Structural basis of the target-binding mode of the G protein–coupled receptor kinase–interacting protein in the regulation of focal adhesion dynamics

Focal adhesions (FAs) are specialized sites where intracellular cytoskeleton elements connect to the extracellular matrix and thereby control cell motility. FA assembly depends on various scaffold proteins, including the G protein–coupled receptor kinase-interacting protein 1 (GIT1), paxillin, and liprin-α. Although liprin-α and paxillin are known to competitively interact with GIT1, the molecular basis governing these interactions remains elusive. To uncover the underlying mechanisms of how GIT1 is involved in FA assembly by alternatively binding to liprin-α and paxillin, here we solved the crystal structures of GIT1 in complex with liprin-α and paxillin at 1.8 and 2.6 Å resolutions, respectively. These structures revealed that the paxillin-binding domain (PBD) of GIT1 employs distinct binding modes to recognize a single α-helix of liprin-α and the LD4 motif of paxillin. Structure-based design of protein variants produced two binding-deficient GIT1 variants; specifically, these variants lost the ability to interact with liprin-α only or with both liprin-α and paxillin. Expressing the GIT1 variants in COS7 cells, we discovered that the two PBD-meditated interactions play different roles in either recruiting GIT1 to FA or facilitating FA assembly. Additionally, we demonstrate that, unlike for the known binding mode of the FAT domain to LD motifs, the PBD of GIT1 uses different surface patches to achieve high selectivity in LD motif recognition. In summary, our results have uncovered the mechanisms by which GIT1's PBD recognizes cognate paxillin and liprin-α structures, information we anticipate will be useful for future investigations of GIT1–protein interactions in cells.

domains (18) (Fig. 1A). By connecting integrin to F-actin and other FA-associated proteins, paxillin plays a critical role in assembly and disassembly of FAs (25). Paxillin was suggested to recruit GIT1 to FA by the binding of the LD4 motif of paxillin to the PBD of GIT1, as the deletion of the LD4 motif or the PBD resulted in the disruption of localization of GIT1 to FA (16,26,27).
Although the PBD is crucial for the subcellular localization and function of GIT1, the molecular mechanisms underlying the PBD-mediated interactions remain elusive. A recent study indicated that the bindings of liprin-␣ and paxillin to GIT1 are mutually exclusive (13). However, without detailed interaction information, it would be difficult to understand how these competitive interactions are coordinated and regulated in FA.
Here, we determined crystal structures of GIT1_PBD in complex with a single ␣-helix (SAH) of liprin-␣2 and the LD4 motif of paxillin. Although liprin-␣2_SAH and paxillin_LD4 associate with the same binding groove on GIT1_PBD, surprisingly, these two helical peptides adopt binding orientations reversed to each other. Cellular analysis using the designed GIT1 mutations indicated that the GIT1/paxillin interaction controls the GIT1's localization to FA and the GIT1/liprin-␣ interaction promotes FA assembly. The structure of the GIT1_PBD/paxillin_LD4 complex reveals a novel LD recognition mode that involves a new surface patch on GIT1_PBD in addition to the canonical LD binding surface found in the FAT domain of FAK.

