A RIAM/lamellipodin–talin–integrin complex forms the tip of sticky fingers that guide cell migration

The leading edge of migrating cells contains rapidly translocating activated integrins associated with growing actin filaments that form ‘sticky fingers' to sense extracellular matrix and guide cell migration. Here we utilized indirect bimolecular fluorescence complementation to visualize a molecular complex containing a Mig-10/RIAM/lamellipodin (MRL) protein (Rap1-GTP-interacting adaptor molecule (RIAM) or lamellipodin), talin and activated integrins in living cells. This complex localizes at the tips of growing actin filaments in lamellipodial and filopodial protrusions, thus corresponding to the tips of the ‘sticky fingers.' Formation of the complex requires talin to form a bridge between the MRL protein and the integrins. Moreover, disruption of the MRL protein–integrin–talin (MIT) complex markedly impairs cell protrusion. These data reveal the molecular basis of the formation of ‘sticky fingers' at the leading edge of migrating cells and show that an MIT complex drives these protrusions.

C ell migration is crucial to diverse processes such as embryonic development, tissue repair, axon extension/ path finding and cancer metastasis. Cells migrating in a mesenchymal mode form actin-driven protrusions, such as lamellipodia and filopodia, at the leading edge 1,2 . These protrusions help to maintain cell polarity and the directional persistence of cell migration [3][4][5][6] Integrins are sensors of the chemical and physical nature of the extracellular matrix and cells can dynamically increase the affinity of integrins for their ligands, which is operationally defined as integrin activation 7 . Activated integrins are enriched at the leading edge of migrating cells 8 where they help to direct migration 9 . A potential connection between integrin activation and actin dynamics in directional migration was revealed by studies showing that activated integrins are associated with polymerizing actin filaments and move transversely in the lamellipodium and along filopodia 10 . Galbraith et al. 10 termed these structures as 'sticky fingers' and proposed that they played a role in directional cell migration and path finding however, the molecular basis of these structures and their functional role remained unclear.
The binding of talins to the b subunit cytoplasmic tail is a final common step in b 1 , b 2 and b 3 integrin activation in vitro 11 and in vivo 7 . In purified systems, talin binding is sufficient to activate integrins 12 . Once activated, integrins engage immobilized ligands, such as components of the extracellular matrix, to form adhesions, a constellation of integrin-driven assemblies of structural and signalling components that dictate many cellular behaviours 13 . The entire spectrum of b 1 , b 2 and b 3 integrin-based adhesions contains talin 14 and talin is required for adhesion formation 13 .
Ras GTPases 15 , in particular, Ras subfamily members, Rap1a and Rap1b, stimulate integrin activation 16 and several Rap1 effectors have been implicated in integrin activation 15,17,18 . Rap1-GTP-interacting adaptor molecule (RIAM) is a Rap1 effector that is a member of the Mig-10/RIAM/lamellipodin (MRL) family of adaptor proteins 17,19,20 that mediates the invasion and migration of melanoma cells. RIAM contains Ras association and pleckstrin homology domains and proline-rich regions, which are defining features of the MRL protein family. Lamellipodin (Lpd) is a RIAM paralogue that is also present in many cells 20 and plays an important role in direction finding during cell migration in vivo and in vitro 21 . MIG-10, the Caenorhabditis elegans orthologue, is known to be involved in axonal path finding 19 . Previously, we exploited the fact that agonists fail to activate recombinant a IIb b 3 expressed in CHO (chinese hamster ovary) cells to develop a synthetic reconstruction of an integrin activation pathway and used it in combination with forward and reverse genetics to dissect the pathway 22 . We found that Rap1-induced formation of a complex containing the Rap1 effector, RIAM and talin, which results in talin recruitment to the plasma membrane and to integrin a IIb b 3 ; this complex represents an early modular component of integrinbased adhesions formed by one of several mechanisms that drive the integrin-talin interaction 23 . Mapping studies identified short amphipathic helices in RIAM or Lpd that bind talin; joining those helical peptides to the membrane targeting sequences of Rap1 led to a minimized Rap-MRL module that was sufficient to recruit talin to activate integrins 24 . Thus, MRL proteins function as a scaffold that in effect connects the membrane targeting sequences in Ras GTPases to talin, thereby recruiting talin to the plasma membrane and activating integrins.
