Ovarian cancer is a highly aggressive disease, making inhibition of adjacent organ penetration/invasion, which facilitates the dissemination of this cancer, an important therapeutic goal. Mesenchymal cell migration accompanied by EMT has been suggested to be the major invasion mechanism of ovarian cancer cells 
. However, recent studies, using 3D culture models, revealed that cancer cells can use additional strategies to invade through tissues 
. As such, cells have been shown to be capable of invading without disrupting cell-cell interactions (i.e., collective cell migration) or to penetrate the ECM without major protease or integrin dependencies (i.e., amoeboid cell migration) 
. Since ovarian cancer cells often acquire a peculiar phenotype which includes up-regulation of E-cadherin expression in contrast to its absence in normal OSE 
, we hypothesized that these cells may penetrate/invade through ECMs using alternative strategies to the well-studied mesenchymal cell migration. Therefore, the matrix remodeling capabilities of a panel of ovarian tumor cell lines was assessed using fibroblast-derived matrices, to mimic ECMs of in vivo
Different ECM topographic changes were induced by various tumor cells. A group of cells, represented by OVCAR5, appeared to have degraded the ECM. They also seemed to maintain their cell-cell contact while they invaded through ECMs () and clearly expressed cell-cell adhesion markers ( & ). The other primary group, represented by OVCAR10, appeared to modify the ECM by a massive rearrangement (e.g., contraction or accumulation) of the ECM (). These cells presented a more loose agglomeration ( & ). A potential third group, i.e., SKOV3 and UPN251, appeared to behave in a single cell manner with less clear impairment or remodeling upon ECMs (). Some of previous studies suggested that cell morphology implicates cell invasion strategies 
. However, in our study, most of tumor cell lines presented an epithelial phenotype in 2D cultures and showed rounded morphology within the matrices suggesting that cell morphology itself may not be sufficient to predict the type of ECM remodeling and subsequently, the penetration/invasion strategy. We conducted similar studies using human ovarian fibroblast-derived matrices. Although it was less apparent, tumor cells (e.g., OVCAR5 and OVCAR10) presented similar ECM modifying effect as observed in N3F- or TAF-derived matrices (data not shown). Human fibroblasts produced matrices visibly very different from N3F-derived matrices when they were fluorescently stained, i.e., substantially thinner and less uniform, making it more difficult to assess cell behavior. In addition, differences in architecture of the ECM as well as protein composition might explain differences between the ECM modifying behavior on N3F- and human fibroblast-derived matrices. For instance, cancer cells may have a limited ability to migrate through a rigid pore 
, therefore, they are likely to rely on ECM-degrading enzymes to invade through rigid ECMs. Also, cells may prefer invading through N3F-derived matrices which lack type I collagen, which is more resistant to enzymatic degradation than other ECM components (e.g., fibronectin), compared to human fibroblast-derived matrices rich in type I collagen (data not shown).
We conducted most of our studies using cells that represented the two distinct phenotypes of collective agglomerates within ECMs, OVCAR5 and OVCAR10. These cells induced strikingly different topographic changes to the ECM while making their way through these substrates ( & ). We believe that the matrix remodeling phenotype could be used to assess penetration/invasion strategies. For example, based on matrix remodeling phenotypes, OVCAR5 cells fit into a ‘collective cell migration’ category where cells migrate as groups and use ECM-degrading proteases for their invasion 
. In fact, this phenotype was also observed when OVCAR5 cells were evaluated as spheroids invading through a monolayer of mesothelial cells 
, thus confirming the relevance of our approach. Collective cell migration has also been demonstrated in colon cancer cells by Nabeshima and colleagues 
. Their study showed that the expression of ECM-degrading proteases, such as MT1-MMP and MMP-2 at the leading edge of the cell aggregates are critical for cohort migration, a type of collective cell movement 
. More recently, it was reported that podoplanin, a plasma membrane glycoprotein, induces tumor cell migration and invasion without disrupting E-cadherin-mediated cell-cell junction both in vitro
and in a transgenic model of carcinogenesis in vivo 
. In contrast, ECM rearrangement mediated by OVCAR10 cells, similar to patterns of collagen contraction induced by mesenchymal cells 
did not seem to either rely on proteases or present tight cell-cell interaction as seen in OVCAR5. In addition, protein localization patterns of E-cadherin, β1
-integrin, and F-actin of the two cell lines suggested that OVCAR5 and OVCAR10 cells indeed present epithelial vs.
