Purpose: The Krüppel-like transcription factor KLF4/GKLF induces both malignant transformation and a slow-growth phenotype in vitro. Although KLF4 expression is increased in most cases of breast cancer, it was unknown whether these cases represent a distinct subtype with a different clinical outcome.
Experimental Design: We examined expression of KLF4 by immunostaining 146 cases of human primary infiltrating ductal carcinoma of the breast. Staining patterns were correlated with clinical outcome and with established prognostic factors.
Results: Subcellular localization exhibited case-to-case variation. Tumors with high nuclear staining and low cytoplasmic staining were termed type 1. For patients with early-stage disease (i.e., stage I or IIA), type 1 staining was associated with eventual death because of breast cancer (hazard ratio, 2.8; 95% confidence interval, 1.23–6.58; P = 0.011). The association was stronger in patients with early-stage cancer and small primary tumors (i.e., ≤2.0 cm in diameter; hazard ratio, 4.3; 95% confidence interval, 1.75–10.62; P < 0.001). For patients with early-stage disease, multivariate analysis indicated that type 1 staining was independently associated with outcome (adjusted hazard ratio 2.6; 95% confidence interval, 1.10–6.05; P = 0.029). Type 1 staining was also associated with high histological grade (P = 0.032), increased expression of Ki67 (P = 0.016), and reduced expression of BCL2 (P = 0.032). In vitro, KLF4 was localized within the nucleus of transformed RK3E epithelial cells, consistent with a nuclear function of this transcription factor during induction of malignant transformation.
Conclusions: The results suggest that localization of KLF4 in the nucleus of breast cancer cells is a prognostic factor and identify KLF4 as a marker of an aggressive phenotype in early-stage infiltrating ductal carcinoma.
Appropriate regulation of cellular proliferation and differentiation is important for embryonic development, tissue morphogenesis, and homeostasis and requires integration of diverse extracellular signals that program transcription and DNA replication within the nucleus. Within epithelial cells, pathways that transforming these signals are frequent targets of genetic alteration during progression to carcinoma (1 , 2) . A shared property of the transforming growth factor (TGF)-β, WNT, and Sonic Hedgehog pathways is the regulation of transcription factors by tethering within the cytoplasm. In normal tissues, these secreted morphogens signal through distinct receptors to induce translocation of latent transcription factors from the cytoplasm to the nucleus, resulting in activation of specific target genes such as cyclin D1 and c-MYC (3 , 4) . Genetic alterations found in cancers can mimic signaling by a morphogen, resulting in inappropriate nuclear accumulation of oncogene-encoded transcription factors such as GLI, β-catenin, or γ-catenin/plakoglobin (5, 6, 7, 8, 9, 10) .
As with these transcription factors, a zinc finger protein of the Krüppel family termed KLF4 (or GKLF) exhibits potent transforming activity when expressed in cultured RK3E epithelial cells (11) . A functional screen identified multiple independent cDNA clones of KLF4 and c-MYC, implicating these molecules as major transforming activities in the human cancer cell lines examined. As shown by mRNA in situ hybridization analysis of tumor and adjacent normal tissue, KLF4 expression is up-regulated in a majority of infiltrating ductal carcinomas of the breast and in virtually all squamous cell carcinomas of the oropharynx (11 , 12) . In contrast, expression in carcinomas of the prostate was unchanged or was down-regulated. Genetic studies in mice revealed that Klf4 is dispensable for development in utero but critical for differentiation of skin and gut epithelium and for function of the skin as a water barrier in postnatal animals (13) . Consistent with these results, in mouse and humans KLF4 is normally expressed in nondividing, differentiated epithelial cells (11 , 14 , 15) . The apparent deregulation of this potent transforming activity in tumors suggested that KLF4 may contribute to tumor progression when inappropriately expressed in proliferation-competent cell types such as stem cells or transit amplifying cells within the epithelium (11 , 12) .
Another property of KLF4 is the slow-growth phenotype induced in cultured cells, one of the early functional observations made for this zinc finger protein (15, 16, 17, 18) . Candidate mechanisms related to the slow-growth include transcriptional up-regulation of WAF1/CIP1 and repression of CYCLIN D1 or ornithine decarboxylase. We observed the same slow-growth phenotype (11) , as well as induction of Waf1/Cip1, in RK3E cells that were transformed by KLF4.6 These malignant cells proliferate with a doubling time of 28 h, longer than for parental RK3E cells (12 h), for c-MYC- or GLI-transformed cells (18 h each), or for RAS-transformed cells (9 h). In summary, KLF4 induces a complex phenotype that includes a reduced proliferation rate, induction of a spindled, mesenchymal-like cell morphology, loss-of-contact inhibition, and malignant transformation as indicated by tumorigenicity in mice (11) .