Because RIAM or Lpd can drive the formation of an integrintalin complex containing activated integrins 24 and because both paralogues are enriched at the protruding leading edge of migrating cells 17,20 , we hypothesized that an MRL protein-integrin-talin (MIT) complex forms the tip of the 'sticky fingers' in migrating cells. Here we develop and validate an indirect bimolecular fluorescence complementation (BiFC) method to visualize the MIT complex in living cells. We find that the MIT complex is enriched at the tips of growing actin filaments in lamellipodial and filopodial protrusions corresponding to the tips of 'sticky fingers'. Formation of the complex requires that talin bridges integrin and MRL proteins. Moreover, disruption of the MIT complex using a RIAM mutant defective in talin binding results in impaired cell protrusion. Thus, we reveal the molecular basis of formation of 'sticky fingers ' at the leading edge of migrating cells.

Results
An MIT complex is at the tips of sticky fingers. Active integrins can localize to the tips of actin-based protrusions to form 'sticky fingers' that probe the extracellular matrix 10 . Activated Rap1 causes RIAM to bind to talin, resulting in the association of talin with integrins and consequent integrin activation [24][25][26] . We used BiFC 27,28 to specifically visualize the complex of RIAM with talin bound to integrins (Fig. 1a). The N-terminal b-sheet moiety of the Venus (VN) was joined to the N-terminus of RIAM and the C-terminal b-sheet moiety of the Venus (VC) was attached to the cytoplasmic tail of the integrin a-subunit. We reasoned that, since the interaction of RIAM with the integrin requires an endogenous talin bridge 23,24 , the presence of BiFC would reveal the complex of RIAM/talin/integrin (Fig. 1a). U2-OS cells expressing VN-RIAM and a IIb -VCb 3 in combination with Lifeact (to visualize filamentous actin 29 in live cells) exhibited BiFC at the tips of growing actin filaments within lamellipodia and filopodia ( Fig. 1b; Supplementary Movies 1 and 2). To exclude potential path length effects in this striking enrichment of BiFC, we examined its distribution in comparison with markers for bulk cytoplasm (blue fluorescent protein (BFP)) or membrane (BFP-Ras-CAAX). Line scans along the filopodia from the base to their tips demonstrated that the BiFC signal was selectively enriched at the tip ( Supplementary Fig. 1a,b). Similarly, when we performed this experiment in NIH-3T3 (Supplementary Movie 3) or Ptk1 (Supplementary Movies 4 and 5) cells which form large lamellipodia, we observed localization of BiFC at the edge of protruding lamellipodia, particularly at the tips of actin filaments traversing the lamellipodia. Thus the RIAM/integrin/talin complex is present at the tips of growing actin filaments that either traverse lamellipodia or drive filopodial extension.
We also observed co-localization of BiFC with Myosin-X, an established marker of filopodia 30 at the tips of filopodia ( Supplementary Fig. 2a). Furthermore, we compared the BiFC signal versus the overall abundance of RIAM and b 3 integrin. We observed that BiFC was prominent at the plasma membrane in protrusions, but was neither associated with the bulk of cytoplasmic RIAM nor with intracellular vesicles containing integrin b 3 , indicating that BiFC represented only a fraction of total cellular RIAM and integrin ( Fig. 1c and d; Supplementary  Fig. 1c and d).
BiFC depends on the interaction of talin with the integrin. As noted above (Fig. 1a), BiFC between RIAM and the integrin should require a talin bridge. In support of this idea, we observed that talin was usually present when BiFC was present in protrusive structures ( Fig. 2a; Supplementary Fig. 3a; Supplementary Movie 6), although the vast majority of talin was either cytoplasmic or present in focal adhesions. As noted above, Myosin-X, an authentic marker of filopodial tips, was co-localized with BiFC. Similarly, in cells co-transfected with enhanced green fluorescent protein (EGFP)-Myosin-X and mCherry-talin, there was clear localization of a minor fraction of talin in the Myosin-X-tagged filopodial tips (Fig. 2b) in addition to the expected prominent focal adhesion localization. Importantly, talin is prominent in vinculin-containing mature focal adhesions 14 but BiFC was much reduced in these more mature focal adhesions ( Supplementary  Fig. 3b,c; Supplementary Movie 7), consistent with the previous finding that vinculin inhibits RIAM binding to talin 23,31 , thereby displacing RIAM from adhesions. Conversely, even though vinculin can induce talin association with and activation of integrins 23   To determine when RIAM and talin reside in a common complex, we used cross-variance analysis 32,33 . This method detects the presence of molecular complexes using the covariance of fluorescence fluctuations arising from dynamic exchange of molecules in adhesions. In CHOK1 cells, talin-mGFP and RIAM-mCherry localize in nascent adhesions as the cell edge protrudes (Fig. 2c). We observed positive cross-variance between talin and RIAM as the adhesion starts to assemble and persists even after the adhesion forms (Fig. 2d). Thus, RIAM and talin reside in a common molecular complex as nascent adhesion form 32,33 .
