Regulation of paxillin family members during epithelial-mesenchymal transformation: a putative role for paxillin δ

Gastrulation, neural crest cell migration, organogenesis and tissue repair, as well as tumorigenesis and metastasis each involve the morphological transformation of cells and tissues (Birchmeier and Birchmeier, 1993; Birchmeier et al., 1996; Duband et al., 1982). The transition of cells from primarily immobile epithelia to motile mesenchyme or vice versa can be triggered by multiple stimuli, which include different extracellular matrices and/or soluble factors, and is mediated by multiple internal cellular cues (Birchmeier and Birchmeier, 1993; Gumbiner, 1992). Integrin-extracellular matrix (ECM) engagement and growth factor receptor activation are well-characterized mechanisms of signal transmission mediated through multiple phosphorylation/dephosphorylation events that trigger and modulate internal cellular pathways critically involved in gene expression changes and/or cellular reorganization (Giancotti and Ruoslahti, 1999; Rozengurt, 1995; Schwartz et al., 1995). Subsequent phenotypic changes include altered cell adhesion and migration that are hallmarks of epithelial-mesenchymal transformation (EMT). The coordinated destabilization of cell-cell associations and stimulation of cell-matrix associations through complex intracellular events involving adaptor and signaling proteins is essential for the regulation of the dynamic processes of EMT (Boyer et al., 2000; Savagner, 2001).

The focal adhesion adaptor protein paxillin has been shown to be intricately involved in linking both scaffolding and signaling molecules from sites of integrin and growth factor receptor engagement to the internal actin cytoskeleton (Rozengurt, 1995; Turner, 2000; Turner et al., 1990). Furthermore, paxillin is essential for embryonic development (Hagel et al., 2002) and the elevated levels of paxillin tyrosine phosphorylation are probably important in facilitating cell migration and tissue remodeling within the developing embryo (Turner, 1991; Turner, 2000).

Paxillin is a 68 kDa phosphoprotein originally identified as a substrate for the non-receptor tyrosine kinase Src in Rous sarcoma virus-transformed fibroblasts (Glenney and Zokas, 1989; Turner et al., 1990). Structurally, paxillin consists of an N-terminal region containing five LD protein-protein interaction motifs, a proline-rich region possibly involved in Src-SH3 binding, the SH2 binding phosphotyrosine residues 31 and 118; and a C-terminal region containing four LIM domains that are responsible for focal adhesion targeting and binding to the phosphotyrosine phosphatase PTP-PEST and the microtubule protein tubulin (Brown et al., 1998a; Brown et al., 1996; Cote et al., 1999; Herreros et al., 2000; Schaller and Parsons, 1995; Tumbarello et al., 2002; Turner, 2000).

Functionally, paxillin has been implicated in the regulation of cell adhesion, spreading and motility, muscle differentiation and gene expression through its ability to directly interact with multiple structural and signaling proteins involved in coordinating these events, such as tubulin, p120RasGAP, PKL, PTP-PEST, FAK, Src, Crk and Csk (Brown et al., 1996; Cote et al., 1999; Herreros et al., 2000; Sabe et al., 1994; Schaller and Parsons, 1995; Tsubouchi et al., 2002; Turner et al., 1999). Importantly, paxillin Y31 and Y118 phosphorylation mediates the interaction with Crk and p120RasGAP (Schaller and Parsons, 1995; Tsubouchi et al., 2002). These interactions are associated with cytoskeletal regulation through modulation of the Rho GTPases, Rac1 and RhoA respectively, and appear to perform cell specific roles in regulating integrin signaling and migration (Lamorte et al., 2003; Petit et al., 2000; Tsubouchi et al., 2002). Importantly, both RhoA and Rac1 activity are implicated in actin cytoskeletal remodeling during EMT owing to their involvement in the dissociation of cell-cell adhesion and the formation of cell-ECM adhesion (reviewed by Lozano et al., 2003; Savagner, 2001).

A larger paxillin superfamily exists that includes the paralogues Hic-5/Ara55 (Fujimoto et al., 1999; Shibanuma et al., 1994; Thomas et al., 1999) and leupaxin (Lipsky et al., 1998). These proteins maintain the general structure of paxillin with well-conserved N-terminal LD motifs and C-terminal LIM domains (Tumbarello et al., 2002). Interestingly, a growing literature purports a role for Hic-5 as a natural antagonist of paxillin function presumably by virtue of its capacity to quench signaling through paxillin by competition with shared LD-binding partners, PTP-PEST (Nishiya et al., 1999) and GIT1 (Nishiya et al., 2002), and through suppression of signaling downstream of paxillin tyrosine phosphorylation (Nishiya et al., 2001).

Three paxillin isoforms, α, β and γ produced by alternative splicing have been described (Mazaki et al., 1997; Turner and Miller, 1994). Interestingly, examination of the paxillin nucleotide sequence (Turner and Miller, 1994) reveals an internal AUG codon with a conserved Kozak sequence present downstream of the primary start codon (Kozak, 1987). Subsequent cloning of paxillin cDNAs from multiple species including human, mouse, frog, zebrafish and fly reveal that this downstream `alternative' translation site is evolutionarily conserved.

Herein, we provide evidence for the existence of a truncated form of paxillin which is produced from an internal translation initiation site within the full-length paxillin mRNA. The internally translated form of paxillin α, named paxillin δ to conform to current nomenclature, differs in its N-terminal region by lacking the LD1 domain, the proline-rich region, and the SH2-binding tyrosine residues at amino acid positions 31 and 118. Paxillin δ protein is preferentially expressed in a well-differentiated epithelium, whereas expression is reduced following transforming growth factor β1 (TGF-β1)-induced transition to a mesenchymal phenotype. In contrast, Hic-5 exhibits a reciprocal protein expression profile. Paxillin δ localizes efficiently to focal adhesions where it can suppress the tyrosine phosphorylation of full-length paxillin α, implicating it as a possible negative regulator of integrin signaling. Indeed forced protein expression of paxillin δ inhibited cell migration in normal murine mammary gland (NMuMG) epithelial cells, whereas overexpression of Hic-5 stimulated cell migration. Finally, evidence is presented indicating that tyrosine phosphorylation of paxillin during TGF-β1-induced EMT in NMuMG cells promotes an interaction with the adapter protein Crk, rather than p120RasGAP. We suggest that paxillin δ protein expression in epithelial cell populations serves as an internal competitive inhibitor of integrin signaling through full-length paxillin α to limit inappropriate trans-differentiation and cell migration as occurs during epithelial cell metastasis.