KLF4 can function within the nucleus to regulate expression of specific target genes that contain promoter-proximal KLF4 response elements similar to the binding sequence 5′-G/A-G/A-G-G-C/T-G-C/T-3′ (19) . However, few functional studies have demonstrated a requirement for nuclear localization because the nuclear localization signal is partially contained within the zinc fingers of the DNA binding domain (20, 21, 22) . We previously used a mouse monoclonal antibody to human KLF4 (anti-KLF4) to analyze expression in multiple normal tissues and cancers (12) . Specificity of the antibody was supported by analysis of the tissue sections by mRNA in situ hybridization and by virtue of cell-type and cancer-type specific patterns of expression in vivo. We observed that the subcellular localization is mixed, with prominent expression in both the nucleus and cytoplasm. More recently, similar results were obtained using a different anti-KLF4 antibody (23) .
In the current study, we show that KLF4 exhibits distinct patterns of subcellular localization in different primary breast tumors. Preferential nuclear localization of KLF4 in surgically excised tumors of patients with early-stage disease correlated with eventual death because of breast cancer and with other parameters previously associated with increased risk of recurrence or death. Small primary tumors with preferential nuclear localization of KLF4 were much more likely to lead to death from breast cancer and may be distinct with respect to mechanisms of pathogenesis, mechanisms of metastasis, or response to specific therapies.
MATERIALS AND METHODS
Tissue Procurement and Study Patients.
Archived paraffin blocks of formalin-fixed tissue corresponding to 146 cases of infiltrating ductal carcinoma of the breast, diagnosed between 1983 and 1995, were obtained from the Department of Pathology at the University of Alabama at Birmingham (Table 1)⇓ . Approval for the study was obtained from the University of Alabama at Birmingham Institutional Review Board. Tumor size, axillary lymph node status, stage at diagnosis, treatment, and demographic information were obtained by review of pathology reports and clinical charts maintained in the Departments of Pathology and Surgery, respectively. All charts were independently reviewed by each of two coauthors (M. Chakravarthy, J. M. Ruppert), and discordances were reconciled. Follow-up information and cause of death was obtained from the University of Alabama at Birmingham Hospital Tumor Registry, which maintains an updated database through annual contact with primary care physicians. Stage at diagnosis was based upon tumor size (T), axillary lymph node status (N), and the presence of distant metastases (M) as described by the American Joint Committee on Cancer (Table 1⇓ ; Ref. 24 ).
Tissues were fixed in neutral-buffered formalin and embedded in paraffin. To avoid antigen decay, sections were cut to 5-μm thickness 1 day before immunostaining. Sections were attached to the slide by heating in a 60°C oven for 1 h. Deparaffinized tissue sections were treated for 5 min in a 3% aqueous solution of hydrogen peroxide, blocked in PBS with 3% goat serum (Sigma) for 1 h at room temperature, and then incubated for 1 h at room temperature with anti-KLF4 monoclonal antibody IE5 at 1.0 μg/ml in binding buffer (PBS containing 1% bovine serum albumin, 1 mm EDTA, and 0.01% sodium azide; Ref. 12 ). Anti-KLF4 was stored in aliquots at −85°C, and was stable through multiple freeze/thaw cycles. Activity is lost within weeks when stored at 4°C. Slides were washed in 50 mm Tris-HCl (pH 7.6), 150 mm NaCl, 0.01% (v/v) Triton X-100. Immunodetection was performed using a biotinylated secondary antibody, streptavidin-horseradish peroxidase (Signet Pathology Systems), and the chromogenic substrate diaminobenzidine (BioGenex). Sections were counterstained with Harris’ hematoxylin (Surgipath). As controls, histological sections of each case were processed without the addition of primary antibody for each antigen retrieval method along with positive/negative, multi-tissue control sections.
