Identification of a Subtype of Poorly Differentiated Invasive Ductal Carcinoma of the Breast Based on Vimentin and E-cadherin Expression

Objective The use of molecular markers can identify a subgroup of tumors with distinct recurrence patterns. The present study aimed to characterize the immunohistochemical expression of vimentin (VIM), of E-cadherin (CDH1), and of cytokeratin 5 (CK5) in patients with invasive ductal carcinomas (IDCs). Methods We have constructed a tissue microarray (TMA) from 87 patients with IDC of the breast. Immunohistochemistry (IHC) was performed to study the expression of estrogen and progesterone receptors (ER and PgR), human epidermal growth factor receptor 2 (HER2), VIM, CDH1, CK5, and Ki67. The tumors were classified as luminal A and B (n = 39), HER2 enriched (n = 25), and triple-negative (TNBC) (n = 23), based on the IHC expression. Results We have observed that luminal A and B tumors lack the VIM+/CDH1-/low phenotype. This phenotype was observed in 16.5% of the HER2+ tumors and in 60% of the TNBC tumors (p = 0.0001). Out of a total of 20 TNBC tumors, the CK5 (basal-like marker) was positive in 11 of them. The VIM+/CDH1-/low phenotype was observed in 5 CK5+ TNBC tumors (45%) and in 7 out of 9 CK5- TNBC tumors (78%) (p = 0.02). The median Ki67 index in the VIM+/CDH1-/low tumors was 13.6 (range: 17.8–45.4) compared with 9.8 (range: 4.1–38.1) in other tumors (p = 0.0007). The presence of lymph node metastasis was less frequent in patients with VIM+/CDH1-/low tumors (23% versus 61%; X2 test; p = 0.01). Conclusion Our findings suggest that the expression of VIM and CDH1 can identify a subset of IDCs of the breast with a mesenchymal phenotype associated with poor prognosis, high-grade lesion, and high mitotic index.


Introduction
Based on its genomic profile, breast cancer can be classified in six subtypes (luminal A, luminal B, human epidermal growth factor receptor 2 (HER2)-enriched, basal-like, claudin-low, and normal-like). 1,2 This classification stratifies patients with adverse prognostic characteristics, and the basal-like and claudin-low subtypes confer the shortest overall survival. [3][4][5][6] Immunohistochemical phenotyping based on estrogen and progesterone receptors (ER and PgR), HER2, and Ki67 expression has been proposed as an alternative to classify breast cancer into different subtypes. According to this immunophenotyping approach, breast cancer can be divided into luminal A (HR þ , HER2-, low Ki67 index), luminal B (HRþ with HER2þ or HER2-plus high Ki67 index), HER2-enriched (HR-, HER2 þ ) and triple negative (TNBC) (HR-and HER2-). [7][8][9][10] Triple negative tumors tend to have a worse prognosis than luminal A, luminal B, and HER2-enriched tumors, and no specific targeted TNBC therapy has been developed yet. [11][12][13][14] Although TNBC has been characterized as a pathological entity, there is significant biological heterogeneity within these tumors. 13,15 The basal-like phenotype is reported as the main TNBC subtype. 8,16,17 However, all the other molecular subtypes can occur. 16,17 The use of immunohistochemistry (IHC) to identify different subgroups of TNBC tumors has been proposed. Cytokeratin 5 (CK5) and epidermal growth factor receptor (EGFR) expressions have been reported as markers of basal-like carcinomas. 7,8,14,15,18 In fact, CK5 expression has been reported to be the most accurate basal-like marker. 7,[19][20][21][22] Breast cancers that have arisen from basal cells are supposed to have a mesenchymal phenotype. 17,23,24 The claudin-low group lacks cell-cell junction proteins, and has features resembling the epithelial-mesenchymal transition (EMT) activation. 2,13,17,25 Additionally, some studies have suggested that there is a subgroup with a quintuple-negative profile (HR-, HER2-, CK5-, and EGFR-) that exhibits a worse prognosis when compared with CK5þ or EGFRþ TNBC. 18,26 Thus, the use of EMT markers can identify an undifferentiated subgroup with a possibly highly aggressive clinical behavior. 4,17,27 Most investigated and reported EMT markers include the Ecadherin (CDH1), a cell surface protein responsible for the adhesion of epithelial cells. Downregulation in CDH1 is highly associated with EMT, and partial or total loss of CDH1 expression is associated with more aggressive behavior and poor prognosis. [28][29][30] Often coupled with low CDH1expression, vimentin (VIM), a type III intermediate filament protein that is expressed in mesenchymal cells, has been used to investigate EMT activation in epithelial tissues. The VIM expression in epithelial cells is associated with a migratory phenotype and consequent invasiveness and metastasis. 29,[31][32][33] The aim of the present study was to determine the VIM and CDH1 expression in invasive ductal carcinomas (IDCs) of the breast, and to analyze their association with CK5 expression and clinical and pathological features.

