Design of Potent b -Lactamase Inhibitors by Phage Display of b -Lactamase Inhibitory Protein*

b -Lactamase inhibitory protein (BLIP) binds tightly to several b -lactamases including TEM-1 b -lactamase ( K i 0.1 n M ). The TEM-1 b -lactamase/BLIP co-crystal structure indicates that two turn regions in BLIP insert into the active site of b -lactamase to block the binding of b -lactam antibiotics. Residues from each turn, Asp 49 and Phe 142 , mimic interactions made by penicillin G when bound in the b -lactamase active site. Phage display was used to determine which residues within the turn regions of BLIP are critical for binding TEM-1 b -lacta-mase. The sequences of a set of functional mutants from each library indicated that a few sequence types were predominant. These BLIP mutants exhibited K i values for b -lactamase inhibition ranging from 0.01 to 0.2 n M . The results indicate that even though BLIP is a potent inhibitor of TEM-1 b -lactamase, the wild-type sequence of the active site binding region is not optimal and that derivatives of BLIP that bind b -lactamase extremely tightly can be obtained. Importantly, all of the tight binding BLIP mutants have sequences that would be predicted theoretically to form turn structures. b

␤-Lactamase inhibitory protein (BLIP) binds tightly to several ␤-lactamases including TEM-1 ␤-lactamase (K i 0.1 nM). The TEM-1 ␤-lactamase/BLIP co-crystal structure indicates that two turn regions in BLIP insert into the active site of ␤-lactamase to block the binding of ␤-lactam antibiotics. Residues from each turn, Asp 49 and Phe 142 , mimic interactions made by penicillin G when bound in the ␤-lactamase active site. Phage display was used to determine which residues within the turn regions of BLIP are critical for binding TEM-1 ␤-lactamase. The sequences of a set of functional mutants from each library indicated that a few sequence types were predominant. These BLIP mutants exhibited K i values for ␤-lactamase inhibition ranging from 0.01 to 0.2 nM. The results indicate that even though BLIP is a potent inhibitor of TEM-1 ␤-lactamase, the wild-type sequence of the active site binding region is not optimal and that derivatives of BLIP that bind ␤-lactamase extremely tightly can be obtained. Importantly, all of the tight binding BLIP mutants have sequences that would be predicted theoretically to form turn structures.
␤-Lactam antibiotics such as the penicillins and cephalosporins are among the most frequently used antibiotics. The major mechanism of bacterial resistance to ␤-lactam antibiotics is the production of ␤-lactamase. ␤-Lactamase catalyzes the hydrolysis of the amide bond in the ␤-lactam ring to create an ineffective antimicrobial (1). There are four classes (AϪD) of ␤-lactamases based on primary sequence homology (2). TEM-1 ␤-lactamase is a class A enzyme and is the most prevalent plasmid-encoded ␤-lactamase in Gram-negative bacteria (3). It is capable of hydrolyzing penicillins and many cephalosporins (1). Because it is widespread and has a broad substrate profile, TEM-1 ␤-lactamase poses a significant threat to antibiotic therapy.
One strategy for overcoming ␤-lactamase-mediated drug resistance is the use of small molecule, mechanism-based ␤-lactamase inhibitors such as sulbactam, tazobactam, and clavulanic acid (4). These inhibitors are used in conjunction with existing ␤-lactam antibiotics to successfully treat infections by bacterial pathogens that produce ␤-lactamase. In recent years, however, bacteria resistant to the ␤-lactam/␤-lactamase inhib-itor combination have evolved. The resistance is due to mutations in TEM-1 or the related SHV-1 ␤-lactamase that create enzymes that do not react efficiently with the small molecule inhibitors but retain the ability to hydrolyze ␤-lactam antibiotics (5,6).
