doi:10.1006/jmbi.2000.4035 available online at http://www.idealibrary.com on
J. Mol. Biol. (2000) 301, 1135±1147
High-density Mutagenesis by Combined DNA Shuffling and Phage Display to Assign Essential Amino Acid Residues in Protein-Protein Interactions: Application to Study Structure-Function of Plasminogen Activation Inhibitor 1 (PAI-I) A. Allart Stoop1, Laurent Jespers2, Ignace Lasters3, Eric Eldering1 and Hans Pannekoek1* 1
Department of Biochemistry Academic Medical Center University of Amsterdam Amsterdam, The Netherlands 2
Medical Research Council, Centre for Protein Engineering Hills Road, Cambridge, UK 3
AlgoNomics N.V., GentZwijnaarde, Belgium
The identi®cation of speci®c amino acid residues involved in protein-protein interaction is fundamental to understanding structure-function relationships. Supported by mathematical calculations, we designed a high-density mutagenesis procedure for the generation of a mutant library of which a limited number of random clones would suf®ce to exactly localize amino acid residues essential for a particular protein-protein interaction. This goal was achieved experimentally by consecutive cycles of DNA shuf¯ing, under error prone conditions, each followed by exposure of the target protein on the surface of phages to screen and select for correctly folded, functional mutants. To validate the procedure, human plasminogen activator inhibitor 1 (PAI-1) was chosen, because its 3D structure is known, many experimental tools are available and it may serve as a model protein for structure-function studies of serine proteinases and their inhibitors (serpins). After ®ve cycles of DNA shuf¯ing and selection for t-PA binding, analysis of 27 randomly picked clones revealed that PAI-1 mutants contained an average of 9.1 amino acid substitutions distributed over 114 different positions, which were preferentially located at the surface of the protein. This limited collection of mutant PAI-1 preparations contained multiple mutants defective in binding to three out of four tested anti-PAI-1 monoclonal antibodies. Alignment of the nucleotide sequence of defective clones permitted assignment of single dominant amino acid residues for binding to each monoclonal antibody. The importance of these residues was con®rmed by testing the properties of single point mutants. From the position of these amino acid residues in the 3D structure of PAI-1 and the effects of the corresponding monoclonal antibodies on t-PA-PAI-1 interaction, conclusions can be drawn with respect to this serpin-serine proteinase interaction. # 2000 Academic Press
*Corresponding author
Keywords: DNA shuf¯ing; phage display; PAI-1; protein-protein interactions; epitope mapping
Introduction Abbreviations used: ASA, accessible surface area; mAb, monoclonal antibody; PAI-1, plasminogen activator inhibitor 1; PBS, phosphate-buffered saline; serpin, serine proteinase inhibitor; SPOP, shuf¯ed proteins on phages; SPR, surface plasmon resonance; t-PA, tissue-type plasminogen activator; u-PA, urokinase-type plasminogen activator. E-mail address of the corresponding author:
[email protected] 0022-2836/00/051135±13 $35.00/0
The identi®cation of speci®c amino acid residues, involved in protein-protein interaction, is an essential step towards understanding the function and the mechanism of a target protein. At present, the analysis of consistent co-crystals is the best approach to unambiguously assign amino acid residues involved in interaction and to establish the nature of the interaction # 2000 Academic Press
1136 (reviewed by Janin & Chothia, 1990; and Jones & Thornton, 1996). However, co-crystallization of proteins is frequently unsuccessful either due to intrinsic properties of the complex or to inability to ®nd appropriate conditions for crystallization. Therefore, in most cases, either random, alanine scanning or site-directed mutagenesis is performed, followed by comparing the functional properties of the corresponding mutant protein with its wild-type counterpart (reviewed by Wells, 1991). Obviously, site-directed mutagenesis requires thorough prior knowledge of the structure and function of the protein of interest, extensive alanine scanning mutation analysis is labor intensive, whereas random mutagenesis demands screening of many individual clones. Consequently, to perform detailed protein-protein interaction studies in the absence of suitable cocrystals, a comprehensive procedure would be useful that requires analysis of only a limited number of clones and lacks the need for a 3D structure of the protein of interest. To meet this objective, the mutant library should: (i) consist of mutants, with multiple amino acid substitutions at the surface of the protein, likely to be hampered in a particular protein-protein interaction, but otherwise fully active; (ii) be devoid of mutants with amino acid alterations that distort proper folding of the protein; and (iii) lack introduced translation termination codons that result in truncated proteins. The basis to create such a library would be the powerful method of DNA shuf¯ing, that has been designed for molecular evolution of proteins in vitro (Stemmer, 1994a,b). DNA shuf¯ing is an iterative procedure to introduce and recombine mutations by homologous recombination, under error-prone PCR conditions. The major advantage of DNA shuf¯ing versus random mutagenesis is the possibility to remove deleterious mutations in a consecutive round, while mutations are only additive in a next round of random mutagenesis. So far, spectacular results have been reported using DNA shuf¯ing, e.g. in altering substrate speci®city (Zhang et al., 1997) or in enhancing enzymatic activity (Crameri et al., 1997) although only a minor fraction of the diversity generated by DNA shuf¯ing was assessed. Display of a repertoire of variant proteins on phages enables the simultaneous analysis of >109 proteins in a single assay (reviewed by Rodi & Makowski, 1999). Therefore, to exploit fully the diversity created by DNA shuf¯ing, we combined it with phage display to select functional proteins. Thus, we developed a stepwise procedure that yielded consecutive libraries of highly mutated but functionally active proteins, of which individual clones could be screened by negative selection to provide valuable information on amino acid residues involved in a certain protein-protein interaction. We designated this method ``shuf¯ed proteins on phages'' (SPOP). To validate SPOP human plasminogen activator inhibitor 1 (PAI-1) was chosen, because its 3D
Mapping Sites of Protein-Protein Interaction
structure is known (Mottonen et al., 1992; Sharp et al., 1999), many PAI-1-speci®c experimental tools are available and it may serve as a model protein for serine proteinase inhibitor (serpin) structurefunction studies. The serpin PAI-1 is the major physiological regulatory protein of the ®brinolytic system (reviewed by van Meijer & Pannekoek, 1995) and is also involved in cell adhesion and migration (reviewed by Loskutoff et al., 1999). PAI1 regulates the activity of its target serine proteinases tissue-type (t-PA) and urokinase-type plasminogen activator (u-PA) by rapidly forming irreversible equimolar complexes, in a reaction typi®ed by a suicide substrate mechanism (Wilczynska et al., 1995; Lawrence et al., 1995; van Meijer et al., 1997). To this end, the exposed reactive-center loop of PAI-1 serves as a pseudo-substrate for the cognate serine proteinase. The subsequent cleavage and entrapment of the proteinase induces a large conformational change, leading to insertion of the reactive-center loop into the central b-sheet A of PAI-1 (reviewed by Gils & Declerck, 1998). Insertion of the reactive-center loop also occurs spontaneously in the absence of a serine proteinase and yields the inactive, latent conformation (Mottonen et al., 1992). To probe the various conformations and functions of PAI-1, we have employed a set of well-characterized antihuman PAI-1 monoclonal antibodies (mAbs). These mAbs act either by preventing the interaction between PAI-1 and t-PA (BjoÈrquist et al., 1999), or by converting PAI-1 into a substrate (Debrock & Declerck, 1998), or by inducing a latent conformation of the inhibitor (Verhamme et al., 1999). In addition, anti-PAI-I mAbs are available that bind to PAI-1, but do not affect either one of the functions mentioned above (Keijer et al., 1991). Therefore, precise mapping of the interaction sites of these different mAbs may provide detailed insight into the mechanism of action of PAI-1. Here, we report the design, implementation and validation of the SPOP procedure, to assign amino acid residues of PAI-1 that are essential for the interaction with other proteins. For that purpose, a high density of mutations was created within the PAI-1 protein by ®ve consecutive rounds of DNA shuf¯ing, each followed by phage display of the mutant proteins and selection for functional PAI-1. Due to the selection pressure for functional PAI-1 proteins, mutations were preferentially located on the exposed surface of the protein. By analyzing only a limited number of randomly picked clones from a library of PAI-1 mutants, a direct assignment could be unambiguously made of single amino acid residues which are essential for the interaction with three out of four classes of antiPAI-I mAbs.
