Structural and Computational Characterization of the SHV-1 β-Lactamase-β-Lactamase Inhibitor Protein Interface*

β-Lactamase inhibitor protein (BLIP) binds a variety of class A β-lactamases with affinities ranging from micromolar to picomolar. Whereas the TEM-1 and SHV-1 β-lactamases are almost structurally identical, BLIP binds TEM-1 ∼1000-fold tighter than SHV-1. Determining the underlying source of this affinity difference is important for understanding the molecular basis of β-lactamase inhibition and mechanisms of protein-protein interface specificity and affinity. Here we present the 1.6Å resolution crystal structure of SHV-1 ·BLIP. In addition, a point mutation was identified, SHV D104E, that increases SHV ·BLIP binding affinity from micromolar to nanomolar. Comparison of the SHV-1 ·BLIP structure with the published TEM-1 ·BLIP structure suggests that the increased volume of Glu-104 stabilizes a key binding loop in the interface. Solution of the 1.8Å SHV D104K ·BLIP crystal structure identifies a novel conformation in which this binding loop is removed from the interface. Using these structural data, we evaluated the ability of EGAD, a program developed for computational protein design, to calculate changes in the stability of mutant β-lactamase ·BLIP complexes. Changes in binding affinity were calculated within an error of 1.6 kcal/mol of the experimental values for 112 mutations at the TEM-1 ·BLIP interface and within an error of 2.2 kcal/mol for 24 mutations at the SHV-1 ·BLIP interface. The reasonable success of EGAD in predicting changes in interface stability is a promising step toward understanding the stability of the β-lactamase ·BLIP complexes and computationally assisted design of tight binding BLIP variants.

␤-Lactamase inhibitor protein (BLIP) binds a variety of class A ␤-lactamases with affinities ranging from micromolar to picomolar. Whereas the TEM-1 and SHV-1 ␤-lactamases are almost structurally identical, BLIP binds TEM-1 ϳ1000fold tighter than SHV-1. Determining the underlying source of this affinity difference is important for understanding the molecular basis of ␤-lactamase inhibition and mechanisms of protein-protein interface specificity and affinity. Here we present the 1.6 Å resolution crystal structure of SHV-1⅐BLIP. In addition, a point mutation was identified, SHV D104E, that increases SHV⅐BLIP binding affinity from micromolar to nanomolar. Comparison of the SHV-1⅐BLIP structure with the published TEM-1⅐BLIP structure suggests that the increased volume of Glu-104 stabilizes a key binding loop in the interface. Solution of the 1.8 Å SHV D104K⅐BLIP crystal structure identifies a novel conformation in which this binding loop is removed from the interface. Using these structural data, we evaluated the ability of EGAD, a program developed for computational protein design, to calculate changes in the stability of mutant ␤-lactamase⅐BLIP complexes. Changes in binding affinity were calculated within an error of 1.6 kcal/ mol of the experimental values for 112 mutations at the TEM-1⅐BLIP interface and within an error of 2.2 kcal/mol for 24 mutations at the SHV-1⅐BLIP interface. The reasonable success of EGAD in predicting changes in interface stability is a promising step toward understanding the stability of the ␤-lactamase⅐BLIP complexes and computationally assisted design of tight binding BLIP variants.
Class A ␤-lactamases are a major cause of ␤-lactam resistance in Gram-negative bacteria. These enzymes catalyze the hydrolysis of ␤-lactam antibiotics, such as penicillins and cephalosporins, rendering them inactive. ␤-Lactamase inhibitor protein (BLIP), 6 which is secreted by the Gram-positive soil bacterium Streptomyces clavuligeris, inhibits a variety of class A ␤-lactamase enzymes with a wide spectrum of affinities. Its binding partners include Escherichia coli TEM-1, Klebsiella pneumoniae SHV-1, Serratia marcescens SME-1, Bacillus anthracis BlaI, and Proteus vulgaris K1, among others. BLIP is able to inhibit K1 with picomolar affinity and TEM-1, SME-1, and BlaI with nanomolar affinity. However, it inhibits SHV-1 with only micromolar affinity (1,2). Whereas SHV-1 shares 67% sequence identity with TEM-1 (Fig. 1), and the crystal structures of the unbound ␤-lactamases overlay with an ␣-carbon r.m.s. deviation of 1.4 Å, BLIP exhibits a 1000-fold difference in affinity for the two (3). This poses an interesting question of binding specificity and affinity. How does BLIP bind multiple targets, and what is the source of variation in binding affinity? Recent alanine scanning mutagenesis has provided insight into the origins of BLIP affinity and specificity for an array of ␤-lactamases, including TEM-1 and SHV-1 (2,4). However, interpretation of these data has been limited, because only the structure of the TEM-1⅐BLIP complex has been solved by x-ray crystallography (5).
