|Reles, Angela : MOLECULAR GENETIC ALTERATIONS IN OVARIAN CANCER The Role of the p53 Tumor Suppressor Gene and the mdm2 Oncogene |
Our study includes a large cohort of primary ovarian carcinomas and addresses several issues relevant to p53 sequence alterations and protein overexpression. First, we analyzed, as in our previous study (Wen et al. 1999), frozen tissue in order to avoid problems related to loss of antigenicity that is observed with immunohistochemistry in paraffin-embedded tissue. Secondly, we assessed the frequency of mutations throughout the entire coding region of the gene by analyzing exons 2-11 by SSCP and subsequent sequencing of cases with alterations, as opposed to most other studies which are limited to exons 5-8. Thirdly, information on adjuvant treatment was available and the results of the p53 sequence analysis and immunohistochemistry were compared to the response to chemotherapy, in order to evaluate the role of p53 for the sensitivity of platinum based chemotherapy. Finally, clinical follow-up information concerning outcome was available in 96% of the cases and information about the progression of the disease was available in 78% of the patients, which allows the evaluation of the prognostic relevance of p53 alterations in a sufficient number of cases.
We found a total of 100 p53 mutations in 99 out of 178 ovarian carcinomas (56%), which is higher than the percentage in most other studies. The frequency of p53 mutations in studies using frozen tissues including our previous study is 50% and ranges from 20% to 73% (Mazars et al. 1991, Kihana et al. 1992, Kohler et al. 1993a, Milner et al. 1993, Sheridan et al. 1994, Fujita et al. 1994, Lee et al. 1995, Casey et al. 1996, Sakamoto et al. 1996, Skilling et al. 1996, Schuyer et al. 1998, Wen et al. 1999). In studies using paraffin-embedded tissues, the frequency of mutations was 47%, ranging from 26% to 79%. (Teneriello et al. 1993, Kupryjanczyk et al. 1993, Wertheim et al. 1994, Niwa et al. 1994, Zheng et al. 1995, Kappes et al. 1995).
Several studies have analyzed part of the p53 coding region for mutations and found mostly between 20% and 52% mutations (Milner et al. 1993, Teneriello et al. 1993, Wertheim et al. 1994, Fujita et al. 1994, Niwa et al. 1994, Zheng et al. 1995, Kappes et al. 1995) with the exception of Kohler et al. (1993a) who found a percentage of 73% and Jacobs et al. (1992) who found 66% mutations. Three studies analysing exon 4-9 found 20-37% mutations (Kim et al. 1995, Lee et al. 1995, Sakamoto et al. 1996) and one study found 26% mutations in exon 4-8 (Niwa et al. 1994).
126The mutations were located predominantly (92%) in exons 5-8 and their associated splice junctions. Eight mutations (8%) though, were found outside of exons 5-8. These eight mutations were located in exon 4 and the exon 4/intron 4 splice junction (4 mutations) and in exon 9 and the intron 8/exon 9 splice junction (4 mutations). These results demonstrate that the evaluation of codons 126 to 307 (182 codons) provided 92 of 100 (92%) of the mutations, while eight mutations (8%) could only be identified by evaluation of codons 1 to 125 and 308 to 393 (211 codons).
The percentage of mutations in studies analysing the entire coding region of p53 is overall 59% (207/349) and varies from 52% to 79%, which is consistent with our own results (Kihana et al. 1992, Kupryjanczyk et al. 1993, 1995, Skilling et al. 1996, Casey et al. 1996, Wen et al. 1999). The frequency of mutations in exons 2-4 and 9-11 was reported to be between 0% and 20% with an overall average of 11% (24/258) (Kihana et al. 1992, Kupryjanczyk et al. 1993, 1995, Skilling et al. 1996, Casey et al. 1996, Wen et al. 1999). This means that studies analysing only exons 5-8 of the p53 gene will miss approximately 11% of the mutations.
In our previous study (Wen et al. 1999) exons 2-11 were first screened by SSCP in 105 ovarian carcinomas and cases with suspicious band patterns were sequenced. Out of 50 cases with normal SSCP band patterns, 42 were subjected to complete sequence analysis of the open reading frame by automated sequencing (Wen et al. 1999, Wang-Gohrke et al. 1998). A total of 60 mutations (57%) were identified in this study with only 5% mutations outside exons 5-8. Skilling et al. (1996) identified 61% mutations in 64 cases with 34/39 mutations located in exon 5-8 and only 5 mutations (13%) outside these exons. Kihana et al. (1992) reported that two of 10 mutations (20%) identified in 14 ovarian carcinomas were outside exons 5-8, although the codon for one of the two mutations was not specified. Kupryjanzcyk et al. (1993) analyzed 38 ovarian carcinomas and found 31 mutations in 30/38 tumors (79%) by PCR-SSCP and sequencing, which is the highest percentage so far reported in the literature. Three of these mutations were located outside exon 5-8.
Excluding nonsense and frameshift mutations, 222 of the 393 codons of the human p53 gene have been described as the targets of at least 698 different mutations (Beroud et al. 1996). While some codons, such as codon 190, are the target of only one mutational event, others such as codon 179 have been the target of multiple mutational events. In our study, the codons most frequently mutated were codons 175, 179, 234, 242, 248, 273, 275, and 281, which were each mutated in three or more different carcinomas. No mutations have been identified in codons 92-112, the region which has been recently identified as the degradation signal in p53/MDM2 interaction (Gu et al. 2000).
127In our study we identified 72% missense mutations. This is consistent with the results of most other authors who found between 62% and 85% missense mutations (Milner et al. 1993, Kupryjanczik et al. 1993, Niwa et al. 1994, Skilling et al. 1996, Casey et al. 1996, Righetti et al. 1996, Schuyer et al. 1998). Deletions are usually described in a percentage of 7%-13% (Kupryjanczik et al. 1993, Milner et al. 1993, Righetti et al. 1996, Schuyer et al. 1998, Wen et al. 1999), which is similar to our findings of 10% deletions. But some authors found deletions in a percentage as high as 22% by SSCP of exon 2-11 and sequencing (Skilling et al. 1996) and 27% of all mutations by sequencing exon 2-11 (Casey et al. 1996). Insertions have been found less often and have been described by Skilling et al. (1996) as 2.5%, Casey et al. (1996) as 6%, and Milner et al. (1993) as 3% of the mutations in ovarian cancer. Several studies with smaller case numbers do not describe insertions at all. We identified 3 insertions (3%), one of which was combined with a deletion. Splice junction mutations are considered rare mutations and so far only 12 splice junction mutations have been described in the ovarian cancer literature including our previous study, with a percentage of usually 1-5% (Kupryjanczik et al. 1993, Niwa et al. 1994, Skilling et al. 1996, Casey et al. 1996, Righetti et al. 1996, Schuyer et al. 1998). Including three previously published splice junction mutations (Wen et al. 1999) we found eight splice junction mutations accounting for 8% of all mutations identified, which is higher than in any of the other studies.
The majority of mutations in this study (61/99 carcinomas, 62%) were located within highly conserved domains of p53 (domains II, III, IV and V) (Cho et al. 1994, Soussi and May, 1996). These domains are almost equivalent to the loop-sheet-helix (domains II and V), loop 2 (domain III), and loop 3 (domain IV) regions of p53 (Cho et al. 1994, Soussi and May, 1996). The loop-sheet-helix is responsible for direct DNA interactions with the major groove; loop 3 is responsible for direct interactions with the minor groove of DNA, and loop 2 and 3 together are responsible for maintaining the needed 3-dimensional conformation (Cho et al. 1994, Prives 1994, Soussi and May, 1996, Arrowsmith and Morin, 1996). Mutations in these conserved domains have been observed to confer more aggressive behavior in patients with breast carcinomas or colon carcinomas than mutations in other regions of p53 (Bergh et al. 1995, Goh et al. 1995). Similarly, in our study, comparisons of overall survival for women with mutations in highly conserved domains together did show a statistically significant difference in survival compared to those with mutations in non-conserved regions or wildtype p53 sequence (p = 0.007).
