The relationship between expression of VIMENTIN and CD146 genes in breast cancer

Objectives: CD146 is an adhesive molecule that was originally reported on malignant melanoma cells as a protein crucial for cell adhesion. It is now known that high expression of the CD146 protein is not only characteristic of melanoma, but it occurs on a number of cancers, contributing to worse prognosis and increased aggressiveness. Independent in vitro studies in breast cancer have shown that CD146 protein alone can induce a change in epithelial to mesenchymal transcriptional pro ﬁ le, which is the basis of the tumor aggressiveness and metastasis. Methods: In the following work, the correlation coefficients were analyzed between the genes of the mesenchymal profile and the CD146 gene in 10 independent transcriptomic data of breast cancer patients. Results: The analysis confirmed the relationship between CD146 expression and mesenchymal pro ﬁ le genes, point-ing VIMENTIN as the gene which expression is most strongly correlated with the CD146 , suggesting that both genes, CD146 and VIM may be directly controlled by the same mechanism or regulate one another. Conclusions: The analysis points a potential route for research on the CD146 gene expression, which may lead to understanding of its regulation in breast cancer, contributing to the development of new therapeutic strategies targeting highly metastatic breast cancer cells.


Introduction
CD146 was initially described on melanoma cells as an adhesion molecule contributing to the invasiveness of cancer cells. Of note, the currently published studies have reported the involvement of CD146 in progression and poor prognosis of several cancers, including breast cancer [1]. Interestingly, in vitro and in vivo data determined that CD146 introduction into breast cancer cells with epithelial characteristics trigger the changes in transcriptional profile leading to epithelial to mesenchymal transition (EMT) [2,3]. The mesenchymal cancer cells are characterized by poor cell-cell adhesive interaction and increased migratory and invasiveness potential, which altogether leads to enhanced aggressiveness and promote the progression of the disease. A key event in the epithelial-mesenchymal transition is the change in gene expression profile from epithelial to mesenchymal [4,5]. Although based on in vitro experiments, there is no doubt that overexpressed CD146 contributes to the aggressiveness of tumors by promoting cadherin switch, mesenchymal profile and malignant cell motility, still little is known about the regulation of this gene in cancer cells. So far none of the published study reported translocation, amplification or mutation in CD146 gene sequence in cancer cells [6]. From the other site, our preliminary data strongly suggest that epigenetic mechanism, exactly DNA methylation, may play a crucial role in the regulation of CD146 expression in breast and prostate tumors [7][8][9]. Regardless of the mechanism regulating the expression of CD146, the research data indicates that overexpression of this protein in vitro triggers signaling to lead to an alteration in the epithelial expression profile into the mesenchymal one in breast cancer cells [2]. In line with this observation, we show here that CD146 expression is significantly correlated with mesenchymal markers in several independent breast cancer patient's transcriptomic datasets. The analysis of the correlation coefficients performed in the current study showed that the VIMENTIN gene is the most strongly correlated mesenchymal marker with CD146, which may indicate that these two genes are regulated directly by the same overriding mechanism or regulate one another, which may be crucial for the initiation of the epithelial-mesenchymal transition. Our analysis indicates the direction for in vitro research, the results of which can significantly broaden our understanding of breast cancer progression.

Data and statistical analysis
To analyze the relation between CD146 and mesenchymal markers, we used R2 database developed at the Department of Oncogenomics in the Academic Medical Center (AMC) in Amsterdam, Netherlands (http://r2. amc.nl). In the following study, we applied a tool "Correlate Genes" in the panel of 10 breast cancer patients transcriptomic datasets (GSE7390-u133p2 and GSE46563-ilmnhwg6v3, composed of only lymph nodenegative patients and eight datasets GSE1456-u133a, GSE12276-u133p2, GSE2109-u133p2, GSE3494-u133a, GSE102484-u133a, GSE29271-u133p2, GSE69031-u133a, GSE36771-u133a), where patients were not selected  based on lymph node-negative status. To analyze correlation between CD146 gene expression and expression of 11 selected mesenchymal profile genes, p-value of the Spearman correlation was considered significant below 0.0045 according to Bonferroni correction (0.05/ 11). In addition, using R2 database we analyzed gene expression data from 51 breast cancer cell lines (GSE 12777). All statistical analysis described in the article were performed using Statistica 12 (Statsoft Polska). All affymatrix gene expression array data were normalized using the MAS5 algorithm.

