1 min; and plate reading for 10 s (plate reading temperature was set 2.5°C below the pre-established melting temperature of each PCR product). Annealing temperatures were optimized by running temperature gradient PCR reactions (Fig. 1,C). Annealing and melting temperatures for each PCR product are shown in Table 3. Melting curves for each PCR product were then generated; an example is shown in Fig. 1 D. Data were collected using the DNA Engine Opticon (MJ Research, Waltham, MA), a 96-wellplate-based continuous fluorescence detection system. Each sample was tested in duplicate, and a negative control was included on every plate. An increase in fluorescence above baseline indicated the accumulation of a specific PCR product. Its threshold was set above the baseline and in the exponential phase of the PCR. The computed tomography value was defined as the cycle number at which the fluorescence crossed the threshold.

To determine the absolute copy numbers of p73 isoforms, PCR products for each isoform were cloned into the pCR II vector using the TA Dual Promoter Cloning Kit (Invitrogen). Constructs were verified by sequencing. A pBR322-based vector containing the 28S rRNA cDNA insert was a gift from G. Zambetti (St. Jude Children’s Research Hospital). Cloned plasmids were linearized, and their concentrations were carefully measured by UV spectrophotometry to allow copy number determination. Ten-fold serial dilutions were made ranging from 10 to 10 million copies for each p73 isoform plasmid and from 1000 to 1000 million copies for the 28S rRNA containing plasmid. Our assay allowed the reproducible detection of as low as 10 copies of p73 mRNA transcripts for each isoform; thus the dynamic range of the standard curves spanned at least seven orders of magnitude (Fig. 1 E). After normalization of each sample to its own 28S rRNA transcript level, data were expressed as absolute copy numbers per 154 ng of total RNA.

Frozen pulverized tumor tissue was suspended in 1.5 ml of TENN buffer [50 mm Tris, 5 mm EDTA, 150 mm NCL, 0.5% NP40 (pH8.0)] containing a protease inhibitor mixture (Roche) and homogenized by sonication. Lysates were centrifuged twice at 14,000 rpm for 15 min, and supernatants were loaded onto 10% SDS-PAGE gels. Membranes were immunoblotted with the p73β antibody GC15 (Calbiochem) at 1:40. For comparison, a normal tissue pool was generated by mixing equal ratios of 12 normal ovarian tissues. Normal ovarian tissue samples were also analyzed separately and found to show little differences in expression level from sample to sample.

To detect inactivating mutations in the p53 gene, the functional yeast-based assay was used as described previously (27). Briefly, the functional analysis of separated alleles in yeast is based on transcriptional activity of wild-type versus mutant p53 alleles in a yeast reporter system. p53 mRNA species from tumors were reverse transcribed, amplified by PCR, and cotransformed into Saccharomyces cerevisiae together with a linearized yeast homologous recombination expression vector carrying the 5′ and 3′ ends of the p53 open-reading frame. Wild-type p53, which activates transcription of the yeast ADE2 gene that encodes the phosphoribosylaminoimidazole carboxylase results in white colonies, whereas mutant alleles lack transcriptional activity and result in smaller, red colonies.

The data distribution of normalized expression levels for both the ovarian cancer group and the normal tissue group was non-Gaussian. Therefore, nonparametric tests were applied for data analysis. The fold up-regulation of each p73 isoform in tumors was calculated in reference to the median expression level of the entire normal tissue group. Up-regulation in a given tumor was defined as expression levels above the 75th percentile of median expression level of the normal tissue group. Correspondingly, down-regulation in a given tumor was defined as expression levels below the 25th percentile of the median expression level of the normal tissue group. In all classes of p73 isoforms, a few cases fell into the upper or lower “outlier” group in both the normal and the tumor group. The latter was defined as 1.5-fold above or below the 75th and 25th percentile values, respectively, and are indicated (Fig. 3,A, black circles). The Mann-Whitney U test was used for comparing p73 isoform expression of tumor and normal tissues and for correlation analysis of p73 isoform expression with p53 mutational status (Figs. 3,A and 5). For correlations among the various p73 isoforms, data were logarithmized that produced a Gaussian distribution, and the Spearman correlation test was applied. P values < 0.05 were considered statistically significant.

