Inhibition of the Class C β-Lactamase from Acinetobacter spp: Insights into Effective Inhibitor Design

Abstract

The need to develop β-lactamase inhibitors against class C cephalosporinases of Gram-negative pathogens represents an urgent clinical priority. To respond to this challenge, five boronic acid derivatives including a new cefoperazone analog were synthesized and tested against the class C cephalosporinase of Acinetobacter baumannii (Acinetobacter-Derived Cephalosporinase, ADC). The commercially available carbapenems antibiotics were also assayed. In the boronic acid series, a chiral cephalothin analog with a meta-carboxyphenyl moiety corresponding to the C₃/C₄ carboxylate of β-lactams showed the lowest K_i (11 ± 1 nM). In antimicrobial susceptibility tests, this cephalothin analog lowered the ceftazidime and cefotaxime minimum inhibitory concentrations (MICs) of Escherichia coli DH10B cells carrying bla_ADC from 16 → 4 μg/ml, and 8 → 1 μg/ml, respectively. On the other hand, each carbapenem exhibited a K_i < 20 μM, and timed electrospray ionization mass spectrometry (ESI-MS) demonstrated the formation of adducts corresponding to acyl-enzyme intermediates with both intact carbapenem and carbapenem lacking the C₆ hydroxyethyl group. To better understand the interactions between the β-lactamase and the inhibitors, we constructed models of ADC as an acyl-enzyme with: i) the meta-carboxyphenyl cephalothin analog; and ii) the carbapenems imipenem and meropenem. Our first model suggests that this chiral cephalothin analog adopts a novel conformation in the β-lactamase active site. Further, the addition of the substituent mimicking the cephalosporin dihydrothiazine ring may significantly improve affinity for the ADC β-lactamase. In contrast, the ADC: carbapenem models offer a novel role for the R₂ side group, and also suggest that elimination of the C6 hydroxyethyl group by retroaldolic reaction leads to a significant conformational change of the acyl-enzyme. Lessons from the diverse mechanisms and structures of the boronic acid derivatives and carbapenems provide insights for the development of new β-lactamase inhibitors against these critical drug resistance targets.

Acinetobacter spp. are Gram-negative pathogens responsible for an increasing number of serious nosocomial infections including hospital-acquired pneumonia, urinary tract infections, and bacteremia (1-4). This non-fermentive, aerobic pathogen harbors multiple antibiotic resistance determinants, including chromosomal AmpC β-lactamase enzymes, OXA carbapenemases, metallo-β-lactamases, and multidrug resistance (MDR) efflux pumps (5, 6). In addition, changes in outer membrane proteins decrease permeability to antimicrobials (7). Besides intrinsic resistance, Acinetobacter spp. possess the ability to acquire new resistance determinants through gene mutations, derepression, and transfer from other organisms. These remarkable attributes can lead to infections resistant to all available β-lactam antibiotics (8, 9). Consequently, treatment of patients with Acinetobacter spp. infections is very challenging, and therapeutic options are severely limited for MDR strains (10-12).

One strategy for restoring the efficacy of β-lactam antibiotics is the development of novel β-lactamase inhibitors. Boronic acid derivatives are compounds that replace the β-lactam ring with boronic acid. The boron atom forms of a reversible, dative covalent bond with the active site serine of class A and C β-lactamases, assuming a geometry that resembles the tetrahedral transition state of the β-lactamase hydrolytic reaction (Figure 1) (13, 14). By modifying the boronic acid substituents to resemble in structure, distance, and stereochemical arrangement the R₁ side chains of natural substrates, affinities in the nM range against class C enzymes of Escherichia coli are achieved (15-17).

Figure 1.

Schemes illustrating the interactions of a serine β-lactamase with: (A) the β-lactam cephalosporin ceftazidime; (B) the boronic acid ceftazidime analog, compound 2; and (C) the carbapenem imipenem.

A second approach to counteracting β-lactamase mediated antibiotic resistance is the design of β-lactams that resist hydrolysis. Through the combined efforts of natural product screens and medicinal chemistry, β-lactamase-stable penem and cephem derivatives have been modified and synthesized. The most potent β-lactams are the derivatives of thienamycin (i.e., imipenem, meropenem, ertapenem, and doripenem). Carbapenems act as inhibitors of class A, C, and certain class D β-lactamases by forming a prolonged acyl-enzyme intermediate with the β-lactamase that is very slowly hydrolyzed (18-25).

The Acinetobacter-Derived Cephalosporinases (ADCs) are class C β-lactamases found in Acinetobacter baumannii and Acinetobacter genomospecies 3 and are responsible for resistance to penicillins, cephalosporins, and β-lactam-β-lactamase inhibitor combinations (26). These AmpC β-lactamases demonstrate a remarkable k_cat for first-generation cephalosporins and relatively low affinity for the commercially available β-lactamase inhibitors (26). Therefore, ADC enzymes can serve as important targets for the design of new mechanism-based inactivators. To date, the search for effective inhibitors for the ADC β-lactamase remains challenging. To this end, we synthesized and tested a panel of boronic acid derivatives with specific side chains to serve as chemical probes. We also designed a novel boronate that contained the R₁ side chain of cefaperazone. Concurrently, we explored the role of the R₂ side chain of four different carbapenems in the inhibition of ADC. Taken together, our results indicate that scaffold and side chain modifications can be optimized to yield inhibitors with favorable kinetic properties against ADC β-lactamases.

Materials and Methods

Antibiotics and inhibitors

The chemical structures of antibiotics and inhibitors studied are shown in Figure 2. Ceftazidime and cefotaxime were purchased from Sigma (St. Louis, MO). The boronic acid ceftazidime analog and cephalothin analogs were synthesized as previously described (17). The chiral cephalothin analogs 4 and 5 were obtained in the enatiomerically pure form (27). Imipenem and ertapenem were obtained from Merck & Co. Inc. (Whitehouse Station, NJ). Meropenem was purchased from AstraZeneca Pharmaceuticals (Wilmington, DE) and doripenem from Ortho-McNeil Pharmaceutical Inc. (Raritan, NJ).

