Proliferation, Apoptosis, and Survival in High-Level Microsatellite Instability Sporadic Colorectal Cancer¹

On the basis of the level of DNA MSI,³ three distinct subgroups of CRC have been identified (1, 2, 3). The majority of sporadic CRCs (70–80%) show allelic loss (LOH) with no MSI, and these are defined as MSS. The other two CRC subgroups defined as MSI-L and MSI-H, each comprising 10–15% of sporadic CRC, develop small insertion and deletion mutations at an accelerated rate in repetitive DNA sequences. Genotypically, MSI-L cancers are distinguished by low-level DNA instability at <40% of microsatellite loci, chromosomal instability (allelic loss), and mutation of the tumor suppressor genes APC and TP53 and the oncogene K-ras(3). In contrast, the MSI-H subgroup characterized by MSI at >40% of microsatellite loci rarely shows allelic loss (3, 4, 5, 6). However, because of defective DNA mismatch repair caused by hypermethylation-induced “silencing” of the promoter region of one of the mismatch repair genes, hMLH1(7), sporadic MSI-H cancers often acquire frameshift mutations in the repeat regions of genes implicated in tumor progression, including TGF-βRII(8), BAX(6, 9), and insulin-like growth factor type II receptor (10). Phenotypically, MSI-H cancers are predominantly right-sided (located in the proximal colon) and more likely to be poorly differentiated, mucinous, and larger at presentation (2, 4, 11). In addition, we have shown recently that >70% of sporadic MSI-H cancers are characterized by the presence of TILs, and these TILs are predominantly CD8⁺ T cells (12).

MSI-H in sporadic CRC is associated with a less aggressive phenotype and improved survival when compared with stage-matched MSS and MSI-L cancers (13, 14, 15, 16). Proliferative activity also has prognostic significance in various types of solid tumors including CRC (17, 18, 19, 20, 21, 22, 23, 24, 25, 26). In one study, a higher PI correlated with the presence of lymph node metastases in CRC (25). To date, the reason for the association between MSI-H and favorable clinical outcome remains unclear, and it is not known whether lower proliferative activity explains the MSI-H survival advantage, because previous studies assessing proliferative activity in CRC did not subdivide tumors according to the level of MSI. In the current study, we measured the PI by immunohistochemical examination with an anti-Ki-67 antibody in sporadic CRC classified according to MSI status. The Ki-67 antibody detects a nuclear matrix-associated antigen that is present only in proliferating cells (27), and Ki-67 immunoreactivity correlates with tumor proliferative activity (27, 28).

CRC growth is regulated by the balance between proliferation and apoptosis (29, 30). We have shown recently that MSI-H cancers have an increased AI within the tumor cell population compared with MSS cancers, whereas MSI-L cancers are intermediate (12). The aim of this study was to evaluate PI and AI:PI ratio in the same series of 100 sporadic CRCs of known MSI status to determine the relationship between proliferative and apoptotic activities and to clarify the prognostic impact of these biological properties on patients with MSI-H CRC.

This study was carried out on a selected series of 100 sporadic colorectal adenocarcinomas surgically resected from 99 patients at the Royal Brisbane Hospital between 1989 and 1999. Of these, 91 tumors were derived from a cohort of 303 cancers characterized previously for clinicopathological and molecular features including microsatellite instability (2), and 9 MSI-H tumors were added to increase the statistical power of comparison within this group. Microsatellite instability was originally defined by PCR at six microsatellite loci including MYCL, AT3, D2S123, F13B, BAT-26, and BAT-40(2). Analysis at additional loci ACTC, BAT-25, BAT-34C4, D5S346, D10S197, D17S250, and D18S55 from the National Cancer Institute panel (31) has not altered the classification of any of these tumors. Altogether, 31 cancers were MSI-H (with bandshifts in at least three of six markers), 29 were MSI-L (with bandshifts in one or two microsatellite markers), and 40 were microsatellite stable with no bandshifts.