1137
Mapping Sites of Protein-Protein Interaction
Results Characteristics of phage-displayed PAI-1 mutant libraries We devised a procedure to create a PAI-1 library with a high density of mutations to perform detailed protein-protein interaction studies. The procedure, depicted in Figure 1, is based on the sequential use of error-prone DNA shuf¯ing, phage display of mutant libraries, and functional selection. Amino acid alterations should be preferentially selected at the surface of the target protein. The characteristics of the successive phage-displayed mutant PAI-1 libraries, and the procedure to select phages that properly express PAI-1 was assessed by analyzing the proteins before and after two rounds of t-PA binding. An antigen assay on ®lters was applied to determine the percentage of fully translated proteins, using an anti-PAI-1 mAb (MAI-12) that binds to an epitope in the reactive center which is located close to the carboxyl terminus of the protein (Keijer et al., 1991) (Table 1). Two rounds of stringent selection for PAI-1 mutants, that are fully able to bind t-PA, increases the percentage of properly translated PAI-1 clones for each cycle of DNA shuf¯ing from 20-50% to 9598 %, respectively. The ®fth cycle of DNA shuf¯ing results in a library of which only 5 % of the colonies react with anti-PAI-1 mAb MAI-12, whereas two selection rounds for t-PA binding yields a library that consists of about 70 % MAI-12-reactive colonies. Insertion of the corresponding full-length PAI-I cDNAs into the expression vector pMBL11-N allowed the subsequent synthesis and puri®cation of 27 randomly selected clones. Each of these PAI-1 preparations similarly inhibited the chromogenic activity of t-PA and formed SDS-stable
Figure 1. Schematic representation of consecutive steps in SPOP. Full-length PAI-1 cDNA is mutagenized by a round of DNA shuf¯ing, yielding PAI-1*. These mutants are inserted into the phagemid vector pComb3 and transformed into E. coli, resulting in a library PAI1*-pComb3. This PAI-1-mutant library is expressed on phage (f-PAI-1*-pComb3) and two rounds of stringent selection for t-PA binding is performed (t-PA-f-PAI-1*pComb3). This complete procedure is repeated an additional four times to obtain the desired mutation frequency.
t-PA/PAI-1 complexes, as determined by SDSPAGE (representative example in Figure 2). As a consequence, it is deduced that phage display selection is a powerful tool to select active PAI-1 mutants in a library of shuf¯ed PAI-1 proteins. Distribution of mutations in the mutant PAI-1 library DNA sequence analysis revealed that the mutation rate of every round of DNA shuf¯ing was 0.3-0.4 %. Therefore, ®ve rounds of DNA shuf¯ing should result in a ®nal library with 1.5-2.0 % nucleotide mutations, corresponding to an accumulation of approximately ten amino acid substitutions per PAI-1 protein. This expectation was born out by sequencing full-length PAI-1 cDNA (1137 bp), isolated from 27 independent clones. It was found that the mutant PAI-1 proteins contain an average of 9.1 amino acid substitutions that are encountered at 114 different positions in PAI-1 (Figure 3(a)). However, certain mutations at speci®c amino acid residues, and the relative high frequency at which they appear, are indicative for the presence of two different populations of mutations (Figure 3(b)). The ®rst one is composed of mutations that occur one to three times and ®ts within a regular Poisson distribution. The second population of mutations occurs with higher frequency (greater than three times) and obviously does not ®t Poisson distribution, which suggests a selective advantage for mutations at these positions. It was anticipated that the selection during panning was not only directed towards PAI-1 mutants that have unaltered capacity to bind to tPA, but simultaneously towards PAI-1 variants that display increased stability at the elevated temperatures, used during overnight production of phages. This explanation is supported by a determination of the half-life of 12 random PAI-1 mutants. This analysis revealed that the half-life of these PAI-1 preparations at 37 C is two- to fourfold longer than that of wild-type PAI-1 (Figure 4). We conclude that the population of most frequently observed mutations, that do not obey Poisson distribution, most likely contributes to stabilizing the active conformation. To verify the assumption that mutations were preferentially selected at exposed residues, we calculated the solvent accessibility of each amino acid Ê radius (as residue to a probe with a 1.4 A described in Materials and Methods), using the coordinates of the 3D structure of active PAI-1 (Sharp et al., 1999). The values for accessible surface area (ASA) were expressed as percentage of maximal ASA for each residue when positioned in an extended dipeptide. Based on their ASA, PAI-1 residues were then divided into three groups of comparable sizes: ``buried'' residues (0-5 % ASA), moderately accessible residues (5-35 % ASA) and highly accessible residues (>35 % ASA) (Miller et al., 1987). About threefold more positions were mutated in the group of highly accessible residues
1138
Mapping Sites of Protein-Protein Interaction
Table 1. Characterization of ®ve consecutive generations of mutant PAI-1 libraries before and after two rounds of t-PA selection Consecutive generations of PAI-1 mutant libraries 1 2 3 4 5
Library size (cfu)
MAI-12 reactive colonies in library (%)
MAI-12 reactive colonies after two rounds of t-PA panning (%)
1.0 107 5.0 106 3.0 106 4.0 107 1.0 107
25 50 30 20 5
98 98 95 95 70
Five consecutive generations of PAI-1 mutants were made by DNA shuf¯ing of PAI-1 cDNA that had been subjected to two rounds of phage display selection for t-PA binding. Indicated are colony forming units (cfu) after transformation of shuf¯ed pComb3/PAI-1 cDNA ligation mixture to Escherichia coli XL-1 blue cells. Before and after phage display selection for t-PA-binding PAI-1 mutants, the percentage of PAI-1 expressing clones was assayed by colony lifting, combined with Western blotting, as described in Materials and Methods. The percentages refer to the number of colonies that bind to anti-PAI-1 mAb MAI-12 compared to the total number of colonies.