Clinical data of 96 ovarian carcinoma patients were available. All patients were treated at the Department of Obstetrics and Gynecology, University of Innsbruck, Austria, between May 1991 and June 2001. Ninety-four of the 96 patients underwent tumor resection, whereas two patients were treated with carboplatin monotherapy only because of advanced age and impaired general health. Surgical treatment included total hysterectomy and bilateral salpingo-oophorectomy in all but two patients. Eighty of the 94 patients underwent additional omentectomy, and partial bowel resection was necessary in 13 patients. Lymph node evaluation involved pelvic lymphadenectomy in 37 of 94 patients.

The median age at the time of diagnosis was 62 (range, 24 to 90) years. Patients presented with International Federation of Gynecologists and Obstetricians’ stage I (n = 20), stage II (n = 5), stage III (n = 58), and stage IV (n = 11) at the time of diagnosis. In two cases, the International Federation of Gynecologists and Obstetricians’ stage was unknown. To classify tumor grade, the widely used system based on architectural and nuclear grade and mitotic activity was applied. Distant metastases were seen in 11 patients at the time of diagnosis [lung (n = 1), pleura (n = 5), liver (n = 3), skin (n = 1), and others (n = 3)]. With the exception of seven patients with stage I carcinomas, all patients received adjuvant platinum-based chemotherapy. This involved four to six courses of carboplatin monotherapy (n = 24) or a combination therapy of cisplatin and cyclophosphamide (n = 35); cisplatin, carboplatin, and cyclophosphamide (n = 9); carboplatin and paclitaxel (n = 20); and cisplatin, paclitaxel, and cyclophosphamide (n = 1). During follow-up, 42 of 96 patients suffered recurrent ovarian cancer, and 54 of the 96 patients died. The median time of follow-up was 43 months (ranging from 3 to 154 months); the mean time of follow-up was 80 ± 30 months.

Survival probabilities were calculated by the product limit method of Kaplan and Meier. Differences between groups were tested using the log-rank test. The results were analyzed for the end point of overall and recurrence-free survival. Overall survival was defined as the time between diagnosis and death, regardless of the cause. Patients who had not died were censored at the last date they are known to have been alive. Recurrence-free survival was calculated from the day of diagnosis until the date when progressive disease, relapse, or death was reported, whichever occurred first. Patients who had not experienced any unfavorable event were censored at the last date they were known to have been alive. The Cox proportional hazards model was used for multivariate analysis to assess the independence of different prognostic factors.

After validation of the isoform-specific real-time RT-PCR (examples are shown in Fig. 1), we determined the p73 expression profiles of TAp73 and all four ΔTA isoforms in 100 primary ovarian carcinomas and compared them to 48 normal ovarian and other tissues (Fig. 3,A, linear scales, left, and logarithmic scales, right). Of note, we showed previously in 50 different adult and fetal tissues including ovaries that total p73 levels varied only minimally (<4-fold; Ref. 28). All data were normalized to the tissues’ own internal control of 28S rRNA cDNA. Moreover, the 28S rRNA cDNA levels in tumors were very similar compared with the normal tissue group (Fig. 3 A, top box plot). Therefore, the raw data changed only slightly after normalization for 28S rRNA. Absolute copy numbers, expressed per standard input total RNA (154 ng), were then calculated for all isoforms.