To directly test whether BiFC required talin-integrin interaction, we introduced the b 3 (Y747A) mutation that inhibits the binding of talin, filamin and other cytoplasmic proteins or the b 3 (L746A) mutation that selectively reduces integrin interactions only with talin 11 (Fig. 3a). In cells expressing either a IIb -VCb 3 (L746A) or a IIb -VCb 3 (Y747A) there was a dramatic reduction in the BiFC signal as assayed by quantitative flow cytometry (Fig. 3b) and by spinning disk confocal microscopy ( Fig. 3c). Thus, we conclude that BiFC reports the presence of an MIT complex.
The MIT complex is regulated by Rap1. We exploited the observation that the activity of Rap1 (refs 24,25) regulates the co-precipitation of RIAM with talin and integrins to test whether BiFC reported the physiological regulation of the MIT complex in living cells. Suppression of endogenous Rap1 activity by expression of Rap1GAP1 markedly reduced the BiFC signal as assayed by quantitative flow cytometry (Fig. 4a,b) directly demonstrating Rap1 regulation of the RIAM-MIT complex. Conversely, expression of activated Rap1(G12V) induced a dramatic increase (Fig. 4a,b) in BiFC. In sharp contrast, activated H-Ras(G12V) suppressed the BiFC signal (Fig. 4a,b), suggesting that the previously described H-Ras inhibition of integrin activation 34 is ascribable to disruption of an MIT complex.  In sum, these data show that, the observed BiFC signal can report the physiological regulation of formation of the MIT complex in living cells.
The MIT complex co-localizes with activated integrins. We stained non-permeabilized cells with an anti-a IIb b 3 antibody to test whether the MIT complex was associated with cell surface integrins, resulting in staining of most of the BiFC-positive structures in the plasma membrane ( Supplementary Fig. 2b,c). Furthermore, the BiFC-positive surface structures contained activated integrins, as they stained with PAC1, an activationspecific anti-a IIb b 3 (ref. 35; Fig. 4c). However, the PAC1 staining was less apparent on the tips of filopodia suggesting that at this location some of the integrins were occupied by ligands that would displace the PAC1. Indeed, the BiFC-positive tips of filopodia were brightly stained with anti-LIBS1, an anti-a IIb b 3 that recognizes the activated and ligand-bound integrin 36,37 ( Supplementary Fig. 5). As shown above ( Fig. 1d; Supplementary Fig. 1d), BiFC reports only a subset of total integrins ( Supplementary Fig. 5). Taken together, these data indicate that BiFC reports the presence of an MIT complex that co-localizes with cell surface-activated ligand-bound and unoccupied integrins.
BiFC identifies diverse MIT complexes in multiple cell types.
Our initial analysis was conducted with the complex formed with RIAM and a IIb b 3 . We asked whether this approach could be used to visualize complexes of RIAM with other integrins, such as a   Fig. 6a) and was also true with a 5 b 1 -RIAM ( Supplementary Fig. 6b). Therefore, MIT complex formation and localization is similar in multiple cell and integrin types. We also found that the talin-binding RIAM paralogue, Lpd 19,24,38 , formed a complex with a 5 b 1 ( Fig. 5b; Supplementary Movie 9) and a IIb b 3 ( Fig. 5c; Supplementary Movie 10) at the tips of actin-based protrusions. In summary, either mammalian MRL protein can form a complex with talin and activated integrins localized at the tips of 'sticky fingers' and they can do so in multiple cell types.
Assembly and localization of the MIT complex without BiFC.
