Involvement of the TGF-β and β-Catenin Pathways in Pelvic Lymph Node Metastasis in Early-Stage Cervical Cancer

Standard treatment of early-stage cervical cancer patients consists of radical hysterectomy and pelvic lymphadenectomy. For this group of patients, the presence of lymph node metastases is the most important prognostic factor (1). Early-stage cervical cancer patients with negative lymph nodes have a 5-year survival of 90% versus only 65% in patients with lymph node metastases (2). Patients with lymph node metastases are therefore treated with adjuvant (chemo)radiation. However, the combination of surgery and (chemo)radiation is associated with severe morbidity (3). If the presence of metastatic lymph nodes could be predicted prior to treatment, primary chemoradiation could be considered, which is equally effective, but associated with a different treatment-related morbidity pattern.

Several histopathological characteristics such as tumor size, lymph vascular space involvement, and depth of invasion have been associated with lymph node metastases in cervical cancer but none of these is of sufficient clinical relevance (4). Furthermore, various molecular tumor markers like the expression of VEGF and p16 have been reported to be related with lymph node metastases in cervical cancer (5, 6), but presently no markers are available to predict lymph node status with high sensitivity and specificity. Non- and minimal invasive diagnostic techniques, such as sentinel lymph node biopsy are currently being explored to better identify patients with disease outside the cervix (7).

Little is known about biological pathways involved in lymph node metastasis in cervical cancer. Metastasis is a complex, multistep process involving decreased cell–cell interaction, increased cell migration, disruption of the basal membrane, intravasation into the circulation, survival of direct exposure to the immune system and extreme mechanic forces in the bloodstream, and finally extravasation and growth in metastatic sites (8). Apart from tumor-specific changes, many processes in the tumor microenvironment of the primary tumor have shown also to be important for initiation of the metastatic potential at the primary site (9).

Gene expression profiling has provided tools to identify patterns of biological differences between different tumor types, cancers with diverse clinical outcome or treatment responses (10, 11). To get insight into the mechanism of lymph node metastasis in head and neck (12), colorectal (13), and cervical cancer (14–17), gene expression profiling has been used. However, in most studies little overlap was found between differentially expressed genes, which may be due to a variety of methodological issues (18). Explanations that have been debated extensively in the literature are the use of different microarray platforms (18, 19) and the restricted number of samples used to select genes from a large pool of probes (20). Therefore, comparing gene expression profiles with gene sets that represent unique pathways may provide more insight into the mechanism of lymph node metastasis. Different pathway analysis methods have been developed, including Gene Set Enrichment Analysis (GSEA). GSEA is used to determine whether predefined gene sets available for example in the Kyoto Encyclopedia of Genes and Genomes (KEGG; ref. 21) and Biocarta data bases (http://www.biocarta.com/), show significant, concordant differences between 2 phenotypes (22). Another method has recently been developed by Bild and colleagues (23). Experimentally generated expression signatures using human primary mammary epithelial cell cultures (HMEC) that reflect the activation of various oncogenic signaling pathways (c-Myc, H-Ras, c-Src, E2F3, and β-catenin) can be used to assess the activation probability of the oncogenic pathways in individual expression profiles. Both methods have not been applied previously for differentiating between lymph node negative and positive cervical cancer patients.

The aim of this study was to identify cellular tumor pathways associated with pelvic lymph node status in patients with early-stage cervical cancer. Apart from obtaining more insights on the molecular processes of lymph node metastasis in early-stage cervical cancer, our findings might contribute to individual treatment strategies. To identify such pathways, expression array analysis was performed on a well defined series of cervical squamous cell carcinomas of patients with histologically confirmed lymph node metastases (N₊) versus patients with histological and clinically confirmed negative lymph nodes (N₀). Potential markers representing the predictive value of pathways were validated in a large consecutive series of early-stage cervical cancer patients by immunohistochemistry on tissue microarrays (TMA).

