Colorectal Mucosal Expression of MSH2 as a Potential Biomarker of Risk for Colorectal Neoplasms

Despite noticeable advances in cancer diagnostics and treatment, colorectal cancer remains the second leading cause of cancer mortality in the United States (1). It is a multifactorial disease that appears to be the result of interacting lifestyle and genetic factors (2-9). The adenomatous polyp, the precursor to the vast majority of so-called “sporadic” colorectal cancer, currently is the only accepted reliable biomarker of risk for colorectal cancer (8, 9), and its removal markedly reduces risk of cancer development. However, colonoscopy and sigmoidoscopy, the procedures currently used to screen for adenoma, are costly and labor-intensive, require highly qualified personnel, and are not well accepted by many patients and physicians. Preneoplastic biomarkers or profiles of biomarkers of risk for colorectal neoplasms will help address these challenges. The developed biomarkers can be used to individualize colonoscopy recommendations and schedules and to monitor the efficacy of preventive interventions.

The mismatch repair pathway is one of the two main molecular pathways of colorectal cancer development, accounting for ∼15% of colorectal cancers (10). The DNA mismatch repair system is a highly conservative mechanism that involves a complex set of proteins that identifies and repairs mismatch errors that occur during DNA replication (8, 11, 12). A crucial part of this system is the MutS-homologue 2, colon cancer, nonpolyposis type 1 (Escherichia coli) [Homo sapiens] (MSH2) gene, which is located at chromosome 2p22 (13). The product of this gene, MSH2 protein, recognizes DNA mismatches by forming two functional heterodimers: MSH2-MSH6, which recognizes single-base mismatches and short insertion-deletion loops, and MSH2-MSH3, which recognizes larger loops (13-16).

In the DNA mismatch repair mechanism, MSH2 protein closely cooperates with the other key protein, MLH1; whereas decrease or loss of MLH1 protein expression usually occurs due to hypermethylation of the MLH1 gene, MSH2 expression may change because of other factors (17). MSH2 heterodimers recognize DNA mismatches and recruit the hMLH1α heterodimer (MLH1-PMS2), which facilitates mismatch correction. Thus, low expression of MSH2 may lead to ineffective recognition and correction of DNA mismatches even when MLH1 expression is adequate. Earlier animal and human studies suggest that MSH2 protein may have a role in apoptotic signaling involving bcl-2 and p53 proteins (18, 19). Decrease or loss of MSH2 expression in the colorectal mucosa may predispose to malignancy independently of loss MLH1 expression. Expression of MSH2 protein in the colon cell is likely to indicate the functional level of the DNA mismatch repair mechanism, which makes MSH2 expression in normal colorectal mucosa a promising candidate as a biomarker of risk for colorectal cancer.

To the authors' knowledge, there is no literature addressing the distribution of MSH2 protein in normal-appearing colorectal mucosa, its potential as a biomarker of risk for colorectal cancer, or its associations with risk factors for colorectal neoplasms. To address this, we conducted a case-control study of incident, sporadic colorectal adenoma in which we measured the overall expression and distribution of MSH2 protein within the crypts of the normal-appearing colorectal mucosa and estimated their associations with colorectal adenoma and known risk factors for colorectal neoplasms as a first step in evaluating MSH2 as a prospective biomarker of risk for colorectal neoplasms.

The Markers of Adenomatous Polyps II (MAP II) study is a pilot case-control study (51 cases and 154 controls) designed to investigate potential biomarkers of risk for incident, sporadic colorectal adenomas. Participants were recruited from people scheduled for elective outpatient colonoscopy at Consultants in Gastroenterology, a large gastroenterology practice in Columbia, South Carolina. To be considered eligible for the study, participants must have been 30 to 74 years old, of both sexes and all races, English-speaking, and capable of providing informed consent. Specific exclusion criteria included history of previous colorectal adenomas, history or findings consistent with familial adenomatous polyposis or hereditary nonpolyposis colon cancer syndromes, inflammatory bowel disease, bowel resection, history of cancer other than nonmelanoma skin cancer, and medical contraindication to colorectal mucosal biopsies (medically unstable, bleeding disorders, and cannot stop warfarin or aspirin), and polyethylene glycol colon-cleansing preparations.

