p21/WAF1 expression in human colorectal carcinoma: association with p53, transcription factor AP-2 and prognosis

p21/WAF1 expression was studied in a series of 162 colorectal carcinoma patients and its relation to p53- and activator protein (AP)-2 expressions and to stage as well as survival was assessed. p21 expression was moderate or intense in 33% of the tumours, and 53% of the tumours had moderate or strong p53 staining intensity. Eighty-nine percent of the tumours showed a weak cytoplasmic AP-2 signal. As expected, p21 and p53 stainings were inversely related to each other (P < 0.001). There was a significant positive association between p21 and AP-2 expression levels (P = 0.01). p21 intensity and percentage were higher in Dukes' A and B stages (P < 0.001). The cancer-related survival and recurrence-free survival (RFS) rates were significantly lower among patients with a low signal for p21 (P < 0.001) and low p21 percentage in tumour epithelium (P < 0.001). High p53 staining intensity in tumour epithelium predicted poor survival (P = 0.01) and RFS (P = 0.003). In the multivariate analysis, p21 percentage distribution independently predicted cancer-related survival in all cases, and p21 expression intensity in T1–4/N0–3/M0 and T1–3/N0/M0 cases. p21 percentage distribution was an independent predictor of RFS in all and T1–3/ N0/M0 cases. AP-2 staining did not reach any prognostic significance. These results suggest that the immunohistochemical detection of cyclin-dependent kinase inhibitor p21 could be used to predict more precisely the outcome of colorectal cancer patients. © 1999 Cancer Research Campaign

radiographs, bone scans, computerized tomography, colography, endoscopy (recto-sigmoidoscopy and colonoscopy) and laboratory tests reflecting the possible metastasis. All 162 patients underwent operations and, of these, 22 were additionally treated with chemotherapy and 17 with radiotherapy. The follow-up was done according to the standard practice used in our clinic by the same team of gastroenterologists. The pertinent data of the patients are summarized in Table 1.

Histology
The tumour samples obtained in the operation were immediately fixed in 10% buffered formalin (pH 7.0), and later embedded in paraffin. Several original sections from each of the primary tumours were re-examined by two observers unaware of the clinical data or the disease outcome, and the most representative tissue block was selected, cut at 5-µm thickness and stained with haematoxylin and eosin (H & E). Tumours were graded as well, moderately, or poorly differentiated (WHO grade). The patientsÕ histopathological data are shown in Table 1.

