2-Mercaptomethyl thiazolidines (MMTZs) inhibit all metallo-β-lactamase classes by maintaining a conserved binding mode

Abstract

Metallo-β-lactamase (MBL) production in Gram-negative bacteria is an important contributor to β-lactam antibiotic resistance. Combining β-lactams with β-lactamase inhibitors (BLIs) is a validated route to overcoming resistance, but MBL inhibitors are not available in the clinic. Based on zinc utilization and sequence, MBLs are divided into three subclasses, B1, B2 and B3, whose differing active-site architectures hinder development of BLIs capable of “cross-class” MBL inhibition. We previously described 2-mercaptomethyl thiazolidines (MMTZs) as B1 MBL inhibitors (e.g. NDM-1) and here show that inhibition extends to the clinically relevant B2 (Sfh-I) and B3 (L1) enzymes. MMTZs inhibit purified MBLs in vitro (e.g. Sfh-I, K_i 0.16 μM) and potentiate β-lactam activity against producer strains. X-ray crystallography reveals that inhibition involves direct interaction of the MMTZ thiol with the mono- or di-zinc centers of Sfh-I/L1, respectively. This is further enhanced by sulfur-π interactions with a conserved active site tryptophan. Computational studies reveal that the stereochemistry at chiral centers is critical, showing less potent MMTZ stereoisomers (up to 800-fold) as unable to replicate sulfur-π interactions in Sfh-I, largely through steric constraints in a compact active site. Furthermore, in silico replacement of the thiazolidine sulfur with oxygen (forming an oxazolidine) resulted in less favorable aromatic interactions with B2 MBLs, though the effect is less than previously observed for the subclass B1 enzyme NDM-1. In the B3 enzyme L1, these effects are offset by additional MMTZ interactions with the protein main chain. MMTZs can therefore inhibit all MBL classes by maintaining conserved binding modes through different routes.

Graphical Abstract

Introduction

β-Lactams are the most prescribed antibiotic class worldwide, but their efficacy is increasingly challenged by the growing problem of antibiotic resistance.¹ The production of β-lactamases in Gram-negative bacteria is the major resistance mechanism in the clinic, as members of this large enzyme family are capable of inactivating all β-lactam antibiotics.² β-Lactamases can be divided into two mechanistic groups, the serine-β-lactamases (Ambler classes A, C and D^{3, 4}, SBLs) and the zinc ion dependent metallo-β-lactamases (class B⁴, MBLs). SBL-catalyzed hydrolysis involves attack of a nucleophilic serine on the β-lactam ring, and occurs with formation and resolution of a covalent acyl-enzyme intermediate via labile tetrahedral species.⁵ Conversely, MBLs utilize an active site water/hydroxide to hydrolyze the β-lactam ring without formation of a covalent intermediate (Figure 1A).⁵

Figure 1. MBL-catalyzed β-lactam breakdown and MBL inhibitors.

(A) Penicillin breakdown by MBLs, resulting in penicilloic acid product formation via a proposed high-energy tetrahedral intermediate. (B) Inhibitors based on proposed species in β-lactam hydrolysis. Left, bisthiazolidine (BTZ) scaffold, R=H/CH₃; middle, bicyclic boronates (e.g. taniborbactam, QPX7728); right, 2-mercaptomethyl thiazolidine (MMTZ) scaffold, R=H/CH₃. (C) 2-Mercaptomethyl thiazolidines studied here. The absolute R/S configurations of the chiral carbon centers (C-2 and C-4, labelled) are shown. L-anti-1a (brown), D-anti-1a (orange), L-anti-1b (green), L-syn-1b (cyan), D-anti-1b (purple) and D-syn-1b (pink). (C) is adapted from Rossi et al.⁶