Since 1980 clinicopathological characteristics of all cervical cancer patients referred to the Department of Gynecological Oncology of the University Medical Center Groningen are prospectively collected in a database. For the present study, patients with stage IB-IIA disease, primarily treated with surgery between 1980 and 2004 were selected (n = 337). Follow-up data were collected for at least 5 years. Staging was performed according to FIGO guidelines. Primary treatment consisted of type 3 radical hysterectomy and pelvic lymph node dissection. In case of poor prognostic factors, such as lymph node metastases or positive resection margins, patients were treated with adjuvant radiotherapy or chemoradiation. From these patients paraffin-embedded, formalin-fixed primary tumor tissue was collected. All tumor tissues were histological revised and only tumor specimens with sufficient tumor cells were included in the study for construction of the TMA. In 274 cases, sufficient pretreatment paraffin-embedded tissue was available for TMA construction. Of 274 patients, 112 (41%) received adjuvant (chemo)radiation. Median follow-up time for patients on the TMA was 5.5 years (range 0.3–18.6). Since 1990, when sufficient material was available, pretreatment fresh frozen tumor tissue was stored. For the microarray experiment, we selected fresh frozen primary cervical cancer tissue, containing at least 80% tumor cells, of patients with histologically confirmed N₀ (n = 20) and of patients with N₊ (n = 19). The N₀ and N₊ groups were matched for age, FIGO stage, and histology (all squamous cell carcinoma). However, as expected the groups differed regarding presence of lymphangioinvasion (P = 0.024) and infiltration depth (P = 0.001). Patient and tumor characteristics are summarized in Table 1. In the University Medical Center Groningen, clinicopathologic and follow-up data are prospectively obtained during standard treatment and follow-up and stored in a computerized registration database. For the present study, all relevant data were retrieved from this computerized database into a separate, anonymous database. Patient identity was protected by study-specific, unique patient numbers. Codes were only known to 2 dedicated data managers, who also have daily responsibility for the larger database. In case of uncertainties with respect to clinicopathologic and follow-up data, the larger databases could only be checked through the data managers, thereby ascertaining the protection of patients' identity. Using the registration database all tissue specimens were identified by unique patient numbers and retrieved from the archives of the Department of Pathology. Therefore, according to Dutch law no further Institutional Review Board approval was required (http://www.federa.org/).

From the frozen biopsies, 4 10-μm-thick sections were cut and used for standard RNA isolation. After cutting, a 3-μm-thick section was stained with hematoxylin/eosin for histological examination and only tissues with more than 80% tumor cells were included. RNA was isolated with TRIzol reagent (Invitrogen) according to manufacturer's protocol. RNA was treated with DNAse and purified using the RNeasy mini-kit (Qiagen). The quality and quantity of the RNA was determined by Agilent Lab-on-Chip analysis. For labeling, 10 μg of total RNA was amplified by in vitro transcription using T7 RNA polymerase. Labelled RNA samples were hybridized according to a randomized design to the human genome U133 plus 2.0 microarrays (Affymetrix). The microarrays were loaded with 200 μL of hybridization cocktail solution and then placed in Genechip Hybridization Oven 640 (Affymetrix) rotating at 60 rpm at 45°C for 16 hours. After hybridization, the arrays were washed on Genechip Fluidics Station 400 (Affymetrix) and scanned using Genechip Scanner 3000 (Affymetrix) according to the manufacturers' procedure. Labeling of the RNA, quality control, the microarray hybridization, and scanning were performed by ServiceXS (Leiden, http://www.serviceXS.com) according to Affymetrix standards. Preprocessing of CEL files was performed with Affymetrix Expression Console software. Probe set expression summary was done using the Robust Multi-array Average (RMA) algorithm. Quality of the microarray data was checked using histograms, box plots, and a RNA degradation plot. Principal component analysis (PCA) was performed for controlling the quality of the hybridizations (24). The MIAME-compliant microarray data are available at http://www.ncbi.nlm.nih.gov/geo/ under accession number GSE26511.