Alanine-scanning mutagenesis and comparison of unbound crystal structures suggested that class A ␤-lactamase residue 104 (Ambler numbering system) (6), which is a glutamate in TEM-1 and an aspartate in SHV-1, plays a key role in mediating BLIP affinity. It has been hypothesized that the reduced volume of aspartate disrupts the interfacial salt bridge seen between Glu-104 TEM-1 and Lys-74 BLIP and moves the carboxylate group of Asp-104 SHV-1 closer to Glu-73 BLIP , introducing an electrostatic clash (2,3). Interestingly, the TEM E104K mutation decreases affinity for BLIP from 0.11 to 140 nM (7). By contrast, we found that the analogous D104K mutation in SHV has little effect on binding affinity (582 nM for SHV D104K⅐BLIP versus * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.  (Table 1). To understand the basis of these affinity differences, we solved the structures of the SHV-1⅐BLIP and SHV D104K⅐BLIP complexes to 1.6 and 1.8 Å, respectively ( Table 2).
Although crystal structures provide a physical model that can be visualized, a thorough understanding of binding specificity and affinity also requires analysis of the energetics underlying interface stabilization. In turn, by understanding the physical forces involved in protein complex formation, the design of tight binding protein partners with desired specificities should become feasible. The TEM-1⅐BLIP interface has already been demonstrated to be an excellent system for dissecting and reengineering protein-protein interactions; both the Schreiber and Tidor groups (8 -10) have examined the effects of electrostatic interactions on the stability of this complex. To further interpret the mutational data in the context of structure, we used EGAD, our protein design algorithm, to calculate changes in binding affinity for an extensive set of experimentally characterized TEM-1⅐BLIP and SHV-1⅐BLIP mutants (2,7,9,(11)(12)(13)(14). This energy function was previously shown to predict relative affinities for a large set of protein interface mutational data, including a reduced set of TEM-1 mutants, correctly (15). The present calculations extend this work to a larger set of TEM-1⅐BLIP mutations as well as to the SHV-1⅐BLIP interface. The reasonable success of these predictions indicates that EGAD can be used to guide further mutational analysis of the ␤-lactamase⅐BLIP complexes, including creation of tighter binding variants. Whereas the ␤-lactamase⅐BLIP complexes were chosen as model systems, a combined approach of structural, computational, and mutagenic techniques is emerging as a powerful method for engineering protein variants with potential for medical and other scientific applications.
SHV-1 and SHV D104E were expressed in pBC SK(Ϫ) in E. coli DH10B cells grown overnight, without induction. SHV D104K was expressed in E. coli BL21(DE3) cells by induction with 0.2 mM isopropyl-␤-D-thiogalactopyranoside at A 600 ϭ 0.8 for 3 h at 37°C. For all enzymes, cells were harvested by centrifugation at 4°C and frozen overnight. ␤-Lactamase was liberated using stringent periplasmic fractionation with lysozyme and EDTA as previously described (18). Preparative isoelectric focusing was performed with a Sephadex granulated gel and ampholines in the pH range of 3.5-10 (Amersham Biosciences). The protein was eluted with 20 mM diethanolamine buffer, pH 8.3. An additional HPLC purification step was performed on a Waters high pressure liquid chromatograph using a Sephadex Hi Load 26/60 column (Amersham Biosciences) and eluted with phosphate-buffered saline (pH 7.4). This expression and purification scheme yielded ϳ10 mg of pure protein/liter of culture for SHV-1 and SHV D104E and 1 mg for SHV D104K.

