Guanylate kinase domains of the MAGUK family scaffold proteins as specific phospho‐protein‐binding modules

To understand the molecular basis underlying the bindings of the SH3‐GK tandems of DLG4/PSD‐95 and DLG2/SAP102 with the linker domain of LGN/mPins reported earlier (BellaIche et al, 2001; Sans et al, 2005), we performed a series of biochemical binding experiments, including analytical gel filtration chromatography and pull‐down assays, using bacterially expressed, highly purified SH3‐GK tandems of PSD‐95 and SAP97 and the linker region of LGN (residues 374–479). To our disappointment, we could not detect any bindings between the SH3‐GK tandems and the LGN linker in these in vitro assays (data not shown). However, in HEK293T cells co‐expressing GFP‐tagged full‐length mouse LGN and Myc‐tagged full‐length rat SAP97, LGN could be coimmunoprecipitated with SAP97 as reported earlier (BellaIche et al, 2001; Sans et al, 2005). Interestingly, addition of a cocktail of phosphatase inhibitors (100 mM sodium fluoride, 10 mM β‐glycerophosphate, 20 mM PNPP, and 50 μM sodium vanadate) into the cell lysate led to a two‐fold enhancement of the amount of LGN precipitated by SAP97 (Figure 1A), suggesting that the interaction between LGN and SAP97 might be regulated by phosphorylation. This finding is encouraging, as recent studies have shown that Ser401 in the linker region of LGN is phosphorylated by aPKC during mitosis in polarized epithelial cells (Hao et al, 2010), and that Ser436 of Drosophila Pins (which is the equivalent of Ser401 in LGN) is phosphorylated by Aurora‐A kinase in S2 cells (Johnston et al, 2009). Elimination of the phosphorylation of LGN Ser401 (or Pins Ser436) led to mitotic spindle orientation defects (Johnston et al, 2009; Hao et al, 2010). To test whether Ser401 of LGN could be phosphorylated by aPKC directly, we performed an in vitro phosphorylation assay on the purified LGN linker (aa 374–479) by GFP‐aPKC. Indeed, the LGN linker was robustly phosphorylated by aPKC. Substitution of Ser401 with Ala largely eliminated the aPKC‐mediated phosphorylation of the LGN linker (Figure 1B), indicating that Ser401 is the major aPKC phosphorylation site of the LGN linker. We next tested whether the previously reported bindings of DLG SH3‐GK tandems to the LGN linker (Sans et al, 2005) might be directly regulated by the aPKC‐mediated phosphorylation of Ser401. As shown in Figure 1C, the interaction between SAP97 SH3‐GK and the LGN linker could only be observed when the LGN linker was treated with aPKC. Additionally, the S401A mutant of the LGN linker could hardly bind to SAP97 SH3‐GK even in the presence of aPKC, confirming our hypothesis that the SAP97 SH3‐GK/LGN interaction requires the phosphorylation of Ser401 in the LGN linker. Amino acid sequence analysis of the linker region of LGN from different species revealed a highly conserved, short peptide fragment (aa 395–414) surrounding the aPKC phosphorylation site Ser401 (Figure 1D). We asked whether this short peptide fragment of LGN is sufficient to bind to SAP97 SH3‐GK upon phosphorylation. To test this hypothesis, we assayed the binding between SAP97 SH3‐GK and an 18‐residue synthetic phospho‐LGN peptide (‘GRRHpSMENLELMKLTPEK’, referred to as p‐LGN18) using fluorescence spectroscopy, and found that p‐LGN18 binds to SAP97 SH3‐GK with a high affinity (Kd ∼0.22 μM; Figure 1E). Importantly, the unphosphorylated LGN18 peptide showed a ∼500‐fold weaker binding (Kd ∼102 μM) to SAP97 SH3‐GK compared to the p‐LGN18 peptide (Figure 1E). Taken together, the above biochemical results demonstrate that the direct interaction between the SAP97 SH3‐GK tandem and the LGN linker requires the phosphorylation of Ser401, an event possibly mediated by aPKC ((Hao et al, 2010); and this study), or other kinases such as Aurora‐A kinase (Johnston et al, 2009) and PKA or PKB/AKT as predicted by Scansite (http://scansite.mit.edu/).