Immunohistochemistry and detection of p21, p53 and AP-2
Five-micrometre-thick paraffin sections were rehydrated and washed twice for 5 min with phosphate-buffered saline (PBS). The sections were heated in a microwave oven in 0.05 mol l Ð1 TrisÐHCl buffer (pH 9.7) for 2 × 5 min. Endogenous peroxidase activity was blocked with 5% hydrogen peroxide for 5 min, followed with a wash for 2 × 5 min with PBS. The tissue sections were incubated overnight at + 4°C with a p21-specific mouse monoclonal antibody (NCL-WAF-1, Novocastra Laboratories Ltd, UK) at a working dilution of 1:10. Samples were washed twice for 5 min with PBS and incubated for 30 min with biotinylated secondary antibody (Vectastain ABC Elite Kit, Vector Laboratories, Burlingame, CA, USA) in PBS. After two 5-min washings in PBS, the sections were incubated for 40 min in avidinÐbiotin peroxidase detection solution. Samples were washed 2 × 5 min with PBS, developed with diaminobenzidine tetrahydrochloride (DAB; Sigma, St Louis, MO, USA) for 5 min, slightly counterstained with MayersÕs haematoxylin, dehydrated, cleared and mounted with DePex (BDH Supplies, Poole, UK).
The p53 immunoreactivity was demonstrated by means of the same staining protocol. We used a monoclonal DO7 (Dako, Denmark) antibody, known to be specific for both mutant and wild-type forms of p53 protein at a working dilution of 1:1000. For antigen retrieval the samples were boiled in a microwave oven twice for 5 min in citrate buffer (pH 6.0).
The staining protocol for AP-2 has been published previously (Hietala et al, 1997). After eliminating non-specific staining with 1.5% normal goat serum for 30 min, the tissue sections were incubated overnight at +4°C with a primary antibody. We used a rabbit polyclonal AP-2α (C-18) (Santa Cruz Biotechnology, Santa Cruz, CA, USA) antibody, raised against a peptide corresponding to amino acids 420Ð437 mapping at the carboxy terminus of AP-2 of human origin, at a working dilution of 1:2500. After washings, bound antibody was localized using the same method as described above for p21. In each staining batch a section of human cutaneous malignant melanoma (p21, AP-2) or colorectal carcinoma (p53) with a known positivity for antibodies was used as a positive control for p21, p53 and AP-2. Omission of the antibody served as a negative control.
All slides were evaluated simultaneously with a dual-head microscope (field diameter 490 µm) the observers (KMR and VMK for p21 and AP-2; TP and VMK for p53) being unaware of the clinical data. Disagreement in the assessment of staining was found in fewer than 10% of the slides examined and consensus was reached on further review. The staining intensity of p21, p53 and AP-2 in tumour epithelium and per cent distribution was analysed from the most representative tissue section. p21 and p53 staining intensities in tumour epithelium were scored as follows: Absent staining was 0 If the staining intensity of tumour cell nuclei was clearly weaker than the positive control, the staining intensity was considered weak (1) Moderate or strong (2) expression was defined as dark brown staining comparable to the positive control AP-2 intensity was either negative (0) or weak (1).
For evaluation of AP-2 expression, positive inflammatory cells within the tumour and a section of human cutaneous melanoma were used as positive controls. The fraction of positive tumour cells was primarily analysed in a continuous scale but, for statistical calculations, the percentage distribution of p21 and AP-2 was divided into two groups using median as a cut-off level: ² 10%, > 10%, ² 30%, > 30% respectively. Finally, the percentage distribution of p53 was scored as previously: ² 20%, > 20% (Hirvikoski et al, 1997).

Analysis of lymphocyte density
The density of tumour-infiltrating lymphocytes (TILs) was graded by two observers as absent, weak, moderate or dense, as described previously (Ropponen et al, 1997). In brief, the TIL level was quantified from ten microscopic fields (×40), and the mean TIL value was calculated. TIL density was classified as absent (grade 0) when < 10 lymphocytes were observed per high-power field (HPF). TIL density was weak (grade 1) when scanty lymphocytes were seen in the stroma and the inflammatory cell reaction around the tumour was mild, i.e. 10Ð50 lymphocytes per HPF. TIL density was moderate (grade 2) when peri-and/or intra-tumoural lymphocyte infiltrate was intermediate between grades 1 and 3, i.e. lymphocyte level 51Ð100 per HPF. TIL density was dense (grade 3) when the tumour margins and stroma contained > 100 lymphocytes per HPF.

Mitotic indices
Mitotic figures were counted in the area of the highest mitotic frequency, and ten consecutive high-power microscopic fields (×40) were selected (field diameter 490 µm). The volumecorrected mitotic index (M/V index) method was used in evaluating the mitotic frequency as originally described (Haapasalo et al, 1989). The M/V index expresses the number of mitotic figures per mm 2 of neoplastic epithelium in the microscopic image.

Statistical analysis
In statistical calculations, the SPSS-X programme package was used in an IBM computer. The differences between the means of continuous variables were tested by analysis of variance (ANOVA), and the frequency distributions by the χ 2 test (Mantel and Haenszel, 1992). The univariate survival analysis was based on the life-table method (log-rank analysis) with statistics by Gehan. Cancer-related survival was measured from the date of surgery to the end of follow-up or death. Recurrence-free survival (RFS) was defined as the time elapsed between the primary treatment (date of surgery) and the date of recurrent tumour. Cases with metastases at diagnosis were not included when the RFS was analysed. The causes of death were verified from patient files and death certificates. Multivariate survival analysis (CoxÕs regression analysis) was performed with the SPSS-X programme package in a step-wise manner (Cox, 1972), and P ² 0.05 and P ³ 0.01 were statistically significant for entry and removal limits respectively. CoxÕs proportional hazard model was used to assess the contribution of the following base-line covariates: age, sex, tumour grade, tumour site (tumours involving rectum n = 78, or not n = 84), DukesÕ classification, TNM stage, TILs, p21, p53, AP-2 and M/V index. In the survival analyses, low p53-and p21 intensities (0 and 1) were considered as one group, and moderate and strong intensity (2) was considered as another group.

