Cancer Cell Invasion in Three-dimensional Collagen Is Regulated Differentially by Gα13 Protein and Discoidin Domain Receptor 1-Par3 Protein Signaling*

Background: The role of Gα13 in mediating invasion in three-dimensional collagen is not well known. Results: Gα13 promotes invasion in three-dimensional collagen by disrupting cell-cell adhesion, whereas DDR1-Par3 signaling counteracts the effects of Gα13 by enhancing cell-cell adhesion. Conclusion: Gα13 and DDR1-Par3 differentially regulate cell-cell junctions to mediate three-dimensional invasion. Significance: Better understanding of how cells invade in three-dimensional collagen may identify potential therapeutic targets. Cancer cells can invade in three-dimensional collagen as single cells or as a cohesive group of cells that require coordination of cell-cell junctions and the actin cytoskeleton. To examine the role of Gα13, a G12 family heterotrimeric G protein, in regulating cellular invasion in three-dimensional collagen, we established a novel method to track cell invasion by membrane type 1 matrix metalloproteinase-expressing cancer cells. We show that knockdown of Gα13 decreased membrane type 1 matrix metalloproteinase-driven proteolytic invasion in three-dimensional collagen and enhanced E-cadherin-mediated cell-cell adhesion. E-cadherin knockdown reversed Gα13 siRNA-induced cell-cell adhesion but failed to reverse the effect of Gα13 siRNA on proteolytic invasion. Instead, concurrent knockdown of E-cadherin and Gα13 led to an increased number of single cells rather than groups of cells. Significantly, knockdown of discoidin domain receptor 1 (DDR1), a collagen-binding protein that also co-localizes to cell-cell junctions, reversed the effects of Gα13 knockdown on cell-cell adhesion and proteolytic invasion in three-dimensional collagen. Knockdown of the polarity protein Par3, which can function downstream of DDR1, also reversed the effects of Gα13 knockdown on cell-cell adhesion and proteolytic invasion in three-dimensional collagen. Overall, we show that Gα13 and DDR1-Par3 differentially regulate cell-cell junctions and the actin cytoskeleton to mediate invasion in three-dimensional collagen.

Pancreatic ductal adenocarcinoma (PDAC) 2 is a highly lethal malignancy, in part because of the presence of distant metastases at the time of diagnosis (1,2). PDAC tumors in particular demonstrate a pronounced collagen-rich stromal reaction (3,4) that, as we have shown previously, facilitates tumor invasion by inducing the expression of the key collagenase membrane type 1 matrix metalloproteinase (MT1-MMP) (5)(6)(7). Although cancer cells in high-grade tumors may invade as single cells, most low-grade tumors invade as a cohesive group of cells ("collective migration"), which can range from strands of only one or two cells in diameter to a cluster of cells (8). It has been postulated that collective migration requires the regulation of cellcell junctions to transmit force between cells as well as the coordination of cytoskeletal reorganization within a group of cells to drive organized migration (9). Because collective migration requires cells to maintain intact cell-cell junctions, proteins such as E-cadherin maintain the integrity of cell-cell junctions. The collagen-binding protein discoidin domain receptor 1 (DDR1) has been shown to localize to cell-cell junctions and mediate cell-cell adhesion (10,11). In addition to regulating E-cadherin stability at the cell surface (12,13), DDR1, through its interaction with the Par3 polarity complex, can also regulate actin cytoskeleton dynamics (11).
G␣ 13 , a member of the G 12 family of heterotrimeric G proteins, activates Rho signaling downstream of G protein-coupled receptor activation (14 -16) and has been implicated in cell migration downstream of both G protein-coupled receptors and receptor tyrosine kinases (17). For example, mouse embryonic fibroblasts from G␣ 13 -null mice are deficient in lysophosphatidic acid-stimulated cell migration (18). G␣ 13 , but not the closely related G␣ 12 , also mediates EGF-driven migration (18). In addition to regulating cell migration, G␣ 13 can weaken adherens junctions by binding to E-cadherin and displacing ␤-catenin from adherens junctions (19,20). Recently, G␣ 13 has also been shown to disrupt endothelial adherens junctions by binding to VE-cadherin and causing VE-cadherin internalization (21). Significantly, G␣ 13 is up-regulated in metastatic breast and prostate cancers (22,23) and has been shown to play a role in pancreatic cancer cell invasion (24). However, the extent to which G␣ 13 contributes to pancreatic cancer cell invasion in three-dimensional collagen is not well known.
In this study, we show that G␣ 13 is up-regulated in human PDAC tumors and, using a quantitative method to track cell invasion in three-dimensional collagen gels, we demonstrate that knockdown of G␣ 13 decreased MT1-MMP-driven proteolytic invasion of PDAC cells in three-dimensional collagen and enhanced E-cadherin-mediated cell-cell adhesion. Although E-cadherin knockdown reversed the effect of G␣ 13 siRNAinduced cell-cell adhesion, E-cadherin knockdown failed to reverse the effect of G␣ 13 siRNA on proteolytic invasion. Instead, concurrent knockdown of E-cadherin and G␣ 13 led to an increased number of single cells rather than groups of cells. Significantly, knockdown of DDR1 reversed the effect of G␣ 13 knockdown on cell-cell adhesion and proteolytic invasion. Additionally, knockdown of the polarity protein Par3, which can function downstream of DDR1, reversed the effect of G␣ 13 knockdown on cell-cell adhesion and proteolytic invasion. Overall, we show that G␣ 13 and DDR1-Par3 signaling differentially regulate cell-cell junctions and the actin cytoskeleton to mediate invasion in three-dimensional collagen.

