Colorectal cancer is estimated to be the second leading cause of cancer death worldwide.1 Differences in etiology, clinical behavior, pathologic features and genetic alterations apply to right-sided vs. left-sided colorectal tumors, which indicate that different mechanisms drive tumorigenesis at these sites.2, 3 Approximately 35% of colorectal cancers are located in the rectum, and rectal cancer affects males at a rate of 1.5:1 compared to females, with 75% of cases diagnosed after 65 years of age. Treatment advances such as the standardized surgical technique TME, preoperative radiotherapy and adjuvant chemotherapy have reduced the previously high local recurrence rate and improved survival in rectal cancer patients. Despite these advances, about 40% of patients still die from disseminated disease.4, 5, 6 Hence, there is a need for novel markers that allow improved identification of high-risk patients who would benefit from adjuvant treatment.7
Colorectal cancer is one of the best-characterized tumor types at the molecular level. Numerous genetic alterations in oncogenes, tumor-suppressor genes and DNA MMR genes are acquired during tumorigenesis, which occurs through successive waves of clonal expansion according to 2 major mutational pathways, the CIN and MSI pathways.8, 9 Although the sequential accumulation of genetic changes that characterize the adenoma–carcinoma sequence of colorectal cancer applies also to rectal cancer, few studies have separately analyzed these tumor types. Rectal cancer has been associated with high proliferative activity, frequent aneuploidy and loss of heterozygosity at 17p and 18q, mutation/overexpression of p53 and consequently low frequencies of diploid tumors and MSI.2, 3, 10, 11, 12 Since the frequencies of the different genetic alterations vary according to tumor location within the bowel, studies investigating a possible prognostic role of biologic parameters should optimally analyze these tumor types separately.
Several tumor-associated proteins may be relevant prognostic markers in rectal cancer. However, no single molecular marker has been demonstrated to provide consistent and independent prognostic information, and despite advances in our understanding of the pathogenesis of rectal cancer, molecular markers have not gained utility in clinical decision making.7, 13, 14 The TMA technology allows high-throughput analysis of molecular markers and thereby facilitates studies of multiple markers in large tumor series.15 Our aim was to apply TMA to investigate the expression patterns of multiple markers and their prognostic correlations in a large series of rectal cancer patients. These markers, which included Ki-67, p53, Bcl-2, EGFR, β-catenin, E-cadherin and the MMR proteins MLH1 and MSH2, were selected either because they are altered in rectal cancer or because they have been suggested to represent prognostic factors in colorectal cancer.
Diagnostic tumor biopsies obtained before any treatment were utilized for the study. The 269 patients were diagnosed between 1994 and 2001 and included 173 men (64%) and 96 women (36%) at a mean age of 68 (range 32–92) years. Tumor location within the rectum was low (0–5 cm) in 97 patients (36%), middle (>5–10 cm) in 120 (45%) and high (>10–15 cm) in 52 (19%). Short-course, 25 Gy in 5 fractions, preoperative radiotherapy was given to 217 (81%) patients; prolonged preoperative radiotherapy, 50 Gy in 25 fractions, was given to 30 (11%) patients, 12 of whom also received concomitant chemotherapy; and 22 (8%) patients were not treated with radiotherapy because of small tumors or metastatic disease. Surgery was performed using TME. Histopathology and tumor stage were based on the original histopathologic reports and the patients' clinical records. All patients had a proven primary adenocarcinoma of the rectum. Tumor stage was Dukes' A in 63 patients (23%), Dukes' B in 88 (33%), Dukes' C in 79 (29%) and Dukes' D in 39 (15%). Data on tumor differentiation grade were available from 266 patients: 5 tumors (2%) were well differentiated, 182 (68%) were moderately differentiated and 79 (30%) were poorly differentiated or undifferentiated. Metastasis developed in 95/247 (38%) patients from whom data on metastatic disease were available. During follow-up (until September 2002), 105/269 (39%) patients died from all causes. Six patients died from other causes than tumor (e.g., postoperative complications, sepsis, myocardial infarction and a ruptured aortic aneurysm) within 2 months of surgery, and these patients were excluded from the survival analysis. Median follow-up was 2.9 years (range 0.2–7.8) for all patients and 3.5 years (range 1.7–7.8) for survivors. In the analyses of time to distant metastasis, the 39 patients with metastases at diagnosis (Dukes' D) were excluded. The Lund University Ethics Committee granted permission for the study.