(45 %) than in the group of buried residues (17 %), while one-third (33 %) of the moderately accessible residues were mutated. Comparison of the number of non-buried residues (235) with the number of calculated mutable residues (247) (see Materials and Methods) revealed similar numbers, although the corresponding residues are only partially overlapping. In conclusion, the results emphasize that our library preferentially contains mutations at exposed residues and, therefore, at potential protein-protein interaction sites. Application of mutant PAI-1 library to study protein-protein interaction A comprehensive analysis of mAbs against human growth hormone demonstrated that the number of dominant amino acid residues involved in antigen-antibody binding amounts to about three non-linear residues (Jin et al., 1992). We assumed that the same number is also valid for the interaction of PAI-1 with speci®c anti-PAI-1 mAbs and calculated that the probability (A) for our group of 27 mutants to harbor a mutation in a dominant amino acid residue involved in antibody
recognition would be 78 % (see Materials and Methods). To experimentally validate the utility of the constructed mutant PAI-1 library, we investigated binding to four different classes of anti-PAI-1 mAbs. The following mAbs were selected (Keijer et al., 1991): (i) I-201, a mAb which binds to PAI-1, but does not affect the interaction with t-PA; (ii) CLB-2C8, a mAb which induces the substrate behavior of PAI-1 (BjoÈrquist et al., 1999); (iii) M-5, a mAb which converts active PAI-1 into an inactive (latent) conformation (A.A. S. & H.P., unpublished data); and (iv) MAI-12, a mAb which directly blocks the interaction between PAI-1 and t-PA (Debrock & Declerck, 1997). The binding of each mutant PAI-1 preparation to the indicated mAbs was determined by surface plasmon resonance (SPR). Therefore, the mAbs were immobilized in separate channels on a sensor-chip surface of a BIAcore 2000 instrument. Subsequently, the mutant PAI-1 preparations were presented simultaneously to the different mAbs and binding was quanti®ed by a rise in the (arbitrary) response units (RU). After the injection period, the PAI-1 preparations are replaced by buffer and dissociation of the
Figure 2. Complex formation of representative mutant and wild-type PAI-I preparations with t-PA. Puri®ed mutant or wild-type PAI-1 preparations (1 mg) were analyzed prior to and after incubation with t-PA (2 mg) by non-reducing SDS-10 % PAGE, followed by Coomassie brilliant blue staining, as described in Materials and Methods. Numbers indicate randomly picked mutant PAI-1 preparations. wt, wild-type.
Mapping Sites of Protein-Protein Interaction
1139
Figure 3. Distribution of mutations found in 27, randomly picked, mutant PAI-1 clones after ®ve rounds of DNA shuf¯ing followed by two rounds of t-PA selection. (a) For every amino acid position in PAI-1 (379) the presence of a mutation in one or more of the 27 mutant PAI-1 clones is indicated. DNA sequence analysis for all mutant clones was performed as described in Materials and Methods, and compared to wild-type PAI-1. Within this collection of mutants, positions mutated more than ®ve times have been speci®ed. (b) Distribution of mutations according to frequency. Indicated are the number of amino acid residues that have been mutated with the indicated frequency, e.g. 74 amino acid residues have been mutated once. The 265 positions, that have not been mutated, are not shown.
protein from the mAb is visualized. Finally, before the injection of the subsequent preparation, the surface is regenerated and the RU signal returns to baseline levels. The analysis by SPR of a representative set of mutant PAI-1 preparations is shown in Figure 5 and demonstrates that af®nity for a single mAb could be reduced, while interaction with other mAbs is unaffected. From these experiments a number of conclusions can be drawn. First,
among the collection of 27 mutant PAI-1 preparations none displays a substantially altered binding to mAb MAI-12. It is conceivable that this property is due to a similar structural requirement for t-PA binding and mAb MAI-12 binding. Second, we found different mutant PAI-1 proteins that display a strongly reduced af®nity for at least one of the three remaining anti-PAI-1 mAbs tested. Collectively, we found that, among the 27 mutant PAI-1 preparations, ten preparations have a substantially reduced af®nity for either one of the four anti-PAI-1 mAbs: ®ve proteins with lower af®nity for I-201 (i.e. clones 7, 12, 40, 72 and B-13), ®ve with reduced binding to CLB-2C8 (i.e. 12, 16, 22, 29 and B-5) and three with reduced af®nity for M-5 (i.e. 12, 50 and B-5). We conclude that our ®rst object, namely to identify clones defective for a particular interaction of PAI-1 with another protein by analyzing a limited number of random clones, has been reached. Aligning of mutations for direct assignment of dominant PAI-1 amino acid residues for the interaction with various anti-PAI-1 monoclonal antibodies
Figure 4. Determination of half life of PAI-1 activity for a representative set of mutant PAI-1 preparations. Puri®ed mutant and wild-type PAI-1 preparations were incubated at 37 C for up to 25 hours. Aliquots were taken at the indicated times, and incubated with a ®xed amount of t-PA. Residual t-PA activity was determined in a chromogenic assay, as described in Materials and Methods. Symbols: ~, wild-type (t1/2 2.5 hours); !, clone 16 (t1/2 5.7 hours); ^, clone 61 (t1/2 7.5 hours); & , clone 24 (t1/2 7.5 hours); *, clone 7 (t1/2 10 hours).
Our second objective is to localize exactly the dominant amino acid residues in PAI-1 that are responsible for the interaction with another protein, e.g. a monoclonal antibody. Therefore, the amino acid sequences of the ten different PAI-1 cDNAs, that code for mutant PAI-1 proteins hampered in the interaction with either one of the antiPAI-1 mAbs I-201, CLB-2C8 and M-5, were aligned (Figure 6). Each mutant PAI-1 cDNA contained multiple nucleotide mutations (11-21), leading to between eight and 12 amino acid substitutions.
1140
Mapping Sites of Protein-Protein Interaction
Figure 5. SPR analysis of binding of three mutant and wild-type PAI1 preparations to three anti-PAI-1 mAbs. The anti-PAI-1 mAbs I-201 (red), CLB-2C8 (black) and M-5 (blue) were immobilized to a sensor chip of a BIAcoreTM 2000 instrument, as described in Materials and Methods. The baseline signal is obtained when buffer passes the chip surface. Upon presentation of puri®ed mutant PAI-1 preparations 7, 29, 50 and wildtype (wt) to these antibodies an increase in the response units (RU) demonstrates binding of the protein to the mAb. After association and dissociation (120 s) of the PAI-1 preparation and prior to presentation of the next sample, the chip surface was regenerated to baseline level by injection of 100 mH H3PO4. Wild-type PAI-1 is included as a positive control for binding to the anti-PAI-1 mAbs.
