Integrated Proteomics and Genomics Analysis Reveals a Novel Mesenchymal to Epithelial Reverting Transition in Leiomyosarcoma through Regulation of Slug*

Leiomyosarcoma is one of the most common mesenchymal tumors. Proteomics profiling analysis by reverse-phase protein lysate array surprisingly revealed that expression of the epithelial marker E-cadherin (encoded by CDH1) was significantly elevated in a subset of leiomyosarcomas. In contrast, E-cadherin was rarely expressed in the gastrointestinal stromal tumors, another major mesenchymal tumor type. We further sought to 1) validate this finding, 2) determine whether there is a mesenchymal to epithelial reverting transition (MErT) in leiomyosarcoma, and if so 3) elucidate the regulatory mechanism responsible for this MErT. Our data showed that the epithelial cell markers E-cadherin, epithelial membrane antigen, cytokeratin AE1/AE3, and pan-cytokeratin were often detected immunohistochemically in leiomyosarcoma tumor cells on tissue microarray. Interestingly, the E-cadherin protein expression was correlated with better survival in leiomyosarcoma patients. Whole genome microarray was used for transcriptomics analysis, and the epithelial gene expression signature was also associated with better survival. Bioinformatics analysis of transcriptome data showed an inverse correlation between E-cadherin and E-cadherin repressor Slug (SNAI2) expression in leiomyosarcoma, and this inverse correlation was validated on tissue microarray by immunohistochemical staining of E-cadherin and Slug. Knockdown of Slug expression in SK-LMS-1 leiomyosarcoma cells by siRNA significantly increased E-cadherin; decreased the mesenchymal markers vimentin and N-cadherin (encoded by CDH2); and significantly decreased cell proliferation, invasion, and migration. An increase in Slug expression by pCMV6-XL5-Slug transfection decreased E-cadherin and increased vimentin and N-cadherin. Thus, MErT, which is mediated through regulation of Slug, is a clinically significant phenotype in leiomyosarcoma.

The adhesion protein E-cadherin, encoded by CDH1, plays a central part in the process of epithelial morphogenesis. The down-regulation of E-cadherin is associated with a process called epithelial to mesenchymal transition (EMT) 1 that accounts for increased invasion and metastasis during tumor progression in multiple carcinomas of epithelial origin (1)(2)(3)(4)(5). Altered expression of E-cadherin has been shown to be regulated through several transcriptional factors such as Snail, SIP1, Twist, and Slug (encoded by SNAI2) and non-coding RNA such as miR-200 and let-7 (3)(4)(5)(6)(7)(8)(9). EMT of epithelial cancer cells is characterized by acquisition of fibroblast-like properties with reduced intercellular adhesion and increased motility in vitro as well as metastasis (2,5,10). Recently, a similar but reverse process called mesenchymal to epithelial reverting transition (MErT) has been observed and reported (11). Ecadherin is also a key indicator of MErT during the metastatic seeding of disseminated carcinomas (11). In synovial sarcoma, the fusion protein SYT-SSX1 has been shown to induce MErT through the regulation of SNAI1 and Slug (12). These investigations suggest that MErT might be an important biological process for tumors of mesenchymal origin, although E-cadherin expression is infrequent in most sarcomas examined thus far (13).
Leiomyosarcomas and gastrointestinal stromal tumors (GISTs), two of the most common mesenchymal tumors, share remarkably similar phenotypic features but are molec-ularly and clinically distinct (14 -16). In the present study, proteomics profiling analysis of these two types of sarcoma using a reverse-phase protein lysate array surprisingly revealed increased expression of E-cadherin in a subset of leiomyosarcomas. To our knowledge, no previous study has comprehensively evaluated leiomyosarcomas for E-cadherin expression. In this study, we validated the expression of epithelial cell markers, including E-cadherin, using an array of assays in a subset of leiomyosarcomas, determined the clinical significance of the MErT phenotype, and characterized the regulatory mechanism responsible for this MErT.