Preparation of TMAs
Fresh 5 μm sections were obtained from paraffin-embedded tumor blocks and stained with hematoxylin–erythrosin. Viable, representative areas of tumor specimens were marked by one of the investigators (E.F.). Core needle biopsies were retrieved from the original tumor blocks using a manual arrayer (Beecher Instruments, Sun Prairie, WI)15 and positioned in a recipient paraffin array block. We aimed at obtaining at least 3 core biopsies from each biopsy specimen, and the number obtained was mainly determined from the size of the biopsy. Between 2 and 5 core biopsies were obtained: 2 biopsies from 2 cases (1%), 3 biopsies from 108 cases (40%), 4 biopsies from 153 cases (57%) and 5 biopsies from 6 cases (2%). Overall, 970 core biopsies were obtained from the 269 patients. Of these, 79 cylinder sections (8%) were lost, folded or severely damaged and 84 (9%) were nonrepresentative due to normal tissue or necrosis. All TMA sections were lost or otherwise nonevaluable in 9/269 (3%) cases. Depending on the amount of well-preserved tissue, 1 or 2 TMA sections were required for evaluation of the staining pattern in each tumor. A variable number of TMA sections could be evaluated in the remaining tumors, and the number of evaluable cases thus varied between 249 and 257 for the different markers. An example of TMA sections from the same tumor stained for different markers is given in Figure 1.
Fresh 5 μm sections from TMA blocks were transferred to glass slides (ChemMate Capillary Gap Microscope Slides, 75 mm; Dako, Glostrup, Denmark). Sections were dewaxed and rehydrated. To achieve antigen retrieval, sections were pretreated in 10 mM citrate buffer (pH 6.0) or 1 mM EDTA buffer (pH 9.0) in a microwave (750 W) for 15 min or in 0.4% pepsin in 0.01 M HCl at 37°C for 60 min (Table I). An automated immunostainer (TechMate 500Plus, Dako) was used for the staining procedure, with Dako ChemMate Kit peroxidase/3-3′-diaminobenzidine. All primary antibodies used were commercially available mouse monoclonal IgGs, and detailed data on the antibodies and staining evaluations are given in Table I. After counterstaining with hematoxylin, slides were dehydrated in ascending concentrations of ethanol and mounted.
Table I. Data On IHC Markers and Evaluation Criteria
San Diego, CA.
Percentage of nuclei stained: <30% (low), 30–60% (medium), >60% (high)
Cut-off values and scoring for tumor immunostaining are summarized in Table I, and representative aberrant/normal patterns are depicted in Figure 2. In summary, MIB-1 staining for the proliferation marker Ki-67 was scored according to the number of positively stained tumor cell nuclei, with <30% stained nuclei representing low proliferation, 30–60% medium high and >60% high. p53 staining was considered positive if >5% of nuclei were stained, irrespective of staining intensity. Cytoplasmic staining for Bcl-2 and membranous and cytoplasmic staining for EGFR were evaluated as positive or negative. Staining for β-catenin was independently evaluated in the cell membrane and the nucleus (as strong, weak or absent) and in the cytoplasm (as present or absent). E-cadherin staining was evaluated in the membrane (as strong, weak or absent) and in the cytoplasm (as absent or present). Nuclear staining for MLH1 and MSH2 was evaluated as present or absent, but to classify the staining as absent, retained staining in the tumor stroma was required. Slides were independently evaluated by 2 of the investigators (E.F. and M.N.), and a third investigator (M.D.) reviewed 14% of the samples. Discrepancies in the evaluation between the first 2 reviewers and the pathologist (M.D.) were present in 3–6% of TMA sections for the markers Ki-67, p53, MLH1, MSH2 and EGFR and in 10–15% of sections for Bcl-2, β-catenin and E-cadherin. A consensus agreement was reached for each case, and for the markers with high (10–15%) disagreement, all TMA sections were reevaluated. Expression patterns for each marker were correlated with development of metastasis. Markers that did show correlations with metastasis were further analyzed with respect to metastasis-free survival.
