Identification of pathways associated with invasive behavior by ovarian cancer cells using multidimensional protein identification technology (MudPIT)

Abstract

Proteomic profiling has emerged as a useful tool for identifying tissue alterations in disease states including malignant transformation. The aim of this study was to reveal expression profiles associated with the highly motile/invasive ovarian cancer cell phenotype. Six ovarian cancer cell lines were subjected to proteomic characterization using multidimensional protein identification technology (MudPIT), and evaluated for their motile/invasive behavior, so that these parameters could be compared. Within whole cell extracts of the ovarian cancer cells, MudPIT identified proteins that mapped to 2245 unique genes. Western blot analysis for selected proteins confirmed the expression profiles revealed by MudPIT, demonstrating the fidelity of this high-throughput analysis. Unsupervised cluster analysis partitioned the cell lines in a manner that reflected their motile/invasive capacity. A comparison of protein expression profiles between cell lines of high (group 1) versus low (group 2) motile/invasive capacity revealed 300 proteins that were differentially expressed, of which 196 proteins were significantly upregulated in group 1. Protein network and KEGG pathway analysis indicated a functional interplay between proteins up-regulated in group 1 cells, with increased expression of several key members of the actin cytoskeleton, extracellular matrix (ECM) and focal adhesion pathways. These proteomic expression profiles can be utilized to distinguish highly motile, aggressive ovarian cancer cells from lesser invasive ones, and could prove to be essential in the development of more effective strategies that target pivotal cell signaling pathways used by cancer cells during local invasion and distant metastasis.

---

Introduction

Among gynecological cancers, epithelial ovarian cancer is the most frequent cause of death in North American women. It has a poor prognosis as it is typically diagnosed subsequent to intraperitoneal spread, and treatment efforts are undermined by the eventual recurrence of chemo-resistant disease. Ovarian cancer metastasis occurs locally, within the peritoneal cavity, through surface shedding of tumor cells. Malignant ascites, which often accumulates during disease progression, contains inflammatory factors that promote the peritoneal adhesion, motility and invasion of tumor cells into the collagen I-rich sub-mesothelial stroma, such that a self-promoting cycle of metastasis ensues. The attachment and spread of these cancer cells onto peritoneal surfaces leads to adhesions that cause malfunctioning of vital peritoneal organs, and mortality. Hence an understanding of the molecular events that promote metastasis is paramount to the development of therapeutic strategies that will prolong patient survival.

Proteomic analyses have recently emerged as valuable strategies for identifying molecular alterations in a variety of disease states, including cancer¹ and mass spectrometry-based approaches have proven to be the method of choice for detecting unknown proteins and comparing expression profiles.² We have previously demonstrated MT1-MMP (also known as MMP-14) to be a critical determinant of collagen I invasion by ovarian cancer cells.³ Whereas MT1-MMP expression and activity were found to be essential for collagen I degradation, factors that promote additional determinants of metastasis including peritoneal adhesion and motility have yet to be identified.

In this study, multidimensional protein identification technology (MudPIT) analysis was used to reveal the proteomic profile of six ovarian cancer cell lines. Cell migration and Matrigel penetration in these cells were evaluated and their protein expression profiles compared. Unsupervised cluster analysis separated the cell lines in a manner that correlated with their motile/invasive capacity. This unique application of MudPIT, in combination with bioinformatics, indicated an up-regulation of proteins that were associated with promoting actin cytoskeletal dynamics, integrin function, and proteolysis in the invasive/motile cell lines. A subset of these regulated proteins interacted in a functional network that would promote aggressive ovarian cancer cell behavior.

Results

Identification of proteins expressed by ovarian cancer cell lines

MudPIT analysis identified 2365 proteins with high confidence in whole cell lysates from HOC-7, HEY, ES-2, OVCA429, SKOV-3 and OVCAR-3 ovarian cancer cell lines (Table 1). Several proteins mapped to the same gene and these were collapsed, resulting in 2245 unique gene products for quantitative comparison by spectral counting (SpC) and comparative proteomics analysis (Supplementary Table S1/S2/S3).†

Table 1 Number of proteins detected with at least 2 unique peptides in ovarian ovarian cancer cell lines by MudPIT analysis

Cell line	Proteins	Unique peptides	Reverse proteins
HOC-7	1294	8122	3
HEY	1471	9514	5
OVCAR-3	1441	8775	4
OVCA429	1260	7374	2
SKOV-3	1503	9600	6
ES-2	1500	10 185	1
Total	2365	18 900	21

---

Hierarchical cluster analysis correlates with motile phenotype

Unsupervised hierarchical cluster analysis of the proteomic datasets separated HOC-7, HEY, ES-2 and OVCA429 cells (group 1) from SKOV-3 and OVCAR-3 cells (group 2) in the first branching of the dendrogram (Fig. 1A). This clustering pattern was observed using several algorithms, including Spearman rank correlation, Euclidean distance, uncentered correlation, and Kendall’s tau. This finding was also independently corroborated by binary tree-structured vector quantization (BTSVQ) analysis,⁴ both by k-means clustering (data not shown) and by self-organizing maps (SOMs) (Fig. 1B). BTSVQ combines SOMs and partitive k-means clustering in a complementary fashion—samples are clustered using k-means, while proteins are organized into clusters using SOMs. The analysis starts by taking full profiles for each cell line. The algorithm then partitions the data using the standard k-means algorithm in sample space, where k is kept constant at 2. Iteratively applying the algorithm and using evaluation of variance as a stopping criterion, a binary tree is generated. The SOM algorithm is then used to cluster the protein space. The cluster structure in protein space is visualized using component planes of the already computed SOMs, and mapping them to a colour scheme. Furthermore, as evident from the component planes of SOMs in Fig. 1B, replicate runs were highly similar, providing evidence of robust profiling and high specificity of MudPIT analysis. The proteomic profile of HOC-7 cells was sufficiently different to cause its segregation at a subsequent level of clustering. This likely relates to the highly epithelial character of HOC-7 cells, which is in contrast to the fibroblast-like morphology of the other three cell lines (OVCA429, HEY, and ES-2) with which it had initially partitioned (Fig. 1C). Notably, the grouping revealed in the present study mirrored the differential capacities of the cell lines for motility and Matrigel penetration (Fig. 1D). We have previously reported a similar separation of these cell lines based upon their ability to invade an acid-extracted collagen type I matrix.³

Fig. 1 Unsupervised hierarchical clustering of the six ovarian cancer cell lines based on their proteomic profiles. (A) Hierarchical cluster analysis based on normalized spectral counts for proteins identified with high confidence. (B) Self-organizing maps generated by BTSVQ analysis provides a visual comparison of complete proteomic profiles for each of the cell lines and each of the MudPIT runs. (C) Cell morphology as shown by phase-contrast microscopy. (D) Capacity for cell migration as determined by scratch-wound healing and by transwell Matrigel™ penetration. Bars represent means ± SE. of 3–8 replicates, normalized to HEY cells. Bars with different letters (Matrigel) or Roman numerals (scratch migration) are significantly different from one another as determined by Tukey’s HSD test (p < 0.05).

