Expression of the Tumor Suppressor Krüppel-Like Factor 4 as a Prognostic Predictor for Colon Cancer Free

Despite substantial advances in the early diagnosis and treatment of colorectal cancer, it remains a disease with high morbidity and mortality. Approximately 150,000 new cases of colon cancer were diagnosed last year in the United States, making it the fourth most common cancer diagnosed in men and women (1). It is also the second most common cause of cancer-related death despite a recent reduction in mortality due to increased screening and improvements in treatment for late-stage cancer (2). Studies have identified mutations in tumor suppressor genes and oncogenes that result in dysregulation of numerous pathways leading to colorectal carcinogenesis (3, 4). Nonetheless, future research in colorectal cancer depends on the ability to conduct research relative to potential applications of available data, translate the mechanistic data into the clinical arena, and evaluate markers independent of established clinicopathologic predictors of the disease.

Biomarker research in colorectal cancer is becoming increasingly popular for a variety of clinical and research applications. Precise biomarkers may be useful as a surrogate endpoint in preliminary studies, for the stratification of patients in clinical trials, and in the refinement of disease prognosis. The debate surrounding the role of chemotherapeutics in American Joint Committee on Cancer (AJCC) stage II colon cancer provides an example of such an application. Within the past decade, a number of new systemic treatments for colon cancer, including oral fluropyrimidines, oxaliplatin, and irinotecan, have been shown to improve overall and disease-free survival in stage III cancer patients (5). Unlike patients with third-stage cancer, current recommendations by the American Society of Clinical Oncology do not promote the routine use of chemotherapeutics in stage II cancer (6). Although no clinical trial has proven overall benefit of adjuvant therapy in stage II, it is hypothesized that up to 20% of patients with a risk for recurrence similar to that of stage III disease would have benefited from chemotherapeutics (5). Currently, the American Society of Clinical Oncology suggests that patients with inadequately sampled nodes, T₄ lesions, perforation, or poorly differentiated histology should at least be considered for such treatment at the discretion of the treating physician and patient (6). However, no studies have evaluated the benefit of such a recommendation or whether the suggested criteria correctly predict recurrence risk. Biomarkers may prove especially useful in further stratifying stage II patients into those who have a high risk for recurrence and those who do not, the former group being more likely to benefit from chemotherapeutics. Several biomarkers such as S-phase fractions and vascular endothelial growth factor expression have been evaluated in this regard (7, 8).

Krüppel-like factors (KLF) are a family of evolutionarily conserved mammalian zinc finger transcription factors named for their homology with Krüppel, a Drosophila melanogaster protein (9). KLFs are involved in a diverse array of fundamental biological processes, including proliferation, differentiation, development, and apoptosis (10-12). Among the KLF family, KLF4 (also called gut-enriched KLF or GKLF) was one of the first identified (13, 14). In addition to regulating many important physiologic processes, KLF4 has been shown to play a role in pathologic conditions such as cancer and inflammation (15-20). More recently, KLF4 was shown to play a crucial role in the reprogramming of somatic cells into induced pluripotent stem cells (21-25).

Expression of KLF4 is enriched in epithelial tissues, including the gut (13, 14). In the intestine, KLF4 is highly expressed in the postmitotic, terminally differentiated epithelial cells at the luminal surface (26, 27). In vitro, KLF4 inhibits proliferation by blocking cell cycle progression at the G₁/S boundary (28). In addition, KLF4 mediates the cell cycle checkpoint functions of the tumor suppressor p53 following DNA damage (29-31). Moreover, loss of expression of KLF4 due to several reasons, including loss of heterozygosity, promoter hypermethylation, and loss-of-function mutations, has been documented in a small set of colorectal cancer specimens (32). This tumor suppressive effect of KLF4 is supported by in vivo evidence in which mice heterozygous for Klf4 manifest increased tumor burden when bred to the Apc^Min mice that are genetically predisposed to intestinal adenoma formation (17). From a mechanistic perspective, KLF4 inhibits Wnt signaling, a key pathway in colorectal carcinogenesis, by inhibiting the activity of β-catenin, a mediator of Wnt signaling (33).