TGF-β was purchased from R&D systems. Antibodies used in these studies were anti-phospho-paxillin Y31 and Y118, anti-phospho-FAK Y397 polyclonal antibodies (Biosource International), phosphotyrosine 4G10 (Upstate), Hic-5, paxillin (clone 165), PKL, N-cadherin, p120RasGAP, p190RhoGAP, Crk, p130Cas, E-cadherin, ILK, FAK (clone 77) and β-catenin monoclonal antibodies (Transduction Laboratories), p120RasGAP clone B4F8 (Santa Cruz), and paxillin-specific monoclonal antibody PXC10 (Sigma). The avian-specific anti-paxillin polyclonal antibody (Pax1) has been described previously (Turner et al., 1990).

Avian paxillin cDNA constructs were cloned into pcDNA3.1 (Invitrogen). Full-length paxillin plasmid with an internal Kozak mutation was generated by site-directed mutagenesis utilizing an oligonucleotide (5′-GAACCATCTCCCTCACTGACCAGCACC-3′) for the insertion of two site-specific nucleotide mutations introducing the amino acids serine at position 132 and leucine at position 133. The paxillin δ plasmid, starting from methionine residue 133 of the avian paxillin cDNA, was generated by PCR and sequenced in its entirety. The pEGFPc1 vector was used in cotransfection studies for the identification of the transfectants.

Chinese hamster ovary (CHO.K1) cells were maintained in modified Ham's F-12 (Mediatech) supplemented with 10% (v/v) heat-inactivated, certified FBS (Atlanta Biologicals), 50 U/ml penicillin and 50 μg/ml streptomycin (Sigma-Aldrich). Normal murine mammary gland (NMuMG) cells purchased from ATCC were maintained in Dulbecco's modified Eagle's medium (Mediatech) supplemented with 10% (v/v) heat-inactivated FBS, 1 mM Glutamine, 50 U/ml penicillin and 50 μg/ml streptomycin. Cells were maintained at 37°C in a humidified chamber with 5% CO₂. Transfection of CHO.K1 cells was performed with Fugene 6 (Roche Molecular Biochemicals) according to the manufacturer's instructions. For EMT induction of NMuMG cells, 3×10⁵ cells were plated in six-well dishes 24 hours prior to stimulation in complete DMEM, 10% FBS. Cells were either stimulated with vehicle alone (4 mM HCl, 0.1% BSA) or with 2 ng/ml TGF-β1 for 48 hours.

The pLEGFPc1 retroviral vector (BD clontech) was provided by B. Pawlikowski (SUNY Upstate Medical University). XhoI and BamHI restriction sites were generated by PCR flanking the full-length paxillin α (containing internal Kozak mutation) (GFP-FLpax), paxillin δ and Hic-5 cDNA for subsequent cloning into the pLEGPc1 vector. Production of retroviral supernatant was performed as described on the Nolan Laboratory website (http://www.stanford.edu/group/nolan/). Briefly, Phoenix 293 packaging cells (obtained from B. Pawlikowski, SUNY Upstate Medical University, NY) were transfected with the retroviral constructs by calcium phosphate-mediated transfection followed by collection of retrovirus conditioned media 24-72 hours post-transfection. NMuMG cells, plated in six-well dishes, were infected by treatment with retroviral conditioned media supplemented with 4 μg/ml polybrene (Sigma-Aldrich) followed by centrifugation at 1000 g for 20 minutes at 32°C.

Modified Boyden chamber migration assays were performed as previously described (Riedy et al., 1999). An 8 μm polycarbonate filter (Neuroprobe) was pre-coated with 100 μg/ml gelatin type B (Sigma-Aldrich) at room temperature for 16 hours. Briefly, 10,000 cells were plated in the upper chamber of a Boyden chamber apparatus and the cells were allowed to migrate to 10 μg/ml fibronectin in serum-free media (Sigma-Aldrich) for 16 hours at 37°C. The polycarbonate filter was then fixed in methanol and stained with Giemsa (Sigma-Aldrich). Cells migrating to the underside of the membrane were quantified by obtaining an absorbance value at 540 nm. Values are the mean of three experiments. The migration of the GFP control was set to 100%, and the other cell types were measured against this value. Statistical analysis was performed using Student's t-test.

3D collagen I gels were prepared on ice using equal volumes of Vitrogen 100 (Cohesion) and 2× HEPES-buffered salt solution [50.4 mM HEPES, pH 7.4, 162.6 mM NaCl, 10.6 mM KCl, 88.2 mM NaHCO₃, 1.6 mM Na₂HPO₄, and 11 mM D(+)-glucose] (Saelman et al., 1995) yielding a concentration of 1.3 mg/ml following addition of culture medium. The collagen gel solution (0.7 ml) was added to each 35 mm tissue culture dish and allowed to gel at 37°C for 30 minutes. NMuMG cells were collected in a single cell suspension and resuspended in the collagen gel solution at a concentration of 3×10⁵/ml followed by the addition of 300,000 cells onto the preformed collagen gel. The collagen mixture was allowed to gel at 37°C for 30 minutes before the addition of 1.5 ml culture medium. Culture medium was replenished every 2-3 days until day 7, at which time complex structures had formed. For 2D culture, cells were seeded at 300,000 cells/35 mm tissue culture dish in growth media and were cultured alongside 3D collagen gel cultures. Hoffman modulation contrast microscopy was performed with a Nikon Eclipse TE-300 microscope equipped with a 20× objective lens and a Spot™ RT Monochrome camera (Diagnostic Instruments). The images were processed using SPOT™ RT Software v3.0 (Diagnostic Instruments).