Validation of protein profiles supports the high fidelity of MudPIT analysis

The reproducibility of the MudPIT technique was evident from the similarities between the six MudPIT runs performed for each cell line, and we made use of these multiple replicates for semi-quantitative comparison based on normalized spectral counts. Repeat MudPIT analyses allowed for the robust comparison of the proteome of the six analyzed cell lines (i.e., use of a t-test comparison) and increased the opportunity for detection of the less abundant proteins (i.e., random sampling effect).⁵

Western blot validations for selected proteins were consistent with the expression patterns determined by MudPIT in every case analyzed, indicative of the reliable nature of relative quantification based on the use of normalized spectral counts. A more precise quantification could have been achieved by labeling with stable isotopes, such as ICAT (isotope-coded affinity tags),⁶ICPL (isotope-coded protein label),⁷SILAC (stable isotope labeling with amino acids in cell culture)⁸ and iTRAQ (isotope tags for relative and absolute quantitation).⁹ Although these methodologies provide a more reliable quantification, quantitative proteomics by spectral counting has been used extensively by us^10,11 and others^12,13 and we decided to use spectral counting for its simplicity.

For the present study, relative quantification using spectral counting was sufficient for the elucidation of differences in expression patterns between the two groups of cell lines. We have validated seven proteins (Fig. 2). Zyxin was validated using real-time RT-PCR because the two commercially available antibodies selected for Western blot analysis yielded numerous non-specific bands. In addition, we have previously characterized the expression patterns of MT1-MMP, E-cadherin and N-cadherin in these cell lines, which are reflected by the MudPIT results.³

Fig. 2 Validation of proteomic profiles of selected proteins by Western blot analysis or real time RT-PCR. (A) Bar graphs of normalized spectral counts generated by MudPIT for the six ovarian cancer cell lines for those proteins selected for validation. Bars represent the mean ± SE of the six MudPIT replicates. (B) Western blot analysis of the corresponding proteins . β-tubulin verifies equal loading. (C) Validation of zyxin expression by real-time RT-PCR. Bars represent the mean ± SE of three samples per cell line.

Differential expression of proteins associates with cell motility, adhesion, integrin signaling, and known invasive cancer cell pathways

We have used the GOFFA system (gene ontology for functional analysis) to determine biological processes for the proteins that were differentially expressed between the two groups of cells. Differentially expressed proteins were defined as those in which the average spectral counts for each group exceeded a 2-fold difference, with a p-value less than 0.01 (2-tailed t-test). Based on this criterion, 196 proteins were found significantly upregulated in group 1 and 104 proteins were upregulated in group 2 (or downregulated in group 1) (Supplementary Table S4A† ). The gene ontology analysis of these significantly differentially expressed proteins between group 1 and group 2 is shown in Supplementary Table S4B.† These proteins are graphically represented in black (Fig. 3A) as a contrast to the other 1945 proteins identified that were not differentially expressed (white). Similar strategies for the visualization of spectral counting as a measure for relative quantification have recently been reported in the literature.^14,15 The proteins unique to each group fall along the 0.01 value on each axis (the denominator value substituted for 0-values in the data analysis to avoid division by zero). GO-terms associated with cell motility, cell adhesion and integrin signaling were significantly enriched only within the subset of proteins upregulated in group 1 (Fig. 3B), suggesting that the differential expression of proteins promoting invasive behavior are an important contributor to the clustering pattern (Fig. 1). Table 2 shows differentially expressed proteins that may contribute to the enhanced invasive phenotype of group 1 cell lines, identified either through GO analysis or literature search.

Fig. 3 Comparison of protein profiles between cell groups derived from unsupervised hierarchical cluster analysis . (A) Scatter plot of the average normalized spectral counts (group 1 versus group 2) for all proteins . The input values of the plot are the average of the normalized spectral counts of group 1 cells versus that of group 2 cells. Proteins that were significantly differentially expressed (≥2-fold change, p < 0.01) are indicated in black. Proteins that were uniquely expressed by only one group fall along the 0.01 value (substituted in place of 0 in the original data set) of its respective axis. (B) Heat map analysis generated by hierarchical cluster analysis of proteins that were significantly changed between the two groups (≥2-fold change, p < 0.01). Within each cell line, the average of the normalized spectral counts from each MudPIT run (per corresponding protein ) was expressed as a ratio of the total (from all cells), such that their total added to 1. These subsets of proteins were analyzed for significantly enriched gene ontology terms (biological processes, p < 0.01, E > 2.0) using GOFFA. Gene ontology terms associated with cell motility, adhesion, and integrin signaling were found only in the subset of proteins enriched in group 1.

Table 2 Proteins associated with cell motility are upregulated in Group 1 ovarian cancer cells. Selected examples and their normalized spectral counts (±SE) are shown

Accession	Group 1				Group 2		Protein
	HOC-7	HEY	ES-2	OVCA429	SKOV-3	OVCAR-3
IPI00018274	1.34 ± 0.33	3.36 ± 0.79	1.25 ± 0.24	4.38 ± 0.32	ND	ND	Isoform 1 of EGFR precursor
IPI00218398	0.35 ± 0.22	2.56 ± 0.46	ND	2.06 ± 0.49	ND	ND	MMP-14 precursor
IPI00221224	7.28 ± 0.73	23.85 ± 1.75	7.33 ± 1.37	9.87 ± 1.32	0.89 ± 0.21	ND	Aminopeptidase N
IPI00009342	3.15 ± 0.67	4.22 ± 0.55	1.65 ± 0.34	1.68 ± 0.57	ND	ND	IQGAP1
IPI00418169	54.50 ± 7.06	42.98 ± 4.34	46.31 ± 4.86	101.44 ± 7.67	10.59 ± 0.70	22.03 ± 1.68	Annexin A2
IPI00645194	14.87 ± 0.40	24.66 ± 1.76	13.97 ± 1.17	15.59 ± 1.32	5.14 ± 0.68	4.98 ± 1.07	Integrin-β1 isoform 1A precursor
IPI00020513	0.71 ± 0.44	5.95 ± 0.88	9.11 ± 1.62	8.72 ± 1.69	2.17 ± 0.74	2.54 ± 0.79	Zyxin
IPI00294578	13.00 ± 1.16	21.12 ± 1.92	13.54 ± 1.69	18.70 ± 1.20	2.23 ± 0.23	0.40 ± 0.26	Isoform 1 of tTG2
IPI00013976	1.53 ± 0.35	2.77 ± 0.58	5.08 ± 1.04	4.02 ± 0.47	ND	ND	Laminin-β1 chain precursor
IPI00013808	23.82 ± 1.68	30.65 ± 1.18	18.93 ± 2.54	32.37 ± 1.04	10.61 ± 1.72	15.97 ± 1.35	α-Actinin-4
IPI00297160	9.85 ± 1.14	10.80 ± 1.26	7.10 ± 0.63	4.19 ± 0.33	3.80 ± 0.62	2.44 ± 0.76	CD44 Antigen isoform 4 precursor
IPI00219365	12.87 ± 1.87	37.14 ± 3.89	33.93 ± 2.44	16.10 ± 0.95	7.62 ± 0.62	10.86 ± 1.27	Moesin
IPI00219301	1.56 ± 0.52	3.61 ± 1.05	1.48 ± 0.59	2.91 ± 0.79	ND	ND	MARCKs
IPI00215995	6.70 ± 0.70	17.15 ± 1.96	4.55 ± 0.69	6.24 ± 1.24	2.45 ± 0.67	1.67 ± 0.33	Integrin-α3A precursor
IPI00163187	6.89 ± 0.56	18.70 ± 1.23	8.15 ± 0.95	18.48 ± 1.15	6.66 ± 0.60	4.65 ± 1.37	Fascin
IPI00178352	0.53 ± 0.23	2.46 ± 0.77	11.94 ± 1.84	8.10 ± 0.70	ND	ND	Isoform 1 of Filamin-C
IPI00013744	12.00 ± 1.35	1.35 ± 0.49	2.28 ± 0.67	3.43 ± 0.66	ND	ND	Integrin α2 precursor
IPI00298281	0.68 ± 0.21	1.10 ± 0.45	1.91 ± 0.54	3.74 ± 0.73	0.50 ± 0.23	ND	Laminin-γ1 chain precursor
IPI00015102	13.99 ± 1.50	12.01 ± 1.71	8.17 ± 0.77	19.22 ± 1.88	2.33 ± 0.64	3.94 ± 1.13	CD166 antigen precursor
IPI00215997	3.82 ± 0.62	2.75 ± 0.33	1.30 ± 0.36	4.80 ± 0.63	ND	ND	CD9 antigen

