Unusual Binding Properties of the SH3 Domain of the Yeast Actin-binding Protein Abp1

Abp1p is an actin-binding protein that plays a central role in the organization of Saccharomyces cerevisiae actin cytoskeleton. By a combination of two-hybrid and phage-display approaches, we have identified six new ligands of the Abp1-SH3 domain. None of these SH3-mediated novel interactions was detected in recent all genome high throughput protein interaction projects. Here we show that the SH3-mediated association of Abp1p with the Ser/Thr kinases Prk1p and Ark1p is essential for their localization to actin cortical patches. The Abp1-SH3 domain has a rather unusual binding specificity, because its target peptides contain the tetrapentapeptide +XXXPXXPX+PXXL with positive charges flanking the polyproline core on both sides. Here we present the structure of the Abp1-SH3 domain solved at 1.3-Å resolution. The peptide-binding pockets in the SH3 domain are flanked by two acidic residues that are uncommon at those positions in the SH3 domain family. We have shown by site-directed mutagenesis that one of these negatively charged side chains may be the key determinant for the preference for non-classical ligands.

The actin cytoskeleton plays a key role in many essential cellular processes, such as motility, endocytosis, secretion, and membrane recycling (1)(2)(3). As a consequence, its organization and dynamic rearrangements need to be tightly controlled spatially and temporally. A thorough understanding of the interaction network connecting all the actin-associated proteins, the scaffolds and the anchoring proteins, is likely to help to clarify the mechanisms underlying its coordinated regulation (4).
Most of the components of the yeast cell cytoskeleton have homologues in mammals where they often play similar roles (5)(6)(7). Abp1p 1 (actin-binding protein) is a Saccharomyces cer-evisiae protein homologue to the mouse mAbp1p-SH3P7, which is a Src kinase target involved in polarized cell growth and motility (8,9). The yeast Abp1 protein is 592 amino acids long and includes an actin depolymerizing factor-homology domain at the N terminus and a SH3 domain at the C terminus of the protein (10). Abp1p is found concentrated in the actin patches that are enriched at the sites of polarized cell surface growth in the bud of budding yeast and in the mating projection of mating yeast. The overexpression of Abp1p disturbs the actin cytoskeleton and leads to an aberrant budding pattern and cortical actin assembly (11). Deletion of the ABP1 gene, on the other hand, does not cause any apparent cytoskeletal defect (12). Abp1p, however, is found to be essential when any of the genes encoding for Sac6p (the actin filament-bundling protein, fimbrin), Sla1p, Sla2p, or Prk1p is deleted (13,51). These proteins, which functionally interact with Abp1p, are also localized in the cortical actin cytoskeleton (11,13). The observed negative synergism in yeast cells carrying combinations of deletions in ABP1-SH3 and in the SLA1, SLA2, and SAC6 genes suggests that these gene products perform redundant essential function(s) that require the presence of an intact Abp1p-SH3 domain. Moreover, this synthetic lethality is not rescued by Abp1p devoid of its SH3 domain, suggesting a significant role for this protein interaction module (14).
To further characterize the role played by Abp1p, we have set out to identify its binding partners. More specifically, we have explored the role played by the SH3 domain in the assembly and regulation of these actin structures. The SH3 family of protein-protein interaction domains includes Ͼ1200 modules 60 -70-amino acids long sharing an identity of at least 30%. More than 1000 proteins containing SH3 domains are found in many cell compartments and are involved in different functions (e.g. cell cycle control, signal transduction, or cytoskeleton organization). Typically, an SH3 domain recognizes and binds to polyproline-rich regions, which adopt a type II left-handed helix conformation (15,16). The SH3 peptide recognition surface includes an hydrophobic cleft that binds to the proline-rich core and is flanked on one side by variable loops (RT and the N-Src loops) that contribute to the recognition specificity and determine ligand register and orientation (17,18). A typical SH3 ligand contains a PXXP motif that is hosted in the hydrophobic cleft. Depending on the presence of a positive charge at the N or C terminus, the peptides bind in either of the two possible orientations and are classified as class 1 (RXXPXXP consensus) or class 2 (PXXPXR consensus). Recently, the recognition repertoire of the SH3 domain family has been shown to be more diverse than originally thought including atypical ligands, which do not contain positively charged residues that are proline-independent or tyrosine-based (19 -21).