Several p53 constitutional polymorphisms have been reported. These are located in codon 21 (exon 2) (Ahuja et al. 1990), codon 36 (exon 4) (Felix et al. 1994), codon 72 (exon 4) (Harris et al. 1986, Matlashewski et al. 1987), codon 213 (exon 6) (Carbone et al. 1991), intron 1 (Ito et al. 1994), intron 2 (Pleasants and Hansen 1994, DiCioccio and Piver 1996, Ge et al. 1998, Verselis and Li, 2000), intron 3 (Lazar et al. 1993) and intron 6 (Peller et al. 1995).
p53 polymorphisms can cause abnormal bands in single-strand conformation polymorphism (SSCP) analysis, which is widely used as a screening technique for p53 mutations and can therefore be mistaken for mutations, if they are not confirmed by DNA sequencing (Carbone et al. 1991).
Among the 178 ovarian carcinomas, which were analyzed by SSCP and DNA sequencing, we identified 100 mutations (Table 8) and 45 polymorphisms in exons 4, 6, and 7, and in the intron sequences, which were flanking exons 3, 6, 10, and 11. We observed four polymorphisms in codon 36 (CCG to CCA), codon 213 (CGA to CGG), codon 224 (GAG to GAA), and codon 231 (ACC to ACA). The frequency of each of these polymorphisms was 0.6% in our study cohort. The codon 36 polymorphism was first described by Felix et al. 1994 and was present in the heterozygous state in 4% of 100 individuals. Carbone et al. (1991) first described the polymorphism at codon 213 (CGACGG), which doesn´t result in an amino acid exchange but causes the loss of a Taq I restriction site. It was found in a frequency of 3.2% in lung cancer and breast cancer and caused an SSCP alteration (Carbone et al. 1991). The sequence alterations at codon 224 (exon 6) and 231 (exon 7) have not been previously described in the literature. They may be polymorphisms or silent mutations since they do not result in an amino acid exchange. But in these cases, no normal tissue DNA was available, so that we could not verify whether this is a polymorphism.
Five percent (9/178) of all the cases had a CCGCCC polymorphism at codon 72 resulting in an arginine to proline amino acid exchange. Codon 72 of the p53 gene has been described to be the site of a primary structure polymorphism with an amino acid residue which could be an arginine, proline or cysteine (Harris et al 1986, Matlashewski et al. 1987). The majority of Caucasian individuals express the arginine-containing p53 protein as opposed to African-Americans who express predominantly the proline-containing p53 protein (Weston et al. 1992, Själander et al. 1995). We therefore considered the arginine-72 p53 as the normal sequence and the proline-72 p53 as the polymorphic sequence.
129Overall, the percentage of polymorphisms identified in our study seems low compared to the literature. This may be due to the fact that this cancer cohort was not systematically screened for p53 polymorphisms. Only cases with SSCP band alterations were sequenced. Therefore polymorphisms, which do not cause SSCP alterations, may have been missed.
The codon 72 p53 polymorphism is presently discussed to play a role for cancer predisposition. In breast cancer patients, a significant increase in the codon 72 proline allele frequency was observed, which was most pronounced in highly differentiated breast cancer (Själander et al. 1996). In ovarian cancer, probands homozygous for the arginine allele have been found to develop ovarian cancer at an earlier age and have a survival advantage compared to arginine/proline heterozygote and proline homozygote patients (Buller et al. 1997). When a loss of heterozygosity occurred in patients with invasive ovarian cancer, the proline allele was lost preferentially and tumors which retained a proline allele were significantly more prone to mutation than tumors without a proline allele (Buller et al. 1997).
Storey et al. described a significant overrepresentation of homozygous arginine-72 p53 in human papilloma-virus-associated cervical carcinoma compared with the normal population which makes individuals homozygous for arginine 72 about seven times more susceptible to HPV associated tumorigenesis than heterozygotes (Storey et al. 1998). They conclude that the arginine-encoding allele represents a significant risk factor in the developement of human papilloma-virus-associated cervical carcinoma. This finding was corroborated in two independent studies (Zehbe et al 1999, Agorastos et al. 2000), however, refuted by others (Helland et al. 1998, Hildesheim et al. 1998; Josefsson et al. 1998, Lanham et al. 1998). Inter-laboratory variation in p53 genotyping may have contributed to the inconsistent findings across studies (Makni et al. 2000). Makni et al. observed odds ratios of 8.0 (95% CI: 2.3-28.5) for the Arg/Arg genotype and cervical cancer when they excluded individuals which were differently genotyped in three independent laboratories (Makni et al. 2000). Recently a differential effect of the p53 codon 72 variants on interaction of mutant p53 with p73 has been published (Marin et al. 2000). The binding of the arginine isoform to p73 tended to be stronger than that observed for the proline isoform, with a neutralizing effect on p73-induced apoptosis. The presented data also suggested the codon 72 polymorphism to modify the response of selected tumor cells to chemotherapy. In some tumors, such as squamous cell cancers, the authors found evidence for preferential mutation and expression of the arginine form of p53 in arginine/proline heterozygotes (Marin et al. 2000).
Compared to the expected frequency of the arginine and the proline allele, the arginine allele was clearly overrepresented in our study population which consisted
130predominantly of caucasian individuals. A correlation between the proline allele and p53 overexpression was not identified.
Various intron alterations of unknown significance have been found in the p53 gene. A polymorphism in intron 3, which consists of a single repeat of 16 nucleotides starting at nucleotide 11951 of the p53 gene, was first described by Lazar et al. (1993) and was identified in its heterozygote form in 28% of 82 individuals in lymphozyte DNA. The polymorphism was not associated with a predisposition to breast cancer. We found this 16 bp repeat polymorphism in 14% of the ovarian cancer cases. The presence of the polymorphism was not associated with a higher percentage of p53 overexpression, but was found in eight of nine cases with the codon 72 proline polymorphism. According to Själander et al. (1995), significant ethnic differences have been found for the codon 72 and the intron 3 16 bp repeat polymorphism (Själander et al. 1995). The most common finding in a northern european population was the codon 72 arginine allele combined with absence of the 16 bp repeat polymorphism and was therefore considered the wild-type haplotype (Själander et al. 1995).
The biological significance of the intron 3 16 bp repeat polymorphism remains unclear. Homozygosity of the p53 intron 3 polymorphism was suspected to be highly associated with sporadic ovarian cancer by Runnebaum et al. (1995). The authors found an overall frequency of 24.1% heterozygous and 1.7% homozygous PIN 3 polymorphism in normal controls, while the frequency was 30.6% and 11.3% respectively in ovarian cancer patients, which was a 6.7 fold increase of the homozygotes in ovarian cancer (Runnebaum et al. 1995). However, no such association was observed by others in 82 ovarian carcinomas (Lancaster et al. 1995). The frequency of this polymorphism in our study cohort was lower than that mentioned by Runnebaum et al. (1995). In a study of breast cancer patients and normal controls, no significant difference was found for the frequency of the 16 bp repeat polymorphism (Själander et al. 1996).