GSEA was performed with the software package GSEA 2.0, developed by the Broad Institute of MIT and Harvard (22). Each gene was ranked according to its relative difference in expression between the N₀ and N₊ group using the Student's t statistic. Ranked expression data for all annotated 20,606 genes (in case of more than one probe per gene, the probe with the highest intensity was considered) were compared against a large collection of biological gene sets to determine whether genes both at the top or bottom of the ranked list were enriched in these functional gene sets. GSEA analysis was performed separately with a total of 155 gene sets in the KEGG (21) and 125 gene sets in the Biocarta data base. The gene sets used are available at the Molecular Signature Database (http://www.broadinstitute.org/gsea/msigdb/). Statistical enrichment was determined using an empirical phenotype-based permutation test based on 1,000 permutations. Furthermore, for each functional set the false discovery rate (FDR) and nominal P- value were calculated. P values of less than 0.05 were considered statistically significant.

In addition, oncogenic pathway activation analysis was performed using experimentally generated expression signatures from HMECs that reflect the activation of various oncogenic signaling pathways (c-Myc, H-Ras, c-Src, E2F3, and β-catenin; ref. 23). Publicly available software implementing these models (BinReg; ref. 23) was used to assess the activation probability of the oncogenic pathways in our 39 cervical tumor samples. Principal Component Analysis (PCA) was used to correct for variances due to possible unreliable activation probabilities (24–27). The oncogenic pathway activation analysis and PCA are described in detail in the supplementary data.

Class comparison was performed using the software package BRB Array Tools 3.7.0, developed by the Biometric Research Branch of the US National Cancer Institute (http://linus.nci.nih.gov/BRB-ArrayTools.html). Differentially expressed probe sets were identified using a parametric 2-sample t test (with random variance model) with a significance threshold of P < 0.001. In addition, for each probe set the FDR was determined (28). Finally, a global test was performed to assess the probability of getting the observed number of identified significant probe sets by chance, that is, under the assumption that there is no difference in expression between the N₀ and N₊ group. Differentially expressed genes were ranked according to lowest FDR and lowest parametric P value.

Immunohistochemistry of the relevant proteins (Smad2, pSmad2, Smad4, β-catenin, E-cadherin, and p120) was first performed on whole tumor slides of a small series of 20 randomly selected cervical cancer tissues (see supplementary data for more details). Only if a homogeneous staining pattern was found, immunostaining was performed on TMAs. TMAs were constructed as previously described (29). For immunohistochemistry, 3-μm sections from the TMAs were immunostained with antibodies directed against β-catenin, p120, Smad4, and pSmad2. Normal cervical epithelium was used as a positive control. Scoring was performed by 2 independent observers without knowledge of clinical data. A concordance of more than 90% was found for all stainings. The discordant cases were reviewed and scores were reassigned on consensus of opinion. Staining intensity was semiquantitatively scored as negative (0), weak positive (1), moderate positive (2), and strong positive (3). Also the percentage of positive cells was recorded. Positive Smad4 expression was defined as presence of both more than 50% moderate/strong positive nuclear and moderate/strong positive cytoplasmic staining (30). β-catenin and p120 positivity was defined as membranous staining at any intensity (1–3) in more than 50% of cells (31).

Statistical analysis was performed with SPSS 16.0 for Windows (SPSS Inc.). Associations between immunostainings and lymph node metastases were compared using logistic regression models, in which immunostainings were used as dependent factors and the clinicopathological characteristics as independent factors. P values of less than 0.05 were considered statistically significant.

GSEA using biological pathway definitions according to KEGG and Biocarta data bases revealed that 5 pathways (TGF-β, NFAT, ALK, BAD, and PAR1 pathway) were significantly enriched in the N₀ group, whereas only 1 pathway (Glycosphingolipid Biosynthesis Neo Lactoseries pathway) was enriched in the N₊ group (Table 2). The ALK pathway is defined by genes that are also present in the β-catenin and TGF-β pathways such as WNT1, CTNNB1, TGFB2, TGFR2, and SMADs (http://www.broadinstitute.org/gsea/msigdb/).

Analyzing the association between oncogenic pathways and lymph node status, using expression signatures that reflect the activation of 5 major oncogenic signaling pathways (c-Myc, H-Ras, c-Src, E2F3, and β-catenin) revealed that the activation probabilities of the oncogenic β-catenin pathway correlated highly significantly with N₊ (P = 0.001). Supplementary Table 1 shows the predicted probabilities for all 5 oncogenic pathways. A scatter plot of the activation probability of β-catenin for our 39 cervical tumor samples shows that tumor samples with a low or high probability of β-catenin activation are predominantly N₀ or N₊ tumor samples, respectively (Fig. 1).