Correlation of p21, p53, and AP-2 to other prognostic factors
The expressions of p53 and p21 correlated inversely. As hypothesized, the AP-2 and p21 expressions were positively associated with each other (Tables 3 and 4). The percentage of p53-reacting cells correlated inversely with p21 intensity (ANOVA, F = 5.24, P = 0.006). The percentages of p53-and p21-reacting cells were inversely interrelated (ANOVA, F = 4.23, P = 0.04). There were no significant correlations between tumour grade and p21, p53 or AP-2 expressions. In DukesÕ A and B tumours, p21 expression intensity and percentage distribution were higher in tumour epithelium than in DukesÕ C and D tumours (Table 3). p53 expression intensity and percentage distribution in tumour epithelium and AP-2 percentage distribution did not correlate with DukesÕ. No significant associations were found between p53, AP-2 and TILs or M/V index. p21 expression correlated with M/V index; low p21 signal was associated with high M/V index (Table 4).

Univariate survival analysis
In univariate survival analysis patients with negative or weak (0 and 1) p21 expression intensity in tumour epithelium had lower survival and RFS rates as compared to patients with moderate or strong p21 expression intensity (2) (Figures 2 and 3). p21 percentage distribution in tumour cells correlated with survival and RFS as well. Patients with over 10% distribution of p21 had higher survival and RFS rate as compared to patients with ² 10% distribution of p21 (Figures 4 and 5). Patients with negative or weak (0 and 1) p53 expression intensity in tumour epithelium had both higher survival and RFS rate as compared to patients with moderate or strong p53 expression intensity (2) in their tumour epithelium (χ 2 = 5.9, P = 0.01; χ 2 = 8.9, P = 0.003 respectively). p53 and AP-2 percentages were not related to cancer-related or RFS (data not shown). In addition, the combination of the results of p21 and p53 expression (p53Ð/p21Ð, p53+/p21Ð, p53Ð/p21+, p53+/p21+) had no effect on survival or RFS in all patients or in those with DukesÕ B stage tumours, probably due to noncomparable group sizes (data not shown). T1Ð3/N0/M0 subgroup of patients with moderate or strong p21 expression intensity (2) in tumour epithelium had higher survival rate (90% survived 10 years) as compared to patients with negative or weak p21 expression intensity (0 and 1) in their tumour epithelium (35% survived 10 years) (χ 2 = 19.3, P < 0.001). Low p21 expression intensity in tumour epithelium (0 and 1) predicted poorer RFS rate (35% survived recurrence-free 10 years) as compared to patients with moderate or strong p21 expression  (70) 23 (20)  2 (87) 26 (64) 46 (67) 15 (28) 10 (14) χ 2 = 34.6, P < 0.001 a F = 8.3, P < 0.001 b a χ 2 test; b Analysis of variance.  The survival of 162 patients categorized according to the p21 expression intensity in tumour epithelium. The difference between the curves is significant (χ 2 = 48.7, P < 0.001, Curve 0 and 1 = p21 intensity 0 and 1, n = 109; curve 2 = p21 intensity 2, n = 53) intensity (2) in tumour epithelium (80% survived recurrence-free 10 years) (Χ 2 = 14.0, P < 0.001). Similarly patients with low percentage fraction of p21 (² 10%) had lower RFS rate (35% survived recurrence-free 10 years) as compared to patients with high percentage fraction of p21 (> 10%) (70% survived recurrence-free 10 years) (Χ 2 = 10.0, P = 0.002). Patients with negative or weak p53 expression intensity in tumour epithelium (0 and 1) survived longer as compared to patients with moderate or strong (2) p53 expression intensity (Χ 2 = 13.5, P = 0.004 and Χ 2 = 4.2, P = 0.04 for RFS). Chemotherapy or radiation therapy in connection with primary surgical therapy had no effect on survival. Chemotherapy or radiation therapy for treatment of the recurrent disease had no prognostic value since the therapies were given at a very late stage of the disease as a palliative measure. In addition the type of surgery had no effect on survival or RFS.