Experimental Procedures
Antibodies and Reagents-General tissue culture supplies were purchased from VWR (West Chester, PA). Antibodies against MT1-MMP (catalog no. ab38971) and G␣ 13  Oncomine Dataset Analysis-Datasets analyzing mRNA from both normal and PDAC tumor samples were selected from the publicly available Oncomine database.
Human PDAC Tissue Analysis-The human pancreatic tissue microarray (TMA) was generated by the Pathology Core Facility at Northwestern University using specimens obtained on an institutional review board-approved protocol. The TMA was composed of 2-mm cores from 17 PDAC specimens and 15 normal pancreatic tissue specimens. Additional TMA slides (TMA-1421) with 2-mm cores from three normal pancreatic tissue samples and 20 PDAC cases were purchased from Protein Biotechnologies (Ramona, CA). Slides with 4-m sections of the TMA were deparaffinized and stained with H&E or trichrome-stained using a Masson kit by the Pathology Core Facility (Northwestern University). The extent of fibrosis was determined by the blue trichrome stain in the TMA specimens: Ͻ25% of the TMA core with blue stain (low) or Ͼ25% of the TMA core with blue stain (high).
For immunostaining, TMA slides were deparaffinized in xylene and rehydrated in a decreasing ethanol gradient, and then antigen retrieval was performed using a decloaking chamber at 110°C for 10 min in BD Retrievagen A (pH 6.0) target retrieval solution, followed by blocking with Protein Block (Dako) and Peroxidase Block (Dako). The TMA slides were incubated with the primary rabbit monoclonal antibody against G␣ 13 (EPR5436, catalog no. ab128900, Abcam) overnight at 4°C at 1:500 dilution in Dako antibody diluent. Following multiple washings with Dako wash buffer, the slides were incubated with anti-rabbit HRP-labeled polymer (Dako) at room temperature for 1 h and incubated with 3,3Ј-diaminobenzidine substrate (Dako) for 1 min, and then the slides were counterstained with hematoxylin. As specified on the Abcam web site for the ab128900 antibody, positive staining for G␣ 13 was indicated by the presence of a brown precipitate in the cytoplasm. The TMA cores were evaluated and scored as follows: 0, no staining or background staining; 1ϩ, mild specific staining of malignant cells; 2ϩ, moderate staining of malignant cells; 3ϩ, the most intense staining of malignant cells.
The specificity of the rabbit monoclonal ab128900 antibody for G␣ 13 was validated by knocking down G␣ 13 in CD18tet-MT cells, creating a cell block and then embedding the cell block in paraffin. Control and G␣ 13 knockdown CD18-tet-MT cell blocks were then sectioned and stained as above for the TMA specimens. In additional controls, human PDAC tumor sections were processed for immunostaining using a rabbit IgG monoclonal antibody (catalog no. ab172730, Abcam) as an isotype control for the anti-G␣ 13 ab128900 antibody.
Cell Culture-CD18, Panc1, and SCC25 cells were purchased from the ATCC, whereas UMSCC1 cells were provided by Dr. E. Lengyel. MT1-MMP-expressing cancer cells were generated and maintained as described previously (26).
Down-regulation of G␣ 13 , E-cadherin, DDR1, and Par3-Cells were transfected with 25 nM siRNA using Lipofectamine RNAiMAX (Invitrogen). Transfected cells were allowed to recover for 48 h, trypsinized, and plated in collagen or suspended in hanging drops. For immunofluorescence, cells were plated on glass coverslips prior to transfection.
Three-dimensional Collagen Invasion Assay-MT1-MMPexpressing cancer cells were transfected with siRNA, allowed to recover for 48 h, grown in three-dimensional type I collagen gels (2.2 mg/ml) as described previously (7), and then treated with doxycycline and various inhibitors. Cells grown in collagen gels were examined using a Zeiss Axiovert 40 CFL microscope, and pictures were taken with a Nikon Coolpix 4500 camera. Collagen gels were then fixed in 10% formalin for 24 h and embedded in paraffin by the Pathology Core Facility at Northwestern University. Six 8-m sections from each of the collagen-paraffin blocks were cut from the top of the block to the same depth, stained with H&E, and examined using a Zeiss Axioskop 40 microscope and camera. Areas of cleaved collagen ("collagenolytic tracks") were measured from a minimum of 10 fields/condition at ϫ10 magnification using ImageJ.
Two-dimensional Collagen Migration-Migration of PDAC cells over collagen-coated surfaces was assessed as described previously (6). PDAC cells were plated onto thin-layer type I collagen overlaid with colloidal gold. Cells were allowed to migrate for 24 h, phagokinetic tracks were monitored by visual examination using a Zeiss Axiovert 40 CFL microscope, and pictures were taken with a Nikon Coolpix 4500 camera.
Extraction of PDAC Cells from Collagen Gels-PDAC cells were grown in collagen gels as indicated above. After 72 h, cells were extracted from the gels using collagenase and lysed in radioimmune precipitation assay buffer containing protease and phosphatase inhibitors for Western blotting (27,28).
