Broad spectrum antibiotic-degrading metallo-β-lactamases are phylogenetically diverse

Antibiotic resistance has emerged as a major threat to global health; multi-drug resistant bacteria already kill more patients in the United States each year than HIV/AIDS, Parkinson’s disease, emphysema and homicide combined (Laxminarayan et al., 2013). Among the most effective bacterial resistance mechanisms are β-lactamases, a family of enzymes that are divided into four distinct classes. Classes A, C and D (serine-β-lactamases, SBLs) use a catalytic site serine residue to initiate inactivation of the antibiotic, while Class B (metallo-β-lactamases, MBLs) relies on a Zn-activated hydroxide (Walsh et al., 2005; Bush and Jacoby, 2010; Mitic et al., 2014; Lisa et al., 2017). Clinically relevant inhibitors of Class C and D SBLs are available and in use (e.g., clavulanic acid (CA), Drawz et al., 2010), but for MBLs the search for such inhibitors has remained challenging (McGeary et al., 2017). MBLs are divided into three subgroups, i.e. B1, B2 and B3 (Bush and Jacoby, 2010). Enzymes of the B1 subgroup constitute the majority of MBLs associated with antibiotic resistance (Khan et al., 2017). Fewer B2-type MBLs are currently known; they are phylogenetically related to B1 MBLs but are characterized by a preference for “last line” carbapenem substrates (Sun et al., 2016). While B3-type MBLs share low sequence similarity to B1 and B2 enzymes (<20% amino acid (aa) identity), they have a substrate range similar to that of B1 MBLs (Selleck et al., 2016; Lee et al., 2019). MBLs contain catalytic centres that can accommodate two closely spaced Zn ions bound in the α and β sites with similar yet distinct sequence motifs (B1: His116, His118, His196 and Asp120, Cys221, His263 (i.e., HHH/DCH) for the α and β sites, respectively; B2: NHH/DCH; B3: HHH/DHH). For B3-type MBLs two variations of the canonical active site motif have been observed, QHH/DHH in GOB-1/18 from the opportunistic pathogen Elizabethkingia meningoseptica and HRH/DQK in SPR-1 from Serratia proteamaculans (variations shown in bold) (Vella et al., 2013; Moran-Barrio et al., 2016). The discovery of atypical active sites in B3-type MBLs may have important implications for the design of clinically useful MBL inhibitors. We thus probed the evolutionary history and diversity of B3-type MBLs by searching for homologs in the release 02-RS83 of the Genome Taxonomy Database (Parks et al., 2018) comprising 111,330 quality-filtered bacterial and archaeal genomes. A total of 1,449 B3 MBL proteins were identified in 1,383 genomes (representing 1.2% of all analyzed genomes), of which 1,150 have the characteristic B3 active site residues (HHH/DHH), 162 the QHH/DHH and 47 the HRH/DQK motifs. In addition, we also discovered 90 proteins with another single aa variation in the α-site (EHH/DHH). Phylogenetic inference of a representative subset of 761 of these proteins indicates that each of the three motif variants originate from within the B3 radiation when using Class D SBLs as the outgroup (Fig. 1). We therefore propose to use the active site aa changes as a means of distinguishing the variants (i.e., B3-RQK, B3-Q, B3-E). B3-RQK appears to have only arisen once, likely because the ancestral change required at least four nucleotide (nt) substitutions to produce the three aa changes. By contrast, the B3-Q and B3-E variants have a single aa difference in position 116 requiring only one and two nt changes, respectively. The B3-Q variant appears to have arisen on at least six independent occasions and reverted back to the B3 motif on at least three occasions as a result of the need for only one nt change. No archaeal genomes harbored B3-type MBLs, and the majority were found in just four bacterial phyla; the Proteobacteria, Actinobacteria, Bacteroidetes and Firmicutes (Figs. 1 and S1). While this reflects to some extent the current over-representation of these phyla in the genome database (Fig. S2), it also suggests that the host range of B3 MBLs is relatively restricted. Between two and five B3 genes were found in 57 genomes, with the most copies being present in an as-yet-uncultured member of the Acidobacteria (Table S1). Numerous instances of native B3 enzymes cooccurring with B3-E and B3-Q were identified, however, only one instance of a B3 and B3-RQK was found (in a member of the Enterobacteriaceae) possibly indicating functional incompatibility of these enzymes. The phylogenetic analysis of the B3 MBL family indicates a large and diverse reservoir of bacterial species potentially able to degrade β-lactam antibiotics. Standard B3 MBLs are potent β-lactamases as shown by in vitro β-lactam antibiotic degradation assays and their ability to confer ex vivo resistance to Escherichia coli (Yong et al., 2012). The only characterized representatives of B3-RQK (SPR-1) (Vella