Sample Selection
We have selected all patients (n ¼ 175) with IDC of the breast subjected to surgery at Hospital das Clínicas of the Faculdade de Medicina de Ribeirão Preto between January 2005 and December 2007. The selection was based on the histological and immunohistochemical diagnosis in the files of the patients. The staging system was based on the American Joint Committee on Cancer (7 th edition) classification. We have used pathological staging for patients subjected to primary surgery, and clinical staging in cases of neoadjuvant treatment. According to the IHC reports, 27 IDCs were TNBC, 39 were HER2þ (with a score of 3 þ), 36 were HER2þ (score of 2 þ), and 73 were HRþ (for estrogen or progesterone receptors) and HER2-(with a score of 0 or 1 þ). We were able to retrieve 94 paraffin blocks, and 82 were suitable for tissue microarray (TMA) construction. The total sample comprised 24 TNBCs, 9 HERþ and HR-, 14 HER2þ and HR þ , and 35 HRþ and HER2-. All samples were reevaluated, and the histological diagnosis and tumor grade were reported. ►Table 1 summarizes the characteristics of the patients.

Tissue Microarray
Core biopsies (diameter ¼ 1mm) were punched from 2 representative areas of each tumor of each of the 87 donor paraffin blocks and arrayed into a new recipient paraffin block using a manual tissue arrayer (Beecher Instruments, Sun Prairie, WI, USA). Three 1 μm-thick sections were cut from a tissue microarray (TMA) paraffin block using a paraffin tape-transfer system (Instrumedics Inc., Saint Louis, MO, USA). The slides were dipped in paraffin to prevent oxidation. One section was stained with hematoxylin and eosin (H&E) to confirm the presence of the tumor by light microscopy.

Immunohistochemistry
The immunohistochemical staining was performed using the Novolink Max Polymer Detection System (Leica Biosystems, Wetzlar, Germany). The sections were deparaffinized in xylene and rehydrated through a series of graded alcohols. The endogenous peroxidase activity was blocked for 30 minutes in a solution containing 0.3% of hydrogen peroxide to block non-specific immunoassaying. The sections were then placed in a 10 mM citrate buffer and submitted to heat retrieval using a vapor lock for 40 minutes. After the antigen retrieval, the specimens were allowed to cool for 30 minutes, and then incubated at 4°C overnight with a primary antibody. The dilution and source of the primary antibodies used in the present study were: anti-human ER (1:100, clone 6F11) (Leica Biosystems, Wetzlar, Germany), anti-human PgR . After overnight incubation with the primary antibody, the slides were incubated with a postprimary solution for 30 minutes and then incubated with the polymer for another 30 minutes (both provided by the Novolink Max Polymer Detection System). The reaction was developed with diaminobenzidine (DAB), followed by hematoxylin counterstaining. The slides were then dehydrated in an ethanol series and mounted with Permount (Fischer, Fair Lawn, NJ, USA). The DAKO Herceptest (Agilent Technologies, Santa Clara, CA, USA) was used for HER2 protein staining following the protocol of the manufacturer.
Digital analysis: In brief, the Ki67 TMA slide was scanned at 20 Â magnification on an Aperio Scanscope CS (Aperio Technologies Inc., Vista, CA, USA). The Aperio Scanscope CS is a whole-slide imaging system that scans entire tissue sections and registers the image stripes into one file. The image was uploaded onto an Aperio Spectrum Plus server, and then we used the TMA Laboratory module to break the image into its logical components of sectors, rows, columns, and cores. The Nuclear V9.1 (Aperio Technologies, Vista, CA) image analysis algorithms were used for the quantification of the Ki67 staining. In general, this algorithm begins by splitting the color data in the image into a maximum of 3-color channels. In this case, we separated the channels into a blue hematoxylin channel and a brown DAB channel. The

Statistical Analysis
The expression of VIM and CDH1proteins and other categorical variables were compared using the standard X 2 test or the Fisher exact test. The difference in the Ki67 staining index was evaluated using the median test, while the difference in age between groups was assessed using the t-test. The disease-free and overall survival interval was calculated from the date of diagnosis, and the survival curves were derived from Kaplan-Meier estimates and compared by logrank tests. The JMP version 10 software (SAS, Cary, NC, USA) was used for the statistical analyses. The significance was established as p < 0.05 (two-sided). Comparing tumors with the VIM þ /CDH1 -/low phenotype with the other tumors, there was no difference in the age, menopausal status, clinical stage, and the presence of lymph node invasion in the patients (X 2 test; p ¼ 1.0). The frequency of lymph node metastases was lower in patients with VIM þ /CDH1 -/low tumors than in other tumors (►Table 2).