Using this comprehensive data set, we first analyzed whether the entire five-variable set of p73 isoforms correlates with the presence or absence of disease. A few individuals were excluded because of incomplete data sets for all five variables, leaving 94 patients and 45 normal controls. To this end, a correlation matrix was calculated as follows: first, patients and controls were combined into one group; next, each individual (i) in the combined group was assigned a five component vector V(i), composed of v_1i = TAp73, v_2i = ΔN′p73, v_3i = ΔNp73, v_4i = Ex2p73, and v_5i = Ex2/3p73. To look for similarities or dissimilarities within that combined group, a correlation matrix was calculated. This quantifies the similarity of parameters v_1–5, between every possible pair of individuals (i) and (j). Fig. 2,A shows the contour plot of the resulting correlations that are higher than 0.990. Axes X and Y plot individuals for pairwise comparison (positions 1–94 are patients, ordered by increasing survival time from left to right or bottom to top; positions 95–139 are normal controls). As is clearly apparent in Fig. 2,A, the set of p73 isoform values within the control group are strongly intercorrelated (top right block, characterized by solid contours). This quantifies the low person-to-person variation in levels of TAp73, ΔN′p73, ΔNp73, Ex2p73, and Ex2/3p73 isoforms within the control group. In contrast, the patient group is characterized by a high person-to-person variability in levels of TAp73, ΔN′p73, ΔNp73, Ex2p73, and Ex2/3p73 (lower left block). The high individuality of isoform levels within the patient group is revealed by the scattered “dot” character of the in-group correlations. Moreover, the patient group is uncorrelated with the control group, as revealed by virtually empty off-diagonal areas in positions 94–139. Interestingly though, the patient matrix does reveal a higher similarity in the p73 isoform contours within those patients who survive longest, compared with the remaining patients with shorter survival (subgroup of patients 70–94; see small top right subset within the patient block). Only nine patients exhibit some similarity to the control group (Fig. 2 A, see the off-diagonal contours in the 95–139 area), indicating that in ∼10% of cases, assaying for the set of p73 isoform levels alone might not reveal the presence of disease without additional indicators. Taken together, values of the five-variable set of p73 isoforms alone, without any additional clinical or pathological information about a person, clearly are able to discriminate the presence of ovarian cancer from its absence in about 90% of individuals.

To further analyze if there are interdependencies among the five-variable set of p73 isoforms within patients, the Principal Component Analysis with Varimax rotation was used, which is an established method of variance analysis. This analysis reveals that there are only three independent and statistically significant bits of information within the five components of the assay results (v_1i = TAp73, v_2i = ΔN′p73, v_3i = ΔNp73, v_4i = Ex2p73, and v_5i = Ex2/3p73; Fig. 2 B). TAp73 and ΔN′p73 represent one independent component, ΔNp73 represents a second component, and Ex2/3p73 and Ex2p73 represent the third component. This statistical grouping correlates well with the predicted molecular grouping that derives from the two promoters and the gene architecture.

Next, individual isoforms were analyzed. Consistent with previous reports in other tumor types, significant tumor-specific up-regulation was found for TAp73 (Fig. 3 A, second row box plots). In contrast, expression levels of p73 in normal tissues were very low albeit somewhat scattered, with a median of 34 copies of TAp73 (P < 0.05). The median up-regulation in tumors was 11-fold (median of 357 copies of TAp73) compared with normal tissues, with a range up to 233-fold above the median normal. Thus, TAp73 up-regulation, defined as tumor levels ranging above the 75th percentile of normal tissue levels, was seen in 35% of ovarian cancers.

We next measured ΔN′p73. This isoform contains an additional 198 bp of the 3′ portion of exon 3′ via an aberrant splice from exon 3 into exon 3′ (Fig. 1,A). This leads to an upstream premature stop, but a second translational start site within exon 3′ is productive. The ΔNp73 transcript contains an additional 78 bp region at the 5′ end of exon 3′ that is missing from ΔN′p73. However, at the translational level, both ΔNp73 and ΔN′p73 transcripts encode the same protein product, for which an oncogenic potential has been described (15, 18). Significantly, in our series of 100 ovarian cancers we found an almost universal prevalence of tumor-specific up-regulation of ΔN′p73, compared with normal tissue (P < 0.05; Fig. 3 A, third row box plots). In the normal tissue group, the median copy number of ΔN′p73 transcripts was zero, and the range was rather narrow. In fact, in 59% (28 of 44) of normal tissues, no ΔN′p73 transcripts could be detected. In tumors, the median ΔN′p73 copy number was 65, ranging up to 1,325 copies. Thus, up-regulation of ΔN′p73, defined as tumor levels ranging above the 75th percentile of normal levels, was present in 95% (94 or 99) of ovarian cancers.