MBLs can be further subdivided into the B1, B2 and B3 subclasses, based on active site architecture and zinc utilisation.^{5, 7} The B1 enzymes,⁸ such as NDM-1, VIM-2 and IMP-1, are plasmid mediated and widespread, produced in Gram-negative bacteria such as Klebsiella pneumoniae, Escherichia coli and Pseudomonas aeruginosa. B1 enzymes have a wide substrate specificity and can hydrolyze almost all β-lactam antibiotics, including the carbapenems, such as imipenem and meropenem, that were once considered antibiotics of last-resort. The active site of B1 enzymes is conserved, containing two Zn(II) ions, termed Zn1 and Zn2, that are coordinated by three amino acids each. Zn1 is coordinated by three histidine residues (His116, His118 and His 196; standard MBL numbering throughout⁹) and Zn2 by Asp120, Cys221 and His263. The Zn(II) ions are bridged by a water/hydroxide that is thought to act as the nucleophile in β-lactam hydrolysis. B2 MBLs are mono-zinc enzymes, containing a zinc ion coordinated as in the Zn2 site of B1 MBLs. To date, they have been identified on the chromosomes of the Serratia fonticola (Sfh-I¹⁰), Aeromonas sp. (CphA¹¹), and Pseudomonas sp. (PFM¹²), organisms capable of causing a range of opportunistic infections in the clinic. B2 MBLs are carbapenemases but have a narrower substrate profile than the B1 enzymes, as they are unable to hydrolyze most penicillins and cephalosporins.^13–16 In contrast, in the di-zinc B3 enzymes, Zn1 is coordinated as in B1 MBLs, but Zn2 by Asp120, His121 and His263, resulting in a different active site architecture to the B1 and B2 MBLs. They are found on the chromosomes of numerous environmental and pathogenic Gram-negative bacteria, including Stenotrophomonas maltophilia (L1^{17, 18}), Elizabethkingia meningoseptica (GOB^{19, 20}), and Pseudomonas otitidis (POM-1²¹). Some have now been shown to be incorporated into the chromosomes of pathogenic bacteria via a mobile genetic element,^{22, 23} indicating the potential for wide dissemination. Taken together with their wide substrate spectrum, that includes efficient carbapenem hydrolysis, B3 enzymes could pose an increasing clinical concern.

Inhibitors closely mimicking β-lactam molecules (e.g. clavulanic acid and tazobactam), diazabicyclooctanes (DBOs, e.g. avibactam) and vaborbactam (a cyclic boronate) are now used in the clinic to inactivate some serine-β-lactamases through covalent attachment to the nucleophilic serine.^24–26 However, the mechanistically distinct MBLs are not inhibited and can hydrolyze the β-lactam based inhibitors.²⁷ This has prompted extensive investigations into MBL inhibitor development, with bicyclic boronates currently representing the most promising compounds (Figure 1B).^{28, 29} For example, taniborbactam (formerly VNRX-5133), currently in late-stage clinical development, is a potent inhibitor of most SBLs and B1 MBLs.^28–31 In addition, QPX7728 is in phase 1 clinical trials and has an improved spectrum of activity compared to taniborbactam, particularly against SBLs and the B1 MBL IMP-1.^{32, 33} Despite these significant advances in MBL inhibitor development, there remains a need for further exploration of compounds active against all MBL subclasses that will increase our available armamentarium to combat antibiotic resistance and guard against the possible future dissemination of enzymes beyond subclass B1.

We considered that common features of species populated during antibiotic hydrolysis by diverse MBLs could be exploited for inhibitor development.^{34, 35} To this end, we previously reported on the synthesis of a series of bisthiazolidines (BTZs) that were designed as bicyclic substrate mimics, containing a carboxylate and additional free thiol as a zinc-binding moiety (Figure 1B).³⁶ We demonstrated that BTZ stereoisomers inhibited all MBL subclasses by adopting multiple binding modes in the structurally different active sites.³⁷

More recently, we described a series of 2-mercaptomethyl thiazolidines (MMTZs) that were designed to exploit common features of the binding of hydrolyzed β-lactam products to the MBL active site (Figure 1B).⁶ These compounds contain a thiazolidine ring and two chiral carbon centers with carboxylate and free thiol groups (Figure 1C). Some MMTZs also include a gem-dimethyl group that is also present in penicillins and some BTZs. They inhibit the clinically important B1 enzymes NDM-1, VIM-2 and IMP-1 by adopting consistent binding modes that exploit thiol-Zn coordination, and an interaction of the thiazolidine sulfur with a conserved aromatic residue in the active site.

Here, we utilize enzyme kinetics, microbiology, X-ray crystallography and computational chemistry techniques to demonstrate that MBL inhibition by MMTZs extends to the subclass B2 and B3 enzymes, Sfh-I and L1. These data further explore the roles of conserved interactions in the active site, including the sulfur-π interaction, in maintaining a conserved binding mode. The results highlight that exploiting common features of the MBL active site can represent a tractable route towards achieving cross-class inhibition and new drug development.