Cell Surface Biotinylation-CD18-tet-MT cells were labeled with cell-impermeable Sulfo-NHS-LC-Biotin (Pierce) in icecold PBS as described previously (29). Equal amounts of protein were then incubated with streptavidin beads (Pierce) to isolate biotinylated cell surface proteins. The biotin-labeled samples were analyzed for MT1-MMP by Western blotting using Transferrin receptor as a labeling control.
Three-dimensional Collagen Contractility Assay-Collagen contractility assays were performed as described previously (31,32). Briefly, PDAC cells were embedded in three-dimensional collagen (2.2 mg/ml) at 2 ϫ 10 5 cells/plug (for inhibitor treatment) or 5 ϫ 10 5 cells/plug (after transfection with siRNA) to form a cylinder-shaped plug. The plug was then inserted in an additional layer of collagen (2.2 mg/ml) to allow the formation of a core of cells surrounded by collagen. After contraction over a period of 5 days, pictures of the contracted cores were taken using an 8-megapixel iPhone 5S camera.
Hanging Drop Assay-PDAC cells were transfected with the indicated siRNAs and suspended in hanging drops as described previously (33). For cells treated with the E-cadherin functionblocking antibody HECD-1, cells were incubated with 2 M IgG or HECD-1 per drop for 30 min at room temperature prior to suspension in drops. After 18 h, cells were imaged, and pictures were taken using a Zeiss Axiovert 40 CFL microscope and a Nikon Coolpix 4500 camera.
Immunofluorescence Imaging-PDAC cells were plated on glass coverslips and transfected with the indicated siRNAs for 72 h. Cells were fixed in 4% paraformaldehyde and permeabilized with 0.01% Triton X-100 and then stained for E-cadherin and phalloidin. Images were taken on a Zeiss LSM510 Meta confocal microscope at the Northwestern University Center for Advanced Microscopy (Chicago, IL). The relative intensity of E-cadherin at cell-cell junctions was determined using Image J.
Statistical Analysis-All statistical analyses were performed using GraphPad Prism 5 (San Diego, CA).

G␣ 13 Is Up-regulated in Human PDAC
Tumors-Cell invasion requires precise coordination of the actin cytoskeleton (31,34). Because G␣ 13 , the ␣ subunit of the heterotrimeric G protein G 13 , is a major regulator of the actin cytoskeleton (35,36), we examined the extent to which G␣ 13 levels are altered in human PDAC tumors using the publically available Oncomine database (37)(38)(39)(40). As shown in Fig. 1A, there is an ϳ2-fold increase in GNA13 mRNA levels in human PDAC tumors compared with normal tissue. Furthermore, immunohistochemical staining of human pancreatic TMAs showed that there is increased cytoplasmic staining of G␣ 13 protein in tumors compared with normal pancreas (Fig. 1, B and C). Because PDAC tumors are associated with a pronounced collagen-rich stromal reaction, as demonstrated by increased trichrome staining (Fig.  1B), we examined the relationship between fibrosis and G␣ 13 protein expression. We show that G␣ 13 protein levels are associated with increased fibrosis in human PDAC tumors (Fig. 1C).

Establishment of a Quantitative Method to Track MT1-MMP-dependent Cell Invasion in Three-dimensional Collagen
Gels-To examine the role of G␣ 13 in PDAC progression, we established a quantitative method to track MT1-MMP-dependent cell invasion in three-dimensional collagen gels using inducible expression of MT1-MMP in PDAC cells (hereafter referred to as PDAC-tet-MT cells; Fig. 2, A and B). Induction of MT1-MMP in CD18-tet-MT ( Fig. 2C) and Panc1-tet-MT adenocarcinoma cells (data not shown) in three-dimensional type I collagen resulted in tracks generated by clearing of the pericellular collagen by invading cells, which was inhibited by the broad-spectrum MMP inhibitor GM6001. We also examined the extent to which our system can track MT1-MMP-dependent invasion of squamous cell cancer (SCC) cells in threedimensional collagen. Induction of MT1-MMP in SCC cells (UMSCC1 and SCC25 cells) also resulted in tracks generated by clearing of the pericellular collagen by the invading cells that was also inhibited by GM6001 (data not shown). We additionally validated this quantitative method of tracking cells in threedimensional collagen by examining the effect of EGF on cell invasion. EGF treatment enhanced the invasion of MT1-MMPexpressing CD18 (Fig. 2D) and Panc1 cells (data not shown). Significantly, there were morphological changes in PDACtet-MT cells upon induction of MT1-MMP expression. CD18 cells maintained their cohesive growth with a slight ellipsoid morphology (Fig. 2D) that morphed into a half moon-shaped group of cells following EGF treatment (Fig. 2D). Induction of MT1-MMP in established colonies also caused polarization of CD18-tet-MT cells, as demonstrated by increased numbers of half moon-shaped cell clusters, and increased invasion (Fig. 2E).