Characterization of the SHV-1⅐BLIP Interface
TEM-1 cloned into pET24a(ϩ) with a N-terminal OmpA secretion signal was provided by Stéphane Gagné (Université Laval). TEM-1 was expressed and purified as in Ref. 19. The BLIP construct was a generous gift from Susan Jensen (University of Alberta). The BLIP bli gene was cloned into pET26b (Novagen, Madison, WI), with the native S. clavuligeris signal sequence at the N terminus. BLIP was expressed in E. coli BL21(DE3) cells by inducing with 1 mM isopropyl-␤-D-thiogalactopyranoside at A 600 ϭ 0.5 for 3 h at 30°C. Cells were harvested by centrifugation, resuspended in 20 g/100 ml sucrose, 1 mM EDTA, 30 mM Tris, pH 8.0, and incubated for 15 min. The cells were pelleted and resuspended in 5 mM MgCl 2 . After centrifugation at 7,000 rpm for 10 min, the periplasmic fraction was retained, and a Complete protease inhibitor mixture tablet (Roche Applied Science) was added. That fraction was subsequently clarified by additional centrifugation and loaded onto a HiPrep 16/10 Q-XL anion exchange column (Amersham Biosciences) equilibrated in 25 mM Tris, pH 8.4, 1 mM EDTA. A 250-ml gradient from 0 to 500 mM NaCl was used to isolate BLIP, which elutes at ϳ150 mM NaCl. Fractions were pooled and concentrated to a final volume of ϳ2 ml and subsequently passed through a HiLoad 26/60 Superdex 75 preparation grade gel filtration column (Amersham Biosciences) equilibrated in 50 mM Tris, pH 8.4, 100 mM NaCl. After purification, BLIP was concentrated to 1 mg/ml and stored at Ϫ80°C. This expression and purification scheme yielded roughly 0.5 mg of pure protein/liter of culture.
Crystallization, Data Collection, and Structure Solution-SHV-1 or SHV D104K was mixed 1:1 with BLIP in 20 mM BisTris, pH 7.25, 50 mM NaCl, concentrated to 8.7 mg/ml, and dialyzed overnight against 20 mM BisTris, pH 7.25, 50 mM NaCl. Crystals were grown at 19°C in microbatch format under Al's oil (Hampton Research, Aliso Viejo, CA) by mixing 1 l of protein with 1 l of well solution. The SHV-1⅐BLIP crystals were grown in a well solution of 60% ammonium sulfate, 50 mM cacodylate, pH 6.5; SHV D104K⅐BLIP crystals were grown in a well solution of 8% polyethylene glycol 8000, 50 mM cacodylate, pH 6.5. For harvesting, a cryoprotectant solution containing well solution plus 25% glycerol for SHV-1⅐BLIP or well solution plus 30% xylitol for SHV D104K⅐BLIP was added directly to the drop, and the crystals were immediately looped and flash-frozen in liquid nitrogen.
Data sets were collected on Beamline 8.3.1 at the Advanced Light Source at Lawrence Berkeley National Laboratory (20). A preliminary 2.1 Å data set for SHV-1⅐BLIP was collected on an R-Axis IVϩϩ at the University of California Berkeley (Rigaku/ MSC, The Woodlands, TX). Data were indexed and reduced with HKL2000 (21) or ELVES (22) using MOSFLM (23). For the 1.8 Å high resolution SHV-1⅐BLIP structure, molecular replacement was performed using a partially refined structure from the 2.1 Å data set. Initial maps for the SHV D104K⅐BLIP structure were generated by molecular replacement with PHASER (24) using polyalanine models of SHV-1 (coordinates taken from Protein Data Bank entry 1SHV) and BLIP (coordinates taken from Protein Data Bank entry 1JTG) (3,5). Manual rebuilding was carried out with O (25). Refinement was carried out with Refmac (26) using ARP to automatically place ordered waters, followed by TLS refinement (27). Additional details of the data collection and refinement are provided in Table 2. Structural alignments and r.m.s. deviations were calculated with LSQMAN (28). All molecular figures were created with PyMOL (29).
Inhibition Assays-All kinetic determinations were performed using nitrocefin (BD Biosciences) as the indicator substrate to measure hydrolysis rates of SHV-1, SHV D104K, SHV D104E, and TEM-1 ␤-lactamases with and without inhibitor. BLIP and lactamase were incubated for 2 h at room temperature in 10 mM sodium phosphate-buffered saline containing 1 mg/ml bovine serum albumin. Initially, 7 nM enzyme was used for all assays, but this was reduced to 4 nM for SHV D104E and 2 nM for TEM-1 to obtain an accurate measurement of the low K d values. Reactions were initiated with nitrocefin at the K m for the enzyme (25 M for SHV-1, 15 M for SHV D104K, 25 M for SHV D104E, and 150 M for TEM-1). Final reaction volumes were 1 ml. All measurements were performed in triplicate. Hydrolysis rates were determined at ϭ 482 nm using the extinction coefficient, ⑀ ϭ 17,400 M Ϫ1 cm Ϫ1 for the hydrolyzed form of nitrocefin (30).