Figure 2.

Chemical structures of: (A) commercially available inhibitors and cephalosporin substrate cephalothin; (B) boronic acid derivatives; and (C) carbapenems used in this study. Cephalothin structure is labeled with accepted ring numbering system. The C₆ hydroxyethyl group of imipenem, which may be eliminated after formation of the acyl-enzyme, is circled in dashed red lines.

Synthesis of cefoperazone analog

The cefoperazone analog 1 was synthesized according to the general protocol for the other boronic acid derivatives by acylation of pinacol bis-(trimethylsilyl)-aminomethaneboronate with the commercially available cefoperazone acid, promoted by isobutyl chloroformate (17). Triethylamine (231 μL, 1.66 mmol) and isobutyl chloroformate (216 μL, 1.66 mmol) were added to a solution of (2R)-2[(4-ethyl-2,3-dioxopiperazinyl)carbonylamino]-2-(4-hydroxyphenyl)acetic acid (530 mg, 1.66 mmol) in anhydrous tetrahydrofuran (THF, 60 mL) at 0 °C and allowed to react under argon atmosphere for 1 h. A solution of bis-(trimethylsilyl)-aminomethaneboronate (500 mg, 1.66 mmol) in anhydrous THF (5 mL), previously treated for 30 min with anhydrous methanol (1.74 mmol), was added at the same temperature. After 20 min, the cooling bath was removed and the mixture was allowed to react overnight at room temperature. Thereafter, the reaction mixture was diluted with diethylether (60 mL) and the precipitate (trithylammoium chloride) was removed by filtration. The solvent was distilled under reduced pressure and the solid residue crystallized from ethyl acetate, affording the title compound as a whitish solid (54% yield). [α]_D −59.0 (c 0.8, CH₃OH). The pinacol ester spontaneously hydrolyzes in the phosphate buffer, generating the corresponding cefoperazone boronic acid.

¹H and ¹³C NMR spectra of compound 1 were recorded on a Bruker DPX-200 or Avance 400 spectrometer. The chemical shifts (δ) are reported in parts per million downfield from the internal standard, tetramethyl silane (s, singlet; d, doublet; t, triplet; q, quartet; m, multiplet). Coupling constants (J) are recorded in hertz. Mass fragmentations were determined on a Finnigan MAT SSQ A electron impact mass spectrometer (EI MS, 70 eV).

1H NMR (200 MHz, DMSO-d6): δ 1.35 (3H, t, J = 7.2 Hz, CH₂CH₃ ), 1.15 (12H, s, CH₃ pic), 2.44 (2H, d, J = 3.7 Hz, NCH₂B), 3.38 (2H, q, J = 7.2 Hz, CH₂CH₃), 3.48–3.62 (2H, m, CH₂ pip), 3.82–3.96 (2H, m, CH₂ pip), 5.32 (1H, d, J = 7.3 Hz, CHα), 6.71 (2H, d, J = 8.5 Hz, meta), 7.19 (2H, d, J = 8.5 Hz, ortho), 8.41 (1H, t, J = 3.7 Hz, NH), 9.14 (1H, s, OH), 9.64 (1H, d, J = 7.3 Hz, NH).

¹³C NMR (50 MHz, DMSO-d6): δ 12.3, 25.1, 25.2, 40.8, 42.1, 43.3, 56.5, 83.3, 115.5, 128.7, 129.2, 152.3, 155.9, 157.5, 159.8, 170.6.

EI MS: m/z (%) 474 (M+, 0.2), 458 (0.4), 400 (2), 374 (2), 332 (14), 317 (10), 275 (16), 274 (96), 273 (36), 216 (19), 173 (9), 148 (18), 142 (58), 121 (35), 120 (46), 99 (100), 83 (15).

Genetic constructs and host strains

For large-scale protein expression and β-lactamase characterization, the bla_ADC gene (specifically, bla_ADC-7) was cloned into pET24a (+) vector (kanamycin resistant, Novagen, Madison, WI) following a previously published method (26). After sequencing verification, the correct construct was maintained in E. coli DH10B cells and transformed into E. coli BL21(DE3) cells for protein expression. For MIC determinations, bla_ADC was directionally cloned into the pBC SK(+) phagemid vector (chloramphenicol resistant, Stratagene, La Jolla, CA) as previously described (26). Briefly, the pET24a (+) bla_ADC construct was digested with XbaI and BamHI in Multi-Core buffer (Promega, Madison, WI) preserving the 5′ upstream flanking region from the pET24a (+) vector in front of the insert when ligated into pBC SK (+).

Antimicrobial susceptibility (MICs)

E. coli DH10B cells expressing the bla_ADC gene were phenotypically characterized by lysogeny broth agar dilution MICs. The MICs for various antibiotics were determined using a Steers Replicator™ that delivered 10 μl of a diluted overnight culture containing 10⁴ colony forming units. The cephalothin analogs 3 and 5 were tested at a constant concentration of 4 μg/ml in combination with either ceftazidime or cefotaxime.

β-lactamase purification

The ADC β-lactamase was prepared from E. coli BL21(DE3) cells after induction with isopropyl-β-D-thiogalactopyranoside (IPTG). Five hundred milliliter cultures were induced at an optical density at 600 nm of 0.5 to 0.8 (final IPTG concentration, 0.2 mM) at 37 °C for 4 hr in lysogeny broth. These cells were pelleted and resuspended in 50 mM Tris (pH 7.4) and β-lactamase liberated with lysozyme and EDTA per established methods (28). Accordingly, ADC protein was purified by preparative isoelectric focusing and fast protein liquid chromatography (FPLC) with a Sephadex Hi Load 26/60 column (Pharmacia, Uppsala, Sweden) (29). The enzyme was quantified, purity was assessed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis, and size verified by intact protein mass spectrometry (26).