The mean postoperative follow-up in this study was 42 months (range, 5–60 months). The vital status of each patient was determined at the date of last follow-up or at the end of a 5-year follow-up period, and if deceased, the cause of death was ascertained from the medical record and/or death certificate. The cause of death was classified as secondary to, or unrelated to, colon carcinoma. Only CRC deaths were considered as outcome events; all other end points were considered as censored.

Immunohistochemistry was performed on archival tissue obtained from the Department of Pathology, Royal Brisbane Hospital. Paraffin sections (4 μm) were pretreated with DAKO high pH antigen retrieval system (DAKO, Carpinteria, CA) using a domestic 600 kW microwave oven. Endogenous peroxidase activity was quenched in 1% H₂O₂, 0.1% sodium azide in TBS. Nonspecific antibody binding was blocked by incubating sections in 4% commercial skim milk powder followed by 10% nonimmune goat serum (Zymed Corp., San Francisco, CA). Primary antibody (DAKO polyclonal anti-Ki-67) was applied at a 1:75 dilution overnight at room temperature. Sections were then incubated with prediluted goat antirabbit secondary antibody (Zymed Corp.) for 30 min. Ki-67 localization was demonstrated with prediluted streptavidin-horseradish peroxidase (Zymed) and 0.05% 3,3′-diaminobenzidine in TBS, with H₂O₂ as the substrate. All sections were lightly counterstained with hematoxylin. A negative control, omitting primary antibody, was included in all staining runs.

Apoptosis staining using the novel M30 CytoDEATH antibody was described in detail previously (12, 32). Briefly, dewaxed sections were pretreated by microwave heating in 0.01 m citrate buffer (pH 6.0), and M30 CytoDEATH monoclonal antibody (Boehringer Mannheim, Mannheim, Germany) was applied overnight at a 1:200 dilution. Prediluted goat antimouse secondary antibody (Zymed) followed by streptavidin-horseradish peroxidase (Zymed) was used, and the product was developed in 0.05% 3,3′-diaminobenzidine in the presence of hydrogen peroxide as described above for Ki-67 staining.

We confirmed previously that all 31 MSI-H sporadic CRCs in this series showed loss of hMLH1 expression by immunohistochemistry (34).⁴

Ki-67 scoring was performed by two independent observers (J. M. M-R., L. E. R.) with differences resolved using a conference microscope. The PI was defined as the percentage of tumor nuclei showing Ki-67 staining per total of 1000 neoplastic cells counted in five fields of 200 tumor cells. Consistent with previous reports (22, 25), considerable intratumor heterogeneity was observed in the distribution of Ki-67-stained neoplastic cells in each lesion. Ki-67-positive tumor cells were therefore scored in areas with the highest proliferative activity. Tumor cells were considered positive for Ki-67 whenever diffuse or punctate brown nuclear staining could be identified.

M30 CytoDEATH scoring was also performed by two independent observers (J. M. M-R., A-E. B-H.) as described previously (12). The AI was defined as the mean percentage of cells expressing cytoplasmic M30 staining per 2000 neoplastic epithelial cells counted in 10 randomly selected fields of 200 tumor cells.

DNA extraction from fresh tumor samples was described previously (2). BAX and TP53 mutations were detected previously in this cohort using SSCP analysis (6). Determination of TP53 LOH on chromosome 17p was described and analyzed previously (3). Mutations in repeat sequences of the TGF-βRII of MSI-H cancers were identified as originally described by Markowitz et al.(8). Detection of point mutations in codons 12 and 13 of the K-ras proto-oncogene using nonradioactive PCR-RFLP was also described and analyzed previously in this laboratory (33).

Continuous variables are summarized in terms of mean ± SE. Nonparametric methods were used for bivariate analyses, because the primary variables of interest (Ki-67 and apoptosis:Ki-67 ratio) were not normally distributed (Fig. 2). In general this type of analysis is more conservative than equivalent parametric approaches. Mean and median PI and AI:PI were compared across MSI levels using a Kruskal-Wallis nonparametric ANOVA. Multiple linear regression was used to compare PI and AI:PI across MSI levels after adjusting for potential confounding by other variables such as age, tumor stage, and grade. Spearman’s nonparametric correlation coefficient was used to measure the degree of association between PI and tumor size.