The foregoing studies established that BiFC could be used to observe the localization, dynamics and regulation of assembly of the MIT complex in living cells. The interaction of VN and VC domains, required to form BiFC, can increase the stability of the observed complex 28 . We performed a proximity ligation assay to visualize the localization of the MIT complex in the absence of BIFC. To perform this assay, we used a rabbit anti-Lpd antibody to stain endogenous Lpd in U2-OS cells and stained Flag-tagged recombinant integrin a5 with a murine anti-Flag antibody (Fig. 6b). We observed localization of the proximity ligation signal to the ends of actin filaments in protruding areas at the cell edge and at the tips of filopodia (Fig. 6a). Consistent with previous studies reporting that this method only detects a subset of protein complexes 39 , only a fraction of filopodia exhibited a signal at their tips. We also used tandem affinity purification to biochemically assess the formation of a RIAM/integrin/talin complex in the absence of BiFC. Sequential affinity chromatography under non-denaturing conditions isolated Flag-tagged RIAM followed by streptavidin affinity chromatography isolated integrin a IIb -SBPb 3 in a complex containing talin. The formation of this complex required the talin-integrin interaction because none of these components were identified when the tandem affinity purification was performed from cells bearing talin-binding defective a IIb -SBPb 3 (Y747A) mutant (Fig. 6c). Finally, when we examined the effect of BiFC on cellular behaviour, we observed BiFC-positive cells exhibited similar numbers and lengths of finger-like protrusions to cells lacking BiFC ( Supplementary Fig. 7). In summary, the presence of BiFC, which signals the formation of the MIT complex did not alter protrusion formation and, in the absence of BiFC, the MIT forms is localized to the ends of actin filaments in cell protrusions.
An MIT complex forms without ligand binding. One of the hallmarks of 'sticky fingers' is that they contain activated unoccupied integrins. As noted above, BiFC that reported formation of the a IIb b 3 MIT complex co-localized with PAC1 (Fig. 4c), a ligand-mimetic antibody, suggesting that the MIT complex forms without ligand binding. To directly test this idea, we used b 3 (D119A), a ligand binding-defective mutant 40 . Even though cells expressing a IIb -VC b 3 (D119A) did not spread on fibrinogen, they still formed actin-containing microspikes that exhibited bright BiFC at their tips ( Fig. 7a; Supplementary Movie 11). Alternatively, we examined cells expressing a IIb -VCb 3 plated on non-adhesive bovine serum albumin (BSA). Cells on fibrinogen spread well (Fig. 7b), but cells on BSA did not. Nevertheless, BiFC was still evident at the tips of actin-based protrusions in cells (Fig. 7c) plated on BSA. Thus, the MIT complex forms without ligand engagement to generate activated but unoccupied integrins at the tips of actin-based protrusions.  (Fig. 8a). Blocking the RIAM-talin interaction led to a dramatic reduction in RIAM recruitment to paxillin-staining adhesions (Fig. 8b). As previously reported 23 , silencing RIAM markedly inhibited lamellipodial protrusion; rescue with wildtype RIAM restored protrusion whereas the RIAM(4E) mutant had no effect ( Fig. 8c; Supplementary Movie 12). Therefore, RIAM in the MIT complex supports cellular protrusion.

Discussion
Rapidly translocating activated integrins associated with growing actin filaments are proposed to guide cell migration and have been termed 10 'sticky fingers' at the leading edge. Here we have characterized the molecular basis of these sticky fingers using an indirect BiFC approach to visualize an MIT complex of MRL proteins with talin and activated integrins in lamellipodia and filopodia. The MIT complex is enriched at the tip of growing actin filaments in protrusions. Formation of this complex required the capacities of integrins to bind talin and the complex was regulated by the activity of a Ras family GTPase. Moreover, disruption of the MIT complex using a RIAM mutant defective in talin binding led to a severe impairment in cell protrusion. In summary, these data reveal the molecular basis of formation of 'sticky fingers' at the leading edge and show that the MIT complex is important in forming these protrusions that guide cell migration.