In this study, we have determined the three-dimensional structure of the Abp1p-SH3 domain, which accounts for its unusual recognition specificity. Furthermore, we have identified as Abp1-SH3 partners several proteins that were previously implicated in processes related to endocytosis. For two of these Abp1-SH3 partners, the serine/threonine kinases Ark1p and Prk1p, we have proved that in vivo Abp1p acts via its SH3 domain as an adaptor, which is essential for correct localization of the two kinases to cortical patches. Because Prk1p functions as a negative regulator of endocytosis, our results suggest that Abp1p links the cortical actin cytoskeleton to the control of endocytosis.
The yeast genomic library was cloned into pACT, AmpR, ColE1 ORI, LEU2, and GAL4AD (amino acids 768 -881) fused to yeast genomic DNA fragments [CLONTECH].The selection was performed in Hf7c or in Y-190 [CLONTECH]. The ␤-galactosidase test in liquid was performed after transformation into Y-187 according to the protocol by the manufacturer (CLONTECH). The positive control plasmids were from CLONTECH, such as pCL1 expressing GAL4 full-length, pVA3-1 expressing GAL4BD (amino acids 1-147) fused to murine p53 (amino acids 72-390), and pTD1-1 expressing GAL4AD (amino acids 768 -881) fused to SV40 large T antigen (amino acids 87-708). All prey plasmids were found to be negative when tested with pLAM5 expressing GAL4BD fused to human lamin C protein. The positive interactors in Fig. 1 showed a ␤-galactosidase activity that is 100 -400 times higher than that of the negative control (0.003 units).
Plasmid pyArc40 was used for the test with Arc40p from pACT2 and ARC40 inserted into BamHI-XhoI, which was fused to GAL4AD using the oligonucleotides R804, AATTTggatccCATGTCATTTTCCAATTC TAAAGAC, and R805, AATTTaagcttTTAAATTGTATAAATGACGA TCTTA.
The yeast strains used were HF7c (MATa ura3-52 his3-200  ade2-101 lys2-801 trp1-901 leu2-3112 gal4 -542 gal80 -538  LYS2::GAL1 UAS -GAL1 TATA -HIS3, URA3::GAL4  Affinity Selection of a Yeast Genomic Library Displayed on Phage Capsid-Fragments (50 -1000 bp) of yeast DNA obtained by random priming with partially degenerated oligonucleotides were cloned into the phagemid vector Pc89 between the EcoRI and BamHI sites at the 5Ј end of M13 gene VIII, which encodes the major coat protein (22). The library that we obtained (10 8 independent clones representing Ͼ50 times of the whole genome) was a collection of recombinant phage particles displaying yeast protein fragments of which average length was 40 amino acids. These particles were affinity-selected for their ability to bind to the full-length yeast Abp1 protein expressed as a GST fusion and linked to glutathione S-Sepharose.
10 g of GST-Abp1p were incubated for 1 h at 4°C with 10 10 recombinant filamentous phage particles displaying yeast protein fragments. Unbound phages were discarded by washing the resin with 1ϫ Trisbuffered saline containing 0.1% Tween 20. Affinity-selected clones were amplified in a bacterial host DH5␣ and subjected to a new affinity selection cycle. After the third cycle, selected clones were tested by ELISA for their capability of interacting with GST-Abp1p, and the sequence of the yeast DNA fragments were determined by automatic sequencing of phage single-stranded DNA using universal primer Ϫ40. The Abp1-SH3 fusion to GST was in PYEX(ABP1-SH3), a pGEX2T derivative in which the SH3 domain (Pro-535-Asn-592) was inserted into EcoRI-SacI, with oligonucleotides R403 and R404. The oligonucleotides R649 (TATATggatccATGGCTTTGGAACCTATT) and R650 (TCT-GTgaattcGTTGCCCAAAGACACATA) were used to insert the ABP1 gene cloned into pGEX6P-1 (EcoRI-BamHI) to determine the synthesis of the bait protein Abp1p as a GST fusion. The GST-Abp1p fusion protein produced in the bacterial strain HB101 was immobilized on a glutathione-Sepharose resin. The bacterial strains used were DH5␣ (FЈ::supE44 (lacZYA-argFV169) deoR endA1 gyrA96 hsdR17 (r k Ϫ mk ϩ ) recA1 relA1) and HB101 (supE44 hsdS20 (r-B m-B) recA13 ara-14 proA2 lacY1 galK2 rplS20 xil-5 mtl-1 leuB).
Panning of Phage-displayed Peptide Libraries-Phage panning is performed as described previously (23)(24)(25)(26). A nonapeptide repertoire of random sequence exposed on the pVIII coat protein of the filamentous phage M13 (10 10 particles) were challenged with 10 g of the purified GST-SH3 domain of Abp1p. After three rounds of affinity selection (8 h at room temperature in 1-2 ml of PBS), washing (10 ml of PBS ϩ 0.005% Tween 20), elution (10 min at room temperature in 100 l of 0.1 M HCl, pH 2.2), and neutralization (10 l of Tris-HCl 2 M, pH 9.0), the specifically bound phages were recovered by infection. The singlestranded DNAs of the affinity-selected phage were sequenced to determine the sequence of the selected peptides.
