A phage display-based method for determination of relative affinities of mutants. Application of the actin-binding motifs in thymosin beta 4 and the villin headpiece.

We propose phage display combined with enzyme-linked immunosorbent assay as a tool for the systematic analysis of protein-protein interactions by investigating the binding behavior of variants to a partner protein. Via enzyme-linked immunosorbent assay we determine both the amount of fusion protein presented at the phage surface and the amount of complex formed, the ratio of which is proportional to the affinity. Hence this method enables us to calculate the relative affinities of a large number of mutants. As model systems, we investigated actin-binding motifs conserved in a number of proteins binding monomeric or filamentous actin. The hexapeptide motifs LKKTET, present in thymosin beta4, and LKKEKG, present in the villin headpiece, were mutated, and the variants were analyzed. Study of the positional tolerance allows postulating that the motifs, although similar in primary structures adopt different conformations when bound to actin. In addition, our data show that the second and the fourth amino acid of the thymosin beta4 motif and the first three residues of the villin headpiece motif are most important for actin binding. The latter result challenges the charged crown hypothesis for the villin headpiece filamentous actin interaction.

As a consequence of various genome-sequencing projects, novel proteins are being identified, many of which may take part in new interactions. Efforts are going on to tackle the discovery of protein-protein interactions globally (1), but there is also need for novel methods for the analysis of these interactions, preferably easy, reliable, and high through-put techniques. Several means to probe amino acids that contribute to the binding of two proteins have been developed. X-ray crystallography and NMR may yield superior information about the interface of proteins at atomic resolution (2,3). However, both methods are intrinsically difficult (even for noncomplexed proteins) and time-consuming and require expensive and sophisticated equipment, as well as large amounts of biological material. Cross-linking of interacting proteins followed by mass spectroscopy or conventional sequence determination of covalently coupled peptide fragments has limited application (4 -6). Probing the accessible surface in protein complexes by deuterium exchange has also been employed (7). But by far the two most popular methods are the use of peptide mimetics either in solution (8) or on membranes (9,10) and especially site-directed mutagenesis (11,12). However, in the latter powerful technique to study functions of proteins or interactions between proteins, one often faces the difficulty of choosing the position and type of amino acid exchange to be introduced. To circumvent this, one may perform a saturation mutagenesis at several positions thought to be important for the interaction. Obviously this creates a new problem, i.e. screening a large number of mutants in a systematic way.
We propose to combine saturation mutagenesis, phage display and ELISA 1 for the systematic screening and quantification of protein-protein interactions. Phage display is traditionally used for the selection of stronger binding ligands against target molecules (13); here we employ it differently. We generate defined libraries of mutants and subsequently analyze all of the different recombinant phages using ELISA. In this ELISA we determine both the amount of fusion protein on the phage and the amount of complex formed. The ratio of both gives information about the binding strength of the different mutants. We apply this method to the presentation of thymosin ␤4, analyzing monomeric actin (G-actin) binding, and of the villin headpiece, probing both G-actin and filamentous actin (F-actin) binding.
We selected these actin-binding modules because for thymosin ␤4 the actin binding information is relatively well characterized, but its structure is poorly defined. By contrast the villin headpiece has a well resolved NMR structure, but information on those residues that interact with actin is still elusive. Thymosin ␤4 interacts with actin via residues in an ␣-helix and a conserved hexapeptide motif (LKKTET) (14 -16). NMR studies of thymosin ␤4 reveal no unique structure for this motif; however, it is evident that the motif must become structured upon binding actin (17)(18)(19). A seemingly related sequence (LKKEKG) is present in a set of C-terminal headpiece domains (15,16) implicated in F-actin binding (20). NMR of the villin headpiece showed that the first five amino acids of the motif are in a ␣-helical conformation (21,22), and some of these are part of a charged crown suggested to be important for actin binding (20). Although both modules appear to have analogous motifs in their primary structure, they display different specificities for G-and F-actin. This suggests that local structural changes in these actin-binding units govern the recognition of the different conformational states of actin.
Here we show that a combination of mutagenesis, phage display, and ELISA may be a powerful new tool to systematically investigate protein-protein interactions by evaluating the relative binding strengths of mutants. Next to mutational tolerance at a position, structural information on the bound conformation can be inferred from mutants displaying increased affinity. In addition, our results show that some of the charged crown residues in the villin headpiece, hypothesized to be important for actin interaction, are dispensable for actin binding.

