Identification of two distinct peptide-binding pockets in the SH3 domain of human mixed-lineage kinase 3

Mixed-lineage kinase 3 (MLK3; also known as MAP3K11) is a Ser/Thr protein kinase widely expressed in normal and cancerous tissues, including brain, lung, liver, heart, and skeletal muscle tissues. Its Src homology 3 (SH3) domain has been implicated in MLK3 autoinhibition and interactions with other proteins, including those from viruses. The MLK3 SH3 domain contains a six-amino-acid insert corresponding to the n-Src insert, suggesting that MLK3 may bind additional peptides. Here, affinity selection of a phage-displayed combinatorial peptide library for MLK3's SH3 domain yielded a 13-mer peptide, designated “MLK3 SH3–interacting peptide” (MIP). Unlike most SH3 domain peptide ligands, MIP contained a single proline. The 1.2-Å crystal structure of the MIP-bound SH3 domain revealed that the peptide adopts a β-hairpin shape, and comparison with a 1.5-Å apo SH3 domain structure disclosed that the n-Src loop in SH3 undergoes an MIP-induced conformational change. A 1.5-Å structure of the MLK3 SH3 domain bound to a canonical proline-rich peptide from hepatitis C virus nonstructural 5A (NS5A) protein revealed that it and MIP bind the SH3 domain at two distinct sites, but biophysical analyses suggested that the two peptides compete with each other for SH3 binding. Moreover, SH3 domains of MLK1 and MLK4, but not MLK2, also bound MIP, suggesting that the MLK1–4 family may be differentially regulated through their SH3 domains. In summary, we have identified two distinct peptide-binding sites in the SH3 domain of MLK3, providing critical insights into mechanisms of ligand binding by the MLK family of kinases.

The human interactome has been estimated to contain ϳ650,000 interactions (1) with many protein-protein interac-tions facilitated by modular domains (2)(3)(4). Among different techniques used to map the interaction of domains, phage display has proven successful at mapping their peptide binding preferences (5)(6)(7)(8). This approach is based on the observation that peptides isolated through affinity selection of combinatorial peptide libraries usually bind to biologically relevant sites on the domain. Thus, it is possible to predict the candidate interacting partners of the domain in a particular target protein through a database search with the primary structure of the selected peptide ligands (9 -12). Furthermore, the isolated peptides have the potential to serve as inhibitors or activators of specific cellular interactions in proof-of-principle experiments for drug development (13)(14)(15).
A small number of SH3 domains, such as the Pex13p (31), p67 phox (32), and Vav (33) SH3 domains, bind noncanonical peptides via a different surface on the domain. More specifically, the SH3 domain of Pex13p (31) can independently bind two substrates with Pex14p interacting with the canonical site via PXXP motif and Pex5p interacting with a second noncanonical ligand-binding site in a proline-independent fashion. In the case of the p67 phox SH3-p47 phox interaction, both canonical and noncanonical binding surfaces facilitate interactions with respective PXXP and nonproline motifs present within the same p47 phox -derived peptide sequence; this bivalent interaction results in a high-affinity and specific protein-protein interaction (32). Finally, the N-terminal SH3 domain of Vav cannot bind Pro-rich ligands because its RT loop (located between ␤1 and ␤2 strands of SH3 domain) contains a tetraproline insertion, which blocks accessibility of the canonical PXXP binding pocket; instead, it uses a different surface to bind the C-terminal SH3 domain of Grb2 (33).
Mixed-lineage protein kinase 3 (MLK3) is widely expressed in a variety of normal and cancerous tissues, such as brain, lung, liver, heart, breast, kidney, pancreas, and skeletal muscle (34 -36). MLK3, also known as MAP3K11, is a serine/threonine protein kinase involved in regulating the JNK, p38, and ERK pathways (37)(38)(39)(40)(41)(42). It is composed of several modules: an SH3 domain, a kinase domain, tandem leucine zippers, a Cdc42/Rac interactive binding (CRIB) motif, several linker regions, and a proline-rich C-terminal tail (39,43). According to the BioGRID interaction database (44), MLK3 is involved in 36 unique interactions with several being mediated through the SH3 domain of MLK3. The SH3 domain has been implicated in the autoinhibition of the kinase domain through an intramolecular interaction via a nonconsensus single proline motif located between the leucine zipper and CRIB motif in the full-length MLK3 protein. Disruption of that intramolecular interaction results in increased activity of MLK3 (43). The SH3 domain of MLK3 also plays role in several other protein-protein interactions. The hepatitis C virus nonstructural 5A (HCV NS5A) protein was found to interact with the SH3 domain of MLK3 via a PXXPXR motif. Replacement of the two key prolines to alanine significantly decreases the interaction and NS5A's ability to block MLK3-induced apoptosis (45). The SH3 domain of MLK3 was also shown to interact with proline-rich regions in a hematopoietic progenitor kinase 1 (HPK1) (46,47) and a germline center kinase (GCK) (42). Both kinases regulate the JNK pathway in an MLK3-dependent manner (42,46).
Recently, MLK3 has been implicated in intracellular vesicle transport and motor neuron disease (48,49), and given that the MLK3 SH3 domain contains a six-amino-acid insert, corresponding in position to the n-Src loop (50), we wondered whether the MLK3 SH3 domain might have novel binding properties. Here, we report the discovery of a novel noncanonical binding site on the SH3 domain of MLK3 that is topographically different from previous publications on noncanonical recognition by SH3 domains (31-33). The newly identified binding surface is composed of an extended n-Src loop (located between ␤2 and ␤3 strands) of the MLK3 SH3 domain, which opens up to bind a peptide ligand, named "MLK3 SH3interacting peptide" (MIP), which lacks the canonical PXXP motif. Interestingly, when MIP binds to this site, it competes with binding of the HCV NS5A-derived peptide (PXXPXR) to the canonical binding pocket on the SH3 domain of MKL3. As the extended n-Src loop is conserved among the four members of the MLK1-4 protein family, we hypothesize that cellular interacting proteins exist that bind the SH3 domains of the MLK family of proteins in a unique MIP-like manner.