Here we modified BiFC, a well-established method to study direct protein-protein interactions 41 , to identify the ternary MIT complex and to study its localization and regulation. We validated the indirect nature of BiFC between the MRL protein and the integrin by showing that it was critically dependent on the capacity of each of them to bind to a talin bridge. The desire to image a multi-molecular complex led us to employ BiFC rather than Förster Resonance Energy Transfer, whose signal decays exponentially above a distance of 10 nm between the interacting proteins 42 , because the talin head domain itself has a diameter exceeding 10 nm (ref. 43). In contrast, BiFC can work over longer distances and is less reliant on the orientation of the interactors. Although the interaction of VN and VC domains can prolong complex stability 28 , the talin-dependence of BiFC reported here establishes its specificity. Moreover, we ensured that the BiFC itself does not enforce the formation and the localization of the MIT complex. First, we used a split tandem affinity approach to biochemically demonstrate a RIAM/integrin/talin complex without the BiFC moieties. Second, we used a proximity ligation assay in fixed cells to demonstrate the formation of the MIT complex between a 5 b 1 integrin and endogenous lamellipodin, and showed that it is enriched at the cell edge and the tips of actin bundles in protrusions. In addition, since BiFC required expression of recombinant proteins, we took care to analyse cells whose morphology and morphodynamics were not divergent from control untransfected cells to exclude overexpression artifacts as before 44 . Thus indirect BiFC can be used to visualize higher order protein-protein interactions and we have here established that it provides a means to identify the MIT complex in living cells.
The MIT complex was largely co-localized with a subset of cell surface-activated integrins, as most of the integrins were accessible to antibodies in intact cells. Importantly, the fact that some of the BiFC-associated integrins bound PAC1 shows that they were activated and unoccupied, since PAC1 is a true ligandmimetic antibody, that is, it will not bind to a ligand-occupied integrin 35 . Furthermore, the MIT complex formed with ligand binding-defective b 3 (D119A), substantiating that it can form with unoccupied integrins and suggesting that the MIT complex is the molecular identity of the 'adhesion precursor complex' proposed by Choi et al. 45 and forms the talin in nodes at the edge of the lamellipodium described by DePasquale and Izzard 46 . Similarly, ligand binding does not disassemble the MIT complex, since BiFC co-localized with an occupancy-dependent antibody, anti-LIBS1 (ref. 36), in filopodial tips. Importantly, we observed bound PAC1 not associated with BiFC, suggesting that either these activated integrins were generated by endogenous MRL proteins or by MRL protein-independent mechanisms of activation, for example, vinculin 23,31 . Furthermore, the absence of detectable vinculin from the tips of protrusions highlights the value of selective visualization of integrins activated by a specific mechanism. Previous methods for detecting activated integrins utilized either binding of natural ligands or their fragments or conformation-specific antibodies 47 and could not differentiate between different activation pathways. Since BiFC enables visualization of only a subset of integrins activated by a specific mechanism, it can form the basis of convenient screens for modulators of specific integrin activation pathways. Finally, we note the surprising absence of substantial vesicular BiFC, in spite of the extensive integrin trafficking to and from the plasma membrane 48 . Because BiFC maturation is time dependent 49 we cannot exclude the possibility of intracellular sites of assembly of the MIT complex. In sum, the detected MIT complex is largely at the plasma membrane, can contain both occupied and unoccupied activated integrins, and represents a molecularly defined 'modular' subset of such integrins. Association of activated translocating integrins with the tips of growing actin filaments are a hallmark of 'sticky fingers' 10  Arp2/3 complex to proline-rich motifs 17,21,25 . Thus, the MRL protein N terminus can recruit and stabilize talin at the membrane in a Ras GTPase-dependent manner to activate integrins 22 and the MRL protein C terminus that can promote actin polymerization 17,21,25 to drive the rapid translocation of the integrin (Fig. 8d). Together, these two biochemical functions of the MRL proteins result in the formation of the 'sticky fingers' at the leading edge that direct protrusion during cell migration. Plasmids. BiFC Venus vector encoding for a IIb -VC was described previously 