EXPERIMENTAL PROCEDURES
Construction of the Libraries and Rescue of the Recombinant Phages-Enzymes and reagents for molecular cloning were purchased from New England Biolabs or Invitrogen and were used following the manufacturer's instructions. The oligonucleotides were synthesized by Eurogentec. The phagemid vector pCANTAB5E, the detection module, and the Escherichia coli strain TG1 were from Amersham Biosciences. The helper phage M13KO7 was from Promega. The Sequenase version 2.0 kit was from U.S. Biochemical Corp., and [␣-35 S]dATP was from ICN.
We used the phagemid pCANTAB5E harboring the wild type human thymosin ␤4 or human villin headpiece cDNA, inserted between the SfiI and NotI restriction sites, as starting material to create the libraries. We constructed this library by independently mutating each of the six codons of both hexapeptide motifs into the 63 other possible ones, using overlap extension PCR mutagenesis (23) with synthetic oligonucleotides completely degenerated at the desired positions. The obtained PCR products were SfiI-and NotI-ligated in pCANTAB5E. The resulting phagemid contains under control of the lac promotor, the gene III (gIII) signal coding sequence, the thymosin ␤4 or the villin headpiece gene, the E-tag coding sequence, an amber codon (TAG), and the rest of the gIII coding region. The ligation was electroporated in TG1 cells. Phagemid DNA of individual transformants, was purified with Wizard TM minipreps (Promega) and sequenced. These isolated transformants were also used to produce phages, displaying thymosin ␤4 or villin headpiece variants, according to the protocol available from the company (Amersham Biosciences). We determined phage titers as described previously (24). Rescued phages were also used in the ELISAs described below.
Biotinylation of Actin on Cys 374 or Gln 41 -Actin was prepared from rabbit skeletal muscle (25) and isolated as calcium-G-actin by chromatography over Sephadex G-200 in G buffer (5 mM Tris-HCl, pH 7.7, 0.1 mM CaCl 2 , 0.2 mM ATP, 0.2 mM dithiothreitol, and 0.01% NaN 3 ). For biotinylation on Cys374, we applied actin to a NAP10 column (Amersham Biosciences) and eluted it with G buffer without dithiothreitol. We immediately derivatized actin, on Cys 374 , with 5-fold molar excess Immunopure Iodoacetyl-LC-Biotin (Pierce) in the dark at room temperature for 2 h. The biotinylated actin was dialyzed four times against G buffer with dithiothreitol to remove excess of labeling reagent. We biotinylated actin (1 mg) on Gln 41 using transglutaminase (1 unit; Sigma) as in Ref. 26 except that we used biotin cadaverin (Molecular Probes). In both cases we checked the incorporation of biotin by a Western blot using streptavidin-conjugated alkaline phosphatase (Sigma).
Theoretical Considerations Relating Absorbance to Relative Affinity-The set up of the method relies on monovalency of the phage display system and on the possibility to measure two absorbance values in an ELISA. The first criterion is fulfilled (see main text and Refs. 24 and 27). In wells, coated with E-tag antibody, one measures absorbance A 0 , which is proportional to the total amount of recombinant phages, ergo to the total amount of presented protein [Y tot ]. In wells coated with target protein (here G-or F-actin), one measures absorbance A, which is proportional to the amount of protein-protein complex [ for the presented protein is significantly lower than the K d value of the coated protein for the presented protein. We can then write: K d ϭ [X tot ](kA 0 Ϫ kA)/kA or (K d /[X tot ])ϩ 1 ϭ A 0 /A, i.e. the K d value is in a linear correlation with the ratio A 0 /A. Although our results, presented below, indicate that this correlation exists, the formula cannot be used to calculate K d values because [X tot ] represents a solution concentration, and we here use a solid phase assay. Coating of functional biotinylated monomeric actin or filamentous actin was reproducible.
In a previous paper (24), we used a linear correlation between the K d and the ratio of phage titer (T) over the absorbance (A) measured in a similar way as here, with a formula of the following form: (K d /[X tot ]) ϩ 1 ϭ T/ANk (where N is Avogadro's number and k is a proportionality factor). This formula is related to the one above because T/N is equal to kA 0 . Note that both formulae are also modified Scatchard equations.