NMuMG cell lysates were prepared in RIPA buffer: 1% NP-40, 0.1% SDS, 1% sodium deoxycholic acid, 150 mM NaCl, 20 mM Tris-HCl pH 7.6, 5 mM EDTA, 1 mM PMSF, 10 μg/ml leupeptin (Sigma-Aldrich), and phosphatase inhibitors (2 mM sodium fluoride, 2 mM sodium pyrophosphate, 1 mM Na₃VO₄; Sigma-Aldrich). For immunoprecipitation, cell lysates were prepared in lysis buffer: 1% Triton X-100, 0.1% sodium deoxycholic acid, 96 mM NaCl, 10 mM Tris-HCl pH 7.6, 1 mM EDTA, 1 mM PMSF and 10 μg/ml leupeptin. BioRad D_c Protein Assay was performed on cell lysates. For NMuMG detergent-soluble lysates, 25 μg lysate was mixed with an equal volume of 2× SDS-sample buffer. For Pax1 immunoprecipitation, 200-400 μg protein lysate was incubated with the Pax1 antiserum and Protein A/G agarose at 4°C for 1 hour with rotation, followed by washing in lysis buffer. Samples were boiled in SDS sample buffer for 5 minutes prior to being loaded and run on 7.5% SDS-PAGE and transferred to Immobilon NC (Millipore). Western blot analysis was performed as previously described (West et al., 2001).

Prior to cell lysis, cells were washed once with ice-cold PBS. Cells were lysed in 1% Triton X-100, 10 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 10% Glycerol, 10 μg/ml leupeptin, and phosphatase inhibitors. Cell lysates were centrifuged at 14,000 g for 15 minutes at 4°C. 800-1000 μg cell lysate were utilized for coimmunoprecipitation experiments. Cell lysates were either incubated with 1 μg Crk, p120RasGAP (clone B4F8), paxillin (clone 165) or control IgG (Santa Cruz) antibodies for 2 hours at 4°C followed by the addition of 12.5 μl Protein A/G Agarose (1:1 suspension) for 1 hour at 4°C with rotation. Bead complexes were washed twice with cold lysis buffer and boiled with 2× SDS sample buffer. Western blot analysis was performed as previously described (West et al., 2001).

For respreading assays, CHO.K1 cells were removed from tissue culture dishes by washing with PBS followed by incubation for 3-5 minutes with PBS supplemented with 1 mM EDTA. Cells were collected and washed twice with Ham's F-12, 10% FBS and once with serum-free Ham's F-12 containing 1% BSA. Cells were resuspended in serum-free media supplemented with 1% BSA and placed on a rocker in suspension for 1 hour at 37°C. Cells were replated on fibronectin-coated coverslips (10 μg/ml) in serum-free media supplemented with 1% BSA and then subsequently processed for indirect immunofluorescence microscopy at the indicated time points as previously described (West et al., 2001). Imaging was performed utilizing a Zeiss Axiophot photomicroscope equipped with epifluorescence illumination using a SPOT™ RT slider camera (Diagnostic Instruments). Images were processed using Adobe Photoshop™ v6.0.1.

The 68 kDa form of paxillin, paxillin α, is ubiquitously expressed. Two slightly larger, alternatively spliced forms β and γ exhibit a more restricted distribution (Mazaki et al., 1997; Mazaki et al., 1998). Additionally, several reports have described the existence of a 46 kDa immunoreactive paxillin band speculated to be the result of proteolytic cleavage (Sorenson and Sheibani, 1999; Surin et al., 2000; Yamaguchi et al., 1994). However, examination of the paxillin nucleotide sequence reveals the existence of an internal Kozak sequence conforming to the consensus GCCRCCaugG (R, purine), with a potential initiation methionine corresponding to amino acid 133 in paxillin α (human and avian) and amino acid 134 in murine paxillin α. An alignment of all paxillin sequences currently available reveals that this putative internal initiation site is evolutionarily conserved from humans to flies (Fig. 1A). The predicted molecular weight of a protein initiated at this internal start site is 46.9 kDa. This raises the intriguing possibility that in certain cases the lower molecular weight form of paxillin previously described may be the product of an internal translation mechanism.

Examination of the domain structure of paxillin α indicates that an internally translated paxillin form, from here on referred to as paxillin δ, would lack several important functional domains of the full-length protein including the LD1 motif involved in ILK and actopaxin binding (Nikolopoulos and Turner, 2000; Nikolopoulos and Turner, 2001), the putative Src SH3 domain-binding proline-rich region (amino acids 46-53) (Weng et al., 1993) and the tyrosine residues at amino acids 31 and 118 that when phosphorylated are utilized in the binding of Crk and p120RasGAP (Schaller and Parsons, 1995; Tsubouchi et al., 2002) (Fig. 1B). It should be noted that this domain structure would be generated in all paxillin δ orthologues except for Drosophila. In this case, the initiation codon resides downstream of the LD2 domain. Therefore, in the case of flies, translation at this internal site would lead to a product lacking this particular domain thereby potentially affecting associations with FAK and vinculin (Brown et al., 1996). How this might result in differential function for paxillin δ in flies is unknown. Lastly, as the paxillin β and γ splice sites are downstream of the internal initiation start site (see Fig. 1B), there is a possibility that multiple isoforms of paxillin δ could be generated. This, along with phosphorylation, may explain the diffuse nature of the 46 kDa band observed upon western blotting of cell lysates (see Fig. 3) as has been previously described for full-length paxillin α (Turner et al., 1990).

Expression in CHO.K1 cells of the wild-type avian paxillin cDNA (wt pax) containing both the primary and internal Kozak sequences generates two protein products of 68 kDa and 46 kDa as detected with the paxillin-specific monoclonal antibody, PXC10 (Sigma), which recognizes an epitope within the LIM domains (C.E.T., unpublished observations) (Fig. 2A). Both products are also recognized by the paxillin 165 and 349 antibodies (Transduction Labs) (data not shown). Two paxillin protein products, also of 68 and 46 kDa, are produced from the same paxillin cDNA following in vitro transcription/translation (Nikolopoulos and Turner, 2000).