To elucidate the molecular mechanism underlying the SAP97 SH3‐GK/p‐LGN18 interaction, we solved the crystal structure of the SAP97 SH3‐GK/p‐LGN18 complex at 2.7 Å resolution (Table I). The SAP97 SH3‐GK tandem is composed of a split SH3 domain and a GK domain (Figure 2). The SH3 domain is split by an elongated HOOK/hinge‐region composed of an α‐helix (αA) and a ∼50 residue unstructured hinge sequence (Figure 2A). The βE strand after the HOOK region and the βF strand at the very C‐terminal following the GK domain form an antiparallel sheet, which completes the SH3 domain fold (Figure 2A). The overall structure of SAP97 SH3‐GK is highly similar to that of the PSD‐95 SH3‐GK structures solved earlier (Figure 2C; McGee et al, 2001; Tavares et al, 2001), which is not suprising as the amino acid sequences of the SH3‐GK tandems of the two DLGs (except for the unstructured HOOK regions) are highly similar (Supplementary Figure 1). Beside the N‐terminal first residue and the C‐terminal a few residues, the electron densities of the rest of the p‐LGN18 peptide are clearly defined in the complex (Figure 2B). The p‐LGN18 peptide occupies the elongated concave groove formed by the NMP‐binding domain and the helical core sub‐domain of GK (Figure 2A and B; Supplementary Figure 2), although the vast majority of the p‐LGN18 contacting sites are limited to the residues in the NMP‐binding sub‐domain of GK (Figure 3A and B). The centre of the p‐LGN18 peptide forms a one‐turn α‐helix and followed by a 4 residue β‐strand conformation in the complex. There is no direct contact between the p‐LGN18 peptide and the SH3‐Hinge domains, implying that the GK domain is sufficient for the p‐LGN18 binding (Figure 2A). Indeed, the isolated GK domain from either SAP97 or PSD‐95 binds to p‐LGN18 with indistinguishable affinities compared to their SH3‐GK counterparts (Figure 3D). We also show that truncation of the last 14 residues of SAP97 SH3‐GK (‘SAP97 SH3‐GKΔ14C’ in Figure 3D, the equivalent of the fly DLG^SW mutant, (Woods et al, 1996)) has no impact on the p‐LGN18 binding, further demonstrating that the SH3 domain is not required for the DLG/LGN interaction. Structural analysis further showed that the structure of SAP97 SH3‐GK in the SAP97 SH3‐GK/p‐LGN18 complex is slightly closer to the PSD‐95 SH3‐GK structure crystallized in the GMP‐bound form (RMSD=0.625 Å) than that crystallized in the GMP‐free form (RMSD=0.704 Å; Figure 2C). However, this difference is minor.