Both liprin-␣ and paxillin bind with GIT1_PBD via a single ␣-helix
To understand the assembly mechanism of GIT1, liprin-␣, and paxillin, we investigated the binding of GIT1 to liprin-␣ and paxillin, which is mediated by the paxillin-binding domain of GIT1 (GIT1_PBD). To narrow down the minimal PBD-binding region in liprin-␣2, various fragments were expressed in Escherichia coli and purified with high quality (Fig. 1A). As indicated by an isothermal titration calorimetry (ITC)-based assay, a small fragment containing residues 642-671 in liprin-␣2 interacts with GIT1_PBD with a binding affinity of ϳ3 M (Fig. 1D).
Neither an extension at the N-terminal nor at the C-terminal part of the fragment could significantly increase the binding affinity (Fig. S1A), suggesting that the 30-residue fragment contains essential and sufficient elements for the binding of liprin-␣2 to GIT1_PBD. Although this fragment was predicted as a coiled-coil (14), the molecular weight of the fragment measured by multiangle static light scattering matches the theoretical molecular weight of its monomeric state ( Fig. 1C and Fig.  S1B). Combining the CD spectrum of the fragment (Fig. S1C), we concluded that the fragment folds as a single ␣-helix. There- The color-coding of the regions is applied throughout the entire paper except as otherwise indicated. B, sequence alignment of five LD motifs in paxillin and the SAH sequence in liprin-␣2 showing the consensus sequence in the LD core region. The unique FK sequence found in the LD4 motif is highlighted by a purple box. C, the molecular weight of liprin-␣2_SAH was measured using analytical gel filtration chromatography-coupled multi-angle static light scattering. The theoretical molecular mass of monomeric Trx-tagged liprin-␣2_SAH is 18.2 kDa. D and E, ITC-based measurement of the bindings of GIT1_PBD to liprin-␣2_SAH (D) and paxillin_LD4 (E), respectively. The thermodynamic parameters of the ITC titrations as shown in this and following figures are summarized in Table S1.

Overall structures of GIT1_PBD in complex with liprin-␣2_SAH and paxillin_LD4
As the bindings of liprin-␣ and paxillin to GIT1_PBD are mutually exclusive (13) and liprin-␣ contains a LD-like motif in the SAH region (Fig. 1B), do liprin-␣ and paxillin share a similar LD-binding model, interacting with GIT1? To address this question, we aimed to solve the complex structures of GIT1_PBD/liprin-␣2_SAH and GIT1_PBD/paxillin_LD4 by using X-ray crystallography. Gel filtration analysis showed that GIT1_PBD forms a stable complex with either liprin-␣2_SAH or paxillin_LD4 in solution (Fig. S2, A and B). Crystallization trails for the two complex samples yielded high-quality crystals. The crystal structures of the two complexes, GIT1_PBD/liprin-␣2_SAH and GIT1_PBD/paxillin_LD4, were successfully determined at 1.8 and 2.6 Å resolutions, respectively, by using the molecular replacement method (Table 1). Most residues of liprin-␣2_SAH (29 of the 30 total residues) and paxillin_LD4 (21 of the 24 total residues) in these two structures were clearly assigned (Fig. 2, A and B). As a FAT-like domain, GIT1_PBD adopts the four-helix-bundle conformation. The overall structures of GIT1_PBD in the two complexes are essentially same as its apo-form structure (overall root mean square deviations of 0.9 and 1.0 Å, respectively), indicating that GIT1_PBD does not undergo conformational change upon target binding. Consistent with our biochemical analysis, both liprin-␣2_SAH and paxillin_LD4 form single helices to interact with GIT1_PBD on the same surface composed of the first helix (H1) and the last helix (H4) (Fig. 2, C and D), well explaining the competitive binding of liprin-␣ and paxillin to GIT1_PBD (13).
Paxillin_LD4 interacts with GIT1_PBD using the similar LDbinding mode for FAT-like domains, in which the LD4 helix is parallel to the H1 helix (19, 28 -30) (Fig. 2D and Fig. S3). However, to our surprise, despite containing an LD-like motif, liprin-␣2_SAH binds to GIT1_PBD with the reverse orientation, in which the SAH helix is antiparallel to the H1 helix (Fig. 2C). Although the interacting orientations are different for the two PBD-binding peptides, both liprin-␣2_SAH and paxillin_LD4 use their hydrophobic sides in their amphipathic helices to tightly pack with the H1/H4 groove of GIT1_PBD through hydrophobic interactions, which are further strengthened by hydrogen bondings (Fig. 2, E and F). In support of our findings, previous structural studies reported that hydrophobic interactions involving the GIT1_PBD/paxillin_LD4 interaction are critical for complex formation (15,16). Additionally, Asp-267 in paxillin_LD4 forms a salt bridge with Lys-758 in GIT1_PBD (Fig. 2F), which is a characteristic interaction in the LD-binding mode (15,16,20). Sequence analysis further shows that the interface residues in GIT1, liprin-␣2, and paxillin are highly conserved across species (Fig. 2, G-I), suggesting that the two different GIT1-binding modes observed for liprin-␣2_SAH and paxillin_LD4 are likely to be shared by other members of the liprin-␣ family and HIC-5, respectively.