MIP isolated from a phage-displayed combinatorial peptide library
We utilized a GSH S-transferase (GST) fusion to the SH3 domain (amino acids 43-104) of human MLK3 (Fig. S1) for affinity selection of a phage-displayed combinatorial 12-mer peptide library (51). Following three rounds of affinity selection, we isolated a single clone displaying a 12-mer peptide, NH 2 -IRINPNGTWSRQ-COOH. As confirmed by phage ELISA, the clone bound to the wildtype (WT) form of the MLK3 SH3 domain (Fig. 1). Early experiments suggested that residues beyond the C terminus of the 12-mer peptide might contribute to binding, so we decided to characterize the binding properties of a 19-mer peptide, NH 2 -AIRINPNGTWSRQAET-VES-COOH, named "MIP." Truncation of MIP revealed that the entire 19-mer sequence is necessary for binding to GST-MLK3 SH3 (Fig. S2); for instance, deletion of Ala-1 (Fig. S2A) or deletion of several residues from the C terminus (Fig. S2B) resulted in lack of target binding.

Peptide ligands for the MLK3 SH3 domain
intermolecular interaction with GST-MLK3 SH3(43-104) (Fig.  2). Surprisingly, mutation of Pro-6 to Ala had no effect on binding ( Fig. 2), indicating a noncanonical form of peptide binding to the MLK3 SH3 domain. Alanine replacements of Gln-13 through Ser-19 also had no effect on SH3 binding (Fig. 2); however, deletion of any of the C-terminal residues compromised binding to the MLK3 SH3 domain in the phage-displayed format ( Fig. S2B) but not as a fusion protein (see Figs. 4 and S4). We interpret this as an indication that the C-terminal peptide region (amino acids 14 -19) of MIP serves as a linker necessary to facilitate binding in the phage-displayed format.