For the detection of estrogen receptor (ER), progesterone receptor (PR), p27KIP1, and Ki67, histological sections were subjected to low temperature antigen retrieval with enzymatic pretreatment. This consisted of predigestion in 0.1% trypsin (type II-S from porcine pancreas; Sigma) in PBS for 15 min at 37°C, followed by incubation in 10 mm citrate buffer (pH 6) for 2 h at 80°C (25) . The antibodies used were the anti-ER mouse monoclonal antibody, clone ER88 (0.33 mg/ml total protein; BioGenex) at 1:30 dilution, the anti-PR receptor mouse monoclonal antibody, clone PR88 (0.33 mg/ml total protein; BioGenex) at 1:30 dilution, anti-Ki67 mouse monoclonal antibody, clone MIB-1 (0.37 mg/ml total protein; BioGenex) at 1:30 dilution, and the anti-p27KIP1 mouse monoclonal antibody, clone 1B4 (8.0 μg/ml IgG; Novocastra Laboratories Ltd.) at 1:30 dilution. Immunostaining for BCL2 was preceded by incubation of histological sections in boiling 10 mm sodium citrate buffer (pH 6.0) for 10 min in a microwave oven (25) . Anti-BCL2 (clone 124; Genosys Biotechnologies, Inc.) was used at 12.5 μg/ml. Anti-ERBB2 (clone 3B5; Oncogene Research Products) was used at 0.25 μg/ml. Anti-p53 (clone BP53.12; Oncogene Research Products) was used at 0.25 μg/ml.
The intensity of immunostaining of individual cells was scored on a scale of 0 (no staining) to 4 (strongest intensity), and the percentage of cells with staining at each intensity was estimated. For ERBB2, only membranous staining was assessed. The proportion of cells at each intensity was multiplied by the corresponding intensity value, and these products were added to obtain an immunostaining score (immunoscore) ranging from 0 to 4 (26 , 27) . All slides were examined and scored independently by two investigators (A. R. Frost, D. C. Chhieng, or J. M. Ruppert), large discordances were reconciled by reexamination of the slide, and the scores were then averaged.
The Nottingham modification of the Bloom and Richardson histological grading system was used to categorize carcinomas as high grade, corresponding to a total score of 8 or 9, or low-to-moderate grade (referred to as low), corresponding to a total score of <8 (28) .
Median immunoscores were used to group cases with high or low expression levels of KLF4, Ki67, BCL2, ERBB2, or P27KIP1. For ER, PR, and p53, tumors with >10% positive cells were scored as positive. Associations of clinical, pathological, and demographic factors with KLF4 staining patterns were evaluated using the Mantel-Haenszel χ2 test or, where appropriate, Fisher’s exact test (29) . Kaplan-Meier methods were used to compare overall survival rates and significance was assessed using the log-rank test (30) . Survival time was defined as the interval from the date of diagnosis to the date of death. Patients who were alive at the last date of contact, died from unknown causes or died from causes other than breast cancer were censored at the date of last contact.
Multivariate Cox proportional hazards models were performed to evaluate the effect of KLF4 staining patterns on survival while controlling for the effects of extraneous factors (31) . The analysis was performed using a stepwise selection technique. Included in the full model were KLF4, race, chemotherapy, stage, histological grade, ER, and PR. Significance level to stay in the model was set at 0.050. In addition, interaction terms were included in the model to evaluate the multiplicative effect of KLF4 staining patterns and various clinical or pathological factors on survival. All significance tests were two-sided with α = 0.05.
Plasmid Construction and Analysis of in Vitro Transforming Activity.
The human KLF4 cDNA was modified at the NH2 terminus with the hemagglutinin (HA) epitope (32) and cloned into the Moloney murine leukemia virus vector pLJD (obtained from L. T. Chow, University of Alabama at Birmingham). Cell culture, retroviral transduction of RK3E cells (33) , and assay of transforming activity were performed as described previously (11) . For focus assays, transduced cells were maintained for 4 weeks in nonselective growth media. For colony morphology assays, transduced cells were selected in 400 μg/ml G418, and colony morphology was scored 4 weeks later. A population derived from >1000 independently transduced cells was passaged in selective medium and then assayed for expression of HA-KLF4 by immunofluorescence.
Subcellular Localization of Epitope-Tagged KLF4.