Pull Down Assay from Yeast Cell Extract-The yeast extracts were prepared from the kinase null strains as follows. The prk1 null strain DDY1558 was transformed with the following plasmids: pDD558 from pRS316, CEN, URA3, 6Myc-PRK1 (13) expressing 6Myc-Prk1p under its own promoter; pDD561 from pTS408, GFP-PRK1-kinase-dead K56A, Gal promoter, CEN, URA3 (13) expressing a GFP-tagged Prk1p under GAL promoter; and pDD961 from GFP-PRK1-kinase-dead K56A in which the P752PPK755 of Prk1p was mutated into alanines. The ark1 null strain DDY2115 was transformed with the following plasmids: pDD382 from pRS315, CEN, LEU2, ARK1-6Myc (13) expressing a Ark1p-6Myc under its own promoter and pDD962 from pRS315, CEN, LEU2, ARK1-6Myc kinase-dead K56A in which the sequence Lys-608-Pro-626 of Ark1p-6Myc was substituted by two alanines. To perform the pull down experiment using mutations in the SH3 targets, the ark1 null strain was transformed with pDD962 expressing the Ark1p-6Myc devoid of its polyproline tail, otherwise cells were transformed with pDD961 encoding the GFP-Prk1p with a mutagenized proline stretch. pDD961 and pDD962 were both created using a PCR-stitch technique as follows. To make pDD961, two first-round of PCR reactions were performed using pDD554 (13) as template with primer pairs 1) AF19 and JC74 and 2) JC95w and JC75. The products from these reactions were pooled and amplified in a second round using AF19 and JC95w. pDD561 was digested with AflII and XbaI, and the removed fragment was replaced by similarly cut product from PCR round 2. A similar procedure was used to make pDD962 using pDD1160 (13) as template: SY47 and JC77 and JC76 and SY51 in the first round, SY47 and SY51 in the second round, and replacing an NcoI fragment in pDD1160. All mutations were verified by sequencing.
As a bait for the pull down assay, the Abp1p-SH3 domain (Pro-535-Asn-592) in PYEX(ABP1-SH3) was expressed in Escherichia coli HB101 and purified as a fusion to GST. When using the mutated bait, pYEX(ABP1-SH3 W569A) was used. The conserved tryptophan 569 in the SH3 peptide-binding pocket was changed into alanine using the oligonucleotide R591 (TAGTTCCCCTAGCCACGCGTCATCGTCGAC AAA). The GST-fused bait proteins were immobilized on glutathione S-Sepharose (Amersham Biosciences, Inc.) (30 g) and incubated at 4°C for 2 h with the concentrated extract (10 mg/ml) equivalent to 100 ml of cells at A 600 ϭ 0.5. After extensive washing, the pellet is suspended in SDS and run on PAGE. The bands visible after immunoblotting with anti-Myc or anti-GFP antibodies were obtained loading the input extract from 10exp8 cells or the pulled down materials from 10exp9 cells. The Myc-tagged proteins were detected using the anti-Myc antibody (Invitrogen), The GFP-tagged proteins were detected using the anti-GFP antibody (Invitrogen). PYEX was a derivative of pGEX2TK with an expanded multiple cloning site. The yeast strains used in this study were DDY1558 (MATa ELISA-Affinity selections with polyproline peptides were performed in microtiter plates (Nunc) coated overnight at 4°C with 5 g/ml of streptavidin (Sigma) in 100 l of PBS (10 mM Na 2 HPO 4 /KH 2 PO 4 , pH 7.2, 150 mM NaCl). The coated plates were washed 10 times with PBS and 0.05% Tween 20 and incubated for 30 min at 25°C with biotinylated peptides (10 M in PBS). Plates were washed again and blocked for 1 h at 25°C with 4% bovine serum albumin in PBS. The GST-SH3 fusion proteins were added (1 g). After 10 washes, anti-GST serum in goat (Amersham Biosciences, Inc.) was added in bovine serum albumin and 4% PBS for 1 h at room temperature followed by incubation with anti-goat alkaline-phosphatase-conjugated antibody (Sigma). The ELISA-utilizing phage-exposed peptides were performed using 10exp9 phage particles/well.
