Evaluation of p53 protein expression as a marker for long-term prognosis in colorectal carcinoma.

Mutation of the p53 gene is reported to be of prognostic importance in colorectal carcinomas. Immunohistochemical staining of the accumulated p53 gene product may be a simple alternative for p53 mutation analysis. Previous studies addressing the prognostic importance of p53 expression, however, yielded contradictory results. Therefore, we evaluated the importance of p53 expression as a marker for long-term prognosis in a well-characterised study population of 109 colorectal carcinomas. After antigen retrieval with target unmasking fluid (TUF), immunostaining of p53 was performed with both monoclonal antibody DO7 and polyclonal antibody CM1. Objective quantification of the p53 signal was assessed by a computerised image analyser. p53 expression was higher in non-mucinous tumours than in mucinous tumours (p53 labelling index = 30% and 17% respectively, P = 0.05), and in metastatic tumours compared with non-metastatic tumours (p53 labelling index = 37% and 22% respectively, P = 0.05). Other histopathological features were not related to p53 expression. In multivariate analysis, Dukes' stage (P = 0.02) and histological grade (P = 0.05) stood out as independent markers for prognosis. p53 expression was not an independent marker for prognosis. At present, p53 expression is not a useful marker for long-term prognosis. Further insight into the relationship between p53 mutations and p53 expression is needed to elucidate more precisely the clinical relevance of p53 alterations.

The p53-suppressor gene is the most frequently altered gene in solid human malignancies (reviewed in Lane, 1992;Levine, 1992aLevine, , 1992bOren, 1992;Vogelstein and Kinzler, 1992). It is located on the short arm of chromosome 17 in the region 17pl3 and encodes a 53 kDa nuclear phosphoprotein that serves as a transcription factor El-Deiry et al., 1993). The p53 protein indirectly regulates cell growth and inhibits cells with mutagenic damage from entering the S-phase by arresting the cell cycle in GI, during which DNA repair can proceed.
In colorectal cancer, p53 mutations are frequently accompanied by allelic loss of 17p (Baker et al., 1989;Rodrigues et al., 1990). Both p53 mutations and allelic deletion of 17p occur late in tumour progression (Baker et al., 1990) and are reported to have prognostic value after surgery (Kern et al., 1989;Laurent-Puig et al., 1992;Offerhaus et al., 1992;Hamelin et al., 1994). However, both detection of p53 mutations at the DNA level and detection of allelic deletion of 17p by restriction fragment length polymorphism (RFLP) analysis are cumbersome procedures and therefore not feasible in routine diagnosis.
Recent reports describe a strong relation between p53 gene mutations and mutant p53 protein expression (Rodrigues et al., 1990;de Angelis et al., 1993;Baas et al., 1994). The mutant p53 protein is characterised by a conformational change resulting in prolonged half-life and stability, enabling its detection by routine immunohistochemical (IHC) techniques (Finlay et al., 1988). Therefore, immunostaining of the p53 protein may be an important surrogate test for p53 mutation analysis.
In solid neoplasms including carcinomas of the breast (Barnes et al., 1993), stomach (Martin et al., 1992;Starzynska et al., 1992), lung (Quinlan et al., 1992), ovary (Bosanr et al., 1993) and pancreas (DiGuiseppe et al., 1994), p53 expression has been correlated with shortened survival. In colorectal carcinomas, however, a correlation between survival and nuclear p53 expression has not been consistently observed (Scott et al., 1991;Remvikos et al., 1992;Starzynska et Yamaguchi et al.. 1992;Bell et al., 1993;Bosari et al., 1994;Nathanson et al., 1994). The contradictory results of these studies might be partly due to the variability in IHC techniques used. Moreover, most followup studies lacked statistical power owing to relatively small patient populations or limited follow-up periods. Therefore, in this study we analysed the value of p53 protein expression for long-term prognosis in a large, wellcharacterised study population with over 20 years of followup. p53 expression was evaluated by two different anti-p53 antibodies, which in a previous study stood out as being most accurate for p53 protein detection and association with p53 gene mutation (Baas et al.. 1994). p53 expression was objectively scored by a computerised image analyser. In addition, we evaluated the relationship between p53 expression and other histopathological parameters known to be of importance in colorectal cancer.