Because of its putative tumor suppressor function, KLF4 may serve as a prognostic indicator for colon cancer. Here, we measured KLF4 expression levels in a large cohort of primary colon cancer specimens in an attempt to correlate its expression with clinical parameters, including survival and recurrence.

A retrospective case control study was conducted to evaluate the association between KLF4 and colon cancer while adjusting for a number of covariates. A tissue microarray bearing a large number of colon cancers across various stages was analyzed by immunohistochemistry for KLF4. Each tissue section was processed in duplicate because 2-fold redundancy permits accurate analysis of protein expression (34).

The paraffin-fixed colon tissue microarray was constructed between 1989 and 1996 by the National Cancer Institute's Cancer Diagnosis Program of the National Cancer Institute (35). The microarray was assembled using 367 cores of colon cancer (49 stage I, 122 stage II, 144 stage III, and 52 stage IV), 37 cores of adenomatous polyps, 34 cores of normal colon tissues matched to tumor sections, and an additional 40 normal colon sections from individuals with diverticulosis. Of the colon cancer cases, 5-year follow up data were complete on 96 stage II tumors (26 recurred) and 125 stage III tumors (65 recurred). Of all stage II and III tumors, 45 were censored before the 5-year follow up was complete. Noncolon tissue cores were embedded on the microarrays for internal control of staining. In addition, cores from colon cancer cell lines were embedded as additional internal controls. These cell lines were characterized and authenticated by the National Cancer Institute under American Type Culture Collection guidelines (35). Of all patients, 418 were Caucasian, 12 were African American, and 11 identified with another race. Two hundred and seven subjects were male, and 233 were female (gender of one individual was unknown). Mean age was 68.62 (SD, 12.48). Additional covariate data were collected on tumor depth, nodal status, metastasis, histology, location, and degree of dysplasia as assessed by an independent pathologist.

Selection of patients across various stages was done to ensure enough power to detect differences in recurrence within stage II and stage III cancer independently. The cases were chosen to detect a difference in the prevalence rate of 0.35 within stage II or III tumors that recurred and those that did not within the 5-year follow-up period. To detect differences in a binary outcome marker across various stages of disease, enough stage I and IV tumors were also included, so >80% power was available to detect a 0.30 difference in prevalence rate of KLF4 (see Guidance for Statistical Analysis of Biomarker Data Generated from the NCI Colon Tissue Microarray; ref. 36).

The microarrays were treated with xylene for deparaffinization and rehydrated with ethanol. Endogenous peroxidase activity was blunted with 10% H₂O₂ in methanol. Antigen retrieval was done using 10 mmol/L citrate buffer (pH 6.0) at 120°C for 15 minutes. The sections were then incubated for 1 hour in blocking buffer (2% nonfat dry milk, 0.001% Tween 20, and 10% normal horse serum in PBS). Vector Laboratories avidin/biotin blocking kit was used in conjunction with blocking buffer as directed by the manufacturer to reduce background and nonspecific secondary antibody binding. Sections were stained with KLF4 (goat anti-human KLF4; R and D Systems) at a dilution of 1:1,000 in blocking buffer for 1 hour. Detection of primary antibody and color development was done using Biocare Medical Betazoid DAB development kit. Sections were counterstained with Mayer's hematoxylin (Invitrogen), dehydrated, and covered with cover slip. Images were acquired with an Axioskop 2 plus microscope (Zeiss) with an AxioCam MRc5 CCD camera (Zeiss).

Images were graded by two investigators blinded to tissue stage as assessed by an independent pathologist and other covariate information. Tissues were graded either negative (<10% staining) or positive (≥10% staining) in a manner similar to previous reports (37). Intensity of nuclear KLF staining was compared between histopathologic stages using the χ² test or Fisher's exact test as appropriate. A binary logistic model was created to assess the role of stage, age at diagnosis, race, and gender in determining odds ratios for KLF4. All two-way interaction terms were evaluated using a Wald test for inclusion into the model. At each step of modeling, the most insignificant interaction term was dropped, and the model was then evaluated. After a covariate was dropped, assessment about whether previously dropped covariates could be reentered into the model was done. Once all covariates were evaluated, the final model was constructed.