Abp1-SH3 domain was crystallized by the sitting-drop technique using 3.2 M ammonium sulfate in 100 mM bis-Tris-propane at pH 8.0 as a precipitating agent and 12 mg/ml Abp1-SH3 domain in the purification buffer. The crystals were optimized by seeding techniques and reducing the protein concentration to 1.2 mg/ml. X-ray data were collected at 100K at the Beamline BW7B (EMBL/ DESY, Hamburg) using a 345-MAR research image plate as detector. The high concentration of ammonium sulfate was sufficient to cryoprotect the crystal. The data were processed with the Denzo and Scalepack package (27). The crystal belongs to the P2 1 2 1 2 1 space group with unit-cell parameters a ϭ 24.0 Å, b ϭ 38.1 Å, and c ϭ 59.1 Å. The Wilson plot, calculated in the resolution range of 3.9 -1.3 Å using the program TRUNCATE from the CCP4 package (28) yielded an overall isotropic temperature factor of 9.8 Å 2 . The statistics of the data processing are reported in Table I.
Structure Solution and Refinement-The structure was phased by molecular replacement with AMoRe (29) using the 1.5-Å resolution structure of c-Crk SH3 domain (PDB entry code 1CKA), 35% sequence homology, and 20% sequence identity (16) as a search model. The observed data between 15 and 3.5 Å resolution were used employing an integration sphere of 12 Å. The final solution had a correlation coefficient of 32.7 and a R-factor of 51.9%.
The structure was built using an ARPwARP package (30) and refined using REFMAC5 (31). The refinement program used the data from 15 to 1.3 Å. The data set contained 13,742 unique reflections from which 687 were excluded for the calculation of the R free (32). One molecule of 58 residues (497 non-hydrogen protein atoms) is present in the asymmetric unit. Anisotropic refinement of the individual atoms was applied, which further improved the statistics.
The weighted 3F o Ϫ 2F c electron density map displayed 11 residues in double conformations. The solvent molecules presenting reasonable hydrogen bonds (2.3-3.3 Å) were added using the option Arp solvent from the ARPwARP package. Two large tetrahedral-like electron density peaks were interpreted as sulfate ions from the crystallization conditions. Finally, the structure was validated by the program PROCHECK (33). A summary of the refinement statistics is given in Table II.

Identification of Ligands of the SH3 Domain of Abp1p-
The yeast two-hybrid approach is the most popular selective method to search for protein partners. However, we have recently demonstrated the feasibility of achieving similar goals by panning phage-displayed libraries in which cDNA or genomic fragments are fused to phage capsid genes (34). We have observed that the two methods are partly complementary, and that a search carried out with a specific bait domain by both methods results in a better coverage of the ensemble of potential partners (35).
By applying such a combined approach to the SH3 domain of Abp1, we have identified seven prey proteins, which were selected by either method (Table III). Our results confirm that the Abp1-SH3 binds to the adenylyl cyclase-associated protein Srv2 (12). In addition, our two-hybrid screen also identifies the open reading frame product Yir003wp, the unconventional myosin Myo5p, and the kinase Prk1p as putative partners of Abp1p. Using the entire Abp1p as a bait, we have also screened a yeast genomic library displayed on filamentous phage capsids obtained by fusing random fragments of the yeast genome to the 5Ј end of f1 gene VIII. This approach permitted us to identify a second Ser/Thr kinase, Ark1p, as an Abp1-SH3 ligand. The 247 amino acid N-terminal catalytic domain of Ark1p is 73% similar to the catalytic domain of Prk1p. The two kinases aside from a similar C-terminal proline-rich sequence are otherwise rather divergent (13). The Ark1p peptide region selected by phage display corresponds to amino acids Arg-561-Leu-627 (Fig. 1) in which two putative SH3-binding motifs are present. Because the fragment of Ark1p was selected by panning phage-displayed libraries, we can exclude the possibility that the interaction between the Ark1p kinase and the SH3 domain is bridged by a third protein.
Another putative partner of Abp1p-SH3 discovered by phage display is Inp52p/Sjl2p, which is similar to synaptojanin, a mammalian synaptic inositol 5-phosphatase that is suggested to have a role in the regulation of membrane trafficking and actin cytoskeleton organization (36). In the short yeast synaptojanin region selected as Abp1-SH3 ligand, we identify a putative SH3-binding motif 1115 PPVVKKP 1121 . Finally, a protein of unknown function, Ynl094wp, was repeatedly found by screening phage display genomic libraries. The shorter selected region comprises the residues from 471 to 504.