Architecture of the SH3 domain of MLK3
The sequence alignment of the SH3 domain of MLK3 and five other human SH3 domains (Src, Fyn, Eps8, p67 phox , and Grb2) revealed that the n-Src loop sequence of MLK3 SH3 contains an elongated insert of five residues (Fig. 4A). We solved the X-ray structure of the apo MLK3 SH3(44 -103) domain at 1.5-Å resolution and refined it to an R work/ R free of 18.6%/21.7% ( Table 1). The final model has all residues in the allowed region of the Ramachandran plot. The overall structure reveals a fivestranded antiparallel ␤-barrel, a characteristic of all known SH3

Peptide ligands for the MLK3 SH3 domain
domains, and three loops (RT, n-Src, and 3 10 helix), commonly involved in the recognition of both proline-rich and nonproline ligands (16 -18) (Fig. 4B). The residues that extend the MLK3 n-Src loop form a 3 10 helix, which is not commonly present in SH3 domains (Fig. 4B).

The structure of MLK3 SH3-MIP complex reveals a novel binding site on the SH3 domain
To determine the structure of the MLK3 SH3-MIP complex, we constructed a fusion protein MLK3 SH3(41-105)-MIP(1Ј-19Ј) where the prime denotes residues of the peptide. This con-

Peptide ligands for the MLK3 SH3 domain
struct was expressed in bacteria and purified as described under "Experimental procedures." Because the fusion protein elutes as a higher order oligomer from a gel-filtration column (Fig.  S3), we expected the interaction to occur between two separate molecules of SH3-peptide fusion rather than an intramolecular interaction. The structure of MLK3 SH3(41-105)-MIP(1Ј-19Ј) was solved to 1.2-Å resolution and refined to an R work/ R free of 13.6%/15.8% (Table 1). The final model has all residues in the allowed region of the Ramachandran plot. As shown in Fig. 4C, the crystallographic asymmetric unit (gray box) contains two protomers that are labeled Chain A (cyan, domain; blue, peptide) and Chain B (pink, domain; red, peptide). The interaction of the SH3 domain of MLK3 with MIP occurs between two protomers that are related by crystallographic symmetry. The first residue in the SH3 sequence (Tyr-41) and the C terminus of MIP (13Ј-19Ј) have no clear electron density and, therefore, were not modeled.
The MIP binds in a pocket formed between the n-Src loop and ␤-strands 2-4 ( Fig. 4D). As shown in Fig. 4E, multiple residues are involved in the binding of MIP to the SH3 domain of MLK3. Most notably, several hydrogen bonds contact Trp-10Ј in the MIP sequence. The backbone amide group of Trp-10Ј interacts with Asp-74, and the carbonyl group interacts with the backbone amide of Ala-76. Notably, both Asp-74 and Ala-76 are found within the MLK3-unique insert in n-Src loop of the SH3 domain. The Trp-10Ј indole nitrogen atom hydrogen bonds with the carbonyl group of Trp-83 located on the edge of ␤3 strand. Gly-8Ј and Arg-12Ј also contribute to hydrogen bonding; the carbonyl group of Gly-8Ј interacts with the hydroxyl group of Ser-72 in the SH3 domain, and the guanidine moiety of Arg-12Ј forms a salt bridge with the side-chain carboxylate of Asp-80.

Trp-10 and Arg-12 in MIP are critical for the intermolecular interaction with the MLK3 SH3 domain
To evaluate the importance of MIP residues Trp-10Ј and Arg-12Ј for binding to the MLK3 SH3 domain, we generated three fusion proteins composed of the SH3 domain linked via its C terminus to MIP(1-13)W10A, MIP(1-13)R12A, and the truncated version of the WT ligand, MIP(1-13). These purified proteins were analyzed by size-exclusion chromatography to determine their oligomeric state. Although the WT construct MLK3 SH3(41-105)-MIP(1-13) eluted as a higher order oligomer, both mutant constructs elute at a volume corresponding to the size of a monomeric fusion protein (Fig. S4). Mutation of either Trp-10Ј or Arg-12Ј to Ala resulted in a loss of binding to MLK3 SH3 (Fig. S4), consistent with alanine scanning results of phage-displayed MIP (Fig. 2).