Kinetics

Steady state kinetics were performed on an Agilent 8453 diode array spectrophotometer (Palo Alto, CA). Each continuous assay was performed in 10 mM phosphate-buffered saline at pH 7.4 at room temperature. K_i values was calculated by measuring the initial velocity in the presence of a constant concentration of enzyme (3 nM) and increasing concentrations of the inhibitors competed (ranging from 50 nM to 500 μM) against the indicator substrate nitrocefin (NCF) (BD Biosciences, San Jose, CA) (Δε₄₈₂ = 17,400 M⁻¹ cm⁻¹). The K_i values were corrected to account for the affinity of NCF for ADC using the following equation (30):

$$K_{\mathrm{i}}\left(\text{corrected}\right)=K_{\mathrm{i}}\left(\text{observed}\right)∕[1+(\left[\mathrm{S}\right]∕K_{\mathrm{m}\mathrm{NCF}}\left)\right]$$ Equation 1

Due to time-dependent inhibition of chiral boronic acid derivatives, compounds 4 and 5 were preincubated with enzyme for 5 min in phosphate-buffered saline before initiating the reaction with the addition of substrate, as described previously (15, 27, 31-33). In earlier experiments, preincubation of the achiral compound 3 with enzyme did not affect the K_i determination (data not shown).

Electrospray Ionization (ESI) Mass Spectrometry (MS)

Intact protein mass spectrometry was performed to determine products of inactivation. We incubated 14 μM of ADC for 15 min with and without compounds 2 and 5 and each carbapenem at an inhibitor: enzyme ratio of 20:1. Each reaction was terminated by the addition of 0.1% trifluoroacetic acid and immediately desalted and concentrated using a C₁₈ ZipTip (Millipore, Bedford, MA) according to the manufacturer’s protocol. Samples were then placed on ice and analyzed within 1 hr.

Spectra of the intact ADC: inhibitor proteins were generated on a Q-STAR XL Quadrupole-Time-of-Flight mass spectrometer (Applied Biosystems, Framingham, MA) equipped with a nanospray source. Experiments were performed by diluting the protein sample with 50% acetonitrile/0.1% trifluoroacetic acid to a concentration of 10 μM. This protein solution was then infused at a rate of 0.5 μl/min and the data were collected for 2 min. Spectra were deconvoluted using the Analyst program (Applied Biosystems). All measurements have an error of ± 3 atomic mass units (amu).

Molecular representations

The ADC model was generated by the SWISS-MODEL automated protein structure homology-modeling server, available at http://swissmodel.expasy.org, using the deposited GenBank ADC-7 protein sequence AY648950 and the Enterobacter aerogenes CMY-10 β-lactamase as a template (Protein Data Bank entry 1ZKJ) (34, 35). We optimized the generated model by energy minimization using Discovery Studio 2.1 software (Accelrys, San Diego, CA). The minimization was performed in several steps, using Steepest Descendent and Conjugate Gradient algorithm to reach the minimum convergence (0.02 kcal/mol*A). The protein was immersed in a water box, 7 Å from any face of the box, and the solvation model used was with periodic boundary conditions (PBC). The force-field parameters of CHARMm were used for minimization and the Particle Mesh Ewald method was used to treat long-range electrostatics. The bonds that involved hydrogen atoms were constrained with the SHAKE algorithm. Following equilibration, two separate 2 fs molecular dynamics simulations (Heating/Cooling and Production) at constant pressure and temperature (300 °K) were carried out for the ADC model. The trajectories were analyzed, and the minimum energy conformation was chosen.

To verify the quality of the ADC β-lactamase model, we used the Protein Structure and Model Assessment Tools available at http://swissmodel.expasy.org (see Supporting Information, Figures 1-3). The atomic empirical mean force potential (ANOLEA) evaluation of the model’s packing quality showed that 98% of the amino acids were in the favorable energy environment (36). We validated the stereochemical quality of the ADC model using the Procheck program which compares the geometry of protein residues with the stereochemical parameters of well-refined, high-resolution structures (37). Additionally, 97.5% of the non-proline, non-glycine residues in the ADC model were in the most favorable region of a Ramachandran Plot.

The Align Multiple Sequences function of Discovery Studio 2.1 allowed us to compare the generated ADC protein structure with that of the deposited crystal structure coordinates for the Enterobacter cloacae P99 enzyme (PDB entry 1XX2) and E. coli AmpC (PDB entry 2BLS). The program uses a method based on CLUSTAL W which aligns multiple sequences using a progressive pair-wise alignment algorithm (38). A multiple sequence alignment is generated, and secondary structure matches are graded as identical, strong, weak or non-matching are based on the calculated alignment score.

The minimized and equilibrated ADC model was used for constructing the acylation complexes of the ADC β-lactamase and the chiral cephalothin analog 5, imipenem, and meropenem ligands. The ligand structures were built using Discovery Studio Fragment Builder tools. The CHARMm force field was applied; the molecule was solvated with PBC and minimized using a Standard Dynamics Cascade protocol (one minimization using Steepest Descendent algorithm, followed by Adopted Basis Newton-Raphson algorithm and three subsequent dynamics stages at NVT and 300 °K).

The minimized ligands were docked in the active site of the enzyme using LibDock (39). The generated conformations (30-40) were manually analyzed and the most favorable ones chosen. The complex between the ligand and the enzyme was created, solvated, and energy minimized. The acyl-enzyme complex was created by making a bond with Ser64 and the assembly was further minimized using Conjugate Gradient algorithm with PBC to 0.001 minimum derivative. To reach the minimum equilibrium, the complexes were equilibrated using Molecular Dynamic Simulations.

Results

Kinetics

Table 1 summarizes our kinetic analysis of the inhibition of ADC β-lactamase. To establish a comparison, we list the previously reported K_is of the commercially available β-lactamase inhibitors against ADC; these K_is are not in the range that would translate into effective inhibition in MIC testing (26).

Table 1.

K_i s of inhibitors in direct competition assays with ADC

Inhibitor	Ki (μM)
Commercially available class A inhibitors
Clavulanate	4,275 ± 253a
Sulbactam	109 ± 3a
Tazobactam	91 ± 21a
Transition state analogs
Compound 1	0.60 ± 0.06
Compound 2	0.31 ± 0.03
Compound 3	0.78 ± 0.02
Compound 4	0.036 ± 0.008
Compound 5	0.011 ± 0.001
Carbapenems
Imipenem	1.3 ± 0.1
Ertapenem	5.8 ± 0.2
Doripenem	12.2 ± 0.4
Meropenem	19 ± 2

^aK_i values obtained previously by same methodology (26).