Identification of the Abp1-SH3 Consensus Target Motif-The Abp1p-SH3 targets selected by the two-hybrid approach contain several proline-rich regions. However, sequence alignment does not permit the definition of a clear consensus peptide ligand. In preliminary experiments, we had already shown that the SH3 domain of Abp1p does not bind to any of the typical SH3 ligands that we had characterized in our laboratory.
To be able to identify in each protein the peptide that is bound by the Abp1p-SH3 domain, we characterized its recognition specificity by panning peptide libraries of random sequence displayed on filamentous phage. We have initially selected peptides from a repertoire of nonapeptides displayed by fusion to the major coat protein pVIII. After three selection cycles, four clones, which were found to be positive in ELISA, were characterized, and the amino acid sequence of the displayed peptide ligands was deduced from the DNA sequence of the hybrid coat gene (Fig. 1A). However, compared with other SH3 selections carried out in our laboratory, fewer clones were selected, and they were found to bind to the bait Abp1-SH3 less tightly (optical density of 0.2-0.4 in ELISA, whereas typical SH3 showed an optical density of Ͼ1.4 when tested with their partner peptides in similar experimental conditions). This result indicates that the SH3 domain of Abp1p has a higher specificity, and that for tight binding, it requires a larger number of specific residues in fixed positions and possibly a longer peptide. The consensus sequence of the selected peptides, RPX-PXXPXK, could be interpreted either as an extended class 2 peptide, containing an extra positively charged residue at the amino side, or an extended class 1 peptide, containing an extra positively charged residue at the C terminus.
Because this first panning experiment did not yield a sufficient number of ligand sequences, we have also panned a second library of pentadecapeptides containing a fixed PXXP motif displayed at a lower density by the filamentous phage pIII receptor protein (37). The three selected pentadecapeptides reveal a slightly different consensus, PXXPXRPXW# (where # represents a hydrophobic residue), that can be interpreted as an extended class 2 with a requirement for a proline after the conserved positively charged and hydrophobic residues at the C terminus.
We used the consensus sequences obtained by the two panning experiments to identify the best matches on the selected protein partners. Whenever visual inspection was not sufficient, we used a position-specific profile derived from the phage display experiments as described by Tong et al. (35). The results are illustrated in Fig. 1B. Although no protein partner contains an exact match of either consensus, the predicted peptides identify a third consensus, ϩXXPXXPXϩPXX#, that can be interpreted as a merge of the two phage-displayed consensus.
Ark1p, Prk1p, Srv2p, Inp52p, and Yir003wp contain a peptide that matches most of the extended consensus. Myo5p and Ynl094p contain a peptide in which the conserved KP motif is preceded by a long stretch of prolines.
Mapping of the SH3 Target Peptides on Prk1p and Ark1p-To verify by an independent method the physical interaction between the SH3 domain of Abp1p and the Prk1p kinase, we have used Abp1-SH3 fused to glutathione S-transferase to pull down Prk1p fused to an N-terminal 6Myc-tag expressed in prk1 null yeast cell. As shown in Fig. 2A, ϳ5% 6Myc-tagged Prk1 protein is pulled down by the Abp1p-SH3 domain. By contrast, most of the binding is abolished when the conserved tryptophan in the SH3 domain (Trp-569) is changed to an alanine. Several bands can be identified with an anti-Myc-tag antibody in yeast extracts expressing the Myc-tagged Prk1 protein, suggesting that this protein is degraded either in vivo or during the preparation of the protein extracts. However, only the full-length protein is pulled down by the SH3 domain of Abp1p. This is consistent with our identification of the SH3binding site with the proline-rich region at the C terminus of the Prk1 protein.
Similarly, we confirmed by a pull down experiment the in-  (37) were panned for three consecutive cycles with a GST-Abp1-SH3 domain bound to gluthathione-Sepharose. The sequence of the peptides displayed by the selected ligands were deduced from the DNA sequence of the hybrid capsid genes, and from the alignment, an extended consensus was deduced. B, inferred Abp1-SH3 target peptides obtained by comparison of the sequence of the Abp1-SH3 protein partners with the phage-displayed consensus. teraction between Abp1p and the Ark1p tagged with 6Myc at the C terminus (Fig. 2B). The Ark1p-6Myc was efficiently pulled down by GST-Abp1-SH3. By contrast, Abp1-SH3 (W569A) mutated in the conserved tryptophan does not bind to Ark1p. Furthermore, the SH3 domain seems to be as efficient as the full-length Abp1p protein in the pull down assay. These results further stress the primary role of the SH3 domain of the scaffold protein Abp1p in binding to the two regulatory kinases. Moreover, we have used the pull down assay to confirm that the peptides identified by matching the phage display consensus to the Ark1p and Prk1p protein sequences indeed mediate the interaction with the SH3 domain. To this end, we have modified the two kinase genes by altering the sequences encoding the putative SH3-interacting motifs. As a result, four residues in the polyproline stretch ( 752 PPPK 755 ) at the C terminus of GFP-Prk1p were changed into alanines, and the 19 residues ( 608 KPTPPPKPSHLKPKPPPKP 626 ) in the C-terminal region of Ark1p-6Myc were deleted and substituted with two alanines. The two kinases, altered in their putative SH3-binding motifs, no longer bind to the Abp1-SH3 as judged by the pull down experiments (Fig. 2C).