The effect of PI and AI:PI on patient survival was assessed in a number of ways. The subjects were split based on the median Ki-67 score, and a Kaplan-Meier survival analysis was conducted to compare the patient survival in each group. A similar analysis was also conducted for the AI:PI ratio. The effects of continuous PI and AI:PI scores on patient survival were assessed using Cox’s Proportional Hazard Modeling, which also allowed adjustment for confounding by age, stage, and MSI status. The results of the survival analysis are presented as hazard ratios along with 95% confidence intervals.

Clinicopathological data from the selected series of 100 primary CRC are shown in Table 1. Patients’ ages ranged from 27 to 87 years with a mean of 69.7 ± 10.3 years. Patients with MSI-H cancer were older (72.8 ± 5.9 years; range, 59–87 years) than patients with MSI-L or MSS cancer (68.2 ± 11.5 years; range, 27–87 years). Patients with any family history suggestive of hereditary nonpolyposis colon cancer syndrome were excluded from this study.

Ki-67 nuclear protein expression was quantified by immunohistochemistry (Fig. 1) and analyzed with respect to patient age and sex, tumor MSI status, tumor site, presence of TILs, clinicopathological stage, level of differentiation, tumor type, allelic loss at TP53, proliferation, point mutation in codon 12 or 13 of K-ras, and frameshift mutations in the TGF-βRII receptor or BAX genes (Table 1). The Ki-67 index was significantly associated with MSI status, with MSI-H tumors exhibiting a higher mean Ki-67 labeling index (90.1 ± 1.2%) than the MSI-L (69.5 ± 3.1%) or MSS (69.5 ± 2.2%) subgroup, which had similar Ki-67 indices (P < 0.0001; Table 1 and Fig. 1). In addition, a high Ki-67 index was significantly associated with proximal (right-sided) location, TIL positivity, poor differentiation, mucinous histology, and absence of TP53 LOH (P < 0.05), all of which are characteristic features of MSI-H cancers. When multiple linear regression was performed adjusting for all variables (age, sex, MSI, side, stage, TIL status, tumor grade, type, and TP53 status), only MSI status remained significant (Table 1), indicating that MSI-H is the only variable independently associated with a high proliferative activity in CRC. There was a trend toward lower proliferative activity in Dukes’ stage D (metastatic) tumors (Fig. 1,F), and this negative association between Ki-67 expression and tumor stage approached significance when corrected for all other variables in the multivariant analysis (Table 1). The PI, as determined by Ki-67 staining, was not related to patient age or sex, or K-ras status, in any MSI subgroup. Although both high PI and frameshift mutations in TGF-βRII and BAX are associated with MSI-H cancers, the PI was not associated with TGF-βRII or BAX status within the MSI-H subgroup (Table 1). There was a trend toward increased Ki-67 staining with larger tumor size, although this did not reach significance (Spearman correlation coefficient for tumor size versus Ki-67 index was 0.136; P = 0.18). Furthermore, the extent of Ki-67 staining was not related to the site of the tumor per se. In all three MSI subgroups, left-sided tumors (located distal to the splenic flexure) had similar Ki-67 scores to right-sided (proximal) tumors (Table 2).

When assessed by immunohistochemistry, tumors exhibited a wide range of Ki-67 expression from 31.0 to 99.0% (Fig. 2,A), indicative of a large variation in proliferative activity. Overall, the median and mean (± SE) proliferative indices were 78.6% and 69.7% ± 10.3% respectively. Apoptosis in the same set of tumors was assessed previously by immunohistochemistry in this laboratory (12), using the M30 CytoDEATH antibody that binds to the caspase-cleaved fragment of cytokeratin-18 during the early stages of epithelial cell apoptosis. Median and mean apoptotic indices were 1.9% and 2.5% ± 2.0%, respectively, and ranged from 0.2 to 8.2% (Fig. 2 B). The PI and AI were strongly correlated with a Spearman rank coefficient of 0.51 (P < 0.001).