Structural comparison of the LD4 and SAH-binding modes of GIT1_PBD
As the bindings of liprin-␣2_SAH and paxillin_LD4 to GIT1_PBD are mainly mediated by hydrophobic interactions (Fig. 2, E and F), we carefully analyzed the interfaces by comparing the buried hydrophobic residues of the two bound peptides. In the LD4 peptide, Leu-266 LD4 , Leu-269 LD4 , and Leu- Table 1 Statistics of data collection and model refinement Numbers in parentheses represent the value for the highest-resolution shell.

Data collection
Data set GIT1_PBD/liprin-␣2_SAH GIT1_PBD/paxillin_LD4 Space group P2 1 where T is a test data set of about 5% of the total reflections randomly chosen and set aside prior to refinement.
Complex structures of GIT1/liprin-␣ and GIT1/paxillin interact with the hydrophobic groove formed by the H1 and H4 helices of GIT1_PBD, which is a general LD-binding feature identified in other FAT-like domains (Fig. S3). Interestingly, by aligning the two peptide-bound GIT1_PBD structures together, we found that the side chains of Ile-660 SAH , Leu-657 SAH , and Leu-653 SAH in the SAH peptide could be overlapped precisely with those of the three leucine residues in the LD4 peptide (Fig. 3A), despite the different helical orientations between the two peptides. Likewise, the reversed amino acid sequence of the SAH peptide can be aligned to the LD4 sequence with the hydrophobic residues matched (Fig. 3B). This suggests the central roles of these hydrophobic residues in the binding of liprin-␣2_SAH and paxillin_LD4 to GIT1_PBD.
In addition to the three overlapped residues, a few C-terminal residues of both of the two peptides significantly contribute to the intermolecular hydrophobic interactions. As shown in Fig. 3A, Ile-664 SAH and Ile-667 SAH in the SAH peptide interact with the upper part of the H1/H4 groove, formed by the N terminus of H1 and the C terminus of H4, whereas the bulky

Complex structures of GIT1/liprin-␣ and GIT1/paxillin
side chain of Phe-276 LD4 in the LD4 peptide occupies the H1/H4 groove in the lower part. In contrast, the corresponding positions of Ile-664 SAH and Ile-667 SAH in the LD4 peptide (Ala-262 LD4 and Ala-259 LD4 ) and of Phe-276 LD4 in the SAH peptide (Ala-650 SAH ) are all alanine residues (Fig. 3, A and B), of which small side chains barely touch the H1/H4 groove, thereby contributing little to the bindings of either paxillin_LD4 or liprin-␣2_SAH to GIT1_PBD. Thus, replacing Ala-650 SAH with a phenylalanine in the SAH peptide may artificially create an additional PBD-binding site by mimicking the binding of Phe-276 LD4 to GIT1_PBD. In line with this hypothesis, the A650F mutation of liprin-␣2_SAH largely enhances the binding (Fig.  S4). Based on the above analysis, we conclude that GIT1_PBD can specifically recognize liprin-␣2_SAH and paxillin_LD4 by combinatorial usage of different surface patches in the H1/H4 groove, in which the upper and lower parts of the groove provide the binding specificities for the SAH and LD4 peptides, respectively, whereas the middle part of the groove creates the common binding surface for both of the two peptides (Fig. 3A).
The binding mode differences between paxillin_LD4 and liprin-␣2_SAH allow us to design GIT1 mutations, A754Q and I655Q, for disruption of specific target interactions with GIT1_PBD. The A754Q mutation largely decreases the hydrophobicity of the overlapped binding surface and imposes steric hindrance for interactions, thereby disrupting the bindings of GIT1_PBD to both liprin-␣2_SAH and paxillin_LD4. By contrast, as Ile-655 locates on the SAH-binding surface on GIT1_PBD, the I655Q mutation presumably prevents GIT1_PBD from binding to liprin-␣2_SAH while it retains the ability of GIT1_PBD to bind with paxillin_LD4. Fully consistent with our design, neither liprin-␣2_SAH nor paxillin_LD4 shows detectable binding to the A754Q mutant, whereas the I655Q mutant interacts with paxillin_LD4 but not liprin-␣2_SAH, as indicated by ITC-based measurements (Fig. 3C) and analytical gel filtration analysis (Fig. S2, C-F).