Overall survival was defined as time from diagnosis to death of patients. Disease-free survival was defined by time between diagnosis of colon cancer and recurrence of disease. The association between KLF4 expression and survival was assessed by Kaplan-Meier survival analysis. Differences between curves were assessed using a log-rank test. To evaluate KLF4 expression as an independent prognostic factor for overall and disease-free survival, a Cox regression model was applied, and hazard ratios were estimated. In a similar fashion to logistic model building, all possible two-way interaction terms were evaluated after adjustment of the model to fulfill the proportional hazards assumption. P < 0.05 was considered indicative of statistical significance. The statistical software package SAS 9.2 was used for statistical analysis and graphics.

The general characteristics of subjects enrolled in the study are shown in Table 1. Most participants were White for cancer and noncancer cases. Gender and age were relatively evenly distributed in the two groups of subjects. Table 2 shows the characteristics of tumor sections among AJCC stages. Lymph nodes were not examined in seven stage IV patients, and status was unknown in one stage IV patient. Distal margins were involved in one stage I patient, otherwise no enrolled patients had either proximal or distal margin involvement. In five stage IV patients, margins could not be assessed.

Figure 1 shows a representative example of KLF4 immunostaining of the colon cancer tissue microarray. To assess the relationship of certain covariates with KLF4, a univariate analysis was first undertaken (Table 3). Cancer tissues had significantly less overall KLF4 expression in comparison to noncancer tissues (P < 0.0001). The proportions of KLF4-positive tumors were significantly different among men and women (P = 0.0432). Proportions of KLF4-positive and -negative tumors were also significantly different among stage I (P = 0.0341) and stage III (P = 0.0438) tumors. In contrast, no significant difference was noted among age or race groups or among stage II and stage IV tumors.

A multivariate logistic model was created to assess the relationship between covariates independent of possible confounders and KLF4. All possible two-way interaction terms were tested, and in concert with results from the univariate analysis, none were found to significantly contribute to the model. As such, the final model accounts for age at diagnosis, gender, race, and stage as possible predictors for KLF4 status (Table 4). Among all possible predictors, only stage III, as compared with stage I in the odds ratio, is statistically significant (P = 0.0211). However, a trend, although not significant, showing a decreased odds ratio at higher stages with reference to stage I cancer is evident.

A Kaplan-Meier curve representing univariate survival analysis is shown in Fig. 2. For all included participants, overall survival was significantly better for individuals that retained KLF4 expression as compared with those that did not (P = 0.0437). Stage-specific survival curves show no independent differences in survival between individuals with KLF4-positive and -negative tumors (not shown). Kaplan-Meier curves were also constructed to determine difference in recurrence between KLF4-positive and KLF4-negative tumors. Figure 3 shows overall time to recurrence or disease-free survival is significantly greater in KLF4-positive patients (P = 0.0001). Stage-specific curves (not shown) show only a significant difference among stage III tumors (P = 0.0046), in which KLF4-positive tumors have significantly improved disease-free survival.

Finally, a Cox model for survival was created that included stage, age, race, gender, and all possible two-way interactions among the factors as possible confounders in addition to KLF4 status. In this model, KLF4 is a significant predicator for overall survival (P = 0.0427) and disease-free survival (P = 0.0486).

KLFs and, notably, KLF4 are important regulators of the intestinal epithelial cell homeostasis and tumorigenesis (27, 38). However, the clinical use of KLF4 as a prognostic marker in colon cancer has not been established. In this study using tissue microarrays from a large cohort of colon cancer cases, we convincingly showed that KLF4 expression level is significantly downregulated in cancer and that loss of KLF4 is an independent predictor of survival and disease recurrence.

We first showed that KLF4 is grossly reduced in histologic colon cancer sections as compared with normal colonic sections. This is consistent with the brunt of data from our laboratory and others that suggests KLF4 is a putative tumor suppressor (17, 32, 39). One study showed that, in a set of 30 colorectal cancer sections, KLF4 mRNA transcripts were reduced by 50% compared with matched normal tissue (32). This reduction paralleled the reduction in p21^Waf1/Cip1, a cell cycle inhibitor gene, suggesting that the reduction in the latter may be a direct consequence of loss of KLF4 expression (32). This loss of KLF4 expression is thought in part due to the loss of heterozygosity of the KLF4 loci and promoter hypermethylation (32). In the current study, we provided corroborative evidence that the level of KLF4 protein is significantly reduced in colon cancer specimens as detected by immunohistochemistry. This result is consistent with that of a similar study showing loss of KLF4 protein expression in colorectal cancer (39), although in contrast to our study, that study failed to show a statistically significant correlation between KLF4 expression and several clinicopathologic parameters, including stage and lymph node metastasis. The reason behind this difference is unclear, although it could be due to the different sample size, statistical methodology, or criteria by which KLF4 positivity was assessed between the two studies.