Knockdown of G␣ 13 Decreases MT1-MMP-driven Proteolytic Invasion of Cancer Cells in Three-dimensional Collagen-We next determined whether G␣ 13 was involved in MT1-MMPdriven invasion in three-dimensional collagen. Initially, we evaluated the effect of G␣ 13 knockdown on MT1-MMP expression and cell surface localization. Knockdown of G␣ 13 did not affect the levels of the catalytically active, 55-kDa form or the 43-kDa autodegradation form of MT1-MMP (Fig. 3A). G␣ 13 knockdown also did not affect cell surface expression of MT1-MMP in CD18-tet-MT cells (Fig. 3A). However, knockdown of G␣ 13 inhibited both basal and EGF-stimulated invasion of MT1-MMP-expressing CD18 (Fig. 3B) and Panc1 cells (Fig.  3D). Knockdown of G␣ 13 also inhibited invasion of MT1-MMP-expressing SCC cells (data not shown). In contrast, knockdown of G␣ 13 did not inhibit two-dimensional migration of PDAC cells over collagen-coated surfaces (Fig. 3C). Moreover, knockdown of G␣ 13 did not affect proliferation or apoptosis of PDAC cells in three-dimensional collagen (data not shown).
Knockdown of G␣ 13 Does Not Affect ERK1/2 Phosphorylation but Enhances MLC Phosphorylation and ROCK Activity in Three-dimensional Collagen-To understand the mechanism by which G␣ 13 regulated MT1-MMP-driven invasion in threedimensional collagen, we examined the effect of G␣ 13 knockdown on ERK signaling. Notably, G␣ 13 knockdown has been shown previously to inhibit ERK1/2 phosphorylation in Jurkat T cells (41). Initially, we examined the extent to which EGFR-ERK1/2 signaling mediated invasion of MT1-MMP-expressing PDAC cells in three-dimensional collagen. Treatment of MT1-MMP-expressing CD18 cells with the EGFR kinase inhibitor AG1478 or the MEK inhibitor U0126 reduced invasion of MT1-MMP-expressing CD18 cells (Fig. 4A). Additionally, cells treated with AG1478 or U0126 blocked the EGF-induced morphological changes, indicating that EGFR-MEK-ERK signaling mediated the invasion of MT1-MMP-expressing CD18 cells. However, G␣ 13 knockdown in CD18-tet-MT cells did not affect ERK1/2 phosphorylation (Fig. 4B), suggesting that the inhibition of invasion by G␣ 13 knockdown was not due to its regulation of the ERK signaling pathway.
Because a balance of Rho and Rac activity is required for the efficient invasion of cells (6,42), we investigated the role of Rho and Rac signaling in regulating invasion in three-dimensional collagen using NSC23766, which inhibits the interaction between Rac1 and its guanine nucleotide exchange factor Tiam1 and the ROCK1/2 inhibitor Y27632 (43,44). As shown in Fig. 4C, Y27632, but not NSC23766, blocked basal and EGFdriven invasion of MT1-MMP-expressing CD18 cells, indicating that ROCK1/2 signaling mediates the invasion of PDAC cells in three-dimensional collagen. However, G␣ 13 knockdown in three-dimensional collagen did not decrease ROCK1/2 activity, as measured by myosin light chain phosphorylation (phospho-MLC). Instead, G␣ 13 knockdown in three-dimensional collagen increased MLC phosphorylation (Fig. 4D). Because this was an unexpected result, we further evaluated the effect of G␣ 13 knockdown on ROCK1/2 activity using a collagen contraction assay. Because both MT1-MMP proteolytic activity and ROCK1/2 signaling mediate the contraction of collagen gels by PDAC cells (Fig. 4, E and F) (31), we examined the effect of G␣ 13 knockdown on collagen contraction by CD18tet-MT cells. G␣ 13 knockdown enhanced collagen contraction, and this effect could be reversed by Y27632 (Fig. 4G), indicating that G␣ 13 knockdown in three-dimensional collagen increases ROCK1/2 activity. This suggests that the inhibitory effect of G␣ 13 knockdown on three-dimensional collagen invasion cannot be due to its effect on ROCK signaling. A, the relative expression of GNA13 mRNA was evaluated in human PDAC tumors (T) relative to normal (N) pancreas using the publicly available Oncomine database. Only databases that contained at least 10 normal specimens were analyzed. Overall statistical analyses of Oncomine data were performed using Student's t test. ***, p Ͻ 0.001. B and C, human pancreatic TMAs generated by the Pathology Core Facility at Northwestern University or purchased from Protein Biotechnologies were H&E-stained, trichrome-stained (blue ϭ fibrosis), or immunostained with anti-G␣ 13 antibody (B). The relative staining was graded as low (0 or 1ϩ) or high (2ϩ or 3ϩ) as detailed under "Experimental Procedures." The relative expression of G␣ 13 in tumor tissue and adjacent normal was assessed by Fisher's exact test (C). The extent of fibrosis was determined by the blue trichrome stain in the TMA specimens: Ͻ25% of the TMA core with blue stain (low) or Ͼ25% of the TMA core with blue stain (high). The relationship between G␣ 13 expression in the tissue samples and the extent of fibrosis was assessed by Fisher's exact test (C). IHC, immunohistochemistry. Scale bars, 50 m.