The binding of p‐LGN18 to the GK domain is mainly mediated by extensive polar (charge–charge and hydrogen bonding) and hydrophobic interactions (Figure 3A and B; Supplementary Figure 3). The GK/p‐LGN18 interface can be divided into two sites: the phospho‐Ser‐binding site (Phospho‐site) and the Specificity‐site (Figure 3C). At the Phospho‐site, the phosphate group of pSer at the 0 position from the p‐LGN18 peptide forms an extensive charge–charge and hydrogen bonding interaction network with Arg755, Arg758, and Glu761 from the β3/β4‐loop, Tyr767 from β4, and Tyr796 from β6 of the GK domain. The side chain of Arg at the ‐3 position (R‐3) of p‐LGN18 forms salt bridges with Glu761 and Asp766 from the β3/β4‐loop of GK and a hydrogen bond with the phosphate group from pSer(0) (Figure 3A and Supplementary Figure 3). The extensive interactions between p‐LGN18 and the GK domain in the Phospho‐site provide a clear mechanistic explanation as to why Ser(0) needs to be phosphorylated for LGN to bind to DLG1 SH3‐GK. Consistent with these structural data, substitution of two Arg residues of GK domain in the Phospho‐site (Arg755&Arg758 in SAP97, Arg568&Arg571 in PSD‐95) with Gly led to a >70‐fold decrease in p‐LGN18 binding (Figure 3D). A quadruple substitution mutant (R755,758G/Y767,796A) of SAP97 SH3‐GK had no detectable binding to p‐LGN18 (Figure 3D). Replacement of Arg(‐3) with an Ala caused a 5‐fold decrease in p‐LGN18's binding to PSD‐95 SH3‐GK. Finally, although substitution of Ser(0) of LGN18 with Glu led to a ∼14‐fold affinity decrease between the DLG SH3‐GK/LGN complex formation, the S(0)E‐LGN18 mutant peptide still displays a respectable binding affinity to the SH3‐GK tandem (Kd ∼4.6 μM, Figure 3D). This finding points to the possibility that some MAGUK SH3‐GK tandems may also recognize target proteins in a phosphorylation‐independent manner if the residue corresponding to the Ser(0) of LGN18 in such targets is a negatively charged amino acid.

The interactions at the Specificity‐site are mainly hydrophobic. The side chains of Met(+1) and Leu(+4) from p‐LGN18 project into a pocket formed by the side chains of Cys749, Pro751, Tyr767, Tyr796, and Thr798 from GK domain. Met(+7) and Leu(+9) from p‐LGN18 make hydrophobic contacts with the side chains of Ile780, Ala788, and Leu795 from the GK domain (Figures 3A and B; Supplementary Figure 3). Again, entirely consistent with the above structural findings, removal of the C‐terminal four residues from the p‐LGN18 peptide, none of which form any direct contact with the GK domain, did not alter its binding to the SH3‐GK tandems of both SAP97 and PSD‐95 (Figure 3D, ‘p‐LGN14’). Further truncation of the p‐LGN14 peptide by removing the hydrophobic Met(+7) and Leu(+9) led to a >50‐fold decrease in its binding to SH3‐GK (Figure 3D, ‘p‐LGN9’). Thus, the residues C‐terminal to pSer(0) play critical roles in establishing the specificity of the interaction between LGN and DLG SH3‐GK. In agreement with the above analysis, an 16‐residue synthetic peptide corresponding to residues 405–420 of mouse AGS3, equivalent to p‐LGN18 in LGN (Figure 1D), displayed a very weak binding to DLG SH3‐GK (Kd ∼63 μM, Figure 3D). This much weaker binding of the p‐AGS3 peptide to SH3‐GK can be explained by the differences of several critical residues between p‐LGN18 and p‐AGS3. These differences are: the ‐3 residue is a Gln instead of Arg, the +1 position is a small Ala instead of a more bulky Met, and the +4 position is a very bulky Trp instead of a Leu in AGS3 (Figure 1D). Consistent with the above analysis, substitution of Gly789 in the shallow hydrophobic pocket accommodating Met(+1) and Leu(+4) of p‐LGN18 with a bulky Tyr decreased SAP97 SH3‐GK's binding to p‐LGN18 by ∼100‐fold (Figure 3B and D). In agreement with the above structural and biochemical analysis, it has been shown that AGS3 cannot replace LGN in regulating mitotic spindle orientations during asymmetric cell divisions (Williams et al, 2011) and DLG/LGN complex‐regulated NMDA receptor trafficking (Sans et al, 2005).

The amino acid sequence conservation map of GK domains of the DLG‐, CASK‐, MPP‐, and MAGI‐sub‐family MAGUKs showed that the residues forming the phospho‐peptide‐binding groove (the Phospho‐site in particular) are highly conserved (Figure 3C), suggesting that the GK domains from these MAGUKs share similar phosphor‐peptide‐binding properties.