Clavulanic acid is produced by the soil bacterium Streptomyces clavuligerus (7). This bacterial species also synthesizes a protein inhibitor of ␤-lactamase called the ␤-lactamase inhibitory protein (BLIP) 1 (8). BLIP is a 165-amino acid protein that has been shown to be a potent inhibitor of class A ␤-lactamases including TEM-1 ␤-lactamase (K i ϭ 0.1 nM) (9,10). The cocrystal structure of BLIP with TEM-1 ␤-lactamase indicates that a ␤-hairpin including residues 46 -51 of BLIP inserts into the active site of ␤-lactamase (11). An aspartic acid residue at position 49 of the hairpin is positioned in the active site to form hydrogen bonds with four catalytic residues of ␤-lactamase. In addition, a phenylalanine at position 142 on the other loop occupies a position in the active site similar to the position that the benzyl group of ␤-lactam antibiotic penicillin G occupies during substrate binding and catalysis (11). Replacement of Asp 49 with Ala lowers the binding affinity approximately 80fold, whereas substitution of Phe 142 with Ala results in a 300fold decrease in binding affinity (10). Thus, Asp 49 and Phe 142 make important contributions to the stability of the inhibitory complex.
The display of proteins on the surface of filamentous phage is a powerful method to select variants of a protein with desired binding properties from large combinatorial libraries of mutants (12). The display of a protein of interest can often be achieved by fusing it to the N terminus of the M13 gene III coat protein (13,14). High affinity variants can be obtained by direct selection of combinatorial libraries of mutant proteins displayed on the phage against a corresponding immobilized receptor protein (12,14). Recently, we demonstrated that functional BLIP can be displayed on the surface of M13 phage as a fusion to the gene III coat protein (15). In this study we have created libraries of random mutants at positions including residues 46 -51 of the ␤-hairpin of BLIP as well as within a region containing the important Phe 142 residue. The libraries have been used to select BLIP derivatives that are potent inhibitors of TEM-1 ␤-lactamase.
The pG3-BLIP vector encodes chloramphenicol resistance and contains BLIP as a fusion to the N terminus of the gene III coat protein (15). The secretion of the BLIP-gene III protein fusion is directed by the ␤-lactamase signal sequence, and transcription of the fusion is controlled by the constitutive ␤-lactamase promoter (15). For this study, the pG3-BLIP vector was altered by the insertion of a His 6 tag after the secretion signal sequence. The resulting vector, pTP154, directs the expression of BLIP with a His 6 tag at the N terminus to the surface of the phage. Both the pG3-BLIP and pTP154 vectors contain an amber codon between BLIP and gene III. Therefore, it is necessary to propagate phage in E. coli strains containing an amber suppressor gene.
In the first round, an amplification was performed with the primers BLIPXHOI and L46 -48-Bot, and a separate amplification was performed with the primers L46 -48-Top and BLIPXBAI in a total volume of 100 l for each reaction. In the second round, 1 l of each of the two reactions from the first round was used as a template in an amplification using the BLIPXHOI and BLIPXBAI primers. The L46 -48-Top and L46 -48-Bot primers have complementary sequences, and therefore the PCR products from the first round have overlapping complementary sequences that allow an extension reaction to occur between the PCR products from round 1. The BLIPXHOI and BLIPXBAI primers serve to amplify the overlap extension products. The resulting PCR product was purified using a Qiaquick column (Qiagen) and digested with the restriction enzymes XhoI and XbaI. The digested DNA fragment was gel-purified and ligated with the pTP154 vector, which had been digested with SalI and XbaI. The ligation reaction was used to transform E. coli XL1-Blue cells by electroporation. A total of 2.9 ϫ 10 6 colonies was pooled for the 46 -48 library, giving a Ͼ95% probability of all possible amino acid combinations represented in the libraries (19). The 49 -51 and 141-143 libraries consisted of 3.9 ϫ 10 6 and 2.3 ϫ 10 6 pooled transformants, respectively.