Paxillin and liprin-␣ play distinct roles in GIT1-mediated FA dynamics
As the core component of FA, paxillin is indispensable for FA assembly (25, 31). Considering the localization of GIT1 and liprin-␣ to FA for controlling FA dynamics (23,24,32,33), the competitive bindings of liprin-␣ and paxillin to GIT1 raise a question of how GIT1 coordinates with liprin-␣ and paxillin to fulfill their functions in FA. To approach this question, the subcellular localization of GIT1 was analyzed by overexpressing the GFP-tagged GIT1 in COS7 cells. The GFP tag was fused to the C terminus of GIT1 to avoid a potential inhibitive effect of N-terminal fused GFP on GIT1's activity (11, 34). As reported previously (16), GIT1-GFP is localized to paxillin-positive FA, whereas GFP alone shows a diffused distribution in cells (Fig.  4A). Consistent with our structural and biochemical analysis, the I665Q mutant of GIT1 remains largely co-localization with paxillin in FA, suggesting that liprin-␣ is unlikely to play an important role in the FA localization of GIT1. Unlike the I665Q mutant, the A754Q mutant of GIT1, presumably abolishing its binding to paxillin, completely lost its localization to FA (Fig. 4A). These observations indicate that paxillin is required for recruiting GIT1 to FA via the PBD-mediated interaction.
Next, we measured the potential changes of FA in the GIT1overexpressed COS7 cells by quantification of areas for matured FAs (Fig. 4, B-D). Compared with cells transfected with GFP alone, cells transfected with GIT1-GFP have enlarged FA area, suggesting that the recruitment of GIT1 to FA by paxillin promotes FA assembly. Consistently, as the A754Q mutant of GIT1 fails to localize to FA, neither FA area nor FA number in cells expressing this mutant shows significant changes compared with cells expressing GFP alone. However, GIT1 carrying the liprin-␣ binding-deficient mutation, I665Q, although remaining localized to FA, shows severely decreased FA numbers in COS7 cells (Fig. 4C), indicating the critical roles of liprin-␣ in regulating FA dynamics. Because the averaged FA area was reduced in cells expressing the I665Q mutant (Fig. 4D), it is likely that the I665Q mutation impairs the GIT1-promoted FA assembly. Taken together, the above cellular data revealed the distinct roles of GIT1 in the two competitive interactions, in which GIT1 is recruited to FA by binding to paxillin and promotes FA assembly by binding to liprin-␣.
Considering the similar binding affinities of GIT1/liprin-␣ and GIT1/paxillin (Fig. 1, D and E), it would be interesting to know how these two competitive interactions are switched from each other. Protein phosphorylation is a common mechanism in regulation of protein-protein interactions (35,36). By analyzing the GIT1_PBD sequence, we found that Thr-765, located on the SAH-specific binding surface (Fig. 3A), is a potential phosphorylation site (Fig. 5A). It is likely that the phosphorylation of Thr-765 specifically blocks the binding of GIT1 to liprin-␣ by introducing a highly negatively charged phosphoryl group to the hydrophobic interface. Indeed, mutating Thr-765 to a negatively charged glutamate, mimicking its phosphorylated state, almost eliminates the GIT1_PBD/liprin-␣2_SAH interaction while maintaining the binding of GIT1_PBD to pax-illin_LD4 (Fig. 5, B and C). We further tested the effects of this phosphorylation-mimicking mutation on the FA dynamics. Consistent with our biochemical analysis, the Thr-765 mutation does not affect the localization of GIT1 to FA (Fig. 5D), but it significantly decreases the FA numbers and areas of matured FAs (Fig. 5, E-G), presumably due to the disruption of the GIT1/liprin-␣ interaction. Thus, the Thr-765 phosphorylation of GIT1 may act as a switch to regulate the two PBD-mediated interactions in vivo.