Although the GK domain of PSD‐95 displays a very weak GMP binding (Olsen and Bredt, 2003), the PSD‐95 SH3‐GK tandem could be crystallized with a bound GMP in the presence of a very high concentration (5 mM) of GMP in the crystallization buffer (Tavares et al, 2001). A comparison of the structures of the SAP97 SH3‐GK/p‐LGN18 complex, the PSD‐95 SH3‐GK/GMP complex, and the yeast GK/GMP complex illustrates how the DLG GK domains have evolved into specific pSer/pThr‐binding modules (Figure 4). All of the residues involved in binding to the phosphate group of GMP in yeast GK (Arg39, Arg42, Glu45, Tyr51, and Tyr79) are retained in DLG GKs (Arg 568, Arg571, Glu574, Tyr580, and Tyr609 in PSD‐95 GK). The structure of GMP‐bound SH3‐GK illustrates that the mechanisms of GMP and peptide phosphate group recognitions by DLG GKs are essentially the same. Interestingly, the GMP‐bound PSD‐95 SH3‐GK structure contains a guanidine group from the crystallization buffer (Figure 4B). This guanidine group occupies the same position as the Arg(‐3) side chain guanidine in the SAP97 SH3‐GK/p‐LGN18 structure (Figure 4A), illustrating the importance of Arg(‐3) of p‐LGN18 in binding to DLG GKs. The three critical residues responsible for the guanine ring binding in yeast GK (Ser35, Glu70, and Asp101) are not present in DLG GKs (Figure 4C), explaining why the GMP binding affinities of DLG GKs are so low (Olsen and Bredt, 2003). Instead, a number of conserved, hydrophobic residues from the NMP‐binding domain of DLG GKs function to recognize several hydrophobic residues C‐terminal to the pSer in p‐LGN18 (Figure 3A and Supplementary Figure 3). Taken together, the above structural analysis indicates that the GK domains of DLG have evolved from GMP‐binding domains into specific pSer/pThr peptide‐binding modules.

Among previously identified DLG GK‐binding proteins, we have been particularly interested in DLGAP1 (Kim et al, 1997), as DLGAP1 is a major post‐synaptic scaffold protein with an abundance similar to that of PSD‐95 and each accounts for >5% of total PSD proteins (Walikonis et al, 2000; Peng et al, 2004; Sheng and Hoogenraad, 2007; Feng and Zhang, 2009). Understanding of the molecular basis govening the PSD‐95/DLGAP1 interaction is important for understanding the functional roles of DLGAP1 and its isoforms in synaptic plasticities, the potential involvement of DLGAP3 in obsessive‐compulsive disorder caused by the deletion of DLGAP3 (Welch et al, 2007), and DLGAP‐medaited neuronal polarity development (Manneville et al, 2010). However, biochemical evidence supporting the direct binding between DLGAP1 and PSD‐95 SH3‐GK has eluded us in the past, although the PSD‐95 GK‐binding domain of DLGAP1 has been narrowed to the N‐terminal five 14 aa‐repeats (Kim et al, 1997). With the discovery of the specific phsophorylation‐dependet interaction of LGN to DLG SH3‐GK tandems, we revisited the DLGAP1/PSD‐95 SH3–GK interaction. Amino acid sequence alignment analysis immediately revealed many similarities between the 5 N‐terminal 14 aa‐repeats and LGN18 (Figure 5A). Specifically, each DLGAP1 14 aa‐repeat contains absolutely conserved Ser and Arg residues, which are aligned perfectly with Ser(0) and Arg(‐3) of LGN18. Additionally, each DLGAP1 14 aa‐repeat also contains two hydrophobic residues immediately following Ser(0). These two hydrophobic residues can align well with Met(+1) and Leu(+4) of the LGN18 peptide if we assume that the corresponding sequences in 14 aa‐repeats adopt extended structures instead of an α‐helix, as in p‐LGN18 (Figure 5A). The above analysis suggests that the DLGAP1 14 aa‐repeats are likely to bind to PSD‐95 SH3‐GK in a phosphorylation‐dependent manner. We chose two repeats (DLGAP1‐R2 and 3) to test this prediction. Exactly as we predicted, both the p‐DLGAP1‐R2 and p‐DLGAP1‐R3 peptides bind to PSD‐95 SH3‐GK with affinities comparable to that of p‐LGN18 (Figure 5B). In sharp contrast, the non‐phosphorylated DLGAP1‐R2 peptide bound very weakly to PSD‐95 SH3‐GK (Kd ∼143 μM). On the basis of the structure of the SAP97 SH3‐GK/p‐LGN18 complex, we built a structure model of the SAP97 SH3‐GK/p‐DLGAP1‐R2 complex (Figure 5C). In this model, pSer(0) of p‐DLGAP1‐R2 occupies the GMP‐binding pocket of SAP97 GK as pSer(0) of p‐LGN18 does. Residues immediately C‐terminal to pSer(0) adopts an extended structure, allowing Tyr(+1) and Leu(+2) of p‐DLGAP1‐R2 to occupy positions in the shallow hydrophobic pocket of GK analogous to Met(+1) and Leu(+4) of p‐LGN18 (Figure 5C). Thr(+5) and Pro(+7) of p‐DLGAP1‐R2 make additional hydrophobic contacts with GK, just as Met(+7) and Leu(+9) of p‐LGN18 do (Figure 5C). To validate the structural model of the SAP97 SH3‐GK/p‐DLGAP1‐R2 complex, we showed that the changes of the residues at the Phospho‐site (the ‘R755,758G/Y767,796A’ mutant) and the shallow hydrophobic pocket (the ‘P751K,G789Y’ mutant) of SH3‐GK eliminated its binding to p‐DLGAP1‐R2 (Figure 5C and D). We have tested the role of the hydrophobic residues in the +2 and +5 positions of the DLGAP1‐R2/3 in binding to PSD‐95 SH3‐GK by substituting the Leu(+2) and Val(+5) of DLGAP1‐R3 with Gln. The mutant p‐DLGAP1‐R3 peptide binds to PSD‐95 SH3‐GK with a ∼10‐fold lower affinity then the WT peptide does (Supplementary Figure 5). Finally, amino acid sequence analysis reveals that all members of the DLGAP family contain highly similar DLG SH3‐GK‐binding sequence motifs in their N‐termini (Figure 5E), suggesting that the phosphorylation‐dependent binding to DLGs is a property common to all isoforms of DLGAPs.