Phage Preparation and Panning-E. coli cultures containing the pooled colonies from the library constructions were grown to A 600 of 0.6 in 25 ml of 2YT (16 g of tryptone, 10 g of yeast extract, 5 g of NaCl per liter) supplemented with 12.5 g/ml chloramphenicol. Approximately 1 ϫ 10 9 VCS M13 helper phage were then added, and the culture was grown overnight with shaking at 37°C. The E. coli cells were removed by centrifugation, and the phage were precipitated from the supernatant with a 0.2 volume of 20% polyethylene glycol, 2.5 M NaCl. The phage were pelleted by centrifugation and resuspended in 0.01 original culture volume of STE (0.1 M NaCl, 10 mM Tris-Cl (pH 8.0), 1 mM EDTA (pH 8.0)). The phage titer was determined by making serial dilutions of a 0.1-ml total volume and adding 0.2 ml of E. coli TG1 cells. Aliquots of 0.15 ml were plated on Luria-Bertani agar supplemented with 12.5 g/ml chloramphenicol. After overnight growth at 37°C, the number of colonies was determined, and the titer was calculated.
Panning was performed by coating wells in a 96-well microtiter plate with 0.2 ml of a 10 g/ml solution of purified TEM-1 ␤-lactamase or bovine serum albumin in 0.05 M Na 2 CO 3 (pH 9.6) overnight at 4°C. The wells were then blocked for nonspecific binding with 0.2 ml of SUPER-BLOCK (Pierce) for 1 h. The wells were washed four times with 0.2 ml of wash buffer (Tris-buffered saline containing 1 mg/ml bovine serum albumin and 0.5 g/liter Tween 20) (27). A total of 1 ϫ 10 11 phage from the library to be panned was added to the ␤-lactamase-coated well in a volume of 0.2 ml and allowed to bind for 2 h at room temperature. Wells were washed 10 times with 0.2 ml of wash buffer, and phage retained in the wells were eluted with 0.2 ml of elution buffer (0.1 M glycine HCl (pH 2.2), 100 mM KCl, 1 mg/ml bovine serum albumin, 0.05% Tween 20). The mixture was neutralized by adding 20 l of 1 M Tris-Cl, pH 8.0. The number of phage eluted was determined by titering for colony-forming units using E. coli TG1 as described above. The eluted phage were amplified for additional rounds of panning by adding 0.15 ml of the neutralized elution mixture to 1 ml of E. coli TG1 cells. After 30 min of incubation at room temperature, 25 ml of 2YT medium were added along with 10 9 VCS M13 helper phage. The phage were precipitated as described above after overnight incubation with shaking at 37°C.
BLIP and ␤-Lactamase Expression and Purification-Wild-type BLIP and mutant BLIP derivatives contain a His 6 tag and were purified using metal affinity chromatography and gel filtration chromatography as described previously (10). The mutant BLIP genes were transferred from the phage display vector to the pTP123 expression vector by digestion with SacI and XbaI to release the BLIP gene followed by insertion of this fragment into pTP123 that had been digested with SacI and XbaI. The pTP123 expression vector provides for isopropyl-1-thio-␤-D-galactopyranoside-inducible expression of BLIP (10). BLIP Inhibition Assay-Inhibition assays were performed as described previously (10). The error listed in Table I is the error of the fit of nonlinear regression analysis to the curves shown in Fig. 5.

Selection of BLIP Variants from Combinatorial Libraries-
Previously, we showed that functional BLIP can be displayed on the surface of M13 phage as a fusion to the gene III protein using a phagemid system (15). For the experiments described here the phagemid was modified to include a His 6 tag following the secretion signal sequence at the N terminus of mature BLIP. It is known that the addition of a His 6 tag to soluble BLIP does not affect its function (10). Enzyme-linked immunosorbent assay experiments confirmed that phage displaying the His 6 -tagged BLIP have the same binding properties as phage displaying wild-type BLIP (data not shown).