---

The utility of pathway analysis in combination with expression data has recently been emphasized^16,17 to reveal potential functional interactions between multiple candidate proteins . Functional interactions between the two groups of differentially expressed proteins were examined using the KEGG pathway database (http://www.genome.ad.jp/kegg)¹⁸ and the interologous interaction database ver. 1.71 (I²D; http://ophid.utoronto.ca/i2d) of known and predicted protein –protein interactions.¹⁹ The rational for this analysis was to obtain a systems-wide overview of affected signaling pathways, potentially regulated in the more invasive/motile cell lines (Fig. 4). First, the panel of the 196 proteins significantly up-regulated and the 104 proteins significantly down-regulated were compared against the KEGG database to identify the most significantly enriched signaling pathways, taking into account the number of proteins in the KEGG pathway, as well as a probability that a protein can map to a KEGG pathway by chance alone [see Experimental Procedures for details]. The KEGG pathway analysis showed a significant enrichment (p < 0.05 using Student t-test) of 13 signaling and 6 cancer pathways out of 37 pathways tested. The most significantly enriched signaling pathways include extra-cellular matrix receptor signaling (p = 2.4E − 24), focal adhesion signaling (p = 1.9E − 15), and actin cytoskeleton (p = 9.6E − 15), followed by adherens junctions (p = 6.0E − 09), tight junctions (p = 5.1E − 06), PPAR signaling (p = 1.1E − 05), calcium signaling (p = 2.6E − 05), phosphatidylinositol (p = 2.8E − 05), cell adhesion (p = 0.0002), p53 (p = 0.003), and gap junction pathways (p = 0.02). The most significantly enriched cancer pathways included small cell lung cancer (p = 1.4E − 18), glioma (p = 4.7E − 10), and colorectal cancer (p = 5.9E − 05), followed by endometrial cancer (p = 0.0005), prostate cancer (p = 0.009), and thyroid cancer pathways (p = 0.05) (Fig. 4A). The proteins of the three most enriched cell signaling pathways were used to search the I²D database¹⁹ for known and predicted protein –protein interactions. This combined network includes 80 of the 196 significantly up-regulated proteins ,²³ of the 104 significantly down-regulated proteins , and 328 proteins from the three selected cell signaling pathways. Networks from significantly enriched pathways were visualized using NAViGaTOR ver. 2.0 (network analysis, visualization and graphing, Toronto; http://ophid.utoronto.ca/navigator). To decrease graph complexity, we have reduced the final network to show only the direct interactors of the 103 significantly up- and down-regulated proteins and the three selected signaling pathways. Supplementary Table S5 lists the proteins and their corresponding SwissProtein identifiers for proteins overlapping with the 3 most enriched KEGG signaling pathways.†

Fig. 4 KEGG pathway and protein network analysis indicated functional interplay between proteins upregulated in group 1 cells. (A) The 196 significantly up-regulated proteins were compared against the KEGG database of 37 key cell signaling and cancer pathways. The analysis showed an enrichment of several pathways, and the most significant are depicted here. In order to estimate the significance of this result, we have considered the size of each KEGG pathway (i.e., number of proteins that form it) and also generated 1000 lists of 196 random proteins each and mapped them to the same 37 KEGG pathways (background) to identify significantly enriched pathways. (B) The key component proteins of the three most enriched cell signaling pathways from the KEGG database (actin cytoskeleton, ECM receptor, and focal adhesion) were used to search the I²D database ver. 1.71 for known and predicted protein –protein interactions. The resulting network was then visualized using NAViGaTOR ver. 2.0. Only the direct interactors of 80 of the 196 significantly up-regulated proteins and 23 of the 104 significantly down-regulated proteins and the three signaling pathways (partially translucent gray edges) are displayed. Direct interactions among significant proteins are highlighted as thicker blue edges. Blue nodes represent the key elements of the actin cytoskeleton pathway, yellow nodes represent the ECM receptor pathway, and the purple nodes represent the focal adhesion pathway. The common elements of the focal adhesion and ECM receptor pathways are represented by pink nodes, while the common elements of the focal adhesion and actin cytoskeleton pathways are represented by turquoise nodes. The orange nodes represent proteins that overlap all three pathways. Overlapping proteins from the main list (103 of the 300) are highlighted in green triangles (down-regulated) and red triangles (up-regulated), and labeled with SwissProtein protein names, and those belonging to the key components of the KEGG signaling pathways (328 proteins when combined) are color-coded to show pathway membership.

Discussion

Cancer cell behavior may become altered with the passage of time, and thus characterization of tumourgenicity at the time cell lines were derived (often decades earlier) may not reflect their current behavior. In addition to our characterization of cell motility and Matrigel penetration in the current study, we have recently also characterized the in vitrocollagen I invasion of the six cell lines studied,³ and a recent study provides information on the intraperitoneal metastatic potential of five of these cell lines.²⁰ Notably, cluster analysis mirrored the differential capacities of the cell lines for motility and Matrigel penetration (Fig. 1D). This clustering pattern is also consistent with the invasive capacities that we have documented for these cell lines: in contrast to HEY, OVCA429 and ES-2 cells, the SKOV-3 and OVCAR-3 cells were unable to degrade and invade a collagen I barrier.³ Moreover, the recent in vivo behavior of these cell lines is also reflected in this grouping, as intraperitoneal injection of HEY, ES-2, or OVCA429 cells consistently formed lethal tumors, whereas SKOV-3 occasionally (1/3), and OVCAR-3 never formed intraperitoneal tumors.²⁰ Histological subtype was not a determinant of the clustering pattern of our six cell lines, as tumors formed by both OVCA429 and SKOV-3 cells were of clear-cell histology,²⁰ yet OVCA429 clustered with the HEY and ES-2 cell lines, which formed undifferentiated intraperitoneal tumors. The fact that unsupervised clustering reflected the invasive capacity of the cell lines rather than the histological subtype implies that proteins involved in metastatic behavior comprise a significant subset of those identified, and likely result from an up-regulation of pro-metastatic signaling cascades.