To determine if the 46 kDa product is the result of initiation of translation at the internal methionine residue at position 133, a double point mutant designated FL pax KM, was generated in which the internal Kozak sequence and the initiating methionine were ablated (see Materials and Methods for details). Expression of this paxillin construct eliminated production of the 46 kDa protein band (Fig. 2A), confirming that this product was the result of internal initiation of translation and not proteolysis. Next, we produced a truncated paxillin cDNA that would generate a translation product beginning at amino acid 133. Expression of this truncated paxillin construct, pax δ, produced a protein with a relative mobility identical to the low molecular weight paxillin product generated from the cDNA encoding wild-type paxillin (Fig. 2A). These data, along with the presence of a highly conserved internal Kozak consensus sequence, and the lack of evidence suggesting the presence of an alternatively spliced mRNA transcript that may encode paxillin δ (Mazaki et al., 1998; Salgia et al., 1995), further indicates that paxillin α contains a functional internal translation initiation site.

The LIM 3 domain of paxillin has previously been shown to be essential for paxillin localization to focal adhesions (Brown et al., 1996). The LIM domains are unaltered in paxillin δ. Thus, to confirm that the absence of the N-terminal portion of paxillin (amino acids 1-132) has no detrimental effect on the focal adhesion targeting of paxillin δ, exogenous expression studies followed by immunofluorescence microscopy were performed in CHO.K1 cells. Expression of avian paxillin δ in CHO.K1 cells followed by staining with the avian-specific anti-paxillin antibody (Pax1) (Turner et al., 1990), demonstrated robust localization of paxillin δ to vinculin-rich focal adhesions (Fig. 2Bc,f), similar to wild-type paxillin (Fig. 2Ba,d). As expected, the full-length paxillin protein, corresponding to paxillin α and derived from the cDNA containing mutations in the internal Kozak region (FL pax KM), also targeted efficiently to focal adhesions (Fig. 2Bb,e).

To further test for the natural occurrence of paxillin δ, the protein expression profile was examined in a series of cell lines utilizing the paxillin-specific antibody, PXC10 (Fig. 3). Interestingly, although paxillin α (68 kDa) was expressed in all cell lines examined, including those of both mesenchymal (lanes 1-4) and epithelial origin (lanes 5-7), the 46 kDa product corresponding to paxillin δ was predominantly expressed only in the epithelial cells, NMuMG, MDCK and NRK. In striking contrast, Hic-5, detected using a Hic-5-specific antibody (Transduction Labs) was primarily expressed in mesenchymal cells (Fig. 3, lanes 1-4). Our results indicate a suppression of Hic-5 protein expression in normal epithelial cells and although Hic-5 mRNA may be present in certain transformed epithelia such as HeLa cells (Zhang et al., 2000) and in selected prostate cancer cells (Mestayer et al., 2003), protein expression was not confirmed in these reports. Regardless, our data indicate that paxillin δ and Hic-5 exhibit a reciprocal expression pattern in a variety of normal epithelial and mesenchymal cell lines.

A natural transition between epithelial and mesenchymal phenotypes (EMT) occurs during tissue remodeling associated with embryonic development, wound repair and during tumor metastasis (Birchmeier and Birchmeier, 1993; Duband et al., 1995; Thiery, 2002). We were interested in determining if the reciprocal expression pattern exhibited by paxillin δ and Hic-5 could be recapitulated during EMT induced in culture. The murine mammary gland epithelial cell line, NMuMG, can be induced to undergo an EMT when treated with TGF-β1 (Miettinen et al., 1994). In the epithelial state, NMuMG cells exhibit a typical `cobblestone' morphology. These cells have relatively few focal adhesions, a cortical arrangement of actin filaments, and there is a substantial cytoplasmic pool of paxillin (Fig. 4Aa,c,e) (Nakamura et al., 2000). Following stimulation with 2 ng/ml TGF-β1 for 48 hours, a morphologic change is induced which is characterized by a dissociation of cell-cell contacts and the appearance of a flattened, elongated fibroblastic phenotype in which paxillin is localized predominantly to focal adhesions at the ends of well-organized actin stress fibers (Fig. 4Ab,d,f) (Nakamura et al., 2000).

Consistent with previous reports, the TGF-β1-induced EMT coincides with the loss of cell-cell adherens junctions and the downregulation and relocalization of E-cadherin to the perinuclear region (Fig. 5A; data not shown). Conversely, N-cadherin is upregulated and becomes localized at the cell membrane (Fig. 5A; data not shown). Biochemical characterization of the paxillin profile during TGF-β1-induced EMT indicates that full-length paxillin α exhibits a substantial upward shift in electrophoretic mobility resulting from multiple phosphorylation events including phosphorylation of Y31 and Y118 (Fig. 5B) (Nakamura et al., 2000). There may also be a modest increase in protein expression, as reported following TGF-β1 treatment of malignant astrocytoma cells (Han et al., 2001). In striking contrast, paxillin δ expression is highest when cells are maintained in the epithelial state and is significantly reduced after the TGF-β1-induced transition to the mesenchymal phenotype (Fig. 5A,B).

As noted earlier, Hic-5 expression was found to be selectively elevated in cells of mesenchymal origin (Fig. 3). In addition, Hic-5 was initially identified as a TGF-β1-inducible gene (Shibanuma et al., 1994). Using the NMuMG model, we wanted to determine whether Hic-5 might be upregulated during the induction of an EMT, concomitant with the loss of paxillin δ. Interestingly, Hic-5 protein expression is very low when NMuMG cells are maintained in the epithelial state but increases significantly during EMT (Fig. 5A). Consistent with this modulation in protein expression, immunocytochemistry reveals that Hic-5 staining is barely detectable in NMuMG cells prior to TGF-β1stimulation (Fig. 4Ba), but exhibits robust localization to focal adhesions following the TGF-β1-induced EMT (Fig. 4Bb). Thus, the loss of paxillin δ expression and the gain of Hic-5 expression probably represent an important reciprocal functional relationship in the context of TGF-β1-induced EMT in NMuMG cells.