It is known from the crystal structure of the BLIP-␤-lactamase complex as well as from mutagenesis results that the residues of the 46 -51 ␤-hairpin as well as a second ␤-hairpin containing Phe 142 of BLIP make important interactions with the active site of TEM-1 ␤-lactamase (10,11). Therefore, these regions were targeted for random mutagenesis to create libraries of amino acid substitutions. BLIP residues were randomized in blocks of three to create libraries containing all possible amino acid combinations for positions 46 -48, 49 -51, and 141-143 (Fig. 1). Overlap extension PCR was used to create each of the random libraries as described under "Experimental Procedures" (18). All of the libraries contain greater than 10 6 clones and therefore have a Ͼ95% probability of containing all possible amino acid combinations over each three-residue window (19).
Each random library contains 20 3 or 8000 possible amino acid sequences over the region randomized. The libraries were sorted for variants that bind tightly to TEM-1 ␤-lactamase by immobilizing ␤-lactamase and allowing phage particles that display the set of BLIP sequences to bind. After extensive washing, the bound phage were eluted from the wells and used to infect E. coli. A portion of the infected culture was spread on agar plates to determine the number of eluted phage particles, whereas the remainder of the culture was used to amplify the phage for additional rounds of binding enrichment.
After one round of panning, the BLIP 46 -48 random library was not enriched for any specific sequences (Fig. 2). After three rounds of panning, however, the library was strongly enriched for the sequence Pro 46 -Ser 47 -Asn 48 (Fig. 2). To ensure that the Pro 46 -Ser 47 -Asn 48 enrichment was not due simply to chance, the 46 -48 library panning experiment was repeated two additional times. In the second set of panning experiments, the Pro 46 -Ser 47 -Asn 48 sequence was not observed after three rounds of panning. However, the closely related Pro 46 -Ser 47 -Ala 48 and Pro 46 -Ser 47 -Ser 48 sequences were found. In a third set of panning experiments, the Pro 46 -Ser 47 -Ala 48 sequence was again found. These results suggest that variants containing proline and serine at positions 46 and 47 are enriched by panning. Furthermore, an examination of the sequences of clones from round three of all of the trials indicates a strong bias for serine at position 47 irrespective of the presence of proline at position 46.
Panning of the 49 -51 library on ␤-lactamase resulted in random sequences after one round of panning (Fig. 3). After three rounds, however, only clones with the sequence Asn 49 -Ser 50 -Tyr 51 were observed, suggesting that this sequence en-codes a BLIP variant with tight binding affinity. The Asn 49 -Ser 50 -Tyr 51 sequence was again detected among enriched phage populations when the panning experiment was repeated two additional times. The sequence Asn 49 -Gly 50 -Tyr 51 was also observed multiple times in the repeated panning experiments. Among the clones obtained after three rounds of panning in each of the trials, it is apparent that asparagine is strongly preferred at position 49. This result is surprising because aspartate at position 49 is thought to make critical interactions with the active site of ␤-lactamase. A possible explanation is that asparagine improves the stability of the type IIЈ turn and that this improved stability compensates for altered interactions with the active site.
Several types of sequences were selected after three rounds of panning of the 141-143 library (Fig. 4). The strongest consensus was observed for position 143, where asparagine was present in 18 of the 29 clones that were sequenced. Large hydrophobic residues similar to the wild-type phenylalanine were observed at position 142 among the phage enriched for binding. Previous studies indicate that substitution of the wildtype phenylalanine with alanine at position 142 results in a 300-fold loss in binding affinity (10). These results suggest that large hydrophobic residues can effectively substitute for phenylalanine at this position. At position 141 the wild-type glycine and serine were the predominant sequences observed. Taken together, the sequencing results for the three positions suggest that the exact wild-type 141-143 sequence is not essential for binding.