Comparison of the apo and MIP-bound MLK3 SH3 structures reveals closed and open conformations of the n-Src loop modulated by the ligand
Comparison of the unbound and ligand-bound structures of MLK3 SH3 domain revealed a closed and open conformation of the n-Src loop, respectively (Fig. 5A). As modeled in Fig. 5B, the closed conformation of the loop precludes the peptide from the groove. To accommodate MIP as described previously, the n-Src loop undergoes a dramatic conformational change. In particular, residues Ile-77 and Ser-78, which participate in the 3 10 -helical structure, as well as the following Gly-79 undergo

Peptide ligands for the MLK3 SH3 domain
dramatic spatial movements required to facilitate binding of MIP to the SH3 domain of MLK3 as demonstrated by the change of positions of the ␣-carbons in these residues (Fig. 5C). The rest of the SH3 structure does not undergo any significant conformational changes upon binding with an r.m.s.d. of 0.41 over 40 residues (Fig. 5A).

MIP binds to a noncanonical binding site on the SH3 domain of MLK3 in proline-independent manner
Based on our structure, MIP fits into a novel binding site facilitated by a single n-Src loop and the edge of the ␤3 strand (Fig. 4D). In contrast, interactions with the canonical binding site commonly require involvement of three loops: RT loop, n-Src loop, and 3 10 helix loop (20,24,31). The SH3 domain of MLK3 was previously reported to interact with the HCV NS5A protein (UniProt entry W8GG88) (45). The authors mapped the interaction to a PXXPXR motif, KAPTPPPRRRR (aa 2320 -2331), present at the C terminus of the NS5A protein. To learn whether this peptide binds a site distinct from MIP, we fused the NS5A motif, APPIPPPR from a different HCV genotype (UniProt entry A0A076MER2; aa 2322-2333) previously crystallized with the SH3 domain of c-Src (Protein Data Bank (PDB) code 4QT7), to the SH3 domain of MLK3 and prepared crystals for X-ray diffraction. The crystals diffracted to 1.5 Å with an R work /R free of 18.8%/22.8%. Fig. 6A shows the overlay of the NS5A-bound structure compared with the MIP-bound structure. The NS5A sequence crystallized here is from a different genotype of HCV but carries the same PXXPXR motif that is conserved across all HCV genotypes (54,55). Alignment of both complexes (Fig. 6A) shows that the two binding sites are on the opposite faces of the SH3 domain.
The NS5A peptide interacts with the canonical polyprolinebinding site (Fig. 6D) by forming contacts with residues Asp-58 and Glu-59 of the RT loop, residues Asn-98 and Tyr-99 of the 3 10 helix, residues Asp-80 of the n-Src loop, and residue Trp-83 of the ␤3 strand from MLK3 SH3 (Fig. 6, B and C). Conversely, MIP fits into a novel binding site defined by the n-Src loop and the edge of the ␤3 strand (Figs. 6E and 4D).
Interestingly, Asp-80 and Trp-83 of the MLK3 SH3 domain participate in binding both peptides; however, in the case of Trp-83, the side chain is implicated in interactions with the NS5A peptide (Fig. 6B), whereas its backbone carbonyl interacts with MIP (Fig. 4E). The Asp-80 side chain is the only one involved in both interactions (Figs. 6, A and B, and 4E); it is oriented differently depending on which peptide is bound and interacts with arginine residues present in both peptides (Fig. 6A).

MIP and NS5A peptide bind the SH3 domain of MLK3 in competitive fashion
Because Asp-80 in the SH3 domain of MLK3 participates in binding both peptide ligands, competition experiments were performed to determine whether MIP and NS5A peptide can bind the SH3 domain in an exclusive or simultaneous manner. The GST-MLK3 SH3(43-104) protein was first incubated with increasing concentrations of soluble MIP peptide (AIRIN-PNGTWSRQAETVES) or NS5A peptide (KKAPTPPPRRRR-GGG-K) and then transferred to a 96-well microtiter plate coated with biotinylated MIP peptide (Fig. 7A) or biotinylated NS5A peptide (Fig. 7B), which was captured on the microtiter plate well surface via immobilized NeutrAvidin. It was seen that the free NS5A peptide competes with immobilized MIP for binding to GST-MLK3 SH3(43-104) with an IC 50 value of 12.9 Ϯ 0.6 M and that the free MIP peptide competes with immobilized NS5A for binding GST-MLK3 SH3 with an IC 50 value of 6.6 Ϯ 1.6 M. This supports the crystallographic data in which Arg-12Ј of MIP and Arg-8Ј of NS5A would suffer a steric clash if bound at the same time as they induce a different side chain orientation in Asp-80 of the MLK3 SH3 domain (Fig. 6A). Additionally, previously described experiments noted the importance of Arg-12Ј in the MIP peptide for overall binding (Fig. S4).