Over a 5-month period, 351 patients were identified for recruitment; 232 (76%) of these agreed to participate in the study and 205 (51 cases and 154 controls) met final eligibility criteria and were included in the study. Due to limited tissue and financial resources, biopsies from only 92 participants (45 cases and 47 controls) were processed for MSH2 expression and used for the analysis reported here.

Data on medical history, family history of cancer, diet, lifestyle, and anthropometrics were collected using mailed questionnaires, including a modified Willett Food Frequency Questionnaire, before the colonoscopy visit and knowledge of case-control status.

The colon site and in vivo size and shape of all polyps found were recorded. All polyps were removed and examined by one study index pathologist who identified polyp type, subtype, and degree of atypia according to criteria established by the National Polyp Study (20).

After a 12-h fast and polyethylene glycol bowel cleansing preparation, biopsies of normal-appearing mucosa were collected according to a standard protocol by gastroenterologists using standard-cup flexible endoscopy forceps during usual care colonoscopies. Six sextant pinch biopsies, ∼1 mm thick, were obtained from the rectum (10 cm above the anus) on all participants, and from the mid-sigmoid and proximal ascending (immediately distal to the cecum) colon on 20% of participants, for a total of up to 18 biopsies. No biopsies were taken within 4.0 cm of a polypoid lesion.

Biopsy specimens were fixed by 10% normal buffered formalin for 24 h and then transferred to 70% ethanol. Within a week, the biopsies were processed and embedded in paraffin blocks with three biopsies per colon site per participant per block.

Within 7 days of being embedded in paraffin blocks, 3.0 μm thick sections taken 30 μm apart were cut from each block with a microtome such that five slides with four levels each (yielding a total of 20 levels) were prepared per colon site per person.

The immunohistochemistry protocol was calibrated to produce the darkest biomarker labeling staining possible short of yielding nonspecific background staining (21). The slides were immunohistochemically processed using a DAKO Automated Immunostainer and Leica H&E Autostainer (Leica Microsystems). First, MSH2 antigen was unmasked via a heat-induced epitope retrieval procedure by placing the slides in a preheated Pretreatment Module (Lab Vision) with DAKO TBS (DAKO S1968) and steamed for 40 min. Then, the slides were immunohistochemically processed using an anti-MSH2 antibody (Oncogene NA27) in a 1:500 dilution, a DAKO Envision+ detection system (DAKO K4007), and 3,3′-diaminobenzidine (DAKO K3466) as the chromogen. No counterstaining was used and all stained slides were glass coverslipped with a Leica automated coverslipper (Leica Microsystems).

All five slides per colon site per person were included in one staining batch of up to 48 slides that also included negative and positive control slides. A surgical specimen of normal colon was used for the control slides; the negative and positive control slides were treated identically to study participant slides, except that antibody diluent was used rather than primary antibody on the negative control slide.

MSH2 is expressed in a density gradient along the crypt (Fig. 1) and cannot be quantified by eye (e.g., by counting cells). Therefore, MSH2 expression density, detected by immunohistochemical staining, was quantified in the stained slides using image analysis densitometry methods (22, 23). The procedure was conducted by one trained “scorer” using a light microscope (Olympus BX40), digital camera (Polaroid DMC Digital Light Microscope Camera), digital drawing tablet, and a custom-developed plug-in to ImagePro Plus (Media Cybernetics) image analysis software. The scorer was blinded to case-control status and colon site.

The imaging and analysis unit was a “hemicrypt,” defined as one side of a colonic crypt bisected from base to colon lumen surface (Fig. 1). Intact (at most two contiguous cells missing) hemicrypts extending from the muscularis mucosae to the colon lumen were considered eligible for quantitative image analysis (“scorable”).

For each patient, two of the three biopsies from each colon site with the greatest number of “scorable” hemicrypts were selected for quantitative image analysis (“scoring”). Intact hemicrypts were “scored” in order from the first hemicrypt on the first biopsy from left to right. The goal was to find at least 16 “scorable” hemicrypts per biopsy (32 per patient; ref. 21). If the 16th hemicrypt was reached before the level was finished, the scorer continued scoring until either the level was finished or the 20th hemicrypt was scored, whichever came first. No more than 20 hemicrypts per biopsy were scored. If the two best biopsies from a colon site on a patient had less than 32 “scorable” hemicrypts, an attempt was made to cut more slides. If that did not solve the issue, scoring was completed if the two best biopsies had ≥16 “scorable” hemicrypts between them. All three biopsies harvested from the same colon site were scored only if there was less than a total of 16 “scorable” hemicrypts between the two best biopsies.