Interaction of the SH3 Domain of Abp1p with the C-terminal Proline-rich Peptide of Prk1p and Ark1p Is Essential for Localization of the Kinases to Actin Cortical Patches-Cope et al. (13) have shown that the localization to the actin cortical patches of Prk1p and Ark1p is dependent on Abp1p. To verify whether the in vivo co-localization of Abp1p with both Prk1p and Ark1p is mediated by the direct physical interaction that we have identified in vitro, we looked at the localization of GFP-tagged kinases in abp1 null yeast cells expressing different derivatives of the Abp1 protein. In this experiment, we used both kinases carrying an inactivating mutation in their enzymatic domain (kinase-dead), because elevated levels of either Ark1p or Prk1p result in abnormal cell morphology and ultimately in cell death (13). Each of these two strains was further transfected with one of three different plasmids expressing (i) a wild type Abp1p, (ii) an Abp1p-deleted of the SH3 domain, or (iii) an Abp1p with a mutated (W569A) SH3 domain. Normal cortical spots of kinase localization are visible only in cells that are transfected with a plasmid expressing the wild type Abp1p, demonstrating that the SH3 domain is essential for proper localization of the two GFP-tagged kinases (Fig. 3, upper panel). However, we did notice a residual cortical localization of Prk1p even in the absence of Abp1p, whereas Ark1p was completely delocalized.
To further characterize the SH3 interaction in vivo, we examined the localization of a GFP-tagged Prk1p kinase that was altered in its Abp1-SH3-binding motif in the C-terminal region. The cells were transformed with a plasmid encoding a GFP-PRK1 gene fusion in which the four residues ( 752 PPPK 755 ) had been changed into alanines. As a result, the Prk1p mutated at the polyproline site is no longer localized to the cortical patches (Fig. 3, lower panel). These results identify the SH3 domain of Abp1p and the polyproline motifs of Prk1p and Ark1p as essential regions for the proper cellular localization of both kinases to the cortical actin patches. Three-dimensional Structure of the Abp1-SH3 Domain-The interaction of a standard SH3 ligand and its receptor domain can be schematically split into two contributions. The first is rather nonspecific and derives from PXXP contacting the two hydrophobic pockets, while the second derives from further often electrostatic interactions in the so-called specificity pocket. As outlined above, the SH3 domain of Abp1p displays a peculiar recognition specificity, and differently from most SH3 domains described so far, it does not bind tightly to any short class 1 or class 2 peptide that we have tested. Abp1-SH3 ligands are characterized by two positive side chains separated by seven of eight residues containing a PXXP motif. To understand the structural basis of this novel specificity, we have determined the three-dimensional structure of the Abp1-SH3 domain at 1.3-Å resolution. The high resolution provides a detailed model in which double conformations and numerous solvent molecules could be determined accurately (Fig. 4A). The overall structure is similar to other SH3 domains consisting of a compact ␤-barrel of five anti-parallel ␤-strands forming two orthogonal ␤-sheets (Fig. 4B). At first sight, the SH3binding pockets, which normally host the PXXP-binding motif, do not reveal any peculiar characteristics that would readily explain the unusual binding properties of Abp1p. The residues that normally flank the hydrophobic cavities can be superimposed to the equivalent residues of the Abl and Src SH3 domains for instance (Fig. 5).
By considering the multiple alignment of the SH3 domains of known specificity and by concentrating on the residues that in the three-dimensional structure flank the ligand-binding pocket, the most striking peculiarity observed in Abp1-SH3 is Glu-6 that precedes the conserved tyrosine at position 7. The vast majority of SH3 domains have a hydrophobic residue at the Glu-6 position. Although this side chain is not in direct contact with the ligand in the SH3-peptide complexes of known structure, it is possible that the presence of a negative charge in an otherwise hydrophobic surface results in a decrease of the affinity for a "typical" PXXP peptide. Possibly because of the suboptimal interaction with the PXXP motif, the Abp1-SH3 physiological partners extend their contacts beyond the first hydrophobic pocket that is normally boxed by the aromatic ring of Tyr-7. Notably, just beyond the Tyr-7 side chain, the surface of Abp1-SH3 displays a second glutamate residue (Glu-21) in a position that in most SH3 domains is occupied by a positively charged residue. Because until now we have been not successful in our attempts to co-crystallize the Abp1-SH3 domain with its target peptides, we have designed site-directed mutants to test this model. In one mutant, Glu-6 was changed into Leu as found in most SH3 domains, and in a second one, Glu-21 was substituted by Lys.