The above derivation relies on an excess of functional target protein, coated in the wells, over the amount of phage presented protein. The amount of biotinylated actin in the neutravidin-coated wells (see below) was 1.4 pmol/well. This is 10 6 -fold higher than the amount of presented thymosin ␤4 (in general attomol/well). We performed a similar assay coating monomeric actin (5 g/100 l) directly to the wells. Although 29 pmol of actin was coated, the read-out in the ELISA, using the same phage preparations, consistently resulted in lower signals (not shown), indicating most of the directly coated actin is not functional. Therefore, and because biotinylation on Cys 374 of actin does not interfere with thymosin ␤4 interaction, 2 we used the biotinylated form in all of the thymosin ␤4-G-actin interaction assays. For the villin headpiece we used a previously documented procedure of F-actin coating (28).
ELISA for Thymosin ␤4 or Villin Headpiece-We coated microtiter plates (Nunc Maxisorp) either with 400 ng/100 l Immunopure neutravidin (Pierce) or 1 g/100 l of E-tag antibody (Amersham Biosciences) overnight at 4°C. Excess binding sites were blocked with 200 l of blocking buffer (2% solution of skimmed milk powder in phosphatebuffered saline). Neutravidin-coated wells were incubated with 1 g/ 100 l biotinylated actin for 1 h at room temperature. We washed the microtiter plate three times with phosphate-buffered saline containing 0.05% Tween 20 (washing buffer) and added 10 10 rescued phages in 200 l of blocking buffer to the biotin-actin-neutravidin wells or a 1:100 dilution of the phages to the E-tag antibody-coated wells. After 2 h of incubation at room temperature, we washed the plates four times with washing buffer and supplied the wells with 200 l of 1:2500 diluted anti-M13 antibody, conjugated with horseradish peroxidase (Amersham Biosciences) for 1 h at room temperature. We washed the wells three times with washing buffer, incubated the enzyme with the substrate 2Ј,2Ј-azinobis(3-ethylbenzothiazoline-6) sulfonic acid (Amersham Biosciences) and measured absorbance at 405 nm using a microtiter plate reader (Dynatech). For the villin headpiece mutant analysis, we coated 12 M polymerized actin overnight at 4°C (28). The ELISA was carried out as above with similar phage dilutions as used for thymosin ␤4. Based on the higher A 0 values, the amount of presented villin headpiece was a factor 2-3-fold higher than that of thymosin ␤4.
Other Binding Assays-The K d values of chemically synthesized thymosin ␤4 variants (24) were determined as in Ref. 15 except that we used phosphate-buffered saline, the buffer also used in the ELISA. The genes coding for wild type or mutant villin headpiece were recloned from the pCANTAB5E phagemid in pET11d (Novagen), with a new stop codon inserted at the 3Ј-end of the villin head-piece coding sequence and expressed in E. coli MC1061 containing pSCM26 (29). The proteins, thus lacking the E-tag sequence, were purified following the protocol in Ref. 22. The K d values for the villin headpiece mutants were determined in a sedimentation assay. We incubated 12 M F-actin with various molar ratios of the villin headpiece, or mutant, for 30 min at room temperature and sedimented the F-actin with an Airfuge (Beckman) for 15 min at 100,000 ϫ g. The supernatant was removed, and the F-actin pellet was washed with G buffer supplemented with 100 mM KCl and 1 mM Mg 2 Cl. Aliquots of the supernatant and pellet were analyzed using a 10% Tricine gel. After staining the gels, we used densitometric scanning to determine the amount of bound and free villin headpiece. We used these values in a Scatchard plot to calculate the K d . The K d for the WT-villin headpiece F-actin interaction is ϳ10-fold higher than previously reported using slightly different buffer conditions at 4°C (20). The K d of the E-tag antibody for the E-tag sequence was determined using a Biacore X. The antibody was coated to a CM5 chip (Biacore, Sweden) using amine coupling with 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride and N-hydroxysulfosuccinimide chemistry according to the manufacturer's instructions. Various concentrations of puri-fied thymosin ␤4, expressed from pCANTAB5E in HB2151 cells and thus containing the E-tag sequence, were passed over the sensorchip in HEPES-buffered saline (Biacore, Sweden). We derived the K d value (100 nM) from fitting the association and dissociation curves. Note that this is 100-fold lower than the K d for the actin thymosin ␤4 interaction in phosphate-buffered saline.

