Is there a relationship with the basal-like subtype?
Colan M. Ho-Yen MBChB,
Centre for Tumour Biology, Barts Cancer Institute, Barts and the London School of Medicine and Dentistry, London, United Kingdom
Corresponding author: Colan M. Ho-Yen, MBChB, Centre for Tumour Biology, Barts Cancer Institute, Barts and the London School of Medicine and Dentistry, London EC1M 6BQ, United Kingdom; Fax: (011) +44-207 882 3884; firstname.lastname@example.org.
Basal-like (BL) breast cancer is an aggressive form of breast cancer with limited treatment options. Recent work has identified BL breast cancer as a biologically distinct form of triple-negative breast cancer, with a worse outlook. The receptor tyrosine kinase c-Met is a novel therapeutic target associated with reduced survival in breast cancer. Few studies have specifically addressed the association between c-Met and molecular subtype of breast cancer, yet this is a key consideration when selecting patients for clinical trials. The aim of this study is to evaluate c-Met expression in a large cohort of invasive breast cancers and in particular, its correlation with molecular subtype.
Immunohistochemistry for c-Met was performed and evaluated on 1274 invasive breast cancers using tissue microarray technology. The c-Met scores were correlated with molecular subtype, survival, and other standard clinicopathological prognostic factors.
Multivariate logistic regression showed c-Met was independently associated with BL status (odds ratio = 6.44, 95% confidence interval = 1.74-23.78, P = .005). There was a positive correlation between c-Met and Her2 (P = .005) and an inverse correlation with tumor size (P < .001). C-Met was an independent poor prognostic factor at Cox regression analysis in all subtypes (hazard ratio = 1.85, 95% confidence interval = 1.07-3.19, P = .027) and there was a trend toward reduced survival in BL tumors overexpressing c-Met, but this was not significant.
Basal-like (BL) breast cancer is an aggressive form of breast cancer and one of the 5 main subtypes identified at gene expression profiling, along with the luminal A, luminal B, human epidermal growth factor receptor 2 (Her2)-overexpressing, and “normal-like” subtypes.[1-3] BL breast cancer accounts for 10% to 25% of breast cancers and is more common in women of African descent and those harboring BRCA1 mutations.[4-6] Patients with BL breast cancer are of particular clinical interest because they currently lack any form of targeted therapy, with chemotherapy being the only option for systemic treatment.
Several studies have developed surrogate immunohistochemical profiles, allowing the identification of BL tumors on archival material.[8-10] These studies define BL tumors as negative for estrogen receptor (ER), progesterone receptor (PgR), and Her2, a tumor status referred to as triple negative (TN), but positive for at least one basal marker, such as cytokeratin (CK) 5/6, CK14, or epidermal growth factor receptor (EGFR).[8-10] Importantly, survival analysis shows that patients with these BL tumors do worse than women with TN tumors that are negative for basal markers (referred to here as “unclassified”),[8-10] suggesting that new molecular therapies may be of particular relevance to the BL category.
C-Met is a receptor tyrosine kinase mainly expressed on epithelial and endothelial cells. Upon binding of its ligand (hepatocyte growth factor [HGF]), the receptor undergoes dimerization and autophosphorylation of the tyrosine residues in the kinase domain (1234 and 1235) followed by phosphorylation of the 1349 and 1356 tyrosine residues in the multifunctional docking site near the C-terminus. Subsequent docking of various transduction molecules (including growth factor receptor-bound protein 2 [GRB2] and GRB2-associated binding protein 1 [GAB1]) initiates downstream signaling pathways with diverse biological functions that include cell survival, proliferation, migration, and invasion.
The HGF/c-Met signaling axis has an important role in embryological development and tissue repair[13, 14] but has also been implicated in cancer. Mutations in the MET gene have been documented in both sporadic and hereditary papillary renal cell carcinoma as well as a high proportion of cancers of unknown primary origin. MET amplification or c-Met protein overexpression have been described in lung, pharyngeal, gastric, and breast carcinomas.[18-21]
In breast tissue, there is a range of c-Met expression, with lowest levels in normal/benign tissue, higher levels in ductal carcinoma in situ, and highest expression in invasive carcinoma. Overexpression of c-Met is seen in 20% to 30% of invasive breast cancers, and several studies have shown c-Met overexpression to be an independent predictor of poor outcome in breast cancer.[23-25] The potential involvement of c-Met in the progression of breast and other cancers has led to considerable interest in the clinical application of anti-c-Met therapy, and several clinical trials are ongoing (www.clinicaltrials.gov). The c-Met kinase inhibitors constitute the bulk of the compounds currently under development and include ARQ197 (ClinicalTrials.gov identifier NCT01542996) and Cabozantinib/XL184 (ClinicalTrials.gov identifier NCT01738438), which are being evaluated for the treatment of metastatic TN breast cancer.