MATERIALS AND METHODS
Patients and Samples-All 31 leiomyosarcoma samples and 38 GIST samples were obtained from surgical specimens at The University of Texas M. D. Anderson Cancer Center and stored at the Institutional Tumor Tissue Repository with patient consent and according to an Institutional Review Board-approved protocol. All the tissues were snap frozen within 20 min of surgical resection. The pathology evaluation was described previously (14) and confirmed for the current study by a co-author who is a board-certified sarcoma pathologist (A. J. F. Lazar). Briefly, diagnosis was made on the basis of cellular features by light microscopy; immunohistochemical staining for KIT, CD34, and SMA; and clinical observations, including the site of the primary tumor, the pattern of metastatic spread, and the efficacy of systemic therapy. Leiomyosarcoma was diagnosed when a tumor manifested intersecting fascicles of elongated spindle cells with cigar-shaped, elongated nuclei with amphophilic cytoplasm with at least five of 10 high power field mitotic figures and positive SMA and after negative CD34 immunostaining in patients with the appropriate clinical setting. GIST diagnosis was made when a patient had the clinical presentation consistent with GIST and the tumor was composed of spindled, epithelioid, or mixed cell proliferations with positive KIT, positive cytoplasmic CD34, and negative SMA. The few tumors that were negative for KIT, SMA, and CD34 were classified on the basis of the light microscopic features, clinical pattern of disease, and the two-gene classifier described (14).
RNA Isolation and Whole Genome Microarray Experiments-Total RNA of 31 leiomyosarcoma samples was isolated and quantified. The transcriptional microarray experiments used whole human genome oligo arrays with 44,000 60-mer probes (Agilent Technologies, Palo Alto, CA) as described previously (14). The microarray data are available as supplemental material.
Reverse-phase Protein Lysate Array and Data Evaluation-Protein isolation from leiomyosarcoma and GIST tissues (28 leiomyosarcomas and 38 GISTs) and the detection of protein expression by reverse-phase protein lysate array were performed according to methods described previously (17,18). Briefly, the frozen tissue was ground in liquid nitrogen and lysed with protein lysis buffer (20 mM Tris, pH 7.6, 150 mM NaCl, 5 mM EDTA, 0.5% Nonidet P-40) freshly supplemented with 0.02 mM leupeptin. The concentrations of the lysate solution were determined using the Bradford assay according to the manufacturer's protocol (Bio-Rad) and adjusted to 20 mg/ml with lysis buffer. The lysate solution was then serially 2-fold diluted six times with lysis buffer. The serially diluted protein lysates were printed on PVDF-coated glass slides in triplicate using a robotic spotter (G3, Genomics Solutions) as described previously (18). The data evaluation and clustering analysis were performed according to the reported methods (17,18).
Immunohistochemical Staining on Tissue Microarray-A tissue microarray (TMA) consisting of leiomyosarcoma (31 samples), GIST (38 samples), and their normal control tissues (24 samples) was con-structed with the use of punch cores measuring 0.6 mm in diameter. Immunohistochemical staining of tissue sections cut from the TMA block was performed with the use of a panel of antibodies with various dilutions, and the results were analyzed by one of the authors (A. J. F. Lazar) as described previously (19).
Cell Culture-The leiomyosarcoma cell line SK-LMS-1 was obtained from ATCC and was maintained in Eagle's minimum essential medium supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin solution. Cells were incubated at 37°C in a humidified atmosphere of 5% CO 2 .
Slug siRNA Transfection and Plasmid Slug DNA Transfection-Slug siRNA (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) was used to knock down Slug expression as described previously (4,20). The pCMV6-XL5-Slug containing the Slug cDNA clone NM_003068 (OriGene Technologies, Rockville, MD) was used to overexpress Slug in SK-LMS-1 cells as reported previously (4). shows negative E-cadherin and positive vimentin (right panel). C, heat map from clustering analysis of the protein expression in leiomyosarcomas and GISTs by protein lysate array profiling. Ecadherin protein expression is significantly higher in a subset of leiomyosarcomas as indicated by the red color code. PDGFR, PDGF receptor; p-mTOR, phosphorylated mammalian target of rapamycin.