Culture of epithelial cells in a 3D collagen I gel provides a suitable environment to support the generation of well-differentiated epithelium in the form of spherical cysts and/or tubules (Hall et al., 1982) (reviewed by Zegers et al., 2003). Previous reports have demonstrated the downregulation of specific focal adhesion proteins, FAK, talin and p130Cas, following the culture of MDCK cells on a collagen I gel and consequently a reduction in integrin-mediated signaling through these proteins (Wang et al., 2003). We therefore sought to determine if paxillin δ was preferentially expressed in cells cultured in a 3D environment, consistent with our predicted role for paxillin δ in the maintenance of an epithelial phenotype. NMuMG cells were cultured under normal 2D culture conditions or cultured within collagen I gels. Following a 7-day incubation period, NMuMG cells cultured under 2D conditions exhibit a typical cobblestone-patterned monolayer (Fig. 6Aa), whereas cells grown under 3D conditions form complex structures indicative of cysts and tubules (Fig. 6Ab). The expression profiles of various focal adhesion proteins were assessed by western blot analysis. Interestingly, the expression level of paxillin δ was maintained in 3D cultured cells at levels comparable to that detected in cells cultured under 2D conditions (Fig. 6B,C). In striking contrast, the protein expression of other focal adhesion components including vinculin, p130Cas and FAK were significantly decreased (Fig. 6B). Importantly, similar results were obtained utilizing MDCK cells cultured under identical conditions (data not shown).

In addition to a reduction in FAK expression, phosphorylation at Tyr397 (an indicator of FAK activity) was correspondingly reduced in the 3D culture conditions, consistent with an attenuation in integrin signaling under these conditions (Fig. 6B). Although paxillin δ expression remained unchanged, full-length paxillin α expression in cells cultured under 3D conditions was reduced along with a significant reduction in paxillin Y118 phosphorylation, as determined following paxillin immunoprecipitation (Fig. 6C). Immunodetection of paxillin α in the total lysates was not possible with the 3D cultures owing to the presence of large amounts of co-migrating matrix/serum proteins. In accordance with previous reports, the decrease in tyrosine phosphorylation of FAK and paxillin in cells cultured under 3D conditions is not likely to be the result of inefficient extraction of protein (Wozniak et al., 2003). Paxillin tyrosine phosphorylation is tightly coupled to FAK/Src activity (Turner, 2000). Thus, the reduction in paxillin phosphorylation in both 2D and 3D cultures provides further evidence for a suppression of this particular axis of the integrin signaling pathway. Furthermore, under 3D culture conditions that more closely resemble the physiologic environment of epithelial tissues by permitting complete cell differentiation, paxillin δ expression, in contrast to a number of other focal adhesion proteins including paxillin α, is maintained at a relatively high level. These results further suggest a role for paxillin δ in maintaining differentiated epithelia by possibly preventing EMT-associated cell migration through the suppression of integrin signaling to paxillin α.

To evaluate how paxillin δ may influence normal integrin signaling to paxillin α, exogenous expression studies were performed in CHO.K1 cells. Cells were transiently transfected with either full-length paxillin Kozak mutant (FL pax KM) or paxillin δ constructs. Following protein expression, the cells were harvested and then replated on fibronectin to induce cell spreading and thereby integrin signaling (Fig. 7A). At time points ranging from 60 to 240 minutes, overexpression of the FL pax KM construct did not have any significant effect (<12%) on paxillin Y118 phosphorylation in focal adhesions (as measured using a phospho-paxillin Y118 antibody) (Fig. 7Ba,b and C) when compared to vector control (data not shown), consistent with this paxillin construct functioning in an identical fashion to the endogenous protein. In contrast, cells transiently transfected with the paxillin δ cDNA showed a substantial reduction in focal adhesion staining for endogenous paxillin Y118 phosphorylation at all time points (Fig. 7Bc,d and C). Similar results were obtained using anti-phospho Y31 antibodies (data not shown). Interestingly, the paxillin δ transfectants showed no obvious spreading defect, consistent with previous reports indicating that phosphorylation of Y31 and Y118, although critical for Crk-mediated cell migration in certain cell types (Petit et al., 2000), is not required for cell adhesion (Brown et al., 1998b) or cell spreading (Wade et al., 2002). Unfortunately, antibodies that can discriminate between full-length paxillin and paxillin δ are not available. Thus, we were unable to determine whether the loss of phospho-Y118 staining in focal adhesions is due to the exclusion of paxillin α from focal adhesions or potentially due to paxillin δ sequestering paxillin-binding proteins such as FAK/Src and thereby abrogating their ability to efficiently phosphorylate full-length paxillin at Y31 and 118 within focal adhesions. Nevertheless, these results clearly demonstrate that paxillin δ can suppress integrin signaling to paxillin α.

A retroviral-based expression system was utilized to overexpress GFP, GFP-full-length paxillin α Kozak mutant (GFP-FL pax) and GFP-paxillin δ (GFP-pax δ) in NMuMG cells. Cells expressing GFP, GFP-FL pax and GFP-pax δ each exhibited a morphology indicative of cells with an epithelial phenotype (Fig. 8A; data not shown). However, GFP-pax δ overexpression resulted in reduced levels of endogenous paxillin Y118 phosphorylation in focal adhesions located along the peripheral edges of epithelial cell islands (Fig. 8Ab,d,f; indicated by arrows) and also suppressed paxillin α phosphorylation in response to TGF-β1, compared to GFP or GFP-FL pax expressing cells (Fig. 8B). Additionally, phosphotyrosine (4G10 clone) immunoblot of ectopic paxillin δ immunoprecipitates indicated a very low level of tyrosine phosphorylation (Fig. 8C, arrow). As paxillin δ lacks Y31 and Y118, this indicates paxillin δ may be phosphorylated at other potential tyrosine residues, specifically Tyr181, Tyr434 and Tyr488 (Nakamura et al., 2000) (corresponding to Tyr182, 436 and 490 of the avian cDNA). The significance of this low stoichiometry of phosphorylation remains to be determined. Interestingly, paxillin δ overexpression almost completely blocked the induction of Hic-5 expression in response to TGF-β1, whereas the downregulation of E-cadherin was unaffected (Fig. 8B). Full-length paxillin α overexpression also caused a modest attenuation in Hic-5 expression in response to TGF-β1 (Fig. 8B), possibly indicating that the overall level of paxillin α and paxillin δ is important in controlling Hic-5 levels.