A key point that remains unclear is whether c-Met expression correlates with specific breast cancer molecular subtypes or indeed a specific group within a subtype, for example, the BL subset of TN. This is important because effective approaches to patient stratification are a crucial part of assessing the efficacy of new molecular therapies.
The aim of this study was to analyze c-Met expression in a large cohort of well-characterized invasive breast cancers, with particular attention to the relationship with molecular subtype.
MATERIALS AND METHODS
This study was based on tumors from 1896 patients who form a cohort of the previously characterized Nottingham Tenovus Primary Breast Carcinoma series, a collection of primary invasive breast cancers diagnosed between 1986 and 1998.[28, 29] Clinical and pathological data were available on these samples and included: age at presentation, tumor size, tumor grade, histological subtype, lymph node status, and presence/absence of vascular invasion. Overall survival time was defined as the time (in months) between diagnosis and the date of death/date of last follow-up. Patients were treated according to local guidelines. This study had approval from the Nottingham Research Ethics Committee 2 (title: “Development of a Molecular Genetic Classification of Breast Cancer”).
Tissue microarrays (TMAs) were constructed from whole tissue blocks as described. Cores of tumor tissue measuring 0.6 mm in diameter were extracted from “donor” blocks and inserted sequentially into a new “recipient” block using a TMA device (Beecher Instruments, Sun Prairie, Wis). Sections were then cut at a thickness of 4 μm and placed onto glass slides.
Immunohistochemistry (IHC) for c-Met was performed using the standard streptavidin-biotin (sABC) horseradish peroxidase method. Briefly, paraffin sections were dewaxed in xylene and then rehydrated through graded alcohol solutions. Endogenous peroxidase activity was blocked by incubating the sections in 3% hydrogen peroxide solution. Antigen retrieval was performed by microwave treating the sections in 0.1 M citrate buffer (pH 6) for 10 minutes. Normal goat serum (Vectastain ABC kit; Vector Laboratories) was diluted to 1:75 in 1% bovine serum albumin/phosphate-buffered saline (BSA/PBS) and deposited on the sections to block nonspecific antibody binding. The primary anti-c-Met antibody (CVD13; Invitrogen) was then incubated with the sections for 1 hour at room temperature at a dilution of 1:100 in 1% BSA/PBS. After washing in PBS, the secondary biotinylated antibody (Vectastain ABC kit; Vector Laboratories) was incubated for 40 minutes at room temperature at a dilution of 1:200 in 1% BSA/PBS. The sections were washed in PBS and then incubated with avidin-biotinylated peroxidase complex for 30 minutes. The reaction was developed for 2 minutes using a DAB kit (Vectastain, Vector Laboratories), after which sections were counterstained with hematoxylin, rinsed and dehydrated through alcohol to xylene, and mounted with cover slips. Negative controls consisted of breast cancer samples, with omission of the primary antibody.
IHC for other biomarkers was also available as described[28, 29, 32] and included ER, PgR, Her2, CK5/6, CK14, EGFR, p53, and MIB1.
Scoring, Cut-Point Selection and Molecular Subtyping
C-Met reactivity was scored using a semiquantitative approach combining numerical scores for intensity (0 = negative, 1 = weak, 2 = moderate, and 3 = strong) and area of reactivity (0 = < 1%, 1 = 1%-25%, 2 = 26%-50%, 3 = 51%-75%, 4 > 75%). Both the cytoplasmic and membrane compartments were scored and summed to give a total tumor score between 0 and 14. C-Met IHC was scored blindly by a single observer on 2 separate occasions. A proportion of cases were also scored by a second observer to ensure reproducibility. C-Met scores were kept as an integer score between 0 and 14, except for the purposes of survival analysis where samples were dichotomized into c-Met “high” (total score ≥ 7) and “low” (total score < 7) using the X-Tile bioinformatics tool for outcome-based cut-point selection. X-Tile defines subpopulation divisions in a “training” set and validates these on a “validation” cohort to provide a robust approach to optimal cut-point selection. This software has been tested on known prognostic factors in breast cancer (including tumor size and number of positive lymph nodes), indentifying cut-points comparable to established staging protocols.
The cut-points for ER, PgR (any tumor reactivity confers positivity), EGFR, Her2, CK5/6, and CK14 (≥ 10% reactivity) were as previously described. Scores for p53 and MIB1 were kept as continuous variables (percentage of positive tumor cells in whole tissue sections).