Results and Discussion

MMTZs are cross-class MBL inhibitors in vitro.

To determine the inhibition profile of MMTZs against MBLs, we first measured inhibitory constants (K_i values) against the clinically relevant B2 and B3 enzymes Sfh-I and L1 by monitoring imipenem hydrolysis in the presence of MMTZs (Figures S1 and S2). Inhibition profiles could then be compared with our previously collected K_i data for the B1 MBLs (Table 1).

Table 1.

MMTZ inhibition constants (K_i, μM) against purified, recombinant metallo-β-lactamases

	B2 MBL	B3 MBL	B1 MBLsa,b
Inhibitor	Sfh-I	L1	NDM-1	VIM-2	IMP-1
L-anti-1a	0.16 ± 0.03	10.0 ± 0.8	5.2 ± 0.7	0.38 ± 0.05	1.0 ± 0.2
D-anti-1a	20 ± 2	4.0 ± 0.5	2.5 ± 0.5	0.39 ± 0.04	1.3 ± 0.1
L-anti-1b	1.3 ± 0.1	20 ± 2	0.44 ± 0.06	0.75 ± 0.09	0.46 ± 0.05
L-syn-1b	100 ± 10	28 ± 3	8 ± 1	3.6 ± 0.4	6.0 ± 0.6
D-anti-1b	130 ± 10	1.4 ± 0.2	3.1 ± 0.3	0.9 ± 0.1	0.93 ± 0.08
D-syn-1b	1.0 ± 0.1	4.0 ± 0.6	0.6 ± 0.05	1.9 ± 0.1	2.0 ± 0.2
L-BTZ-1b	0.26 ± 0.03	12 ± 1	7 ± 1	2.9 ± 0.4	8 ± 2
D-BTZ-1b	26 ± 3	10 ± 1	19 ± 3	3.2 ± 0.4	6 ± 1
L-BTZ-2b	0.36 ± 0.04	11 ± 2	18 ± 3	6 ± 1	15 ± 3
D-BTZ-2b	29 ± 3	10 ± 1	12 ± 1	14 ± 3	10 ± 2

^aB1 MMTZ data are from Rossi et al⁶ following the same protocol as used here.

^bAll BTZ data are from Hinchliffe et al, calculated following the same protocol as used here.³⁷ BTZs −1 and −2 are with and without a gem-dimethyl group, respectively. See Figure S3A for chemical structures of the L-/D-BTZ stereoisomers.

Our results show that MMTZs potently inhibit all MBL subclasses, although there are clear differences between different enantiomers. This is most apparent with Sfh-I for which K_i values range from 0.16 μM to 130 μM, an ∼800-fold difference. In particular, there are ∼100-fold reductions in K_i between L/D-anti-1a, L/D-anti-1b, and L/D-syn-1b. Further, compounds in which C-2 of the thiazolidine ring is in the (R)-configuration have poor potency compared to the (S)-isomer, with K_i values of 20, 100 or 130 μM (D-anti-1a, L-syn-1b and D-anti-1b, respectively). We also found BTZs to be 100-fold less potent against Sfh-I when the chiral carbon center bearing the mercaptomethyl group is in the S-configuration, i.e. the D-enantiomer (Figure S3A). However, unlike the BTZs, the presence of a gem-dimethyl group in MMTZs also adversely affects their potency against Sfh-I, with a c. 10-fold reduction in K_i between L-anti-1a/1b and D-anti-1a/1b. Variation in the inhibitory activity of MMTZs against L1 is less apparent (as is also the case for BTZs), although a slight preference for the D-over the L-isomer is observed (2.5-/7-/14-fold K_i increases for anti-1a, anti-1b and syn-1b L/D stereoisomers, respectively). The two compounds most active against the range of MBLs tested here, and previously,⁶ are L-anti-1a and D-syn-1b. In contrast, D- and L-captopril, compounds with a free thiol and pyrrolidine ring (Figure S3B), display particularly poor potency against a B2 enzyme (K_i’s 72/950 μM against CphA), and are up to 20-fold less potent than MMTZs against B3 enzymes (K_i’s 20/8.8 μM against L1/SMB-1) (Table S1).