Western Blot Analysis-Western blot analysis was performed according to a standard procedure (17). The primary antibodies Ecadherin, EMA, cytokeratin AE1/AE3, and pan-CK were purchased from Santa Cruz Biotechnology, Inc., and the Slug and N-cadherin were from LifeSpan Biosciences (Seattle, WA).
Statistical Analyses-Data from the protein lysate array, mRNA microarray, and TMA were analyzed according to methods described previously (14 -15, 17-19). A hierarchical clustering algorithm based on a robust least square estimator, which could provide the most accurate quantification of the protein lysate microarray data, was used to identify significantly different protein expression between leiomyosarcomas and GISTs according to previous description (17,18).
For mRNA expression data of leiomyosarcoma samples, K-means clustering was used to determine the epithelial-like (E) group and mesenchymal-like (M) group according to the mRNA expression level of different gene markers (Table I). Specifically, samples were repeatedly clustered into high and low expression groups for each gene in the table using K-means with a squared Euclidean distance measure. Samples that were grouped into high expression clusters for epithelial markers more frequently than into low expression clusters were assigned to the E group; similarly, samples displaying frequent high expression of mesenchymal markers were assigned to the M group. Kaplan-Meier survival analysis was used to quantify survival outcomes, and the log rank (Mantel-Cox) test was used to determine whether there was a significant difference in survival probability between the E and M groups. Pearson's pairwise correlation was used to measure the correlation of E-cadherin with other factors (Table II) according to their mRNA expression levels across all leiomyosarcoma samples.
Kaplan-Meier survival analysis was also used to show the different survival of patients with different E-cadherin protein expression on TMA. The correlation of the E-cadherin and Slug protein expression on TMA was determined by Spearman correlation. Analysis of vari-ance was performed to determine the cell proliferation, invasion, and migration in cell line experiments. p Ͻ 0.05 was considered statistically significant for all tests.

E-cadherin Is Expressed in a Subset of Leiomyosarcomas,
Suggesting MErT-When we compared E-cadherin protein expression between leiomyosarcomas and GISTs by reversephase protein lysate array, we surprisingly found that a fraction of leiomyosarcomas showed E-cadherin-positive expression (Fig. 1A). Because leiomyosarcoma is a mesenchymal tumor type and there was no report of E-cadherin expression in this sarcoma type, we first sough to confirm this observation. Western blotting of the same tissue samples was performed. The results showed that these leiomyosarcoma samples had E-cadherin expression or both E-cadherin and vimentin expression but no c-KIT expression, whereas GISTs showed vimentin and c-KIT expression but no E-cadherin expression (Fig. 1B).
Consistently, hierarchical clustering of protein lysate array data revealed that E-cadherin expression was significantly higher in leiomyosarcomas than in GISTs (p Ͻ 0.05) where E-cadherin expression was rarely detected (Fig. 1C). We next sought to confirm that the E-cadherin protein expression was not due to possible epithelial cell contamination in the protein lysate used for protein lysate array and Western blotting.
For this purpose, we constructed a TMA with the same leiomyosarcoma and GIST samples to detect E-cadherin protein immunohistochemically. To enhance the validity of the evidence, other epithelial cell markers such as EMA, pan-CK, and AE1/AE3 were also detected by immunohistochemical staining on the serial TMA sections. In concordance with the results from the protein lysate array and Western blotting, many leiomyosarcoma tissues on the TMA showed expression of epithelial cell markers ( Fig. 2A). Kaplan-Meier analysis of the survival data from leiomyosarcoma patients with positive E-cadherin protein expression revealed that these patients had significantly better survival duration than patients with negative E-cadherin expression (Fig. 2B).