38 . The sequence encoding the streptavidin binding peptide (SBP) was amplified by PCR and then inserted in-frame into the 3 0 end of a IIb after removing VC in pcDNA3.1-a IIb -VC to form a IIb -SBP. To generate the a 5 -VC construct, a plasmid template encoding human integrin a 5 was amplified by PCR to place appropriate restriction sites, and the PCR product was inserted in-frame at the 5 0 end of VC in pcDNA3.1 (Life Technologies). The N-terminal half of Venus was amplified to place appropriate restriction sites, and the PCR product was inserted in-frame into the 5 0 end of RIAM after removing EGFP in pEGFP(C1)-RIAM to form VN-RIAM. The sequence encoding for the 3xFlag tag (DYKDHDGDYKDHDI DYKDDDDK) was amplified to place appropriate restriction sites, and the PCR product was inserted in-frame into the 3 0 end of RIAM to generate the VN-RIAM-Flag construct. To generate the VN-Lpd construct, the sequence encoding for human Lpd was amplified to place appropriate restriction sites, and the PCR product was inserted in-frame into the 3 0 end of VN after removing RIAM in the vector encoding for VN-RIAM. VN-RIAM was subcloned into the pTRE2-hyg expression vector (Clontech) by standard techniques. Integrin b 1 , b 3 , b 3 (L746A), b 3 (Y747A), b 3 (D119A) and b 3 (D119A, Y747A) were constructed in pcDNA3.1 by PCR-based cloning and site-directed mutagenesis. cDNAs encoding b 3 , b 3 (L746A) and b 3 (Y747A) were then amplified by PCR and subcloned into PRRL-CMV 44 . RIAM, Rap1A(V12), Rap1GAP1, H-Ras(V12) and ubiquitin were amplified by PCR and subcloned into p3xFLAG-CMV-7.1 (Sigma-Aldrich). K-Ras-CAAX was amplified by PCR and ligated into both pTagBFP-N (Evrogen, Moscow, Russia) and PRRL-CMV-mCherry. A cDNA encoding shRNA-resistant RIAM was generated by PCR and cloned into pTagBFP-N. Talin-binding defective RIAM(4E) 24 mutant (M11E, F12E, L15E and L16E) was generated by PCR-based cloning and in situ mutagenesis using BFP-RIAM(WT) 23 as template. The mutations were verified by DNA sequencing. Plasmids encoding shRNA for mouse RIAM and the human U6 promoter 25 were amplified by PCR to place Xba1 and Xho1 sites, and PCR products were cloned into lentiviral vector FG12. Promoter and shRNA sequences were confirmed by DNA sequencing. Control shRNA was obtained from a scrambled sequence of mRNA derived from Rock1 (ref. 11). RIAM, vinculin and talin were fused to mCherry and inserted into pcDNA3.1 by PCR-based cloning. Human full-length talin was fused to a HA-tag at the N terminus 51 . Integrin b 3 was amplified by PCR and then inserted at the 5 0 end of mCherry into pcDNA3.1. mCherry-LifeAct was a gift from O Pertz (University of Basel, Switzerland). EGFP-Myosin-X construct was a gift from R Cheney, University of North Carolina. Myosin-X cDNA was fused to mCherry and inserted into pcDNA3.1 by PCR-based cloning.
U2-OS and NIH-3T3 cells were transiently transfected with FuGene HD (Promega) or TransIT-LT1 (Mirus) according to the manufacturer's protocol. Transient transfection of MEF, PtK1 and HuVEC were carried out using Amaxa nucleoporator (Lonza). Cells were washed in PBS and resuspended in 100 ml of PBS containing 3-5 mg of plasmid constructs and electroporated using programs A034 (HuVEC), A023 (MEF) or T020 (PtK). The cells were allowed to spread for 1-2 h before they were washed with fresh medium and allowed to grow overnight.
FG12 lentiviral constructs encoding RIAM shRNA and control have been previously described 22 . VSV-G pseudotyped lentiviral vectors were produced by transient transfection of HEK293T (human embryonic kidney) cells. Viral particles from cell culture supernatant were used for infection of U2-OS or NIH-3T3 cells overnight. The cells were assayed at least 4 days after transduction.
To generate U2-OS inducible cell lines, a total of 1 Â 10 6 U2-OS Tet-On cells (Clontech) expressing the tetracycline-controlled transactivator were plated into a 100 mm dish and transfected with pTRE2-hyg-VN-RIAM using TransIT-LT1 (Mirus). Forty-eight hours after transfection, 500 mg ml À 1 of Hygromycin B (Life Technologies) was added. Transformants were selected and single cells cloned using the limiting dilution method. To test for induced expression, 200 ng ml À 1 doxycycline (Sigma-Aldrich) was added to the medium and VN-RIAM-positive clones were selected by flow cytometry analysis using a polyclonal anti-EGFP antibody (cross-reacts with VN). Selected clones were transduced with lentiviral particles encoding a IIb -VCb 3 (WT), a IIb -VCb 3 (L746A) or a IIb -VCb 3 (Y747A) and assayed as a polyclonal population. BiFC-positive cells expressing a IIb -VCb 3 (WT) were single cell cloned by automated cell deposition unit using a BD FACSAria after induction of VN-RIAM expression with 200 ng ml À 1 doxycycline overnight.