In contrast, we found that the boronic acid derivatives containing the R₁ side chain of cephalosporins bind the class C ADC with K_is in the nM range. The ceftazidime analog and the cefoperazone analog, compounds 1 and 2, respectively, show ≤ 1 μM K_is. Compound 5 with the cephalothin R₁ side chain and the meta-carboxyphenyl ring, which has a carboxylate that is designed to mimic the geometry and distances of the conserved C₄ carboxylate of cephalosporin β-lactams (Figure 3), had the lowest K_i for ADC (11 ± 1 nM). We interpret the 70-fold difference in K_i between compounds 3 and 5 to mean that the meta-carboxyphenyl moiety contributes significantly to the binding of this inhibitor with the ADC β-lactamase. However, compound 4, which lacks only the meta-carboxylate as compared to 5, also had a low K_i of 36 ± 8 nM.

Figure 3.

Overlay of the molecular coordinates for the E. coli AmpC covalently bound to cephalothin substrate (colored by atom, PDB entry 1KVM) and boronic acid chiral cephalothin analog, compound 5 (colored green, PDB entry 1MXO). The position of cephalothin’s dihydrothiazine ring and C₄ carboxylate is shown relative to the meta-carboxyphenyl group of compound 5, which is designed to mimic in stereochemistry and geometry the conserved β-lactam carboxylate.

Because inhibition reactions with the boronic acid derivatives are reversible, we quantified the binding energy contribution of these substituents by using K_i as an equilibrium constant in the Gibbs free energy equation (15):

$$\Delta \Delta G=-RT\ln \left(\right(K_{\mathrm{i}}5)∕(K_{\mathrm{i}}3\left)\right)$$ Equation 2a

Compared to the achiral cephalothin analog 3, we determined that the meta-carboxyphenyl group on the chiral cephalothin analog 5 contributes 2.5 kcal/mol in binding energy to ADC. The presence of the meta-carboxylate provides 0.7 kcal/mol of the 2.5 kcal/mol provided by this substituent:

$$\Delta \Delta G=-RT\ln \left(\right(K_{\mathrm{i}}5)∕(K_{\mathrm{i}}4\left)\right)$$ Equation 2b

Based on these calculations, we maintain that the presence of the phenyl group, which approximates the cephalosporin’s dihydrothiazine ring, is largely responsible for the low K_is of compounds 4 and 5.

Carbapenems are highly effective β-lactam antibiotics in the treatment of Gram-negative bacteria. Furthermore, carbapenems form prolonged acyl-enzymes with class C β-lactamases which can effectively inhibit the enzyme. We chose the four commercially available carbapenems to explore the determinants that contribute to the inactivation of ADC β-lactamase. Carbapenems, as ADC AmpC inhibitors, demonstrated low K_is (ranging from 1.3 ± 0.1 to 19 ± 2 μM). Comparing the carbapenems with different R₂ side chains, we see that the least substituted, imipenem, has the lowest K_i. The penem scaffold on which these β-lactams are constructed is similar; thus we assign the differences in K_i among these carbapenems to the interactions of the β-lactamase with the R₂ side chain.

ESI-MS and the nature of inactivation products

We performed timed ESI-MS with ADC, compounds 2 and 5, and the carbapenems to detect covalent intermediates in the inactivation pathway. As shown in Table 2 and Figure 4, analysis of the ADC: 2 and ADC: 5 reactions using ESI-MS shows that the β–lactamase is unmodified. This result is expected as boronates undergo reversible inhibition. In contrast, when ADC was reacted with the carbapenems, the predominant mass adduct formed corresponded to the sum of the molecular weights of the enzyme and the inhibitor, suggesting the formation of a non-fragmented covalent acyl-enzyme product. This result is consistent with MS data of the carbapenems forming acyl-enzymes with the class A SHV-1 β-lactamase (21). In addition, the ESI-MS analysis of each ADC: carbapenem spectra included a small adduct which was the mass equivalent of the carbapenem and β-lactamase minus 43 ± 3 Da, an observation made previously (40). We advance that there is a retroaldol elimination of the ligand’s C₆ hydroxyethyl substituent (see Figure 5 and discussion below).

Table 2.

ESI-MS analysis (atomic mass units, amus) of ADC alone and incubated with inhibitors^a

	Predicted molecular weight of β-lactamase or inhibitor	Species observed in deconvoluted spectra	Δ difference from β-lactamase molecular weight
ADC alone	40,631	40,638	7
ADC + Inhibitor
Compound 2	330	40,637	1
Compound 5	319	40,638	0
Imipenem	299	40,936	298
		40,893	255
Ertapenem	476	41,112	474
		41,069	431
Doripenem	420	41,058	420
		41,015	377
Meropenem	383	41,021	383
		40,979	341

^aAll measurements have an error of ± 3 amu.

Figure 4.

Deconvoluted mass spectra of: (A) ADC β–lactamase alone; (B) ADC after 15 min incubation with compounds 2 and 5; and (C) ADC β-lactamase after 15 min incubation with imipenem, ertapenem, doripenem, and meropenem. The peak in each ADC: boronate spectrum corresponds to the unmodified ADC enzyme. The major peak in each of the ADC: carbapenem spectrum indicates covalent attachment of the β-lactam with a minor additional peak corresponding to the acyl-enzyme without the carbapenem’s C₆ hydroxyethyl substituent. All measurements have an error of ± 3 atomic mass units (amu).

Figure 5.

Proposed mechanism of the retroaldolic reaction leading to elimination of C₆ hydroxethyl substituent from the β-lactamase: carbapenem acyl-enzyme. Glu272, supported by Lys315, may serve as the base to deprotonate the alcoholic function β–hydroxy carbonyl moiety of the C₆ substituent. Alternatively, Glu272 may abstract a proton from Lys315 which subsequently deprotonates the C₆ group (mechanism shown in red). The negative charge on the β-lactam carbonyl could be supported by Tyr150 and a Lys67.