Further intron polymorphisms have been described in the literature to be associated with cancer predisposition. An intron 6 polymorphism, a G to A transition at position 61, has been found significantly more frequently in ovarian carcinomas as compared to normal controls (Mavridou et al. 1998). This was confirmed in a second study, where an association of this intron 6 polymorphism and the strongly linked 16 bp polymorphism in intron 3 with ovarian cancer has been established, however, only in women not carrying BRCA1 or BRCA2 germline mutations (Wang-Gohrke et al. 1999). Furthermore Peller et al. (1995) described a 8bp polymorphic sequence in intron 6 of the p53 gene, which was identified in the heterozygotic form in 32% of normal blood samples. Patients with gastric cancer and breast cancer demonstrated a
131higher incidence of heterozygosity (50%). Therefore the authors suggested an association between this polymorphism and cancer predisposition and susceptibility (Peller et al. 1995).
We describe three intron alterations, which to our knowledge have not been described in the literature. One g c alteration was identified in intron 6 at position 13964 of the p53 gene in two ovarian cancer cases. Both cases had shown SSCP band shifts. In one case, a p53 mutation was identified. Since no normal tissue was available in these two cases, we can not proove that this is a polymorphism, but we found this alteration in only 1% of the cancer cases, while the other cases, which were sequenced because of SSCP band shifts, revealed the normal sequence at this location.
We identified two new polymorphisms in intron 10, which have not previously been described. An a t polymorphism was found at position 17708 of the p53 gene in two cases in the flanking region of exon 10. In one of the cases, normal tissue was available and DNA sequencing confirmed the polymorphism. In three other ovarian cancer cases which showed only a weak SSCP band shift in the exon 10 and flanking intron region, a normal sequence at nucleotide 17708 was identified.
The other intron 10 polymorphism was found in the flanking sequence of exon 11. Four out of eight cases with SSCP band shifts revealed the polymorphic sequence, a c t nucleotide exchange at position 18550 of the p53 gene. In three out of these four cases, the polymorphic sequence could be confirmed in normal cervical tissue of the patient. We therefore conclude, that both intron 10 sequence alterations are polymorphisms. Since none of the cases with intron 10 alterations had a mutation in exon 10 or exon 11, we assume that the SSCP band shift was caused by the polymorphism. These two polymorphisms have not been described in the literature so far.
Interestingly, out of eight cases with intron 6 or intron 10 alterations, seven showed p53 overexpression, though p53 mutations and p53 wildtype, respectively, were evenly distributed among these cases. Though the case number was very small, the result almost reached statistical significance. Intron polymorphisms may affect transcription and splicing processes and may therefore cause alterations in protein expression, stability and/or activity. For variations in p53 introns functional relevance has been suggested (Beenken et al. 1991; Avigad et al. 1997). The actual allele frequencies and possible association with p53 overexpression of the here newly identified intron alterations need to be clarified by screening larger case numbers.
The biological significance of p53 polymorphisms is unclear at this point, since conflicting results regarding cancer predisposition have been reported. But there is increasing evidence, that some polymorphisms or defined constellations of polymorphisms (haplotypes) may be important for processes relevant for cancer risk
132and treatment outcome, as has been suggested for the codon 72 polymorphism as intragenic modifier of mutant p53 behaviour (Marin et al. 2000). It might therefore be worthwhile, to screen a large cohort of patients with ovarian cancer in comparison to unaffected individuals for several p53 polymorphisms to possibly define risk and prognostic factors for ovarian carcinoma.
Several investigators have used frozen ovarian carcinoma specimens for immunohistochemical analysis of p53 and found p53 overexpression in 32% - 84% with an average percentage of 51% (Marks et al. 1991, Kihana et al. 1992, Eccles et al. 1992, Kohler et al. 1993a, Kiyokawa et al. 1994, Henriksen et al. 1994, Sheridan et al. 1994, Lee et al. 1995, Casey et al. 1996, Skilling et al. 1996, Buttitta et al. 1997, Geisler et al. 1997, 2000, Schuyer et al. 1998, Wen et al. 1999). We found p53 overexpression in 62% of the cases. A few authors have performed both immunohistochemical and mutation analysis (Marks et al. 1991, Kihana et al. 1992, Sheridan et al. 1994, Lee et al. 1995, Casey et al. 1996, Skilling et al. 1996, Buttitta et al. 1997, Wen et al. 1998). Only four of the latter studies involved more than 50 carcinomas (Casey et al. 1996, Skilling et al. 1996, Buttitta et al. 1997, Wen et al. 1999).
Many studies have used formalin-fixed, paraffin-embedded material for immunohistochemistry. In those studies, which analyzed more than 50 cases of FIGO I-IV ovarian carcinomas, the frequency of p53 overexpression ranged from 26% to 70% with an overall frequency of 48% (Bosari et al. 1993, Sheridan et al. 1994, Inoue et al. 1994, Niwa et al. 1994, Hartmann et al. 1994, Kupryjanczyk et al. 1994, van der Zee et al. 1995, Klemi et al. 1995, Herod et al. 1996, Reles et al. 1996, Viale et al. 1997, Dong et al. 1997, Eltabakkah et al. 1997, Röhlke et al. 1997, Anttila et al. 1999, Ferrandina et al. 1999, Werness et al. 1999). In one study, p53 overexpression was identified immunohistochemically in only 14% of the cases (Marx et al. 1998).
With the exception of four studies (Kupryjanczyk et al. 1993, Casey et al. 1996, Lee et al. 1995, Teneriello et al. 1993), three of which were of a small number of cases, the frequency of p53 overexpression is consistently higher than the frequency of p53 mutations. Our observations of a 62% immunostaining rate and a 56% mutation rate for p53 were consistent with these generalizations.
A strong correlation was observed between p53 immunostaining and p53 mutations (p < 0.001), but this was mainly an effect of the high rate of immunostaining of missense mutations (94%). However, only 46% of nonmissense mutations were identified by immunostaining and this percentage was not significantly different from the percentage of staining in wildtype p53 cases (38%) (p=0.43). Therefore it can be concluded that immunohistochemistry is not suitable for the detection of nonmissense mutations, which is consistent with the findings of other authors (Casey et al. 1996, Skilling et al. 1996, Schuyer et al. 1998). The low rate of immunostaining in cases with nonmissense mutations is assumed to be caused by alteration and truncation of the protein due to introduction of a stop codon in nonsense mutations, frameshifts caused by deletions or insertions, and alterations of transcribed RNA in splice site mutations. Since the antibody used for immunohistochemistry (DO-7) recognizes an epitope near the amino-terminus of the protein between amino acid 19 and 26, one would expect the antibody to have the potential to recognize nearly all of the mutant p53 proteins. But p53 protein may not only be qualitatively altered but also quantitatively reduced, possibly because of destruction of the messenger RNA as a result of sequence changes.
A high percentage of cases with wildtype p53 sequence (38%) showed overexpression of the p53 protein. Since this can not be explained by sequence alterations, the stabilization and accumulation of a presumably normal p53 protein is most likely caused by other genes interacting with p53. Alterations of the mdm2 gene, which downregulates p53 in an autoregulatory feedback and promotes degradation of the p53 protein (Kubbatat et al. 1997, Haupt et al. 1997, Kubbutat et al. 1999) may play an important role in these cases. In fact, we found mdm2 alternative splicing in a high percentage of cases with p53 overexpression despite wild-type p53 sequence, and p53 overexpression was significantly correlated with the presence of a 654 bp splice variant (mdm2-b), which has lost most of the p53 binding sequence (see chapter 5.2.8).
Cell culture experiments have shown that the sensitivity of tumor cells to various chemotherapeutic agents depends on the efficient induction of apoptosis mediated by a functional p53 protein and that loss of p53 can enhance resistance to chemotherapy (Lowe et al. 1993, Vasey et al. 1996, Vikhanskaya et al. 1997).