The alignment shows that speci®c PAI-1 mutants, which have in common a reduced af®nity for a particular anti-PAI-1 mAb, also share a speci®c mutation that is not present in clones with unaffected binding. All ®ve PAI-1 mutants hampered in binding to I-201 harbor an amino acid substitution of glutamine at position 56 (Q56), which is not found in any of the I-201 binding mutants. In a similar way, all three mutants with lower af®nity for M-5 have a common alteration at D181. Slightly more complex is the situation for CLB-2C8, ®ve clones were found to have a substantially decreased af®nity for CLB-2C8. Three out of these ®ve share a mutation at position K154 (statistically this chance is 6 %), while the remaining two clones have mutations in other dominant residues. Based on these results, it is likely that K154 is a dominant residue. However, no conclusions can be drawn on the other dominant residues, even though the two remaining clones do contain a cluster of mutations in the direct vicinity of K154. Of special interest is mutant clone 12, which contains all three indicated mutations and, consequently, does not bind any of the three mAbs. Taken together, these observations suggest that positions Q56, K154 and D 181 represent dominant residues for the interaction with mAbs 1-201, CLB-2C8 and M-5, respectively. To verify these deductions, we constructed, expressed and puri®ed the corresponding mutants, containing single mutations (denoted PAI-1 Q56R, PAI-1 K154D and PAI-1 D181A). These mutants were assayed by SPR to measure binding to immobilized anti-PAI-1 mAbs I-201, CLB-2C8 and M-5, respectively (Figure 7). Indeed, each single mutant displayed reduced binding to the expected mAb, therefore unambiguously con®rming that the substituted amino acid constitutes a dominant residue for the interaction with the respective mAb. By using the coordinates of the 3D structure of active PAI-1 (Sharp et al., 1999), it is clear that Q56, K154 and D181 are
located in different areas of the protein (Figure 8(a)). This observation is in agreement with SPR data, demonstrating that all three mAbs can simultaneously bind PAI-1 (Figure 8(b)). Finally, the data con®rm that an alignment of altered amino acids of PAI-1 variants, that share the property of defective interaction with another protein, allows direct identi®cation of dominant residues.
Discussion Here, we describe the design and experimental validation of a method, SPOP, which is intended to delineate the amino acid residues involved in speci®c protein-protein interactions. The method combines error-prone DNA shuf¯ing, phage display and selection for functionality of mutant proteins. The iterative character of this combination makes it possible to stepwise increase the mutation level in the target protein, in order to create mutant libraries that harbor the desired high density of altered amino acid residues. In addition, the number of clones tested from this library can be increased. Hence, the method can be extended along these lines to obtain an even larger and systematic set of mutations. Nevertheless, two major requirements have to be met for SPOP: (i) the target protein must be functionally displayed on phage. Although not all proteins can be displayed on phage, there are multiple successful reports that describe the display of a wide variety of proteins on phage (e.g. enzymes, SH-3 domains and antibodies; Kay & Hoess, 1996), (ii) A stringent selection for proper protein function of the target protein must be available. Once these requirements are met then the generation of a high density of mutations in the target protein should be feasible. The application of phage display to determine the binding epitopes for antibodies has been described, previously (Jespers et al., 1997). However, in that
Mapping Sites of Protein-Protein Interaction
1141
Figure 6. Mapping of antibody-binding sites by consensus analysis. DNA sequences of ten mutant PAI-I cDNAs that encode proteins which share reduced af®nity for the same anti-PAI-1 mAb were aligned. Search for consensus amino acid substitutions (indicated with x) in mutants with impaired binding revealed positions that are solely altered in these variants (depicted with boxes). Of special interest is clone 12, which has reduced af®nity for all three tested antibodies. None of these mutations were found in any of the clones with unaffected anti-PAI-1 mAb binding.
study only a single round of mutagenesis was applied, in combination with a negative selection. In contrast, our ultimate objective was to develop a high density of mutations in PAI-1. Therefore, comparison of only a limited number of clones would be necessary to assign amino acid residues that are crucial for the interaction with a PAI-1binding protein (e.g. an anti-PAI-1 mAb). This aim was born out since the analysis of PAI-I protein preparations, puri®ed from 27 randomly picked colonies, demonstrated that ten out of 27 preparations were severely hampered in the interaction with three different anti-PAI-1 mAbs. A straightforward alignment of the mutated positions of variants, which do not bind to one of these mAbs,
suf®ced for the unambiguous assignment of single, dominant amino acid residues. Selection of highly mutated PAI-1 variants that still interact with t-PA is thus feasible and therefore we assume that the mutations do not disturb the overall folding of the protein. We anticipated that mutations would preferentially be located at the surface of the protein, which could be veri®ed with the available 3D structure of the protein (Sharp et al., 1999). By a limited analysis of only 27 mutant PAI-1 cDNAs, we detected substitutions at 114 amino acid residues, predominantly located on the surface of the protein. Most striking are mutations at 45 % of all highly accessible residues. However, no mutations were found at highly accessible
Figure 7. SPR analysis of binding of single point mutant and wild-type PAI-1 protein preparations to three antiPAI-1 mAbs. The anti-PAI-1 MoAbs I-201 (red), CLB-2C8 (black) and M-5 (blue) were immobilized to a sensor chip of a BIAcoreTM 2000 instrument, as described in Materials and Methods. Binding to the single PAI-1 mutants Q56R, K154D and D181A is determined by an increase in response units (RU) upon presentation of the mutant. Regeneration of the signal to baseline level is performed by injection of 100 mM H3PO4. Wild-type PAI-1 is included as a positive control for binding to the anti-PAI-1 mAbs.
1142
Mapping Sites of Protein-Protein Interaction
Figure 8. Localization of dominant amino acid residues on PAI-1 for interaction with the anti-PAI-1 mAbs I-201, CLB-2C8 and M-5. (a) Ribbon diagram of the structure of active PAI-1 (Sharp et al.. 1999) in which the dominant residues for I-201 (Q56), CLB-2C8 (K154) and M-5 (D181) are indicated by the arrows. (b) SPR analysis demonstrating that the mAbs I-201, CLB-2C8 and M-5 can simultaneously bind PAI-1. The anti-PAI-1 mAb I-201 was immobilized to the sensor chip surface, subsequently wild-type PAI-1 was presented and bound as shown by a rise in response units (RU). The rise in RU seen upon presentation of both CLB-2C8 and M-5 demonstrate that the binding sites for the three mAbs do not overlap.
positions known to be vital for the interaction between t-PA and PAI-1, notably, in the reactive center loop at P1 (Arg347), P40 (Glu350) and P50 (Glu351) (Huber & Carrell, 1989; Madison et al., 1990; van Meijer et al., 1996), in the gate region 240 to 246 (Tucker et al., 1995), and at Glu283 in bstrand 6A (Sharp et al., 1999). These ®ndings stress the necessity of these areas for PAI-1-t-PA interaction and prompt further research into the importance of especially the gate region for PAI-1 function. Overall, SPOP effectively introduces mutations at tolerable and accessible positions in PAI-1, while conserving essential amino acid residues for the reaction with t-PA as well as residues indispensable for structural integrity. The distribution of the mutations in the mutant PAI-1 library into two populations stresses the effect of performing consecutive rounds of DNA shuf¯ing in which multiple parents are used (Moore et al., 1997). For a given position in PAI-1, each round of shuf¯ing combines a mutant sequence with wild-type sequence at that position. Since the wild-type sequence is by far more abundant than an a-speci®c mutation, the recombination event will preferentially lead to restoration of the wild-type sequence, and in that sense be considered a back-cross event. However except for newly introduced mutations, the overall mutation rate should stay the same during the DNA shuf¯ing process. Therefore, mutations present in the parent cDNAs at a higher frequency will have more chance of being incorporated into the off-
spring and will, eventually, accelerate the evolution of the protein toward the selection criteria. In the case of PAI-1, at least three explanations can be advanced for the preferential occurrence of a mutation in parental cDNA. First, the mutation may favor any of the intermediate steps that lead to a higher production of phages. Second, the mutation may increase the stability of the active conformation of PAI-1. Finally, the mutations may enhance the interaction with t-PA. However, the second-order rate constant of inhibition for t-PA by PAI-1 is virtually diffusion limited (2.0 107 Mÿ1 sÿ1) and therefore, further enhancement of this interaction is not feasible. Our preliminary analysis on the half life of mutant PAI-1 preparations indicates that mutations occur which increase the stability of the protein. Therefore, mutations found at a higher frequency than expected on the basis of a Poisson distribution are at least partially involved in increasing PAI-1 stability and have for that reason been enriched during consecutive rounds of DNA shuf¯ing and selection. The explicit data reported here are that the PAI-1 residues Q56, K154 and D181 constitute dominant residues for the epitopes that bind to the anti-PAI-1 mAbs I-201, CLB-2C8 and M-5, respectively. The signi®cance of these data for the structure and function of PAI-1 can be deduced using the 3D structure (Sharp et al., 1999) (Figure 8(a)). The implications for the inhibition of t-PA by PAI1 can be summarized as follows.