In vitro kinetic data and ex vivo plate tests of selected B3 metallo-b-lactamases. As a positive control, genes of known B3 MBLs (L1, FEZ-1, AIM-1, MIM-1 and MIM-2) were included in the analysis. a. All plate tests were performed in the present study and some in vitro kinetic data were obtained from published studies. b. The signal peptide was removed from all proteins (~25 aa). An additional 30 to 50 aa were removed from the N-terminus of the B3-RQK variants (see text); SPR-1 (49 aa), CSR-1 (32 aa), SER-1 (35 aa). c. Aggregate resistance scores of the MBLs were calculated as the sum of individual resistance scores for each of nine substrates. The scoring system was 1 for resistance (R; zone of inhibition <18 mm), 0.5 for marginal resistance (r; inhibition zone 18-26 mm) and 0 for sensitivity (S; inhibition zone ³27 mm) based on zones observed around the negative control. d Previously called B4 (Vella et al., 2013).   (Drawz et al., 2010), and in the only MBL for which a Ki for clavulanic acid (67 μM) has been reported (the B1type MBL SPM-1; Murphy et al., 2003).  1), and the type of active site is noted at the base of the figure. The top panel shows cartoon representations of the overall structure of the selected enzymes, with active site ligands in orange, Zn 2+ ions in grey, and the N-terminal loop in red. The bottom panel shows surface representations of the five structures highlighting occlusion of the active site by the N-terminus in CSR-1. A conspicuous difference between the four enzymes is the relative position of their N-termini. In L1, the N-terminus protrudes from the structure thereby making the active site accessible, whereas in AIM-1, a disulphide bridge between Cys32 and Cys66 locks its N-terminus in a position away from the catalytic centre (Leiros et al., 2012;Hou et al., 2017). Similar to L1, GOB-1/18 has a protruding N-terminus exposing the active site (Horsfall et al., 2011;Moran-Barrio et al., 2016).

Fig. S5|
Experimental ITC data collected for the variants of CSR-1 from the subgroup B3-RQK. Data for CSR-1trunc (a), CSR-1trunc, sm (b), CSR-1trunc,dm (c) and CSR-1trunc,tm (d) were fitted using a nonlinear algorithm to minimise c 2 values using equations derived from equilibrium-binding models simulating either one or two independent binding sites.  Table S4| ITC parameters collected for CSR-1trunc and its mutants. CSR-1trunc binds two Zn 2+ ions with moderate to weak affinities (αKd ~2 µM and βKd ~150 µM). Two Zn 2+ ions also bind to CSR-1trunc,sm, but the affinity of the α-site (αKd ~0.1 µM) is significantly enhanced, while the β-site (βKd ~180 µM) is unaffected by the mutation. In the double and triple mutants both the α and β metal binding sites have high affinity (αKd = βKd ~0.14 µM and αKd ~0.06 µM and βKd ~0.4 µM, respectively). For comparison, similar ITC measurements with AIM-1 result in an αKd and βKd of ~0.17 µM (Selleck et al., 2016) indicating that the double and triple mutants of CSR-1 fully restore the characteristic ability of native B3 enzymes to bind two Zn 2+ ions tightly. (a) Mathematical model applied to fit the experimental data. N is the stoichiometry of an interaction measured by ITC, and Kd is the dissociation constant.  Table S5. In vitro kinetic data and ex vivo plate test of CSR-1 and its mutants. The effect of the single mutation in the α-site (i.e. R118H) on the catalytic properties is modest; CSR-1trunc,sm displays a slightly enhanced catalytic activity towards most of the substrates tested (Table S6), but this positive effect is largely offset by reduced substrate binding. The effect of the double and triple mutations in CSR-1trunc, however, is striking. A marked increase in catalytic efficiency was observed against most antibiotics accompanied by an increase in ex vivo resistance to both antibiotics (resistance score 8.5 vs 5 for CSR-1trunc) and clavulanic acid.