The ability of paxillin δ to suppress integrin signaling to paxillin α in both CHO.K1 and NMuMG cells, as evidenced by reduced paxillin α phosphorylation, along with previously published data regarding the importance of paxillin α phosphorylation in cell migration (Nakamura et al., 2000; Petit et al., 2000; Tsubouchi et al., 2002), suggests a potential role for paxillin δ in suppressing a migratory phenotype. Thus, we used Boyden chamber transwell assays to evaluate the impact of paxillin δ on cell migration. NMuMG cells overexpressing GFP-pax δ exhibited a significant reduction in migration to fibronectin compared to levels in the GFP control and GFP-FL pax expressing cells (Fig. 8D; P<0.001). Importantly, cells overexpressing Hic-5 exhibited a significant stimulation of migration to fibronectin (Fig. 8D; P<0.001). The small reduction in migration seen with the GFP-FL pax overexpressing cells (Fig. 8D; P<0.05) is consistent with previous reports (Yano et al., 2000). Together these data reinforce the notion that paxillin δ may be functioning to preserve an immobile epithelium through a suppression of full-length paxillin tyrosine phosphorylation and Hic-5 expression, and thereby limiting the ability of the cell to migrate in the absence of an appropriate physiologic stimulus.

As indicated, paxillin δ downregulation following TGF-β1-induced EMT of NMuMG cells correlated with a robust increase in paxillin phosphorylation at Y31 and Y118. In order to delineate the biochemical interactions associated with paxillin following tyrosine phosphorylation, coimmunoprecipitation experiments were performed with NMuMG cells to evaluate the interactions of previously described binding partners. The Y31 and Y118 phosphorylation sites mediate direct associations with the SH2 domains of Crk and p120RasGAP and thereby initiate signaling events associated with actin cytoskeletal remodeling. For instance, in NBT-II bladder carcinoma cells a paxillin-Crk association is essential for migration on collagen (Petit et al., 2000). In contrast, Sabe and colleagues could not detect this interaction in NMuMG cells and instead suggested that p120RasGAP binding to paxillin contributed to p190RhoGAP-mediated Rho inhibition to promote lamellipodial extension and motility (Tsubouchi et al., 2002).

To evaluate which of the endogenous paxillin binding proteins is utilized in our model system, a series of coimmunoprecipitation experiments were performed with NMuMG cells cultured under conditions promoting epithelial monolayers or following treatment with TGF-β1 to induce EMT. Interestingly, Crk is able to coimmunoprecipitate the tyrosine-phosphorylated form of paxillin α under both conditions, but the amount of paxillin α precipitated increased following EMT (Fig. 9A) concomitant with the increase in tyrosine phosphorylation. Increased levels of p130CAS and FAK (probably via an indirect association with Cas or paxillin) co-precipitating with Crk were also observed following TGF-β1 stimulation (Fig. 9A). Of particular note is that the increase in paxillin α association with Crk correlates with the decrease in paxillin δ expression (Fig. 9A). Also, owing to the lack of the primary SH2-binding consensus phosphorylation sites on paxillin δ, it does not interact with Crk (Fig. 9A).

Parallel experiments were performed to evaluate a possible paxillin-p120RasGAP interaction. We were unable to detect any paxillin co-precipitating with p120RasGAP before or after TGF-β1 stimulation (Fig. 9B). Similarly, immunoprecipitation of paxillin from TGF-β1-treated NMuMG cell lysates failed to coimmunoprecipitate p120RasGAP while successfully coimmunoprecipitating the well-characterized paxillin binding partner, paxillin kinase linker (PKL) (Fig. 9C). The association of p120RasGAP with p190RhoGAP occurs through the identical SH2 domain regions of p120RasGAP that are involved in paxillin binding. Consistent with the absence of any paxillin binding to p120RasGAP we failed to detect any decrease in levels of the p120RasGAP/p190RhoGAP complex following TGF-β1 treatment (Fig. 9B). Together, these results suggest that, as in other epithelial cell types, a functional paxillin-Crk interaction probably contributes to integrin signaling events in NMuMG cells following the induction of EMT. Paxillin δ expression may contribute to suppression of this association thereby limiting the ability of epithelial cells to migrate inappropriately.

We report the characterization of a new 46 kDa paxillin isoform, paxillin δ. Production of paxillin δ is initiated from an internal downstream `alternative' translation start site (Fig. 1). Importantly, the internal translation start site and consensus Kozak sequence are conserved across species from human to flies (Fig. 1). Paxillin δ expression is elevated in epithelial cells whereas the protein is absent from mesenchymal cells (Fig. 3). It is enriched relative to other focal adhesion proteins following differentiation (Fig. 6), and is regulated during the distinct morphological changes that occur during TGF-β1-induced EMT (Figs 4 and 5). Expression of paxillin δ in CHO.K1 cells inhibits paxillin α tyrosine phosphorylation during cell spreading on fibronectin (Fig. 7) similar to that described for the paxillin superfamily member Hic-5 (Nishiya et al., 2001). Interestingly, paxillin δ exhibits a reciprocal protein expression profile to that of Hic-5; i.e. epithelial cells preferentially expressed paxillin δ over Hic-5, whereas the reverse was true of fibroblasts (Fig. 3). Further, Hic-5 protein expression is upregulated during EMT whereas paxillin δ is downregulated (Figs 4 and 5). In addition, ectopic expression of paxillin δ in NMuMG cells inhibited cell migration to fibronectin, whereas Hic-5 overexpression resulted in elevated levels of cell motility. Taken together, our data indicate that paxillin δ exists as a regulated product of internal translation and potentially functions as a physiologic inhibitor of full-length paxillin signaling. If paxillin δ functions in this capacity, it would, most likely, have to directly compete at focal adhesions to disrupt downstream signaling. Even though there are higher levels of full-length paxillin α compared to paxillin δ in epithelial NMuMG cells, only a relatively small pool of total paxillin is present in focal adhesions, with a majority of the localization being cytoplasmic (Fig. 4A). Following TGF-β1-induced EMT this balance is shifted to a predominant focal adhesion localization and an increase in full-length paxillin α tyrosine phosphorylation and Hic-5 expression, concomitant with a significant reduction in paxillin δ expression. This may indicate that when cells are cultured as an epithelium the elevated level of paxillin δ may be sufficient to compete with full-length paxillin at focal adhesions thereby preserving the integrity of the epithelia through suppression of certain downstream signaling events.