Associations in 2 × 2 tables were evaluated with Fisher's exact test, whereas a test for 0 slope in a linear regression model was used if one or both factors had 3 or more ordered categories. The latter test is also known as the χ2 test for trend. Linear scores were used for ordered variables. Time from surgery to metastasis or death was analyzed using Kaplan-Meier estimates, log rank tests and Cox's proportional hazards regression analysis. Proportional hazards assumptions were checked graphically, and no gross violations were observed. All tests were 2-sided, and the significance level was set at 5%. The software package Stata 7.0 (StataCorp, College Station, TX) was used for statistical analyses.
Representative expression patterns for the different markers are shown in Figure 2, and the data are summarized in Table II.
Table II. Staining Patterns For the Different Markers and Their Relation to Metastasis
Fisher's exact test was used for comparison of 2 proportions, whereas a test for linear trend was used when staining was categorized into 3 groups.
A high Ki-67 proliferative index was detected in the rectal cancers studied; 23% of tumors had <30% nuclei staining, 41% had a nuclear Ki-67 expression in 30–60% of tumor cells and 36% had >60% of nuclei with positive Ki-67 staining. Hence, 77% of the tumors showed >30% staining for Ki-67.
Overexpression of p53 was observed in 205/253 (81%) tumors. In 179/205 (87%) of the p53+ tumors, p53 staining was widespread and intense, whereas the remaining 26/205 (13%) tumors showed heterogenous staining, although >5% of the tumor nuclei were stained.
The anti-apoptotic Bcl-2 protein showed cytoplasmic overexpression in 165/252 (65%) of tumors, and in 43/252 (17%) tumors, strong staining for Bcl-2 was found. A combined analysis of p53 and Bcl-2 expression revealed 68 tumors with a p53+/Bcl-2– phenotype and 30 tumors that were p53–/Bcl-2+.
Staining for EGFR was observed in 133/255 (52%) tumors, and in 78/255 (31%) tumors, strong staining was found.
Cell adhesion proteins β-catenin and E-cadherin.
Aberrant staining for β-catenin was observed in the majority of tumors, with reduced or absent membranous staining in 117/252 (46%), increased cytoplasmic staining in 238/252 (94%) and increased nuclear staining in 146/257 (57%). The cell adhesion molecule E-cadherin showed reduced membranous staining in 175/254 (69%) tumors, and 43/175 (25%) of these tumors showed complete lack of membranous staining, whereas cytoplasmic staining for E-cadherin was found in 220/252 (87%) tumors.
MMR proteins MLH1 and MSH2.
Loss of staining for the MMR proteins MLH1 and MSH2 was found in a low fraction of tumors: loss of MLH1 was found in 8/253 (3%) tumors and loss of MSH2 in 2/252 (1%) tumors. Notably, 8 of these 10 tumors were early-stage (4 were Dukes' stage A and 4 were Dukes' B).
Associations between dichotomized staining patterns
Significant associations were found between the expression patterns for β-catenin, E-cadherin and MLH1. All associations were evaluated using Fisher's exact test. For β-catenin, there was an inverse association between nuclear and membranous staining (p = 0.001); 63% of tumors with retained membranous staining showed nuclear accumulation of β-catenin. A positive correlation between nuclear and cytoplasmic staining for β-catenin was found (p = 0.001). Indeed, 99% of the tumors that showed nuclear accumulation of β-catenin also showed cytoplasmic staining. There was also a positive association between cytoplasmic staining for β-catenin and E-cadherin (p < 0.01). The staining patterns for E-cadherin showed a positive association between cytoplasmic and membranous staining (p = 0.01). Finally, a positive association was found between loss of MLH1 staining and lack of nuclear staining for β-catenin (p < 0.001).
Distant metastases developed in 95/247 (38%) patients, and the development of metastatic disease was correlated to the different expression patterns for each marker (Table II). Some staining patterns were significantly correlated with metastatic disease, and these included lack of cytoplasmic staining for β-catenin (p = 0.04), reduced membranous staining for β-catenin (p = 0.04), reduced membranous staining for E-cadherin (p = 0.02) and absence of cytoplasmic staining for E-cadherin (p = 0.02) (Table II). None of the markers Ki-67, p53, Bcl-2 and EGFR showed any significant association with metastatic disease. When data for p53 and Bcl-2 expression were combined, tumors with a p53+/Bcl-2– phenotype showed the highest rate of metastasis (27/64, 42%), patients with p53–/Bcl-2+ tumors showed the lowest fraction of metastasis (7/29, 24%) and the remaining tumors showed an intermediate fraction of metastasis (54/134, 40%). However, the groups were not significantly different regarding development of distant metastasis (p = 0.16). Expression patterns for the MMR proteins MLH1 and MSH2 were not correlated with metastasis or survival since only a few tumors showed aberrant expression patterns.