We next sought to determine whether MMTZs could potentiate antibiotic activity against recombinant E. coli expressing NDM-1, Sfh-I and L1, by measuring their effect on the minimal inhibitory concentration (MIC) of imipenem (chosen as Sfh-I predominantly hydrolyzes carbapenems, and for consistency with our previous work⁶) in broth microdilution assays (Figure 2).

Figure 2. Effect of MMTZs on imipenem minimum inhibitory concentrations of MBL-expressing E. coli.

The minimum inhibitory concentration of imipenem (IMI) +/− 100 μg/mL MMTZ was determined for the E. coli strain DH10B expressing NDM-1, Sfh-I or L1 (see Methods for details). At this concentration, in the absence of antibiotic, MMTZs do not have a detrimental effect on bacterial growth. An asterisk denotes the inhibitor had a four-fold effect on MIC (i.e. two dilutions). A two-fold effect (i.e. one dilution factor) is not considered significant. Results are the mode of three biological replicates.

At a concentration of 100 μg/mL, MMTZs display variable effectiveness in lowering the imipenem MIC of MBL-expressing E. coli DH10B. Although some of the compounds only cause a two-fold reduction in imipenem susceptibility, a number result in a more significant four-fold reduction (i.e. a factor of two dilutions, see asterisks in Figure 2). Indeed, compound L-anti-1a was the most effective, causing a 4-fold MIC reduction against E. coli producing NDM-1, Sfh-I or L1. The compound that was most potent against purified, recombinant MBLs, D-syn-1b, surprisingly showed the least effect against these MBL-expressing bacteria. To test whether lower concentrations of inhibitor could potentiate activity, we tested the effect of 50 and 75 μg/mL on the imipenem MIC against the L1-expressing E. coli DH10B (Table S2). These data indicate that 50 μg/mL of L-anti-1b, D-syn-1b and L-syn-1b all caused a four-fold reduction in imipenem MIC, indicating lower concentrations of inhibitor can potentiate imipenem activity against a lab strain carrying an efficient carbapenemase. However, we note that concentrations of 100 μg/mL were required for effective imipenem potentiation against a number of clinical strains expressing B1 MBLs.⁶

We have previously reported that MMTZs are relatively stable in solution (PBS), with 12% forming disulfides after 6 hours, and 88% remaining intact.⁶ It is possible that, over the time course of an MIC assay, this may have a small, but limited, effect on the MIC profiles, contributing to the discrepancy between K_i values and MICs, but we consider it unlikely that this is the sole explanation. We note that penetration into the bacterial periplasm may also be a contributing factor, as our previously determined in cell IC₅₀ values against NDM-1 were also slightly elevated compared to in vitro K_i values.⁶

MMTZs therefore inhibit all MBL subclasses and can potentiate carbapenem activity against recombinant E. coli expressing enzymes from all three MBL classes. These in vitro data also highlight the importance of considering inhibitor stereochemistry for cross-class inhibition, with stereochemical preferences particularly marked in the case of the B2 enzymes which have a narrower substrate specificity. In contrast, inhibition of the B1 MBLs was less affected by stereochemistry, with the B3 enzymes displaying moderate selectivity towards the different stereoisomers.

X-ray crystallography.

Having previously established that MMTZs adopt a consistent binding mode to B1 MBLs,⁶ we next sought to determine how they interact with the active sites of both Sfh-I and L1 by X-ray crystallography. We therefore soaked pre-formed MBL crystals in MMTZs and obtained high-resolution structures of L-anti-1a bound to Sfh-I and D-syn-1b to L1 (Table S3). These represent the two compounds that can be considered the most potent over all three MBL subclasses tested (Table 1). Sfh-I crystallized in space group P2₁ with two molecules in the asymmetric unit (ASU) and L1 in space group P6₄22 with one molecule in the ASU. In both cases, bound MMTZ could be modelled into clearly defined F_o-F_c electron density (Figure S4) with real space correlation coefficients calculated by the PDB after refinement of 0.96/0.97 (Sfh-I, chains A and B) and 0.95 (L1).

MMTZ binding involves interaction of the thiol directly with the mono-zinc center of Sfh-I, or bridging the di-zinc center of L1 and displacing the catalytic water/hydroxide (Figure 3). In Sfh-I, there is a weak hydrogen bond between the ethyl ester carbonyl of L-anti-1a and the side chain nitrogen of Asn233 (3.2 A). In contrast, D-syn-1b binding to L1 is stabilized by stronger hydrogen bonds with the side chain oxygens of Ser223 (2.5 Å) and Tyr33 (2.9 Å).