1143
Mapping Sites of Protein-Protein Interaction
First, no mutant PAI-1 preparation was found that displayed substantially altered binding to mAb MAI-12. However, the epitope for MAI-12 was reported to cover the reactive center of PAI-1 that is obviously required for stable interaction with t-PA (Keijer et al., 1991). Therefore, it is likely that selection for mutant PAI-1s with unaffected tPA binding, at least partially, coincides with preserving binding to mAb MAI-12. This observation may indicate that the same amino acid residue(s), which is (are) involved in t-PA inhibition, also constitutes a dominant part of the epitope for MAI-12. Second, residue Q56, which is essential for binding to mAb I-201, is located on helix C. mAb I-201 can bind all forms of PAI-1 (unreacted, complexed, cleaved and latent) and does not affect the PAI-1 - t-PA interaction. These ®ndings imply that none of the intermediate reactions between PAI-1 and t-PA, namely docking of the proteinase on the inhibitor, subsequent translocation to the central bsheet A and ®nal entrapment of the proteinase, actually occurs at or near Q56. This conclusion is in agreement with results reported by other investigators who used ¯uorescence resonance energy transfer to study the interaction between trypsin and a1-proteinase inhibitor (Stratikos & Gettins, 1999). In those studies it was observed that residue T85 of the inhibitor, located at an equivalent position as residue Q56 of PAI-1 (Huber & Carrell, 1989), is remote from the trypsin moiety during and after covalent complex formation. Third, in this report we established that residue K154, present in the large loop that connects helix F and b-strand 3A, is an essential part of the epitope for anti-PAI-1 mAb CLB-2C8. Actually, this observation contrasts our previous conclusion (Van Zonneveld et al., 1995). Initially, it had been deduced that the epitope is localized between residues 110 and 145, using immuno-precipitation of in vitro translated deletion mutants of PAI-I (Keijer et al., 1991). It should be mentioned that, by virtue of the design of those experiments, a straightforward immuno-precipitation of a peptide solely composed of residues 110 to 145 with mAb CLB2C8, was not feasible (Keijer et al., 1991). In a subsequent study, using phage-displayed fragments of PAI-1, the epitope was localized between residues 128 and 166 (van Zonneveld et al., 1995). Taking these two studies together, it was deduced that the epitope would be localized between amino acid residues 128 and 145. However, in view of the present data, we believe that the deduction is not justi®ed and that the epitope for mAb CLB-2C8 is located between residues 128 and 166, encompassing the dominant residue K154 detected in the present study. In addition, in this study two mutants (numbered 16 and B-5) display a reduced af®nity for mAb CLB-2C8, but lack the K154 mutation (Figure 6). However, these PAI-1 variants did contain other unique, altered amino acid residues, located between 128 and 166. As mentioned before, the anti-PAI-1 mAb CLB-2C8 converts PAI-1 to a substrate for t-PA. Furthermore, this mAb binds
equally well to active, latent and cleaved PAI-1 as well as to complexes of PAI-1 and t-PA (BjoÈrquist et al., 1997). These properties can be explained by assuming that binding of CLB-2C8 to residue K154 would prevent or delay insertion of the reactive center loop of the inhibitor into the central b-sheet A by steric hindrance. In accordance, it is conceivable that binding of the antibody to K154 would again be possible after covalent complex formation, subsequent to full insertion of the reactive center loop into the central b-sheet A. This explanation would oppose a model for the interaction between the inhibitor and the proteinase in which partial insertion of the reactive-center loop occurs (Wilczynska et al., 1997) and support a model of full loop insertion leading to positioning of the proteinase at the far end of the serpin, opposed to the docking site (Wright & Scarsdale, 1995; Stratikos & Gettins, 1999). Finally, a dominant epitope for the anti-PAI-1 mAb M-5 was mapped to D181, located at the top of the protein in a loop connecting b-strand 3A with b-strand 4C, which is located directly below the reactive center loop. It is curious that residue D181 is moderately accessible (25 % ASA) in active PAI-1 and highly accessible (580 % ASA) in latent PAI-1 (Sharp et al., 1999; Mottonen et al., 1992). Therefore, it is conceivable that upon binding mAb M-5 displaces and forces insertion of the reactive center loop into the central b-sheet A, resulting in an inactive (latent) PAI-1 protein. In aggregate, we conclude that SPOP is a wellsuited procedure to study protein-protein interactions. To validate the method, the dominant binding sites for three MoAbs on PAI-1 were determined. However, the method is not restricted to antibodies and, in the case of PAI-1, should also be applicable to map binding sites of physiological relevance, e.g. with vitronectin or the LDL-receptor related protein (LRP) (J. G. Neels, A.A.S. & H.P., unpublished results). Furthermore, we had the advantage of a 3D structure of PAI-1 to interpret the results. This was particularly valuable to correlate the binding sites of the mAbs with their corresponding effect on t-PA - PAI-1 interaction. Nevertheless, for SPOP a 3D structure is not a prerequisite, since the assignment of dominant residues was based on consensus analysis of alignedmutant sequences. In conclusion, once the target protein can be functionally displayed on phage and an adequate selection for functional mutants is available, SPOP provides a means to create highdensity mutagenesis. Subsequent screening for loss of function of individual clones and alignment of corresponding clones should reveal binding sites on the target protein for many different ligands.