Univariate analysis using distant metastasis as the clinical end point revealed significantly increased HRs for Dukes' stage, tumor differentiation grade and reduced membranous staining for β-catenin (HR = 1.8, p = 0.04) (Fig. 3a) and E-cadherin (HR = 2.2, p = 0.04) (Fig. 3b, Table III). Sex, age and tumor location within the rectum did not correlate with development of metastases. A lower frequency of distant metastases was found in tumors with cytoplasmic staining for β-catenin (HR = 0.32, p = 0.02) and in tumors with cytoplasmic staining for E-cadherin (HR = 0.47, p = 0.03).
Table III. Univariate and Multivariate Analyses with Metastasis-Free Survival As End Point (n > 208)
Multivariate analysis (adjusted for stage and differentiation grade)
B vs. A
C vs. A
Low vs. high
Male vs. female
High vs. low
Positive vs. negative
Positive vs. negative
Positive vs. negative
Reduced/absent vs. retained
Present vs. absent
Present vs. absent
Reduced/absent vs. retained
Present vs. absent
Multivariate analysis including the factors Dukes' stage and tumor differentiation grade showed that cytoplasmic staining for β-catenin was associated with a decreased risk of metastasis (HR = 0.32, p = 0.02) and revealed a higher risk of metastasis in tumors with reduced or lost membranous expression of β-catenin and E-cadherin. Tumors with reduced/absent membranous expression of β-catenin had HR = 1.7 (p = 0.06) (Table III). Tumors with reduced/absent membranous staining for E-cadherin had HR = 2.1 (p = 0.06) (Table III). When overall survival was used as the clinical end point in the multivariate analysis, these factors did not reveal any significant independent prognostic information (data not shown). Indeed, tumors with normal, retained membranous staining for both β-catenin and E-cadherin (n = 49) had the lowest risk of metastasis, 21% compared to 43% among tumors with reduced/absent staining for at least one of these markers (p = 0.009), which corresponds to HR = 0.3 (95% CI 0.12–0.93, p = 0.04).
Characterization of a molecular phenotype in primary tumors that signifies aggressive tumor behaviour would be clinically valuable in the identification of patients who would benefit from adjuvant treatment and constitutes a basis for the development of targeted therapies. Colorectal cancer is one of the best-characterized tumor types at the molecular level. Consequently, many studies have investigated these different genetic aberrations in relation to prognosis, but although several molecular markers have in a few studies shown prognostic correlations, no single molecular marker has provided consistent and independent prognostic information.12, 13, 14 Our aim was to characterize the IHC patterns in a large series of rectal cancers by applying TMA analysis and to assess the prognostic importance of the IHC expression profiles. TMA has successfully been applied for immunostaining of several of the markers included in our study, such as p53, MLH1, MSH2 and β-catenin.12, 16, 17, 18 When applying TMA, nuclear staining patterns and immunostaining evaluated based on the presence or absence of staining can generally be determined from one TMA section containing representative tumor tissue of good quality. In contrast, markers like EGFR, Bcl-2, β-catenin and E-cadherin, which produce cytoplasmic or heterogeneous staining patterns or are evaluated in 3 or more categories, will result in a higher number of nonassessable tumors due to discordant readings and may thus require 2 or more TMA sections to obtain an acceptable level of reproducibility.12, 19 Loss of TMA sections during the array process is another reason for obtaining 3–4 replicate biopsies from each tumor. Most TMA series report loss of about 10–15% of the sections, due to either empty spots on the slide or sections of poor technical quality. If 3 core biopsies are obtained from each tumor, 98% of the tumors are estimated to be successfully analyzed using immunostaining.12 The recommendation of using triplicate TMA sections for analysis is in accordance with the results of other investigators validating the TMA technique and will minimize loss of data, ensure concordant readings in most cases and thereby provide a reliable IHC expression profile. TMA thereby increases the number of markers that can be investigated within the same tumor set and contributes to tissue preservation. We analyzed 2–5 core biopsy sections from each tumor. In total, 17% of TMA sections were nonevaluable, but due to redundancy only 3% of the tumors were nonevaluable.