The HA-KLF4 cDNA was inserted into pRK5 adjacent to the cytomegalovirus immediate early promoter-enhancer. HEK293 embryonic kidney epithelial cells were plated on poly-l-lysine coated coverslips, grown to 50% confluence, then transfected using the lipid reagent transIT-LT1 (Mirus). To enable identification of transfected cells, a green fluorescent protein expression vector (pEGFP; Clontech) was included in the transfection mixture. Cells were fixed in 4% paraformaldehyde in PBS for 10 min at room temperature, treated with 0.5% Triton X-100 in PBS for 10 min at 4°C, and then blocked in 50% (v/v) goat serum in PBS (blocking solution). Anti-HA monoclonal antibody 12CA5 (Roche) was used at 4.0 μg/ml in blocking solution for 45 min in a humidified chamber. Bound antibody was detected using goat antimouse IgG conjugated to Alexa Fluor 594 (Molecular Probes). Where indicated, the cytoplasm was stained using Alexa Fluor 488 phalloidin (Molecular Probes). Nuclei were stained using 4′,6-diamidino-2-phenylindole at 0.3 μm in PBS, rinsed briefly, mounted using Prolong Antifade medium (Molecular Probes), and then stored in the dark at −20°C for subsequent examination.
RK3E epithelial cells stably transduced with pLJD-HA-KLF4 (see above) were analyzed for expression of the transgene in similar fashion. Fluorescence of the secondary antibody, cotransfected green fluorescent protein, phalloidin, or 4′,6-diamidino-2-phenylindole were visualized using an Axioplan 2 Imaging microscope equipped with an external filter wheel (Zeiss). Black and white images were collected using an AxioCam HRc digital camera, and the pseudo-colored images were merged using Axiovision software (version 3.1).
Characteristics of the Patients.
We previously reported that expression of KLF4 is up-regulated in most cases of infiltrating ductal carcinoma of the breast (12) . To determine whether expression of KLF4 is associated with other properties of breast tumors, we identified 146 cases of infiltrating ductal carcinoma that were well characterized for clinical and pathological parameters, including surgical management, stage at diagnosis, histological grade, postoperative therapy, and cause of death (Table 1)⇓ . A total of 134 of the 146 patients (92%) underwent axillary lymph node dissection with at least five lymph nodes sampled, and ≥10 lymph nodes were sampled for 116 patients (79%). The median follow-up from the time of diagnosis was 7.1 years. Patients with early-stage disease (i.e., stages I and IIA) exhibited a 5-year disease specific survival rate of 87% (e.g., see below, Fig. 2B⇓ ), similar to that observed in larger studies (34 , 35) . Likewise, patients with stage IIB, stage III, or stage IV disease exhibited 5-year survival rates (75, 37, or 20%, respectively) similar to rates observed for larger groups. We analyzed expression of KLF4 and other prognostic or predictive factors, including the steroid hormone receptors ER and PR, the receptor tyrosine kinase ERBB2, the proliferation marker Ki67, the tumor suppressor p53, and two markers associated with favorable clinical outcome, BCL2 and the cyclin-dependent kinase inhibitor p27KIP1 (36) .
Distinct Patterns of KLF4 Subcellular Localization in Breast Tumors.
Overall, expression was detected in >90% of cases. Primary tumors varied greatly in their relative staining of the nucleus and cytoplasm (Fig. 1A)⇓ . The pattern of subcellular localization within individual tumors was quite uniform across a histological section, and was similar in invasive and in situ components within the same section (Fig. 1A⇓ and data not shown). Scatterplot analysis demonstrated the spectrum of staining patterns observed in these tumors (Fig. 1B)⇓ . On the basis of the four quadrants defined by the median immunostaining scores, tumors were classified as type 1, 2, 3, or 4. Type 1 tumors exhibit higher than median nuclear staining and lower than median cytoplasmic staining. Type 2 tumors have lower staining in each compartment. Type 3 tumors have predominantly cytoplasmic staining, and Type 4 tumors have increased staining in each compartment.
As KLF4 is likely to function in the nucleus rather than in the cytoplasm, we evaluated the impact of preferential nuclear expression of KLF4 on survival (Fig. 2A)⇓ . We compared type 1 tumors to all other tumors combined (referred to as type 2–4). Although type 1 tumors appeared to be more often associated with death from breast cancer, this trend was not statistically significant (log-rank test, P = 0.090). No significant differences were obtained by comparison of type 3 tumors versus all others (P = 0.227). Likewise, tumors with high versus low nuclear expression exhibited similar outcomes (i.e., using the median score as cutoff; P = 0.601), as did tumors with high versus low cytoplasmic expression (P = 0.157).