Materials and Methods Materials The marine anti-human PAI-1 mAbs I-201 and M-5 were obtained from Monozyme (Charlottenlund,
1144
Mapping Sites of Protein-Protein Interaction
Denmark) whereas MAI-12 was from Biopool (UmeaÊ, Sweden). The murine anti-PAI-1 mAb CLB-2C8 and the murine anti-t-PA MoAb CLB-16 have been described previously (van Zonneveld et al., 1987; Keijer et al. 1991). Two-chain human melanoma t-PA was purchased from Biopool and the chromogenic t-PA substrate, CH3SO2-DHHT-Gly-Arg-para-nitroanilide (Pefachrome t-PA) was from Pentapharm (Basel, Switzerland). The Escherichia coli strain XL-1 Blue [F0 ] and the VCSM13 helper phages were obtained from Stratagene (La Jolla, CA, USA). The construction of the plasmid pMBL11-N has been described previously by van Meijer et al. (1996). Primers used were from Pharmacia Biotech (Roosendaal, the Netherlands). All DNA restriction and modifying enzymes, and bacterial growth media, were purchased from Gibco BRL (Paisley, UK) unless otherwise stated. Ampicillin, tetracyclin, and kanamycin were from Sigma (St. Louis, MO, USA). DNA sequencing was done with the Thermo Sequenase kit (Amersham Pharmacia Biotech, Uppsala, Sweden) and the ALF express automatic sequencer (Pharmacia Biotech). Analysis of DNA sequences was performed with the GCG Wisconsin package 10.0 (Genetics Computer Group, Madison, WI, USA).
more mutations are found is: (1 ÿ the probability of no mutations) 1 ÿ eÿl. Therefore, the expected number of positions that have one or more mutations is: W
1 ÿ eÿl f 362
1 ÿ eÿ126=
f 362 f 362 97 Numeric analysis gives f 0.6350 and consequently E(I) 0.6350 362 230. Based on these calculations, we assume that PAI-1 contains E E(I) E(II) 230 17 247 mutable positions. The probability (A) that the collection of 27 mutant PAI-1 clones contains one or more clones that have a mutation in at least one dominant residue for antibody binding, can be computed as follows. Since we select for well-folded molecules (selection for t-PA binding) it is assumed that any implied dominant residue is a mutable residue. When considering any dominant residue for the binding with a given antibody, this residue can be either of the above two types of mutatable positions. Hence, the probability that this dominant residue is a type I or a type II mutatable residue is respectively 230/247 and 17/247. The probability (a) that this dominant residue is substituted in one or more clones can be expressed as: a
230=247 eÿ126=230
17=247 1
Calculations Mathematical calculations were performed to support the theoretical considerations. PAI-1 consists of 379 amino acid residues: a de®ned number of residues can be mutated without loss of t-PA inhibitory activity. To calculate the percentage of mutable residues, we used the DNA sequence data obtained from a collection of 27 mutant PAI-1 clones (described here). This collection contains 244 mutations distributed at 114 different positions (W) (Figure 2(a)). These mutations can be divided in two types: (i) according to a Poisson distribution: 126 mutations in 97 positions where each position is substituted one to three times and (ii) a non-Poisson distribution: 118 mutations in only 17 positions where each position is substituted more than three times. The total number of mutatable positions (E) in PAI-1 (379) can be expressed as: E E
I E
II where E(I) and E(II) denote the number of mutatable positions for type I and type II positions, respectively. It is assumed that all positions (17) of type II (highly substituted) have been identi®ed in the collection of 27 clones. Consequently, the above expression for E can be rewritten as E E(I) 17. To estimate E(I), we express E(I) as a fraction (f) of total number positions in type I: E
I f
379 ÿ 17 f 362 The probability (P) that a mutable position of type I has been mutated in our collection is: P 126=
27 f 362 The numerator represents the total number of mutations in type I, the denominator is the number of mutable positions of type I in the 27 clones. The expected number of mutations E(I) to be found at any mutable position of type I is: l P 27 126=
f 362 The probability that, at a given type I position one or
where we have assumed, as before, that all positions of type II (high frequency of substitutions) have one or more substitutions in the set of 27 clones. We can now estimate A by the following formula: A 1 ÿ aS where S is the number of dominant residues for antibody recognitions. Using S 3 (Jin et al., 1992), it is found that A 78 %. To calculate the ASA of every amino acid residue in PAI-1, the coordinates of active PAI-1 were used (Sharp et al., 1999). This active PAI-I is a quadruple mutant with a prolonged half life, containing amino acid substitutions at: N150H, K154T, Q319L and M354I (Berkenpas et al., 1995). Calculations were performed using a probe with a Ê radius and the Brugel modeling program (Delhaise 1.4 A et al., 1984). The ASA was expressed as a percentage of maximal ASA, when an amino acid residue is placed in an extended dipeptide. Construction and screening of mutant PAI-1 libraries DNA shuf¯ing of PAI-1 was performed essentially as described (Stemmer, 1994a; Zhao & Arnold, 1997). Typical PCR-ampli®ed PAI cDNA was randomly digested by pancreatic DNase I in the presence of 1 mM MgCl2 (Promega, Madison, WI, USA) and fragments of 40-80 bp were isolated from 10 % (w/v) polyacrylamide gels. Fragments were reassembled in a PCR reaction using Taq I polymerase in the absence of primers (40 cycles: 94 C, 30 seconds; 55 C 30 seconds; 72 C, one minute ®ve seconds; per cycle), followed by PCR-ampli®cation with PAI-1 speci®c primers (forward: 50 -GTGCAGCTCGAG-CTGCACCATCCCCCATCCTAC-30 , reverse: 50 GCCACCACTAGTGGGTTCCATCACTTGGCCC-AT-30 ) to obtain full-length PAI-1 cDNA. After digestion with restriction enzymes XhoI and SpeI (cleavage sites are contained within the PAI-1 speci®c primers), the ``shuf¯ed'' PAI-1 cDNA was inserted into the phagemid pComb3 (Barbas et al., 1991) and used to transform E. coli XL-1 Blue cells. The resulting library of PAlI-1 mutants was
1145
Mapping Sites of Protein-Protein Interaction expressed on the surface of phages, after infection with helper phages and overnight growth at 30 C. Selection of non-truncated, properly folded active PAI-1 mutants, was performed as described (van Meijer et al., 1996). In brief, approximately 1011 phages of the pComb3/PAI-1 phage library were incubated for 1 h at 37 C with 0.3 nM t-PA in phosphate-buffered saline (PBS). PAI-1expressing phages that ef®ciently form complexes with t-PA were captured by binding to 3 mg of anti-t-PA MoAb CLB-16, coated to Nunc Maxisorb plates (Gibco BRL). Bound phages were extensively washed, eluted, ampli®ed and subjected to a second round of selection for binding to t-PA. Next, phagemids were isolated and t-PA- binding mutant PAl-1 cDNAs were ampli®ed, and subjected to a consecutive round of DNA shuf¯ing. The procedure of DNA shuf¯ing and subsequent phagedisplay selection for t-PA binders was performed ®ve times. DNA sequence analysis of mutant clones revealed that DNA shuf¯ing introduced an average mutation rate of 0.3-0.4 % per round.