In contrast, tumors showed no median up-regulation of ΔNp73, although some individual tumors did exhibit increased levels (Fig. 3 A, fourth row box plots). The median tumor level was similar to the median normal level (15 copies in tumors versus 12 copies in normal tissues). Only 26% (26 of 100) of tumors exhibited up-regulation of ΔNp73 (above the 75th percentile of the normal group). Within this up-regulated subgroup of tumors, the median up-regulation was 17-fold, with levels ranging up to 246-fold above the median normal. In contrast, the median up-regulation in normal tissues above the 75th percentile was only 11-fold. Within tumors, 6% (6 of 100) showed ΔNp73 expression levels that exceeded the highest ΔNp73 values found in the normal group.

Tumors did not up-regulate Ex2p73 (Fig. 3 A, bottom row box plots). Although the median level in the tumor group was slightly lower (2-fold) than in the normal group, this difference was not significant (21 copies in tumors versus 46 copies in normal, P = 0.132). None of the ovarian carcinomas expressed Ex2p73 levels that exceeded the highest normal value.

For Ex2/3p73 transcripts, tumors exhibited strong and significant down-regulation compared with normal tissues (median of 393 copies in tumors versus 3162 copies in normal; P < 0.05; Fig. 3 A, fifth row box plots). Interestingly, Ex2/3p73 showed a very wide range of expression within the normal group, varying 7,027-fold from the lowest to highest level measured. Median expression levels in the normal group were 8-fold higher than in the tumor group. Down-regulation (below the 25th percentile of the normal group) was seen in 57% (57 of 100) of tumors with a maximum down-regulation of 132-fold below the mean normal.

Thus, taken together, the strongest and most prevalent tumor-specific increase in expression was seen for ΔN′p73, with the great majority of tumors showing up-regulation. In ovarian cancer, this parameter stands out among all other p73 isoforms and serves as a molecular marker of this disease. About one-third of the tumors also exhibit concomitant up-regulation of the antagonistic TAp73. Only a small subgroup of tumors showed ΔNp73 up-regulation. In contrast, Ex2p73 and Ex2/3p73 levels either tended to be slightly lower in tumors (Ex2) or showed significant tumor-specific down-regulation (Ex2/3).

To verify that tumor transcript levels translated into respective protein levels in tumors, select tumors were analyzed by Western blot with the TAp73β-specific antibody GC15. Among all available p73 antibodies, GC15 is the most sensitive. As shown in Fig. 3 B, protein levels correlated semiquantitatively with RT-PCR data for TAp73. On a transcript level, tumor 1 showed no overexpression, whereas tumors 2 and 3 showed strong and moderate overexpression of TAp73 transcripts, respectively. However, the complete set of NH₂-terminal p73 isoforms cannot be measured on a protein level because no specific antibodies of sufficient quality for any of the ΔTAp73 forms currently exist.