The wildtype p53-expressing A2780 human ovarian cancer cell line acquired cross resistance to Cisplatin and Doxorubicin by transfection with a dominant negative mutant p53 gene, while it retained sensitivity to Taxol (Vasey et al. 1996). In another
134study (Vikhanskaya et al. 1997), Cisplatinum caused strong induction of p53, WAF1 and Bax in the Cisplatin sensitive A2780 cell line, while there was no such effect in a Cisplatin resistant cell line A2780-DX3, which furthermore showed a significant proportion of potentially inactive p53 protein located in the cytoplasm instead of the nucleus.
However, some of the studies using cell culture experiments report contradictory results. In a panel of cell lines, cisplatin was more efficient against mutant/null p53 cell lines than wildtype cell lines, while the novel platinum analogue DACH-aceto-Pt was considerably more toxic in wild-type p53 cell lines (Hagopian et al. 1999). Similarly, in isogenic A2780 human ovarian cancer cell lines that differ only in p53 function by transfection of HPV-16 E6, the p53-deficient cell line was more sensitive to cisplatin and the novel platinum agent ZD0473 (Pestell et al. 2000).
Recently, genetic suppressor elements (GSEs) have been identified which correspond to various regions within the p53 gene and can act as dominant negative peptides or antisense RNA molecules (Gallagher et al. 1997). A synthetic peptide, representing the predicted amino acid sequence of this GSE, conferred resistance to Cisplatin when introduced into A2780 cells and inhibited the sequence specific DNA binding activity of p53 protein in vitro. This indicates that inactivation of p53 function confers Cisplatin resistance in these ovarian tumor cells.
The hypothesis that ovarian cancer cells with functional p53 are more sensitive to Cisplatin is furthermore supported by the findings of gene therapy studies. Introduction of wildtype p53 via adenovirus gene transfer into A2780/CP cisplatinum resistent cells significantly sensitized these cells to platinum cytotoxicity, indicating that p53 was involved in resistance to cisplatin (Song et al. 1997).
Further strong evidence for the importance of a functional p53 protein for the efficacy of Cisplatin and Carboplatin is given by a database of the National Cancer Institute on drug activity in cell lines (Weinstein et al. 1997). This database compares activity patterns of chemotherapeutic agents and possible targets or modulators of activity in the cells, such as oncogenes, tumor-suppressor genes, drug-resistance-mediating transporters and others for more than 60.000 compounds against a panel of 60 human cancer cell lines. The results show a strong correlation between p53 wildtype sequence respectively p53 function in a yeast based assay and efficacy of Cisplatin and Carboplatin (Weinstein et al. 1997). As opposed to this, antimitotic tubulin-active agents such as Taxol show a strong negative correlation between p53 wildtype as well as p53 function and activity of the drug (Weinstein et al. 1997).
Since a dysfunctional p53 can not mediate the apoptotic process, tumors with p53 mutations or altered p53 protein may become resistant to platinum-based chemotherapy (Lowe et al. 1994). In our study in a subgroup of 72 patients, who
135received Cisplatin or Carboplatin / Cyclophosphamide combination therapy, we found a significant correlation between p53 protein overexpression and resistance to platinum-based chemotherapy (p=0.001). If all p53 mutations were taken into account, there was only a trend but not a significant association with treatment response (p=0.071). Interestingly though, we found that missense mutations, when compared to wildtype p53 or other mutations, were correlated with a high percentage of chemotherapy resistance in this cohort (p=0.008). This is consistent with the results of two other studies which have demonstrated a correlation between p53 alterations and resistance to platinum-based chemotherapy in ovarian cancer (Righetti et al. 1996, Buttitta et al. 1997). One study analyzed p53 overexpression and mutations in 32 cases of ovarian cancer FIGO stage III and IV and found a higher frequency of chemotherapy-resistance both in p53 overexpressing tumors and in tumors with missense mutations (Righetti et al. 1996). Another study, analysing 33 cases of advanced ovarian cancer FIGO III and IV found a significant association between chemotherapy-resistance and p53 immunostaining, as well as SSCP alterations (Buttitta et al. 1997). These results are further supported by two immunohistochemical studies. p53 overexpression was found to be associated with poor response to either paclitaxel/platinum or cyclophosphamide/ platinum chemotherapy in 54 stage III and IV patients with results approaching statistical significance (Goff et al. 1998). A higher frequency of early tumor progression was found in 28/59 patients who had received either cisplatin and treosulfan or treosulfan alone (Petty et al. 1998). In one immunohistochemical study, a dose-dependent response to platinum-based chemotherapy was found only in p53 negative, but not in p53 positive tumors (Marx et al. 1998). And a more recent study, which analyzed 168 primary stage III-IV ovarian carcinomas showed p53 overexpression to be significantly correlated with resistance to a platinum based chemotherapy in those patients who underwent pathologic assessment of response (Ferrandina et al. 1999). However, three other studies using immunohistochemistry (van der Zee et al. 1995, Herod et al. 1996, Viale et al. 1997) and one study using temperature gradient gel electrophoresis and immunohistochemistry (Röhlke et al. 1997) did not find a difference in treatment response between patients with or without p53 alterations.
Interestingly, missense mutations seem to have a different effect on p53 function in terms of apoptosis than other mutations. Recently, it was shown that certain types of mutations such as the p53His175 mutant and the p53His179 mutant substantially reduce the rate of etoposide-induced apoptosis, whereas other mutations had a much milder effect (Blandino et al. 1999). This suggests that certain types of mutation may have a selective gain of function, which may compromise the efficacy of cancer chemotherapy.
136Recently, small synthetic molecules have been identified which not only promote the stability of wild-type p53 protein but also allow mutant p53 to maintain an active conformation (Foster et al. 1999). Further work on these compounds may open new perspectives to overcome an impaired p53 function in combination with chemotherapy.
In contrast to these findings, sensitivity to Taxol appears to be increased through the absence of functional p53 protein because of increased G2M arrest and p53 independent apoptosis (Wahl et al. 1996, Vikhanskaya et al. 1998). Interestingly, though numbers are very small, the only patient in our study who was sensitive to Taxol/Carboplatin had high overexpression of p53 protein and a missense mutation in exon 5, while three out of four patients with resistant or refractory disease were immunohistochemically negative and two had wild type p53 sequence.
To date, only p53 protein overexpression but not p53 mutations have been shown to be a predictor of poor clinical outcome in ovarian cancer. Several studies have found a correlation between p53 overexpression and shortened survival (Bosari et al. 1993, Henriksen et al. 1994, Hartmann et al. 1994, van der Zee et al. 1995, Klemi et al. 1995, Levesque et al. 1995, Herod 1996, Viale et al. 1997, Eltabakkah 1997, Geisler et al. 1997, 2000, Röhlke et al. 1997, Anttila et al. 1999, Werness et al. 1999), but only few studies have identified p53 as an independent prognostic factor for overall survival (Klemi et al. 1995, Herod et al. 1996, Geisler et al. 1997, 2000, Röhlke et al. 1997) respectively recurrence-free survival (Anttila et al. 1999) in multivariate analysis. However, p53 protein expression was not a significant predictor of poor outcome in seven other studies (Marks et al. 1991, Kohler et al. 1993a, Kupryjanczyk et al. 1993, Sheridan et al. 1994, Niwa et al. 1994, Reles et al. 1996, Ferrandina et al. 1999, Wen et al. 1999). Only five studies found a correlation between p53 overexpression and time to recurrence (van der Zee et al. 1995, Levesque et al. 1995, Röhlke et al. 1997, Werness et al. 1999, Anttila et al. 1999).
p53 overexpression is relatively rare in tumors of low malignant potential but has been identified in 13% of advanced (stage II and III) borderline ovarian tumors and was found to be strongly associated with increased probability of recurrence or progression (Gershenson et al. 1999).