The AI:PI ratio was also examined in relation to the same clinicopathological variables tested against PI in Table 1. In both univariate and multivariate regression analysis, MSI status was significantly correlated with the AI:PI ratio (Table 3). The AI:PI ratio was not associated with any other variable tested including patient age, sex, tumor location, TIL status, stage, tumor grade and type, TP53 LOH, K-ras, TGF-βRII or BAX status (Table 3 and data not shown).

To determine whether the PI is related to prognosis in any MSI subgroup, we compared patient survival and tumor PI as determined by Ki-67 staining. A survival advantage, which approached significance (P = 0.09), was associated with higher Ki-67 expression among MSI-H cancers but not MSI-L or MSS cancers (Table 4). Using the median Ki-67 score of 78.6% as a cut point, neither MSI-L nor MSS tumors showed a trend toward reduced survival in cancers with a higher Ki-67 index (data not shown). The AI as assessed by staining with the M30 CytoDEATH antibody was not associated with survival either overall (P = 0.41) or within MSI subgroups (data not shown).

Because the AI:PI ratio gives an indication of net cell production, we postulated that the AI:PI might be more informative than either index alone. Overall, a low AI:PI ratio of <0.26 (using the median value as the cutoff) was associated with a more favorable prognosis, compared with a high AI:PI ratio of >0.26 (Fig. 3), and this difference approached significance (P = 0.087). This survival difference did not appear to be particular to any one of the three MSI subgroups.

Immunostaining with the Ki-67 antibody is widely accepted as a reliable and reproducible method for evaluating proliferative activity in human tumors. Although the PI estimated by Ki-67 staining has been shown to produce a consistent overestimate of the growth fraction of tumors and is therefore not an exact measurement of tumor growth (35), the Ki-67 index correlates well with the predicted growth fraction (27, 35, 36). In another independent study, the Ki-67 index was significantly associated with the mitotic activity in tumor tissue (21), and there is good evidence that Ki-67 immunoreactivity correlates well with other markers of cell proliferation (37, 38, 39).

In this study, we found that MSI-H cancers exhibit a significantly higher Ki-67 index (90.1 ± 1.2%; P < 0.0001) than MSI-L or MSS cancers that had similar mean Ki-67 indices (69.5 ± 3.1% and 69.5 ± 2.2%, respectively). To our knowledge, this is the first study to show that the PI is significantly correlated with MSI status in sporadic CRC. Cycling cells are more likely to be undifferentiated, and MSI-H cancers are more likely to show poor differentiation. This could explain the higher proportion of cycling cells in the MSI-H subset. Consistent with our findings, Aoki et al.(25) reported recently that mucinous or poorly differentiated CRCs were characterized by an increased PI, both these features being characteristic of MSI-H cancers (2). Similarly, in the present study a high Ki-67 index was significantly associated with mucinous lakes, poor differentiation, proximal location, the presence of TILs, and wild-type TP53, all characteristics of MSI-H cancers. None of these associations remained significant after multiple linear regression analysis, indicating that MSI status was the primary predictor of PI in this series of cancers.

Consistent with previous reports, no correlation was found between the level of Ki-67 staining and age or sex (36, 37, 40, 41). Interestingly, there was a negative association between Ki-67 expression and stage that approached significance when corrected for all other significant variables in the multivariate analysis (P = 0.054). This trend suggesting that a lower PI is associated with Dukes’ stage D colorectal tumors is in contrast to previous findings by Kyzer and Gordon (42), who found a positive correlation between the Ki-67 proliferative index and stage D tumors; however, only 4 of 30 patients had metastatic disease. Consistent with our findings, the Kyzer study found no correlation between Ki-67 score and Dukes’ B or C cancers. Conversely, an earlier study by Suzuki et al.(24) that included 58 colorectal carcinomas found that the mean Ki-67 labeling index was higher in Dukes’ B and Dukes’ C tumors than in Dukes’ A tumors. Similarly, Aoki et al.(25) reported recently a significant association between lymph node metastases (Dukes’ stage C) and a higher Ki-67 index. In apparent contradiction, five additional studies found no correlation between the Ki-67 index and tumor stage (36, 37, 40, 42, 43). One reason for these discrepancies could be differences in tumor sampling. The sample size in these studies ranged from 30 to 255 tumors. Our observations are consistent with others that noted marked heterogeneity of Ki-67 expression within different tumor stages (22). Taken together, these results suggest that the Ki-67 index is not a reliable indicator of tumor stage.