BLIP inhibition curves were graphed using Origin 7.5 SR software and fit to the following equation, where E free represents the remaining free enzyme concentration calculated based on activity, [E 0 ] is initial enzyme concen-  (7). K* d was corrected for the presence of substrate using the following equation.
Computational Calculation of Free Energies of Dissociation-EGAD_lib, 7 a Cϩϩ implementation of the EGAD protein design energy function, was employed to calculate dissociation free energies of the complexes (15,31). The function consists of an OPLS-AA derived force field (32) alongside a solvent-accessible surface area term and the generalized Born model to describe solvation. The force field terms include a linearized van der Waals (vdW) potential, a coulombic electrostatics term, and a torsional potential. Minimization was conducted using Monte Carlo simulated annealing followed by a heuristic quench step (15).
An initial model for the bound state of the complex was created by rotamer optimization of the corresponding crystal structure (Protein Data Bank code 1JTG for TEM-1⅐BLIP, Protein Data Bank code 2G2U for SHV-1⅐BLIP, or Protein Data Bank code 2G2W for SHV D104K⅐BLIP) (45). The unbound state model was generated by separating the two chains in the complex by 25 Å, followed by rotamer optimization of each. For these complexes, calculations performed with the rotamer-optimized initial model were not significantly different from those using the Protein Data Bank structure directly. Typically, we use initial rotamer optimization in interface calculations as a precaution against improperly refined rotamers, although the high resolution of the ␤-lactamase⅐BLIP structures makes this step less important.
The free energy of dissociation is defined as the energy difference between the complex and free states.
The free energy used for comparison with experimental values is as follows.
Here is a scaling factor used to normalize the predicted changes in binding affinity to have a slope of 1 when compared with the experimental energy changes (15). For ⌬⌬G d, mut calculated with the TEM-1⅐BLIP backbone coordinates, ϭ 0.18; for both SHV-1⅐BLIP and SHV D104K⅐BLIP, ϭ 0.30. The mutant free energies were calculated by rotamer optimization of the mutant structures, allowing only a reduced set of positions to move during optimization. This set was restricted to residues at the interface (residues that undergo a change in solvent-accessible surface area upon binding) and their neighbors.

RESULTS
Structure of the SHV-1⅐BLIP Complex-The structure of the SHV-1⅐BLIP complex is globally similar to that of TEM-1⅐BLIP; alignment of the ␣-carbon atoms of TEM-1 and SHV-1 yields an r.m.s. deviation of 0.44 Å between the two ␤-lactamases and an r.m.s. deviation of 1.04 Å between the associated BLIPs (Fig. 2). Alignment of all ␣-carbon atoms yields an overall r.m.s. deviation of 0.61 Å for the two complexes. As in the TEM-1 complex, the concave ␤-sheet region of BLIP latches onto a protruding loophelix region of SHV-1 (residues 99 -112), and two loops of BLIP insert into the ␤-lactamase active site. Residues Asp-49 (at the end of the first loop) and Phe-142 (at the end of the second loop) form an approximate mimic of the ␤-lactamase substrate, penicillin G (5). Comparison of the bound and unbound forms of SHV-1 reveals only minimal global changes in conformation upon binding, with the exception of the H10 helix (residues 218 -230) (Fig.  3). In the SHV-1⅐BLIP structure, the H10 helix partially unravels at the N terminus, assuming a structure similar to the H10 helix as observed in both the bound and unbound forms of TEM-1. The H10 helix conformation in the crystal of unbound SHV-1 is possibly affected by proximal binding of a detergent molecule, Cymal-6 (3). Previously, it was observed that the conformation of the H10 helix interacting with detergent in the SHV-1 unbound structure was similar to the conformational change seen when TEM-1 interacts with an allosteric cryptic site inhibitor (33).