Library screening for PAI-1 expression To determine the percentage of colonies, that express an intact PAI-1 protein, we performed a Western blotting-like screening (Persson et al., 1991). In brief, E. coli XL-1 Blue cells were infected with multiplicity of infection of 100-300 phages and spread on plates containing IPTG for optimal induction of protein synthesis. Colonies were transferred to nitrocellulose ®lters, the cells were lysed by saturated chloroform vapor, and treated with 1 U/ml pancreatic DNaseI in 150 mM NaCl, 50 mM Tris (pH 8.0), 5 mM MgCl2, 3 % (w/v) bovine serum albumin and 400 mg/ml lysozyme. The ®lters were incubated with the anti-PAI-1 MoAb MAI-12, which binds to the carboxyl terminus of PAI-1 (Keijer et al., 1991). Subsequently, the immune complexes were incubated with rabbit anti-mouse immunoglobulins, conjugated to alkaline phosphatase and, color development was monitored after NBT/BCIP (Sigma) treatment. After each incubation, the ®lters were extensively washed in PBS containing 0.5 % (v/v) Tween-20.
PAI-1 protein purification To characterize individual mutant PAI-1 proteins, the corresponding full-length PAI-1 cDNA was inserted into the expression vector pMBL11-N and PAI-1 was puri®ed as described (van Meijer et al., 1996). In brief, single colonies were grown for ten hours at 37 C in 2 YT medium, pelleted and suspended in 10 ml M9 medium, supplemented with 0.2 % (w/v) casaminoacids, 0.2 % (w/v) glucose, 2 mM MgSO4, 0.1 mM CaCl2 and 50 mg/ml ampicillin and, grown overnight at 30 C. Cells were pelleted, suspended in 1 ml 20 mM sodium acetate (pH 5.6), 200 mM NaCl, 0.01 % Tween-20, and lysed by sonication. Cellular debris was removed by centrifugation and the supernatant was incubated for 1 h at 4 C with CMSephadex C-50 beads. After extensive washing of the beads with sodium acetate buffer, PAI-1 was eluted from the beads by incubation with 20 mM sodium acetate (pH 5.6), 1 M NaCl and 0.01 % Tween-20. In this buffer, PAI-1 can be stored for several weeks at 4 C without loss of activity. The protein preparations consisted of 595 % pure PAI-1 as judged by SDS-PAGE and Coomassie brilliant blue staining.
Analysis of mutant PAI-1 proteins The t-PA-inhibitory activity of puri®ed PAI-1 mutants was assessed by two methods. First, serial dilutions of PAI-1 mutant preparations were incubated for 15 minutes at 37 C with 3.4 nM t-PA and the residual t-PA activity was determined. For that purpose, chromogenic substrate Pefachrome t-PA (®nal concentration 0.5 mM) was added and the absorbance at 405 nm was recorded continuously for one hour at 37 C, using a Titertek Twinreader (Flow Laboratories, Irvine, UK). Second. t-PA (2 mg) and PAI-1 mutant (1 mg) were incubated for ten minutes at room temperature in 20 ml PBS and the formation of complexes of t-PA and PAI-1 was analyzed by non-reducing SDS 10 % PAGE, followed by Coomassie Brilliant Blue staining of the protein bands. To determine the half-life of the various protein preparations, 100 nM mutant PAI-1 protein in PBS was incubated for 25 hours at 37 C. At regular time intervals sub-samples were taken, and residual PAI-1 activity was determined by the ®rst method described.
Surface plasmon resonance The binding of puri®ed PAI-1 mutants to speci®c murine anti-human PAI-1 MoAbs was determined by surface plasmon resonance (SPR), using a BIAcoreTM 2000 instrument (BIAcore AB, Uppsala, Sweden). Typically, antiPAI-1 MoAbs were coupled to a NHS/EDC-activated CM-5 sensor chip, using free amine groups, to yield approx. 3000 resonance units (RU). Subsequently, mutant PAI-1 proteins (90 nM) in HBS buffer (20 mM Hepes (pH 7.4), 150 mM NaCl, 3.4 MM EDTA and 0.005 % (v/v) P20) were injected for 90 seconds at a ¯ow rate of 20 ml/min and the interaction was monitored in real time. Regeneration of the sensor chip surface was done with 5 ml 100 mM H3PO4. All measurements were performed at 25 C.
Site-directed mutagenesis Overlap extension PCR was used to construct three, single residue mutant PAI-1 cDNAs (Higuchi et al., 1988). Therefore, the following ``mutagenic'' primers were used 50 -GAAACCCAGCGGCAGATTCAAGC-30 , 50 -CTTGCTTGGGGAAGGAGCCGTG-30 and 50 -ACCCT TCCCCGCCTCCAGCACC-30 , together with their corresponding complementary oligonucleotides and the mutants were designated Q56R, K154D and D181A, respectively. These mutagenic primers were combined with primers that are located at 50 and 30 end of the open reading frame of PAI-1. DNA sequence analysis con®rmed the introduction of the intended mutations and the absence of additional mutations.
Acknowledgments We thank Dr R. J. Read who kindly provided us with the 3D coordinates of active PAI-1 and Dr A.J.G. Horrevoets for critically reading the manuscript. This work was supported by grants from the European Union (grant no. ERB-BIO4-CT96-0389) and from the Dutch Heart Foundation (90.251 and M93.007).
1146
References Barbas, C. F., III, Kang, A. S., Lerner, R. A. & Benkovic, S. J. (1991). Assembly of combinatorial antibody libraries on phage surfaces: the gene III site. Proc. Natl Acad. Sci. USA, 88, 7978-7982. Berkenpas, M. B., Lawrence, D. A. & Ginsburg, D. (1995). Molecular evolution of plasminogen activator inhibitor-1 functional stability. EMBO J. 14, 2969-2977. BjoÈrquist, P., Ehnebom, J., Inghardt, T. & Deinum, J. (1997). Epitopes on PAI-1 important for binding to t-PA. Biochim. Biophys. Acta, 1341, 87-98. BjoÈrquist, P., Ehnebom, J. & Deinum, J. (1999). Protein movement during complex formation between t-PA and PAI-1. Biochim. Biophys. Acta, 1431, 24-29. Crameri, A., Dawes, G., Rodriguez, E., Jr, Silver, S. & Stemmer, W. P. (1997). Molecular evolution of an arsenate detoxi®cation pathway by DNA shuf¯ing. Nature Biotechnol. 15, 436-438. Debrock, S. & Declerck, P. J. (1997). Neutralization of plasminogen activator inhibitor-1 inhibitory properties: identi®cation of two different mechanisms. Biochim. Biophys. Acta, 1337, 257-266. Debrock, S. & Declerck, P. J. (1998). Identi®cation of a functional epitope in plasminogen activator inhibitor-1, not localized in the reactive center loop. Thromb. Haemost. 79, 597-601. Delhaise, P., Bardiaux, M. & Wodak, S. (1984). Interactive computer animation of macromolecules. J. Mol. Graph. 2, 103-106. Gils, A. & Declerck, P. J. (1998). Structure-function relationships in serpins: current concepts and controversies. Thromb. Haemost. 