Figure 3. Interactions of MMTZs in the active site of L1 and Sfh-I.

Left, thiol-Zn interactions and hydrogen bonds in MMTZ:MBL complexes (distance labelled). Right, interactions with hydrophobic residues lining the active site (blue sticks). (A) L-anti-1a (brown) binding to Sfh-I; (B) D-syn-1b (pink) binding to L1.

The binding mode of D-syn-1b to L1 has similarities with its binding to the di-zinc B1 enzymes NDM-1 and VIM-2, particularly with respect to the positioning of the thiol and carboxylate groups (Figure S5). However, the C-2 ethyl ester side chain is rotated 180°, and therefore pointing in the opposite direction in L1. Furthermore, in the B1 enzymes there is only a single (c. 2.6 – 3 Å) interaction with the backbone nitrogen of Asn233, in comparison to the hydrogen bonds D-syn-1b makes with two amino acid residues in L1.

Structure-activity relationships and conserved sulfur-π interactions.

X-ray-crystallographic data, both presented here and previously,⁶ therefore indicate that MMTZs employ a consistent binding mode to the active sites of diverse MBLs, especially when compared with the diverse modes of binding previously observed for complexes of the bicyclic BTZs. This is highlighted by the identification that, in B1 MBLs, MMTZ binding utilizes a conserved sulfur-π interaction between the thiazolidine sulfur atom and an active site aromatic residue (Trp87 in NDM-1/VIM-2 or Phe87 in IMP-1). We note that this interaction is conserved in the B2 and B3 enzymes Sfh-I and L1, with distances comparable to those observed in the B1 enzymes (Figure 4).

Figure 4. Sulfur-π interactions of MMTZs in MBL active sites.

Views from the active sites of MBLs with dashes showing the interaction (distances labelled) of the MMTZ sulfur with an active site tryptophan (blue sticks) in (A) NDM-1:L-anti-1b (PDB 6zyp⁶), (B) NDM-1:D-syn-1b (PDB 6zyq⁶), (C) Sfh-I:L-anti-1a (PDB 7bj9) and (D) L1:D-syn-1b (PDB 7bj8).

Notably, the sulfur-π interaction is not conserved for the bicyclic thiazolidine stereoisomers (bisthiazolidines, BTZs), for which diverse binding modes allowed cross-class inhibition.³⁷ Indeed, in comparison with their respective MMTZ analogues L-/D-BTZs bind very differently to both Sfh-I and L1 (Figure S6) – the BTZ carboxylate, rather than the thiol, interacts with the Zn(II) ion in Sfh-I, and in L1 the BTZ thiazolidine ring is flipped in comparison to that of the MMTZ. These differences, that are also observed for the B1 enzymes,⁶ perhaps reflect the more constrained bicyclic ring scaffold of BTZs which does not afford these compounds the flexibility to adopt consistent binding modes that exploit the sulfur-π interaction.

Furthermore, based on comparisons with X-ray crystal structures of biapenem bound to the B2 enzyme CphA (PDB 1×8i³⁸), and hydrolyzed penicillin G bound to L1 (PDB 6u0z³⁹) (Figure S7), MMTZ binding does not closely reflect that of β-lactam derived hydrolysis products. To date, there are no crystal structures available for B2 MBLs in complex with penicillin-derived products as penicillin is not hydrolyzed by these enzymes, so direct comparisons cannot be made between the orientations of the respective thiazolidine rings. However, the carboxylate of the biapenem-derived product in the CphA complex points in the opposite direction to the MMTZ carboxylate in Sfh-I, indicative of likely differing binding modes adopted by MMTZs and β-lactams (Figure S7A). The pyrroline ring of hydrolyzed biapenem lies in the same plane as the MMTZ thiazolidine, although interactions of the pyrroline N with Zn2 place the ring more deeply in the active site and closer to the Zn(II) ion than for the MMTZ. In the case of L1 (Figure S7B), both MMTZ and hydrolyzed penicillin G hydrogen bond with active site residues Tyr33 and Ser223. However, the thiazolidine rings lie at 90° to one another, resulting in a much weaker sulfur-π interaction for the hydrolyzed antibiotic, compared to that possible for the MMTZ. We previously observed the same relationship when comparing MMTZ/antibiotic binding in NDM-1.⁶