Because we had transcriptome data from the same set of leiomyosarcoma samples (14), we performed an additional analysis using these data. We compiled lists of genes that are known to be epithelial cell or mesenchymal cell markers (Table I). Kaplan-Meier survival analysis showed that leiomyosarcoma patients with elevated mRNA levels of epithelial marker genes (E group) had higher survival rates than patients with higher mRNA levels of mesenchymal cell marker genes (M group) with the p value approaching a significant level set by 0.05 (Fig. 2C). These combined data showed that some leiomyosarcomas exhibit MErT phenotype, and this transition is linked to improved survival.

TABLE I Epithelial and mesenchymal cell marker genes used in transcriptome analysis
next sought to elucidate the mechanism of MErT in leiomyosarcoma. In synovial sarcoma, glandular epithelial differentiation has been reported, and the potential mechanism for MErT was shown to be the increased expression of E-cadherin due to suppression of Slug by the fusion protein SYT-SSX1 (12). In leiomyosarcoma, this fusion event was not detected (21). To determine which factors regulate the observed MErT in leiomyosarcoma, we examined the transcript levels of 15 genes (Table II), including SNAI1, Slug, and TWIST1, that are commonly involved in epithelial or mesenchymal cell differentiation or EMT and MErT processes (22). Pearson's pairwise correlation analysis of the expression patterns showed that Slug and N-cadherin expression had a significantly inverse correlation with E-cadherin (Fig. 3, A and B). It was not a surprise that there was an inverse correlation between E-cadherin and N-cadherin (CDH2) because N-cadherin is a mesenchymal marker. A published study of synovial sarcoma suggested that MErT was regulated by Slug and SNAI1 (12). Thus, based on our mRNA analysis showing an inverse correlation between Ecadherin and Slug, we hypothesized that MErT in leiomyosarcoma was regulated by Slug.
To test our hypothesis, we used our TMA to determine the frequency and pattern of E-cadherin and Slug expression. As shown in Fig. 3C, protein expression of E-cadherin and Slug also showed a significant inverse correlation in leiomyosarcomas ( 2 ϭ 32.24; p Ͻ 0.001; Spearman correlation, Ϫ0.813) consistent with the results from mRNA analysis. Thus, these concordant results suggested that Slug could be important in the mechanism of epithelial differentiation in leiomyosarcoma.

Exogenous Modulation of Slug Inversely Correlates with E-cadherin Expression and Regulates
MErT-To further validate our hypothesis about the role of Slug in MErT in vitro, we used the leiomyosarcoma cell line SK-LMS-1 to investigate whether exogenous modulation of Slug inversely regulates E-cadherin expression. We first examined the basal expression levels of both endothelial and mesenchymal cell markers by Western blotting. Of interest, we found that the SK-LMS-1 leiomyosarcoma cell line expressed both cell lineage markers, including epithelial cell marker E-cadherin (Fig. 4A), suggesting that cell lineage infidelity already occurs in SK-LMS-1 cells. Colon cancer cell lines SW480 (epithelial lineage) and SW620 (the metastatic cells from the same patient of SW480 with EMT process) and Ewing sarcoma cell line TC71 (mesenchymal lineage) were used as controls. We also detected Slug expression in SK-LMS-1.
First, we down-regulated Slug expression by Slug siRNA transfection to determine whether there would be an alteration of E-cadherin expression. After Slug down-regulation, E-cadherin expression was significantly increased with a concomitant decrease in the mesenchymal cell markers vimentin and N-cadherin (Fig. 4B). We also found that the cellular morphology of SK-LMS-1 changed from a spindled to an oval or round shape (Fig. 4C). Additionally, we found that Slug siRNA treatment resulted in a significant decrease in cell proliferation (24-, 48-, and 72-h time points; p Ͻ 0.05) (Fig. 4C), invasion (24-and 48-h time point; p Ͻ 0.05) (Fig.  4D), and migration (24-and 48-h time points; p Ͻ 0.05) (Fig. 4E).
Next, we overexpressed Slug by transfecting a vector expressing Slug cDNA into SK-LMS-1. Contrary to the results with Slug siRNA interference, increased Slug expression induced a significant decrease in E-cadherin and an increase in the mesenchymal markers vimentin and N-cadherin (Fig. 4F).