In agreement with previous reports, our data suggest that Ki-67 staining is not dependent upon tumor site (22, 36, 42, 44). Therefore, the higher PI associated with MSI-H cancers was not related to the predilection for proximal location. In fact, four left-sided MSI-H tumors in this series had a slightly higher mean PI than the 27 right-sided tumors. Left- and right-sided MSS and MSI-L tumors also had similar Ki-67 scores. Therefore, the increased proliferative activity in MSI-H cancers is probably explained by the high level of MSI, although the mechanism is unclear. Similarly, we found that the level of apoptosis did not differ according to tumor site in the same series (12). Indeed, MSI-H cancers are characterized by both increased proliferation and apoptosis, indicative of higher cell turnover, compared with their MSI-L and MSS sporadic counterparts.

There was a significant correlation (Spearman rank coefficient of 0.51; P < 0.001) between PI and AI, suggesting that some degree of coordinated regulation remains. This finding is in agreement with another study by Baretton et al.(45) that found an elevated Ki-67 index was associated with a high AI. In contrast, two other recent reports found no significant correlation between tumor PIs and AIs (44, 46). However, this discrepancy may be attributable to methodology differences. One of these studies assessed proliferation by flow cytometry, which included only S-phase and G₂-M phases of the cell cycle in the proliferating fraction (44), whereas the Ki-67 antibody binds to nuclei in all phases of the cell cycle except G₀ and early G₁(27). In addition, both studies assessed the level of apoptosis by morphological analysis on H&E-stained sections. We have shown that immunohistochemical assessment with the M30 antibody is more sensitive than morphological measurement (12). We also observed a wide range of Ki-67 expression (31.0–99.0%) and M30 CytoDEATH staining (0.2–8.2%), reflecting marked heterogeneity in both PIs and AIs, respectively. Tumor stage is the most important prognostic variable in CRC (47), but stage-independent variability in clinical outcome is still observed, and this may be a consequence of altered rates of apoptosis and cell proliferation (44). Our findings reinforce the need to stratify according to MSI in studies of CRC.

Another report published recently by Bocker Edmonston et al.(48) included a comparison of Ki-67 staining and microsatellite status in CRC. However, their MSI-H subgroup comprised a majority (84%) of hereditary nonpolyposis cancers that are genetically different and may have quite different growth characteristics than their sporadic counterparts. Furthermore, Ki-67 staining was analyzed semiquantitatively. Taken together, these differences may explain why the Bocker Edmonston study found no significant correlation between the Ki-67 PI and MSI status (48).

Given the strong correlation between the AI and PI, we examined possible correlations with the AI:PI ratio. These included patient age, sex, MSI status, tumor stage, site, grade and type, TP53, K-ras, TGF-βRII and BAX mutation status. Unlike the Ki-67 univariate analysis, only MSI status was significantly associated with the AI:PI ratio, with MSI-H and MSI-L cancers showing a 2-fold higher ratio than MSS cancers This analysis confirms that MSI status is the only variable that independently predicts apoptosis and proliferative activity in CRC.

Ki-67 was identified previously as an independent prognostic marker for patients with CRC in some studies (42, 49, 50, 51) but not others (Refs. 43, 44, 45; Table 5). In the current study, a high Ki-67 index was associated with improved survival in MSI-H cancers (P = 0.09) but not MSI-L or MSS cancers. Stage D MSS/MSI-L tumors in this series tended to have lower proliferative activity (Ki-67, <50%). It is noteworthy that in the four CRC studies that found a positive correlation between a high PI and poor prognosis, the correlations were limited to particular stages (Refs. 42, 49, 50, 51; Table 5). Two studies that measured proliferation by Ki-67 immunostaining in a large CRC series not stratified by MSI found no significant correlation between PI and survival (43, 45). Significantly, diploidy (49, 50, 51) and right-sided location (49) independently predicted better survival in CRC, especially diploidy in proximal cancers (44). These findings are consistent with results obtained in our laboratory, given that MSI-H cancers are predominantly right-sided and diploid (2). It is relevant that none of the previously mentioned studies assessed MSI status (Refs. 42, 43, 44, 45, 49, 50, 51; Table 5). Therefore, the data from the 10 to 15% MSI-H cancers present in those unselected series would have been masked by the results of the other CRC subgroups.