DISCUSSION
In normal intestinal mucosa, p21 immunoreactivity has been mostly observed in the nuclei of the upper third of the crypts and in the surface epithelium Sasaki et al, 1996;Yasui et al, 1997), which is in line with our observation. In colorectal and urothelial cancers, p21 expression has been reported to be heterogeneous and decreased compared to normal tissue Sasaki et al, 1996;Clasen et al, 1998). The majority (75%) of the tumours in our study showed p21 positivity, and 43% expressed p21 moderately or strongly. The overexpression of p21 suppresses tumour growth in several experimental models (Bae et al, 1995;Chen et al, 1995;Li et al, 1995;Tenan et al, 1995). However, p21 overexpression has been associated with higher histological grade in breast cancer and non-small-cell lung cancer (Caffo et al, 1996;Bennett et al, 1998;Rey et al, 1998). In our study, p21 staining did not correlate with tumour grade, which is in accordance with several recent observations in different neoplasms Slebos et al, 1996;Byrne et al, 1997). Overexpression of p21 has also been observed in low-grade (IÐII) urothelial tumours (Clasen et al, 1998). In addition, Doglioni et al (1996), Wang et al (1997) and Yasui et al (1997) described higher p21 expression in stages 0Ð2 colon carcinomas than in stages 3 and 4 carcinomas. There is also evidence supporting the hypothesis that p21 may exert its tumour suppressive effects in the early stages of tumour development (Ogawa et al, 1997;Clasen et al, 1998) to which the present study gives further strengthening. However, some cancer studies have not revealed any association between tumour stage and p21 expression (Slebos et al, 1996;Byrne et al, 1997;Bennet et al, 1998). Taking together the available information it is obvious that additional data are needed to explain p21Õs multiplex functions more precisely during cancer progression.
The relative p21 mRNA levels have been suppressed in tumours with p53 mutations, indicating the necessicity of wild-type p53 for p21 activation (Matsushita et al, 1996). In the present study, p21 overexpression correlated with low p53 level (by DO7 antibody), which is in line with other recent papers Sasaki et al, 1996). Doglioni et al (1996) have suggested that high p21 expression could be related to normal or increased p53 function, and low p21 expression could be related to inactivation of p53 function. This, in other words, represents p53-dependent p21 induction pathway. Similar results were found in breast cancer by Ellis et al (1997) and Bukholm et al (1997). Interestingly, DNAbased mutational analysis did not reveal mutations to be responsible for the changed regulatory effects of p53 in their study (Ellis et al, 1997). It is important to remember that p53 overexpression by immunohistochemistry may not necessarily represent mutated p53, and furthermore, mutated p53 may still retain some trancriptional activity .
Elsewhere p21 induction has been observed to be p53-independent in breast cancer (Rey et al, 1998), ovarian cancer (Elbendary et al, 1996;Anttila et al, 1999) and in human leukaemia . Our results suggested a p53-independent mechanism of p21 activation only in a subgroup of patients (Table 3). This kind of phenomenon (independent and dependent mechanisms) has also been observed by Yasui et al (1997) in their series of colorectal tumours. In addition, previous studies on endometrial (Backe et al, 1997;Ito et al, 1997), ovarian (Werness et al, 1997) and urothelial carcinoma (Clasen et al, 1998) have failed to show a correlation between p21 expression and p53. The significance of intratumoural p21 and p53 staining heterogeneity can be hypothesized as follows: a population of cancer cells with one functional copy of p53 might still be able to activate p21 within a tumour in which the other cells have already lost both functional copies. Accordingly, p21 is weakly expressed in cells with one functional allele of p53, and it is absent in cells with a mutation or loss of both alleles (Sasaki et al, 1996). From previous studies it is known that the immunoreactivity of p53 parallels the degree of mutation, and weak staining of p53 indicates the presence of both wild-type and mutant p53 proteins (Ohue et