To ensure scoring protocol adherence, a scorer was guided through the scoring protocol by the computer software. For each scored slide, background correction images were obtained and automatically used by the computer program to yield background corrected densitometries for all hemicrypts analyzed on that slide. All images were taken at ×200 magnification (the maximum magnification at which full-length colorectal crypts can be completely included in a single visual field) and stored and analyzed as 16-bit grayscale 1,600 × 1,200 pixel images.

The scorer used a digital drawing tablet to manually trace hemicrypts from the crypt base center cell up along the crypt basement membrane to the beginning of the turn of the crypt onto the colonic mucosal surface and then back down along the crypt luminal surface of the epithelial nuclei (Fig. 1; refs. 22, 23). The software program divided the traced hemicrypt into segments corresponding in width to that of an average normal crypt epithelial cell (6.59 μm; ref. 21) and then calculated overall hemicrypt- and segment-specific labeling optical density and entered these data into a Microsoft Access database along with various dimensional parameters of the hemicrypt (22, 23).

For quality assurance, slide sets from 10% of the participants were randomly selected by the statistical team, blinded, and resubmitted to the scorer for rescoring (21).

Statistical analyses were done using SAS 9.2 statistical software (© 2002-2008 by SAS Institute). The entire MAP II study population (51 cases and 154 controls) as well as the subset of participants for whom slides were immunohistochemically processed for MSH2 protein (45 cases and 47 controls) were assessed for comparability using the t test for continuous variables, the Fisher's exact test for dichotomous variables, and the Freeman-Halton extension to Fisher's exact test for categorical variables as appropriate. All labeling optical density means were calculated using linear mixed models. Potential confounders as well as staining batch were included in the models as fixed effects, and correlation among multiple optical density measurements within each patient was accounted for by including a patient variable as a random effect. Mean proportional differences were calculated as the model-predicted mean optical density for cases minus that for controls divided by the mean for cases. Statistical significance of these measurement differences was evaluated by t test. The intraclass correlation coefficient was used to assess slide scoring reliability and found to be r = 0.96.

The distribution of MSH2 protein within colonic crypts was evaluated graphically with the Loess procedure as implemented in SAS version 9 statistical software (24). First, the number of cells within a hemicrypt was standardized to 50 segments (the average number of cells within a column of colonic crypt cells). Then, average Loess model-predicted, segment-specific levels of MSH2 for cases and controls by colon site were plotted in the graphs (Fig. 2) along with smoothing lines to make graphical evaluation easier.

Potential confounders were evaluated based on biological plausibility and whether the variable of interest was associated with the exposure based on existing epidemiologic, medical, and basic science literature. Potential confounders considered in this analysis included age, sex, physical activity, body mass index, family history of colorectal cancer in a first-degree relative, smoking, alcohol consumption, aspirin and nonsteroidal anti-inflammatory drug (NSAID) use, and total intakes of energy, fat, fiber, folate, calcium, and vitamin D. All nutrient values were adjusted for total energy according to the residual regression method (25). Continuous variables were dichotomized based on their distributions in the controls.

The association between MSH2 expression and risk of incident sporadic colorectal adenoma within each colon site was assessed with linear mixed models using individual hemicrypt measurements. Potential confounders and staining batch were entered into the models as fixed effects; a random intercept was added to each model to account for multiple correlated optical density measurements within each subject. The overall association between MSH2 expression in the colorectal mucosa and risk of incident, sporadic colorectal adenoma was evaluated by calculating odds ratios (OR) from repeated measures logistic (generalized estimated equations) models based on average hemicrypt MSH2 expression from all three colon sites within an individual. Both mixed and generalized estimated equations models contained the same set of potential confounders. A 95% confidence interval (95% CI) was calculated for each OR. To build the most parsimonious model that adequately controlled for confounding, first, all a priori identified potential confounding variables were ranked based on published literature on their hypothesized relative contributions to risk for colorectal neoplasms and then again on the strengths of their associations with the biomarkers investigated in this study. Next, a summary rank was calculated and covariates were added to the age- and sex-adjusted model one at a time according to their rank from highest to lowest. The model that adequately controlled for confounding and had the smallest number of parameters was selected as the final multivariable-adjusted model.