The increased level of spontaneous apoptosis noted previously in MSI-H cancers indicates that apoptosis signaling does not appear to be especially defective or deregulated in sporadic MSI-H cancers (12, 52). Despite the higher mean AI and better prognosis in MSI-H cancers, we found that the AI was not significantly associated with survival in CRC, either overall or within MSI subgroups. We also found that a low AI:PI ratio across the entire series, indicating a higher index of cell production, was associated with a favorable prognosis with borderline significance (P = 0.087). The explanation for this finding is not clear, and the relationship was not observed when tumors were divided into subgroups according to MSI status. To our knowledge, the only other CRC study to compare the relationship between AI:PI ratio and patient survival did not find an association (44). However, that study included only lymph node-negative (Dukes’ stage A and B) cancers and measured apoptosis by the less sensitive morphological assessment in H&E-stained sections.

Most patients with CRC die from metastatic disease rather than local tumor recurrence (53), and there is no relationship between tumor size and lymph node metastases (25). Consistent with our findings, the Ki-67 index did not correlate with tumor size in any study to date (24, 25, 36, 43). This study provides further evidence that tumor size is not necessarily related to tumor aggression or invasive potential. MSI-H cancers are larger at presentation (2), yet are less aggressive than their MSI-L and MSS counterparts, rarely progressing to stage D disease with distant metastases (13, 14, 15, 16).

MSI-H cancers are characterized by a higher PI that appears to be associated with an improved survival within this subset. The latter finding may be explained, at least in part, by biological heterogeneity within MSI-H cancers. MSI-H cancers with a low PI may be less typical or show more overlap with non-MSI-H cancers. Furthermore, the higher PI in MSI-H cancers is more than offset by the higher AI, giving a reduced index of cell production compared with MSS but not MSI-L cancers. This implies a greater disruption of apoptosis in MSS versus MSI-H cancers. The intermediate AI in MSI-L cancers is not easily explained because TP53 alterations (mutation, LOH, and protein expression) are similar in MSI-L and MSS cancers (3, 6, 33).

In conclusion, this study is the first to show that sporadic MSI-H cancers are characterized by a significantly increased Ki-67 proliferative index. MSI cancers also show an increased AI:PI ratio. Consistent with previous reports (22, 36, 43, 44), the PI was not related to the site of the tumor per se. Studies from this laboratory have also shown that the AI is significantly higher in MSI-H cancers (12). Apoptosis and proliferative indices were positively correlated in this series of sporadic CRC. In contrast to recent studies in lung, bladder, and breast cancer (17, 18, 20), as well as some studies in unselected CRC series (42, 49, 50, 51), but in agreement with several other CRC studies (43, 44, 45), we did not find that a high mean PI was a poor prognostic indicator. Rather, our survival analysis suggests that a high PI may be associated with the MSI-H survival advantage, whereas in MSI-L and MSS cancers, the PI did not correlate with survival. The increased PI in MSI-H cancers may be explained by the higher proportion of MSI-H cancers that are undifferentiated or poorly differentiated. This finding does not conflict with the known behavior of MSI-H cancer because the apparent increased growth fraction is more than offset by the higher AI. Additionally, an increased PI need not indicate increased tumor aggression because the latter is more directly explained by mechanisms underlying invasiveness and metastasis.

Our thanks to Clay Winterford for excellent photographic reproduction and Kelli Biden, Ron Buttenshaw, Dianne McKeone, and Estelle Schoch for technical expertise. Tumors were obtained from the Walter Paulsen Memorial Tumor Bank; we are indebted to the general surgeons and Professor Jeffrey Searle (Department of Pathology, Royal Brisbane Hospital) for assistance in tumor collection.