The unique LD motif recognition mechanism revealed by the GIT1_PBD/paxillin_LD4 complex structure
The typical binding of the helical LD motif to the FAT domains is mainly determined by the characteristic sequence of LDXLXXXL (where X denotes any residue of the 20 common amino acids) (Fig. 1B). Upon binding to FAT domains, the three leucine residues buried in the H1/H4 groove and the aspartate residue form a salt bridge with a highly conserved lysine in the FAT domains (20). Such interactions are also observed in the GIT1_PBD/paxillin_LD4 complex structure (Fig. 2F). However, unlike the FAT domains in FAK and PYK, which interact with the LD2 and LD4 motifs of paxillin in the same fashion (Fig. S3), GIT1_PBD shows a much higher bind-

Complex structures of GIT1/liprin-␣ and GIT1/paxillin
ing selectivity to LD4 than to LD2 (15) (Fig. S5). It has been largely unclear where the selectivity comes from, as the core regions of LD2 and LD4 are nearly identical.
To decode the LD recognition mechanism by GIT1_PBD, the GIT1_PBD/LD4 structure was overlapped with the FAK_FAT/ LD4 structure. Compared with the short helix of FAK_FATbound LD4, the GIT1_PBD bound LD4 forms a longer helix, in which the C-terminal flanking residues of the LD core are also involved in the helix formation and attach to the H1/H4 groove with a closer distance (Fig. 6A). The helix extension is mainly stabilized by the interaction of Phe-276 LD4 and Lys-277 LD4 with GIT1_PBD via hydrophobic interaction and hydrogen bonding (Fig. 2F). The involvement of the LD core flanking sequence in the binding was not found in other FAT/LD structures (Fig. S3). Therefore, we ask why FAK_FAT cannot interact with the LD core flanking sequence. Detailed analysis showed that the Phe-276 LD4 -packing site, formed by the C-terminal part of H1 and the N-terminal part of H4 in GIT1_PBD, is distorted in FAK_FAT by replacing the last turn of the H1 helix in GIT1_PBD with a short loop (Fig. 6B). Such a conformational distortion in FAK_FAT prevents Phe-276 LD4 from binding to FAK_FAT.
Consistent with our structural findings, the binding of GIT1_PBD to the LD4 motif is either abolished by deleting the LD4 C-terminal sequence (residues 276 -283) or largely diminished by mutating Phe-276 LD4 and Lys-277 LD4 to the corresponding residues (a valine and a glutamine) in the LD2 motif (Fig. 6C), confirming the essential role of Phe-276 LD4 in the GIT1_PBD/paxillin_LD4 interaction. Even the replacement of Phe-276 LD4 with a smaller hydrophobic residue may interfere with the interaction. As Phe-276 LD4 is not conserved in other LD motifs of paxillin, it is very likely that the C-terminal sequences of the LD motifs determine the binding selectivity for GIT1_PBD. In contrast to GIT1_PBD, FAK_FAT lacks an appropriate pocket to hook Phe-276 LD4 . Indeed, the binding of FAK_FAT to the LD4 motif is only mildly decreased by the deletion mutation (Fig. 6D), indicating that the LD core flanking sequence contributes little to the selectivity for FAK_FAT.