Many of the proteins we have identified have not been previously revealed by gene expression profiling of ovarian tumors in studies attempting to identify critical genes predictive of outcome.^21–24 It is important to note that the profiles we have obtained are of malignant epithelial cell lines reflecting differences in the cancer cells themselves, in contrast to published gene array studies performed on tumour biopsies containing up to 30% stromal tissue. The reactive stroma is a highly metabolically active tissue; thus, even small amounts would contribute significantly to whole tumour gene expression, masking differences between the cancer cells. The presence of a reactive stroma distinguishes invasive tumours from benign and non-cancerous ovarian tissue, so its differential presence would be an important contributor when comparing such samples. In fact, there is little agreement or overlap between gene candidates identified within the tumour gene expression studies themselves.¹⁶ In our network analysis we identified several key interacting proteins of the actin cytoskeleton, focal adhesion, and extracellular matrix (ECM) signaling pathways enriched in motile ovarian cancer cells. The actin cytoskeleton interacts with ECM and intracellular molecules via focal adhesions, which are regulators of key signal transduction events.

ECM-Modifying enzymes are enriched in motile ovarian cancer cells

MT1-MMP was expressed only by group 1 cells, consistent with our previous characterization of this panel of cells.³ Specifically, only those ovarian cancer cell lines that expressed MT1-MMP (HEY, OVCA429, ES-2 and HOC-7: group 1 cells) were capable of degrading a collagen I barrier. Moreover, ectopic expression of MT1-MMP in SKOV-3 and OVCAR-3 cells was sufficient to endow them with an invasive phenotype. However, MT1-MMP expression did not alter their motile capacity. While degradation of collagen I-rich stromal matrices is an important component of peritoneal metastasis, cell adhesion and motility are also critical events, yet the pathways coordinating these processes have not been fully defined.

In addition to MT1-MMP, the matrix-modifying enzymes aminopeptidase N (APN/CD13) and protein –glutaminegamma-glutamyltransferase 2/tissue transglutaminase 2 (TG2) were enriched in the motile cells (group 1). APN, an integral membrane ectopeptidase that hydrolyzes bioactive peptides, is overexpressed in human cancers, including ovarian^25,26 and has been reported to promote invasive behavior.²⁷ TG2 had the largest differential expression between the two groups of cell lines. TG2 cross-links and stabilizes ECM components,²⁸ and has been recently shown to enhance peritoneal dissemination by modifying α5β1 integrin interaction with fibronectin in a manner that promotes ovarian cancer cell adhesion and motility.²⁹

The differential expression of E- and N-cadherins was previously eliminated as a factor that could account for the divergent motile capacities in these ovarian cancer cell lines.³ Consistent with this, in the present study the E-cadherin-expressing HOC-7 cells grouped with HEY, ES-2 and OVCA429 cells, which express the N-cadherin isoform that is associated with motile tumor cells having undergone epithelial-mesenchymal transition (EMT). Although SKOV-3 and OVCAR-3 cells also expressed N-cadherin, they exhibited poor motility, which indicates that additional unidentified factors confer a highly motile phenotype to ovarian cancer cells.

Differential expression of ECM receptors that promote peritoneal adhesion

Collagen type I is the preferred substrate for ovarian cancer cell adhesion and migration.^30,31 Tumor cell adhesion within the peritoneum occurs in areas where the collagen I-rich submesothelial matrix is exposed.^32,33 Ligation of the collagen I-binding integrin receptors stimulates signal transduction events that promote MT1-MMP translocation from intracellular stores to the cell surface, where its proteolytic activity promotes invasion.³⁴ Integrins provide a dynamic link between ECM and the actin cytoskeleton. Clustering of β1-integrins initiates signal cascades that promote cytoskeletal rearrangements needed for cell movement.³⁵ Both β1 integrin and the hyaluronan receptor CD44 mediate ovarian cancer cell adhesion to mesothelial monolayers and their associated ECM.³⁶

In addition to its potential promotion of tumor cell adhesion to the mesothelium, CD44 contributes to the invasive process by transporting MT1-MMP to the leading edge of the cell, allowing its proteolytic activity to be focused.³⁷ This translocation requires the coupling of CD44 to the actin cytoskeleton through the Ezrin-Radixin-Moesin (ERM) family of proteins (discussed below) and cytoskeletal rearrangements.³⁷ In the present study, the up-regulation of integrin subunits α2, α3, and β1, which comprise the collagen binding integrins α2β1 and α3β1, and CD44, in group 1 cells may thus contribute to the enhanced ability of these cells to invade collagen I matrices and to form intraperitoneal tumors in vivo.^3,20

Differential expression of proteins that promote cytoskeletal dynamics

An altered regulation of cytoskeletal dynamics promotes cancer cell acquisition of the migratory and invasive behaviors associated with metastasis.³⁸Proteins mediating cytoskeletal dynamics were enriched in group 1 cell lines and included zyxin, fascin, α-actinin-4, annexin 2, filamin C, moesin, myristoylated alanine-rich C-kinase substrate (MARCKS), and IQGAP1 (Table 2). As well, the previously discussed surface receptors CD44, α2β1 and α3β1 integrins interact with the surrounding ECM to mediate signal transduction pathways that stimulate the cytoskeletal dynamics promoting cell adhesion and motility.^39,40

EGFR, a member of the Erb-B family of receptor tyrosinekinases is a central player within the protein interaction network (Fig. 4B) that was up-regulated in the highly motile group 1 cell lines. EGFR functions to enhance the expression of several proteins within this network, and also participates in β1-integrin-mediated signal transduction.⁴¹ EGFR overexpression in cancers including ovarian is associated with tumor progression and metastasis⁴² and EGF stimulation of ovarian cancer cells increases protease expression and Matrigel penetration.⁴³

Several proteins involved in cytoskeletal dynamics were upregulated in the motile/invasive cell lines. Zyxin mediates cytoskeletal remodeling in response to mechanical stress, and confers a motile phenotype in cancer cells.⁴⁴ Zyxin interacts with α-actinin, which cross-links anti-parallel actin filaments in stress fibers.⁴⁵ Fascin bundles and cross-links parallel actin filaments and its overexpression in epithelial cells promotes cell protrusions, disorganized cell–cell contacts, and increased cell motility.⁴⁶ Alpha-actinin-4 localizes to the leading edge of motile cells and promotes cell motility.⁴⁷ That group 1 cells had enriched α-actinin-4 is consistent with another proteomic study indicating its up-regulation in metastatic cancers.⁴⁸ Annexin 2 promotes the plasticity of the membrane-associated actin cytoskeleton in protrusions at the leading edge of motile cells.⁴⁹ Filamins promote formation of the filamentous network of cortical actin in lamellipodia, and may couple cytoskeletal dynamics to signal transduction events through interaction with membrane receptors.⁵⁰

Moesin, an ERM family member, is elevated in poorly differentiated serous papillary ovarian adenocarcinomas.⁵¹ ERM proteins join cytoskeletal actin filaments to plasma membrane-associated proteins , including CD44, and participate in signaling events associated with invasion.⁵² MARCKS, a protein kinase-C (PKC) substrate, mediates actin cytoskeletal events that affect cell shape and motility.³⁸ The scaffold protein IQGAP1 stimulates cytoskeletal rearrangement and associated signaling events critical for cell migration,⁵³ and is overexpressed at the invasive front of ovarian tumors in association with poor prognosis.⁵⁴ We have previously detected IQGAP1 in the cell fraction of ascites fluid.¹⁷

The protein expression profiles and associated network analysis suggest an upregulation of actin cytoskeleton, ECM receptor and focal adhesion signaling is associated with an enhanced migratory/invasive capacity of ovarian cancer cells. Notably, a considerable overlap was evident between proteins prominent in cytoskeletal dynamics and those up-regulated in our group 1 cell lines. Of these proteins , filamin C, integrin-β1, ERM, annexin 2, and TG2 were also up-regulated by TGF-β in a non-small cell lung cancer cell line (A-539), in association with an enhanced invasive capacity.⁵⁵ This suggests that these proteins , perhaps co-regulated by cytokines such as TGF-β, function in a coordinated manner to elicit a motile phenotype. Interestingly, small cell lung cancer that is associated with early onset of local invasion and distant metastases, was identified as a significantly enriched cancer pathway when the proteins up-regulated in group 1 cells were compared against the KEGG database¹⁸ cancer (Fig. 4A). Of note, the KEGG database (http://www.genome.jp/kegg) does not currently include an ovarian cancer annotated pathway.