When these mutant domains were assayed in pull down experiments, they proved to be at least as efficient as wild type in binding to Prk1p (data not shown). This result proves that neither of the residues tested plays a major role in ligand binding. However, consistent with the model, the mutant G6L has acquired the ability to bind to a classical (class 1) peptide, LSSRPLPTLPSP, thus pointing to the important role of the side chain at the position corresponding to Glu-6 in shaping the typical SH3-binding pocket (Fig. 6C). DISCUSSION The large scale two-hybrid screening experiments have failed to confirm the characterized interactions of Abp1p with Rvs167p and Srv2p (11,12,38) and have revealed a new putative partner of Abp1p, Yor284wp (39,40). However Yor284wp does not contain proline-rich regions that are typical signatures for SH3 domain targets.
In this study, by combining yeast two-hybrid screenings and selections of protein fragments displayed on filamentous phage capsids, we have identified six new ligands of the Abp1-SH3 domain that are potentially physiologically relevant. These include the two Ser/Thr kinases, Ark1p and Prk1p. By mutating either a conserved tryptophan on the Abp1-SH3-binding surface or the polyproline target peptides near the C termini of the two kinases, we have shown that this SH3-mediated interaction is responsible for the proper localization of both kinases to actin cortical patches.
Prk1p was suggested to be a negative regulator of endocytosis by several lines of evidence. In vivo, a loss of Prk1p activity suppresses the effects of a loss of function mutation in Pan1p, the yeast homologue of Eps15. By phosphorylating Pan1p on threonine residues, Prk1p hinders the interaction of Pan1p with the endocytosis protein, End3p (41). The formation of the heterotrimeric complex Sla1p-End3p-Pan1p is required for proper cortical actin organization and endocytosis (2,42).
In our two-hybrid screen using Abp1-SH3 as a bait, we also selected the unconventional (type I) myosin Myo5p and the product of an open reading frame of uncharacterized function, Yir003wp. The interaction with Myo5p has the potential to occur in vivo, because this motor protein co-localizes with Abp1p in cortical patches. Another protein that is implicated in endocytosis, the yeast synaptojanin Inp52p/Sjl2p, was also found to bind to Abp1p-SH3 in our screenings. The interaction we have characterized between Abp1p and Inp52p provides a possible mechanism for the localization of Inp52p to actin patches following hyperhosmotic stress (43). Finally the panning of phage-displayed genomic libraries has identified the product of the reading frame Ynl094wp as a partner of Abp1p. Collectively, the newly discovered Abp1p ligands together with the information already available on their function suggest a large web of SH3-mediated interactions that regulate the dynamic assembly of actin and link it to the endocytic process.
Abp1p contains three acidic repeats containing the conserved DDW motif that was shown to mediate the association of cortactin, WASP, and Myo3p to the Arp2⅐Arp3 complex (44,45). One of these motifs is solvent-exposed in the N-Src loop of the SH3 domain. However, our selection experiments have not revealed any Abp1-SH3 ligand that is a component of the Arp2⅐Arp3 complex. Furthermore the full-length protein Arc40p, a component of the Arp2⅐Arp3 complex that binds the acidic motif at the tail of the Myo3p (44), could not bind to Abp1-SH3 in a two-hybrid assay or in an in vitro assay with purified proteins (data not shown). Recently, the two other acidic regions of Abp1p have been proven to be essential for Arp2⅐Arp3-binding and actin polymerization (46).
These data emphasize the role of the yeast Abp1p as an F-actin-associated link between protein complexes involved in actin polymerization and a number of proteins implicated during endocytosis. Moreover, Abp1p can recruit to the actin patches of regulatory kinases, which have an inhibiting role on endocytic proteins. By phosphorylating their targets, Abp1passociated Prk1p and Ark1p may cause a disruption of the endocytic machinery and allow vesicles to pinch off from the membrane and be transported away by actin polymerization. Abp1p in this manner could be coordinating the membrane fission and actin-promoted movement. Fig. 7 illustrates the protein network centered on the Abp1 protein.