Seven of eight studies though which analyzed p53 sequence alterations and clinical follow-up data did not find a significant correlation between p53 mutations and shortened relapse-free or overall survival (Mazars et al. 1991, Kohler et al. 1993a, Kupryjanczyk et al. 1993, Niwa et al. 1994, Sheridan et al. 1994, Buttitta et al. 1997,
137Skomedal et al. 1997). Only our previous study found p53 mutations to be associated with shortened overall survival, but the results reached only marginal statistical significance (p=0.049) (Wen et al. 1999).
Most of the p53 studies examined either p53 expression or p53 mutations but not both in study populations of limited size. In our study p53 overexpression reached only marginal statistical significance as a predictor of poor clinical outcome (p=0.056). p53 mutations and especially those, which were located in highly conserved domains, were clearly correlated with poor overall survival (p=0.014 and p=0.007). An association between p53 mutations and overexpression was observed (p<0.001) and the most favourable prognosis with a significantly longer time to progression and overall survival was seen in patients with neither p53 mutation nor overexpression (p=0.035, p=0.007).
Summarizing these results, we analyzed p53 protein expression and p53 sequence alterations in the entire coding region of the gene in 178 frozen ovarian cancer tissues with complete clinical follow-up information. We could demonstrate that p53 mutations and especially those in evolutionary conserved domains correlate with shortened time to progression and shortened overall survival. Most importantly, evaluation of adjuvant treatment showed that p53 overexpression as well as p53 missense mutations were correlated with resistance to platinum-based chemotherapy. This provides further clinical evidence, that the sensitivity of ovarian cancer cells for Cis- or Carboplatin depends on the efficient induction of apoptosis mediated by a functional p53 protein.
Since mdm2 is upregulated by p53 in response to DNA damage and subsequently inhibits p53 function and promotes p53 protein degradation, it seemed interesting to analyze mdm2 alterations in ovarian cancer cases with known p53 mutations and overexpression.
We analyzed the mdm2 gene by Southern and Northern hybridization but we found neither DNA amplification nor RNA overexpression in any of the 56 ovarian carcinomas analyzed, independently of p53 status. Since accumulation of mutant p53 protein has been shown to depend on a lack of mdm2 induction (Haupt et al. 1997, Kubbutat et al. 1997) one would expect that p53 protein accumulation despite wildtype p53 might induce mdm2 overexpression. Vice versa in cases with p53 mutation but undetectable p53 expression a possible mechanism might be the rapid degradation of p53 through higher levels of MDM2. But none of the p53 alterations in ovarian cancer cases caused mdm2 alterations detectable by Southern or Northern analysis.
This is consistent with findings in other epithelial tumors. Though mdm2 has been found to be amplified in more than 30% of sarcomas and amplification was correlated with overexpression of the MDM2 protein (Oliner et al. 1992), amplification in epithelial tumors is rare (as reviewed by Momand and Zambetti 1997). In breast cancer, gene amplification was noted in only 1.7-7.7% (McCann et al. 1995, Marchetti et al. 1995a, Quesnel et al. 1994), and in HPV negative cervical cancer in 2% (Ikenberg et al. 1995). In ovarian cancer, amplification has not been found in any case (Foulkes et al. 1995). MDM2 protein overexpression is rare in ovarian cancer (Foulkes et al. 1995), but has more recently been described in 40% of borderline ovarian tumors (Palazzo et al. 2000).
RNA overexpression was found in sarcomas which showed amplification of the gene (Oliner et al. 1992) and in 6% of non-small cell lung carcinomas (Marchetti et al. 1995b) but is generally rare in epithelial tumors. In ovarian cancer, absence of mdm2 expression was identified in 18/90 ovarian carcinomas (22%), while 66% showed weak expression and 12% showed strong expression (Tanner et al. 1997). Absence of mdm2 expression was found to be an independent predictor of poor survival in FIGO III and IV ovarian cancer patients (Tanner et al. 1997) and in non-small cell lung cancer (Ko et al. 2000). Low levels as opposed to high levels of mdm2 mRNA were found, furthermore, to be associated with poor clinical outcome in soft tissue sarcomas (Taubert et al. 2000).
139We found three RNA transcripts of approximately 7.4 kb, 5.5 kb, and 2.8 kb by Northern blot analysis in the ovarian cancer cases. In sarcomas and lung cancers, only a single 5.5 kb transcript had been observed (Oliner et al. 1992, Marchetti et al. 1995b) while in mammary epithelial cell lines, multiple messenger RNAs, ranging in size from 4.5 to 12.5 kilobases, were found. (Gudas et al. 1995). These different sizes of RNA transcripts may be the result of alternative splicing which has been described when mdm2 was originally cloned and more recently in ovarian and bladder carcinomas (Oliner et al. 1992, Sigalas et al. 1996).
PCR analysis of the reverse transcribed mRNA and sequencing of the cDNA products revealed numerous mdm2 splice variants, which were present either instead of, or in addition to, the full length mdm2 transcript in ovarian cancer tissue. Only 20% of the ovarian carinomas expressed exclusively the full length mdm2 RNA and some cases (9%) had no expression while the majority of cases (72%) expressed splice variants. We identified 30 different splice variants and the sizes ranged from 52 bp to 888 bp.
Alternative splicing has been previously described for the mdm2 gene (Oliner et al. 1992) but only few splice variants were identified in ovarian cancer (Sigalas et al. 1996). The results were not analyzed with regard to p53 sequencing and immunohistochemistry. Alternative splicing is a common occurrence in a number of different genes and can create isoforms of proteins with extensive sequence overlap but different functions. This has been well described for example for the troponin T gene of rat muscle which exists in an and form depending on splice variants (Lewin, 1994). These splice variants include alternatively, either the or the exon besides three other exons which are identical in both mRNA transcripts. Another example is the APC (adenosis polyposis coli) gene, a tumor suppressor gene altered early in colon carcinogenesis, which is expressed both with and without exon 1 at equivalent levels in brain, heart, and skeletal muscle in humans and mice (Santoro et al. 1997). The resulting APC proteins with different amino-terminal domains potentially have different functions in interaction with other proteins. The expression of APC isoforms is dependent on tissue differentiation indicating that alternative splicing occurs in response to tissue dependent signals as well as signals to differentiate (Santoro et al. 1997).
In ovarian cancer we observed alternative and aberrant splicing. In some of the splice variants, the splice site was located at the exon-intron boundaries, but in others splicing took place within the exon sequence and should therefore be classified as
140aberrant splicing. Aberrant splicing, in contrast to alternative splicing, is the splicing of mRNA which is misdirected and does not occur at de facto splice sites. Both the tumor susceptibility gene 101 (TSG 101) and the fragile histidine triad gene (FHIT) mRNAs show evidence of alternative and aberrant splicing (Gayther et al. 1997).
In ovarian cancer, deletions of several hundred basepairs of the TSG 101 gene, a putative tumor suppressor gene, have been found in 40% of the cases and are thought to be due to aberrant splicing. The breakpoints of these deletions were located at genuine or cryptic splice sites therefore most likely being the result of aberrant splicing (Gayther et al. 1997).
Deletions of the FHIT gene which is also thought to be a tumor suppressor gene, occured in 42% of ovarian carcinomas at exon/intron boundaries and were suspected to be due to alternative splicing (Gayther et al. 1997). Evidence of both alternative and aberrant splicing was furthermore described in breast (Negrini et al. 1996), lung (Fong et al. 1997), head and neck tumors (Virgilio et al. 1996).