Analysis of correlation between mesenchymal genes and CD146 in breast cancer datasets
Initially, using the publicly available R2 database, we calculated Spearman rank correlation coefficient for CD146 and 11 well known genes connected with mesenchymal profile including transcription factors coding genes (SNAI1, SNAI2, ZEB1, ZEB2, TWIST1), matrix metalloproteinases coding genes (MMP2, MMP9), fibronectin, N-cadherin and vimentin coding genes. In Table 1, the mean expression of all analyzed genes in 10 breast cancer datasets is shown. As determined in Table 1A, significant correlation was found between CD146 and mesenchymal markers in all analyzed breast cancer datasets, but the number of correlated genes was different between the different datasets, i.e., in datasets with confirmed lymph node-negativity (only two correlated genes out of 11 tested) in comparison to other datasets (at least seven correlated genes out of 11, where lymph node-negativity was not taken into account during selection of patients material). Subsequently, we also analyzed the correlation between CD146 gene expression and 11 mesenchymal profile genes in dataset composed of 51 breast cancer cell lines (Table 1B). As determined in Table 1C (left panel), most of the mesenchymal markers (eight out of 11) were correlated with CD146 in analyzed cell lines dataset. It is noteworthy that VIM gene was the only gene significantly correlated with the CD146 gene in all analyzed patient's date sets, whereas in cell lines, apparently, the highest correlation coefficient was observed for VIM and CD146 gene (Table 1C, left panel). As breast cancer cell line dataset was composed of the cell lines with luminal and basal characteristics, which resemble, respectively, epithelial and mesenchymal gene expression profile [10], we performed additional analysis separately for basal and luminal breast cancer cell lines (Table 1C, middle panel and right  panel). Interestingly, this analysis revealed that the VIM gene was the only one correlated positively with CD146 gene in breast cancer cell lines with luminal characteristics, what indicates, in agreement with breast cancer patients data, that the relation between CD146 and VIM gene is present regardless of the epithelial or mesenchymal gene expression profile.

Spearman's rank correlation coefficients matrix analysis in breast cancer datasets
Initially, the Spearman's rank correlation coefficients were calculated between the expression level of the specified genes (coding mesenchymal markers) and the level of CD146 gene expression in 10 independent breast cancer datasets. Spearman's rank coefficient enables correlations finding of any monotonic dependence, without assuming linearity. The calculations were performed for each dataset separately. Then each value of the correlation coefficient was modified according to the following formula: r = {r = r for p < 0.0045 r = 0 for p ≥ 0.0045} taking zero as the value of correlation coefficients for   which the significance level p was greater than or equal to 0.0045. The modified correlation coefficients of individual genes with the CD146 gene for selected datasets are presented in Table 2. Next, cluster analysis was performed using the square of Euclidean distance.
d(x, y) = ∑ni = 0(x i − y i ) 2 Subsequently, single linkage method was chosen as a hierarchical method to estimate the distance between clusters. In this method the distance between two clusters is defined by the distance between the closest objects (nearest neighbors) belonging to different clusters. On this basis, the distance matrix was calculated (Table 3A). The distance matrix is visualized on the dendrogram presented in Figure 1 and interpreted based on coefficients mean, coefficients of variation and the values of distance ( Table 3A and Table 3B). Interestingly, it has been observed that the set of correlation coefficients for the VIMENTIN gene with CD146 gene is the most different from the others. It has the highest mean correlation coefficient with the CD146 gene, the lowest coefficient of variation (Table 3B) and the largest distance (Table 3A) from the others what may indicate that VIM gene is regulated directly by the same mechanism as the CD146 gene, or the CD146 and VIM gene directly regulate one another during the induction of the mesenchymal profile. The remaining genes have a lower mean correlation coefficient and a higher coefficient of variation (Table 3B). Moreover, they form completely separate branches on the dendrogram, as determined in Figure 1 in relation to the VIMENTIN gene.