In addition to our data demonstrating the regulated expression of paxillin δ, evidence from others suggests that expression of this 46 kDa paxillin isoform may be regulated in other physiological contexts associated with morphological changes such as those occurring during immune regulation and development. For example, interferon treatment of U937 leukemia cells has been reported to trigger a downregulation of 44-46 kDa paxillin isoforms and concomitant increase in the full-length 68-70 kDa isoforms whereas the opposite is observed with primary monocytes (Surin et al., 2002). During renal morphogenesis where the requirement for cell migration decreases, full-length paxillin expression has been shown to decrease while expression of a lower molecular weight protein of approximately 46 kDa, recognized by paxillin-specific antibodies, is increased (Sorenson and Sheibani, 1999). Additionally, it has been reported that the induction of cell-cell adhesion in Ewing's sarcoma cells by MMP-9 siRNA results in an increase in the expression of a 44-46 kDa paxillin product (Sanceau et al., 2003). In fact, evidence has been provided that the 46 kDa protein could be derived from translation at methionine residue 133 (Wade et al., 2002). These data support the normal and regulated expression of this new paxillin isoform.

Translational control has long been recognized as a principal means to regulate gene expression and cell function and considerable progress has been made in elucidating the pathways that govern control at the level of internal translation initiation (Sonenberg et al., 2000). The precise mechanism of regulation of paxillin δ expression remains to be determined; however, paxillin has a 5′-UTR of over 700 nucleotides with six upstream AUG sequences and is thus categorized as a weak mRNA (Geballe and Morris, 1994). Weak mRNAs that have complex 5′ secondary structure tend to display inefficient translation from the principal translational start and consequently may be outcompeted by strong mRNAs and may also exhibit shunting to a downstream translation start that is less encumbered by secondary structure (Sonenberg et al., 2000). As the 5′ m⁷G-cap binding translation initiation factor eIF4E is limiting in cells, those mRNAs with complex secondary structure can be outcompeted in ribosome binding (Pelletier and Sonenberg, 1987). Activation of eIF4E by growth factor stimulation, inactivation of 4E-BP (Gingras et al., 2001) and/or upregulation of eIF4E, as occurs in many cancers (De Benedetti and Harris, 1999), can override the inefficiency to effect translation of the upstream start at the expense of the downstream site. In such a manner, paxillin δ translation may be regulated as a result of TGF-β1 stimulation. It is also of interest that paxillin was recently found to interact with the poly(A) binding protein 1 (Woods et al., 2002) as it has been established that a functional association between the eIF4E/5′-cap RNA structure and the PABP1/3′-poly(A) RNA tail through eIF4G is a mechanism of translational stimulation that may be regulated through derepression of an inhibitory complex (Sonenberg et al., 2000). Although paxillin δ expression may be regulated at the translational level, there exists the possibility that paxillin δ has a decreased stability under certain conditions relative to full-length paxillin, thereby causing a decrease in its expression. Further work will be required to determine the mechanism of regulation of paxillin δ and Hic-5 expression and whether paxillin may function in a self-regulatory loop.

The expression of isoforms that represent truncated products is an emerging theme associated with signal regulation via focal adhesion proteins. The PKL/GIT2/CAT2 Arf-GAP protein has an alternatively spliced mRNA that generates a truncated product termed KIAA0148/GIT2short that lacks the C-terminal paxillin binding site (Bagrodia et al., 1999; Mazaki et al., 2001; Premont et al., 2000; Turner et al., 1999). Additionally, the CAT2 binding protein, p85Cool-1/βPix, has an alternatively spliced variant termed p50Cool-1 (Bagrodia et al., 1998). Intriguingly, Hic-5 exists in two isoforms that are the products of alternative splicing, `full-length' Hic-5 and a product termed ARA55 that lacks the first 17 amino acids, which includes LD1 (Fujimoto et al., 1999; Shibanuma et al., 2000; Thomas et al., 1999). Similarly, alternative splice products of FAK and PYK2, termed FAK-related non-kinase (FRNK) and PRNK, respectively, have been identified and demonstrated to operate as functional antagonists of the `full-length' isoform in the context of integrin signaling (Schaller et al., 1993; Xiong et al., 1998).

The generation of a paxillin isoform that lacks well-defined protein binding interfaces and the principle sites of tyrosine phosphorylation but maintains other binding sites and the focal adhesion targeting motif may permit differential signaling as compared to the full-length protein. Paxillin α is tyrosine phosphorylated in response to various growth factors such as epidermal growth factor (EGF), angiotensin II and TGF-β1, as well as in response to integrin engagement during cell migration and adhesion (Burridge et al., 1992; Nakamura et al., 2000; Petit et al., 2000; Riedy et al., 1999; Turner et al., 1995). Increased tyrosine phosphorylation of focal adhesion components, including paxillin, has been associated with the transformation of epithelial cells to a more motile and invasive phenotype (Kinch and Burridge, 1995; Mueller et al., 1992). Following integrin engagement, phosphorylation of paxillin at distinct SH2-binding YXXP motifs by the FAK-Src complex leads to the recruitment and formation of a variety of adaptor and signaling proteins that are involved in the transduction of signals through specific Rho GTPases, leading to actin cytoskeleton reorganization during cell migration (Nobes and Hall, 1995; Petit et al., 2000; Schaller and Parsons, 1995; Tsubouchi et al., 2002). In particular, paxillin phosphorylation at Y31 and Y118 by the FAK-Src complex (Schaller and Parsons, 1995) creates binding sites for the Crk family of adapter proteins, the C-terminal Src kinase, Csk, as well as p120RasGAP (Petit et al., 2000; Sabe et al., 1994; Tsubouchi et al., 2002).