The breast cancers were divided into recognized molecular subtypes using the most frequently applied criteria as follows[9, 10]: luminal A (positive for ER and/or PgR, Her2-negative), luminal B (positive for ER and/or PgR, Her2-positive), Her2-positive (ER/PgR-negative, Her2-positive), BL (ER/PgR/Her2-negative, positive for at least one of CK5/6, CK14 or EGFR), and unclassified (ER/PgR/Her2-negative, negative for CK5/6, CK14 and EGFR).
Statistical analysis was performed using SPSS, version 19. Continuous variables were correlated with c-Met scores using Spearman's correlation coefficient, and categorical variables were compared using the Mann-Whitney test. Correlation between prognostic factors/biomarkers and BL status was carried out using univariate logistic regression and multivariate logistic regression with forward stepwise entry. Survival analysis was performed using the Kaplan-Meier method with the log-rank test for univariate analysis and the Cox regression model with forward stepwise entry for multivariate analysis. In order to facilitate comparisons in regression analysis, p53, MIB1, and c-Met scores were rescaled to give a value between 0 and 1. A 2-sided P value < 0.05 was considered significant.
Of the 1896 tumor samples included in the study, 1274 samples could be scored for c-Met reactivity (Fig. 1). The main reason for missing cases was a lack of sufficient tissue/tumor. The clinical characteristics of the whole cohort were comparable to those on which c-Met staining was quantified, including mean age at presentation (both 54 years), tumor size (both 21 mm), tumor grade (48% versus 51% grade 3, respectively), and lymph node status (37% versus 39% were positive in lymph nodes, respectively), suggesting that the c-Met data is representative of the wider cohort. The clinical, pathological, and molecular features of the patient cohort (excluding missing cases) are displayed in Table 1. The most common histological subtype was invasive ductal carcinoma, no special type (IDC-NST; 59% of tumors), and luminal A was the commonest molecular subtype (68%). BL tumors accounted for 13% of the cohort. After a mean follow-up time of 121 months (95% confidence interval [CI] = 23-201 months), 60% of patients were alive and 57% were free from recurrent disease.
Table 1. Clinical, Pathological, and Molecular Features of the Patient Cohort
Mean (95% CI)/Percentage of cases
Abbreviations: CI, confidence interval, IDC-NST, invasive ductal carcinoma, no special type; ILC, invasive lobular carcinoma.
54 years (37-69)
21 mm (9-40)
Lymph node involvement
121 months (23-201)
Correlation Between c-Met Scores and Prognostic Factors
We analyzed the correlation between c-Met scores and factors associated with outcome in breast cancer; continuous variables are shown in Table 2 and categorical variables in Table 3. There was a significant positive correlation between c-Met and Her2 score (P = .005) and a borderline (P ≥ .05) positive correlation with MIB1 and EGFR scores (Table 2). There was an inverse correlation between c-Met and tumor size (Table 2). There was no correlation between c-Met and age at diagnosis or p53 score.
Table 2. Correlation Between c-Met and Prognostic Factors (Continuous Variables)a
Significant correlations are in italics (Spearman's correlation coefficient).
Table 3. Association Between c-Met and Prognostic Factors (Categorical Variables)a
Mean c-Met score (95% CI)
Significant findings are in italics (Mann-Whitney test). Abbreviations: CI, confidence interval; IDC-NST, invasive ductal carcinoma, no special type; ILC, invasive lobular carcinoma.
1 or 2
Lymph node involvement
IDC-NST vs Non-IDC-NST
7.5 (7.3-7.7) vs 7.6 (7.3-7.8)
ILC vs Non-ILC
6.7 (6.1-7.3) vs 7.6 (7.5-7.8)
Tubular carcinoma (TC) vs Non-TC
8.8 (8.1-9.4) vs 7.5 (7.4-7.7)
Mucinous carcinoma (MC) vs Non-MC
7.0 (4.9-9.1) vs 7.5 (7.4-7.7)
Medullary/atypical vs Nonmedullary/atypical
8.0 (6.9-9.1) vs 7.5 (7.4-7.7)
Luminal A vs Nonluminal A
7.4 (7.2-7.6) vs 7.7 (7.5-8.0)
Luminal B vs Nonluminal B
7.4 (6.9-7.9) vs 7.5 (7.4-7.7)
Her2 positive vs Non-Her2
7.6 (7.1-8.1) vs 7.5 (7.4-7.7)
Basal-like vs Non-Basal-like
8.0 (7.6-8.5) vs 7.4 (7.3-7.6)
Unclassified vs Non-Unclassified
7.7 (6.9-8.5) vs 7.5 (7.4-7.7)
C-Met scores were significantly higher in patients with negative lymph nodes (P = .033), tubular carcinoma (P = .003), and BL breast cancer (P = .037). C-Met scores were lower in invasive lobular carcinoma (ILC; P = .001) and luminal A cancer (P = .014); although there was an association between c-Met and Her2 score, there was no significant association with tumors categorized as Her2-positive (ER/PgR-negative, Her2-positive). C-Met scores showed no significant association with tumor grade or the presence of vascular invasion (Table 3).