The elimination of individual proteins that are critical for cytoskeletal dynamics can effectively block cell motility, as has been demonstrated using siRNA to many proteins identified and discussed (e.g., zyxin, IQGAP, ERM family members). This indicates that the absence of individual proteins that participate in a complex multiprotein process can prevent the process, but would be unlikely to explain the differential behaviour observed between the two groups of cell lines, where numerous proteins involved in integrin function and cytoskeletal dynamics are differentially expressed. We did not observe an enhanced motility of SKOV-3 or OVCAR-3 cells in response to TGF-β treatment (data not shown), which may relate to differences in cell sensitivity or intracellular signaling response to this cytokine. It is comprehensible that, whereas knock-down or elimination of an essential protein will effectively block cell motility, the alterations required to endow cells with a highly motile phenotype are more complicated. Nevertheless, the up-regulation of groups of proteins that promote the actin cytoskeletal rearrangements leading to motility, as well as the integrin-ECM interactions that stimulate these rearrangements, suggest alterations in these pathways are a hallmark of highly motile ovarian cancer cells.

In summary, cancer cell invasive behavior is orchestrated by a complex interplay of pathways affecting adhesion, motility and matrix degradation, underscoring the importance of using high-throughput technologies to identify key proteins . The importance of the global MudPIT analysis used in this study is that it has enabled the identification of functional groups of proteins that may promote invasive capacity. MudPIT analysis achieved a confident identification of over two thousand proteins expressed by these ovarian cancer cell lines. The quantitative potential of this technique was sufficient to allow the relative protein expression levels to be compared between numerous cell lines. Three hundred proteins showing differential expression between cells of high versus low motile/invasive capacity were revealed using this approach.

Conclusion

In conclusion, we have presented an extensive proteomics profile of six commonly used ovarian cancer cell lines, which resulted in the confident identification of over 2300 proteins . The present study suggests that the coordinated altered expression of multiple proteins within the integrin and actin cytoskeletal pathways underlies the invasive phenotype associated with some ovarian cancer cells. Understanding how ovarian cancer cells become motile and invasive is essential for developing more effective strategies that target pivotal signaling pathways involved during local invasion and distant metastasis.

Experimental procedures

Cell culture

Human ovarian cancer cell lines HEY, SKOV-3, and HOC-7 (obtained from Dr Alexander Marks, University of Toronto, Toronto, ON), OVCA429 (from Dr Robert Kerbel, Sunnybrook Hospital, Toronto, ON), and OVCAR-3 and ES-2 (from American Type Culture Collection, Manassas, VA) were maintained in α-minimal essential media (α-MEM; GIBCO, Invitrogen Corp., Mississauga, ON) supplemented with 10% fetal bovine serum (FBS; Cansera International Inc, Etobicoke, ON), 0.017% penicillin G and 0.01% gentamycin at 37 °C and 5% CO₂.

Cell lysis and protein extraction for proteomic analysis

The cells were incubated in hypotonic lysis buffer containing 10 mM HEPES, pH 7.4 for 30 minutes on ice. The suspension was briefly sonicated and Triton-X-100 was added to a final concentration of 1.5%. The suspension was incubated at 4 °C for 30 minutes, followed by centrifugation at 14 000 rpm for 30 min at 4 °C. The supernatant was removed and used for proteomic analysis.

In-solution protein digestion

An aliquot of 150 μg of protein from each sample was precipitated overnight at −20 °C with 5-volumes of ice-cold acetone, followed by centrifugation at 21 000 g for 15 min. The protein pellet was solubilized in 8 M urea, 2 mM DTT, 50 mM Tris-HCl, pH 8.5 at 37 °C for 1 h, followed by carboxyamidomethylation with 10 mM iodoacetamide for 1 hour at 37 °C in the dark. The samples were then diluted with 50 mM ammonium bicarbonate, pH 8.5 to ∼1.5 M urea. Calcium chloride was added to a final concentration of 1mM and digested with a 1 : 25 molar ratio of recombinant, proteomics grade trypsin (Roche Diagnostics, Laval, QC) at 37 °C overnight. The resulting peptide mixtures were solid phase-extracted with Varian OMIX cartridges (Mississauga, ON, Canada) according to the manufacturers instructions and stored at −80 °C until further use.

Multidimensional protein identification technology—MudPIT analysis

A fully automated 6-cycle, 6 h MudPIT procedure was set up similar as previously described.^11,56–59 A quaternary HPLC-pump was interfaced with a LTQ linear ion-trap mass spectrometer (Thermo Fisher Scientific, San Jose, CA) equipped with a nano-electrospray source (Proxeon Biosystems, Odense, Denmark). A 100 μm inner diameter fused silica capillary (InnovaQuartz, Phoenix, AZ) was pulled to a fine tip using a P-2000 laser puller (Sutter Instruments, Novato, CA) and packed with ∼7cm of Jupiter™ 4μ Proteo 90 Å C₁₂ reverse phase resin (Phenomenex, Torrance, CA), followed by ∼5 cm of Luna® 5μ SCX 100 Å strong cation exchange resin (Phenomenex, Torrance, CA). Samples were loaded manually on separate columns using an in-house pressure vessel. As peptides eluted from the microcapillary columns, they were electrosprayed directly into the MS. A distal 2.3 kV spray voltage was applied to the microsplitter tee (Proxeon Biosystems). The MS operated in a cycle of one full-scan mass spectrum (400–1400 m/z), followed by 6 data-dependent MS/MS spectra at 35% normalized collision energy, which was continuously repeated throughout the entire MudPIT separation. The MS functions and the HPLC solvent gradients were controlled by the Xcalibur data system (Thermo Fisher Scientific, San Jose, CA).

Protein identification, validation and grouping

Raw files were converted to m/z XML using ReAdW and searched by a locally installed version of X!Tandem⁶⁰ against a human IPI (international protein index; http://www.ebi.ac.uk/IPI) protein sequence database (v3.20).⁶¹ To estimate and minimize our false positive rate the protein sequence database also contained every IPI protein sequence in its reversed amino acid orientation (target-decoy strategy) as recently described.^58,62 Search parameters were: Parent ion Δmass of 4 Da, fragment mass error of 0.4 Da, and complete carbaminomethyl modification of cystein by IAA. Only peptides passing a default log-expectation value of −1 were further evaluated (see below).