RESULTS
Evaluating the contribution of amino acids participating in the interaction of two proteins is often difficult and time-consuming, especially when three-dimensional structures of the individual proteins or of the complex are unknown. In the post-genome era, in which researchers employ bioinformatics to predict new potential interactions, there is a need for relatively fast and easy ways to investigate the binding of two proteins to confirm these predictions. In a previous study, we already explored the possibility of using a combination of three existing methods: PCR, phage display, and ELISA to study protein-protein interactions (24). In the present work we improved this method and applied it to the interaction of thymosin ␤4 with G-actin and of villin headpiece with G-and F-actin using libraries of mutants.
Creation and Completeness of the Libraries of Thymosin ␤4 and Villin Headpiece-Using double cycle PCR saturation mutagenesis reactions (23) on the thymosin ␤4 and the villin headpiece gene, inserted in the pCANTAB5E phagemid, we created libraries for each of the six positions of the hexapeptide motifs of both actin-binding modules. In the case of thymosin ␤4, the motif is known to be important for the interaction with actin (14,15); for the villin headpiece this has been proposed (22,30). For each of the 12 libraries we obtained at least 2-5 ϫ 10 6 transformants/g DNA in E. coli, far in excess over the expected 64 variants for each position in the motif. From each of the six thymosin ␤4 libraries, we isolated on average 70 clones and sequenced them to identify the created mutation. We recovered 103 of the 114 possible mutants at the amino acid level. From the results presented below for thymosin ␤4, it became evident that classes of mutants exist. For this reason we isolated from the six villin headpiece libraries enough mutants to get a member of each class allowing rapid screening of the villin headpiece-F-actin interaction.
An ELISA-based Method for Measuring Protein-Ligand Interactions-The method is schematically depicted in Fig. 1.   FIG. 1. A phage display strategy for studying protein-protein interactions. Wild type and/or mutant genes are ligated in the phagemid, here SfiI-and NotI-restricted pCANTAB5E, between the coding sequence of the gIIIp signal(s) and the coding sequence for the E-tag followed by the remainder of the gIIIp gene. The phagemid is electroporated in the appropriate E. coli strain. Individual clones are grown both for rescue and preparing the phagemid DNA for sequencing. In ELISA one measures, for the obtained recombinant phages, A 0 and A proportional to the total amount of fusion protein (via the E-tag) and the amount of formed complex with the target protein, respectively. The ratios A 0 /A are in linear correlation with the K d value, and [X tot ] is the total target protein concentration, here actin (for details see "Experimental Procedures").  (Table II). B, we did the same for the affinities of villin headpiece wild type and four purified recombinant mutants (K70R, K71E, E72S, and K73F) for F-actin. Linear correlations are 0.989 in A and 0.978 in B. The mutants were selected to represent a wide range of A 0 /A values. Basically, we insert the gene of interest (or a limited library of mutants) in the gIIIp gene coding for the M13 minor coat protein pIII. We opted for the phagemid pCANTAB5E, which has been engineered for monovalent display of heterologous proteins (27), to avoid complications arising from avidity effects during measurements of relative affinity. In addition, this vector carries the coding information for the E-tag sequence, the epitope for the anti-E-tag antibody, between the gene of interest and the gIIIp gene. The presence of a M13 origin of replication in pCANTAB5E permits the packaging of the phagemid DNA when bacteria are infected with a helper phage (e.g. M13KO7) during the so-called rescue. Because we purify by cloning, we avoid laborious purification steps (note that this also avoids possible background absorbance values in the ELISA by religated phages containing no insert but expressing the E-tag). Because the system is statistically monovalent (maximally one of 2,000 recombinant phages will be divalent) (24), the resulting recombinant phages only carry one copy of the heterologous protein and can be used in an ELISA for presentation of this protein to its coated partner. In this ELISA, the measured absorbance (A) is then proportional to the amount of protein complex formed. On the other hand, using the same phage preparations, the total amount of presented protein (A 0 ) can be measured because only the recombinant phages carry the E-tag sequence. As derived under "Experimental Procedures" (see also Fig. 1), the formula predicts that there is a linear correlation between the equilibrium dissociation constant of the interaction and the ratio (A 0 /A) of the number of recombinant phages over the amount of complex formed, provided that the amount of coated target protein is in large excess over the amount of presented protein. As an alternative for A 0 , one can determine the phage titer (T) (24), assessing the sum of the recombinant and nonrecombinant phages. It is, however, important to note that only 1% of the total phage population carries a recombinant protein (31). In a previous paper (24) we determined the T/A values of 18 mutants of Lys 18 in the actin binding motif of thymosin ␤4. Here, we measured the A and A 0 values of the same mutants in an ELISA using biotinylated actin and coated anti-E-tag antibody, respectively. We plotted T/A (24) versus A 0 /A values (this study) and observed a correlation of 0.86, indicating that both measurements yield essentially the same result independent of the manner in which the amount of recombinant protein is determined (data not shown). This is with the exception of those mutants where codon usage in E. coli influences the amount of expressed fusion protein. Indeed, for thymosin ␤4 mutants in which Lys 18 had been mutated to Arg, the T/A values were higher for mutants with a less frequently used codon, whereas their A 0 /A values were all very similar (Table  I). Similarly, A 0 /A values of K18Q mutants, where glutamine is encoded by TAG in the supE strain TG1 (this stop codon is only partially suppressed resulting in lower expression levels) or by the normal CAG codon, are comparable (83.7 and 82.6, respectively). Consequently, the method using titer over absorbance (T/A) can only be employed if one takes into account codon usage in E. coli. Thus, the purely ELISA-based method (measuring A and A 0 ) is superior and exclusively used in the analysis presented below.