Of these 7 pathways, only the β-catenin and TGF-β pathways, or separate components within these pathways, have been implicated in metastasis or tumor progression (32–35). Therefore, in this manuscript we decided to especially validate whether these tumor cell pathways are predictive for pelvic lymph node status in early cervical cancer.

We identified probe sets that were differentially expressed between N₀ and N₊ samples using a random-variance t test. P values, fold changes, and FDRs for all 54,675 probe sets are given in Supplementary Table 2. Using this analysis, we identified 188 probe sets that are differentially expressed at a significance level of P < 0.001 (Table 3). The probability of finding at least 188 significant probe sets by chance, that is, under the assumption that there are no differences between the N₀ and N₊ groups was P = 0.035. These 188 probe sets represented 149 unique genes of which 46 genes were upregulated and 103 genes were downregulated in the N₊ group. Interestingly, 14 probe sets representing 5 unique genes (TCF4, CTNNAL1, DKK3, CTNND1/p120, and WNT5a) belong to the β-catenin pathway. This is in good agreement with our pathway analysis using all genes.

To validate the association between the lymph node status in early-stage cervical cancer and the oncogenic TGF-β signaling and β-catenin pathways, we performed immunohistochemistry using antibodies directed against proteins that are representative of both these pathways. For this purpose, we used a series of pretreatment early-stage cervical cancer tissues of 274 patients.

Phosphorylation of Smad2/3 and concomitant translocation into the nucleus is an important step in transforming growth factor β (TGF-β) signaling and expression of Smad4 is an essential partner of Smad2/3 in the formation of transcriptional complexes (36, 37). To validate whether Smad2, pSmad2, and/or Smad4 staining on the TMA are representative for the whole tumor, first whole tumor slides of a small series of 20 randomly selected cervical cancer tissues were immunostained. This immunostaining revealed that only Smad4 staining was homogeneous (data not shown). Therefore, Smad4 staining on the TMA reflects best the staining of the whole tumor. Thirty-five out of 255 evaluable cervical carcinomas showed positive Smad4 staining (see Supplementary Fig. 1 for representative immunostainings). Univariate logistic regression analysis of various clinicopathological features revealed that Smad4 positivity was not only related to N₀, (OR: 0.20, 95% CI: 0.06–0.66) but also to infiltration depth less than 10 mm (OR: 0.35, 95% CI: 0.16–0.76; Table 4).

To validate whether β-catenin signaling is associated with presence of lymph node metastases in cervical cancer, immunohistochemical staining was performed for β-catenin, E-cadherin, and p120 on whole tumor slides of 20 cervical cancer tissues. This revealed that E-cadherin was not a homogeneous staining. Immunostaining of β-catenin, a key protein in the canonical β-catenin pathway (38), and CTNND1/p120 that is involved in noncanonical β-catenin signaling (35) and was one of the 5 β-catenin related transcripts present in the list of 149 differentially expressed genes (188 probe sets; Table 3), was therefore performed on TMAs. Positive p120 immunostaining was observed in 112 of 268 (42%) and positive β-catenin in 140 of 272 (51%) patients (see Supplementary Fig. 1 for examples). Logistic regression analysis revealed no association between β-catenin protein expression and presence of lymph node metastases (Table 4). However, positive p120 staining was associated with N₊ (OR: 1.79, 95% CI: 1.05–3.05), in agreement with our microarray results.

In the present study, pathways associated with pelvic lymph node metastases in 39 (20 N₀ and 19 N₊) early-stage cervical cancer patients were identified. Our analysis of well-known and novel (n = 285) pathway signatures revealed an association of lymph node metastases with only few gene sets or signatures, including 2 well-known oncogenic biological gene sets. Enrichment of the TGF-β pathway was related to N₀, whereas oncogenic pathway activation of β-catenin was associated with N₊ patients. The association of both the TGF-β and the β-catenin signaling pathway with lymph node metastases was validated in a large consecutive series of early-stage cervical cancer patients by immunohistochemistry. Immunostaining of Smad4 and p120 representing the TGF-β and β-catenin signaling pathway, respectively, confirmed the association with lymph node metastasis in early-stage cervical cancer.