Correlation With BL Status: Univariate Logistic Regression
Using univariate logistic regression, we analyzed which prognostic factors (including c-Met) associated with BL cancer (Table 4). Increasing age at presentation was inversely associated with BL cancer. Increasing tumor size, p53 score, MIB1 score, and c-Met score all were positively associated with BL cancer, as were grade 3 tumors and tumors in the histological categories of IDC-NST and medullary/atypical medullary carcinoma (Table 4).
Table 4. Association Between Prognostic Factors, c-Met and BL Phenotype (Univariate Logistic Regression)
OR (95% CI)
Invasive lobular carcinoma, tubular carcinoma, and mucinous carcinoma are omitted because they were not represented in the BL phenotype.
Abbreviations: CI, confidence interval; IDC-NST, invasive ductal carcinoma, no special type; LR, likelihood ratio; OR, odds ratio.
Correlation With BL Status: Multivariate Logistic Regression
To establish if c-Met expression was an independent marker of BL cancer, we entered c-Met score together with the other 6 factors that associated with BL status at univariate analysis (age at presentation, tumor size, histological subtype [IDC-NST and medullary/atypical medullary carcinoma], grade, MIB1 score, and p53 score) into the multivariate model. Data on these 7 factors were available on 945 samples in the cohort. After forward stepwise entry, the 5 factors that remained in the multivariate model were: MIB1 score (P < .001), p53 score (< 0.001), tumor grade (P < .001), c-Met score (P = .005), and age at presentation (P = .006; Table 5).
Table 5. Multivariate Model Containing Parameters Predictive of the BL Phenotype (Multivariate Logistic Regression With Forward Stepwise Entry, n = 945)
The c-Met scores were dichotomized into “high” and “low” expression to create Kaplan-Meier curves for univariate analysis of survival across all molecular subtypes (Fig. 2A) and for the BL tumors (Fig. 2B). For all molecular subtypes (Fig. 2A), mean survival time was significantly lower in the c-Met high group (163 months, 95% CI = 157-169 months) compared to the c-Met low group (186 months, 95% CI = 174-198 months; P = .005, log-rank test, n = 1154). The BL c-Met high tumors (Fig. 2B), showed a trend toward reduced survival (mean survival time = 148 months, 95% CI = 130-165 versus 168 months, 95% CI = 129-207 for c-Met low tumors, P > .05, log-rank test, n = 137), but this was not significant.
A Cox regression model was constructed to assess whether c-Met was an independent poor prognostic factor. The c-Met score together with 4 other factors associated with survival (tumor grade, lymph node status, age and tumor size) were entered into the model. Data on these 5 factors were available on 1002 samples across all molecular subtypes. Increasing c-Met score was associated with a worse outcome (hazard ratio [HR] = 1.85, 95% CI = 1.07-3.19, P = .027; Table 6). To establish whether the association between reduced survival and high c-Met scores was due to the link with BL tumors, we added in BL status to the Cox regression model. This data was available on 976 cases, and increasing c-Met expression remained a poor prognostic factor (HR = 1.87, 95% CI = 1.07-3.27, P = .029), suggesting high c-Met predicts a poor outcome in non-BL tumors. C-Met was not an independent predictor of reduced survival in BL tumors alone (n = 118).
Table 6. Multivariate Model Containing Parameters Predictive of Reduced Survival (Cox Regression With Stepwise Forward Entry, n = 1002)
In this analysis of c-Met expression in different molecular subtypes of breast cancer, significantly higher levels were identified in the BL subtype only. Notably, the unclassified tumors showed no significant increase in expression of the receptor. Moreover, this is the first study to show that protein expression of c-Met is independently associated with the BL phenotype at multivariate analysis.