Binding of MIP is conserved across most of the SH3 domains of MLKs
MLK3 is one of the four members of the MLK subfamily (MLK1-4) (39,56). To predict the likelihood of MIP crossreacting with the rest of the subfamily, we generated protein homology models for SH3 domains of MLK1, MLK2, and MLK4 using the solved structure of MLK3 SH3 domain bound to MIP as a template (Fig. S6A). Although the primary sequences of the SH3 domains of MLK1, -2, and -4 are ϳ60% identical to that of MLK3 (MLK1, 60.0%; MLK2, 66.15%; MLK4, 60.0%) (Fig. S6B), structurally the domains are predicted to be very similar based on Swiss-Prot modeling (Fig.  S6A). Notably, four key residues involved in binding of MIP (Ser-72, Asp-74, Asp-80, and Trp-83) are conserved (Fig. S6B), and the overall predicted 3D structure of the n-Src loop and the

Peptide ligands for the MLK3 SH3 domain
rest of the molecule is highly preserved (Fig. S6A). To evaluate whether MIP and NS5A peptides can interact with SH3 domains of MLK1, MLK2, and MLK4, we constructed and purified GST fusions to each of these SH3 domains and examined their binding by ELISA. As can be seen in Fig. 8, the NS5A peptide could bind the GST-SH3 domain fusions of all (MLK1-4) purified proteins, whereas MIP bound the SH3 domains of MLK1, MLK3, and MLK4. Interestingly, the MIP peptide did not bind the SH3 domain of MLK2; comparison of the primary structures of MLK2 SH3 with the rest of the SH3 domains of the MLK subfamily (Fig. S6B) revealed a cysteine instead of a serine (MLK1) or alanine (MLK3 and MLK4) in the n-Src loop.

Discussion
To our knowledge, this is the first report presenting insights into mechanisms of ligand binding by the SH3 domain of MLK3. Previously, a crystal structure of another member of the MLK subfamily, the SH3 domain of MLK2 (PDB code 2RF0), was determined. However, no structure of an SH3-ligand complex has been available until now. Our studies revealed a noncanonical binding site on the surface of MLK3 SH3 that is localized at the opposite face of the molecule compared with the site that accepts proline-rich motifs (Fig. 6). An examination of the crystal structures of several SH3 domains bound to noncanonical peptide ligands revealed that all bind the same side of the SH3 domain as do canonical proline-rich peptides (18), suggesting that the MLK SH3 domain has a novel binding site primarily defined by the extended n-Src loop (Fig. 4). In addition, as the key residues involved in the interaction of the SH3 domain with the MIP ligand are conserved across MLK1-4 ( Fig. S6B), we suspected this noncanonical mode of SH3peptide interaction to be potentially conserved across the other members of the MLK subfamily. Indeed, we found that MIP
In this study, we also mapped the location of the canonical binding site on MLK3 SH3 by solving a structure of the SH3 domain of MLK3 bound to the NS5A-derived peptide (APPIP-PPR). Superposition of these two structures (Fig. 6) showed that the canonical PXXP and MIP-binding site are located at opposite faces of the molecule. However, based on the same overlay in Fig. 6, we also hypothesize that certain residues on the SH3 domain of MLK3 could be shared between those two different binding sites. MIP interaction with MLK3 SH3 requires involvement of Ser-72, Asp-74, Asp-80, and Trp-83 ( Fig. 6C  and 4E). Conversely, interaction of NS5A peptide with MLK3 SH3 seems to be dependent on Asp-58, Glu-59, Asp-80, Trp-83, Asn-98, and Tyr-99 (Fig. 6, B and C). Although the Trp-83 residue interactions are different for each peptide, the Asp-80 side chain is involved in interactions with arginines on both peptides that orient the side chain differently depending on the peptide bound (Fig. 6A). As both peptides require the carboxyl side chain of Asp-80 for binding, they could interact with the SH3 domain of MLK3 in competitive fashion as confirmed by competitive ELISA (Fig. 7). Additionally, mutation of a residue conserved among SH3 domains, Y52A, disrupts binding of the MLK3 SH3 domain to the MIP and NS5A peptides (data not shown). Finally, the shift in the n-Src loop required for MIP binding (Fig. 5) likely interferes with the binding of NS5A, further explaining the competition between the two peptides (Fig. 7).
These findings led us to consider earlier reports of proteinprotein interactions via the SH3 domain and specifically whether there is other evidence of noncanonical interactions. Previous reports suggest that two proteins, HPK1 and GCK, interact with the SH3 domain of MLK3. However, as they were assumed to interact with the SH3 domain via their PXXP motifs, only proline-rich stretches were subjected to further analysis (42,46). In the literature, only one known proteinprotein interaction has been reported to occur with the SH3 domain of MLK3 through a noncanonical PXXP motif. The SH3 domain of MLK3 can bind a region located between the leucine zipper and CRIB motif in an intramolecular fashion, resulting in autoinhibition of the kinase domain (43). However, the exact site of this interaction on the SH3 domain still remains to be determined. Our discovery of a noncanonical binding site on the surface of MLK3 SH3 that can accept motifs devoid of any proline residues could shed new light on the mechanism of ligand binding to the SH3 domain of MLK3. Based on findings of phage-displayed peptide ligands corresponding to cellular proteins (5), we suspect that a cellular protein exists that binds the MLK SH3 domain in a manner equivalent to MIP and are actively attempting to discover it. Finally, the presence of a unique binding site makes MLK3 SH3 even more attractive as a potential drug target, offering a new site for design or discovery of a specific inhibitor.