Until now, all studies using microarray platforms for differentiating between patient with and without lymph node metastases in cervical cancer focused on gene profiles and individual genes present in these profiles (14–17). Another approach is to identify biological pathways that are involved in biological differences between cancers, using pathway analysis methods on all genes that are differentially expressed between 2 phenotypes. For example, Lagarde and colleagues identified pathways that differentiated between N₀ and N₊ esophageal adenocarcinomas (39). Furthermore, Crijns and colleagues identified pathways contributing to clinical outcome of serous ovarian cancer (24). Interestingly, many of these pathways were known for being important in carcinogenesis or cancer progression, which indicates the strength of this approach. To our knowledge, we are the first to identify pathways for discriminating between N₀ and N₊ cervical cancer patients using pathway analysis methods.

Our analysis showed that TGF-β is one of the most important pathways affecting the metastatic potential in early-stage cervical cancer. First, of all 280 tested unique pathways (from the KEGG and Biocarta data bases), the TGF-β pathway was significantly enriched in N₀ (Table 2). Binding of the TGF-β ligand to its receptors initiates intracellular signaling by phosphorylation of Smad2 and Smad3. These phosphorylated Smads then bind to Smad4 and translocate into the nucleus, where this Smad complex is involved in regulation of gene transcription (36, 37). Immunostaining using Smad4 of 255 early-stage cervical carcinomas confirmed that TGF-β pathway activation was related to absence of lymph node metastases. Although Smad4 is a key protein in TGF-β signaling (36, 37), it is not known whether Smad4 immunostaining is representative of TGF-β signaling activity throughput. Immunostaining of more members of the TGF-β pathway would enhance our results, however no homogeneous staining was found for Smad2 and pSmad2 and therefore these stainings could not be performed on TMAs. Early in carcinogenesis, the TGF-β pathway contributes to tumor suppression, for example by stimulating apoptosis and inhibition of growth (36, 37) However, later in the process of tumor progression or in invasive cancer, oncogenic activity of TGF-β signaling is predominantly present, including increased migration and invasiveness, which may result in metastases. This transition from a tumor suppressor to an oncogenic pathway can be due to various alterations in TGF-β signaling, such as loss of Smad signaling and activation of Smad-independent, more oncogenic pathways, such as MAPK pathways (36, 37). Furthermore, TGF-β is directly involved in the formation of metastases, as it contributes to the establishment and outgrowth of lung and bone metastases in breast cancer models (32, 34). Smad4 downregulation is associated with TGF-β downregulation and has been implicated in cervical cancer (30) and metastatic mouse models (32). The downregulation of Smad4 in N₊ is consistent with these data and establishes TGF-β as one of the pathways affecting the metastatic potential in early-stage cervical cancer.

In addition to the TGF-β pathway, GSEA revealed that the NFAT, ALK, BAD, and PAR1 pathways are significantly enriched in the N₀ group and the Glycosphingolipid Biosynthesis Neo Lactoseries pathway in the N₊ group (Table 2). Presently, we and others have not characterized these pathways in detail for their possible association with the metastatic behavior of tumor cells. However, as these 5 pathways are also significantly associated with lymph node status, more detailed analysis is warranted to confirm their possible role in lymph node metastasis in early-stage cervical cancer.