Quantitation of BLIP-␤-Lactamase Binding Affinity-The finding that many non-wild-type sequences were enriched after panning of the BLIP libraries suggests that these mutants bind ␤-lactamase as tightly as, or possibly more tightly than, does wild-type BLIP. This hypothesis was tested by expressing and purifying representative BLIP mutants from each of the libraries. Each of the purified proteins was tested for the ability to bind and inhibit purified TEM-1 ␤-lactamase in an in vitro binding assay. The results of the binding assays for the Pro 46 -Ser 47 -Ala 48 , Asn 49 -Gly 50 -Tyr 51 , and Gly 141 -Ile 142 -Asn 143 mutants are shown in Table I. These particular mutants were chosen based on the frequency of occurrence among the selected clones and on the ability to overexpress and purify these proteins. It is apparent that the Pro 46 -Ser 47 -Ala 48 mutant, despite having been selected from the library, binds ␤-lactamase with a similar affinity as does wild-type BLIP. In contrast, the Asn 49 -Gly 50 -Tyr 51 mutant binds ␤-lactamase 10-fold tighter than does wild-type BLIP (Table I and Fig. 5). This result confirms that asparagine can effectively substitute for aspartate at position 49 despite the numerous interactions that aspartate makes with the ␤-lactamase active site. The Gly 141 -Ile 142 -Asn 143 mutant binds ␤-lactamase approximately 2-fold tighter than does wild-type BLIP. Taken together, the results suggest that the panning experiments effectively identify mutants with a range of binding affinities from slightly less to severalfold greater than that of wild type.
Combining BLIP Mutants to Test Additivity-Several studies have shown that mutations that individually increase binding affinity between proteins often act additively when combined into a single molecule (20,21). This hypothesis was tested for the Asn 49 -Gly 50 -Tyr 51 and Gly 141 -Ile 142 -Asn 143 mutants by combining these substitutions into a single BLIP construct. The Asn 49 -Gly 50 -Tyr 51 /Gly 141 -Ile 142 -Asn 143 BLIP mutant was purified and shown to inhibit TEM-1 ␤-lactamase with a K i of 0.02 nM (Table I). Therefore, the Asn 49 -Gly 50 -Tyr 51 and Gly 141 -Ile 142 -Asn 143 substitutions do not act additively. This suggests that the sequence context of the 49 -51 turn influences the interactions of the residues of the 141-143 turn with ␤-lactamase or vice versa.

DISCUSSION
The 46 -51 BLIP turn structure containing Asp 49 is a Type IIЈ ␤-turn (9). It is possible that amino acid substitutions within the turn in BLIP mutants that retain high binding affinity for ␤-lactamase maintain the Type IIЈ turn structure and simultaneously retain contacts with the ␤-lactamase active site. Alternatively, the substitutions may alter the conformation of the turn and promote new interactions with the ␤-lactamase active site. Hutchinson and Thornton (22) examined 3898 ␤-turns and derived sequence preferences for the i, i ϩ 1, i ϩ 2, and i ϩ 3 positions that comprise a Type IIЈ turn. For the BLIP ␤-turn, Ala 47 is at position i, Gly 48 is at i ϩ 1, Asp 49 is at i ϩ 2, and Tyr 50 is at i ϩ 3. According to the derived sequence potentials, tyrosine, valine, and serine are the preferred residue types at position i. The sequence replacement data obtained from the combinatorial phage display libraries indicate a strong preference for serine at position 47 (i). The fact that serine has a higher turn potential than the wild-type alanine has at the i position suggests that the serine substitution acts to stabilize the Type IIЈ turn structure.