Repeated-measures logistic models (generalized estimated equations) were also used to model associations between the level of MSH2 expression in the rectum and various adenoma characteristics such as location, multiplicity, degree of dysplasia, histologic type, and shape. For this analysis, the batch-standardized optical density variable was dichotomized based on the colon site-specific mean in the controls.

The associations of MSH2 expression in the rectum with various demographic, lifestyle, and dietary characteristics were assessed by mixed models. Potential confounders were entered into the model one at a time as fixed effects. The model also included a fixed effect to control for case-control status and an appropriate interaction term to check for potential modification of the effect of each characteristic by case-control status. A random intercept was included in the model to account for lack of independence among hemicrypt measurements within each patient. The number of biopsies from the more proximal portions of the colon was too small for reliable similar analyses.

In sensitivity analyses, we also analyzed data without standardization for batch as well as by using different mathematical transformations; the results from these analyses did not differ materially from those reported.

The subpopulation of subjects whose biopsies were stained for MSH2 and analyzed (45 cases and 47 controls) was compared with the entire MAP II study population (51 cases and 154 controls) and found completely comparable with respect to all considered characteristics (data not shown). Selected characteristics of cases and controls of the population considered in this analysis are shown in Table 1. On average, cases tended to be older, more likely to be male, more likely to be a current smoker and currently consume alcohol, less likely to regularly take a NSAID, and tended to have higher total energy intakes and lower intakes of calcium, vitamin D, and folate than controls, but only the difference for total energy intake was statistically significant. Physical activity, body mass index, aspirin use, and fat and fiber intakes did not differ substantially between cases and controls.

Among cases, 30% had multiple adenomas, 8% had an adenoma that was ≥1.0 cm in diameter, and 14% had moderate or severe dysplasia in their largest or most advanced adenoma. In adenoma cases, the largest or most advanced adenoma was located in the rectum in 26%; in the sigmoid or descending colon in 29%; in the splenic flexure, transverse colon, or hepatic flexure in 28%; and in the ascending colon or cecum in 19% (data not shown).

The distribution of MSH2 protein within colonic crypts in the rectum, sigmoid, and ascending colon by colon site is presented in Fig. 2A to C, respectively. In all three colon sites, MSH2 expression was higher at the base of the crypt and progressively decreased toward the opening of the crypt onto the colon lumen. In the rectum, MSH2 expression was highest in the lower 20% of the crypt and then rapidly decreased and leveled off the rest of the way up the crypt. In the sigmoid and ascending colon, MSH2 expression was highest in the lower 60% of the crypt (the proliferation zone) and lower in the upper 40% of the crypt (the differentiation zone; refs. 2, 3, 21). MSH2 expression in the rectum and sigmoid colon appeared virtually identical in cases and controls; in the ascending colon, MSH2 expression in cases was noticeably lower than in controls. For each of the three colon sites investigated in the study, the MSH2 expression curves for cases and controls closely paralleled each other, indicating that the differences in MSH2 expression between cases and controls were uniform along the full lengths of the crypts and were not confined to a functional zone of the crypt or to distribution differences; therefore, only the results from analyses of total crypt expression are reported below and in the tables.

Table 2 presents “crude,” age- and sex-adjusted, and multivariable-adjusted MSH2 expression in all cases and controls stratified by colon site as well as the combined OR for the association of MSH2 expression with incident, sporadic colorectal adenoma. On average, after adjusting for potential confounders, expression of MSH2 protein in the rectum and ascending colon was 23% (P = 0.06) and 49% (P = 0.01), respectively, lower in adenoma cases than in controls. In the sigmoid mucosa, on the other hand, MSH2 expression was, on average, 25% higher in cases than in controls, but the difference was not statistically significant (P = 0.42). Risk of incident, sporadic colorectal adenomas was inversely associated with MSH2 expression in colonic crypts from all three colon sites combined, but the association was not statistically significant (combined OR, 0.77; 95% CI, 0.38-1.58).