To understand the importance of these conserved interactions, and particularly the structure-activity relationships apparent for Sfh-I, we modelled in silico the MMTZs D-anti-1a, L-syn-1b and D-anti-1b that are ∼125 – 800-fold less potent against Sfh-I than L-anti-1a. We first generated structures of MMTZs docked in the active site as their anion (see Methods) and subsequently optimized these with quantum mechanical/molecular mechanics (QM/MM) simulations. This procedure was validated by closely reproducing the L-anti-1a binding mode to Sfh-I determined crystallographically, particularly maintaining sulfur-π and thiol-Zn interactions (Figure S8). These optimized, crystallographically intractable, complexes resulted in an increase of the sulfur-n interaction from 6.11 Å in the QM/MM optimized crystal structure (PDB 7bj9, L-anti-1a) to 7.50 Å (D-anti-1a) and 6.72 Å (L-syn-1b) (Figure S9A and S9B, respectively). These binding modes most likely arise due to limited space in the active site for the ligand to occupy, preventing conformations that maintain the sulfur-π interaction. This also results in a conformationally constrained L-syn-1b (625-fold less potent than L-anti-1a), in which its carbonyl oxygen is in an energetically unfavorable position, 1.7 Å from the thiol sulfur, further contributing to the poor potency of this MMTZ isomer. In contrast, D-anti-1b binds to the active site with the thiazolidine sulfur pointing away from Trp87, and in the opposite direction to the other docking modes (and the crystallographic structure of the L-anti-1a complex), further highlighting that the sulfur-π interaction cannot be maintained (Figure S9C). These constraints on the ligand geometry are consistent with the narrower substrate profile of Sfh-I compared with most B1 or B3 enzymes. These structural data rationalize the significant structure-activity relationships, particular the impact of the stereochemistry on inhibitory potency, for MMTZs against Sfh-I, where K_i values for compounds with C-2 in the (R)-configuration were significantly poorer than when C-2 is in the (S)-configuration. Furthermore, our crystallographic data support these observations by showing that the addition of a gem-dimethyl group to MMTZs can result in steric clashes with Val67 (Figure 3A, right), contributing to the ∼10-fold reduction in potency of the 1b compounds against Sfh-I (Table 1).

It has previously been shown that potency can be less for inhibitors which have an oxazolidine/triazole rather than thiazolidine/thiadiazole ring,⁴⁰ highlighting the potential importance of the sulfur atom for inhibition. Therefore, to further explore the role of the sulfur-π interaction in B2 and B3 MBLs we selectively replaced the thiazolidine sulfur with an oxygen atom in silico and optimized the complexes with resultant isosteric 2-mercaptomethyl oxazolidines (MMOZs) with QM/MM. This in silico approach was necessary as the oxazolidine compounds were synthetically intractable due to instability issues. Our previous QM/MM simulations with NDM-1⁶ using the same methodology revealed that MMOZs caused an increase in the interaction distance between the O/S of the MMOZ/MMTZ and the conserved active site aromatic residue Trp87 (Figure 5A and 5B). Here, we note that the effect in Sfh-I and L1 is less prominent than in the B1 NDM-1:L-anti-1a complex (Figure 5 and Table S4). In the case of NDM-1:L-anti-1a, the larger effect was caused by significant movement of the MMOZ in the NDM-1 active site, that is not observed in the Sfh-I and L1 simulations, nor in the NDM-1:D-syn-1b complex.

Figure 5. QM/MM optimized structures of MMTZs and their analagous oxazolidines in MBLs.

Views from the active show overlays of QM/MM optimized complex structures for MMTZs (colored as in Figure 3) and their MMOZ analogues (cyan) bound in MBLs. NDM-1 simulations were performed previously.⁶ (A) NDM-1:L-anti-1a; (B) NDM-1:D-syn-1b; (C) Sfh-I:L-anti-1a; (D) L1:D-syn-1b.