The paxillin Y31 and Y118 residues have been shown to be necessary for Crk-mediated cell migration in NBT-II bladder epithelial cells (Petit et al., 2000) and Crk-mediated lamellipodia formation and cell spreading in MDCK epithelial cells (Lamorte et al., 2003). Paxillin tyrosine phosphorylation and its association with Crk probably contributes to Rac1 activation via the Crk/DOCK180/ELMO complex to promote the motile phenotype (Gumienny et al., 2001). Our own results indicate that paxillin tyrosine phosphorylation also promotes association with Crk in the NMuMG cell line, in contrast to an earlier study using these cells, which indicated that Y31- and Y118-phosphorylated paxillin preferentially binds p120RasGAP thereby facilitating cell migration through an inhibition of RhoA activity (Tsubouchi et al., 2002). The authors proposed that this is accomplished by the association of paxillin with p120RasGAP causing a dissociation of p190RhoGAP from p120RasGAP and a resultant activation of p190RhoGAP. However, contradictory evidence has suggested that phosphorylated p190RhoGAP associated with p120RasGAP is the active form of p190RhoGAP (Noren et al., 2003). Regardless of the mechanism of RhoGAP activation, we were unable to detect any association between paxillin and p120RasGAP and no effect on the interaction between p120RasGAP and p190RhoGAP despite the TGFβ1-induced paxillin tyrosine phosphorylation in NMuMG cells. Thus, as reported in other epithelial cell model systems, a regulated Crk association with paxillin is likely to play a critical role in the regulation of morphology and motility in NMuMG cells. Although the reason for the discrepancy between our own results and the other NMuMG studies is unclear, the difference could be attributed to cell-specific differences as a number of clonal lines exhibiting a variety of phenotypes have been derived from the parental NMuMG population (Soriano et al., 1995; Zutter et al., 1999).

Although our results may indicate that the primary role of paxillin δ is its suppression of full-length paxillin signaling mediated through the SH2-binding tyrosine residues, another important distinction between full-length paxillin and paxillin δ is the loss of the LD1 protein interaction motif. The paxillin LD1 motif has been shown to mediate interactions with the actin-binding protein actopaxin and the serine/threonine integrin-linked kinase, ILK (Nikolopoulos and Turner, 2000; Nikolopoulos and Turner, 2001). Actopaxin is involved in the regulation of the actin cytoskeleton during the processes of cell adhesion and motility (Nikolopoulos and Turner, 2000; Tu et al., 2001); whereas ILK has been implicated in the regulation of cell growth, fibronectin-matrix assembly, cell adhesion and EMT (Hannigan et al., 1996; Nikolopoulos and Turner, 2001; Somasiri et al., 2001; Wu et al., 1998). Overexpression of ILK in the intestinal epithelial cell line, IEC-18, leads to the downregulation of E-cadherin and the production of an organized fibronectin matrix, which are characteristic of a mesenchymal phenotype (Wu et al., 1998). Overexpression of ILK in a mammary epithelial cell line, scp2, has also been directly linked to changes characteristic of a transition to a mesenchymal phenotype (Somasiri et al., 2001), which may be related to the capacity of ILK to induce an active LEF-1/β-catenin complex (Novak et al., 1998; Persad et al., 2001). Expression of paxillin δ in epithelial cells may regulate the localization and functional activity of actopaxin and ILK providing an additional mechanism for regulation of cytoskeleton reorganization and the repression of a motile phenotype.

Although several reports have detailed the capacity of Hic-5 overexpression to antagonize paxillin function in cell spreading and motility (Nishiya et al., 1999; Nishiya et al., 2001) the NMuMG epithelial cell line does not necessarily conform to this paradigm. First, paxillin is primarily cytosolic in a hypophosphorylated form and Hic-5 is undetectable when these cells are cultured as an epithelium (Fig. 4). Induction of an EMT by stimulation with TGF-β1 resulted in a robust stimulation of paxillin localization to focal adhesions and an increase in tyrosine phosphorylation, concurrent with a dramatic induction of Hic-5, robust focal adhesion localization and a substantial reduction in paxillin δ expression (Figs 4 and 5). In addition, overexpression of paxillin δ inhibited paxillin α tyrosine phosphorylation during adhesion and spreading on fibronectin (Fig. 7) as well as during TGF-β1-induced EMT in NMuMG cells (Fig. 8A,B). Furthermore, introduction of a scratch wound into a monolayer of NMuMG cells triggers a rapid increase in Hic-5 expression and colocalization with induced tyrosine phosphorylated paxillin in focal adhesions at the leading edge (D.A.T. and C.A.T., unpublished observations). Finally, this report demonstrates, for the first time, a proactive role for Hic-5 in stimulating cell migration. Whether this occurs via a parallel or alternative signaling pathway with respect to full-length paxillin remains to be determined. Together, these results suggest a role for paxillin δ as the primary physiologic competitor to paxillin-mediated integrin signaling in NMuMG cells, with Hic-5 performing a more complementary role to full-length paxillin. Interestingly, a role for Hic-5 as a specific transcriptional coactivator of the Fos gene has also been described (Kim-Kaneyama et al., 2002). Induction or overexpression of Fos can promote events such as tumorigenesis, metastasis and EMT (Hay, 1995; Reichmann et al., 1992) in part through LEF-1/β-catenin (Kim et al., 2002). That observation, combined with our demonstration that Hic-5 is upregulated during the process of TGF-β1-induced EMT provides a potential functional context for Hic-5 regulation during EMT.

The more global role of paxillin δ as a functional antagonist to full-length paxillin, the possible interrelationship with Hic-5 and the role for these proteins in EMT that occurs during processes such as organism and tissue development, wound repair and tumor metastasis awaits further investigation. The presence of a new paxillin isoform independently translated from the primary paxillin transcript increases the complexity and reinforces the importance of the paxillin adaptor protein family in organizing signaling events originating from focal adhesions during cell migration and adhesion.

The authors thank Brad Pawlikowski for helpful assistance with the retroviral expression system. This work is supported by NIH GM 47607 to C.E.T. and an American Heart Association predoctoral fellowship to D.A.T.