When individual tumors were analyzed for coexpression of various NH₂-terminal p73 isoforms, significant co-up-regulation of isoforms was found in tumors (P < 0.05; Fig. 4). In contrast, only select correlations were present within the normal tissue. In tumors, the strongest correlation existed between TAp73 and ΔN′p73 (Fig. 4, top left panel). Here, the higher the TAp73 expression level in a given tumor, the higher the ΔN′p73 expression level within the same tumor (Pearson correlation coefficient r = 0.829; P < 0.001). This was also the case for the relationship between TAp73 and ΔNp73, albeit this correlation was not as strong (Fig. 4, second left panel; Pearson correlation coefficient r = 0.634; P < 0.001). Nevertheless, 17 of 21 tumors with the highest ΔNp73 levels also exhibited high TAp73 levels; in addition, 8 of these 17 tumors were wild type for p53 (see below). The weakest correlations existed between TAp73 and Ex2p73 or Ex2/3p73 (third left panel and data not shown).

In contrast to tumors, normal tissues showed no correlation between TAp73 and ΔN′p73 expression levels, supporting the significance of tumor-specific co-up-regulation (Fig. 4, top right panel). Also, no correlation existed between ΔN′p73 and Ex2p73 or Ex2/3p73 levels. However, in normal tissues, Ex2p73 and Ex2/3p73 levels correlated with each other, and ΔNp73 levels correlated with ΔN′p73, Ex2p73, and Ex2/3p73 levels (data not shown).

In summary, correlation analysis for coexpression of p73 isoforms within the tumor and normal groups highlights the difference in their expression patterns. As shown in Fig. 4, the correlation between TAp73 and ΔN′p73 is positive for tumors but negative for normals; the correlation between TAp73 and ΔNp73 is positive for tumors and for normals; the correlation between TAp73 and Ex2p73 is positive for tumors but negative for normals; and the correlation between ΔN′p73 and Ex2/3p73 is positive for tumors but negative for normals.

We determined the p53 mutation status in our tumor collection via a functional yeast-based assay. Of the 100 cases, we could ascertain the genotype in 93% (93 of 100). Of these, 74% (69 of 93) of tumors were identified as mutant p53, and 26% (24 of 93) were found to be wild type for p53. Of note, significantly higher expression levels of p53-inhibitory ΔTAp73 isoforms were found in p53 wild-type tumors, compared with p53 mutant tumors (Fig. 5). This was the case for ΔNp73, ΔN′p73, and to a lesser extent Ex2/3 p73 (P < 0.05 for all isoforms). Moreover, the prevalence of ΔNp73 and ΔN′p73 up-regulation (above the 75th percentile of the normal group) was higher in p53 wild-type tumors than in mutant tumors (Table 4). Although 23 and 82% of p53 mutant tumors exhibited up-regulation of ΔNp73 and ΔN′p73, 48% and 96% of p53 wild-type tumors exhibited up-regulation of ΔNp73 and ΔN′p73, respectively. On the other hand, no difference in the prevalence of TAp73 up-regulation existed between wild-type and mutant tumors (Table 4). However, in keeping with the isoform interrelatedness shown in Fig. 4, wild-type tumors expressed higher levels of TAp73 than mutant tumors (Fig. 5). No significant difference in Ex2/3 p73 expression was found between p53 wild-type and mutant tumors.

No significant correlation exists between p73 isoform levels in tumors and stage, grade, histological subtype, residual disease after resection, age, and distant metastases (Table 5). In contrast, p53 mutation status correlates significantly with tumor grade, histological subtype, and the presence of distant metastases at the time of diagnosis (Table 6). Although 91% of p53 wild-type tumors show moderate differentiation (grade 2), the majority of p53 mutant tumors (55%) show poor differentiation (grade 3). Also, the majority of p53 wild-type tumors were of mucinous types (64%). In contrast, the most common histological type of p53 mutant tumors was serous (42%), whereas only 34% of them were mucinous. Moreover, endometrioid tumors are 2.6 times more common in mutant than in wild-type tumors. None of the patients with p53 wild-type tumors presented with distant metastases at the time of diagnosis, whereas all 11 metastases occurred in patients with p53 mutant tumors. A strong trend was also seen for p53 mutation status with respect to residual disease after resection. In 87% (19) of patients with p53 wild-type tumors, complete macroscopic removal of tumor mass or at least debulking to <2 cm residual tumor could be achieved. Residual disease >2 cm was only seen in 14% (3) of p53 wild-type patients, in contrast to 39% (24) of patients with mutant tumors.