Phylogenetic Analysis
Putative protein orthologues of the B3 family of MBLs were identified from the Genome  (Katoh and Standley, 2013). Model parameters for phylogenetic inference were evaluated using ModelFinder (Kalyaanamoorthy et al., 2017) in IQ-Tree (Nguyen et al., 2015), selecting the optimal model according to the Bayesian Information Criterion.
Inference was then performed in IQ-Tree using the WAG+F+R10 model of amino acid substitution and 100 bootstrap re-samplings to assess node support. The resulting tree was visualised and annotated with genome metadata using the Interactive Tree of Life resource (Letunic and Bork, 2016).

Ex vivo whole cell plate assays
All genes listed in Table S2, with the exception of FEZ-1, were synthesised by General Biosystems. Genes were cloned into pET22b(+) and transformed into E. coli BL21(DE3) cells.  et al., 2002). Diffraction data were integrated, scaled and merged using the software HKL-2000 (Otwinowski and Minor, 1997). Refinement and model building were carried out using PHENIX 1.8.4 (Adams et al., 2010) and COOT 0.7 (Emsley et al., 2010), respectively.
The CSR-1 and CSR-1trunc structures were determined using the previously published coordinates for L1 (PDB 1SML) (Leiros et al., 2012). The mutants CSR-1trunc,sm, CSR-1trunc,dm and CSR-1trunc,tm were then solved by molecular replacement using the CSR-1 structure as a search model. All atoms were subsequently refined with anisotropic B-factors; most hydrogen atoms were fitted as riding models. Relevant crystallographic data and refinement statistics are summarised in Table S7.

Enzyme Kinetics
All kinetic measurements were performed on a Cary 60 Bio Varian UV-Vis spectrophotometer. In a standard assay, reactions were initiated by the addition of the enzyme and monitored for 60 s at 25°C. The Michaelis-Menten (Segel, 1993) parameters were determined by nonlinear fitting using Graphpad Prism 7. The hydrolysis of ampicillin, penicillin G, carbenicillin, nitrocefin, cefoxitin, cefuroxime, meropenem, biapenem, imipenem and nitrocefin were monitored in 50 mM TrisHCl (pH 8.5) by detecting the hydrolysis of the substrate. Inhibition measurements were obtained by measuring the activity of the enzyme with ampicillin and analysing the effect of the increase in clavulanic acid concentration on the enzyme activity. The data were calculated using the mixedinhibition equation (Segel, 1993).

Metal binding studies
The metal-free apoforms of the CSR-1 enzyme and its mutants were obtained by incubating approximately 3 mg of protein in a 3 mL solution containing 10 mM of EDTA in 20 mM HEPES buffer (pH 7.0 at 4°C). After 24 hours, protein samples were separated from the chelating solution using an Econo-Pac 10DG gel filtration column equilibrated with fresh HEPES buffer treated with chelex resin to remove any residual metal ions. Atomic absorption spectroscopy (AAS) was used to confirm the absence of metal ions in the final protein solutions. Zn 2+ ions were then titrated back into the protein solutions and the heat release measured by isothermal titration calorimetry (ITC) at 25°C using a Nano ITC from TA instruments. The concentration of the metal ion solution was verified against standardised EDTA solutions and the results were compared against AAS data. At least four sets of data were collected per protein and the average calculated. The ITC data were fitted using TRIOS v4.4.0 software from TA instruments. The software uses a nonlinear algorithm to minimise c 2 values to fit the experimental data to equations derived from equilibrium-binding models simulating either one or two independent binding sites ( Table S4) (Pedroso et al., 2014;Pedroso et al., 2017).