KLF4 Expression in Small Primary Tumors.
The trend observed for all cases combined was more pronounced for patients who were diagnosed with early-stage cancer (i.e., stages I and IIA; Fig. 2B⇓ ; P = 0.011). Tabulation of outcome by stage and KLF4 staining pattern suggested an important role for size of the primary tumor in the association of KLF4 with clinical outcome (Table 2)⇓ . Indeed, all 9 of the deaths among patients with type 1 staining and early-stage disease occurred in the setting of a small primary tumor (i.e., ≤2.0 cm in diameter, indicated as T1). For the 7 patients with T1N0M0 (i.e., stage I) disease and type 1 staining, all but 2 succumbed to breast cancer. In contrast, none of the 7 patients who had T2N0M0 (stage IIA) disease with type 1 staining died from breast cancer (P = 0.010). The median follow-up time for these groups was 4.30 years (stage I) and 8.01 years (stage IIA). Thus, patients with early-stage disease and small primary tumors were much more likely to die from breast cancer when the KLF4 staining pattern was type 1 (Fig. 2C⇓ ; P < 0.001).
We used scatterplot analysis to examine the role of tumor size in the association of type 1 staining and death due to breast cancer (Fig. 3A)⇓ . For patients with type 1 tumors that were ≤2.0 cm, 11 of 15 (73%) died from breast cancer. For patients with type 1 tumors in the range of 2.01–3.00 cm in size, only 2 of 9 (22%) died from breast cancer (P = 0.033). Thus, type 1 staining identifies a paradoxical subset of cancers in which larger tumor size is associated with a lower rate of death from breast cancer. No such effect was observed for patients with type 2–4 staining (Fig. 3B)⇓ . This analysis suggests that any increased risk associated with type 1 staining may be limited to tumors ≤2.0 cm in diameter.
Despite the specific association between type 1 staining and outcome in small tumors, there was no difference in the overall staining pattern in small versus large tumors (Figs. 4, A and B)⇓ . As with the distribution of scores in the two-dimensional plot (Fig. 4, A and B⇓ , left panels), the median immunoscores for each subcellular compartment were very similar (for small tumors: cytoplasm = 1.30, nucleus = 0.43; for large tumors, cytoplasm = 1.25, nucleus = 0.45). Survival analysis of all patients in the study, regardless of stage at diagnosis, demonstrated the specific association of type 1 staining and outcome in small tumors (Fig. 4A⇓ , right panel; P < 0.001). For large tumors, there was no trend toward a worse outcome in patients with type 1 staining (Fig. 4B⇓ , right panel; P = 0.398). As a control for the quality of the outcome data for each of the two subgroups, Kaplan-Meier analysis revealed that high histological grade was associated with death caused by breast cancer in both the small tumor (P = 0.002) and large tumor subgroups (P = 0.026; data not shown).
Association of the Type 1 Staining Pattern with Other Parameters.
These results suggest that T1-type 1 tumors are more likely to recur as distant metastatic lesions, often several years later, leading to eventual death from breast cancer. To better characterize this potentially important subset of tumors, we determined whether other known prognostic factors are associated with type 1 staining (Table 3)⇓ . Associations were tested for small tumors alone, for large tumors alone, and for all tumors combined. Compared with patients with T1-type 2–4 lesions, patients with T1-type 1 lesions did not exhibit a significant difference in stage at diagnosis (P = 0.171). However, high histological grade was associated with type 1 staining in small tumors. A high grade was observed in 9 of 15 T1-type 1 cancers (60%). In contrast, only 14 of 49 (29%) T1-type 2–4 tumors exhibited high grade (P = 0.026). For large tumors, no association of histological grade with type 1 staining was observed (P = 0.252). For small and large tumors combined, histological grade was more often high in type 1 than in type 2–4 tumors (P = 0.032).
Two immunohistochemical markers exhibited significantly different expression in type 1 versus type 2–4 tumors (all tumors combined). Expression of the proliferation marker Ki67 was more often high for tumors with type 1 staining patterns (P = 0.016). BCL2, for which higher expression was previously associated with a more favorable prognosis (37) , was often low in type 1 tumors (P = 0.032). The observed associations additionally define the properties of T1-type 1 tumors. In summary, this group of clinically aggressive tumors is more likely to exhibit high histological grade, increased proliferation, and reduced expression of the favorable prognostic marker BCL2.