A rigorous peptide quality control strategy was applied to effectively minimize false positive identifications. Briefly, a Perl-based tool was written to calculate, on the peptide level, a user-defined false positive rate (Perl version 5). Identified peptides are binned into three charge states (+1, +2, +3), and individual X!Tandem expectation values are calculated for each charge state to minimize the number of peptides mapping to decoy sequences to a user defined percentage. For the presented study, we have set the value of total reverse spectra to total forward spectra to 0.5%, resulting in a low number of decoy sequences in the final output (21 reverse proteins in 2365 forward proteins ). The calculated expectation values were: −1.61 for +1-ions, −1.8 for +2-ions, and −2.27 for +3-ions. Only fully tryptic peptides ≥7 amino acids, matching these criteria were accepted to generate the final list of identified proteins . We only accepted proteins identified with two unique peptides per analyzed cell line. To minimize protein inference we developed a database grouping scheme, and only report proteins with substantial peptide information, as recently reported.¹⁷

Spectral counting and data normalization

Spectral counts (SpC) for every confidently identified protein in each individual MudPIT run (six MudPITs per cell line) were normalized by dividing each SpC by the sum of all SpC for that run.⁶³ This value was then multiplied by a universal factor (5655.8) based on the global average of all SpC (for all IPIs in all MudPIT runs). To avoid division by zero, a value of 0.01 was substituted for denominator values of 0. To ensure that all values in the entire dataset were treated the same, they were adjusted by 0.01. Similar normalizing schemes have recently been reported in the literature.⁶³

Normalized spectral count values were used as a semi-quantitative estimate of protein abundance, as recently described.^14,63–67 Although, proteins with high sequence homology were further evaluated as they present greater difficulties for quantitative proteomics by spectral counting.⁶⁸ To avoid this complication, two further precautions were taken into consideration: (1) proteins confidently identified (2365 unique IPI accessions) were mapped to gene identifiers and proteins mapping to the same gene identifier were combined into one single entry (e.g. isoform 1 and 2 of vacuolar protein sorting 29; VPS29). Although, identified peptides suggest the presence of both isoforms in the analyzed samples. (2) Only spectra of unambiguous peptides (e.g. peptides only mapping to one database identifier in the final list of reported proteins per cell line) were used for quantitative proteomics by spectral counting.

Hierarchical cluster analysis

Global cluster analysis was performed using open source software Cluster 3.0 (http://bonsai.ims.u-tokyo.ac.jp/∼mdehoon/software/cluster/software.htm). Normalized, natural log transformed data were clustered using the Spearman rank correlation similarity metric with average linkage and verified using Euclidean distance, uncentered correlation, and Kendall’s tau algorithms. Clusters were visualized using Java TreeView 1.0 (http://jtreeview.sourceforge.net). Unsupervised BTSVQ clustering was performed as previously described,⁴ using full profiles for each cell line.

Determination and analysis of differentially expressed proteins

Cluster analysis separated the cell lines into two groups. To analyze differences between the two groups, the average normalized SpC of group 1 was expressed as a ratio against that of the average normalized SpC of group 2. A protein was considered to be differentially expressed between the two groups if it met the following criteria: a minimum two-fold change in magnitude and a significance of p < 0.01 (2-tailed t-test, unequal variance). Differentially expressed proteins were analyzed using the GOFFA Library in the ArrayTrack open source software (http://edkb.fda.gov/westart/arraytrack) to identify significantly enriched gene ontologies (GO-terms). Criteria for enriched GO-terms were an enrichment score (E-value) minimum of 2.0 and a p-value of <0.01.

KEGG Pathway analysis and protein –protein interactions

The panel of 196 proteins significantly up-regulated in group 1 cells was compared to the KEGG database of 37 key cell signaling and cancer pathways (http://www.genome.ad.jp/kegg). The analysis showed an enrichment of several pathways, with actin-cytoskeleton, focal adhesion signaling and extra-cellular matrix receptor signaling being the most significantly enriched. In order to estimate the significance of this result, 1000 lists of 196 random proteins each were generated and subjected to the same analysis. Using this data, we have selected the three most enriched signaling pathways, as shown in Fig. 4. Proteins of these enriched cell signaling pathways from KEGG database were then used to search the I²D database ver. 1.71 for known and predicted protein –protein interactions (http://ophid.utoronto.ca/i2d).¹⁹ To reduce graph complexity, the final network shows only the direct interactors of the 196 significantly up-regulated proteins , the 104 significantly down-regulated proteins , and the three signaling pathways, that could be mapped into the I²D database. Presented nodes were color coded to match the selected KEGG pathways. Significantly dys-regulated proteins are represented as triangles and labeled with SwissProtein names (red triangles up-regulated proteins and green triangles down-regulated proteins ). The list of the proteins and their SwissProtein identifiers are presented in Supplementary Table S5.†

Western blot analysis and real-time PCR validation

Western blotting was performed as described previously.³ Briefly confluent cell cultures were lysed in 50 mM Tris-HCl, 120 mM NaCl, 0.5% NP-40, pH 7.4, containing protease inhibitor cocktail (Sigma-Aldrich, St. Louis, MO). Total protein was quantified by Bio-Rad Protein Assay (Bio-Rad Laboratories, Mississauga, ON). Equal amounts of protein (10–20 μg) were separated by 10% or 12% SDS-PAGE and transferred onto PVDF membranes (Amersham Biosciences, Oakville, ON). Blots were probed for EFGR (1 : 300, CIBA-Corning, Alameda, CA), APN (1 : 2000, clone 3D8; Santa Cruz), β1-integrin (1 : 1000, clone 12G10; AbD Serotec), β-tubulin (1 : 2000, T5168; Sigma-Aldrich), vimentin (1 : 300, M725; DAKO, Mississauga, ON), IQGAP1 (1 : 500, clone H-109; Santa Cruz), and annexin A2 (1 : 5000; BD Transduction Laboratories). Samples were reduced for all samples other than APN, β1-integrin, and EGF, which were run non-reduced. Secondary HRP-coupled antibodies (Amersham) were diluted 1:3000 and immunoreactive protein bands were detected and visualized with ECL Western Blotting Detection Reagents (Amersham) by exposing the filter blots to X-ray film (Hyperfilm ECL, Amersham Biosciences, Buckinghamshire, UK; or KODAK BioMax XAR Film, Eastman Kodak Company, Rochester, NY).

Whereas all proteins were initially analyzed by Western blotting, real-time PCR eventually validated Zyxin because the two commercially available antibodies we purchased (Aviva and Santa Cruz) yielded numerous non-specific bands. RNA was extracted from confluent cell cultures using the Absolutely RNA ® Miniprep Kit (Stratagene, Mississauga, ON) and reverse transcribed using a First Strand cDNA Synthesis Kit (Invitrogen Corp., Burlington, ON). Real Time PCR for Zyxin was performed with validated probes for human zyxin (Assay ID #Hs00170299_m1) and eukaryotic 18S endogenous control (#4319413E), using The TaqMan® Gene Expression Assay system (Applied Biosystems, Foster City, CA) according to the manufacturer’s instruction.

Scratch migration and transwell Matrigel™ penetration

Confluent cells in 6-well plates were scraped using a P-1000 plastic pipette tip. Images recorded were evaluated by the reduction in distance between opposing edges after 12 h. Transwell Matrigel experiments were conducted in 8 μm transwell chambers (Costar, Corning Inc., Corning, NY). Matrigel (BD Biosciences, Mississauga) was diluted 1 : 70, 100 μl was applied to each transwell and dried down onto the membrane overnight. Cells were resuspended in α-MEM containing 1% FBS, and 2 × 10⁵ cells in 100 μl were seeded into the upper wells. Following a 2 h incubation to allow cell attachment, α-MEM supplemented with 10% FBS was added to the lower well as a chemoattractant (500 μl). Invaded cells were quantified as previously described.³ Cells were seeded for migration assays 48 h post-transfection.