In addition to the inverse correlation between Slug and Ecadherin, the increased Slug increased the cell proliferation, invasion, and migration, although the differences were not as marked as in the siRNA interference experiments. These in vitro data suggest that the exogenous modulation of Slug in SK-LMS-1 inversely regulated E-cadherin expression and cellular phenotype.

DISCUSSION
In this study, proteomics analysis revealed MErT as a novel phenotype in leiomyosarcoma. Immunohistochemical and transcriptome analyses validated this phenomenon at the cellular and molecular levels. We further demonstrated that MErT in leiomyosarcoma is regulated by Slug. To our knowledge, this is the first characterization of MErT in leiomyosarcoma.
Notably, our analyses provided supporting evidence that MErT is also a clinically relevant process. Specifically, MErT is associated with better survival in leiomyosarcoma. This is consistent with the results from in vitro experiments where induction of epithelial differentiation caused leiomyosarcoma cells to have decreased cell migration and proliferation. This is in contrast with EMT where loss of E-cadherin and acquisition of a mesenchymal phenotype are associated with increased cancer cell migration and proliferation and thus tumorigenesis and progression (1,2). A central regulatory process of EMT is modulation of E-cadherin expression by a group of transcrip- tional repressors (Zeb-1, Zeb-2, Twist, Snail, and Slug) that recruit histone deacetylases to E-box elements that are located within the E-cadherin promoter (23,24). Constitutive expression of these E-cadherin repressors has been shown to augment and maintain the mesenchymal and invasive phenotype while ensuring the survival of micrometastatic cells by suppressing premature senescence and apoptosis (23)(24)(25)(26). Acquisition of EMT features has been associated with chemoresistance, which could give rise to recurrence and metastasis after standard chemotherapeutic treatment (23)(24)(25)(26)(27). In the present study, we showed that increased E-cadherin expression and decreased E-cadherin repressor Slug at least in vitro resulted in inhibition of cell migration/invasion and proliferation in leiomyosarcoma, suggesting that E-cadherin may not function as a tumor suppressor gene only in epithelial cells but also in mesenchymal tumors, although the absence of E-cadherin expression in normal mesenchymal cells is a differentiation event.
As introduced earlier, MErT has been reported previously in another type of human sarcoma, synovial sarcoma, which is characterized by two specific gene fusion events between SYT gene and SSX1 or SSX2 genes (12). Saito et al. (12) showed that SYT-SSX1 and SYT-SSX2 interfere with repression of E-cadherin by Snail and Slug. The SYT-SSX1 fusion product has been proposed as a diagnostic and prognostic marker for synovial sarcoma (28,29). Cagle et al. (29) have pointed out a much more favorable prognosis of biphasic "highly glandular" synovial sarcomas where greater than 50% of the tumor surface area has glandular differentiation. Thus, the previous observation in synovial sarcomas by others and our current observation in leiomyosarcoma are consistent in the notion that MErT represents a beneficial clinical prognosis in mesenchymal tumors, although future studies in other sarcoma types will be needed to determine whether this is indeed the case.
Our studies showed that Slug is an important regulator of MErT and thus a potential therapeutic target in leiomyosarcoma. Creighton et al. (30) suggested that targeting proteins involved in EMT may provide a therapeutic strategy for breast cancer patients. Thiery et al. (23) also proposed that regulating the activity of E-cadherin repressors may be an obvious strategy to suppress cancer progression. A recent investigation of neuroblastoma showed that the down-regulation of Slug increased the sensitivity to apoptosis induced by imatinib mesylate, etoposide, and doxorubicin (31). Furthermore, Slug-deficient cells are radiosensitive to DNA damage treatment (32). Our current study suggests that Slug may represent a therapeutic target for leiomyosarcoma, and inhibition of Slug may lead to MErT in leiomyosarcoma and improved response to chemotherapy.