With our successful prediction of DLGAP1 as a phosphorylation‐dependent binder of DLG, we wondered whether we might be able to extend this prediction to other previously identified DLG‐interacting proteins. We carefully analyzed amino acid sequences of two additional DLG‐binding proteins, namely a Rap‐specific GTPase‐activating protein called SPAR (Pak et al, 2001) and a synaptic scaffold protein BEGAIN (Deguchi et al, 1998). To our delight, we found that both of these proteins contain a highly similar DLG GK‐binding sequence motif (Figure 5A). Importantly, all of these predicted DLG GK‐binding sequences are highly conserved in vertebrates (data not shown). We chose the DLG GK‐binding motif of SPAR to test our prediction. Indeed, a phospho‐SPAR peptide (‘PLRRPpSYTLGMKSL') binds to PSD‐95 SH3‐GK with a Kd of ∼15 μM (Figure 5F). The slightly weaker affinity of the phospho‐SPAR peptide for PSD‐95 SH3‐GK is likely due to a Pro at the ‐1 position, as this Pro would introduce an unfavourable backbone torsion angle in restricting Arg(‐3) to stabilize the phosphate group of pSer(0) as shown in Figure 4A. With the above findings, we conclude that the majority of previously identified DLG GK/target interactions probably occur via the GK‐binding sequence motif identified in this study in a phosphorylation‐dependent manner.