Susceptibility testing

Nanomolar affinity inhibitors are clinically useful only if they can penetrate the outer cell wall of Gram-negative organisms and restore susceptibility to partner β-lactams. To this end, we performed MIC testing using compounds 3 and 5. Our results show that when ADC β-lactamase is expressed in the uniform E. coli DH10B background, the cephalothin analogs 3 and 5 lower MICs to ceftazidime and cefotaxime (16 → 4 and 8 μg/ml, and 8 → 1 and 2 μg/ml, respectively) (Table 3).

Table 3.

MIC values (μg/ml) of ceftazidime and ceftazidime in combination with 4 μg/ml of cephalothin analogs

	E. coli DH10B	E. coli DH10B blaADC
Ceftazidime	1	16
Ceftazidime/ Compound 3	1	8
Ceftazidime/ Compound 5	1	4
Cefotaxime	0.06	8
Cefotaxime/ Compound 3	0.06	2
Cefotaxime/ Compound 5	0.06	1

Our previous data showed that E. coli DH10B harboring bla_ADC have MICs of 0.06 μg/ml to meropenem, ertapenem, and imipenem-cilistatin (26). Thus, we did not perform MIC testing with the carbapenems in combination with the cephalothin analogs as a reduction in susceptibility would be difficult to detect.

Molecular representations

To understand the interactions between the carbapenems and high affinity cephalothin analog 5 in the absence of a crystal structure, we constructed a molecular model of ADC β-lactamase from a homology-modeling server. Accurate high resolution protein models can be generated from templates with greater than 50% sequence similarity; our model shared 66% sequence similarity with the template (41).

We first compared our ADC model to the defined crystal structures of E. cloacae P99 and the E. coli AmpC. An alignment based on the predicted, and known, secondary structures of ADC, P99, and E. coli AmpC shows that ADC shares 63% sequence similarity with both the P99 and E.coli AmpC β-lactamases (37% and 40% amino acid identity, respectively) (Supporting Information, Figure 4).

Using the representation of the ADC: 5 acyl-enzyme, we gained insight into how the Acinetobacter cephalosporinase interacts with compound 5. As the crystal structure of the E. coli AmpC in complex with the same boronic acid derivative has been solved (K_i = 1 nM, PDB entry 1MXO), we overlaid this structure on our generated model (Figure 6) (15). The overall tertiary structures of ADC and the E. coli AmpC are similar with conservation of the α-helix and β-sheet domains. The loops and turns between these secondary structures follow slightly different paths, but we note that these deviations may be part of the model construction and are allowable (e.g., the permission of increased flexibility for these strand regions).

Figure 6.

Overlay of molecular coordinates for the E. coli AmpC: 5 complex in yellow (PDB entry 1MXO) and generated ADC: 5 model colored by atom. The position of α–helices and β–sheets is generally preserved between the two proteins, but deviations are observed in the strand turns between these secondary structures. Active site differences are illustrated by the altered conformation of 5 (colored green) in ADC as compared to 5 (colored yellow) bound to the E. coli AmpC.

In the active site, the distance between the backbone amides of Ser64 and Ser318, which form the β-lactamase oxyanion hole or electrophilic center, are ~ 1 Å further apart in the ADC model than the E. coli AmpC structure (42, 43). The position of the backbones and side chains of the catalytically important Tyr150 and Lys 67 also varies by approximately 2 Å between the enzymes (44, 45). Thus, our model suggests that ADC may harbor a unique binding region as compared to the E. coli AmpC. These tertiary features are reflected in the disposition of compound 5 in the ADC acyl-enzyme model; in Figure 7, we show that the inhibitor may adopt different conformations in these class C β-lactamases. Comparing equivalent atoms of the boronic acid derivatives (e.g., the meta-carboxylate carbons or thiophene sulfur atoms) reveals a greater than 5 Å deviation in the configuration of the compounds in the overlaid structures. Crystal structures of boronic acid derivatives with AmpC enzymes typically show that one boronic acid oxygen atom is placed in the oxyanion hole formed by residues 64 and 318, and the other oxygen atom hydrogen bonds with Tyr150 (15, 17, 44, 46). In the ADC: 5 model, one of the boronic acid hydroxyl groups interacts with the Ser64 backbone carbonyl oxygen, but as the boronic acid and chiral substituents on the inhibitor are rotated approximately 120° compared to the E. coli β-lactamase structure, both oxygens are approximately 5 Å from Ser318 or Tyr150. Instead, Tyr150 is within ~ 3 Å of both the carbonyl oxygen and thiophene ring sulfur atom from the cephalothin R₁ group of the boronic acid derivative. Our model also shows a hydrogen bond between this R₁ carbonyl oxygen and Lys67. Thus, the residues contributing to the high affinity of this inhibitor for these two β-lactamases may play different roles in each AmpC.

Figure 7.

Comparison of the binding site interactions between E. coli AmpC: 5 (left panel) and ADC: 5 (right panel). Figures have a perspective view to show positions of the residues in relation to the inhibitor. The boronic acid derivative is bound to Ser64 in both structures, but the relative rotation of the inhibitor in ADC changes the relationships with other active site residues. Specifically, the boronic acid oxygens interact with Ala318 and Tyr150 in E. coli AmpC, but the hydrogen bond with Ser318 is lost in ADC. Also in ADC, the meta-carboxylate of the dihydrothaizine ring analog has no clear interaction with previously identified carboxylate binding residues (e.g, Asn346, Arg349, or like Asn289 in E. coli AmpC). Instead, the group may form a long hydrogen bond with Asn287. The carbonyl oxygen of the R₁ cephalothin analog interacts with Asn152 in E. coli AmpC, but is reoriented towards Lys67 in ADC. Lastly, the R₁ thiophene ring sulfur in ADC: 5 is moved towards Tyr150 as compared to the E. coli AmpC: 5 structure. Overall, these significant active site differences suggest that while ADC may possess novel architecture, the ability to recognize inhibitors and substrates is preserved because of the versatile functions of the binding site residues.