Discussion
FA dynamics are crucial in regulating various cellular processes (5,7,8). The involvement of GIT1, liprin-␣, and paxillin in FA dynamics has been studied extensively (9,24,32,37,38). Due to the lack of structural information, the molecular mechanism of the coordinated bindings of GIT1 to liprin-␣ and paxillin has been largely unknown. In this study, we solved the crystal structures of GIT1_PBD in complex with the short PBDbinding sequences in liprin-␣ and paxillin, respectively, revealing that GIT1_PBD utilizes the H1/H4 groove to recognize its targets by different modes. These high-resolution structures allow us to design specific binding-deficient mutations of GIT1, providing an amenable way to differentiate the cellular effects of these two competitive interactions. Considering the importance of GIT1 and liprin-␣ in neuronal development (14, 33, 39 -42), our study provides structural information for future investigations of the GIT1/liprin-␣ interaction in neurons.
Given the presence of both paxillin and liprin-␣ in FAs, it remains puzzling how GIT1 switches the binding from liprin-␣ to paxillin or vice versa. Because phosphorylation plays an important role in regulation of GIT1's function (43,44), it is possible that phosphorylated GIT1 selectively blocks one of the two bindings. In support of this hypothesis, we identified Thr-765 in GIT1_PBD as a potential phosphorylation site, and the phosphorylation-mimicking mutation T765E disrupts the binding of GIT1 to liprin-␣ but not paxillin. Such phosphorylation may control GIT1 to alternatively interact with the two PBD-mediated interactions. Notably, many protein-protein interactions in FAs are mediated by LD motifs. For example, GIT1_PBD interacts with some LD motifs of paxillin, HIC5, and leupaxin (16,17), whereas paxillin_LD4 binds with FAK (45), PYK2 (28), and parvin (46,47). All of these interactions need to be coordinated to control the focal adhesion dynamics. Thus, it is possible that the potential phosphorylation of Thr-765 may also regulate the binding of GIT1 to other FA proteins.
The previous studies (15, 19, 28 -30) have indicated that the LD core sequence determines the binding of LD motifs to FATlike domains (Fig. S3). Nevertheless, the GIT1_PBD/pax-illin_LD4 structure, for the first time, reveals that the sequence flanking the LD core contributes significantly to the FAT-like domain/LD interaction (Fig. 6). Although the LD core sequences of different LD motifs are highly similar, their flanking sequences vary a lot (Fig. 1B). Considering that different FAT-like domains show certain selectivity to bind with different LD motifs (29,30,45,49), it is tempting to think that the flanking sequences may participate in the bindings of other FAT-like domains. Therefore, by demonstrating that a FATlike domain can develop an additional surface to recognize the sequences near the LD core region in LD motifs, this study provides a plausible mechanism for the selective binding of different LD motifs. Despite the weak binding of LD2 to GIT1 (Fig.  S5), the multisite binding of paxillin to GIT1 through LD2 and LD4 may enhance the recruitment of GIT1 to FAs.
Although the SAH sequence of liprin-␣ contains the LD-like motif, surprisingly, it binds with GIT1_PBD using a different mode by adopting the reverse binding orientation (Fig. 3, A and  B). By aligning the LD-like sequence of SAH to LD4, we found that the corresponding positions of 276 FK 277 LD4 that play a critical role in the binding of the LD4 motif to GIT1_PBD are replaced by "IQ" in SAH (Fig. 1B), implying that even if SAH bound with GIT1_PBD using the LD4 orientation, the binding affinity would be much lower than in the reverse orientation. Our observation of the reverse binding orientation of the LDlike motif indicates that other FAT-like domain may also rec- Figure 3. The different binding modes of liprin-␣2_SAH and paxillin-LD4 to GIT1_PBD. A, the superposition of the two complex structures revealing the different binding modes of GIT1_PBD. The hydrophobic residues of the LD4 and SAH peptide that are involved in the SAH-specific, LD4-specific, and overlapped interactions with the H1/H4 groove of GIT1_PBD are shown as a stick model and indicated by green, purple, and black arrows, respectively, in the left panel. The corresponding interacting residues of GIT1_PBD are indicated by purple, green, and black circles in the right panel. The residues that were analyzed by mutagenesis in this study are labeled with one-letter codes and position numbers. B, pairwise sequence alignment of the LD4 motif and the reversed sequence of the SAH peptide. C, ITC curves of the bindings of two GIT1_PBD mutants to liprin-␣2_SAH and paxillin_LD4.