Correlation between the K d and the A 0 /A Values of the Different Thymosin ␤4 and Villin Headpiece Variants, Proof of Principle-We predict a correlation between the K d and the A 0 /A ratio (see "Experimental Procedures"). To validate this, we determined the A 0 /A values of six phage displayed thymosin ␤4 variants (WT, K18A, K18R, K18Y, K18E, and K19E) for which the K d values of chemically synthesized counterparts were measured by a sequestration assay (15,24). We also measured the K d values of bacterially produced and purified villin headpiece wild type and mutants (K70R, K71D, E72S, and K73F) by a co-sedimentation assay and Scatchard analysis (data not shown). In both cases we plotted the K d values versus the corresponding A 0 /A values of the same phage presented mutants. There is a linear correlation of 0.99 for the thymosin ␤4-G-actin interaction and of 0.98 for the villin headpiece-Factin interaction (Fig. 2). These values suggest that despite the bias favoring dissociation inherent to washing steps in the ELISA, apparent affinities can be measured, illustrating the robustness of the method.
The Relative Affinities of the Different Thymosin ␤4 Variants Are Indicative of the Mutational Tolerance-We made recombinant phages of the 103 mutants of thymosin ␤4, isolated from the six different libraries. All of the variants could be displayed on the phage surface, except for the Met, Arg, and Asp mutants The values are the averages of two independent measurements (i.e. independent phage preparations and independent ELISAs); A and A 0 for each mutant were always measured in the same ELISA. Variation was maximally 5% for the lower absorbance values (this results in the 0.03 interval in Fig. 3) but was usually around 3% (the same applies to Figs. 4 and 5). ND, present in the library but not displayed; -, not recovered from the library. Leu  at position 17, for reasons we currently cannot explain. We performed the modified ELISA in duplicate, starting from two independent phage preparations and calculated the average A 0 /A values (Table II). Mutants with higher A 0 /A values have reduced affinities. In Fig. 3 these affinities are shown relative to wild type. In this representation a positive value indicates stronger binding than wild type, a negative value indicates weaker binding, and wild type is 0 (we consider values between ϩ0.03 and Ϫ0.03 as wild type-like interaction based on the maximal variation observed during our measurements). Two residues of the motif, Lys 18 and Thr 20 , appear to be essential because no variant has a significantly higher affinity than wild type (K18R, T20A, and T20Q have wild type-like affinity). At the other four positions of the motif, we do observe one or more mutants that have a higher affinity than wild type (see "Discussion"). If we take into account the number of mutants displaying reduced affinity and the extent of decreased binding, simple visual inspection of the graphs in Fig. 3 allows deriving the mutational tolerance at each position in the thymosin ␤4 hexapeptide motif. Lys 18 appears to be most important for the interaction with G-actin, followed by Thr 20 , Leu 17 , Glu 21 , Lys 19 , and Thr 22 . We note that this observed tolerance agrees well with the conservation of hexapeptide motifs in various ␤-thymosins (16,32,33), i.e. more mutations are allowed in the C-terminal half. In addition, most of the naturally occurring alterations result in a good binding thymosin ␤4 mutant in our assay.