In this investigation of malignant, as well as benign ovarian tumors, and normal tissue, we found evidence of expression of normal mdm2 (1473 bp) coexisting with both alternative and aberrant splicing. We found alternative splicing at the exon/intron boundary in two variants (888 bp, 654 bp) which were missing exon 4-9 and exon 4-11, respectively. These two splice variants had been described previously in ovarian carcinomas but not in normal ovarian tissue and had been named mdm2-a and mdm2-b (Sigalas et al. 1996). Both variants were found to have transforming ability in NIH 3T3 cells (Sigalas et al. 1996). We found the 654 bp splice variant in almost half of the ovarian carcinomas, but also infrequently in borderline tumors (1/9), and even in one normal ovarian tissue (1/20). The presence of this variant is therefore strongly, but not exclusively, associated with carcinomas.
All other RNA transcripts of 721 bp to 52 bp size resulted from aberrant splicing. Exon 10 and 11 were spliced out completely in all but one of these splice variants and the cryptic acceptor splice site was consistently located within exon 12. The donor splice sites were located at the exon/intron boundaries of exon 4, exon 5, or exon 6 in seven of the splice variants but in the majority of variants within the sequence of exon 3 to exon 9. In one splice variant of 613 bp, the complete exon 5 as well as parts of exon 9, 10 and 11 were deleted, while exon 6,7 and 8 were contained. All splice variants except, for the 888 bp and the 721 bp variant, completely lose the acidic domain which was suggested to be involved in the regulation of ribosome metabolism (as reviewed by Piette et al. 1997).
Only 13/30 splice variants were spliced in frame. This means that no protein can be synthesized from the 5‘ RNA fragment. Among the splice variants, which were in frame, all had spliced out the putative nuclear localization signal which is thought to
141be located at aa 181-185 (Fakharzadeh et al. 1991), and the nuclear export signal (NES) which is thought to be located at aa 191-199 (Fig. 5, Fig. 24, Table 17). This suggests that even if these RNA transcripts were translated into truncated proteins, there is no function of these MDM2 variants in the nucleus, especially no possible MDM2/p53 interaction. Most importantly though, the majority of splice variants (22/30) lack part or all of the p53 binding sequence at the N-terminal end of the protein, therefore most likely being unable to inhibit and degrade p53 protein. Some of the splice variants splice out the zinc-finger-like sequence which is as yet of unknown function. Other splice variants have lost part or all of the RING finger motif, which was shown to specifically bind to RNA and suggested to play a role for cell cycle regulation (Elenbaas et al. 1996, Argentini et al. 2000).
Furthermore, different mdm2 transcripts containing or skipping exon 2 according to different promoters have also been described. Translational enhancement of mdm2 involved a preferential increase in mdm2 transcription that was initiated from an internal p53-responsive promoter region of the gene (Landers et al. 1994).
Occurrence of multiple MDM2 proteins has been noted in different tissues and is thought to result from alternative splicing. Protein variants of sizes between 78 kD and 12 kD have been found in non-small cell lung cancer and oral squamous cell carcinomas (Maxwell, 1994, Ralhan et al. 2000). The p74 protein, which lacks the N-terminus and does not bind to p53 (Olson et al. 1993) is expressed in response to UV light in a p53-dependent manner (Saucedo et al. 1999). Both internal initiation at AUG codon 50 and alternate splicing can give rise to p76 in cells (Saucedo et al. 1999). A p76 MDM2 protein was found to antagonize the ability of MDM2 to stimulate the degradation of p53 and lead to an increase in the level of p53 (Perry et al. 2000).
Since most of the mdm2 splice variants lack various functional domains of the gene such as the p53 binding site and the nuclear localization signal, the purpose of the alternative splicing remains unclear. But the fact that splicing is found in the majority of ovarian carcinomas and splice variants of identical sequence occur throughout this cancer cohort as well as in benign tissues suggests that these splice variants have distinct, as yet unknown functions and are not only a by-product of RNA processing.
The MDM2 protein binds to p53, inhibits its function and promotes rapid degradation of the p53 protein (Momand et al. 1992, Haupt et al. 1997, Kubbutat et al. 1997). The p53 binding domain has been located to an N-terminal region of aa 19-102 which is critical for stable interaction with p53 protein in vitro (Chen et al. 1993).
142Out of 30 splice variants in ovarian tumors, 20 (67%) were partially and 2 (7%) completely missing the sequence coding for the p53 binding region at aa 19-102. Mapping experiments of the MDM2/p53 interaction domain had shown that N-terminal deletions in MDM2 mutants or internal deletions affecting part or all of this entire region resulted in significantly reduced or absent binding to p53 protein (Chen et al. 1993). Similar binding regions were found by Oliner et al. (1993) who identified a protein sequence of codons 1-118 of MDM2 to be able to interact with full length p53 (Oliner et al. 1993). Other experiments located the p53 binding domain to amino acids 14- 154 (Leng et al. 1995).
Further experiments, though, suggested that amino acid sequences outside of the region between residues 19 and 102 contribute to MDM2/p53 binding, since the smallest mutant that could bind to p53 protein contained amino acids 1 to 294 (Chen et al. 1993). The MDM2 protein sequences which were suggested to be additionally necessary for a stable MDM2/p53 interaction may comprise either amino acids 102 to 294 or 294 to 491 (Chen et al. 1993). Other studies though, showed that the N-terminal 154 amino acids of MDM2 are sufficient for p53 binding in vivo (Leng et al. 1995).
Furthermore, p53 binding can not only be prevented by large deletions but also by point mutations in the mdm2 sequence (Freedman et al. 1997). Though the wild-type (1-115) fragment of MDM2 was clearly able to bind specifically to p53, two MDM2 mutants with an amino acid exchange at residue 58 and 77, respectively, did not show any detectable level of binding to p53. By binding to conformation-dependent antibodies it was shown that these two MDM2 mutants are not grossly misfolded. Site directed mutagenesis yielded mutations of two additional amino acids of MDM2 (D68 and V75) that prevented binding to p53 in vitro. In a functional essay MDM2 with the point mutations also failed to regulate p53 dependent transactivation of target genes, which is consistent with the assumption that a physical interaction between the two proteins is required for MDM2‘s inhibition of p53 activity (Freedman et al. 1997).
Given these experimental results, most of the splice variants which we found in ovarian cancers would be expected to be unable to bind p53. Only eight splice variants contained the complete p53 binding sequence. All of these were spliced with a frame shift so that only a 5‘ portion of 316 to 696 basepairs could potentially code for a protein that binds to p53, comprising 105 to 232 amino acids of the N-terminal portion of the protein. These splice variants, though containing the entire p53 binding sequence, still lack additional supportive protein sequences which may be required for MDM2/p53 interaction.
In mapping experiments (Chen et al. 1993) the smallest MDM2 mutant that could bind to p53 comprised aa 1-294, while two smaller mutants comprising aa 1-222
143and 6-204 were unable to bind to p53. In ovarian cancer five mdm2 splice variants have previously been identified that were all lacking part of the p53 binding sequence (Sigalas et al. 1996). Interestingly though, the smallest of the splice variants which retained the largest part of the p53 binding domain consisting of a 5‘ portion of 225 basepairs of the coding region was clearly shown to bind to p53 in co-immunoprecipitation experiments (Sigalas et al. 1996). This is in contradiction to the hypothesis that a complete p53 binding sequence plus additional protein sequences are required for binding (Chen et al. 1993).
Recently, a highly conserved mdm2 exon has been identified in human and dog DNA (Veldhoen et al. 1999). The exon is located between exon 4 and 5 of mdm2 and codes for additional 29 amino acids. Expression of exon disrupts in vitro translation of the p53 binding domain of MDM2. The putative MDM2 protein lacks the N-terminus of MDM2 and shows little if any binding capacity to p53 (Veldhoen et al. 1999).