Flow cytometry. Cells were harvested by trypsinization, washed with PBS and then fixed in 2% paraformaldehyde for 10 min. After washings, cells were transferred into permeabilization buffer (PBS supplemented with 0.5% BSA and 0.5% (wt:vol) saponin) for 20 min. Cells were centrifuged, resuspended in permeabilization buffer containing primary antibodies (anti-Flag or D57 antibodies; 1:200) on ice for 30 min, washed three times in permeabilization buffer, and incubated in permeabilization buffer containing goat anti-mouse antibody coupled to Alexa Fluor 647 (1:300) for 30 min on ice, washed in permeabilization buffer, and then resuspended in PBS to get analysed on a BD FACSCalibur (BD Biosciences, San Diego, CA, USA) using CellQuest software (BD Biosciences). Results were expressed as the average mean fluorescence intensity corresponding to BiFC within the Flag-or D57-positive cell population.
Immunofluorescence microscopy. Glass coverslips were coated with 10 mg ml À 1 fibronectin or fibrinogen at room temperature for 1 h. Cells were adhered to the indicated substrate for the indicated period of time at 37°C. The cells were fixed with 2% formaldehyde, in 100 mM PIPES, 1 mM EGTA, 1 mM MgSO 4 for 30 min at 37°C. When permeabilization is required, cells were incubated with 0.2% TritonX-100 in PBS for 10 min. Following blocking with 10% goat serum in PBS for 10 min, cover slips were incubated with the primary antibody (1:200) for 2-3 h. The cells were washed three times with PBS, incubated with the secondary antibody for 1 h, washed and mounted with Fluorsave mounting reagent (Calbiochem, La Jolla, CA, USA). For PAC1 staining, cells were incubated with serum-free medium for 20 min and PAC1 antibody was added (1:100 dilution) for 10 min before fixing the cells with 4% paraformaldehyde in PBS for 10 min. The cells were then incubated with the secondary antibody (1:300) for 20 min before being washed with PBS and mounted onto a glass slide with Fluorsave mounting reagent.
The slides were imaged with an IX81 inverted microscope (Olympus, Tokyo, Japan) equipped with a CSU-X1-A1 spinning-disk head (Yokogawa, Tokyo, Japan), electron-multiplying CCD 14 bit 1K Â 1K camera (Hamamatsu, Hamamatsu, Japan), four laser lines (405, 488, 560 and 640 nm), and software controlled by Volocity (PerkinElmer, Waltham, MA, USA). When multiple conditions were imaged as part of the same experiment, the laser power, camera sensitivity and camera exposure times were kept constant. Post-acquisition processing was limited to adjusting contrast or correcting registration by Volocity (PerkinElmer, Waltham, MA, USA). Any modifications were kept consistent when multiple conditions were processed as part of the same experiment.
Line scans were performed with ImageJ (NIH). Briefly, a line was drawn along a filopodium from its tip to a few mm beyond the cell membrane. The intensity values were extracted from ImageJ and input into Microsoft Excel to plot as a multi-colour curve.
Spinning disc confocal microscopy. For live cell imaging, the cells were plated on fibrinogen or fibronectin-coated glass cover slips for the desired time. The cover slips were transferred to a Chamlide magnetic chamber (CMB for 25 mm round coverslips; Live cell instruments, Seoul, Korea) and immersed in culture medium. The chamber was incubated with 5% CO 2 at 37°C for live imaging. The cells were imaged with the IX81 inverted microscope as described for immunofluorescence microscopy. Typically the cells were imaged with the Â 60 objective (numerical aperture (NA) 1.42) for 1-3 min at 5 s intervals. Post-acquisition processing was limited to adjusting contrast, image cropping or correcting registration by Volocity (PerkinElmer, Waltham, MA, USA). Modifications were applied equally to all conditions in a given experiment.
Total internal reflection fluorescence microscopy. Cells were plated on cover slips as described above and incubated at 37°C in 5% CO 2 and visualized with the Applied Precision OMX DeltaVision Ring TIRF module (Issaquah, WA, USA) using a Â 100 objective (NA 1.40) with four laser lines (405, 488, 562 and 641 nm). The cells were imaged for 1-3 min at 5 s intervals. After acquisition, the data sets were transferred to a Linux workstation employing SoftWoRx software (Applied Precision) that uses an algorithm experimentally produced on the system from the convolution of a point spread function. All data sets were subjected to eight deconvolution iterations to reduce extraneous light or scattered light captured by the camera. Subsequent image processing and analysis was done with Volocity.