Although MMOZ movement in the NDM-1:D-syn-1b and Sfh-I:L-anti-1a complexes is limited, the distance of the heteroatom to Trp87 still increases compared to its equivalent in the MMTZ complexes. This can be partly explained by the shorter C-O, compared to C-S, bond distance, resulting in the oxygen atom being further from Trp87 than is the sulfur. In L1, however, the distance increase is minimal (0.13 Å), which can be explained by the increased number of hydrogen bonds formed by D-syn-1b in the active site potentially outweighing the effect of the O/S-aromatic interaction. Substitution of the sulfur with an oxygen therefore has variable effects in silico, but the data are suggestive that the sulfur-π interaction exerts greatest influence in the NDM-1:L-anti-1a complex. Indeed, the S to O substitution appears to only affect structures in which the hydrogen bonds in the active site are minimal (NDM-1 and Sfh-I) but has a less substantive contribution when inhibitor binding is stabilized by stronger hydrogen bonds (L1). Further, inhibitor binding in Sfh-I is more likely driven by constraints imposed by the narrow active site, with a reduced contribution from S-π interactions.

Conclusions

In summary, our analysis demonstrates, and provides the basis for, cross-class inhibition of MBLs by MMTZs. Although S-π interactions contribute to MMTZ:MBL interactions in all cases, our data suggest that different factors are responsible for establishing the broadly similar binding modes in the three structurally different active sites. In B1 MBLs such as NDM-1, S-π interactions most likely dominate; B2 enzymes (e.g. Sfh-I) have a constrained active site that can sterically clash with MMTZs, particularly those with a gem-dimethyl group or C-2 in the (S)-configuration; whereas in B3 enzymes, the contribution of hydrogen bonds to binding appears to override the S-π interaction. Differences in potency, and the potential for steric clashes with the active site, highlight the need to consider alternative inhibitor stereoisomers to achieve cross-class MBL inhibition, and more generally in future inhibitor development. MMTZs therefore represent a promising MBL inhibitor scaffold that is synthetically accessible, allowing for the design of future iterations that improve potency in bacterial cells

Material and Methods

Synthesis of 2-mercaptomethyl thiazolidines.

MMTZs were synthesized as previously described.⁶ Purity was determined by HPLC to be >95% as described in Rossi et al.⁶

Inhibition constants.

Inhibition constants were determined as previously described.⁶ In detail, K_i values were determined under steady-state conditions by following imipenem hydrolysis by Sfh-I and L1 at 30 °C. Imipenem breakdown was measured as a decrease in absorbance at 300 nm (Δε₃₀₀ = −9000 M⁻¹cm⁻¹) in a Jasco V-670 spectrophotometer. Reactions were setup in a 0.1 cm path length quartz cuvette with a final volume of 300 μL. Final enzyme concentration was 2 nM. Reaction buffer contained 10 mM HEPES pH 7.5, 200 mM NaCl and 50 mg/L BSA. Reaction buffer for L1 was supplemented with 20 μM ZnSO₄. MMTZs were dissolved in DMSO at a final concentration of 100 mM and diluted in the reaction buffer to the desired concentration. The DMSO concentration (0.07 %) does not affect the enzyme activity (data not shown). The reactions were started and monitored with the addition of the MBLs to a substrate and inhibitor mixture. Linear time courses were observed for all the conditions and less than 5 % of the substrate was consumed after 300 s (Figure S3). Reaction rates were obtained from the slope of these time courses and fitted with the Competitive Inhibition model in GraphPad Prism 5.0, from which the inhibition constant (K_i) could be calculated (Figure S4).

The same protocol was followed for the previously published K_i determination of the MMTZ compounds with B1 enzymes⁶ and the BTZ compounds with B1³⁶, B2 and B3 enzymes.³⁷

Microbiology.

Minimal inhibitory concentrations (MICs) for imipenem were determined by the broth microdilution method in cation-adjusted Mueller Hinton broth according to Clinical and Laboratory Standards Institute (CLSI) guidelines.⁴¹ Imipenem was chosen as a representative carbapenem as Sfh-I is predominantly a carbapenemase. E. coli DH10B cells were transformed with pMBLe (containing bla_NDM-1 or bla_Sfh-I) or pBC-SK (containing bla_L1). NDM-1 and Sfh-I expression was induced with 0.1 mM IPTG. MMTZs were dissolved to 250 mM in DMSO before dilution to 100 μg/mL in serial doubling dilutions of imipenem (from 64 mg/L to 0.025 mg/L). Results presented are the mode of three biological replicates.

X-ray crystallography.