al, 1994). Based on the above cited reports it is reasonable to suggest that p21 activation by p53 is complex, tissue-specific and may vary in different phases of cell differentiation. Macleod et al (1995) showed that many adult tissues expressed high levels of p21, including thymus, brain and intestine as compared to liver, muscle and kidney. It is possible that other cyclinÐcdk inhibitors such as p27 are functioning in these tissues to induce growth arrest prior to differentiation. Alternatively these tissues may have reached a stable maintenance of growth arrest (Macleod et al, 1995). Our results showed no p53 protein expression in epithelial cells of normal colorectal mucosa, suggesting that p21 gene expression may occur by p53-independent pathways in differentiated cells. This finding supports the observation by Parker et al (1995) that increased levels of p21 may also be linked with terminal differentiation of specialized cells.
Induction of p21 is thought to result in growth inhibition of human cancer cells (Zeng et al, 1997). Different isoforms of transcription factor AP-2 have been shown to activate the expression of both positive and negative growth regulators, including c-erbB-2 and p21 WAF1/CIP1 . In addition, in vitro and in vivo studies have suggested that AP-2 may have the ability to restrict tumour growth (Bar-Eli, 1997;Karjalainen et al, 1998). The strongest clinical evidence for AP-2Õs tumour suppressive properties comes from the study by Karjalainen et al (1998). While studying malignant Significance level P < 0.05; β = coefficient of the regression model; s.e. = standard error of β; CI = confidence interval melanomas they were able to show that AP-2 expression progressively decreased as the tumour thickness increased (Karjalainen et al, 1998). Decreasing AP-2 expression was also a strong predictor for poor clinical outcome (Karjalainen et al, 1998). Further evidence for AP-2Õs importance in cancer cell regulation is derived from breast cancer studies where c-erbB-2 has been shown to be regulated by AP-2 (Bosher et al, 1995). In this case overexpression of c-erbB-2 is widely thought to contribute to poor clinical outcome (Bosher et al, 1995). However, in the present series of colorectal carcinoma, AP-2 staining did not reach prognostic significance. Taken together, the data available so far suggest that the effects of AP-2 are diverse and that different isoforms of AP-2 have different tissue specificity.
In the current study, p21 proved to be an antiproliferative factor; it inversely correlated with mitotic index. Similarly, an inverse correlation between p21 and Ki-67 proliferation antigen has been documented recently in colorectal and endometrial cancer Palazzo et al, 1997). On the other hand, several studies have not been able to demonstrate p21Õs antiproliferative action (Backe et al, 1997;Diab et al, 1997;Werness et al, 1997). Accordingly, in breast cancer and malignant melanoma increased cell proliferation has been positively related to p21 expression (Rey et al, 1998;Karjalainen et al, 1999). Possible explanations in these cases include transient p21 expression before entry into S phase, abnormal function of p21, or that tumour cells have become refractory to inhibitory p21 signals.
Newly recognized cell cycle regulators were tested for their relevance in predicting patientsÕ clinical outcome in the present study. Strong intensity and high percentage distribution of p21 in tumour epithelium predicted longer survival and RFS time. In the multivariate analysis, p21 expression intensity and percentage distribution were independent prognostic factors for overall survival and RFS. M/V index also proved to be a significant indicator of survival in T1Ð4/N0Ð3/M0 and T1Ð3/N0/M0 cases. Our results give further support for the previously documented role of p21 as a negative growth regulator in colorectal cancer Sasaki et al, 1996;Yasui et al, 1997). p21 may inhibit cancer spreading by suppressing abnormally accelerated cell cycle, and may therefore be a good candidate for a new therapeutic approach to control tumour cell growth. However, before being ready for clinical testing, further studies explaining p21Õs possible cell type specificity are needed.