Complex structures of GIT1/liprin-␣ and GIT1/paxillin
ognize the different LD-like motif using different binding modes. In support of this possibility, CD4 was found to interact with FAK_FAT by a reverse orientation (50), despite the fact that the poor quality of the electron density of the binding helix provides few hints to determine which amino acid sequence(s) determines the binding orientation.

Expression constructs and site-directed mutagenesis
DNA encoding sequences of rat GIT1_PBD (residues 639 -770), human liprin-␣2_SAH (residues 642-671), human pax-illin_LD2 (residues 140 -161), and LD4 (residues 252-283) was subcloned into a modified pET-32a vector with an N-terminal thioredoxin (Trx)-His 6 tag. Human FAK_FAT (residues 915-1052) was subcloned into modified pET-28a vector with an N-terminal His 6 tag. The GFP-tagged full-length GIT1 constructs for cellular assays were PCR-amplified using plasmids as template and cloned into mammalian expression vectors containing GFP. All point mutations were created using a site-directed mutagenesis kit and confirmed by DNA sequencing.

Protein expression and purification
All of the proteins were expressed in BL21(DE3) E. coli cells at 16°C with 0.2 mM isopropyl 1-thio-␤-D-galactopyranoside induced in LB medium. The protein samples were purified using nickel-affinity chromatography followed by size-exclusion chromatography. For isothermal titration calorimetry and analytical gel filtration chromatography, the protein samples contained an N-terminal tag. For crystallization, the N-terminal Trx-His 6 tag was cleaved by HRV 3C protease and then separated by size-exclusion chromatography.

Crystallization and data collection
The complex samples were prepared by mixing GIT1_PBD with either liprin-␣2_SAH or paxillin_LD4 and were further purified by size-exclusion chromatography. The complexes were concentrated to ϳ30 mg/ml. Crystals were obtained by the sitting-drop vapor-diffusion method at 16°C. To set up a sitting drop, 1 l of concentrated protein solution was mixed with 1 l of crystallization solution with 0.2 M potassium iodide, 0.1 M MES, pH 6.5, and 25% (w/v) PEG 4000 (for the GIT1_PBD/liprin-␣2_SAH complex) or with 0.1 M HEPES, pH 7.5 and 30% (w/v) PEG 1000 (for the GIT1_PBD/paxillin_LD4 complex). Crystals with better qualities were obtained by removing six N-terminal flexible residues in GIT1-PBD and eight N-terminal flexible residues in paxillin_LD4. Before X-ray diffraction experiments, crystals were soaked in the crystallization solutions containing additional 20% (v/v) PEG 400 or 30% (w/v) PEG 1000 for cryoprotection. Diffraction data were collected at the Shanghai Synchrotron Radiation Facility beamlines BL17U1 and BL19U1. Data were processed and scaled using HKL3000 software.

Structure determination and analysis
The initial phase of the complex structure was determined by molecular replacement in PHASER (51) using the apo-structure of GIT1_PDB (PDB code 2JX0) as the search model. The liprin-SAH or paxillin-LD4 peptide was further built into the corresponding model. The model was refined in PHENIX (52). COOT was used for model rebuilding and adjustments (53). In the final stage, an additional TLS refinement was performed in PHENIX. The model quality was checked by MolProbity (54). The final refinement statistics are listed in Table 1. All structure figures were prepared by using PyMOL (Schroedinger, LLC, New York). The phosphorylation site prediction was performed on the NetPhos 3.1 server (http://www.cbs.dtu.dk/ services/NetPhos/) 5 (48).