Peptide ligands for the MLK3 SH3 domain
For all constructs, the cells were initially grown at 37°C in 2ϫ YT medium (16 g of tryptone, 10 g of yeast extract, 5 g of NaCl/ liter), which was supplemented with 50 g/ml carbenicillin (CB) and 12.5 g/ml chloramphenicol, and then diluted (1:100) into fresh 2ϫ YT medium supplemented only with 50 g/ml CB. Protein expression was induced with 0.5 mM isopropyl ␤-Dthiogalactopyranoside (IPTG) when the culture reached an A 600 nm of 0.6, and cells were incubated overnight at 27°C at 250 rpm. Cells were harvested by centrifugation, and the pellets were resuspended in 1ϫ bind/wash buffer (Novagen), which was supplemented with cOmplete TM EDTA-free Protease Inhibitor Mixture (Roche Applied Science), and then lysed by sonication. Subsequent purification steps were carried out using GST⅐Bind TM resin (Novagen) following the manufacturer's protocol. Following the elution step, the proteins were exchanged into 20 mM Tris, 150 mM NaCl, pH 8.0, using Zeba TM Spin Desalting Columns (Thermo Fisher Scientific, Inc.) and then concentrated with Amicon Ultra centrifugal filters (EMD Millipore) and stored at Ϫ80°C. To confirm the purity, the samples were resolved by SDS-PAGE and detected with Coomassie Brilliant Blue staining.

Isolation of peptide ligand to the SH3 domain of MLK3
Screens for peptide ligands were performed following a protocol published previously (11). The purified target protein, GST-MLK3 SH3-WT, was prebound to GSH magnetic beads (Pierce, catalogue number 88821), and nonspecific binding sites were blocked with phosphate-buffered saline (PBS; 137 mM NaCl, 3 mM KCl, 8 mM Na 2 HPO 4 , 1.5 mM KH 2 PO 4 ) containing 3% (w/v) bovine serum albumin (BSA). To isolate potential peptide ligands, the targets were incubated with a 12-mer combinatorial phage-displayed peptide library, ANL7 (51). After three rounds of affinity selection, individual phage clones were isolated, and their binding was evaluated via phage ELISA (11). Positive clones were then sequenced.