80, 531-541. Higuchi, R., Krummel, B. & Saiki, R. K. (1988). A general method of in vitro preparation and speci®c mutagenesis of DNA fragments: study of protein and DNA interactions. Nucl. Acids Res. 16, 73517367. Huber, R. & Carrell, R. W. (1989). Implications of the three dimensional structure of a1-antitrypsin for structure and function of serpins. Biochemistry, 28, 8951-8965. Janin, J. & Chothia, C. (1990). The structure of proteinprotein recognition sites. J. Biol. Chem. 265, 1602716030. Jespers, L., Jenne, S., Lasters, I. & Collen, D. (1997). Epitope mapping by negative selection of randomized antigen libraries displayed on ®lamentous phage. J. Mol. Biol. 269, 704-718. Jin, L., Fendly, B. M. & Wells, J. A. (1992). High resolution functional analysis of antibody-antigen interactions. J. Mol. Biol. 226, 851-865. Jones, S. & Thornton, J. M. (1996). Principles of proteinprotein interactions. Proc. Natl Acad. Sci. USA, 93, 13-20. Keijer, J., Linders, M., van Zonneveld, A. J., Ehrlich, H. J., de Boer, J. P. & Pannekoek, H. (1991). The interaction of plasminogen activator inhibitor 1 with plasminogen activators (tissue-type and urokinase-type) and ®brin: localization of interaction sites and physiologic relevance. Blood, 78, 401-409. Kay, B. K. & Hoess, R. H. (1996). Principles and applications of phage display. In Phage Display of Peptides and Proteins, a Laboratory Manual (Kay, B. K., Winter, J. & McCafferty, J., eds), p. 21, Academic Press, London. Lawrence, D. A., Ginsburg, D., Day, D. E., Berkenpas, M. B., Verhamme, I. M., Kvassman, J. O. & Shore,
Mapping Sites of Protein-Protein Interaction J. D. (1995). Serpin-protease complexes are trapped as stable acyl-enzyme intermediates. J. Biol. Chem. 270, 25309-25312. Loskutoff, D. J., Curriden, S. A., Hu, G. & Deng, G. (1999). Regulation of cell adhesion by PAI-1. APMIS, 107, 54-61. Madison, E. L., Goldsmith, E. J., Gething, M. J., Sambrook, J. F. & Gerard, R. D. (1990). Restoration of the serine protease-inhibitor interaction by protein engineering. J. Biol. Chem. 265, 21423-21426. Miller, S., Janin, J., Lesk, A. M. & Chothia, C. (1987). Interior and surface of monomeric proteins. J. Mol. Biol. 196, 641-656. Moore, J. C., Jin, H. M., Kuchner, O. & Arnold, F. H. (1997). Strategies for the in vitro evolution of protein function: enzyme evolution by random recombination of improved sequences. J. Mol. Biol. 272, 336-347. Mottonen, J., Strand, A., Symersky, J., Sweet, R. M., Danley, D. E., Geoghegan, K. F., Gerard, R. D. & Goldsmith, E. J. (1992). Structural basis of latency in plasminogen activator inhibitor-1. Nature, 355, 270273. Persson, M. A., Caothien, R. H. & Butron, D. R. (1991). Generation of diverse high-af®nity human monoclonal antibodies by repertoire cloning. Proc. Natl Acad. Sci. USA, 88, 2432-2436. Rodi, D. J. & Makowski, L. (1999). Phage-display technology - ®nding a needle in a vast molecular haystack. Curr. Opin. Biotechnol. 10, 87-93. Sharp, A. M., Stein, P. E., Pannu, N. S., Carrell, R. W., Berkenpas, M. B., Ginsburg, D., Lawrence, D. A. & Read, R. J. (1999). The active conformation of plasminogen activator inhibitor 1, a target for drugs to control ®brinolysis and cell adhesion. Structure Fold Des. 7, 111-118. Stemmer, W. P. (1994a). DNA shuf¯ing by random fragmentation and reassembly: in vitro recombination for molecular evolution. Proc. Natl Acad. Sci. USA, 91, 10747-10751. Stemmer, W. P. (1994b). Rapid evolution of a protein in vitro by DNA shuf¯ing. Nature, 370, 389-391. Stratikos, E. & Gettins, P. G. W. (1999). Formation of the covalent serpin-proteinase complex involves transÊ and location of the proteinase by more than 70 A full insertion of the reactive center loop into b-sheet A. Proc. Natl Acad. Sci. USA, 96, 4808-4813. Tucker, M. H., Mottonen, J., Goldsmith, E. J. & Gerard, R. D. (1995). Engineering of PAI-1 to reduce the rate of latency transition. Nature Struct. Biol. 2, 442445. van Meijer, M. & Pannekoek, H. (1995). Structure of PAI-1 and its function in ®brinolysis: an update. Fibrinolysis, 9, 263-276. van Meijer, M., Roelofs, Y., Neels, J., Horrevoets, A. J., van Zonneveld, A. J. & Pannekoek, H. (1996). Selective screening of a large phage display library of plasminogen activator inhibitor 1 mutants to localize interaction sites with either thrombin or the variable region 1 of tissue-type plasminogen activator. J. Biol. Chem. 271, 7423-7428. van Meijer, M., Smilde, A., Tans, G., Nesheim, M. E., Pannekoek, H. & Horrevoets, A. J. (1997). The suicide substrate reaction between plasminogen activator inhibitor 1 and thrombin is regulated by the cofactors vitronectin and heparin. Blood, 90, 1874-1882. van Zonneveld, A. J., Veerman, H., Brakenhoff, J. P., Aarden, L. A., Cajot, J. F. & Pannekoek, H. (1987).
Mapping Sites of Protein-Protein Interaction Mapping of epitopes on human tissue-type plasminogen activator with recombinant deletion mutant proteins. Thromb. Haemost. 57, 82-86. van Zonneveld, A.-J., van den Berg, B. M. M., van Meijer, M. & Pannekoek, H. (1995). Identi®cation of functional interaction sites on proteins using bacteriophage-displayed random epitope libraries. Gene, 167, 49-52. Verhamme, I., Kvassman, J. O., Day, D., Debrock, S., Vleugels, N., Declerck, P. J. & Shore, J. D. (1999). Accelerated conversion of human plasminogen activator inhibitor-1 to its latent form by antibody binding. J. Biol. Chem. 274, 17511-17517. Wells, J. A. (1991). Systematic mutational analyses of protein-protein interfaces. Methods Enzymol. 202, 390-411. Wilczynska, M., Fa, M., Ohlsson, P. I. & Ny, T. (1995). The inhibition mechanism of serpins. Evidence that
1147 the mobile reactive center loop is cleaved in the native protease-inhibitor complex. J. Biol. Chem. 270, 29652-29655. Wilczynska, M., Fa, M., Karolin, J., Ohlsson, P. I., Johansson, L. B. & Ny, T. (1997). Structural insights into serpin-protease complexes reveal the inhibitory mechanism of serpins. Nature Struct. Biol. 4, 354357. Wright, H. T. & Scarsdale, J. N. (1995). Structural basis for serpin inhibitor activity. Proteins: Struct. Funct. Genet. 22, 210-225. Zhang, J. H., Dawes, G. & Stemmer, W. P. (1997). Directed evolution of a fucosidase from a galactosidase by DNA shuf¯ing and screening. Proc. Natl Acad. Sci. USA, 94, 4504-4509. Zhao, H. & Arnold, F. H. (1997). Optimization of DNA shuf¯ing for high ®delity recombination. Nucl. Acids Res. 25, 1307-1308.
Edited by J. Wells (Received 2 March 2000; received in revised form 17 June 2000; accepted 11 July 2000)