Univariate and Multivariate Analysis.
We determined the unadjusted hazard ratio associated with type 1 staining for three groups of patients: all patients regardless of stage at diagnosis; patients with early-stage disease at diagnosis; and patients diagnosed with small primary tumors in the setting of early-stage disease (Table 4)⇓ . Statistical significance was indicated when the 95% confidence interval (CI) of the hazard ratio excluded 1.00. For all patients (n = 146), factors significantly associated with a poorer survival included higher stage at diagnosis (hazard ratio, 5.5; 95% CI, 2.88–10.64), positive axillary lymph node status (hazard ratio 3.2; 95% CI, 1.65–6.22), high histological grade (hazard ratio, 2.8; 95% CI, 1.53–5.24), African-American race (hazard ratio, 2.3; 95% CI, 1.29–4.12), and reduced expression of BCL2 (hazard ratio, 0.4; 95% CI, 0.23–0.83). For patients with early stage cancer (n = 97), only type 1 staining exhibited a significant association with poor outcome (hazard ratio, 2.8; 95% CI, 1.23–6.58). For small tumors in the setting of early-stage cancer (n = 60), type 1 staining (hazard ratio, 4.3; 95% CI, 1.75–10.62), high histological grade (hazard ratio, 3.3; 95% CI, 1.32–8.28), and African-American race (hazard ratio, 2.6; 95% CI, 1.03–6.45) were each significant. In this smaller group of patients, other parameters previously associated with outcome in breast cancer exhibited the expected trend but did not reach statistical significance (e.g., axillary lymph node status, stage, age, BCL2, P27KIP1, ER, and PR).
Multivariate analysis indicated that type 1 staining is independently associated with outcome in patients with early-stage disease. For all patients with stage I or stage IIA disease, KLF4 was the only significant variable remaining, with an adjusted hazard ratio of 2.6 (95% CI, 1.10–6.05; P = 0.029). The failure of other known risk factors such as nodal status, stage, or ER status to exhibit significance is attributed to the small sample size of this initial study and to the exclusion of patients with later stage disease from the model.
Analysis of KLF4 Subcellular Localization in Vitro.
These results indicate that subcellular localization of KLF4 exhibits case-to-case differences in breast cancer and that preferential nuclear localization may be associated with an adverse outcome. Whether KLF4 functions within the nucleus or cytoplasm to induce transformation in vitro is unknown. We therefore asked whether KLF4 exhibits localization to the nucleus in transformed RK3E cells in vitro as observed in the more aggressive subset of early-stage breast cancer (T1-type 1 tumors).
We added a HA epitope to the NH2 terminus of human wild-type KLF4 and confirmed the sequence of the final expression construct. To determine whether HA-KLF4 retains similar transforming activity as wild-type, we generated retroviral supernatants as described previously (11) . Within 3 weeks after transduction of RK3E epithelial cells, wild-type KLF4 and HA-KLF4 each induced numerous transformed foci upon a background monolayer of contact-inhibited RK3E cells (data not shown). Cells transduced by the empty vector served as a negative control. As an additional measure of transforming activity, we used the colony morphology assay (11) . We examined the morphology of cells within established colonies of RK3E cells that survived retroviral transduction and culture in selective growth medium. Unlike the vector control cells, HA-KLF4 cells and KLF4 cells were spindled, highly refractile, and formed dense colonies. These results indicate that HA-KLF4 retains the transforming activity of wild-type human KLF4.
To test whether the epitope enables identification of HA-KLF4, we first examined human embryonic kidney cells, termed HEK293, 48 h after transfection of the expression vector pRK5-HA-KLF4 or a vector control (Fig. 5A)⇓ . HA-KLF4 exhibited two frequent patterns of subcellular localization. In a subset of cells, expression was localized almost entirely within the nucleus. In another subset, representing approximately one-half of all positive cells, nuclear staining was associated with a prominent rim of perinuclear staining (Fig. 5A⇓ , middle panel, inset). KLF4 expression in the cytoplasm was rarely observed to extend throughout the full extent of the cytoplasm, as shown by coexpression of a green fluorescent protein control (Fig. 5A⇓ , right panel). Similar results were obtained by transfection of MCF7 cells, although the perinuclear rim in these cells was somewhat thinner (data not shown). These results provide evidence for a cytosolic anchoring mechanism that may localize KLF4 to the perinuclear region in cultured epithelial cells, analogous to mechanisms that regulate other transcription factor oncogenes.