The Profile of Relative Affinities of the Different Villin Headpiece Mutants Is Different for Interaction with F-and G-actin-
We prepared phages displaying the villin headpiece mutants and measured their relative affinities for F-actin using wild type villin headpiece as reference (Table III and Fig. 4). Because we only analyzed two-thirds of all possible mutants, we interpret our results cautiously. The first three residues of the motif appear more important because most of the analyzed mutants display significantly reduced binding, whereas at positions 72, 73, and 74 one or more mutants have a significantly higher affinity than wild type.
Although the villin headpiece is an F-actin-binding protein, we assayed the same recombinant phages for their relative affinity for G-actin in view of the observed competitive actin binding with thymosin ␤4 (15). For none of the recombinant phages could we observe binding when we used the same ELISA set up as for thymosin ␤4, i.e. with G-actin biotinylated on Cys 374 . Possibly, the villin headpiece-binding site on actin was occluded by the neutravidin moiety because an ELISA with G-actin biotinylated on Gln 41 , which is located at the opposite end of the actin monomer, yielded absorbance values (Table IV and Fig. 5). In general fewer mutants bind to G-actin, and many of the ones that do bind have strongly reduced affinities for monomeric actin consistent with the preference of this module for F-actin. This is especially relevant for mutations in the first three positions of the motif. However, similar to some of the thymosin ␤4 mutants, these results show that very weak interactions can be measured. Intriguingly, stronger G-actin-binding variants are mainly found at position 72, whereas this is at position 73 for F-actin (compare Figs. 4 and 5). DISCUSSION An Alternative Use for Phage Display-Phage display is a widely used technique for the selection of antibodies against antigens (34) or for searching protein variants with a higher affinity than wild type (13,35,36). We here employ it differently, i.e. not for the purpose of selecting stronger binding mutants but rather for systematically investigating the inter-action of two proteins. In combination with PCR mutagenesis and ELISA, it yields a reliable and easy way to determine relative affinities for mutants, even for those cases where partners interact weakly (we still obtained a positive signal for a low affinity interaction with a K d equal to ϳ0.2 mM). In solid phase assays, such as the ELISA used here, there may be bias toward slowly dissociating mutants. Indeed, within the incubation time, equilibrium between the k on and k off values is reached, but during subsequent washing steps the k off plays a major role in determining the amount of complex remaining on the plates. Nevertheless, the observed ratios of A 0 /A (representing the relative affinities in our solid phase assay) correlate very well with the corresponding K d values for the six chemically synthesized thymosin ␤4 variants and for the five recombinant villin headpiece variants tested (Fig. 2).
The relative affinities can be determined by two related methods. Either one calculates the ratio phage titer (T) over the absorbance (A) (24), or one determines the ratio A 0 over A (this study). The second approach is faster, and the correlation between the K d and the measured values no longer contains a proportionality factor that is dependent on a variety of instru-  Table II and the following formula log(A Mut /A 0 Mut ϫ A 0 WT /A WT ). As a result WT is 0, and WT-like activity (ϩ) is between Ϫ0.03 and ϩ0.03. Positive and negative values indicate increased and decreased affinity compared with wild type, respectively. The height of the bar is proportional to the increase or decrease in binding. These values were plotted versus the mutant amino acid (x axis) at each position in the hexapeptide motif (right y axis). Some mutants were not recovered (K18C, K19H, K19C, T20Y, T20D, T20M,  E21Y, E21N, E21M, T22E, and T22N), and some were not displayed (ND). The amino acids are given in the top and bottom panels only. mentation and environmental parameters. The most important improvement, however, is that the results become independent of the fraction of phages presenting the partner protein. An ELISA-based method for protein-ligand interaction was previously developed for the analysis of the interaction of zinc fingers with DNA (37). Similar to our strategy it relies on presenting protein in concentrations much lower than those of the target, and K d values can be determined by coating several concentrations of target ligand (preferably in 100-fold excess of the K d value). For the low affinity interactions studied here (1 M to 200 M), this would have required milligram amounts of actin, which in practice is difficult to achieve (in our assays we coat nanogram amounts). Because, in our method, we measure the amount of displayed partner protein, we avoid this problem.