G␣ 13 Knockdown Enhances E-cadherin-mediated Cell-Cell
Adhesion, but E-cadherin Knockdown Fails to Reverse the Effect of G␣ 13 siRNA on Proteolytic Invasion-It has been shown previously that G␣ 13 can serve as a negative regulator of cadherin function by directly binding to the intracellular domain of Eand VE-cadherin (20,21). Therefore, we examined the effect of G␣ 13 siRNA on cell-cell adhesion of MT1-MMP-expressing CD18 cells utilizing a hanging drop assay, which allows for analysis of cell-cell adhesion independent of cell matrix adhesion (33). As shown in Fig. 5A, MT1-MMP-expressing CD18 cells transfected with control siRNA formed spheroids in the hanging drop assay. Significantly, transfection with G␣ 13 siRNA caused more compact spheroids, indicating that G␣ 13 siRNA enhanced cell-cell adhesion. Although the spheroids formed by control siRNA-transfected cells became larger and less tightly packed after treatment with Y27632, treatment with Y27632 failed to loosen the compact spheroids formed by G␣ 13 siRNA-transfected cells, suggesting that the effect of knocking down G␣ 13 on cell-cell adhesion was independent of its ability to regulate ROCK signaling. In contrast, addition of the func- tion-blocking E-cadherin antibody HECD1 in the hanging drop assay significantly reversed G␣ 13 siRNA-induced compaction of the MT1-MMP-expressing CD18 spheroids (Fig. 5B), suggesting that the increased cell-cell adhesion observed after knockdown of G␣ 13 was mediated through E-cadherin. Consistent with our findings in the hanging drop assay, E-cadherin localization was increased at cell-cell junctions in cells transfected with siRNA against G␣ 13 compared with control siRNA-transfected cells (Fig. 5C). Importantly, there was no significant change in either the E-cadherin mRNA (data not shown) or the E-cadherin protein levels (Fig. 5D) following knockdown of G␣ 13 , indicating that the effects of G␣ 13 on E-cadherin-mediated cell-cell junctions occurs at the posttranslational level.
We next examined the role of E-cadherin downstream of G␣ 13 in mediating PDAC cell invasion in three-dimensional collagen. We hypothesized that inhibiting E-cadherin-mediated cell-cell junctions in G␣ 13 siRNA-transfected cells would restore proteolytic invasion in three-dimensional collagen. However, concurrent knockdown of E-cadherin and G␣ 13 failed to increase proteolytic invasion of MT1-MMP-expressing CD18 cells in three-dimensional collagen (Fig. 5D). Signif-  The effect on phospho-ERK1/2 (pERK1/2) and total ERK1/2 was determined by Western blotting. C, CD18-tet-MT cells were plated in three-dimensional collagen, treated with doxycycline and co-treated with Rac1 inhibitor NSC23766 (NSC, 20 M) or ROCK1/2 inhibitor Y27632 (10 M), and allowed to invade for 72 h in the presence or absence of exogenous EGF (20 ng/ml). The effect on collagen invasion was assessed as described in Fig. 2A. AU, arbitrary units. D, CD18-tet-MT cells were transfected with siRNA for 48 h and allowed to grow in three-dimensional collagen for 48 h in the presence of doxycycline, and then the effect on phospho-MLC (pMLC S19) was examined by Western blotting. E, CD18-tet-MT cells were plated in a plug of three-dimensional collagen and treated with doxycycline, the pan-MMP inhibitor GM6001 (5 M), and/or EGF (20 ng/ml) as indicated. The ability of the cells to contract the collagen gels over 5 days was examined. The contracted gel area was measured using ImageJ. F, CD18-tet-MT cells were plated in a plug of three-dimensional collagen and treated with doxycycline and the Rac inhibitor NSC23766 (20 M) or the ROCK1/2 inhibitor Y27632 (10 M). The ability of the cells to contract the collagen gels over 5 days was examined. The contracted gel area was measured using ImageJ. G, CD18-tet-MT cells were transfected with siRNA against G␣ 13 for 48 h, embedded in a plug of three-dimensional collagen, and treated with doxycycline and the indicated reagents, and then the ability of the cells to contract the collagen gels over 3 days was examined. The contracted gel area was measured using ImageJ. Invasion quantitation statistics were performed using one-way ANOVA followed by Dunnett's post-test. Contractility assay statistics were performed using Student's t test. ns, not significant; ***, p Ͻ 0.001. The results are representative of three independent experiments.
icantly, concurrent knockdown of E-cadherin and G␣ 13 led to an increased number of single cells rather than groups of cells (Fig. 5E). Concurrent knockdown of G␣ 13 and E-cadherin led to dissociation of cell clusters grown on glass coverslips, which was associated with changes in phalloidin organization (Fig. 5E). 13 Knockdown on Cell-Cell Adhesion and Proteolytic Invasion-Because the collagen-binding protein DDR1 can regulate E-cadherin stability at cell-cell junctions and regulate actin cytoskeleton dynamics in skin cancer cells to regulate group invasion (11), we initially evaluated the extent to which DDR1 levels are altered in human PDAC tumors using the same Oncomine databases as described in Fig. 1A (37-40). As shown in Fig. 6A, there were increased DDR1 mRNA levels in human PDAC tumors compared with normal tissue. We next evaluated the extent to which DDR1 counteracts the effects of G␣ 13 in PDAC cells. As shown in Fig. 6B, cells co-transfected with siRNA against G␣ 13 and DDR1 formed looser clusters, suggesting that knockdown of DDR1 can reverse the effects of G␣ 13 siRNA on cell-cell adhesion. We next evaluated whether the effect of DDR1 on cell-cell adhesion was due to changes in E-cadherin expression or localization. Knockdown of DDR1 in MT1-MMP-expressing CD18 cells led to disruption of cell-cell junctions that was associated with diffuse localization of E-cadherin (Fig. 6C) but did not affect E-cadherin protein levels (Fig. 6C). Concurrent knockdown of G␣ 13 and DDR1 led to a further disruption of cell-cell junctions and changes in phalloidin staining at the cell periphery (Fig. 6C).