To determine the clinical impact of clinical parameters, p53 mutation status and p73 isoform expression levels in ovarian carcinoma, survival analyses were performed for recurrence-free and overall survival. As expected, in univariate analysis, tumor stage, grade, residual disease after resection, and the presence of metastases at time of diagnosis were significant prognostic markers for overall and recurrence-free survival (Table 7; Fig. 6,A–C). Young age was significantly correlated with a good prognosis for overall survival but not for recurrence-free survival. On the other hand, histological subtype had no prognostic relevance in our cohort (Table 7). Together, this study confirms well-established clinicopathological parameters as prognostic markers.

Among the molecular markers, the p53 mutation status proved to be a strong prognostic marker for overall and recurrence-free survival in our ovarian carcinoma cohort (P = 0.019 and P = 0.006, respectively; Table 7; Fig. 6,D). Although 17 of 23 patients (74%) with p53 wild-type tumors were still alive at the last follow-up, only 6 of 23 wild-type patients (26%) had died. In contrast, only 28 of 66 patients (42%) with p53 mutant tumors were still alive, whereas 38 of 66 (57%) mutant patients had died. The median overall survival time was 91 months in the p53 wild-type group, compared with only 61 months in the p53 mutant group. TAp73, Ex2p73, and Ex2/3p73 were not correlated with survival (Table 7). Because ΔNp73/ΔN′p73 transcripts produce the same polypeptide and are overexpressed in all tumors, values were pooled, and tumors were arbitrarily divided by median tumor levels into two approximately evenly sized subgroups (above and below the 50th percentile of the range of Np73/ΔN′p73 expression). A trend was seen for better overall survival in ovarian carcinoma patients with low expression of ΔN′p73/ΔNp73, compared with patients with high expression (P = 0.1175; Table 7; Fig. 6 E). However, this trend did not reach statistical significance. Finally, in multivariate survival analysis, only residual disease after resection remained as a significant prognostic marker for overall survival.

Here we provide the largest and most comprehensive tumor expression analysis of all NH₂-terminal p73 isoforms reported to date, encompassing 100 ovarian carcinomas of all histological types, and evaluate their correlation to clinicopathological parameters. We used a highly reproducible quantitative RT-PCR assay optimized for product specificity and sensitivity. The most striking finding is the high prevalence of up-regulation of ΔN′p73 in the vast majority of ovarian carcinomas (95% of cases), providing a molecular marker of this disease. ΔN′p73 is derived from the P1 promoter. The importance of tumor-specific up-regulation of ΔN′p73 was hinted previously in a small semi-quantitative study of hepatocellular carcinomas that found ΔN′p73 up-regulation in six of six tumors (18). In contrast, up-regulation of the P2 promoter-derived ΔNp73 transcript was present in a small subgroup of 26% of ovarian carcinomas. We conclude that ΔN′p73, rather than ΔNp73, is the main contributor to total ΔNp73 up-regulation in ovarian cancers. Thus, the P1 promoter rather than the P2 promoter is primarily deregulated, at least in this cancer type. Our data also reveal differential in vivo regulation among the various P1 promoter-derived isoforms. Although ΔN′p73 is up-regulated in almost all ovarian cancers, and TAp73 is up-regulated in about one-third, Ex2/3p73 levels are actually down-regulated, possibly because of down-regulation of the splice acceptor on exon 4 at the pre-mRNA level (29). Expression data of Ex2/3p73 in human cancers had not been reported previously. Ex2p73 levels remained largely unchanged. Taken together, Ex2 and Ex2/3p73 isoforms do not seem to play a major oncogenic role in ovarian carcinogenesis. Six percent of tumors showed ΔNp73 expression levels that exceeded the highest normal ΔNp73 values, reflecting up-regulation of the P2 promoter at least in some tumors. This increase in tumor ΔNp73 levels might be a reflection of the previously suggested autoregulatory feedback loop of p53 and TAp73 inducing the P2 promoter (17, 22, 30, 31). Consistent with this notion, 17 of 21 ovarian cancers with the highest ΔNp73 levels also exhibited high TAp73 levels; in addition, 8 of these 17 tumors were wild type for p53.