A limitation of GSEA is that pathway activation can not be assessed for an individual patient. Therefore, another strategy was developed in which expression signatures are experimentally generated to reflect activation status of various oncogenic signaling pathways (23). This pathway analysis indicates that N₊ patients had a higher probability of β-catenin pathway activation than N₀ patients, pointing to a role for the β-catenin pathway in formation of lymph node metastases (Fig. 1). Interestingly, the gene set of 188 differentially expressed probe sets between N₀ and N₊, included 5 unique genes involved in the β-catenin-pathway including p120 (CTNND1 or catenin delta 1), CTNNAL1 (catenin alpha-like 1), DKK3 (dickkopf homologue 3), WNT5a, and TCF4 (transcription factor 4), but did not include β-catenin. In good agreement with these findings, immunohistochemistry confirmed the association of p120 and the lack of correlation of β-catenin with N₊. β-catenin is an important member of both the WNT-signaling pathway and the cell–cell adhesion pathway. However, immunohistochemical analysis revealed no relation between β-catenin and lymph node metastases, which is in agreement with other studies (31, 40) and indicates that the canonical Wnt/β-catenin pathway (containing β-catenin, Wnt1, APC) is not involved in mediating the invasive potential in cervical cancer. In normal cervical epithelium, β-catenin is involved in E-cadherin mediated cell–cell adhesion, by binding to the cytoplasmic domain of E-cadherin. Loss of E-cadherin causes disruption of cell adhesion and therefore might contribute to metastases (35, 38). P120 (also referred to as CTNND1 or delta-catenin) is a member of the catenin family and was originally reported to stabilize the cadherin-complex by direct interaction with the proximal domain of E-cadherin (35, 38). On the other hand, p120 (especially p120 isoform 1) promotes cell motility and invasiveness in cancer (33). P120 was reported to exert its effects by modulating the activities of Rho GTPases, for example by inhibiting activity of RhoA and activation of Rac and Cdc42 (33, 41). To our knowledge our study is the first that reports that p120 expression is associated with presence of lymph node metastases in early-stage cervical cancer. The link of p120 to Rho GTPases in activating the metastatic potential might also offer new opportunities for therapy, as invasion has been inhibited successfully using Rho-inhibitors (42).

Thus, both the TGF-β and the p120-associated noncanonical β-catenin pathway are related to lymph node metastases in cervical cancer. This indicates an important role for epithelial to mesenchymal transition (EMT), as both pathways may contribute to EMT. EMT is characterized by loss of the epithelial phenotype of cells and cells adopt a mesenchymal phenotype. It can be induced by alterations in TGF-β signaling, such as loss of Smad4 (43) and EMT is characterized by loss of E-cadherin, with disruption of cell adhesion as a consequence. Furthermore, EMT results in increased motility of cells, and increased invasion. All these processes contribute to the formation of metastases (44, 45). TGF-β signaling and β-catenin also cooperate in EMT. Loss of E-cadherin causes increased β-catenin signaling, which cooperates with autocrine TGF-β signaling to maintain an mesenchymal phenotype (46). Thus, deregulation of both the TGF-β and the β-catenin pathway, as observed in our study, indicates a role for EMT in lymph node metastasis in cervical cancer. Interestingly, miR-200a which is known for inhibition of TGF-β-mediated EMT by maintaining the epithelial phenotype through regulating expression of the E-cadherin transcriptional repressors ZEB1 and ZEB2 (47), was found to be a suppressor of metastasis in cervical cancer (48). This supports the importance of EMT in lymph node metastasis in cervical cancer.

Presence of lymph node metastases is still one of the most important factors in the choice of treatment for early-stage cervical cancer patients. No markers are currently available for accurate prediction of lymph node metastases before primary surgery. By evaluation of the primary tumor, expression levels of proteins such as Smad4 and p120 as representatives for the TGF-β signaling and β-catenin pathway, respectively, also cannot accurately predict presence of pelvic lymph node metastasis. However, more detailed analysis of these pathways might result in the identification of additional markers that will increase the clinical sensitivity en specificity. More importantly, by identifying pathways involved in lymph node metastasis in early-stage cervical cancer, new opportunities for pathway targeted therapy can be considered to inhibit the metastatic potential, as reported for both pathways (49, 50).

No potential conflicts of interest were disclosed.

Dutch Cancer Society (project number RUG 2004-3161). P.D. Moerland acknowledges support by the BioRange programme of The Netherlands Bioinformatics Centre (NBIC), supported by a BSIK grant through the Netherlands Genomics Initiative (NGI). EVLvT was supported by the FP6 European Union Project “Peroxisome” (LSHG-CT-2004-512018).

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.