Discussion
Several independent studies confirmed the existence of correlation between CD146 and highly motile mesenchymal profile of cancer cells [1]. In our current analysis, we confirmed this finding analyzing the correlation between mesenchymal profile genes (the hallmark of EMT) and the expression of CD146 in 10 transcriptomic breast cancer patients datasets. Nevertheless, still little is known about the regulation of CD146 expression in breast cancer and it is not clear if aberrant CD146 is always required in EMT process. Importantly, in vitro studies in breast cancer cells revealed that ectopic expression of CD146 in epithelial line is sufficient to induces the mesenchymal profile, nevertheless we cannot exclude that in vivo CD146 is triggered by the upstream signaling inducing EMT. One note, high CD146 gene expression is observed especially in triple negative tumors (TNBC) [3], what may indicate the role of estrogen, progesterone or Her2 signaling in CD146-induced EMT. Indeed, in analyzed breast cancer datasets the significant negative correlation has  been found (data bot shown) for estrogen receptor (five out of 10 breast cancer patients dataset) and progesterone receptor (four out of 10 breast cancer patients datasets). Moreover, the literature data revealed the decrease in estrogen receptor expression in cells undergoing CD146-induced EMT [11]. In contrast, we did not find any significant correlation between CD146 and Her2 coding gene (ERBB2) in 10 analyzed datasets (data not shown). Most probably the fact that ERBB2 is often amplified, leading to induction of mesenchymal profile [12], and at the same lost in the group of TNBC tumors (characterized by aggressive mesenchymal expression profile and high CD146 expression) makes the potential relation between these two genes difficult to be analyzed based on gene expression profile. Another noteworthy scenario is that described in melanoma cells, where expression of the CD146 gene alone activates the endogenous PI3K/AKT pathway, which in a positive feedback induces CD146 expression. The PI3K/Akt signaling pathway interacts with other pathways, either directly or indirectly inducing EMT, resulting in increased aggressiveness, including proliferation, resistance to apoptosis, invasiveness, metastasis and resistance to chemotherapy. It is therefore possible that the abnormal overexpression of the CD146 gene itself will induce signaling leading to an epithelial-mesenchymal transition via activation of the PI3K/Akt pathway [6]. What is more, the fact that activation of PI3/Akt promotes CD146 expression in a feedback mechanism, may explain why high CD146 expression maybe crucial to maintain the mesenchymal profile in cancer cells. In other studies, CD146 has been shown to mediate E-cadherin-to N-cadherin switch during TGF-beta-induced EMT transition. As the authors of the study showed, the CD146 expression was positively correlated with the STAT3/ Twist activation and the ERK pathway. It has been shown that activation of CD146/STAT3/Twist signaling inhibits E-cadherin, while CD146/ERK enhances N-cadherin expression [13]. Regardless, the regulation scenario in breast cancer our correlation coefficients analysis revealed the closest relationship between CD146 expression and VIMENTIN expression in breast cancer patients datasets, which implies that a potential mechanism of CD146 regulation may be directly related to the regulation of the VIM gene. What is more, the VIM gene was the only one correlated with CD146 in all patients datasets, regardless the lymph node-negativity status and it was also the only gene significantly correlated with CD146 in luminal breast cancer cell lines, which resemble the tumors with epithelial gene expression profile. Altogether, it indicates that the correlation between CD146 and VIM gene expression is present in regardless to gene expression profile, epithelial or mesenchymal. In addition, of particular importance is the fact that both, the expression of vimentin itself and the expression of CD146 itself may lead to the induction of the mesenchymal profile in breast cancer cells [3,14]. The question then remains whether CD146 is necessary for induction of the mesenchymal profile by vimentin and vice versa. In addition, we should take into account the fact that both markers correlate closely with the aggressiveness of breast cancer and understanding the mechanisms of their potential co-regulation may significantly contribute to the development of new therapies targeting mesenchymal cancer cells responsible for metastasis and the same progression of the disease. Based on the results of the analysis, we propose the model of potential CD146/VIMENTIN coregulation, which should be tested in vitro (Scheme 1) in breast cancer cells. This model is based on three assumptions, the first that upstream signaling triggers the expression of CD146 and VIMENTIN which independently induce the mesenchymal profile or that CD146 triggers the expression of VIM gene (or vice versa) leading to the induction of the mesenchymal profile. Of particular importance is the fact that vimentin silencing leads to downregulation of mesenchymal profile genes [14] and the same was observed for CD146 [3]. The crucial question remains if vimentin-dependent induction of EMT