The question then remains, what features are responsible for the differential affinity of BLIP for TEM-1 versus SHV-1? There are eight SHV-1 residues within 6 Å of BLIP that differ in identity from TEM-1 (Fig. 1A). These include R98S, Q100N, D104E, A114T, S133T, T167P, R215K, and T235S (here the residue identity in SHV-1 is listed first). Comparison of the hydrogen bond, salt bridge, and vdW interactions of these residues in the TEM-1⅐BLIP and SHV-1⅐BLIP complexes provides some structural insight. EGAD was used to quantify the extent of vdW contact. Pairwise vdW energies were calculated between all interface residues, and those providing stabilizing interactions in excess of Ϫ1.0 kcal/mol are tabulated in supplemental Table  2. A combination of the REDUCE and BndList programs was used to detect hydrogen bonds and salt bridges (34) (supplemental Table 1). In a few cases, the SHV-1 residue type contributes new interactions: Arg-98 SHV has improved vdW packing with Trp-150 BLIP relative to Ser-98 TEM , and Arg-215 SHV forms two hydrogen bonds with Glu-31 BLIP , whereas Lys-215 TEM forms only one. For positions 133 and 235, there is little difference between the TEM-1 and SHV-1 residues; S133T makes few interfacial contacts in SHV-1 or TEM-1, and Thr-235 SHV forms a hydrogen bond with Asp-49 BLIP analogous to that formed by Ser-235 TEM . However, for the remaining four residue differences, TEM-1 forms more extensive interactions across the interface. The backbone carbonyls of Gln-100 SHV and Asn-100 TEM both form a hydrogen bond with Arg-160 BLIP , although Asn-100 TEM is able to form more extensive vdW contacts with Trp-150 BLIP and Arg-160 BLIP (Fig. 1B). Thr-114 TEM exhibits improved vdW interactions with Tyr-115 BLIP in comparison with Ala-114 SHV (Fig. 1C), and Pro-167 TEM forms more extensive contacts with Trp-162 BLIP ; however, Thr-167 SHV does provide increased vdW interactions with Phe-142 BLIP (Fig.  1D). The reduced volume of Asp-104 SHV eliminates the interfacial salt bridge between Glu-104 TEM and Lys-74 BLIP . A hydrogen bonding interaction between Glu-104 TEM and the Tyr-143 BLIP backbone is not formed with Asp-104 SHV , favorable vdW packing interactions between Glu-104 TEM and Phe-142 BLIP are entirely eliminated in the SHV-1⅐BLIP interface, and the packing between Asp-104 SHV and Tyr-143 BLIP is significantly reduced relative to Glu-104 TEM (Fig. 1E). These dramatic changes between position 104 in SHV-1 compared with TEM-1 suggest that it might be a particularly important hot spot in the interface.

Mutagenesis of SHV Position 104-
The substitution of aspartate in SHV-1 for glutamate in TEM-1 at position 104 shortens the side chain by a single carbon, yet a number of interactions across the ␤-lactamase⅐BLIP interface are eliminated. The loss of the Glu-104 TEM and Lys-74 BLIP salt bridge results in the energetically unfavorable burial of unsatisfied charges at the interface. The extensive contacts removed between position 104 and the Phe-142 loop are consistent with previous mutational data indicating that Phe-142 TEM is a hot spot at the TEM-1⅐BLIP interface, whereas Phe-142 SHV appears to have a reduced role in the SHV-1⅐BLIP interface (2). For these reasons, we further explored the role of position 104 at the SHV-1⅐BLIP interface through mutagenesis.
Remarkably, we find that the point mutant SHV D104E results in a 1000-fold increase in the binding affinity of BLIP ( Table 1). The D104E substitution in SHV-1 may function to restore the salt bridge with Lys-74 BLIP and, perhaps more importantly, provide stabilizing interactions with the Phe-142 loop. We are currently conducting further characterization of the SHV D104E mutation through crystallography and a more extensive double mutant cycle analysis.
In addition to the SHV D104E mutation, the SHV D104K mutation was characterized. The analogous TEM E104K mutation leads to an extended spectrum resistance phenotype and was previously found to decrease TEM-1 affinity for BLIP from 0.11 to 140 nM (7). However, we find that the SHV D104K mutation leads to a moderate increase in affinity, from 1252 to 582 nM. Interestingly, crystallography of this complex identifies an alternate BLIP binding mode.