Materia and methods
Stud) population andfollow-up The original study population consisted of 155 patients with colorectal carcinoma, operated on between 1967 and 1974 in the University Hospital of Leiden. The study population had previously undergone extensive research for a large number of histopathological parameters which have a bearing on tumour biology and prognosis (Bloem. 1983;Offerhaus et al., 1991). For the present p53 immunostudy. tissue blocks were available from 109 patients only. These patients did not differ significantly from the original 155 with respect to age, sex or the histopathological parameters. Histopathological parameters were determined by review of slides; location of the tumour and macroscopic aspect, together with the patient characteristics, were collected from the medical records by review of charts.
In the p53-tested cohort of 109 patients, there were 56 men and 53 women; the median age was 66 years (mean age 65 years, s.d. = 10 years, range 25-96 years). Twenty-two tumours were located in the caecum or ascending colon, eight in the transverse colon or splenic flexure, 46 in the descending colon or sigmoid and 33 tumours in the rectum. Tumours p53 p in.-'mlau-cudImi x~~~~~~~~~~~~~~~J-WR MukIer eta were staged according to the modified Dukes' classification (Dukes, 1932;Turnbull et at., 1%7). Eighteen patets had a Dukes' A carcoma (confired within the muularis propria), 61 patients had a Dukes' B carcinoma (extension through the muscularis propria into the pericolic fat), 27 patients had a Dukes' C carcinoma (positive regional lymph nodes without distant metastases), and three patients had a Dukes' D carcinoma (either invasion of adjacent organs or evidence of distant metastases). Twenty carcinomas were wel differentiated, 67 were moderately differentiated and 22 were poorly differentiated. Twenty-nine of the 109 tumours were mucinous carcinomas (defined as at lat 30% of the volume being occupied by muce lakes) (Mecklin et al., 1986). Of those cases in which the macroscopic aspect was reliably reported, 34 tumours showed exophytic growth and 55 tumours showed ulcerative growth. Fourteen tumours showed 'Crohn's-like' lymphocytic infiltration (Jass, 1986;Graham and Appehtnan, 1990); in five cases the presce or absence of lymphocytic infiltration was not evaluable. In 102 tumours vasoinvasion of tumour cells was stdied by Van Gieson's elastic stain and a factor VIII immunoperoxidase method for the locaisation of endotheial cells (Muklai et al., 1980;Muller et al., 1989;Offerhaus et al., 1991).
Follow-up was obtained through physician contact and ended on 30 September 1993.

Image anatysis
The CAS 200 image analysis system consit of a conventional microscope with mounted television camera which is linked to a computer and colour monitor. The ER/PR software program enabled measurement of the total area of positive nulcear staining in any selected microscopic field, while the methyl green nuclear counterstain enabled measurement of the total nuclear area. The ratio expressed as p53 labelling index (LI) gave an objective value for the percentage of positive-staining nuclei. Baseline was set on p53negative normal mucosa. Negative stromal ekments were controlled for by computing the mean p53 LI for each slide in at least five representative fields at 400 x magification, containing between 100 and 250 tumour nuclei (Baas et al., 1994).

Statistical analysis
Statistical analysis was performed with JMP software (SAS Institute, Cary, NC, USA). For survival analysis p53 expression was divided into three groups: (1) no nuclear p53 expression (LI<1l%), (2) low nuclear p53 expression (LI 1-30%), and (3) high nuclear p53 expression (LI > 30%). This tripartition is based on the results of previous studies by our group (Baas et al., 1994). Survival analysis for the other histopathological parameters was assessd in both the p53-tested cohort and the original cohort. One patient died within 30 days of surgery, and was therefore excuded from survival analysis. Kaplan-Meier survival curves were alculated and tested for significance by an univariate log-rank statistic.
These cures cluded only colorectal cancer-related deaths as events. Deaths from other causes were treated as censored events at time of occurrence. One patient was lost to followup after 11.8 years and treated as a censored event from that time. The independent prognostic value of parameters was tested using the multivariate Cox regression model. Correlation between p53 expression and histopathological parameters was tested using a t-test statistic or analysis of variance (ANOVA) for multiple means.
Rets p53 expression p53 immunostaining was initially evaluated by conventional light microscopy by two authors who were blinded for other   Table I shows that the results of conventional evaluation of p53 expression slightly differ from the amount of p53 protein when quantified by a computerised image analyser. Thirty-one (28%) carcinomas showed no p53 expression (LI <1%), 35 (32%) carcinomas showed low p53 expression (LI 1-30%) and 43 (40%) carcinomas showed high p53 expression (LI> 30%).
p53 expression and histopathological parameters The associations between p53 expression and histopathological parameters are listed in Table II overall increase in p53 expression with advancing Dukes' stage was observed (P = 0.05). Mucinous tumours expressed significantly less p53 protein than non-mucinous tumours (P = 0.05). No significant difference in p53 expression with regard to sex, age, macroscopic aspect, lymphocytic infiltration or vasoinvasion by tumour cells was observed.