Our findings support previous studies describing an association between c-Met and the BL phenotype at a protein/gene expression level.[34-36] In a study of 122 breast tissue samples, Charafe-Jauffret et al showed that a gene expression signature that included MET overexpression separated BL tumors from luminal A/B tumors. High-intensity c-Met positivity as determined by IHC analysis has also been shown to be significantly associated with BL tumors in a univariate analysis of 137 invasive breast cancers. Elsewhere, a large TMA-based study comprising more than 900 tumors quantified IHC expression of c-Met, together with other markers of BL cancer, including CK5/6, p63, and c-Kit. The study's authors found that c-Met overexpression was associated with a worse outcome but a direct relationship to molecular subtype was not made. Importantly, we found that patients with BL cancers in our cohort had many of the characteristics previously described in the literature,[8, 37-40] including younger age at presentation, larger tumor size, predilection for the histological category of IDC-NST, high nuclear grade and mitotic rate, and increased p53 expression. Hence, our cohort appears to be representative of the BL phenotype and therefore our findings with regard to c-Met expression could be extended to the wider population.
We found an inverse correlation between c-Met expression and tumor size. This finding was unexpected, given that BL tumors tend to be larger at presentation. Similar studies have found no relationship between c-Met and tumor size.[22, 24, 41] In a previous study, we have described tumor size as being a poor predictor of outcome in BL cancer and it is interesting to postulate whether c-Met signaling plays a decisive role in the progression of smaller tumors. A more predictable finding was that the lymph node negative tumors in this cohort showed higher c-Met scores. We previously have shown that BL tumors more frequently metastasize via the hematogenous route to distant organs than via the lymphatic route to lymph nodes.
Our comparison of histological subtypes revealed highest expression in tubular carcinomas, with ILC showing the lowest levels. Few studies have looked specifically at c-Met expression in relation to histological subtype, but with regard to ILC, the findings have been variable with some finding higher expression, lower expression, or no significant difference in comparison to IDC. ILCs lack expression of E-cadherin, and there is some data supporting a direct interaction between c-Met and E-cadherin, because forced expression of E-cadherin in a breast cancer cell line normally negative for this protein (BT-549) results in the recruitment of c-Met from the cytoplasm to the membrane. It has been proposed that this stabilization of c-Met at the membrane facilitates binding of HGF to the receptor.
We identified a positive correlation between c-Met and Her2 scores. Others have described coexpression of these 2 receptor tyrosine kinases in breast cancer,[21, 47] and there is growing evidence of cross-talk between c-Met and the EGFR family. MET amplification has been causally linked to gefitinib resistance in non-small-cell lung carcinoma cells by stimulating persistent phosphoinositide 3-kinase signaling independent of EGFR. Inhibition of c-Met kinase activity was necessary to restore sensitivity to EGFR inhibition in these cells. Similarly, work on the SUM229 breast cancer cell line has shown that proliferation in the presence of EGFR inhibition is mediated by a c-Met/c-Src-dependent pathway. A more recent study looking at metastatic Her2-positive breast cancer found that increased MET and HGF gene copy numbers correlated with increased trastuzumab therapy failure, suggesting that MET/HGF amplification contributes to trastuzumab resistance. This raises the possibility that the therapeutic potential of c-Met inhibition is not limited to BL breast cancers and supports the use of multitarget kinase inhibition over single-target strategies.
Overall, our results identify c-Met expression as an independent poor prognostic factor in breast cancer. This is in concordance with earlier studies that found c-Met to be an independent predictor of poor outcome in node-negative and node-positive breast cancer.[21, 23-25, 45, 50] We also studied the survival profiles of c-Met “high” and “low” tumors within the BL category, noting a trend toward worse outcome in c-Met overexpressing tumors, particularly within the first 5 years. The comparatively small study size of our BL cohort may explain the lack of statistical significance, and it would be interesting to extend this analysis to a larger series. A recent evaluation of TN breast cancers recorded a significantly worse outcome for c-Met-overexpressing tumors after correcting for age, tumor size, nodal status, and tumor grade. These investigators did not include basal markers in their study, but our data would suggest that the c-Met overexpressing TN tumors were predominantly BL cancers.
In conclusion, we found a variable relationship between c-Met and established prognostic factors, possibly reflecting the complexity of HGF/c-Met signaling. Crucially, we have shown an association between c-Met and BL breast cancer that is independent of other prognostic factors and biomarkers associated with this aggressive subtype of breast cancer. Our results support the case for distinguishing BL tumors from unclassified tumors and provide further evidence for studying patients with BL breast cancer in particular, when evaluating the clinical potential of c-Met inhibitors.
Dr. Ho-Yen is funded by a Cancer Research UK Clinical Fellowship.