Structure of the SHV D104K⅐BLIP Complex-In the SHV-1⅐BLIP structure, the Asp-49 and Phe-142 loops occupy the active  site as in TEM-1. However, in the SHV D104K mutant complex, the Phe-142 loop swings out of the active site cavity, and the Asp-49 loop is rotated to partially occupy the space vacated by the Phe-142 loop (Fig. 2). Whereas the wild type binding mode buries an extent of surface area comparable with the TEM-1 complex (2757 Å 2 in TEM-1 versus 2624 Å 2 in SHV-1), the mutant buries only 2450 Å 2 . Curiously, despite the loss of interactions with the Phe-142 loop and the decreased amount of interfacial buried surface area, we find the SHV D104K interface to be slightly stabilized relative to the native complex ( Table 1).
The removal of the Phe-142 loop from the SHV active site is the most striking difference between the wild type and D104K complexes (Fig. 2). The loop is somewhat disordered, and the density for BLIP residue 139 is missing entirely in the mutant complex. The loop forms a small number of crystal contacts with a neighboring SHV D104K⅐BLIP complex, consisting of two potential hydrogen bonds between Tyr-143 BLIP and the neighboring SHV Gln-206 and Trp-110. When Lys-104 SHV is modeled into the wild type SHV-1⅐BLIP complex using EGAD, a clash is introduced with Tyr-143 BLIP . By swinging out of the active site, the Phe-142 loop relieves this collision. The removal of the flexible Phe-142 loop from the interface may serve to increase the overall entropy of the complex. However, all vdW contacts between the BLIP Phe-142 loop and SHV are lost when the Phe-142 loop is not present (supplemental Table 2). Instead, these lost interactions appear to be compensated for by increased interactions with the Asp-49 loop.
The interface hot spot residue Asp-49 BLIP appears in almost identical conformations in the TEM-1⅐BLIP, SHV-1⅐BLIP, and SHV D104K⅐BLIP structures (Fig. 2). The average r.m.s. deviation for all possible pairwise comparisons between Asp-49 BLIP in the three structures is 0.98 Å. In each case, Asp-49 BLIP forms two salt bridges with Arg-244 SHV and Lys-234 SHV , as well as making two hydrogen bonds with Ser-130 SHV and Thr-235 SHV (Ser-235 TEM ). This positioning is consistent with its shared role as a hot spot in both the TEM-1⅐BLIP and SHV-1⅐BLIP interfaces (2,12). Despite the highly conserved structural role of Asp-49 BLIP in SHV D104K, the remainder of this loop is able to pivot around this residue, rotating further into the ␤-lactamase active site (Fig. 4). This allows the Tyr-50 BLIP side chain to swing into the space vacated by loop residue Phe-142 BLIP , creating two new interactions: a hydrogen bond between Tyr-50 BLIP and Asn-132 SHV and a possible interaction between Asn-132 SHV and the aromatic face of Tyr-50 BLIP . In concert with the Tyr-50 BLIP movement, residue Tyr-105 SHV must swivel to the other side of the loop to avoid clashing with Tyr-50 BLIP . This rearrangement breaks a hydrogen bond between Tyr-105 SHV and Gly-141 BLIP (located in the Phe-142 loop) and creates a new hydrogen bond between Tyr-105 SHV and Tyr-51 BLIP (supplemental Table 1).
The concerted movement between Tyr-50 BLIP and Tyr-105 SHV also changes the vdW packing at the interface (supplemental Table 2). Tyr-105 SHV no longer forms stabilizing vdW contacts with Phe-142 BLIP but now interacts with Tyr-51 BLIP and Tyr-53 BLIP in the Asp-49 loop, as well as forming more extensive interactions with Tyr-50 BLIP . The repositioning of Tyr-105 SHV is similar to that observed in the TEM1 F142A⅐BLIP structure (13), in which the Asp-49 and Phe-142 loops maintain their positions in the active site, but the area around the Phe-142 loop is unstructured. Tyr-50 BLIP also forms new interactions with Ser-130 SHV , Asn-132 SHV , and Glu-166 SHV .