In our kinetic studies, we noted 15-fold differences between the K_i values of the highest affinity carbapenem (imipenem) and the lowest affinity carbapenem (meropenem). As these compounds differ primarily by their R₂ side chains, we created models of the ADC: imipenem and ADC: meropenem acyl-enzymes to explore the K_i contributions of these substituents (Figures 8A and 8B). Based on the MS results indicating the presence of species corresponding to the acyl-enzyme both with and without the C₆ hydroxyethyl group of the carbapenems, we constructed representations of intact imipenem and meropenem as well as these compounds without their C₆ hydroxyethyl group. Our models predict that when the hydroxyethyl group is present, the carbonyl oxygen from the β-lactam ring of both imipenem and meropenem is located outside of the enzyme’s electrophilic, or oxyanion, hole created by the backbone nitrogen atoms of Ser64 and Ser318 (47, 48). Rather, the imipenem β-lactam carbonyl is hydrogen bonded to Lys67 and the meropenem β-lactam carbonyl is only 1.5 Å from Tyr150. The R₂ side chain for both intact carbapenems is oriented out of the active site in the acyl-enzyme.

Figure 8.

Molecular representation of: (A) ADC: imipenem acyl-enzyme model; and (B) ADC: meropenem acyl-enzyme model. The intact carbapenem is shown in green and the carbapenem without the C₆ hydroxyethyl group in orange. Hydrogens are not shown except on the carbapenem C₆ hydroxyethyl which is likely deprotonated by Glu272, leading to elimination of the group. Removal of this C₆ substituent precipitates significant reorientation of the compound in the active site. Specifically, the β-lactam carbonyl moves back towards the oxyanion hole formed by the backbone nitrogens of Ser 64 and Ser318, approximately 90° rotation for imipenem and entirely into the hole for meropenem. Also after C₆ elimination, the R₂ group of imipenem is repositioned from outside of the binding pocket into a network of interactions with Tyr150, Asn152, Lys67 and Gln120. Less change is observed for the R₂ group of meropenem, which may have implications for the differing K_is of these carbapenems.

In contrast, when the C₆ hydroxyethyl group is removed, the conformation of the carbapenem is significantly changed. Most notably, the β-lactam carbonyl rotates towards the oxyanion hole, approximately 90° in imipenem, and entirely into the electrophilic pocket for meropenem. Further, the R₂ side chain of imipenem flips back towards the active site so that the terminal amide group is ~ 11 Å from its position in the complex with the C₆ group. The imipenem R₂ group now interacts with Tyr150, Asn152, Lys67, and Gln120. Modeling without the hydroxyethyl group for meropenem produces less change in the conformation of the R₂ group, which remains oriented outside of the binding site. For both imipenem and meropenem, the model without the C₆ hydroxyethyl group is more energetically favorable, as calculated by the final potential energy of the complexes (Δ - 520 and Δ - 80 kcal/mol, respectively).

Discussion

Our analysis shows that high affinity inhibition of the ADC β-lactamase, a class C cephalosporinase of increasing medical importance, is a realistic goal. We assayed two types of inhibitors against ADC: i) compounds that resemble the natural substrate for the Acinetobacter cephalosporinase; and ii) the currently available carbapenems. This approach teaches us important lessons about the inhibition of this clinically challenging β-lactamase and elucidates contributions of R₁ and R₂ side chains. We begin with an examination of the data revealing the low μM K_is of the ADC β-lactamase by the carbapenems, and then discuss how the boronic acid derivatives, as chemical probes, yield important insights into the nature of class C enzyme active sites.

After incubation of ADC and each carbapenem, our ESI-MS data reveals the formation of two molecular species. The mass of the predominant species corresponds to the intact carbapenem acylating ADC; the mass of the minor species corresponds to the acyl-enzyme less 43 Da (Table 2). Based on previous MS studies in our and other laboratories, we assign the major peak to the carbapenem acyl-enzyme species (21, 40). Formation of a stable acyl-enzyme is supported by previously defined crystal structures of carbapenems and class A and C β-lactamases (18, 20, 21). Our ADC: imipenem and ADC: meropenem acyl-enzyme models show conformations where the β-lactam carbonyl oxygen is not found in the oxyanion hole or electrophilic center formed by the backbone nitrogens of residues Ser64 and Ser318. This observation is consistent with the X-ray crystal structure of the E. coli AmpC β-lactamase with imipenem where the carbonyl was positioned approximately 180° outside of the oxyanion hole (18). Additional crystal structures of class A β-lactamases in complex with carbapenems have also shown this repositioning of the β-lactam carbonyl (20, 21). This displacement is likely precipitated by steric interactions induced by the carbapenems’ C₆ hydroxyethyl groups, producing a conformational change that forces the carbonyl away from the oxyanion hole and into a position unfavorable for hydrolysis (20, 42, 43). Taken together, this reasoning offers an explanation for the inhibition of the Acinetobacter cephalosporinase by the carbapenems.

Secondly, we observed a minor peak in each ADC: carbapenem spectrum that reflects the elimination of the carbapenem C₆ hydroxyethyl group. This observation was reported previously in the class A Mycobacterium tuberculosis blaC (40). Our molecular representations of imipenem and meropenem in complex with ADC give us insights into how the carbapenems are behaving in the active site of ADC following this elimination. When the C₆ hydroxyethyl group is removed from the carbapenem, both compounds adopt new positions where the β-lactam carbonyl moves to be either entirely in the oxyanion hole (meropenem) or rotated back toward the hole approximately 90° (imipenem). This prediction is similar to the conclusion drawn from the X-ray crystallographic evidence of the class A Asn132Ala TEM enzyme variant which demonstrated that the substitution allowed the β-lactam carbonyl to rotate back into the oxyanion hole (49). We speculate that removal of the C₆ group is an alternative mechanism of alleviating the steric clashes induced by this substituent, allowing repositioning of the acyl-enzyme. When the β-lactam carbonyl is aligned in the oxyanion hole, the conformation is more compatible with hydrolysis, and increased turnover is likely, consistent with the relatively small amount of this species evident on ESI-MS. Furthermore, we observe this process after a 15 min incubation, which is within the bacterial generation time (i.e., 20 minutes) and suggests that this elimination may be occurring in cells.