ITC analysis
ITC experiments were carried out on a VP-ITC Microcal calorimeter (Malvern) at 25°C. All proteins were dissolved in 50 mM Tris buffer containing 100 mM NaCl, 1 mM EDTA, 2 mM DTT at pH 7.5. Each titration point typically consists of injecting 10-l aliquots of the GIT1_PBD, its mutants, or FAK_FAT at concentration of 400 M into the solution of containing liprin-SAH, paxillin-LD2/4, or their mutants at a concentration of 40 M. A time interval of 150 s between two titration points was used to ensure the complete equilibrium of each titration reaction. The titration data were analyzed using the program Origin version 7.0 and fitted by a one-site binding model.

Analytical gel filtration chromatography
Analytical gel filtration chromatography was carried out on an ÄKTA pure system (GE Healthcare). Protein samples at a concentration of 50 M were loaded onto a Superdex 200 Increase 10/300 GL column (GE Healthcare), equilibrated with 50 mM Tris-HCl buffer, pH 7.5, containing 100 mM NaCl, 1 mM EDTA, and 2 mM DTT.

Multi-angle light-scattering analysis
A miniDAWN TREOS (Wyatt Technology Corp.) coupled with an ÄKTA pure system (GE Healthcare) was used for molar mass measurement. The procedure follows the protocol used for analytical gel filtration analysis.

Cell culture, transfection, and fluorescence imaging
COS7 cells were cultured in Dulbecco's modified Eagle's medium (Corning, 10-013-CVR) supplemented with 10% fetal bovine serum (Pan Biotech) and 50 units/ml penicillin and streptomycin. Transfections of either WT or mutant GIT1/ EGFP-expressing plasmids were performed with polyethyleneimine-25,000 (Polyscience) according to the manufacturer's  Figure 5. The potential phosphorylation regulation of the binding of GIT1_PBD to liprin-␣. A, prediction of phosphorylation sites in the H1/H4 groove of GIT1_PBD. Predicted phosphorylation potential value of each serine, threonine, and tyrosine in the sequence is indicated as a bar. The regions of the H1 and H4 helices are indicated. B and C, ITC-based measurements of the affinities for the bindings of the phosphorylation-mimicking mutant T765E of GIT1_PBD to liprin-␣2_SAH (B) and paxillin_LD4 (C). D, cell imaging of exogenous expressed GFP-tagged GIT1 T765E mutation in COS7 cells. E-G, FA areas and numbers were analyzed using the same method as described in the legend to Fig. 4.  1OW7). B, the structural determinant of the binding of Phe-276 LD4 to the H1/H4 groove of GIT1_PBD. Compared with FAK_FAT, the H1 helix of GIT1-PBD has an additional helical turn, highlighted by a dashed circle, to form the hydrophobic surface patch serving for the recognition of Phe-276 LD4 upon LD4 binding. The bulky side chain of Phe-276 LD4 is indicated as sticks and meshes. C, ITC titrations of GIT1_PBD to paxillin_LD4 and its mutations. D, ITC titrations of FAK_FAT to paxillin_LD4 and its mutation.

Complex structures of GIT1/liprin-␣ and GIT1/paxillin
instructions. One day after transfection, the cells were trypsinized, replated on ϳ20 g/ml fibronectin (Millipore)coated coverslips, and cultured for 1 h. After fixation with 4% paraformaldehyde, the cells were stained with anti-paxillin (BD Biosciences) followed by Alexa 594 -conjugated anti-mouse IgG Ab (Invitrogen) and observed under a Nikon A1R confocal microscope. Images were analyzed using ImageJ software (National Institutes of Health).

CD spectroscopy
CD spectra were recorded using a Chirascan spectrometer (Applied Photophysics, Leatherhead, UK). The buffer used contained 20 mM Na 2 HPO 4 /KH 2 PO 4 (pH 7.0), 50 mM NaCl. Protein sample was concentrated to 50 M and loaded into a quartz cell with a 1-mm path length. The measurement was performed every 1 nm with a 0.5-s time per point and averaged from four repeated scans.