Examining the SHV⅐BLIP structures with PROCHECK reveals that Tyr-50 BLIP is sterically strained in the wild type interface, and the altered conformation of the Asp-49 loop in the SHV D104K⅐BLIP structure relieves this strain (35). The Y50A mutation stabilizes both the TEM-1⅐BLIP and SHV-1⅐BLIP interfaces (2). This stabilizing effect may arise from relief of steric strain in the Asp-49 loop.
Summarizing, the Phe-142 loop swings out of the SHV active site cavity, whereas residues in the Asp-49 loop rotate in to occupy some of the missing positions. Newly created interactions from the Asp-49 loop and relief of steric strain may explain the slightly increased affinity of SHV D104K over SHV-1 for BLIP. These large concerted structural rearrangements also illustrate how prediction of the effects of mutations may sometimes be difficult. Nevertheless, since we are interested in using computational methods to guide future mutagenesis studies of the interface, we conducted an extensive examination of the ability of EGAD to predict stability changes arising from mutation.
Calculation of ⌬⌬G d, mut Using EGAD-Dissociation constants have been experimentally determined for an extensive set of both TEM-1⅐BLIP and SHV-1⅐BLIP mutant complexes, including nonalanine mutations and constructs with as many as six mutations (2,7,9,(11)(12)(13)(14). These data provide an excellent opportunity for evaluating the ability of EGAD to calculate changes in the free energy of dissociation (⌬⌬G d, mut ) and, where predictions fall short, a metric for improving the energy function. Experimentally observed ⌬⌬G d, mut values were calculated within an error of 1.6 kcal/mol for a set of 112 TEM-1⅐BLIP mutants (Fig. 5A). By comparison, the experimental error for ⌬⌬G d, mut determined from independent experiments ranges from 0.1 to 1.3 kcal/ mol (average 0.5 kcal/mol) for the TEM-1⅐BLIP interface. Whereas this demonstrates that the program performs reasonably well for the majority of characterized mutations, in several cases, we find the changes in energy to be largely overestimated. ⌬⌬G d, mut is consistently overpredicted for complexes that include BLIP Y50A or TEM Y105A. The calculated changes in affinity are also overpredicted for the double mutant TEM E104K⅐BLIP F142A and for the quadruple mutant TEM E104A⅐BLIP K74A/F142A/ Y143A (Fig. 5B). For the SHV-1⅐BLIP interface, energies were calculated using the backbone conformations from both the SHV-1⅐BLIP and SHV D104K⅐BLIP complex structures. Using the SHV-1⅐BLIP structure, the effects of mutation on binding affinity are predicted within an error of 2.1 kcal/mol, and for the SHV D104K⅐BLIP structure, the error is 2.2 kcal/mol (Fig. 5, C and  D). For both structures, ⌬⌬G d, mut is again overpredicted for complexes that include the mutation BLIP Y50A. Calculations using the SHV-1⅐BLIP structure also overpredicted the effects of mutants BLIP F142A, BLIP Y143A/W112A, BLIP E73A/Y143A, and SHV D104E.

DISCUSSION
Subtle structural differences between SHV-1 and TEM-1 result in a large change in BLIP binding affinity (Fig. 1, B-E). Four SHV-1 residues that differ in identity from TEM-1 result in decreased vdW and electrostatic interactions at the interface according to our calculations. These include Q100N, A114T, T167P, and D104E (SHV amino acid identity is listed first). Importantly, the loss of stabilizing interactions indicated by the structures and quantified with our program is in agreement with previous alanine scanning mutagenesis. For example, EGAD calculations find that the Q100N difference results in less extensive vdW contacts between Gln-100 SHV and both Trp-150 BLIP and Arg-160 BLIP (supplemental Table 2). This provides a possible structural explanation for previous alanine scanning data, which found that the BLIP W150A and R160A mutations were far more destabilizing to the TEM-1⅐BLIP interface than the SHV-1⅐BLIP interface (2). At position Thr-167 SHV , EGAD calculates a decrease in vdW interactions with Trp-162 BLIP relative to Pro-167 TEM (supplemental Table 2). Correspondingly, it was experimentally determined that the BLIP W162A mutation destabilizes the TEM-1⅐BLIP complex by 2.18 kcal/mol but destabilizes the SHV-1⅐BLIP complex by only 0.53 kcal/mol (2). Additionally, the SHV-1⅐BLIP structure confirms that Asp-104 SHV is unable to participate in a salt bridge formed by Glu-104 TEM and Lys-74 BLIP ; experimentally, it was found that the BLIP K74A mutation destabilizes the TEM-1⅐BLIP interface due to the loss of the salt bridge but not the SHV-1⅐BLIP interface, where the salt bridge is absent (2).