DDR1 Knockdown Reverses the Effect of G␣
We next examined the role of DDR1 downstream of G␣ 13 in mediating PDAC cell invasion in three-dimensional collagen. Although G␣ 13 knockdown decreased EGF-driven invasion of CD18-tet-MT cells, knockdown of DDR1 did not affect EGFdriven invasion of MT1-MMP-expressing CD18 cells. DDR1 knockdown also did not affect MT1-MMP expression or cell surface localization in CD18-tet-MT cells (data not shown). However, concurrent knockdown of G␣ 13 and DDR1 reversed the inhibitory effects of G␣ 13 knockdown on proteolytic invasion by MT1-MMP-expressing CD18 cells (Fig. 6D).
Par3 Knockdown Reverses the Effect of G␣13 siRNA on Cell-Cell Adhesion and Proteolytic Invasion-Our findings that CD18 cells invade in three-dimensional collagen in a "polarized" half moon shape (Fig. 2) suggest that the polarity complex may regulate the invasion of PDAC cells in three-dimensional collagen. Significantly, we show that G␣ 13 knockdown CD18-tet-MT cells fail to demonstrate any half moon cell clusters in three-dimensional collagen (Fig. 6D) and that concurrent knockdown of DDR1 can reverse the effects of G␣ 13 knockdown on the number of cell clusters with a half moon shape (Fig. 6D). Importantly, the polarity protein Par3 can function downstream of DDR1 to regulate the actin cytoskeleton in skin cancer cells (11). Therefore, we examined whether Par3 can regulate invasion of CD18-tet-MT cells in three-dimensional collagen. Initially, we examined the effect of DDR1 and G␣ 13 knockdown on Par3 expression and localization. Knockdown of DDR1 caused redistribution of Par3 without significantly affecting total Par3 levels (Fig. 7, A and B). In contrast to the effects of DDR1 knockdown, G␣ 13 knockdown caused increased localization of Par3 at cell-cell junctions (Fig. 7A).
Finally, we analyzed the effect of co-transfecting Par3 siRNA with G␣ 13 siRNA in MT1-MMP-expressing CD18 cells on cellcell adhesion. Similar to the DDR1 siRNA (Fig. 6B), Par3 siRNA reversed the effects of G␣ 13 siRNA on cell-cell adhesion in the hanging drop assay (Fig. 7C). Moreover, although Par3 siRNA did not significantly increase three-dimensional collagen invasion (Fig. 7D) or affect MT1-MMP levels (data not shown), concurrent knockdown of Par3 and G␣ 13 reversed the inhibitory effects of G␣ 13 siRNA on collagen invasion by MT1-MMPexpressing CD18 cells (Fig. 7D). Concurrent knockdown of Par3 also reversed the effects of G␣ 13 knockdown on the number of cell clusters with a half moon shape (Fig. 7D).

Discussion
Patients with pancreatic cancer, the fourth leading cause of cancer-related death, often present with metastatic disease or locally advanced disease with invasion into surrounding structures (1,2). Therefore, there is increasing interest in identifying mechanisms that regulate the invasion of pancreatic cancer cells. Previously, we have shown that the collagenase MT1-MMP mediated invasion of pancreatic cancer cells in threedimensional collagen (5)(6)(7). In this report, we show that G␣ 13 is overexpressed in human PDAC tumors, and, by utilizing a quantitative method to track proteolytic invasion in three-dimensional collagen gels, we show that G␣ 13 functions downstream of MT1-MMP to mediate the invasion of pancreatic cancer cells in three-dimensional collagen. G␣ 13 is known to regulate the actin cytoskeleton, and therefore cell migration and invasion, through activation of the Rho/ ROCK/myosin pathway (14 -16). Previously, it has been shown FIGURE 5. G␣ 13 knockdown enhances E-cadherin-mediated cell-cell adhesion, but E-cadherin knockdown fails to reverse the effect of G␣ 13 siRNA on proteolytic invasion. A, CD18-tet-MT cells were transfected with siRNA for 96 h and then trypsinized and resuspended (2.5 ϫ 10 6 cells/ml of DMEM ϩ 10% FBS ϩ doxycycline) in the presence or absence of Y27632 (10 M). A 10-l drop of the cell suspension was hung on the undersurface of a tissue culture lid, incubated overnight at 37°C, and photographed the next morning. The area covered by the cluster of cells was measured using ImageJ. B, CD18-tet-MT cells were transfected with siRNA for 96 h and then trypsinized and processed for the hanging drop assay in the presence of IgG or the function-blocking E-cadherin antibody HECD1. Hanging drops were incubated overnight at 37°C and photographed the next morning. The area covered by the cluster of cells was measured using ImageJ. C, CD18-tet-MT cells were plated on glass coverslips, transfected with siRNA against G␣ 13 , grown in the presence of doxycycline for 48 h, and fixed with 4% paraformaldehyde and permeabilized with 0.1% TritonX-100. Cells were stained for E-cadherin and phalloidin and imaged using a Zeiss Axiovert LSM510 Meta confocal microscope. The relative fluorescence intensity at cell-cell junctions was quantified using ImageJ. D and E, CD18-tet-MT cells were transfected with siRNA against G␣ 13 , E-cadherin, or both G␣ 13 and E-cadherin. The effect on E-cadherin and G␣ 13 knockdown was determined by Western blotting. Prior to plating cells in three-dimensional (3D) collagen, cells were incubated with IgG or the E-cadherin (Ecad) function-blocking antibody HECD1. Cells were grown in three-dimensional collagen for 72 h with doxycycline in the presence of EGF (20 ng/ml). Gels were then processed to examine the effect on invasion as detailed in Fig. 2, and relative invasion was quantified (D). The number of single cells present in three-dimensional collagen per field in the invasion assay was quantified (E, left panel). CD18-tet-MT cells were plated on glass coverslips, transfected with siRNA against G␣ 13 and/or E-cadherin, grown in the presence of doxycycline for 48 h, fixed, permeabilized, and stained for E-cadherin and phalloidin (E, right panel). The statistics of the invasion assay were performed using one-way ANOVA followed by Dunnett's post-test. The statistics of the hanging drop assay were performed using Student's t test. ns, not significant; *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. The results are representative of four independent experiments. Ctrl, control.