mesenchymal phenotypes, respectively (). β1
-integrin expression at cell-cell contact in OVCAR5 cells () also indicated very tight cell-cell interactions 
Evaluation of protein expression profiles further confirmed epithelial and mesenchymal characteristics of two groups of cells represented by OVCAR5 and OVCAR10. All cells grouped together with OVCAR5, except for PEO4, showed a distinct epithelial phenotype, i.e., expressed E-cadherin and pan-keratin with little or no expression of vimentin, N-cadherin, and ZEB-1, an E-cadherin repressor (). Conversely, cells grouped with OVCAR10 cells expressed a battery of proteins reminiscent to those expressed in mesenchymal cells, i.e., expressed vimentin and ZEB-1 but lacked pan-keratin and E-cadherin. Nevertheless, these cells also presented some epithelial markers such as claudin-1, ZO-1, and occludin-1, which are typically localized at tight junctions and are often lost during EMT 
, suggesting that this cell may be in partial EMT 
. We also found differences in protease profiles among the cells with the abovementioned epithelial and partial EMT phenotypes (). For example, uPA (pro-form) was only expressed in cells that presented the epithelial properties. Most of the ovarian tumor cells expressed MMP-2 and MMP-9 (pro-forms) weakly as previously observed 
; however, the tumor cells with partial EMT phenotypes had higher levels. Therefore, the two groups could be divided by the differential protease profile as well. Nevertheless, inhibitors of these proteases did not seem to have an effect in reversing any of the two phenotypes studied probably due to differences between cellular expression levels and secreted protease activity as discussed below. In contrast, β1
-integrin expression was relatively uninformative in grouping OVCAR5- and OVCAR10-like cells (). However, we did observe that β1
-integrin expression might be useful to discriminate cells degrading ECM as single cells (SKOV3 and UPN251) from those that induce ECM rearrangement (e.g., OVCAR10). Previous studies have reported that β1
-integrin expression is reduced in protease-independent amoeboid cell migration compared to protease-dependent mesenchymal cell migration 
Overall, our studies demonstrated that protein expression profiles of adhesion molecules, cytokeratin, and proteases could predict the strategy of penetration/invasion using an in vitro
3D model, although, there were some exceptions. As discussed, PEO4 cells were grouped with OVCAR5 cells based on their ECM remodeling properties (); however, their protein profile was similar to OVCAR10 cells (). We speculate that the ECM change induced by PEO4 cells might be smaller than other cells presenting OVCAR10-like phenotypes; thereby, they might not be grouped correctly based on detection of fluorescent signal after prolonged culture. Importantly, PEO1 and PEO4 cell lines were obtained from the same patient before and after the onset of resistance to chemotherapy 
, and may be reflective of a major problem in treating ovarian cancer patients, i.e., the subsequent development of drug resistance and recurrence of cancer in spite of initial effectiveness of platinum-based chemotherapies 
. One can speculate that platinum-based chemotherapy might result in the switching of a given invasion mechanism and evolvement of additional strategies. Nevertheless, these types of questions are beyond the scope of this study.
Based on the above observed characteristics, we anticipated that ECM modification induced by OVCAR5 cells would be blocked by protease inhibitors, while ROCK inhibitors would suppress ECM modifications imparted by OVCAR10 cells which may have an ability to generate force to contract ECMs. The contraction of actin filament is known to be primarily induced by small G-protein Rho and its downstream effector, ROCK, responsible for protease independent invasion (e.g., amoeboid) 
. Upon exposure to inhibitors of different proteases, the degradation of TMP by OVCAR5 cells was greatly prevented. However, the effect was protease type-dependent. Aprotinin, a serine protease inhibitor, effectively inhibited OVCAR5 cell-mediated TMP degradation ( & S2A
). In contrast, the effect of leupeptin, which inhibits both serine and cysteine proteases was minimal, as was GM6001, a broad range MMP inhibitor. These results correlated with the detection of massive caseinolytic but negligible gelatinolytic proteases secreted by OVCAR5 cells cultured within N3F-derived matrices (). We expected that amiloride, known to specifically inhibit uPA but not tPA 
, would effectively prevent OVCAR5 cells from degrading the ECM since these cell type expressed higher uPA (). However, amiloride appeared to enhance TMP degradation (Table S1
). In fact, OVCAR5 cell-conditioned media contained caseinolytic proteases not inhibited by amiloride especially at molecular weight above 50 kDa (Figure S4
). Noticeably, no protease inhibitors effectively prevent fibronectin degradation by OVCAR5 cells ( and Table S2
), implying that fibronectin may be more susceptible to enzymatic degradation and substrates of many different types of proteases. These results suggest that different proteases have specificities for different ECM components and that it will be important to determine protease type secreted by both tumor and stromal cells to predict their specific ECM remodeling strategies.
ROCK inhibitors blocked OVCAR10 cell-induced invasion and ECM contraction ( and S2B
). However, cells with an OVCAR10-like phenotype did not appear to have higher ROCK activity as measured by an enzymatic immunoassay (Figure S5
). It has been reported that invasion of cells which did not express ROCK or Rho at high levels were also inhibited by a ROCK inhibitor 
. Therefore, other mechanisms regulating ROCK activity may be responsible for ROCK-dependent ECM modification by these cells. MMP activity has been considered to be implicated in collagen gel contraction and tissue reorganization as well 
. However, the effect of GM6001 was negligible on ECM contraction by OVCAR10 cells in our study, as observed previously in Hey and ES-2 ovarian tumor cells 
. Also, the addition of a cocktail of protease inhibitors did not seem to further enhance the inhibitory effect of the ROCK inhibitors (), and therefore, none of the protease inhibitors we tested appeared to be effective in OVCAR10-induced matrix change. This might be explained by low protease activity derived from these cells (). Importantly, aprotinin and H1152, which effectively inhibited OVCAR5- and OVCAR10-induced ECM modification, respectively, also inhibited ECM changes induced by other cells grouped together (Figure S3A & B
). Therefore, two different cell types divided by epithelial and mesenchymal phenotypes contribute to different ECM phenotypes, degradation and accumulation or contraction of ECMs, respectively, and their ECM modification can be reversed by the use of specific protease and ROCK inhibitors, respectively.
In conclusion, we have shown for the first time using a physiologically relevant 3D model that ovarian tumors cells that present epithelial characteristics tend to penetrate/invade the mesenchymal ECM as clusters or groups and use proteases to degrade the matrix. On the other hand, cells that present partial EMT phenotypes may push through the ECM by mechanisms that are based on Rho-dependent ECM accumulation or contraction and are less dependent on proteolysis for their penetration/invasion. Our study also expands upon the view that many cell types including tumor cells naturally penetrate through ECMs while invading without the use of classic EMT mechanisms. Our study suggests that differential ECM modification mechanisms by various ovarian tumor cell lines may reflect heterogeneity among tumors from different patients or within a given tumor. Therefore, fully characterizing the potential invasion mechanisms may help to design therapies targeting theses ovarian cancer cell behaviors.