To determine where KLF4 is located within transformed RK3E cells, stably transduced HA-KLF4 cells and vector control cells were passaged in culture, stained with anti-HA, and then examined by microscopy (Fig. 5B)⇓ . Similarly as for HEK293 cells, the vast majority of expression was localized to the nucleus of transformed RK3E cells. In these cells, perinuclear or cytoplasmic staining was rarely detected (i.e., in <5 of 100 cells examined). These results are consistent with a nuclear function of KLF4 during induction of malignant transformation in vitro.
Expression of KLF4 mRNA and protein was previously analyzed by immunostaining and by mRNA in situ hybridization analysis of breast cancers (12) . As compared with adjacent, uninvolved epithelium, expression of this potent transforming activity is increased in the vast majority of breast tumors. In contrast, other tumor types such as colorectal or prostatic carcinoma do not exhibit increased expression (12) . In the current study, we examined an additional 146 cases of breast cancer to determine whether expression of KLF4 is associated with specific clinical, pathological, or molecular features. Although overall expression of KLF4 was not associated with clinical outcome, we found that breast tumors exhibit marked case-to-case variation in subcellular localization. Preferential nuclear localization (i.e., type 1 staining) was associated with clinical outcome in early-stage cancer, and this association was only apparent when the primary tumor size was small (i.e., ≤2 cm in diameter, designated as T1).
Although type 1 staining was strongly associated with a poor outcome in this initial study, the high-risk group with T1-type 1 tumors included only 15 patients. In addition, the association between type 1 staining and outcome was not significant for all patients combined but only for the subset of patients with early-stage disease. Although KLF4 appeared to exhibit both sensitivity and specificity as a marker of poor outcome in these cases, a better appreciation of the hazard associated with type 1 staining will require analysis of larger groups of patients. Likewise, larger studies will reveal whether factors such as Ki67 expression, BCL2 expression, or stage at diagnosis may be associated with type 1 staining in small tumors.
Despite the limited study size, several lines of evidence support the association with clinical outcome. Importantly, our observation of poor outcome in T1-type 1 tumors resulted from testing of a hypothesis rather than from analysis of an arbitrary subset. We examined the potential association between outcome and type 1 staining because of the known role of KLF4 as a nuclear transcription factor and transforming oncogene. Patients with early-stage disease represent a subset in which prognostic markers may well exhibit a stronger association with outcome. For example, increased expression of BAG1, a regulator of cell proliferation and apoptosis, is associated with improved survival in patients with early-stage breast cancer (38) . In that study, multivariate analysis revealed a stronger association in the node-negative subgroup (P = 0.002) than for the node-positive subgroup (P > 0.050) or for all stage I–II patients combined (P = 0.008). Keyomarsi et al. (35) reported the association between low molecular weight isoforms of cyclin E and death from breast cancer. In that study, the association was strongest in stage I and weakest in stage IV. In the current study, high histological grade was significantly associated with a poor outcome for both the small and large tumor subgroups, but the association was stronger for patients with small tumors. Furthermore, as knowledge of cancer biology and therapy improves, early-stage patients stand to benefit greatly from molecular staging. Thus, a careful analysis of this subgroup is justified.
In our study, type 1 staining was associated not only with death because of breast cancer, but also with high histological grade in the primary tumor, a well-recognized correlate of survival (39) . The association with grade, as with the association with clinical outcome, was restricted to small tumors. As for clinical outcome and histological grade, the association between type 1 staining and reduced expression of BCL2 was stronger in small tumors.
The current study does not distinguish between active and passive roles for KLF4 in the aggressive phenotype of these early-stage tumors. Additional insight may come from analysis of KLF4-regulated genes (target genes). Increased expression of target genes in type 1 tumors versus type 4 tumors would be consistent with an active role because type 4 tumors likewise exhibit elevated nuclear expression of KLF4 but did not exhibit an aggressive phenotype. On the other hand, if preferential nuclear localization was a consequence of signaling through upstream regulators of KLF4, such signaling might promote the aggressive phenotype independently of KLF4 through parallel effector pathways. In this case, KLF4 would have only a passive role, and transcriptional targets might be similarly expressed in type 1 and type 4 tumors. We are currently using a conditional allele in combination with microarrays to identify target genes of KLF4.