In addition to colon cancer, decreased KLF4 expression has been observed in cancer of the stomach, esophagus, bladder, lung, and T-cell leukemia (40-44). In addition, KLF4 is reduced in precancerous lesions such as adenomas of the colon (45, 46). In a model of familial adenomatous polyposis syndrome, the Apc^Min mouse, the expression of Klf4 is lower in intestinal adenomas when compared with normal-appearing mucosal tissues (45). Conversely, the burden of intestinal adenomas is significantly increased when Apc^Min mice are bred with mice heterozygous for the Klf4 alleles (17). These results are consistent with a tumor suppressive function of KLF4 in the intestine. However, whether KLF4 serves as a prognostic marker in colon cancer has not been determined.

To evaluate which clinical covariates are associated with expression of KLF4, we did a univariate analysis as well as building a multivariate logistic model. In the univariate analysis, gender, but not age or race, was associated with KLF4 status in that men had higher proportions of KLF4-postivite tumors than women (Table 3). A similar trend, although not statistically significant, was noted for gender in a multivariate analysis taking into consideration all available covariates (Table 4). The reason for the potential association of KLF4 expression with gender is unclear. KLF4 expression has previously been associated with prognosis of early-stage breast cancer but not with the status of either estrogen or progesterone receptors (37). The estrogen and progesterone status in the colon cancer of the current study was not assessed.

Also in the multivariate model, a trend was noted in which higher stages of diseases had decreased odds of KLF4 expression relative to stage I colon cancer (Table 4). This suggests a possible role of KLF4 as a marker of stage prognosis in future studies, if it were in fact associated with tumor progression. This would be consistent with observations from the RKO colon cancer cell line, in which KLF4 loss due to hemizygous deletion of KLF4 contributes to aggression, which is suppressed with re-expression of KLF4 (47). Perhaps the true potential of KLF4 as a prognostic marker may be seen when combined with other biomarkers of disease, providing a refined model with which to evaluate patients.

The most important aspect of the current study is that KLF4 represents a significant predictor of overall survival and disease-free survival by Kaplan-Meir analysis (Figs. 2 and 3, respectively). This is different from the only previously published study on KLF4 as a prognostic factor, in which a trend, although not significant, was noted for improved unadjusted survival only among stage III cancer patients (20). In our Kaplan-Meier analysis, we were not able to find significant survival benefits in any particular stage of disease, with the exception of disease-free survival in stage III disease. However, given the much larger combined sample size, we were able to show a significant benefit in survival due to KLF4. The association of KLF4 expression with disease-free survival of stage III cancer patients is consistent with the trend noted in the previous study (20) and is consistent with the results of multivariate analysis showing a significant association of KLF4 expression with stage III relative to stage I disease (Table 4). Although speculative, these data suggest that KLF4 has a particularly important function in tumor suppression during the transition from local to metastatic colon cancer. Most importantly, inclusion of age, gender, race, and stage, as well as appropriate extension and interaction terms, indicated that KLF4 status is a significant predicator for survival and recurrence. Therefore, we conclude that KLF4 is an independent predictor of survival in colon cancer.

Our data served to further corroborate previous studies indicating that KLF4 has a role in tumor suppression in colon cancer. Importantly, our study established that KLF4 as a novel prognostic marker for colon cancer survival, particularly in stage III disease, and suggests that KLF4 could become a potential clinical marker for risk stratification and response to therapy. However, because of the retrospective nature of this investigation, reproduction of the results in prospective studies, ideally with techniques more suitable for clinical laboratories, is probably warranted to assess whether KLF4 could be an effective clinical tool in colon cancer.

No potential conflicts of interest were disclosed.

Grant Support: NIH CA84197, DK52230, and DK64399 (V.W. Yang); RR25010 (N.V. Patel); and CA130308 (A.M. Ghaleb)

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.