Nevertheless, among the splice variants identified in this ovarian tumor cohort, those which were found most frequently were all missing a large portion of the p53 binding sequence making a MDM2/p53 interaction probably impossible. Some rare splice variants contained the full binding sequence, but these sequences are probably still too short to allow a stable MDM2/p53 interaction. Therefore, we assume that the splice variants which we see in ovarian tumors do not play a role for p53 inhibition, but may possibly fulfill other as yet unknown functions.
The pattern of splicing in many of the variants was interesting since the splice donor and acceptor sites were in regions of exact sequence homology. A sequence of up to 10 bp as for example TTCAAATGAT in the 259 bp splice variant was found at the 3‘ end of the donor site as well as at the untranscribed end of the downstream acceptor splice site. The pattern of deletions between short direct repeats was similar to data from a DNA polymerase delta (pol 3) in Saccharomyces cerevisiae (Tran et al. 1996). A mutation of pol 3 caused an approximately 1000-fold increase in 7-61 deletions between repeat sequences. To clarify whether the deletions in mdm2 cDNA could also result from genomic deletions in the DNA sequence, we performed PCR at the expected site of loss, but did not identify any deletions in genomic mdm2 DNA (data not shown).
Interestingly several splice variants, though different in length shared either an identical cryptic acceptor site sequence or donor site sequence. This has been observed for three donor sites and two acceptor sites, each of which are identical for
144two, three or four splice variants. An identical donor site had been described previously for mdm2-a (888 bp) and mdm2-b (654 bp) (Sigalas et al. 1996). Similar observations have been made for splice variants of the TSG101 gene sharing proximal or distal breakpoints in ovarian cancer (Gayther et al. 1997).
In ovarian cystadenomas and ovarian tumors of borderline malignancy alternative and aberrant splicing of mdm2 was also frequent, but the pattern of splicing was different from ovarian carcinomas. Expression of full length mdm2 was usually present except for three cases, two of which had no detectable mdm2 mRNA at all. Besides full length mdm2, expression of small splice variants was frequent. The 221 bp splice variant was even notable in 5/9 LMP tumors (56%) as opposed to 16% of the ovarian carcinomas. The mdm2-b splice variant though, which is not only frequent in ovarian cancer (41%) but also correlated with poor grade of differentiation and other unfavourable prognostic criteria, was rarely seen in borderline tumors (1/9) and in no case of cystadenomas.
Little is known about mdm2 splicing in benign tumors of the ovary and nothing about LMP tumors. Only five cases of benign ovarian tumors have been analyzed for splice variants and one out of these was described to have alternative splicing without mentioning the exact size of the splice variant (Sigalas et al. 1996).
Our results clearly show that the mdm2-b splice variant predominantly, but not exclusively, occurs in ovarian carcinomas, while small splice variants are frequently seen coexisting with the full length transcript in tumors of borderline malignancy and benign ovarian tumors. This suggests that the mdm2-b splice variant is either involved directly in carcinogenesis or that RNA processing is indirectly affected by alterations of other cell cycle related genes, leading to mdm2 alterations.
Surprisingly, mdm2 alternative splicing also occurred in normal ovarian tissue. Full length mdm2 was, like in ovarian cancer, missing in 2/20 cases (10%). In almost half of the cases small splice variants were noted besides the full length transcript. This is in contrast to previous reports which only found the full length mdm2 transcript in normal tissue (Sigalas et al. 1996).
mdm2 is thought to be expressed under normal conditions only during, but not after embryonic development. During early embryogenesis, the expression of xdm2, the xenopus laevis homolog of mdm2, increases from oocyte stage I/II to reach its maximum in oocyte stage V/VI in unfertilized eggs and then becomes undetectable by
145Northern blot analysis in various differentiated normal tissues after embryonic development (Marechal et al. 1997). Since p53 is stored as a cytoplasmic pool until fertilization, xdm2 expression is possibly independent of transcriptional activation by p53 (Marechal et al. 1997). In the normal ovarian tissue cases of our study, mdm2 expression patterns did not correlate with pre- or postmenopausal status. Therefore, presence or absence of oocytes does not seem to have an influence on mdm2 expression and mdm2 splicing.
Alternative splicing in normal tissue is well known under physiological conditions, for example for muscle proteins such as troponin, but has also been noted for other genes which are typically involved in tumorigenesis. Aberrant splicing in up to 50% of normal tissue or cells including peripheral blood, immortalized B-cell lines and lung parenchyma has been described for the TSG 101 gene (Gayther et al. 1997, Lin et al. 1998). Some of the splice variants were found more frequently in normal cells and the spectrum of splice variants seemed to differ between tumor and normal samples (Gayther et al. 1997). Deletions that are thought to result from alternative splicing were also found for the FHIT gene in normal lymphocytes (Druck et al. 1997). However, concerning other genes which are involved in tumorigenesis, alternative splicing still remains to be a rare phenomenon in normal cells. In normal peripheral blood samples, no mis-splicing of BRCA1, BRCA2, BRUSH1, hMSH2, IGF2 receptor PGDB or RB was seen (Gayther et al. 1997).
In our study 221 bp splice variant was found in 40% of the normal ovaries. This is interesting since this variant is rare in ovarian carcinomas and if present, is associated with a more favourable outcome in terms of survival. Remarkably, the splice variant mdm2-b which seems to be clearly associated with a more aggressive type of ovarian carcinomas, was found in abundance and as the sole transcript in the normal ovary of one woman who died of liver failure. Generally, presence or absence of splicing in normal ovarian tissue was not related to age, day of menstrual cycle or menopause. Interestingly though, if the uterus and ovaries had been removed for a benign disease, mdm2 splicing was rare as opposed to those cases in which the contralateral ovary or the uterus had been affected by cancer.
The presence of mdm2 splice variants in normal ovarian tissue suggests that mdm2 splicing is not generally associated with malignant transformation of cells. It remains unclear though, which possible function a splice variant of 221 bp could have that splices out most of the functional regions of the gene and in addition is spliced out of frame. The fact that it occurs predominantly in normal tissue and borderline tumors raises the question whether it is a waste product of RNA processing in an otherwise normally functioning mdm2 gene. Interestingly though, a distinct pattern of splice variants in normal tissue as opposed to ovarian cancer is notable.
Full length MDM2, p53 and mdm2 splice variant sequences were cloned into the pcDNA3 expression vector and transient transfection of vaccinia virus infected HeLa cells with the vector construct was used to express MDM2 and p53 proteins. We were able to express the full length MDM2 and p53 protein as well as a 40 kD protein from the mdm2-b (654 bp) splice variant. Two smaller splice variants, both spliced in frame, did not express a protein detectable on Western blot, possibly due to degradation of the protein. Most of the splice variants were not suitable for these experiments at all because of out of frame splicing.
We attempted to analyze binding of the splice variants to p53 by co-immunoprecipitation experiments and were able to show binding between full length MDM2 and full length p53, which is a requirement for further binding experiments (data not shown). The co-immunoprecipitation experiments of p53 and mdm2 splice variants though, could not be successfully completed yet. Mapping studies have already shown that the N-terminal region between aa 19 and 102 is most likely necessary for p53 binding (Chen et al. 1993). Co-immunoexpression experiments have shown that mdm2-b and other splice variants which lack a major portion of the p53 binding site are unable to bind p53 (Sigalas et al. 1996).