To assess whether MSH2 expression in the rectum may be associated with a subset of cases (especially those with right-sided adenomas), we investigated associations of MSH2 protein in the rectum with various adenoma characteristics (Table 3). The overall inverse association of MSH2 expression with adenomas tended to be stronger for adenomas in the right colon and for those with more advanced characteristics (villous component or moderate/severe dysplasia), but the sample size was small and the findings were not statistically significant.

We also assessed the potential of MSH2 expression in the rectum as a modifiable biomarker of risk by evaluating associations of MSH2 expression with various risk factors for colorectal cancer (Table 4). The results from the similar analyses for the two proximal colon sites are not presented because the very limited sample size for these colon sites prohibited reliable estimation. The only statistically significant finding was that MSH2 expression in the rectal mucosa was 39% (P = 0.04) higher in subjects who took aspirin or another NSAID at least once a week; this association was stronger in cases (54%; P = 0.09) than in controls (25%; P = 0.22; data not shown).

To our knowledge, this is the first study to report on the distribution of MSH2 protein within normal colorectal crypts in humans or on associations of MSH2 expression in normal-appearing colorectal mucosa with risk for incident, sporadic colorectal neoplasms or with risk factors for colorectal cancer. Our preliminary data support the hypothesis that MSH2 expression in the normal colonic mucosa, especially in the ascending colon, is inversely associated with risk of incident, sporadic colorectal adenoma. The data also suggest that MSH2 expression in the normal colon may be associated with modifiable risk factors for colorectal neoplasms. The association between MSH2 expression in the ascending colon and risk of sporadic adenoma reported here is in the same direction as and of similar magnitude to that reported earlier for MLH1 expression (12). This is to be expected because the two proteins closely cooperate in the DNA mismatch repair mechanism. Our findings may be explained by possible inactivation of the MSH2 gene via DNA methylation, which results in a loss of expression of the protein in the mucosa. Seifert and Reichrath showed that most inactivating mutations in the MSH2 gene lead to a lack of expression or the expression of a truncated protein not detectable by antibodies used in many studies (16). Decreased MSH2 expression in adenoma cases may also be the result of missense mutations in the MSH2 gene (26) or downregulation by bcl-2 (18, 27).

Earlier studies found that MLH1 and MSH2 expression may be lost in adenomas that have a high degree of microsatellite instability in patients diagnosed with hereditary nonpolyposis colorectal cancer (28-30). Muller et al. reported that loss of expression of mismatch repair proteins in hereditary nonpolyposis colorectal cancer patients depended on the distance between an adenoma and a malignant lesion; adenomas located closer than 5.0 cm to a carcinoma were less likely to have mismatch repair proteins expressed (28). The results of our study are consistent with these findings. We investigated normal-appearing mucosa in cancer-free individuals and found decreased MSH2 expression in the ascending colon in adenoma cases, which could be an early sign of mismatch repair deficiency. Earlier studies could only detect whether MSH2 was expressed in the tissue or not, whereas our image analysis methodology allowed us to measure and compare its expression between cases and controls. Mismatch repair–deficient colorectal cancers are characterized by early development of microsatellite instability (29), which could explain the lower MSH2 expression we observed in the ascending colon mucosa of adenoma cases.

We also found that, as would be expected from the known function of MSH2, the colon crypt expression distribution curves for MSH2 appear to follow the cell proliferation distribution within colonic crypts (Fig. 2) with higher expression of the protein in the lower 60% of crypts (proliferation zone) and lower expression in the upper 40% of the crypt (differentiation zone; refs. 2, 3, 21).

The rectum is the most practical colon site for chemoprevention trials and potential clinical outpatient applications because the procedures for obtaining rectal biopsies 10 cm above the anus are minimally invasive and do not require fasting or bowel cleansing preparations (3, 31). The data from our pilot study suggest that MSH2 expression in the rectum may be associated with risk for incident colorectal adenomas and that this association parallels that in the ascending colon (Table 2). We also found that lower MSH2 expression in the rectum tended to be more strongly associated with adenomas in the right colon than in the left colon (Table 3). These findings are consistent with previous research indicating that mismatch repair–deficient colorectal neoplasms tend to be located in the right colon (8, 9).