Few markers associated with clinical outcome in breast cancer have been found to exhibit tumor size dependence. Although several possibilities may account for the nonassociation of type 1 staining and clinical outcome in large tumors, this result is particularly interesting given recent insights into KLF4 function as a transforming oncogene that induces a slow-growth phenotype (11) . Potentially, the dual roles of KLF4 as both a transforming activity and an inhibitor of cell cycle progression could dissociate malignant potential from tumor size, leading to an aggressive or metastatic phenotype in tumors with a smaller diameter. Alternative explanations include the possibility that, in large tumors, the localization or transcriptional activity of KLF4 could be influenced by other signaling pathways or by the tumor microenvironment, thus confounding the associations observed in smaller tumors.
A complementary in vitro analysis revealed that KLF4 is virtually confined to the nucleus in the setting of malignant transformation of RK3E cells. This result supports the notion that KLF4 may function within the nucleus as a transcription factor to induce transformation. In addition to the nuclear expression observed in vitro, a prominent perinuclear component of KLF4 was detected following transient transfection. This preliminary observation warrants additional investigation because many transcription factors implicated in neoplasia exhibit regulated subcellular localization. The perinuclear staining we observed is consistent with similar regulation of KLF4, perhaps through tethering to a cytoplasmic protein. Interestingly, human KLF4 contains a putative SH3 domain binding site near the NH2 terminus that could mediate such an interaction (40) .
Although these results suggest that nuclear import and/or export of KLF4 may be regulated, upstream signaling pathways that mediate such regulation have not been identified. One interesting candidate is the TGF-β pathway. In vascular smooth muscle cells, TGF-β or other TGF-β-superfamily members induce the expression of smaller KLF4 isoforms and induce binding of KLF4 to TGF-β control elements found in the regulatory region of marker genes associated with smooth muscle differentiation (23 , 41) . The perinuclear localization of KLF4 that we observed in vitro provides a simple assay that may facilitate identification of such regulatory pathways. The possibility that this perinuclear staining is related to the more diffuse cytoplasmic staining observed in breast tumors also warrants further study.
The potential use of prognostic markers in the diagnosis and treatment of breast cancer has been reviewed (36 , 42, 43, 44) . A benefit of stratification of patients into distinct risk groups by molecular staging is that the use of known prognostic factors or the effectiveness of specific therapies may be enhanced in one of the subsets. For example, tumor size and stage at diagnosis could be enhanced as prognostic factors after identification and segregation of T1-type 1 cases, thus allowing for more effective selection of therapies and more efficient design of clinical trials. Although the current study associates KLF4 nuclear localization and clinical outcome in breast cancer, there is currently little evidence of a functional role for KLF4 in tumor progression. Currently, we are using short-term induction of KLF4 expression in vitro to identify transcriptional target genes and analyzing expression of these putative target genes within type 1 breast tumors and in a novel mouse model of KLF4-induced neoplasia. These studies may lead to a better understanding of signaling pathways that function upstream or downstream of KLF4 and may indicate whether KLF4 plays an important role in human tumor initiation or progression.
We thank Cynthia Moore for assistance in immunostaining of tissue sections, Laura Gallitz for secretarial assistance, Glenda Williamson and Debra Martin for help with clinical data, and John Carpenter and Sam Beenken for critical reading of the manuscript.
Grant support: NIH Grants RO1 CA65686 (J. Ruppert), P50 CA89019 (K. Bland, J. Ruppert), T32 CA91078 (K. Bland), T32 DK07488 (T. Fitzgerald), and by a gift to the Comprehensive Cancer Center from the Avon Foundation.
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.
Requests for reprints: J. Michael Ruppert, Department of Medicine, Room 570 WTI, University of Alabama at Birmingham School of Medicine, Birmingham, AL 35294-3300. Phone: (205) 975-0556; Fax: (205) 934-9511; E-mail:
↵6 Induction of Waf1/Cip1 by KLF4 in RK3E cells is an unpublished observation by S. M. Lobo-Ruppert and J. M. Ruppert.
- Received October 28, 2003.
- Revision received December 12, 2003.
- Accepted December 17, 2003.