Another novel result of this study is the strong correlation between p53 status and up-regulation of dominant-negative ΔN/ΔN′p73 in tumors. The fact that ovarian cancers select ΔN′p73 and ΔNp73 in 95% and 26% of tumors, respectively, is by itself a strong argument for their oncogenic role in vivo. Up-regulation of these isoforms could bestow oncogenic activity on the TP73 gene, because they interfere with the tumor suppressor functions of p53 and TAp73. To test whether the TP73 gene might function to counteract the mutational selection pressure for TP53, we reasoned that if ΔTAp73 isoforms were indeed oncogenic inhibitors of p53 and TAp73 in vivo, their up-regulation should occur preferentially in wild-type p53 tumors. Our results indeed support this notion. We found significantly higher expression levels of ΔNp73, ΔN′p73, and Ex2/3p73 (as well as TAp73) in p53 wild-type ovarian cancers than in p53 mutant cancers (Fig. 5). Moreover, the prevalence of up-regulation for ΔNp73 and ΔN′p73 was higher in wild-type tumors than in mutant tumors (48% and 96% versus 23% and 82%; Table 4). On the other hand, no difference was seen in the prevalence of TAp73 up-regulation between wild-type and mutant tumors (Table 4). This result further solidifies our previous result on a series of gynecological tumors, where we already saw a strong statistical trend between tumor-specific up-regulation of ΔTAp73 isoforms and wild-type p53 status (19). Together, these in vivo data provide compelling evidence for the notion that expression of dominant-negative ΔTAp73 can alleviate the selection pressure for p53 during tumor formation.

What, if any, is the impact of concomitantly increased TAp73 expression in tumorigenesis? Apparently, cancer cells can tolerate increased levels of this proapoptotic protein, because high TAp73 expression in tumors is neutralized by increased levels of transdominant ΔTAp73. We and others recently showed that physical interaction between oncogenic and antioncogenic family members is one of the possible mechanisms of interference with the specific DNA binding activity of wild-type p53 and TAp73. Inactive mixed protein complexes were found between endogenous ΔNp73α or ΔNp73β on the one hand and wild-type p53 or TAp73α or β on the other hand in primary human tumors, cultured tumor cells, and mouse neuronal cultures (14, 17, 19, 32). Of note, a stoichiometry as low as 1:3 of mutant to wild-type p53 molecules in heterozygous tumors abrogates wild-type function, suggesting that a similarly skewed stoichiometry might be effective for mixed ΔNp73 complexes as well. An additional mechanism of inhibition might be direct promoter competition, with ΔNp73 displacing wild-type p53 or TAp73 from their DNA binding sites (21). Thus, a slight decrease in the TA/ΔTA ratio might be sufficient to convert TP73 from a tumor suppressor into an oncogene. Interestingly, a strong prognostic impact of ΔNp73 up-regulation could be shown in neuroblastoma, notably one of the rare tumor types that show no selection pressure at all for p53 (25). In addition to aberrant subcellular localization of wild-type p53 as an important mechanism to circumvent mutational pressure (33, 34), ΔNp73-mediated inhibition could be a second contributing factor to the virtual absence of p53 mutations in neuroblastoma. A third mechanism that allows cancer cells to tolerate high TAp73 levels could be mediated through p53 mutants, because certain tumor-derived p53 mutant proteins associate with and inhibit TAp73 function in cultured cells (6, 35).