Several other tools and methods have been developed to study protein-protein interactions, such as multi-use peptide libraries (8) and spot synthesis (9, 10), alanine (11,12) or cysteine (38) scanning, and two-hybrid analysis (39,40). Our technique has important advantages over these methods. Multi-use peptide libraries and spot methods may yield very similar, but less quantitative, scan information and are limited by the length of peptides that can be chemically synthesized (generally 15-20 amino acids). There may also be size limitations for phage display, especially when using the pVIII coat protein (41); however, presenting small proteins or single domains is usually no problem unless they precipitate as inclusion bodies or are not compatible for translocation through the membrane into the periplasm. In this respect, we recovered three thymosin ␤4 mutants, of which the fusion protein is not presented.
Alanine scanning is a powerful tool to study protein-protein interactions (12), but one may overlook residues participating in the interaction because this type of mutation may be rather neutral. In addition, it is not indicative for the tolerance of a functional residue (42). This is exemplified here in the thymosin ␤4-G-actin analysis where, with the exception of K18A, the mutation to alanine has no dramatic effect on binding, even not for Thr 20 , a residue important for the interaction. Also in twohybrid systems large numbers of mutants can be analyzed simultaneously, but certainly the method described here allows easier quantification.
Another advantage is that the heterologous mutant proteins are purified by cloning and do not require extensive purifica-tion prior to testing their binding characteristics. Recovering and analyzing all possible mutants from a library remains time-consuming. However, this may not always be necessary FIG. 4. Relative affinities of the villin headpiece mutants for F-actin. We used the formula in the legend to Fig. 3 to calculate the relative affinities from the A 0 /A values listed in Table III and employed the same representation as for Fig. 3. WT, WT-like activity (ϩ), mutants that are not displayed (ND), and displayed mutants with no observed binding (Ϫ) are indicated. because our results with thymosin ␤4 indicate that mutants appear in affinity classes. For the villin headpiece we tested two-thirds of all possible mutants, and these same groups can be distinguished, albeit that at each position of the profile is different. Consequently one can for each position analyze clones randomly, measure their relative affinity, and only sequence a few of each class. In this scenario, once optimized, the technique described here is relatively fast and can in principle be improved because several steps are adaptable to automation. As exemplified by the thymosin ␤4 results, our technique allows probing positional tolerance. As such, it can be applied as a screening tool prior to choosing positions that can be mutagenized when constructing libraries from which stronger binding variants will be selected. Probing Structural Information Using Libraries of Mutants-One of the reasons we embarked on this project was to use the binding characteristics obtained for the extensive set of mutants to distill structural information about the mutated regions. For the discussion below, we base our interpretation solely on mutants having similar or increased affinities compared with wild type, thereby assuming they are properly folded and stable.
First, comparing the profiles per position in thymosin ␤4 and villin headpiece allows to discriminate between residues participating in an electrostatic interaction (T␤4, Lys 18 ; VHP, Lys 70 and Lys 71 ) or not (T␤4, Lys 19 and Glu 20 ; VHP, Glu 72 and Lys 73 ). This is because we observe that some charged residues cannot be changed into any other amino acid without negatively influencing the affinity (with the exception of similarly charged side chains), and the most pronounced effect is observed for variants with an opposite charge. By contrast other positions are more tolerant for charge reversals. For thymosin ␤4 Lys 18 an electrostatic interaction is in agreement with previous studies (14,15), and for villin headpiece Lys 70 this is evident from the NMR structure because this residue is involved in making a buried salt bridge with Asp 39 (22). Intriguingly, in the latter case, substitution with Arg results in increased affinity. Perhaps the mutation to Arg yields a more stable domain enabling a better contact with F-actin. The other charged residues in villin headpiece are solvent exposed, and thus Lys 71 is probably involved in an ionic contact with F-actin. Surprisingly, charge reversal at position 73 leads to increased affinity. We suggest that the third lysine residue in the motif is close to the interface with actin and that substituting it with Glu creates a new (electrostatic) interaction. Given the observation that microinjected K73E mutant still binds F-actin in vivo (30), we predict that this happens with a higher affinity. FIG. 5. Relative affinities of the villin headpiece mutants for G-actin. We used the formula in the legend to Fig. 3 to calculate the relative affinities from the A 0 /A values listed in Table IV and employed the same representation as for Fig. 3. WT, WT-like activity (ϩ), mutants that