Fluorescence fluctuation imaging and cross-variance analysis. Cross-variance analysis was performed on TIRF image time series acquired using an Olympus IX71 microscope equipped with a 100 Â 1.45 NA Plan Apo total internal reflection fluorescence microscopy (TIRFM) oil objective, a Ludl controller (Ludl Electronic Products, Hawthorne, NY, USA), and Metamorph Software (Molecular Devices, Downingtown, PA, USA). mGFP and mCherry excitation was produced using the 488 nm and 568 nm lines of an Ar-Kr ion laser (Melles Griot, Albuguergue, NM, USA), respectively. A polychroic mirror (Z488/568 rpc), dual emission filter (Z488/568 nm; Chroma Technology, Bellows Falls, VT, USA), and a Dual-View (Photometrics, Tucson, AZ, USA) were used for simultaneous dual-colour image acquisition. Images were collected on a QuantEM: 512C EMCCD camera (Photometrics) in non-overlapping mode at 3X conversion gain and 700-1,000 EM gain, Image exposure time and delay time between images were 100 and 500 ms, respectively.
To detect molecular interaction using fluorescence fluctuation cross-variance analysis, we calculate the cross-variance (B cc ) parameter at a given pixel location x i ; y j À Á for given time segments as: B cc x i ; y j À Á ¼ s x i ; y j À Á 2 cc ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi I 1 x i ; y j À Á t I 2 x i ; y j À Á t q ð1Þ I 1 x i ; y j À Á t and I 2 x i ; y j À Á t are the fluorescence average intensity for channels 1 and 2. s x i ; y j À Á 2 cc is the fluorescence intensity cross-variance between channels 1 and 2 for time, t, and is defined as: s x i ; y j À Á 2 cc ¼ 1 t : X t I 1 x i ; y j ; t À Á À I 1 x i ; y j À Á t : I 2 x i ; y j ; t À Á À I 2 x i ; y j À Á t ð2Þ I 1 x i ; y j ; t À Á and I 2 x i ; y j ; t À Á are the fluorescence pixel intensity at time, t; for channels 1 and 2, respectively.
If molecules represented by fluorescence intensity in two channels (wavelengths), channel 1 and 2, are non-interacting, their fluorescence fluctuations will be independent and the cross-variance will be centred around zero. In contrast, interacting species will produce positive (B cc ) values. As the negative control, GAP-mGFP and GAP-mCherry, which localize independently to the cell membrane, are not known to interact and therefore do not display positive cross-variance values. As a positive control, mCherry-paxillin-mGFP; an adhesion marker tagged with two fluorescent probes and therefore produces correlated movement and positive cross-variance 33 .
Duolink in situ proximity ligation assay. U2-OS cells expressing either a 5 -3xFlag or a 5 -VC were plated on fibronectin for 2 h, fixed in 4% paraformaldehyde, permeabilized with 0.2% Triton X-100 and stained with rabbit anti-Lpd and mouse anti-Flag antibodies (1:200). Hybridization, ligation, amplification and detection were realized according to the manufacturer's instructions (Olink Bioscience). The cells were imaged with the IX81 inverted microscope with the Â 60 objective (NA 1.42) as described for Immunofluorescence Microscopy. Post-acquisition processing was limited to adjusting contrast, image cropping or correcting registration by Volocity (PerkinElmer, Waltham, MA, USA). Modifications were applied equally to all conditions in a given experiment.
Cell edge tracking. NIH-3T3 cells expressing either control shRNA or RIAM shRNA, each marked by EGFP expression, were transduced with lentiviral particles encoding mCherry-K-Ras-Caax and transiently transfected with vectors encoding BFP, BFP-RIAM(WT) or BFP-RIAM(4E). Cells were plated on fibrinogen-coated glass coverslips for 2 h and imaged with the IX81 inverted microscope, as described for live cell imaging, by an investigator blinded to the conditions. Typically the cells were imaged with the Â 60 objective (NA 1.42) for 3 min at 5 s intervals. Detection analysis for the cell edge at each time point was performed as described previously 23 , using a custom-built software package written in Matlab 52 .
Statistical analyses. Statistical analyses were performed using GraphPad Prism software. Statistical significance was determined by ANOVA with Bonferroni's post hoc test, Mann-Whitney Rank sum test, or Student's t-test. The test used for each experiment is indicated in the corresponding figure legend. All experiments have been reproduced at least three times.