The decreased volume of Asp-104 SHV not only removes a salt bridge but has potential implications for the stability of the Phe-142 binding loop. As described under "Results," a number of contacts with the Phe-142 binding loop are removed; calculating the interfacial vdW interactions for the Phe-142 loop with EGAD reveals that the interactions eliminated between Asp-104 SHV , Phe-142 BLIP , and Tyr-143 BLIP account for 41% of the overall vdW energy stabilizing the loop in the TEM-1⅐BLIP complex. The reduced role of Phe-142 BLIP in the SHV-1⅐BLIP interface is consistent with the appearance of the alternate conformation observed in the SHV D104K⅐BLIP complex and the relative indifference in binding affinity to the removal of Phe-142 BLIP from the active site cavity. The conformational flexibility observed in the Asp-49 and Phe-142 loops may contribute to the ability of BLIP to bind a range of ␤-lactamases. Structural plasticity in loop regions is a common motif in protein interfaces that bind multiple partners with high affinity (e.g. conformational flexibility in antibody loops results in specific recognition of diverse antigens) (36).
Overall, the change in both hydrogen bonding and vdW packing in the SHV D104K⅐BLIP interface reflects a shift toward increased stabilization around the Asp-49 loop and a reduced number of interactions with the Phe-142 loop. This reorganization suggests that SHV-1 may be inhibited by a reduced set of contacts with a single binding loop. This finding is promising for inhibitor development and reinforces the results of previous efforts to develop peptide inhibitors. Peptides corresponding to residues 46 -51 of BLIP (the Asp-49 binding loop) were shown to bind TEM-1 and SHV-1 with affinities of 488 and 420 M, respectively (37). Random fragmentation and phage display of BLIP identified a peptide consisting of residues 30 -49 of BLIP that inhibits TEM-1 with 446 M affinity, and a combination of phage display and peptide arrays identified a 136 M inhibitor that has 50% sequence identity to BLIP residues 46 -51 (38,39). All of these studies point to the central importance of the Asp-49 loop in inhibition.
The ␤-lactamase⅐BLIP complexes represent challenging computational analysis and design problems, since our structures indicate that the BLIP backbone assumes multiple conformations to accommodate mutations. With some notable exceptions (40 -42), many computational design approaches use a fixed backbone approximation to simplify calculations. This approximation allows only side chain mobility, thus enormously reducing the degrees of conformational freedom required for the energy minimization. However, the fixed backbone approxima-tion may introduce significant error into design calculations, since backbone adjustments upon mutation are neglected. Further, the possible coexistence of multiple conformations is ignored. Entropic effects, such as the penalty associated with burying a flexible loop in a protein-protein interface, are not accounted for. Despite the use of these approximations, the ⌬⌬G d, mut values of 85% of the TEM-1⅐BLIP complexes are predicted within 1.6 kcal/ mol. For the SHV-1⅐BLIP complexes, the calculations are less successful. Using the SHV-1⅐BLIP structure as a backbone model, ⌬⌬G d, mut is overestimated for nearly 26% of the mutant complexes.
Mutations involving large changes in volume (e.g. tyrosine to alanine) at amino acid positions that undergo dramatic conformational changes between the SHV-1⅐BLIP and SHV-1 D104K⅐BLIP structures were frequently observed as outliers in both the TEM-1⅐BLIP and SHV-1⅐BLIP correlations. Neglecting backbone motion results in poor modeling of such mutations. Better treatment of loop conformational flexibility, possibly incorporating energy calculations over a backbone ensemble rather than use of a single fixed backbone, may be necessary for higher accuracy calculations (43,44).

CONCLUSION
A combination of structural, mutagenic, and computational analysis of the SHV-1⅐BLIP interface has highlighted several subtle yet important differences between this complex and that of TEM-1⅐BLIP. Notably, the region surrounding ␤-lactamase position 104 appears to be a key determinant of specificity. Creation of BLIP mutants that attempt to compensate for local interface differences near this and other positions may result in tighter binding variants. The reasonable success of EGAD in predicting the experimental mutational data suggests that it will be useful in guiding further mutagenesis efforts.