Several studies have demonstrated a high fraction of Ki-67 staining cells in rectal cancer, but the prognostic value of Ki-67 is uncertain; some investigators have suggested that a high fraction of Ki-67-expressing cells is associated with a better outcome, whereas other studies, including the present one, have not found such an association.12, 20, 21 We also identified a high Ki-67 index in the rectal cancers studied, with 77% of the tumors showing >30% positive nuclei. However, no correlation to the development of distant metastasis or survival was found.
Rectal cancers show a high degree, 60–80%, of immunostaining for p53, and TP53 mutations are found in 40–70% of the tumors.2, 12, 21, 22, 23, 24 In the present study, 81% of the tumors showed immunostaining for p53. Correlations between p53 status and prognosis have been studied by several investigators, with partly contradictory results. In summary, most studies have not supported any prognostic impact using immunostaining for p53, whereas a relationship may exist between mutations in TP53 and prognosis.21, 22, 23, 24 Furthermore, a possible prognostic importance of TP53 mutations appears to apply specifically to tumors in the proximal colon.22 An association between TP53 mutation and/or positive immunostaining for p53 in the tumor tissue and poor response to radiotherapy and/or poor prognosis has been found in some studies of rectal cancer.2, 25, 26 However, several large studies, including the present one, have failed to demonstrate that p53 confers any consistent and significant prognostic impact, which argues against a prognostic role for p53 immunostaining in rectal cancer.12, 27, 28
p53 inhibits expression of the antiapoptotic Bcl-2 protein and mediates apoptosis via the Bcl-2/BAX pathway. A high degree of Bcl-2 expression has been demonstrated in distal colorectal cancers, and Bcl-2 has been implicated as a prognostic factor in colorectal cancer, with Bcl-2 expression signifying a favorable prognosis; especially poor prognosis has been observed for the subset of tumors with the molecular phenotype p53+/Bcl-2–.26, 27, 28, 29 In our study, tumors with the phenotype p53+/Bcl-2– had a worse outcome, but the difference was not statistically significant.
EGFR activation elicits a cascade of responses involved in cell division, proliferation, differentiation, apoptosis and angiogenesis. Expression of EGFR is of prognostic significance in several types of solid tumors, but only a few studies have assessed its prognostic role in colorectal cancer; and, although EGFR expression is found in 50–70% of colorectal cancers, its prognostic importance is unclear.30, 31, 32 We identified EGFR expression in 52% of the tumors, but this finding did not influence the risk of metastasis. EGFR is a target for intense development of targeted therapies, and several agents, including chimeric anti-EGFR MAbs, that target the EGFR are currently under evaluation.33
The β-catenin pathway plays a central role in transcriptional signaling, cell–cell interactions, maintenance of tissue architecture and differentiation in colonic epithelium.34 β-Catenin expression is largely regulated by its 2 binding partners, the membranous protein E-cadherin and the cytoplasmic APC protein. Mutations of APC or of the regulatory domain of CTNNB1 (encoding for β-catenin) lead to stabilization and cytoplasmic accumulation of free β-catenin, which translocates to the nucleus, where it promotes transcriptional activation through interaction with the TCF/LEF transcription factor family. Immunostaining for β-catenin has shown increased cytoplasmic staining in 85% of colorectal cancers and increased nuclear staining in 20–26% of tumors.35, 36, 37 Our study yielded similar results, with cytoplasmic accumulation of β-catenin in 87% and strong nuclear staining in 25% of tumors. Different conclusions have been reached regarding the prognostic importance of increased cytoplasmic or nuclear IHC staining for β-catenin, and in summary, the role of β-catenin as a prognostic marker is not fully known.18, 35–38 Our study showed a correlation between loss of membranous staining for β-catenin and development of distant metastasis (p = 0.02). Reduced/absent membranous expression for β-catenin was an independent prognostic marker in the multivariate analysis (HR = 1.7, p = 0.06). Furthermore, tumors with cytoplasmic accumulation of β-catenin showed a lower risk of metastasis; however, the number of tumors without such staining was small, and these results should therefore be interpreted with caution. Although β-catenin alterations occur early in colorectal tumorigenesis, increased expression of β-catenin has been demonstrated in metastasis compared to primary tumors. Hence, β-catenin represents a marker of possible prognostic importance in colorectal cancer, but data are still scarce and contradictory, which necessitates further studies in independent material.