L1 and Sfh-1 were produced, purified and crystallized as previously described.^{37, 42} L1 crystallized by mixing 1 μL protein (23 mg/mL in 10 mM Tris pH 7, 5 mM ZnSO₄, 100 mM NaCl) with 1 μL crystallization reagent [0.1 M HEPES pH 7.75, 1.5% PEG400, 2 M (NH₄)₂SO₄].³⁷ Sfh-I was crystallized by mixing, in a 3:2 ratio, protein (15 mg/mL, in 50 mM HEPES pH 7.5, 10% glycerol) with 0.2 M sodium acetate and 27% wt/vol PEG 3350.⁴² MMTZs were dissolved in crystallization buffer at 5 mM and soaked into pre-formed crystals for 1 h 20 m (L1) or 1 h (Sfh-I). Diffraction data were collected at Diamond I03 (Sfh-I) or I04-1 (L1). Reflections were indexed and integrated in Mosflm (L1) or Dials⁴³ (Sfh-I) and scaled in Aimless.⁴⁴ Phases were calculated with molecular replacement in Phaser⁴⁵ using 5evd³⁷ or 5ew0³⁷ as search models for L1 and Sfh-I, respectively. Structural models were completed with iterative rounds of refinement in Coot⁴⁶ and Phenix⁴⁷ Ligand restraints were calculated in eLBOW^{47, 48} and modelled into clearly defined F_o-F_c electron density. Figures were created in PyMol.⁴⁹

Computational calculations.

Starting structure preparation

All ligands were prepared as thiolate anions, with the carboxylate also deprotonated. Oxazolidine complexes were built in silico by replacing the sulfur atom with an oxygen atom in the corresponding L1 and Sfh-I crystal structures, PDBs 7bj8 and 7bj9, respectively.

Sfh-I complexes with crystallographically intractable MMTZs were initially docked in Autodock 4.2.6⁵⁰ using the solvent sites biased Autodock4 docking method (SSBMD).⁵¹ Using information from the available crystal structures, we modified the Autodock4 scoring function by adding an energy term for the thiol sulfur atom to the original function. The Sfh-I crystal structure (PDB 7BJ9) was used as the receptor protein with hydrogen atoms added by AutoDockTools4.⁵⁰ Structures of compounds D-anti-1b, L-syn-1b, D-anti-1a were optimized in vacuum at Hartree–Fock/6-31G* level using Gaussian 09⁵² and subsequently prepared in pdbqt format using AutoDockTools4.⁵⁰ The grid map was set to 40 × 40 × 40 points with a grid spacing of 0.375 Å centered on the catalytic site. For each calculation, 100 different docking runs were performed, and the resulting 100 poses were clustered according to a ligand RMSD cut-off of 2 Å. The predicted binding free energy score (ΔG) and population were the criteria used to choose the correct ligand pose, further considering the crystallographic information of similar compounds.

Quantum Mechanics/Molecular Mechanics (QM/MM) optimizations

All starting structures were geometry optimized at the Quantum Mechanics/Molecular Mechanics (QM/MM) level. For hybrid QM/MM calculations, we used a semiempirical method PM6⁵³ to describe the QM region and the ff14SB force field⁵⁴ to describe the MM region as implemented in Amber16.^{55, 56} Hydrogen atoms were added, and each protein was immersed in a truncated octahedral periodic box with a minimum solute-wall distance of 8 Å, filled with explicit TIP3P water molecules⁵⁷, using the AMBER16 leap module⁵⁵. The van der Waals radius, force constants and equilibrium distances, angles and dihedral of the studied inhibitors were taken from the gaff database and partial charges were RESP charges computed using the Hartree–Fock method and 6-31G* basis set.^{55, 58} To accommodate solvent molecules and possible clashes, an initial minimization at the molecular mechanic level of each complex structure was performed, followed by QM/MM geometry optimization. The QM region consisted of both Zn(II) ions plus the coordinated side chains of residues of the first coordination sphere and the inhibitor compound.

Abbreviations

ASU

asymmetric unit

BLI

β-lactamase inhibitor

BTZ

bisthiazolidine

DBO

diazabicyclooctane

MBL

metallo-β-lactamase

MIC

minimum inhibitory concentration

MMOZ

2-mercaptomethyl oxazolidine

MMTZ

2-mercaptomethyl thiazolidine

QM/MM

quantum mechanics/molecular mechanics

SBL

serine-β-lactamase