It was unexpected that associations between MLH1 and MSH2 expression in the rectum and adenoma characteristics were inconsistent. Whereas various adenoma characteristics were positively associated with rectal expression of MLH1 (12), their associations with MSH2 expression were mostly negative. A possible explanation for this discrepancy may be that both functional heterodimers that MSH2 participates in are part of the DNA mismatch repair mechanism and are involved in DNA mismatch recognition, but among the three heterodimers that MLH1 forms, only one, mutLα (MLH1-PMS2), has been confirmed to take part in the mismatch repair (32). mutLγ (MLH1-MLH3) is thought to play a role in meiosis, and the exact function of mutLβ (MLH1-PMS1) is still unclear (11). It is possible that increased MLH1 expression is associated with other functions of this protein in the cell cycle and not only with DNA mismatch repair. Given the modest sample size of this pilot study, a chance finding is a real possibility.

Our preliminary results, if confirmed by a full-scale study, may indicate that MSH2 expression in the rectum may be a good indicator of its expression and thus risk in the proximal colon, which is much less accessible for screening procedures. We also observed that higher MSH2 expression in the rectum may be associated with lower risk for more advanced adenomas; however, these associations were based on a small sample size and were not statistically significant.

Our analyses to assess associations of MSH2 expression with various risk factors for colorectal neoplasms were severely limited by our small sample size. However, we did find that MSH2 expression in the rectum was statistically significantly substantially higher among persons who regularly took a NSAID, one of the clearest modifiers of risk for colorectal neoplasms. A similar but not statistically significant association between higher doses of aspirin and MSH2 expression was observed in vitro (33). The exact mechanisms underlying the chemopreventive effects of NSAIDs against colorectal neoplasms are not yet fully understood. In addition to cyclooxygenase-2 inhibition, NSAIDs may also act through non-cyclooxygenase-2–dependent mechanisms (34, 35). NSAIDs are effective against mismatch repair–deficient colorectal cancers despite the fact that cyclooxygenase-2 expression is often reduced in mismatch repair–deficient colorectal cancer cells (35), so non-cyclooxygenase-2–dependent mechanisms may play a crucial role in this type of cancer.

Because this study was a pilot study, its main limitation was the small sample size. Due to limited resources and the complexity of the procedure, only a fraction of the patients' biopsies were processed for MSH2 expression, further reducing the sample size. For the same reasons, biopsies for all three colon sites were often not available. Using a commercial automated immunostainer did not completely eliminate staining variability between staining batches, which introduced an additional source of variability into the analysis that had to be accounted for. The participants in this study were drawn from people who underwent a colonoscopy, so the results of this study may not be directly applicable to the general population. Data collected by food frequency questionnaires and self-reported data have shortcomings that are well described in the literature. Because these data were collected before case-control status was determined, any possible bias is expected to be nondifferential.

On the other hand, this study had several important strengths: (a) all participants underwent colonoscopy, which ensured accurate identification of cases and controls; (b) all self-reported data (including dietary information) were collected before the case-control status of each participant was determined, thus minimizing possible recall bias; (c) detailed information on potential confounders such as anthropometrics, diet, vitamin and mineral supplements, and medications used was collected; and (d) the rigorous procedures for biopsy collection, processing, and quantitative assessment of the labeling absorbance of MSH2 expression detected by immunohistochemistry using our specially developed software, which minimized possible measurement error.

In summary, we developed a reliable procedure for detecting and describing MSH2 expression in normal colorectal crypts and report, to our knowledge, the first study to describe the distribution of the MSH2 protein within normal colorectal crypts or on associations of MSH2 expression in normal-appearing colorectal mucosa with risk for incident, sporadic colorectal neoplasms or with important risk factors for colorectal cancer. We found that the distribution of the MSH2 protein within normal colonic crypts parallels that of the normal proliferation zone of normal crypts. The data from this preliminary study suggest that lower MSH2 expression in the normal colonic mucosa may be associated with increased risk of incident, sporadic colorectal adenoma as well as with modifiable risk factors for colorectal neoplasms and thus support further investigation of MSH2 expression, alone or in combination with other biomarkers, as a potential modifiable biomarker of risk for colorectal neoplasms.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank the physicians and staff of Consultants in Gastroenterology (131 Summerplace Drive, Columbia, SC 29169) for work on biopsy procurement and all study participants for their time and dedication to the study.