E-cadherin functions as a cell adhesion molecule but also plays an important role in signal transduction and suppression of invasion. Germline mutations in the CDH1 gene, which encodes for E-cadherin, predispose to hereditary diffuse-type gastric cancer, and somatic mutations appear particularly in gastric cancer and breast cancer. However, loss of immunostaining for E-cadherin frequently occurs in carcinomas without such mutations, and epigenetic mechanisms have been suggested to cause the reduced expression.39 IHC changes in E-cadherin expression and localization occur in most invasive tumors, and low expression of E-cadherin has been associated with a higher grade and stage as well as infiltrative growth patterns in, e.g., breast cancer and gastric cancer.38, 39 In colorectal cancer, reduced expression of E-cadherin has been associated with poorly differentiated and mucinous tumors, and E-cadherin has been suggested to play a role in tumor progression. We did not, however, find any significant correlation between loss of membranous staining for E-cadherin and tumor differentiation grade (data not shown). E-cadherin has also shown significant independent prognostic value in several tumor types, including colorectal cancer.40, 41, 42 Clinically, restoration of a functioning catenin–cadherin protein complex is a presumed mechanism for several drugs that promote differentiation and inhibit invasiveness. We found a correlation between reduced membranous E-cadherin expression and development of distant metastasis (p = 0.02, Table II) and with time to metastasis (HR = 2.2, p = 0.04), with the worst outcome for tumors lacking E-cadherin (Fig. 3). Also, cytoplasmic expression of E-cadherin correlated with metastases, the highest risk being for patients without cytoplasmic E-cadherin expression (p = 0.02) in the univariate analysis; but this did not represent an independent marker in the multivariate analysis (Table III). Although β-catenin acts as a transcription factor and closely interacts with E-cadherin, inactivation of the latter protein and the resultant disruption of cellular adhesion do not significantly increase the levels of free β-catenin or transcription of TCF/LEF target genes.43 Interestingly, several cell surface molecules have been correlated with prognosis in colorectal cancer. These include integrins, ICAM-1, the proteoglycan receptor CD44 and VEGF.
The DNA MMR system consists of a complex of proteins that recognizes and directs repair of nucleotide base mismatches and slippage mistakes at simple repetitive sequences termed “microsatellites”. Defective MMR is caused by germline mutations in tumors associated with hereditary nonpolyposis colon cancer and by hypermethylation of the MLH1 promoter in sporadic tumors. Either mechanism leads to IHC loss of expression of the affected MMR protein. The distribution of MSI tumors varies within the colon, with about 30% of MSI tumors in the proximal colon compared to about 5% in the distal colon.2, 3, 11 Our findings of loss of MLH1 in 3% and loss of MSH2 in <1% of tumors confirm these observations and suggest that other mechanisms than defective MMR cause the vast majority of rectal cancers.
In summary, our study fails to demonstrate any significant association between immunostaining for the markers Ki-67, p53, Bcl-2 and EGFR and development of distant metastases and demonstrates only low frequencies of aberrant MMR with loss of staining for MLH1 and MSH2. Considering the complex genetic alterations that characterize rectal cancer with a broad range of abnormalities already in early-stage tumors, single molecular markers may be hard to apply for prognostication in rectal cancer. Consequently, our findings may indicate that techniques that allow extended molecular profiling, such as array-based methodologies, should be applied to identify novel prognostic markers or combinations thereof. Different proteins involved in cell adhesion have been suggested to be of importance in tumor progression. We found that reduced membranous staining for β-catenin and E-cadherin correlated with development of metastatic disease, which suggests a role for cell adhesion defects as prognostic markers in rectal cancer. Data on cell adhesion molecules and their relation to prognosis in colorectal cancer are scarce, and although our findings confirm some previous observations, further investigations in independent tumor material are needed to establish the prognostic role of β-catenin and E-cadherin in rectal cancer.