Sequence analysis of phage-displayed MIP
To evaluate the phage-displayed MIP (NH 2 -AIRINP-NGTWSRQAETVES-COOH; insert (underlined) ϩ linker region), isolated from ANL7 library (51), both WT and modi-fied sequences (see Fig. S2) were cloned into a phage display vector, SAM (51), as an N-terminal fusion to the pIII capsid protein of M13 bacteriophage. All constructs were generated via Kunkel mutagenesis, described in detail elsewhere (11), and facilitated type 3 (pentavalent) display. Phage-displayed recombinants were evaluated for their binding to GST-MLK3 SH3 via phage ELISA as described elsewhere (11).

Alanine scanning of phage-displayed MLK3 MIP
All constructs were generated via Kunkel mutagenesis (11,57,58) and facilitated type 3 (pentavalent) display. The singlestranded DNA template was generated using a modified bacteriophage M13 vector (SAM) containing an amber stop codon (TAG) within the linker upstream of the N terminus of pIII (51). The oligonucleotides used for generation of all clones were synthesized by Integrated DNA Technologies, Inc. and are shown in Table S1. All constructs were verified by DNA sequencing, and phage clones were evaluated for their binding to the target (GST-MLK3 SH3) via phage ELISA (11). A truncated version of MIP (⌬A1) was used as a negative control (see Fig. S2).

Peptide synthesis
All peptides were synthesized at the Research Resource Center, University of Illinois at Chicago, and were of Ͼ90% purity as verified by MS. Two versions of MIP, unlabeled (AIRIN-PNGTWSRQAETVES) and biotinylated at the C terminus (AIRINPNGTWSRQAETVES-K-biotin), were used to determine the peptide's IC 50 . Two versions of NS5A-derived peptide (UniProt entry W8GG88; aa 2320 -2331), unlabeled (KKAPT-PPPRRRR-GGG) and biotinylated at the C terminus (KKAPT-PPPRRRR-GGG-K-biotin), were used to determine the peptide's IC 50 . The biotinylated NS5A and unlabeled MIP peptide were used in competition ELISA. The biotinylated MIP, MIP-R12A (AIRINPNGTWSAQAETVES-K-biotin; as negative control), and NS5A peptide were used in specificity ELISA to determine whether MIP and NS5A can cross-react with MLK1-4.

Peptide ligands for the MLK3 SH3 domain IC 50 determination for MIP binding to the MLK3 SH3 domain
To determine the IC 50 value by competition, the biotinylated form of MIP (AIRINPNGTWSRQAETVES-K-biotin; 100 l, Ϸ2 M) was captured in wells of a Nunc MaxiSorp TM flat-bottom 96-well microtiter plate (Thermo Fisher Scientific), which had been previously coated with NeutrAvidin TM (100 l, 7.5 g/ml; Thermo Fisher Scientific) and blocked with 3% skim milk in PBS. Separately, the GST-MLK3 SH3 fusion protein (5 g/ml) was preincubated (1-1.  Table S1). The subsequent insert contained MLK3 SH3(41-105) with the NS5A fragment (aa 2325-2332) at the C terminus. All coding regions were codon-optimized for E. coli expression. DNA fragments were cleaved with NdeI and BamHI-HF (New England Biolabs) and subsequently ligated into a modified version of the pET14b expression vector where the N-terminal His 6 tag is followed by a SUMO protease cleavage site. The resulting DNA constructs (verified by sequencing) were transformed into E. coli strain BL21 C41(DE3). Cells were grown at 37°C in 2ϫ YT medium, which was supplemented with either 50 g/ml CB or 100 g/ml ampicillin, protein expression was induced with 0.1-0.3 mM IPTG at an A 600 nm of 0.6 -0.7, and cells were cultured overnight at 22°C. Cells were harvested by centrifugation; washed with 200 mM KCl, 50 mM Tris, pH 7.4; and lysed by sonication in 25 mM Tris, pH 7.4, 200 mM KCl, 10 mM MgCl 2 , 10% glycerol, 1% Triton X-100, 1 mM phenylmethylsulfonyl fluoride (PMSF). Clarified supernatant was loaded onto a 5-ml HisTrap TM HP Ni-Sepharose column (GE Healthcare), and the column was washed with 25 mM Tris, pH 7.4, 500 mM NaCl supplemented with 10 and 25 mM imidazole. Protein was eluted in the same buffer, which was supplemented with 250 and 500 mM imidazole, and the fractions were pooled together. The His 6 -SUMO tag was cleaved with SUMO protease while dialyzing against 25 mM Tris, pH 7.4, 500 mM NaCl, 10 mM imidazole, and the tag was removed by loading the sample back onto a nickel column. Collected fractions containing purified protein were concentrated and injected onto an S-200 size-exclusion column (GE Healthcare) equilibrated with 25 mM Tris-HCl, pH 7.4, 500 mM NaCl. To confirm the purity, collected samples were analyzed by SDS-PAGE and detected with Coomassie Brilliant Blue staining. All purified proteins were concentrated and stored at Ϫ80°C.