Several proteins involved in endocytosis and cytoskeleton dynamics contain an SH3 domain (47). Recently, a number of reports have shown that the binding potential of the SH3 FIG. 6. Charge potential surface representation. A, Abp1p-SH3 domain. B, Sem5-SH3 domain. The color code from blue to red corresponds to the charge potential at the surface of the molecule from positive to negative, respectively. Compared with the charge distribution of Sem5-SH3, Abp1p-SH3 presents an additional negatively charged patch in the region opposite to the specificity pocket. The figure was produced using GRASP (50). C, ELISA. Two biotinilated peptides (10 M) were adsorbed to a microtiter well coated with 5 g/ml of streptavidin. 0.3 g of the wild type SH3 domain or of its mutated form fused to GST was soaked in the well, and domain binding was monitored by probing with polyclonal anti-GST goat serum and with anti-goat alkaline-phosphatase-conjugated antibody. The amino acid sequences of the peptides used in the assay are reported at the right of the histogram. The GST fusions of the Abl and SH3 domains are used as controls.
domain family cannot be restricted to peptides containing the PXXP signature (19 -21). The comparison of the sequences selected by Abp1-SH3 from two different phage-displayed peptide libraries reveals two consensus sequences, ϩPX(P/V/L) XXPXϩ or PXXPXRPXW(L/I). This divergence might reflect the differences in peptide length and peptide density in the two phage-displayed repertoires as well as the bias of the fixed PXXP sequence in the pentadecapeptide library. However, by combining the two consensus sequences, we were able to identify, on the sequence of the seven Abp1-SH3 protein partners, peptides that conform to an extended consensus. Most of the natural ligands of the Abp1-SH3 domain contain a peptide that can be aligned to the consensus ϩXXXPXXPXKPXXL (Fig. 1). In the case of Prk1p and Ark1p, we have experimentally proved that this motif indeed mediates the formation of the protein complex in vivo. This motif can be interpreted as a modified class 2 motif with N-and C-terminal extensions containing a positively charged and a hydrophobic side chain, respectively. The Myo5p and the Ynl094wp target peptides do not conform to this consensus but maintain the conserved KP dipeptide preceded by a long stretch of prolines. The SH3 domain family is one of the most numerous classes of protein recognition modules with more than 1000 different domains found in protein data bases. Collectively, these domains play an important role in the formation of the protein networks, which control a number of important cell functions. The ability to predict the preferred peptide ligand of any SH3 domain and consequently infer its physiological targets would represent a substantial step forward in our attempt to decipher the protein interaction network inside a cell directly from genomic data. Although some predictive algorithms have been proposed, they often fail when they are queried with domains that do not bind to typical class 1 or class 2 domains. For instance, we have not been able to predict the recognition specificity of the Abp1-SH3 domain by using the SPOT algorithm (48) or even recognize that its preferred ligand has an atypical sequence. This implies that the peculiarity of the Abp1-SH3 domain cannot be revealed by simple alignment of the domain sequences and comparison of the residues that were seen in contact with the target peptides in the complexes of the known structure. Indeed, by superimposing the high resolution three-dimensional structure of Abp1-SH3 with that of other SH3 domains, like Abl or Scr, it was not immediately apparent why the Abp1 domain should fail to recognize the typical class 1 or class 2 consensus sequences. The binding surface in the two hydrophobic pockets that normally host the PXXP motif are highly comparable (Fig.  5). When the different SH3 domains are represented as surfaces, colored according to charge (Fig. 6), the only difference that stands out is the negative charge of Glu-6 in a position that is occupied by an hydrophobic residue in most SH3 domains. We have shown that this residue, although it does not make contact with the ligand peptides in the SH3 complexes of known structure, contributes to the characteristics of the peptide-binding pockets. In fact, by changing Glu-6 into a Leu, we were able to modify the recognition specificity of the Abp1-SH3 domain and promote the recognition of a classical class 1 peptide (Fig. 6C). This observation provides an important clue to pinpoint those in a multiple alignment of SH3 domains that are not likely to be good receptors of typical PXXP peptides. According to our model, the hydrophobic pockets in Abp1-SH3 do not represent an optimized receptor for classical PXXP peptides. As a consequence, the interaction energy with a peptide of typical length is below a threshold of physiological relevance (10 -100 M). To make up for the extra energy, the natural ligands of Abp1-SH3 would extend their contacts beyond the typical SH3-binding surface. Our site-directed mutagenesis experiments exclude that the "anomalous" glutamate at position 21 of the Abp1-SH3 domain may be involved in binding to this extended peptide target.