Culturing cells on microfabricated ECM substrates allowed us to control cytoskeletal polarity and FA position, which enabled us to predict where lamellipodia were likely to form when cells were stimulated by a soluble motility factor. Using this system, we were able to detect uncoupling of membrane extension from spatial cues and to analyze the role of paxillin subdomains in this motile response. Normal mouse and human fibroblasts plated on single cell-sized, square FN islands formed large FAs primarily in corner regions and preferentially extended lamellipodia from adjacent sites in response to PDGF. In contrast, paxillin-deficient cells formed more and smaller FAs as well as lamellipodia along the cell periphery, with little spatial preference. These results indicate that paxillin is involved in both promoting membrane extension near FAs, as well as suppressing lamellipodia formation at distant sites.
In addition to showing that paxillin is critical for spatially coupling regions of cell distortion and sites of FA assembly to sites where new lamellipodia will form, we found that the N- and C-termini of paxillin play opposing, but complementary, roles in this process (). The N-terminus is critical for suppressing lamellipodia formation and maintaining directional persistence, while the C-terminus actively promotes lamellipodia formation. An unexpected finding was that paxillin mutation also affects the formation of dorsal CDRs, as well as lateral membrane extensions. Most importantly, these studies revealed that in addition to regulating directional migration in 2D, paxillin is a critical mediator of ECM invasion and migration in 3D, and this more complex response correlates with formation of CDRs in 2D cultures.
Phenotypes of paxillin mutants (MEFs).
Cells migrating on ECM substrates that vary in the mechanical compliance (flexibility) move in the direction in which they exert the highest traction forces 
. In square-shaped cells, traction forces are concentrated in corner regions 
, likely due to positive feedback between geometric constraints and contractility-dependent assembly of FAs 
. In addition to containing high concentrations of signaling molecules, FAs may be “permissive zones” for membrane extension in that actin-driven protrusions are not blocked by cortical actin (due to stress fiber insertions). In support of this hypothesis, myosin-mediated cortical tension has been shown to inhibit branching in endothelial cells, and inhibition of myosin II subjacent to the plasma membrane can induce localized membrane protrusion 
At early time-points after PDGF stimulation (5–10 min), cells with and without paxillin formed both dorsal and lateral membrane extensions. This suggests that paxillin is not required for the initial burst of actin-driven ruffling in response to growth factor stimulation, and that this early process may be molecularly distinct from later rounds of lamellipodia formation. We also found that pax −/−cells and cells expressing the paxN truncation mutant had a greater propensity to form CDRs on their apical membranes than pax+ or paxC cells. CDRs frequently form over the leading edges of motile cells and contain many of the same protein components as lamellipodia (e.g., actin, Arp2/3, vinculin, paxillin; Fig. S2B
, C and D); however, they have been shown to be structurally and biochemically distinct protrusive structures 
, which is consistent with our findings.
Although many paxillin domains have been studied, little is known about the conformation of paxillin in vivo
. Vinculin in FAs undergoes a conformational change that relieves an intramolecular association between the head and tail regions, exposing protein-protein interaction domains that are hidden in the cytosolic form 
. Like vinculin, paxillin may adopt different conformations upon FA recruitment that expose or sequester various protein-interaction sites, which could explain the complex effects of the truncation mutants on the formation of different protrusive structures. It is possible that the different effects of the paxN and paxC truncation mutants are due to exposure of binding domains that are usually only available in specific subcellular contexts (e.g. in FAs versus the cytosol). The paxN and paxC truncation mutants may thus act as “dominant negatives”, sequestering proteins away from other binding partners, or “dominant positives” that can interact with proteins that normally would be unavailable in a given subcellular context.
Paxillin binds the ArfGAPs Git1 and Pkl/Git2 via its N-terminal LD4 motif 
, and these proteins have been implicated in directional motility through both positive and negative mechanisms. Git-1 has been reported to either inhibit membrane extension 
or promote cell migration depending on its location within the cell 
, whereas Pkl appears to be involved in control of directional cell migration in fibroblasts 
. Localization of Pkl to FAs is regulated by tyrosine phosphorylation, and its dephosphorylation is mediated by PTP-PEST, which binds to paxillin via its C-terminal LIM domains 
. Thus, the two halves of paxillin may work together to efficiently drive directional migration by mediating complex cycles of these types of molecular associations.
These findings indicate that paxillin may play distinct roles in different subcellular contexts, such as regulating the formation of different kinds of motile processes (e.g., broad or fan-shaped lamellipodia, filopodia, lateral versus dorsal ruffles). These data also suggest cytoplasmic functions for paxillin in controlling CDR extension and membrane trafficking, as well as lamellipodia formation, which may correspond to different modes of migration in vivo 
. Interestingly, CDRs formed by mesenchymal cells in 2D have been compared to invasive protrusions or ‘invadopodia’ formed by epithelial cells 
. Cells in tissue culture have basal membranes that are in contact with ECM and free dorsal surfaces, whereas mesenchymal cells are usually embedded within 3D ECMs in tissues. We found that dorsal protrusion formation in MEFs correlated with the ability of these cells to invade 3D Matrigel plugs. Paxillin-mediated signaling may therefore be critical for determining whether a cell migrates along a planar (2D) basement membrane or through a 3D interstitial matrix, as occurs, for example, during epithelial-mesenchymal transitions in cancer metastasis. Moreover, switching between Rho- and Rac-mediated modes of migration (e.g. ameboid versus mesenchymal) is a common feature of 3D matrix invasion 
. Thus, paxillin may be involved in tailoring a cell's motile response to physical cues in different microenvironments. Importantly, mutation and misregulation of paxillin correlate with metastatic potential in some human breast and lung cancers 
, suggesting that it may be involved in regulating ECM invasion as well. In any case, the different effects of paxillin deficiency in 2D versus 3D migration underscore the importance of the physical microenvironment on cell behavior, and the central role that paxillin normally plays in this process.
In conclusion, detailed examination of fibroblast cells on patterned and unpatterned substrates revealed that they respond to PDGF with an initial round of membrane extension in all directions, followed by progressive spatial fine-tuning that is sensitive to physical cues. Thus, lamellipodia formation becomes preferentially localized to regions of greatest cell distortion (i.e., corners) after 15 min of PDGF stimulation in these artificially polarized cells. We found that paxillin can enhance or suppress membrane extension depending on its subcellular context. The presence of paxillin within FAs appears to spatially constrain where Rac is activated inside the cell, and thereby preferentially stimulates motile process formation to adjacent regions. Loss of paxillin results in deregulated spatial pruning of membrane extensions. The N-terminus appears to suppress lateral membrane extension, and the C-terminus enhances lamellipodia formation, but both halves are required for efficient directional migration in 2D. Furthermore, overexpression of the N- or C-terminus alone can tip the balance between “dorsal” and “lateral” motile process formation in response to PDGF. Taken together, these results shed light on the molecular mechanism by which cell motility is directed by its physical microenvironment, in addition to revealing new functions for paxillin in coordinating cell migration in both 2D and 3D that might be highly relevant for developmental control in vivo.