Molecular docking and QM/MM calculations
For the theoretical studies, the missing Zn 2+ ions in CSR-1trunc and CSR-1trunc,sm were manually placed in the binding site using CSR-1trunc,tm as a template. All water molecules were removed, as well as the artefactual third Zn 2+ in CSR-1trunc,dm (Fig. 2a). Proteins were protonated using the program tleap in the AmberTools16 software package (Case et al., 2016). Metal-ligating histidine residues were protonated at the nitrogen not involved in metal coordination. Lys263 in the β-site was considered in its neutral form such that it can ligate a Zn 2+ ion. Molecular docking of clavulanic acid (both in intact and hydrolysed form) was performed with FlexX (Rarey et al., 1996) within the LeadIT platform version 2.3.2 (https://www.biosolveit.de/LeadIT/). All residues within 10 Å of the geometric centre of the two metal ions were considered as binding site residues. The 'Enthalpy and Entropy' (i.e. Hybrid) approach was used for initial base placement and the clash factor and maximum allowed overlap volume were set to 0.6 and 3.5 Å 3 , respectively.
Selected poses from the initial molecular docking calculations were further optimized using a QM/MM-based potential according to Marion et al. (Marion et al., 2017) The QM/MM calculations were performed using the ChemShell suite (Sherwood et al., 2003;Spencer and Walsh, 2006) with the DL_POLY module for the MM-calculations interfacing with Turbomole version 7.1 for the QM-calculations. The MM-region was described applying the parameters from the Amber ff14SB (Maier et al., 2015) force field. The QM-region comprised the side chains of Asn142, His196, Leu197, Asp208, Ser214, Tyr229, Asn262, Lys263, Glu265, and Arg266, and the loop from Asn114 to Gln121, as shown in Fig. S6. The QM/MM boundary was located at the nonpolar Cα-Cβ bond of the single side chains and at the N-Cα and Cα-C' bond of the backbone, respectively. The QM-calculations were performed at the DFT level together with the D3 dispersion correction (Grimme et al., 2010). The TPSS meta-GGA (Tao et al., 2003) functional was applied using RI-J approximation (Eichkorn et al., 1995) on a multigrid m4 (Eichkorn et al., 1997). The SCF convergence criterion was set to 10 -7 au. The def2-TZVP (Eichkorn et al., 1997;Weigend and Ahlrichs, 2005) basis set was used for the Zn 2+ ions while the remainder of the QM-region was described with the def2-SVP (Schäfer et al., 1992;Eichkorn et al., 1997) basis set. An electrostatic embedding scheme was applied to describe the QM/MM boundary, in which link atoms are placed at the boundary and the charges are shifted away and replaced by a dipole on the recipient atom, as implemented in the polarised coupling scheme (shift) in ChemShell. The geometry optimisation of all atoms in the QM-region and all direct neighbouring MM-residues was performed using the DL-FIND (Kästner et al., 2009) optimiser.
To summarise, calculations based on the bimetallic enzyme predicted that the carboxylate group of clavulanic acid binds to both metal ions and forms hydrogen bonds with Ser214, Asn254 and Arg257 in the substrate binding pocket of CSR-1Trunc (Figs 2b). Once the calculation was performed with single metal occupancy in the α-site (i.e. the higher metal affinity site), we observed a stable enzyme-inhibitor complex only in CSR-1trunc and CSR-1trunc,sm, indicating the ability of clavulanic acid to displace the β-Zn 2+ from its low affinity binding site in these two variants of CSR-1. We therefore infer that low metal binding affinity in the β-site is necessary for the observed inhibition by clavulanic acid. Hydrogen bonding to Lys263 may provide additional stabilization of the enzyme-clavulanic acid complex in monometallic CSR-1trunc and CSR-1trunc,sm, as this interaction cannot be formed in the double and triple mutants, in which His263 strongly interacts with Zn 2+ bound in the β site (Figs 2b). Since the β-lactam bond in clavulanic acid is prone to hydrolysis (for instance by an attack from an efficient nucleophile such as Lys263) we also investigated interactions between the ring-opened form of the inhibitor with CSR-1 and its variants by docking, using the same approach as employed for the computations with the intact form of clavulanic acid (Fig. S6). No significant differences were observed, indicating that clavulanic acid, independent of its state (intact or ring-opened), inhibits the wild-type of CSR-1 and its single mutant.