Authors’ contributions

KLS, AIE, TJB, MJR, IJ and TK were involved in the conceptualization and design of the study. KLS performed tissue culture, protein extraction, real-time PCR, and Western blotting validations. TK Performed MudPIT, TK and AI preformed protein identification, validation, and database grouping schemes. Clustering and network analysis of proteins was carried out by IJ and MA. KLS and AIE contributed equally to writing the manuscript with critical input from all authors.

Acknowledgements

Supported by a start-up grant from the Ontario Cancer Institute (TK) and by grants awarded from the CR Younger Foundation, Canada Institutes of Health Research (MOP-74726) (TJB), Natural Sciences and Engineering Research Council (MJR), US Department of Defense (DOD #W81XWH-05-1-0104) (IJ), Genome Canada through the Ontario Genomics Institute, and IBM Canada (IJ). IJ, TK, KLS and AI were supported by the Toronto Ovarian Cancer Research Network through funds raised by the Toronto Fashion Show. We thank Inga Kireeva for technical assistance. IJ and TK were supported through the Canadian Research Chair program.

References

1. Y. Kuramitsu and K. Nakamura, Proteomics, 2006, 6, 5650–5661.
2. T. I. Williams, K. L. Toups, D. A. Saggese, K. R. Kalli, W. A. Cliby and D. C. Muddiman, J. Proteome Res., 2007, 6, 2936–2962.
3. K. L. Sodek, M. J. Ringuette and T. J. Brown, Br. J. Cancer, 2007.
4. M. Sultan, D. A. Wigle, C. A. Cumbaa, M. Maziarz, J. Glasgow, M. S. Tsao and I. Jurisica, Bioinformatics (Oxford, England), 2002, 18(Suppl 1), S111–119.
5. H. Liu, R. G. Sadygov and J. R. Yates 3rd, Anal. Chem., 2004, 76, 4193–4201.
6. S. P. Gygi, B. Rist, S. A. Gerber, F. Turecek, M. H. Gelb and R. Aebersold, Nat. Biotechnol., 1999, 17, 994–999.
7. A. Schmidt, J. Kellermann and F. Lottspeich, Proteomics, 2005, 5, 4–15.
8. S. E. Ong, B. Blagoev, I. Kratchmarova, D. B. Kristensen, H. Steen, A. Pandey and M. Mann, Mol. Cell. Proteomics, 2002, 1, 376–386.
9. P. L. Ross, Y. N. Huang, J. N. Marchese, B. Williamson, K. Parker, S. Hattan, N. Khainovski, S. Pillai, S. Dey, S. Daniels, S. Purkayastha, P. Juhasz, S. Martin, M. Bartlet-Jones, F. He, A. Jacobson and D. J. Pappin, Mol. Cell. Proteomics, 2004, 3, 1154–1169.
10. A. O. Gramolini, T. Kislinger, R. Alikhani-Koopaei, V. Fong, N. J. Thompson, R. Isserlin, P. Sharma, G. Y. Oudit, M. G. Trivieri, A. Fagan, A. Kannan, D. Higgins, H. Huedig, G. Hess, S. Arab, J. G. Seidman, C. E. Seidman, B. Frey, M. Perry, P. H. Backx, P. P. Liu, D. H. Maclennan and A. Emili, Mol. Cell. Proteomics, 2007.
11. T. Kislinger, B. Cox, A. Kannan, C. Chung, P. Hu, A. Ignatchenko, M. S. Scott, A. O. Gramolini, Q. Morris, M. T. Hallett, J. Rossant, T. R. Hughes, B. Frey and A. Emili, Cell, 2006, 125, 173–186.
12. B. L. Zybailov, L. Florens and M. P. Washburn, Mol. BioSyst., 2007, 3, 354–360.
13. L. Florens and M. P. Washburn, Methods Mol. Biol. (Clifton, N.J.), 2006, 328, 159–175.
14. B. L. Zybailov, A. L. Mosley, M. E. Sardiu, M. K. Coleman, L. Florens and M. P. Washburn, J. Proteome Res., 2006, 5, 2339–2347.
15. A. O. Bailey, T. M. Miller, M. Q. Dong, C. Vande Velde, D. W. Cleveland and J. R. Yates, Anal. Chem., 2007, 79, 6410–6418.
16. R. S. Fehrmann, X. Y. Li, A. G. van der Zee, S. de Jong, G. J. Te Meerman, E. G. de Vries and A. P. Crijns, Oncologist, 2007, 12, 960–966.
17. L. Gortzak-Uzan, A. Ignatchenko, A. I. Evangelou, M. Agochiya, K. A. Brown, P. St Onge, I. Kireeva, G. Schmitt-Ulms, T. J. Brown, J. Murphy, B. Rosen, P. Shaw, I. Jurisica and T. Kislinger, J. Proteome Res., 2008, 7, 339–351.
18. H. Ogata, S. Goto, K. Sato, W. Fujibuchi, H. Bono and M. Kanehisa, Nucleic Acids Res., 1999, 27, 29–34.
19. K. R. Brown and I. Jurisica, Genome Biol., 2007, 8, R95.
20. T. J. Shaw, M. K. Senterman, K. Dawson, C. A. Crane and B. C. Vanderhyden, Mol. Ther., 2004, 10, 1032–1042.
21. J. M. Lancaster, H. K. Dressman, J. P. Clarke, R. A. Sayer, M. A. Martino, J. M. Cragun, A. H. Henriott, J. Gray, R. Sutphen, A. Elahi, R. S. Whitaker, M. West, J. R. Marks, J. R. Nevins and A. Berchuck, Int. J. Gynecol. Cancer, 2006, 16, 1733–1745.
22. M. E. Schaner, D. T. Ross, G. Ciaravino, T. Sorlie, O. Troyanskaya, M. Diehn, Y. C. Wang, G. E. Duran, T. L. Sikic, S. Caldeira, H. Skomedal, I. P. Tu, T. Hernandez-Boussard, S. W. Johnson, P. J. O'Dwyer, M. J. Fero, G. B. Kristensen, A. L. Borresen-Dale, T. Hastie, R. Tibshirani, M. van de Rijn, N. N. Teng, T. A. Longacre, D. Botstein, P. O. Brown and B. I. Sikic, Mol. Biol. Cell, 2003, 14, 4376–4386.
23. D. Spentzos, D. A. Levine, M. F. Ramoni, M. Joseph, X. Gu, J. Boyd, T. A. Libermann and S. A. Cannistra, J. Clin. Oncol., 2004, 22, 4700–4710.
24. A. Berchuck, E. S. Iversen, J. M. Lancaster, J. Pittman, J. Luo, P. Lee, S. Murphy, H. K. Dressman, P. G. Febbo, M. West, J. R. Nevins and J. R. Marks, Clin. Cancer Res., 2005, 11, 3686–3696.
25. Y. van Hensbergen, H. J. Broxterman, S. Rana, P. J. van Diest, M. C. Duyndam, K. Hoekman, H. M. Pinedo and E. Boven, Clin. Cancer Res., 2004, 10, 1180–1191.