Taken together, a unifying interpretation of our findings is that the P1 promoter of TP73 is up-regulated in ovarian cancer, possibly via deregulated E2F activity or other mitogenic oncogenes. E2F1 is a direct transcriptional activator of the P1 promoter of TP73 (7, 8, 9, 10), and deregulated E2F1 activity is an almost universal hallmark of human cancers (36). Furthermore, the biologically relevant target of P1 up-regulation might be the oncogenic ΔN′p73, whereas the opposing TAp73 might be a “side product” that becomes co-up-regulated by default but is functionally neutralized by ΔN′p73. A second route for self-inactivation of TAp73 exists through ΔNp73, produced via the autoregulatory loop. Our finding that ΔTAp73 up-regulation is significantly correlated with wild-type p53 tumor status strengthens the hypothesis that these isoforms provide a pathway for epigenetic inactivation of wild-type p53 in tumors and thus act as oncogenes in vivo.

Evidence from tissue culture and mouse studies is mounting that ΔNp73 protein might function as an oncogene in human cancer (15, 18, 19, 23). Ideally, the strongest endorsement for a tumor-promoting role of any candidate human oncogene is a direct impact on survival. Thus, studies addressing the clinical relevance of ΔNp73 in human cancer will be of high importance. To date, only one study, performed in 52 neuroblastoma patients, has found that the presence of ΔN/ΔN′p73 expression is a significant prognostic marker for overall and recurrence-free survival (P < 0.001), independent of age, tumor stage, primary tumor site, and N-Myc amplification (25). In the neuroblastoma study, combined ΔN/ΔN′p73 was detectable in 30% of tumors. Tumors expressing both TA and ΔN/ΔN′p73 isoforms had variable ratios, with TAp73 being the prevalent variant. Although TAp73 had no effect on survival, the expression of the ΔN/ΔN′p73 had a strong negative impact on survival of these neuroblastoma patients. In contrast, in our cohort of ovarian cancers, up-regulation of ΔN′p73, defined as levels ranging above the 75th percentile of normal levels, was a molecular marker of the tumorous condition per SE, because it occurred in every cancer. Thus, in contrast to neuroblastoma, the presence or absence of ΔN/ΔN′p73 up-regulation does not stratify ovarian cancer patients, because it occurs universally. To determine whether the degree of up-regulation of ΔNp73/ΔN′p73 contained further prognostic information for survival, tumors were arbitrarily divided by median tumor levels into two subgroups. In support of the findings in neuroblastoma patients, a trend was seen for better overall survival in ovarian carcinoma patients with low expression of ΔN′p73/ΔNp73, compared with patients with high expression (Table 7; Fig. 6,E). Moreover, as in neuroblastoma, TAp73 expression levels showed no correlation with survival. Although our study falls short of demonstrating prognostic significance for ΔN′p73/ΔNp73, it is important to keep in mind that in sharp contrast to neuroblastoma, 74% of our ovarian carcinoma patients harbored p53 mutations, and these mutations already contributed to poor clinical outcome (Fig. 6 D). Thus, the “need” for epigenetic inhibition of p53/TAp73 by deregulated ΔN′p73/ΔNp73 in this cohort was diminished. It is therefore tempting to speculate that ΔN′p73/ΔNp73 might be of prognostic value in the subgroup of ovarian carcinomas that retains wild-type p53, such as e.g., the endometrioid type. Our results are also consistent with the two previous p73 ovarian cancer studies reported. Both determined global p73 expression that did not distinguish between TA and ΔTA isoforms, using immunohistochemistry or semi-quantitative RT-PCR. Nevertheless, high p73 expression correlated with advanced tumor stage (37, 38) and poorer survival (37). Using the isoform-specific assay developed here to design larger studies in the future will allow for a definitive answer on whether N′p73/ΔNp73 is an independent prognostic marker for ovarian cancer.

We thank A. Kuehler, B. Rosati, F. Grau, P. Pancoska and E. Schuster for technical assistance.