Data collection and structure determination of apo, MIP-, and NS5A-bound MLK3 SH3
Diffraction data for MLK3 SH3-MIP, MLK3 SH3-NS5A, and apo MLK3 SH3 were obtained at the Life Sciences Collaborative Access Team ID beamline at the Advanced Photon Source, Argonne National Laboratory (wavelength, 0.979 Å; temperature, 100 K) (refer to Table 1 for data collection and refinement statistics). Data processing was executed using XDS (59). The structure of MLK3 SH3-MIP was solved by molecular replacement using MOLREP (60) with the available structure of the MAP3K10 SH3 domain (PBD code 2RF0) as the search model. Likewise, the apo MLK3 SH3 and MLK3 SH3-NS5A structures were solved using MOLREP using the MIP-bound structure as the model. Refinement was accomplished with REFMAC5 (61). Because the MLK3 SH3-NS5A data indicated the C2 space group with a unique ␤ angle close to 120°(i.e. 120.006°), we also analyzed the data using phenix.

Peptide ligands for the MLK3 SH3 domain
xtriage, which confirmed C2 as the correct choice of space group (62). All structural figures were generated with MacPyMOL (PyMOL TM Molecular Graphics System, version 1.6, Schrödinger, LLC).

Competitive ELISA to determine competition between MIP and NS5A
Competition ELISA was performed to determine whether MIP and NS5A peptides compete in binding the MLK3 SH3 domain and to calculate the relative binding strength (i.e. IC 50 ) of the NS5A peptide. Biotinylated NS5A peptide (KKAPTP-PPRRRR-GGG-K-biotin; 100 l, Ϸ 2 M) was captured in wells of a Nunc MaxiSorp flat-bottom 96-well plate, which had been coated with NeutrAvidin (100 l, 7.5 g/ml), and blocked with 3% skim milk in PBS. Separately, the GST-MLK3 SH3(43-104) domain fusion protein (5 g/ml) was preincubated (1-1.5 h) with increasing concentration (0.02-100 M) of unlabeled MIP as competitor and then allowed to interact for 1 h with biotinylated MIP prebound on the ELISA plate (Fig. S7). The assay was then completed as described above for IC 50 determination of MIP.

Specificity ELISA to determine MIP and NS5A peptide binding to other members of MLK subfamily
To determine whether MIP and NS5A can cross-react with other members of the MLK subfamily, SH3 domains of MLK1, MLK2, and MLK4 were purified as GST fusion proteins as described above. To determine their binding properties, 100 l (Ϸ2 M) of biotinylated peptides, MIP (AIRINPNGTWS-RQAETVES-K-biotin), MIP-R12A (AIRINPNGTWSAQAET-VES-K-biotin; as negative control), and NS5A (KKAPTP-PPRRRR-GGG-K-biotin), were first captured in wells of a Nunc MaxiSorp flat-bottom 96-well plate coated with NeutrAvidin (100 l, 7.5 g/ml) and then blocked with 2% skim milk in PBS. The GST-MLK1 SH3, GST-MLK2 SH3, GST-MLK2 SH3, and GST-MLK4 SH3 domain fusion proteins (0.15 M) were incubated with each of the target peptides, MIP, NS5A, or MIP-R12A, for 1-1.5 h. The rest of the assay was performed as described above for IC 50 determination of MIP.

Data resources
The coordinates for apo SH3, SH3-MIP complex, and SH3-NS5A complex have been deposited in the PDB under accession codes 5K28, 5K26, and 6AQB, respectively.