Survival analysis
The results of the univariate survival analysis in the p53tested study group and the original complete cohort are listed in Table III. The median period between surgery and death or last physician contact was 7.3 years in both groups. During follow-up, in the complete cohort 57 (37%) patients died of colorectal carcinoma and 65 (42%) patients died of causes unrelated to colorectal cancer. In the p53-tested In the complete cohort, increase in Dukes' stage ( Figure  la), poorer grade of differentiation, ulcerative growth, lymphocytic infiltration and vasoinvasion was related to worse prognosis. In the smaller p53 study subset, only Dukes' stage (Figure lb) and grade of differentiation reained significnt. The amount of p53 expression in carcinomas showed a tendency towards an association with patient survival (P = 0.08), but the pattern of this relationship is difficult to interpret: the highest and lowest p53 categories showed the poorest survival, whereas the intermediate p53 category showed the best prognosis (Figure ic).
In the complete cohort, multivanate analysis showed that only Dukes' stage (P = <0.001) and differentiation grade (P = 0.01) were independent markers for prognosis. These parameters were also independent predictors of prognosis in the smaler p53-tested cohort (Dukes' stage, P = 0.02; differentiation grade, P = 0.05) ( Table III).
The prognosis after seemingly curative resection of colorectal carcinoma depends largely on the absence or presence of occult metastases, often accounting for mortality. Prediction of outcome is currently based mainly on the stage of colorectal carcinoma at time of resection. However, patients with tumours of the same stage often show dramatically different outcome. Therefore, more specific prognostic markers would provide a rationale to adjust different therapeutic approaches.
Alterations of the p53 tumour-suppressor gene are potentially such a marker (Hamelin et al., 1994), and immunostaining of the p53 protein product could be a valuable test for p53 gene alterations (Rodrigues et al., 1990;de Angels et al., 1993;Baas et al., 1994).
Previous studies addressing the prognostic value of p53 were mostly restricted to short-term follow-up, and in particular the IHC studies yielded variable results (Table IV). This variability might come from several causes. First of all, these study groups may not always be comparable. Moreover, the use of various antibodies against different epitopes of the p53 protein, and sometimes the use of antigen retrieval systems, may also account for some of the variability in the percentages of p53 positivity seen among the different studies (Table IV) (van den Berg et al., 1993;reviewed in Wynford-Thomas, 1992;Hall and Lane, 1994). We previously evaluated various procedures and six different p53 antibodies in relationship with underlying p53 changes and selected the two most accurate procedures for p53 protein detection for use in this study (Baas et al., 1994). Objective quantification of p53 expression was achieved by use of a computerised image analyser. Opfimised IHC techniques combined with computerised quantification yielded p53 positivity in 72% of the carcinomas (Table I), a percentage that exceeds that of all previous studies (Table IV).
-p53 ard kxg4em pgnosis m coa *talcanoma  Scott et al., 1991;Hanski et al., 1992;Remvikos et al., 1992;Starzynska et al., 1992;Yamaguchi et al., 1992;Bell et al., 1993;de Angelis et al., 1993;Kaklamanis et al., 1993;Bosari et al., 1994;Nathanson et al., 1994). Established markers for prognosis such as Dukes' stage and differentiation grade were independently associated with survival, validating this study population. In this long-term follow-up study we found that nuclear p53 protein expression is not related to outcome after surgery. This result is similar to previous studies (Scott et al., 1991;Sun et al., 1992;Bell et al., 1993;Nathanson et al., 1994). Yamaguchi et al. (1992) reported p53 positivity to be of independent prognostic importance, but in their multivariate analysis Dukes' stage was not included. Both Bosari et al. (1994) and Sun et al. (1992) found cytoplasmic p53 staining with PAb CM1 to be of independent prognostic importance, but as in most other studies no reliable cytoplasmic staining was found in our study.
The lack of consistent prognostic value of nuclear p53 protein expression might indicate that the reported importance for prognosis of p53 mutation needs additional study and/or that the relationship between immunostaining of p53 protein and p53 gene mutation might be too much confounded by other biological mechanisms and technical caveats (reviewed in Wynford-Thomas, 1992;Hall and Lane, 1994).
In conclusion, this study indicates that, with current methodology, p53 protein expression does not appear to contribute to the prediction of long-term prognosis after resection of colorectal carcinoma. We emphasise that further insight into the relationship between p53 gene mutations and p53 protein expression is needed to elucidate more precisely their clinical relevance.