MAGUKs can be divided into 11 sub‐families (Velthuis et al, 2007; Mendoza et al, 2010). Every member of the MAGUK family proteins contains one GK domain (Figure 6A). On the basis of the structural and biochemical data obtained from the DLG GK/target interactions described above, we analyzed possible target binding properties of the GK domains from other MAGUKs. We performed a detailed structure‐based amino acid sequence alignment analysis of all MAGUK family GK domains (Supplementary Figure 1), and selected the whole DLG family and one representative member from each of the other MAGUK subfamilies for detailed comparison in Figure 6B. In this sequence alignment analysis, the residues forming the pSer/pThr‐recognition Phospho‐site of the GK domains are boxed and shaded in yellow, and the residues forming the target binding Specificity‐site are boxed and highlighted in sky‐blue (Figure 6B). All of the DLG GKs contain absolutely conserved amino acid residues in the both Phospho‐ and Specificity‐sites (Figure 6), and thus the DLG GKs should share common phospho‐target binding sequence motif shown in Figure 5A. As we have demonstrated that the the residues forming the Phospho‐site are absolutely required for DLG GK/target interactions, the sequence alignment shown in Figure 6B immediately reveals that the ZO‐, DLG5‐, CARMA‐, and β‐subunit calcium‐sub‐families of MAGUK GKs are not likely to bind to phospho‐target proteins, due to the lack of critical phosphate recognition residues in their Phospho‐site. The CASK‐, MPP‐, and MAGI‐sub‐family MAGUKs all contain complete pSer‐binding pockets, and thus these MAGUK GKs are likely to bind to their targets in a phosphorylation‐dependent manner. It is further noted that the residues in the Specificity‐site of the CASK, MPP, and MAGI GK domains are distinctly different from those of DLG GKs, indicating that CASK, MPP, and MAGI GKs are likely to recognize very different amino acid residues surrounding the pSer. On the basis of the above analysis, we predicted a possible CASK SH3‐GK‐binding peptide sequence of whirlin, which is a reported binding target of CASK (Mburu et al, 2006). Consistent with this prediction, this whirlin peptide (‘GQTL(pS)EDSGVDAGEA’) binds to CASK SH3‐GK only when the Ser at the position 0 is phosphorylated (Supplementary Figure 6). It is further noted that the amino acid sequence of the p‐whirlin peptide is quite different from that of p‐LGN18.

Anti‐GFP antibody was purchased from Santa Cruz; anti‐Myc antibody was from DHSB, and anti‐His‐tag antibody was purchased from Sigma.

Rat SAP97 SH3‐GK (aa 578–911), SAP97 GK (aa 719–911); human PSD‐95 SH3‐GK (aa 426–724), PSD‐95 GK (aa 531–724); and the LGN linker (aa 374–479) were individually cloned into the pET‐15b vector. All the mutations used in this study were created using the standard PCR‐based mutagenesis method and confirmed by DNA sequencing. Recombinant proteins were expressed in Escherichia coli BL21 (DE3) host cells at 16°C, and purified by using a Ni²⁺‐NTA agarose affinity chromatography followed by size‐exclusion chromatography. For in vitro biochemical analysis, SAP97 SH3‐GK and PSD‐95 SH3‐GK were expressed as the GST‐fused proteins and purified by GSH‐Sepharose affinity chromatography.

For immunoprecipitation of overexpressed proteins, HEK293T cells were co‐transfected with the GFP‐tagged full‐length LGN and the Myc‐tagged full‐length SAP97. Cells were harvested 24 h post‐transfection and divided equally into two tubes. Cells were lysed in a lysis buffer (50 mM HEPES pH 7.4, 150 mM NaCl, 10% glycerol, 1 mM EGTA, and 1% Triton) with and without a cocktail of phosphatase inhibitors (100 mM sodium fluoride, 10 mM β‐glycerophosphate, 20 mM PNPP, and 50 μM sodium vanadate) for 30 min at 4°C. Each cell lysate was incubated with 50 μl anti‐Myc antibody‐coupled Protein G beads (50% slurry). After incubation for 2 h at 4°C, the beads were washed with the lysis buffer extensively. The captured proteins were then eluted by the SDS–PAGE sample buffer and immune‐detected with specific antibodies.