At position i ϩ 1, glycine has a much higher calculated turn potential than does any other residue (22). Glycine was also selected at position 48 multiple times from the 46 -48 phage display library (Fig. 2). The next highest potentials for the i ϩ 1 position are serine and alanine, which were also selected multiple times from the library (Fig. 2). Therefore, the results are again consistent with the notion that the substitutions selected from the libraries act to stabilize the Type IIЈ turn. It is also of interest that when a residue other than glycine occurs at position i ϩ 1, a proline is almost invariably found at position 46, which is the i Ϫ 1 position (Fig. 2). A proline is also at this position in the leech decorsin protein, which has an RGD motif in a Type IIЈ turn (23). The limited flexibility of proline may provide rigidity or stability to the turn that permits a residue other than glycine at position i ϩ 1.
The residues with the highest calculated Type IIЈ turn potentials at the i ϩ 2 position are asparagine followed by serine and then aspartate (22). Interestingly, asparagine was clearly the preferred residue at this position among the mutants se-  lected from the 49 -51 phage display library (Fig. 3). Thus, a possible explanation for the observed preference for asparagine over aspartate at position 49 is that the Type IIЈ turn is more stable with asparagine at the i ϩ 2 position and that this increased stability compensates for any interactions that are lost by replacement of the wild-type aspartate residue.
Threonine and glycine have the highest Type IIЈ turn potential at the i ϩ 3 position (22). Glycine was repeatedly selected at this position from the 49 -51 phage display library (Fig. 4). Thus, at all of the Type IIЈ turn positions (47-50) the residues most frequently selected from the phage display libraries were those with high Type IIЈ turn potentials (22). This result suggests that stabilization of the Type IIЈ turn is an important factor in the selection for tight binding BLIP mutants from the random libraries. It is possible that a Type IIЈ turn evolved in this region because the sequence requirements for a Type IIЈ turn overlapped the chemical requirements for binding interactions of the BLIP side chains with the ␤-lactamase active site.
The predominant site of interaction between the turn containing residues 141-143 and ␤-lactamase is at phenylalanine 142 (11). The large, hydrophobic side chain is inserted in the ␤-lactamase active site where the chain makes several interactions (11). Consistent with an important role in binding, substitution of position 142 with alanine results in a 300-fold decrease in affinity (10). The results of the selection of binding mutants from the phage libraries suggest that the function of phenylalanine can be fulfilled by other large, hydrophobic side chains.
The Type IIЈ turn consisting of BLIP residues 47-50 is similar to a Type IIЈ turn in matrix proteins such as fibronectin that is used to bind cell surface integrins such as the platelet glycoprotein IIb/IIIa (24). These matrix proteins contain a consensus Arg-Gly-Asp (RGD) sequence within the Type IIЈ turn. The RGD sequence is found at the apex of solvent-accessible, flexible loops of proteins of different structural folds (24). The RGD Type IIЈ turn can be introduced onto other protein scaffolds where it retains binding function. In addition, cyclic peptides consisting of the RGD turn have been shown to retain the ability to bind integrins in the absence of the scaffold protein (25). The equivalent residues in BLIP to the Arg-Gly-Asp sequence are found in the Ala 47 -Gly 48 -Asp 49 sequence. Studies are in progress to determine whether the 47-50 Type IIЈ turn of BLIP retains binding function when grafted onto other protein scaffolds and to determine whether a cyclic peptide containing the 47-50 region can interact with ␤-lactamase.
The results presented here indicate that although wild-type BLIP binds TEM-1 ␤-lactamase with a K i of 0.1 nM, the BLIP amino acid sequence is not optimally evolved for binding TEM-1 ␤-lactamase. This is not surprising because BLIP is produced by the soil bacterium S. clavuligerus and TEM-1 ␤-lactamase is found in enteric bacteria, and therefore these proteins did not co-evolve (8). However, the BLIP molecule must have evolved to bind a protein that closely resembles TEM-1 ␤-lactamase in order to coincidentally bind TEM-1 ␤-lactamase so tightly. The natural substrate of BLIP is unknown. Nevertheless, the finding that BLIP binds to multiple ␤-lactamases (9) and that the binding affinity can be adjusted by protein engineering presents an opportunity to design new inhibitors for these clinically important targets.