that overexpression of constitutively active G␣ 13 increases invasion by breast and prostate cancer cells through Matrigelcoated transwell filters (22,23), whereas inhibition of G␣ 13 using shRNA decreases lysophosphatidic acid-mediated invasion of pancreatic cancer cells through collagen-coated transwell filters (24). The increased invasion by G␣ 13 -overexpress-ing breast and prostate cancer cells was due to increased activation of Rho signaling (23). Even though we show that ROCK signaling mediates MT1-MMP-driven collagen invasion, G␣ 13 knockdown did not significantly attenuate Rho/ ROCK signaling in pancreatic cancer cells. In fact, G␣ 13 knockdown in three-dimensional collagen increased ROCK activity, FIGURE 6. DDR1 knockdown reverses the effect of G␣ 13 knockdown on cell-cell adhesion and proteolytic invasion. A, the relative expression of DDR1 mRNA was evaluated in human PDAC tumors relative to normal pancreas using the Oncomine database described in Fig. 1A. Overall statistics of Oncomine data were performed using Student's t test. **, p Ͻ 0.01. N, normal; T, tumor. B, CD-18-tet-MT cells were transfected with siRNA against G␣ 13 , DDR1, or both G␣ 13 and DDR1. The effect on DDR1 and G␣ 13 knockdown was determined by Western blotting. The cells were also plated in suspension in a hanging drop for 18 h in the presence of doxycycline. Quantitation of the area of cell clusters was performed using ImageJ as detailed in Fig. 5. C, CD18-tet-MT cells were plated on glass coverslips, transfected with siRNA against G␣ 13 and/or DDR1, grown in the presence of doxycycline for 48 h, fixed, permeabilized, and stained for E-cadherin and phalloidin (left panel). CD18-tet-MT cells were transfected with siRNA against G␣ 13 and/or DDR1 for 48 h, plated in three-dimensional collagen in the presence of doxycycline, and allowed to grow for 48 h. Cell lysates were then analyzed for E-cadherin expression by Western blotting (right panel). D, CD18-tet-MT cells transfected with siRNA against G␣ 13 and/or DDR1 were grown in three-dimensional collagen in the presence of doxycycline and allowed to invade for 72 h. Quantitation of invasion was analyzed as detailed in Fig. 2A. The number of cell clusters demonstrating a half moon shape was also quantified. The statistics of the invasion assay and the number of half moon cell clusters were performed using one-way ANOVA followed by Dunnett's post-test. The statistics of the hanging drop assay were performed using Student's t test. ns, not significant; *, p Ͻ 0.05; ***, p Ͻ 0.001. The results are representative of at least three independent experiments. Ctrl, control.
as demonstrated by increased myosin light chain phosphorylation and increased collagen contraction. This suggests that the inhibitory effect of G␣ 13 knockdown on three-dimensional collagen invasion occurs independently of its effect on ROCK signaling.