For the GST pull‐down assays, GST‐tagged DLG SH3‐GKs (50 μl from 1 mg/ml stock solutions) were first loaded to 30 μl GSH‐sepharose 4B slurry beads in an assay buffer (50 mM Tris pH 8.0, 100 mM NaCl, 1 mM DTT, and 1 mM EDTA). The GST fusion protein loaded beads were then incubated with GFP‐aPKC‐treated or ‐untreated potential binding partners for 2 h at 4°C. After three times washing, the proteins captured were eluted by boiling, resolved by 15% SDS–PAGE, and detected by western blot.

About 1 mg/ml purified His‐tagged WT mouse LGN linker (aa 374–479) and S401A mutant were dialyzed in the kinase buffer containing 20 mM MOPS (pH7.4), 15 mM MgCl₂, and 2 mM DTT. GFP‐tagged aPKC was transfected in HEK293T cells and harvested 24 h after transfection. Cell lysate was incubated with GFP antibody‐coupled Protein G beads. Beads were washed with the lysis buffer and the kinase assay buffer, respectively. GFP‐tagged aPKC was eluted with purified GFP protein. In each kinase assay, 10 μl of each protein sample was mixed with 10 μl of affinity purified GFP‐tagged aPKC. The mixture was incubated at 25°C for 1 h in the kinase assay buffer containing 2 μCi [γ‐³²P]‐ATP (PerkinElmer) and 0.1 mM nonradioactive ATP. [³²P] phosphate labelling of bound proteins was detected by SDS–PAGE electrophoresis followed by autoradiography.

Fluorescence assays were performed on a PerkinElmer LS‐55 fluorimeter equipped with an automated polarizer at 25°C. In a typical assay, a FITC (Molecular Probes)‐labeled peptide (∼1 μM) was titrated with each binding partner in a 50 mM Tris pH 8.0 buffer containing 100 mM NaCl, 1 mM DTT, and 1 mM EDTA. The Kd values were obtained by fitting the titration curves with the classical one‐site binding model.

Crystals of SAP97 SH3‐GK tandem in complex with the p‐LGN18 peptide were obtained by the hanging drop vapour diffusion method at 16°C. Freshly purified SAP97 SH3‐GK was concentrated to 10 mg/ml before a saturating amount of p‐LGN18 was added. The SAP97 SH3‐GK/p‐LGN18 complex was grown in 0.2 M ammonium nitrate, 20% w/v polyethylene glycol 3350. Glycerol (25%) was added as the cryo‐protectant. A 2.7 Å resolution X‐ray data set was collected at the beam‐line BL17U1 of the Shanghai Synchrotron Radiation Facility. The diffraction data were processed and scaled by HKL2000 (Otwinowski and Minor, 1997).

Using the structure of the PSD‐95 SH3‐GK tandem (PDB code: 1KJW) as the search model, the initial structural model was obtained using the molecular replacement method in PHASER (McCoy et al, 2007). The model was then refined by the phenix.refinement (Adams et al, 2010). COOT (Emsley and Cowtan, 2004) was used for the peptide modelling and model adjustments. TLS refinement was applied at the final refinement stage. The final structure was validated by the phenix.model_vs_data (Adams et al, 2010) validation tools. The final refinement statistics are listed in Table I. The structure figures were prepared using the program PyMOL (http://www.pymol.org/).

To build the SAP97 SH3‐GK/p‐DLGAP1‐R2 complex model, the residues in the p‐DLGAP1‐R2 were first aligned with the corresponding residues of p‐LGN18 in the SAP97 SH3‐GK/p‐LGN18 structure. After changing the residues of the p‐LGN18 peptide into those of the p‐DLGAP1‐R2 peptide using COOT, FlexPepDocking from Rosetta (Raveh et al, 2010) was used to calculate 200 refined docking models. These 200 structures of the SAP97 SH3‐GK/p‐DLGAP1‐R2 complex model converged well (Supplementary Figure 4), and the top 20 structure models with the lowest energy functions were selected for analyses.

The atomic coordinates of SAP97 SH3‐GK/p‐LGN18 complex have been deposited to the Protein Data Bank under the accession codes 3UAT.

Supplementary data are available at The EMBO Journal online (http://www.embojournal.org).