are not displayed (ND), and displayed mutants with no observed binding (Ϫ) are indicated. Second, NMR experiments showed that the motif in noncomplexed thymosin ␤4 is structurally poorly defined (17,18), whereas algorithms predict a ␣-helical conformation We show that at certain positions in the motif, some substitutions yield better or wild type binders. If we correlate this with the secondary structure propensities of these mutant amino acids (43), we can speculate on the local conformation of the hexapeptide motif in the actin bound configuration. Interestingly at several of the positions glycine and proline substitutions (L17G, K19G, K19P, E21G, E21P, T22G, and T22P) are tolerated or result in better binding. Because these two amino acids are usually not present in ␣-helices or ␤-sheets, this may be an indication that the hexapeptide motif adopts a loop or ␤-turn structure in the thymosin ␤4-actin complex. These results correlate well with the NMR studies of thymosin ␤4 mutants in solution (i.e. not bound to actin), which show that binding requires proper termination of the N-terminal ␣-helix (residues 5-16) before the motif (19). Along the same lines, NMR data on villin headpiece shows a helical structure for the motif. Consistent with this is our observation that at every position in the motif (with the exception of the last residue), glycine or proline substitutions result in weaker binding than wild type. Thus, although the motifs in both proteins are rather similar in sequence, their secondary structures when bound to actin appear to be completely different, and this technique is capable of probing this. We also observed differences in binding to G-and F-actin for the villin headpiece mutants. This is most evident at the last four positions of the mutated region. Better G-actin-binding mutants at positions 71, 72, and 74 have wild type affinities for F-actin, and vice versa, stronger F-actin binders that are mainly found at position 73 behave similarly to wild type for G-actin binding. Possibly, these mutants probe conformational differences between G-actin and F-actin protomers.
Thus, using this technique one can obtain useful hints on structural information such as the type of interaction that occurs at the interface or the secondary structure required for recognition. Both are based on the tolerance of amino acids at particular positions and/or on those mutants displaying higher affinity.
Thymosin ␤4 and Villin Headpiece Interact Differently with Actin-We also show that thymosin ␤4 and the villin headpiece interact differently with G-actin. This is based on our observation that thymosin ␤4 does interact with G-actin linked via biotinylated Cys 374 on neutravidin, whereas the villin headpiece does not. This suggests that the villin headpiece faces actin close to this residue at the barbed end of the actin molecule and that the biotin-neutravidin moiety sterically hinders binding. Thymosin ␤4 is also proposed to contact the barbed end (5), but a model presented recently shows a location of thymosin ␤4 shifted away from the barbed end (44). Given the observed competition between thymosin ␤4 and the villin headpiece (15), our results suggest that the binding sites only partly overlap. In addition, the profiles of allowed substitutions in the C-terminal halves of the hexapeptide for both actin-binding modules are different, indicating that the G-actin binding capacity of the villin headpiece is not the result of a conformational switch to a thymosin ␤4-like structure.
Some Charged Crown Residues in Villin Headpiece Are Dispensable for F-actin Binding-We correlated our results with the prediction that charged residues (Lys 65 , Lys 71 , Glu 72 , Arg 37 , and the terminal carboxyl group of Phe 76 ) form a charged crown necessary for actin binding (20,22). Because the villin headpiece (WT and mutants) can be presented as a Cterminal fusion protein, the charge of the carboxyl function is not absolutely essential for actin interaction. Likewise, the charges of Glu 72 and Lys 73 appear not necessary for binding. Hardly any of the mutants at these positions displays a dramatically decreased affinity, and more curiously, substitutions to aromatic amino acids at the latter position result in an even higher affinity. Of the charged crown side chains studied here, only Lys 71 is important because the tolerance for mutation at this position is very low. We note that substituting RVDN in the protovillin headpiece domain by a charged crown-like sequence, KKEK, does not increase its affinity for F-actin (20). This is entirely consistent with our data, demonstrating that at least three charges in the crown are dispensable for F-actin binding.
Conclusion-We describe a technique based on phage display technology in combination with PCR mutagenesis and ELISA for the systematic investigation of protein-protein interactions. As a model we chose the mutagenesis of the hexapeptide motif in thymosin ␤4 and probed its interaction with G-actin. We also applied the method to a proposed actin-binding motif in the villin headpiece. The method allows for fast scanning of potential binding sites and determining relative affinities of many mutants. With this technique, one may distinguish between residues that are essential for a certain interaction and positions that are more tolerant of mutation.