Different sizes of MDM2 proteins have been noted previously in cells and have been suggested to stem from alternative splicing. Besides the normal size MDM2 protein of 90 kD, four additional polypeptides (p85, p76, p74, and p58-p57) were identified in mouse cells (Olson et al. 1993). Antibody binding experiments revealed that p76/74 have lost the N-terminal epitopes and the p58/p57 protein is missing the carboxy-terminus epitope (Olson et al. 1993). Similar sizes of proteins of 76/78 kD, 57/59 kD and 46 kD which were identified in non small cell lung cancer were associated with different nuclear substructures and may potentially interact with transcription factors other than p53 (Maxwell, 1994). The presence of distinct MDM2 proteins of 54kD-68kD was also confirmed in breast epithelial cells (Gudas et al. 1995). In oral carcinomas three isoforms of MDM2 (90 kD, 76 kD, and 57 kD) were identified and showed differential compartmentalization in the cells (Ralhan et al. 2000).
The splice variants though, which we found by RT-PCR and sequencing, were mostly much smaller in size. One of the larger variants of 654 bp codes for an approximately 40kD protein, but we were unable to express smaller proteins which may have been too unstable.
p53 mutations often lead to accumulation of the altered protein due to a prolonged half-life. However, protein accumulation was also seen in 38% of the ovarian carcinomas which had the wildtype p53 sequence. Since MDM2 promotes nuclear export and degradation of the p53 protein, our hypothesis was that cases with p53 protein overexpression in the absence of mutation might have mdm2 alterations, which cause accumulation of normal p53 protein. In fact, in this group we found mdm2 alterations in 11/13 cases (85%). Cases with p53 mutation and p53 overexpression also had a high proportion of abnormal mdm2 (71%). Cases without normal p53 expression, regardless of whether they had a wildtype or mutated p53 sequence, had a lower percentage of mdm2 alterations (63%), but the results did not reach statistical significance.
p53 mutations were not correlated with any form of mdm2 alternative or aberrant splicing. However, p53 protein overexpression was found significantly more often in cases with the mdm2-b splice variant. Wildtype p53 protein stabilization through mdm2 splicing has also been found in glioblastomas (Kraus et al. 1999). This is consistent with findings in MDM2 mutant experiments. Expression of the MDM2 mutant ( 222-437) not only failed to target degradation of p53, but also resulted in a significant elevation of p53 levels (Kubbutat et al. 1997). Since these cells express endogenous full length MDM2, it was suggested that the MDM2 mutant can, at significantly high levels of overexpression, act in a dominant negative manner, protecting p53 from degradation by endogenous MDM2 protein (Kubbutat et al. 1997).
These results are corroborated by recent findings about p76 MDM2, a protein which lacks the first 49 amino acids of p90 MDM2. Overexpression of p76 MDM2 was shown to antagonize the ability of MDM2 to stimulate the degradation of p53 and lead to an increase in the levels and activity of p53 (Perry et al. 2000). Furthermore, the mdm2-P2 promoter which is a transcriptional target of p53, was found overexpressed in oral carcinomas which both overexpressed MDM2 and p53 proteins despite wild-type p53 sequence. The authors suggest that enhanced translation of mdm2-P2 transcripts may represent an important mechanism of overexpression, subsequent stabilization and functional inactivation of p53 (Ralhan et al. 2000).
Although we observed expression of both full length and splice variant mdm2, especially the mdm2-b splice variant is present in abundant quantity in many cases of ovarian cancer. It may therefore, have a dominant negative effect even if it coexists with the full length mdm2 transcript. Since this splice variant has lost the p53 binding ability, it may cause accumulation of the p53 protein despite p53 wildtype sequence.
The mdm2-b splice variant (654 bp) was not only the most frequent mdm2 alteration in ovarian cancers and rare in borderline tumors as well as normal tissues. It was also significantly correlated with more aggressive tumors of high grade, high S-phase fraction and presence of residual tumor after surgery. All of this is contributing to a poorer outcome of these tumors in terms of survival. This confirms the results of a previous study in which mdm2 splice variants were associated with poor differentiation and advanced stage (Sigalas et al. 1996).
These findings indicate that this specific form of mdm2 splicing plays a role in tumorigenesis and tumor progression. Similar observations were made for astrocytic neoplasms. mdm2 splice variants, predominantly mdm2-b, were significantly more often found in glioblastomas than in lower grade astrocytomas (Matsumoto et al. 1998). The mdm2-b splice variant, as well as other variants, were found to have transforming ability in NIH 3T3 cells (Sigalas et al. 1996).
Given that most of the splice variants lack a major part of the p53 binding domain, we assume that they cannot inhibit p53 function. But since p53 itself is dysfunctional through mutations in most of the ovarian cancer cases, this does not lead to a more stable p53 function for cell cycle control and apoptosis. Mechanisms which cause mdm2 splicing though might be an attempt to disrupt MDM2/p53 interaction in a cell that has undergone DNA damage in order to prevent mdm2 mediated p53 inhibition and degradation.
Absence of the full length mdm2 transcript and presence of splice variants was significantly associated with increased resistance to platinum-based chemotherapy in ovarian cancer. This is, though numbers are small, an interesting but unexpected result. Current knowledge about platinum based chemotherapy and cell cycle regulation suggests that a functional p53 is necessary to induce apoptosis (Lowe et al. 1993). Resistance to chemotherapy has been found in our study and by others more frequently in tumors with p53 overexpression and p53 missense mutations. A functional MDM2 would inhibit p53 and therefore, potentially make the tumor less prone to undergo apoptosis after treatment with chemotherapy. In contrast, mdm2 splice products which have lost p53 binding ability, might indirectly contribute to p53 stability and enhance its function. Recent experiments with fibroblasts from p53/mdm2 null mice have shown, that loss of mdm2 can induce the p53-dependent apoptotic pathway in vivo (de Rozieres et al. 2000).
149However, the opposite was suggested by our results. Tumors with functional mdm2 were more sensitive to chemotherapy than those with splice variants. Among tumors resistant to platinum based chemotherapy is a high percentage of cases with p53 alterations. Mutant p53 though, does not induce mdm2 in the same manner as wildtype p53, possibly leading to undetectable amounts of mdm2 full length transcript. The underlying regulatory mechanisms for this result remains presently unclear. Further studies would be needed to clarify whether chemotherapy-sensitivity actually depends on a functional mdm2, or whether the above described findings are indirectly influenced by p53 alterations.
In terms of prognosis, no correlation was found between presence of full length mdm2 nor presence of mdm2 alterations in general, and clinical outcome. A trend though, towards poorer outcome was noted in cases that expressed the mdm2-b splice variant, which also correlated with clinical and histopathological indicators of poor prognosis such as grading and residual tumor after surgery. The clinical correlations of mdm2-b though, suggest that this splice variant is associated with tumorigenesis and tumor progression.
Interestingly, presence of the splice variant 221 bp, which was more frequently seen in benign and normal tissues, was significantly associated with better outcome. Though this splice variant has lost most of the important functional domains, we assume that its occurrence is associated with a relatively intact cell cycle regulation and less aggressive tumors. However, further studies will be necessary to clarify whether alternative and aberrant splicing of mdm2 is important in tumorigenesis and progression of ovarian cancer, or whether these transcripts are merely a product of altered mRNA splicing fidelity that occurs more commonly and in a different pattern in tumors as compared to normal tissues.
Summarizing these results, we could demonstrate that mdm2 alternative and aberrant RNA splicing is frequent in ovarian tumors as well as normal ovaries. Distinct splicing patterns were identified in ovarian carcinomas as opposed to benign tumors and normal tissue. The most frequent of the splice variants, mdm2-b, correlates with p53 protein accumulation in ovarian cancer as well as features of biologically more aggressive tumors.
DiML DTD Version 2.0||
der Humboldt-Universität zu Berlin
|HTML - Version erstellt am:|
Thu Aug 15 15:17:40 2002