Additionally, we demonstrate that G␣ 13 siRNA enhances cell-cell adhesion, suggesting that G␣ 13 may regulate pancreatic cancer cell invasion in three-dimensional collagen by disrupting cell-cell junctions. G␣ 13 overexpression has been shown previously to disrupt E-cadherin-mediated cell-cell FIGURE 7. Par3 knockdown reverses the effect of G␣ 13 siRNA on cell-cell adhesion and proteolytic invasion. A, CD18-tet-MT cells were plated on glass coverslips, transfected with siRNA against DDR1, G␣ 13 , or Par3, and grown in the presence of doxycycline for 48 h. Cells on coverslips were fixed, permeabilized, and stained for E-cadherin and Par3. B, CD18-tet-MT cells transfected with siRNA against Par3, G␣ 13 , or DDR1 were grown in three-dimensional collagen for 48 h. Collagenase-extracted cell lysates were then analyzed for E-cadherin, Par3, G␣ 13 , and DDR1 expression by Western blotting. C, CD-18-tet-MT cells were transfected with siRNA against G␣ 13 , Par3, or both G␣ 13 and Par3. The cells were plated in suspension in a hanging drop for 18 h in the presence of doxycycline. Quantitation of the area of cell clusters was performed using ImageJ. D, CD18-tet-MT cells transfected with siRNA against G␣ 13 and/or Par3 were grown in three-dimensional (3D) collagen in the presence of doxycycline and EGF (20 ng/ml) and allowed to invade for 72 h. Quantitation of invasion was analyzed as detailed in Fig. 2. The number of cell clusters demonstrating a half moon shape was also quantified. The statistics of the invasion assay and the number of half moon cell clusters were performed using one-way ANOVA followed by Dunnett's post-test. The statistics of the hanging drop assay were performed using Student's t test. ns, not significant; **, p Ͻ 0.01; ***, p Ͻ 0.001. The results are representative of three independent experiments. adhesion in MDA-MB-435 cells (20). Consistent with our findings that the ROCK1/2 inhibitor Y27632 fails to block G␣ 13induced pancreatic cancer cell-cell adhesion, the effect of G␣ 13 on cadherin function in MDA-MB-435 cells was also independent of its effect on Rho signaling (20). G␣ 13 siRNA enhances E-cadherin-mediated cell-cell adhesion in our study. However, E-cadherin down-regulation does not increase group proteolytic invasion but, rather, increases the number of single cells in three-dimensional collagen. In contrast, knockdown of DDR1, which also localizes to cell-cell junctions (10,11), reverses the inhibitory effect of G␣ 13 siRNA on group proteolytic invasion in three-dimensional collagen. Significantly, even though concurrent knockdown of G␣ 13 and DDR1 leads to single cells when grown on two-dimensional surfaces, knockdown cells lacking both G␣ 13 and DDR1 are able to undergo group migration in three-dimensional collagen. As DDR1 knockdown cells continue to express E-cadherin protein, it is very likely that DDR1 knockdown cells, in contrast to E-cadherin knockdown cells, can re-establish cell-cell junctions in three-dimensional collagen for group migration.
DDR1 has been shown to either attenuate or promote invasion depending on the type of cancer. DDR1 overexpression increased invasion of lung cancer cells and is associated with increased lymph node metastasis in human lung cancer tumor samples (45). Recently, it has been shown that DDR1 expression was increased in human PDAC tumor samples compared with normal adjacent tissue and that high DDR1 expression in PDAC tumors is associated with a poor prognosis (46). In contrast, DDR1 expression in breast cancer cells was associated with decreased invasion and was linked with increased breast cancer cell differentiation (47, 48). Moreover, DDR1 expression in breast cancer cells was primarily restricted to cell lines that demonstrate minimal to no evidence of epithelial-mesenchymal transition (13), suggesting that DDR1 in breast tumors functions to maintain cells in a more differentiated state. It has also been shown that knockdown of DDR1 enhanced cell-cell dissociation by decreasing the membrane stability of E-cadherin (12). We also show that DDR1 knockdown in pancreatic cancer cells decreases cell-cell adhesion. Additionally, we show that DDR1 knockdown decreases cortical actin structures and increases stress fiber formation, suggesting that DDR1 regulates cellular invasion through modulation of the actin cytoskeleton. Although we do not exclude the possibility that DDR1 may additionally be acting as a collagen-binding protein in our system, the effect of DDR1 siRNA in the absence of extracellular matrix on the hanging drop assay strongly suggests that its effect on invasion is not through collagen binding. Additionally, a previous report has demonstrated that the DDR1 regulation of collective cancer cell invasion was through its effect on cellcell junctions (11). Moreover, the collagen-binding defective mutant of DDR1 had a similar effect on F-actin organization at cell-cell junctions as the wild-type DDR1 (11), suggesting that the DDR1 regulation of collective cancer cell invasion of squamous cancer cells may be independent of its collagen-binding function.
In this report, we also show that the knocking down the polarity protein Par3, which has been shown previously to function downstream of DDR1 in skin cancer cells (11), counteracts the inhibitory effects of G␣ 13 siRNA on pancreatic cancer cell invasion. Significantly, Par3 has been shown recently to regulate tumor progression. For example, Par3 loss increased Rasinduced skin keratoacanthomas and also cooperated with oncogenes to induce breast cancer invasion and metastasis in vivo (49 -51). Loss of Par3 cooperated with the ErbB2 oncogene to induce invasion of mammary epithelial cells (51). Importantly, Par3 protein levels are either reduced significantly or localized abnormally in a large majority of breast tumors compared with normal tissue (51). Moreover, Par3, like DDR1, also regulates E-cadherin junction stability in breast cancer cells. Loss of Par3 in breast cancer cells compromises E-cadherin junctions and decreases cohesion between cancer cells (51). We also show that Par3 knockdown decreases cell-cell cohesion in pancreatic cancer cells. Significantly, we found that G␣ 13 knockdown, in contrast to DDR1 knockdown, caused increased localization of Par3 at cell-cell junctions, suggesting that G␣ 13 and DDR1 differentially regulate Par3 function in pancreatic cancer cells.
Overall, we show that G␣ 13 regulates the invasion of pancreatic cancer cells by disrupting cell-cell adhesion (Fig. 8). We also show that DDR1 blocks three-dimensional collagen invasion by enhancing cell-cell adhesion. Finally, we show that Par3, which is regulated differentially by DDR1 and G␣ 13 , can also attenuate the effects of G␣ 13 on three-dimensional collagen invasion by enhancing cell-cell adhesion.