Expression of CK5, VIM and CDH1 in IDCs
We have analyzed the disease-free and overall survival according to tumor subtypes, CK5 expression, and VIM þ / CDH1 -/low phenotype. We did not find a significant difference in disease-free and overall survival between luminal A, luminal B, HER2-enriched, and TNBC tumors, and there was no association between CK5 expression and survival (data not shown). In patients with VIM þ /CDH1 -/low phenotype and non-VIM þ /CDH1 -/low phenotype tumors, the 5-year disease-free survival rates were of 61.5% and 83.7%, and the 5-year overall survival rates were 51.2% and 83.5%, respectively. ►Fig. 2 shows the Kaplan-Meier curves for disease-free and overall survival according to the VIM þ / CDH1 -/low phenotype, which were statically significant (p ¼ 0.02 and p ¼ 0.03 [log-rank test], respectively).
Despite the fact there is yet no specific tumor marker to identify the basal-like subset of breast tumors, the development of a targeted therapy as a clinical approach to enhance systemic therapy has received great attention for breast cancer treatment. 11,12,[37][38][39][40] The acquisition of a mesenchymal phenotype by epithelial malignant cells is an important step to cancer invasion and metastasis. [39][40][41][42] Some novel targeted therapies that interfere specifically with the EMT program are being developed. [43][44][45][46][47] Triple negative tumors are considered a specific subtype of breast cancer with early recurrence and poor prognosis. [9][10][11]14 Transcriptome analysis has demonstrated that basal-like breast carcinomas exhibit a mesenchymal molecular profile. 3,7,8,16,17,19,21 Thus, the use of an anti-EMT program therapy could be investigated in the treatment of TNBCs.
The use of CK5 has been proposed to identify a subset of TNBC tumors with a basal phenotype in some reports, demonstrating that CK5 expression in TNBCs could predict between 61 and 95% of basal-like subtypes as defined by transcriptome analysis. 20,48,49 However, some basal-like tumors do not express this cytokeratin. Additionally, the expression of CK5 is observed in some non-TNBCs. [48][49][50] In our study, we have observed CK5 positivity in 10% of luminal A, in 5% of luminal B, and in 12.5% of HER2-enriched IDCs, and we have demonstrated that only 33% of CK5þ tumors have a mesenchymal phenotype as defined by the positive expression of VIM and the reduction or absence of CDH1 expression (VIM þ /CDH1 low/phenotype).
In our study, a VIM þ /CDH1 low/phenotype was identified in 17.8% of IDCs, and this phenotype was observed only in TNBCs. We have found some features that classify the VIM þ /CDH1 low/phenotype as an aggressive IDC. These tumors exhibit a high Ki67 index, are generally poorly differentiated, and have a reduced incidence of lymph node metastasis, suggesting that they display a mesenchymal type of malignant tumor behavior.
Triple negative tumors are associated with a poor prognosis. In fact, some reports have demonstrated that there is a bipolar biological behavior among TNBC patients. There is a group of TNBC patients that develop early and aggressive recurrence, while other TNBC patients present with a more favorable prognosis. [51][52][53] Our results suggest that the use of mesenchymal markers may be important to identify TNBC tumors associated with a poor outcome. In fact, we have observed that patients with the VIM þ /CDH1 -/low phenotype have a higher rate of recurrence and a worse prognosis than patients with the TNBC non-VIM þ /CDH1 -/low phenotype. However, this is a retrospective and single institution study including a relatively small number of samples. A large prospective study is necessary to confirm our findings.

Conclusion
In conclusion, the IHC identification of a subset of IDC with mesenchymal phenotype suggests the selection of an aggressive IDC subtype and should prompt further investigation on this field.

Contributions
Tiezzi D. G. participated in the design of the study, in the elaboration of the draft of the manuscript and performed the statistical analysis. Andrade J. M. and Reis F.J.C. participated in the design and in the coordination of the study and helped to elaborate the draft of the manuscript. Tiezzi M.G. performed the immunohistochemical analysis. Bosland M. and Deaton R. were responsible for the digital analysis. Orlandini L. F. contributed in the elaboration of the draft of the manuscript. Silveira W. A. contributed with the data collection and with the selection of the paraffin blocks.

Conflicts of Interest
The authors have no conflicts of interest to declare. with the support of the Vice Chancellor of Research. This work received a grant from Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP, in the Portuguese acronym), under the grant number 2008/09294-9.