The mechanism of elimination of the hydroxyethyl group is likely to be a retroaldolic-type reaction of the β-hydroxyethyl moiety of the β-lactamase acyl-enyzme (Figure 5). We propose that Glu272, supported by Lys315, may serve as the base to deprotonate directly the alcoholic function of the C₆ substituent. Alternatively, Glu272 may abstract a proton from the amine side chain of Lys315, and then Lys315 would then deprotonate the carbapenem C₆ alcohol group. In either case, the negative charge on the β-lactam carbonyl could be stabilized by Tyr150 and Lys67. Interestingly, this proposes another role for Tyr150, which is already implicated in both acylation and “substrate-activated catalysis” in AmpC enzymes (44, 45, 50, 51). This mechanism is consistent with our molecular representations which show Glu272 in hydrogen bond distance of the C₆ hydroxyethyl group for both imipenem and meropenem; the Lys315, Tyr150, and Lys67 residues are also positioned to support this reaction (see Figures 8A and 8B). Further investigation of these residues and their potential roles in the retroaldolic-type reaction will help elucidate the C₆ elimination mechanism.

Structural and kinetic studies provide evidence that, following acylation of carbapenems by β-lactamases, the acyl-enzyme formed can tautomerize between a Δ²- and Δ¹-pyrroline species which have differing rates of deacylation (19, 52-55). We modeled both the tautomers in our molecular representations with ADC, but the interactions between the enzyme and the carbapenems were not significantly different for either the Δ¹- and Δ²-species. We anticipate that the Δ²-Δ¹ tautomerization exists as both a separate and integrated pathway to C₆ hydroxyethyl group elimination, and plan further examination of the reaction and its implications for inhibition.

We next turn our attention to the contribution of the R₂ side chain in the differing K_is of the carbapenems for the ADC β-lactamase. The four carbapenems tested share a common β-lactam ring scaffold and vary by their R₂ substituents, yet we observed up to 15-fold differences in K_i values. Our models offer insights for how these side groups interact with the ADC enzyme, and suggest that each carbapenem may behave uniquely. In both the imipenem and meropenem models including the C₆ hydroxyethyl group, the R₂ side chain is oriented out of the active site, and does not engage in significant interactions with the enzyme. This outward conformation is also seen in the E. coli AmpC: imipenem, TEM: imipenem, and SHV: meropenem crystal structures (PDB entries 1LL5, 1BT5, 2ZD8, respectively) (18, 20, 21). However, upon removal of the C₆ hydroxyethyl group, the R₂ side chain of imipenem rotates towards the binding pocket and interacts with several active site residues including Tyr150, Asn152, Lys67 and Gln120. In contrast, the R₂ group of meropenem modeled without the C₆ hydroxyethyl group is found in a similar conformation to the intact meropenem acyl-enzyme model. In light of the K_i measurements for these two carbapenems, the molecular representations predict that the interactions between the R₂ of imipenem and ADC may stabilize this form of the compound in the active site, while for meropenem, the interactions with the enzyme are less favorable (e.g., electrostatically or sterically). Notably, the R₂ group of imipenem is the least substituted of the carbapenems, and the additional atoms and ring structures of meropenem, doripenem, and ertapenem may alter the K_is by various mechanisms, e.g., limiting conformational flexibility necessary for rotating back towards the active site to make favorable interactions. We note that the R₂ side chain of imipenem in the crystal structure with TEM Asn132Ala remains oriented out of the active site, despite the alleviation of the steric strain caused by the C₆ substituent (PDB entry 1JVJ) (49). That our model of imipenem without the C₆ hydroxyethyl group leads to significant R₂ conformational change may reflect inherent differences between the inhibition of class A and C β-lactamases by carbapenems, perhaps partly due to the increased size of the active site in class C enzymes (56).

We now highlight the versatile β-lactamase inhibitory activity of the rationally designed boronic acid derivatives (13, 14). All the cephalosporin analogs had K_i values ≤ 1 μM, and likely these values reflect the naturally high affinity of AmpC enzymes for cephalosporin substrates (24). Interestingly, compounds 1-3 which contain only the cephalosporin R₁ side chains, have very similar K_is, suggesting limited differences in the affinity gains of these R₁ structures. The kinetic substrate profile of ADC, and AmpCs in general, includes increasing affinities for the larger side chains of “third-generation” cephalosporins, which would include ceftazidime and cefoperazone (24, 26, 57). “First-generation” cephalosporins, such as cephalothin, typically have lower affinities, but higher hydrolytic rates. Thus, despite these differences in affinity and hydrolytic rates, the “first-“ and “third-generation” cephalosporin analogs have comparable K_is as inhibitors.

However, the introduction of chirality and the substituent resembling the dihydrothiazine ring of the cephalosporin nucleus leads to 22-fold increase in affinity (compound 3 vs 4). The further addition of the C₄ carboxylate gains 70-fold over the achiral counterpart (compound 3 vs 5). Compound 5, which incorporates these multiple structural features of the cephalosporins, displays the lowest K_i for the ADC enzyme. Our results indicating the contribution of the meta-carboxyphenyl ring on the K_i of compound 5 are consistent with previous data demonstrating the importance of this functional group for molecular recognition in class C AmpC β-lactamases (15, 31). The crystal structure of compound 5 in complex with the E. coli AmpC β-lactamase shows a hydrogen bond between the carboxylate of the inhibitor and the amide of Asn289 (15). This interaction is well-studied and thermodynamic cycle experiments revealed that the hydrogen bond contributes 1.7 kcal/mol to the overall binding affinity, a value within the range for ion-dipole interaction (58). However, Asn289 is not a well conserved residue among class C β-lactamases. The K_i of this cephalothin analog increases from 1 nM with the E.coli AmpC to 29 nM with the E. cloacae P99 AmpC which has a Ser289, suggesting that the amino acid at position 289 plays a role for the affinity of compound 5 (47, 58).