26. P. Surowiak, M. Drag, V. Materna, S. Suchocki, R. Grzywa, M. Spaczyski, M. Dietel, J. Oleksyszyn, M. Zabel and H. Lage, Int. J. Gynecol. Cancer, 2006, 16, 1783–1788.
27. D. Fontijn, M. C. Duyndam, M. P. van Berkel, Y. Yuana, L. H. Shapiro, H. M. Pinedo, H. J. Broxterman and E. Boven, Br. J. Cancer, 2006, 94, 1627–1636.
28. E. A. Verderio, T. Johnson and M. Griffin, Amino Acids, 2004, 26, 387–404.
29. M. Satpathy, L. Cao, R. Pincheira, R. Emerson, R. Bigsby, H. Nakshatri and D. Matei, Cancer Res., 2007, 67, 7194–7202.
30. K. M. Burleson, L. K. Hansen and A. P. Skubitz, Clin. Exp. Metastasis, 2004, 21, 685–697.
31. T. L. Moser, S. V. Pizzo, L. M. Bafetti, D. A. Fishman and M. S. Stack, Int. J. Cancer, 1996, 67, 695–701.
32. S. Ghosh, Y. Wu and M. S. Stack, Cancer Treat. Res., 2002, 107, 331–351.
33. Y. Mochizuki, H. Nakanishi, Y. Kodera, S. Ito, Y. Yamamura, T. Kato, K. Hibi, S. Akiyama, A. Nakao and M. Tatematsu, Clin. Exp. Metastasis, 2004, 21, 39–47.
34. S. M. Ellerbroek, Y. I. Wu, C. M. Overall and M. S. Stack, J. Biol. Chem., 2001, 276, 24833–24842.
35. A. L. Berrier and K. M. Yamada, J. Cell Physiol., 2007.
36. K. Lessan, D. J. Aguiar, T. Oegema, L. Siebenson and A. P. Skubitz, Am. J. Pathol., 1999, 154, 1525–1537.
37. H. Mori, T. Tomari, N. Koshikawa, M. Kajita, Y. Itoh, H. Sato, H. Tojo, I. Yana and M. Seiki, EMBO, 2002, 21, 3949–3959.
38. C. Larsson, Cell. Signalling, 2006, 18, 276–284.
39. H. Ponta, L. Sherman and P. A. Herrlich, Nat. Rev. Mol. Cell Biol., 2003, 4, 33–45.
40. J. D. Hood and D. A. Cheresh, Nat. Rev. Cancer, 2002, 2, 91–100.
41. D. Liu, J. Aguirre Ghiso, Y. Estrada and L. Ossowski, Cancer Cell, 2002, 1, 445–457.
42. H. Meden and W. Kuhn, Eur. J. Obst. Gynecol. Reprod. Biol., 1997, 71, 173–179.
43. S. M. Ellerbroek, L. G. Hudson and M. S. Stack, Int. J. Cancer, 1998, 78, 331–337.
44. S. M. Sy, P. B. Lai, E. Pang, N. L. Wong, K. F. To, P. J. Johnson and N. Wong, Modern Pathol., 2006, 19, 1108–1116.
45. A. W. Crawford, J. W. Michelsen and M. C. Beckerle, J. Cell Biol., 1992, 116, 1381–1393.
46. S. Yamashiro, Y. Yamakita, S. Ono and F. Matsumura, Mol. Biol. Cell, 1998, 9, 993–1006.
47. K. Honda, T. Yamada, Y. Hayashida, M. Idogawa, S. Sato, F. Hasegawa, Y. Ino, M. Ono and S. Hirohashi, Gastroenterology, 2005, 128, 51–62.
48. Y. Wang, R. Hanley and R. L. Klemke, J. Proteome Res., 2006, 5, 907–915.
49. J. A. Koziol, A. C. Feng and J. E. Schnitzer, Anal. Chem., 2006, 78, 3203–3207.
50. Y. Feng and C. A. Walsh, Nat. Cell Biol., 2004, 6, 1034–1038.
51. A. M. Martoglio, B. D. Tom, M. Starkey, A. N. Corps, D. S. Charnock-Jones and S. K. Smith, Mol. Med., 2000, 6, 750–765.
52. P. Herrlich, H. Morrison, J. Sleeman, V. Orian-Rousseau, H. Konig, S. Weg-Remers and H. Ponta, Ann. New York Acad. Sci., 2000, 910, 106–118 ; discussion 118–120.
53. L. Y. Bourguignon, E. Gilad, K. Rothman and K. Peyrollier, J. Biol. Chem., 2005, 280, 11961–11972.
54. P. Dong, K. Nabeshima, N. Nishimura, T. Kawakami, T. Hachisuga, T. Kawarabayashi and H. Iwasaki, Cancer Lett., 2006, 243, 120–127.
55. V. G. Keshamouni, G. Michailidis, C. S. Grasso, S. Anthwal, J. R. Strahler, A. Walker, D. A. Arenberg, R. C. Reddy, S. Akulapalli, V. J. Thannickal, T. J. Standiford, P. C. Andrews and G. S. Omenn, J. Proteome Res., 2006, 5, 1143–1154.
56. T. Kislinger and A. Emili, Exp. Rev. Proteomics, 2005, 2, 27–39.
57. T. Kislinger, A. O. Gramolini, D. H. MacLennan and A. Emili, J. Am. Soc. Mass Spectrom., 2005, 16, 1207–1220.
58. T. Kislinger, K. Rahman, D. Radulovic, B. Cox, J. Rossant and A. Emili, Mol. Cell Proteomics, 2003, 2, 96–106.
59. M. P. Washburn, D. Wolters and J. R. Yates 3rd, Nat. Biotechnol., 2001, 19, 242–247.
60. R. Craig and R. C. Beavis, Bioinformatics, 2004, 20, 1466–1467.
61. P. J. Kersey, J. Duarte, A. Williams, Y. Karavidopoulou, E. Birney and R. Apweiler, Proteomics, 2004, 4, 1985–1988.
62. J. E. Elias and S. P. Gygi, Nat. Methods, 2007, 4, 207–214.
63. C. L. Roth, A. L. McCormack, A. Lomniczi, A. E. Mungenast and S. R. Ojeda, Mol. Cell. Endocrinol., 2006, 254–255, 51–59.
64. B. Cox, T. Kislinger, D. A. Wigle, A. Kannan, K. Brown, T. Okubo, B. Hogan, I. Jurisica, B. Frey, J. Rossant and A. Emili, Mol. Syst. Biol., 2007, 3, 109.
65. C. W. Turck, A. M. Falick, J. A. Kowalak, W. S. Lane, K. S. Lilley, B. S. Phinney, S. T. Weintraub, H. E. Witkowska and N. A. Yates, Mol. Cell Proteomics, 2007.
66. B. Zybailov, M. K. Coleman, L. Florens and M. P. Washburn, Anal. Chem., 2005, 77, 6218–6224.
67. S. R. Nagalla, J. A. Canick, T. Jacob, K. A. Schneider, A. P. Reddy, A. Thomas, S. Dasari, X. Lu, J. A. Lapidus, G. M. Lambert-Messerlian, M. G. Gravett, C. T. Roberts, Jr, D. Luthy, F. D. Malone and M. E. D’Alton, J. Proteome Res., 2007, 6, 1245–1257.
68. S. Dasari, L. Pereira, A. P. Reddy, J. E. Michaels, X. Lu, T. Jacob, A. Thomas, M. Rodland, C. T. Roberts, Jr, M. G. Gravett and S. R. Nagalla, J. Proteome Res., 2007, 6, 1258–1268.

---

Footnotes

† Electronic supplementary information (ESI) available: Supplementary tables. See DOI: 10.1039/b717542f

‡ These authors contributed equally to this work.

---