Based on amino acid sequence, the Acinetobacter ADC cephalosporinase is not closely related to other AmpCs, and has a Glu residue at position 289 (Supporting Information, Figure 4). Glu is not a hydrogen bond donor, and our model shows the Glu289 side chain is well outside hydrogen bond distance (~ 7 Å). Thus, we used our molecular representation of the ADC: 5 complex to search for other residues that could be hydrogen bonding with the meta-carboxylate. Previous structural and functional studies of AmpC enzymes suggest that the more conserved sites Xaa343, Asn346, Arg349 or Thr316 interact with the C₃/C₄ carboxylate of the substrate, although none of these residues were involved with the meta-carboxylate in the E. coli: 5 crystal complex (15, 18, 46-48, 59-61). Similarly, each of these residues is at least 5 Å from the meta-carboxylate in our ADC: 5 representation, an unlikely distance for a high energy hydrogen bond with the group. Asn287 is an “ancillary” ADC residue which may be capable of engaging the meta-carboxylate in a hydrogen bond with ion-dipole character, but is positioned approximately 8 Å away in our model (Figure 7). The molecular explanations of the ADC inhibition by the cephalothin analogs 4 and 5 remain to be validated by further study with boronic acid derivatives as molecular probes, site-directed mutagenesis of ADC, and/or crystallography.

AmpC enzyme binding site “hot spots” were previously identified by a comparison of crystal structures in complex with both boronic acid inhibitors and β-lactam substrates (46). Our ADC acyl-enzyme representation reveals that the recognition elements may differ for the ADC β-lactamase. For example, the E. coli AmpC hydroxyl binding site was defined by Tyr150 and its hydrogen bond with one of the boronic acid hydroxyls, displacing the deacylation water (15, 44). Our ADC: 5 model shows significant repositioning of the boronic acid group, making this interaction with Tyr150 unlikely. Rather, the ADC Tyr150 is within ~ 3 Å of the thiophene sulfur and carbonyl oxygen found in the R₁ side chain of compound 5. Further, the R₁ amide recognition site formed by the interaction of the R₁ carbonyl and Asn152 in the E. coli AmpC differs from our ADC: 5 representation, as this same side chain carbonyl hydrogen bonds with Lys67 (46). These consensus binding sites were compiled exclusively from crystal structures of the E. coli AmpC, and our kinetic data and modeling analyses indicate the ADC β-lactamase may interact differently with boronic acid inhibitors than other class C enzymes (15, 46, 58). Our molecular representations of ADC in complex with the inhibitors were useful for developing hypotheses; however, we remain cognizant of modeling limitations, such as the lack of active site flexibility and the removal of water molecules during the ligand docking protocol.

That the chiral cephalothin analog 5 can maintain low nM K_i for several phylogenetically divergent AmpC β-lactamases reflects not only the potency of this inhibitor, but also what may be an important plasticity of AmpCs (26, 46, 58). The significant repositioning of the boronic acid derivative revealed in our ADC: 5 model may be an indication of this enzyme’s versatility, causing the β-lactamase: ligand interactions to have different molecular correlates. We posit that compounds 4 and 5 benefit from the presence of an additional side chain which more closely resembles the dihydrothiazine ring of the natural substrate, cephalosporins. The stereochemistry and conformation of the chiral inhibitors may create a better “fit” for this enzyme. In part, this improved fit may be due to approximation of the deacylation transition state of the cephalosporinase, a theory which has been previously offered to explain the high affinity of chiral boronic acid derivatives (33, 44).

Hence, the notion of dedicated AmpC enzyme R₁ and R₂ binding sites may be especially fluid and adaptable in ADC, permitting the β-lactamase to change recognition elements depending on the ligand (46, 56). These novel interactions may reflect that differences in primary sequence can be compensated by common secondary and tertiary structures, allowing the enzyme to use multiple ancillary residues to make contact with substrates and inhibitors. Alternatively, ADC may have subtle differences in its deacylation mechanism which is suggested by the unanticipated position of the boronic acid oxygen atoms in our model. This AmpC structure-function redundancy merits further study, as it could both lie at the core of why these β-lactamases have evolved as versatile “traps” of cephalosporin substrates, but also aid the careful design of broad-spectrum inhibitors (62, 63).

Despite the description of boronates as β-lactamase inhibitors since the 1970’s, boronic acid derivatives have not yet been developed for clinical use in combination with a β-lactam (64). Concerns about the safety and efficacy of boron-containing therapeutics are currently being addressed by clinical studies sponsored by the pharmaceutical industry (65). The data presented in this paper encourages the in vivo study of boronates as β-lactamase inhibitors.

Conclusion

In summary, we provide important insights into the interaction of two types of inhibitors with the Acinetobacter and other clinically relevant cephalosporinases. Firstly, we present kinetic data and molecular representations that explain why carbapenems are effective inhibitors of class C enzymes, including formation of a stable acyl-enzyme and a role for the compounds’ R₂ side groups. Our results add to increasing evidence supporting the activity of carbapenems as broad-spectrum β-lactam antibiotics, “slow substrates,” and inactivators of class A and C β-lactamases (18-21, 23-25). Secondly, our ADC model suggests that inhibitors designed to mimic the structure of natural substrates (i.e., boronic acid analogs) may adopt unique conformations in different class C active sites. Despite significant sequence and structure dissimilarity between ADC and the E. coli AmpC, the chiral cephalothin analogs attain similar K_is for both enzymes. This versatility may reflect an important plasticity of this cephalosporinase β-lactamase. Our data offers promise for the development of compounds that have an extended inhibition profile across, and within, β-lactamase classes, and specifically against this challenging Acinetobacter spp. target.

ABBREVIATIONS

ADC

Acinetobacter-Derived Cephalosporinase

amu

atomic mass units

EI MS

electron impact mass spectrometer

ESI-MS

electrospray ionization mass spectrometry

IPTG

isopropyl-β-D-thiogalactopyranoside

MIC

minimum inhibitory concentration

MDR

multidrug resistant

NCF

nitrocefin

PDB

periodic boundary conditions

PDB

Protein Data Bank

THF

tetrahydrofuran