NMR-filtered virtual screening leads to non-metal chelating metallo-β-lactamase inhibitors† †Electronic supplementary information (ESI) available: All experimental details, crystallographic data collection and refinement statistics, details of chemical synthesis, additional figures and tables. See DOI: 10.1039/c6sc04524c Click here for additional data file.

NMR-filtered virtual screening led to the identification of non-Zn(ii)-chelating metallo-β-lactamase inhibitors, which mimic interactions made by the bicyclic β-lactam antibiotic substrates as they initially bind to the enzymes.


Introduction
One of the most important mechanisms of resistance to b-lactam antibacterials involves their hydrolysis as catalysed by serine-b-lactamases (SBLs) and metallo-b-lactamases (MBLs) (Fig. 1). 1,2 Although clinically useful inhibitors of the SBLs are established, there are no such MBL inhibitors. [3][4][5][6][7] Inhibitors of human MBL-fold proteins are also of medicinal interest, including for the human DNA cross-link repair enzymes SNM1-A/B, in order to combat resistance to major anti-cancer drugs such as cisplatin. 8,9 To date, almost all reported MBL inhibitors work via zinc ion chelation ( Fig. S1 and S2 †). [10][11][12][13] Development of new types of MBL-fold enzyme inhibitors that do not work via metal chelation is presently desirable, in part because this may enable improved selectivity than (readily) achievable with zinc ion chelation. Here we report how a virtual screening approach combined with NMR ltering, led to the identication of nonmetal chelating inhibitors of the clinically relevant Verona Integron-encoded MBL (VIM)-2. As revealed by crystallographic, NMR, and biochemical studies, the new inhibitors bind via a mode that does not involve direct zinc chelation, but which may mimic interactions made by intact b-lactam substrates as they initially bind to VIM-2. VIM-2 is a clinically important representative of the class B1 MBLs (which also includes the imipenemase (IMP)-1, and New Delhi MBL (NDM)-1), that have a broad-spectrum substrate prole that includes penicillins, cephamycins, cephalosporins, oxacephamycins, and carbapenems. 14 The B1 subfamily MBLs are di-Zn(II) utilizing enzymes, with both the Zn1 and Zn2 ions having crucial roles in catalysis, with respect to b-lactam substrate binding, and hydrolytic water activation. [15][16][17]  We began by carrying out a virtual screen with VIM-2 for which several high-resolution (<1.5Å) crystal structures are available. 12,[18][19][20] A customized virtual screening method, which combines molecular docking simulations with a molecular interaction ngerprints (IFPs)-based ltering approach, was used to identify compounds that are likely to interact with catalytically important active site residues including Arg228, Asn233, Phe61, Tyr67, and Asp120 (using the standard BBL numbering scheme for class B b-lactamases 21 ) as well as zinc ions. Eight types of protein-ligand interactions (hydrogenbonding donor, hydrogen-bonding acceptor, positively charged, negatively charged, face-to-face p-p stacking, edge-to-face p-p stacking, hydrophobic, and metal-ligand interactions) as dened in our previous work 22,23 were used to generate the IFPs (for details of the virtual screening methods see ESI Experimental section SE. 1 and Fig. S3 †). Although our strategy included the identication of potential zinc ion binding inhibitors, since we have found experimentally that metal ion chelation can serve to 'template' ligands to the active sites of metallo-enzymes, [24][25][26] we were particularly interested in the identication of non-Zn chelating inhibitors. We subsequently screened selected compounds identied in the virtual screen for activity against VIM-2 using a uorescence-based assay. 27 We then used ligand-observe 1 H Carr-Purcell-Meiboom-Gill (CPMG) NMR spectroscopy 28 to test for binding to both the apo-VIM-2 and catalytically active di-Zn(II) VIM-2, with the aim of establishing whether the zinc ions are required for inhibitor binding.

Results
Application of the virtual screening method led to the identication of a number of fragment-sized compounds, mostly containing acidic groups, that are likely to interact with catalytically important active site features, replicating interactions involved in binding the carboxylate present in b-lactam antibacterials (i.e. with Arg228, Fig. S4-S6 †). Of the 20 experimentally tested compounds, 15 exhibited inhibitory activity against VIM-2, and 8 compounds manifested IC 50 values <400 mM, including compounds 6,7,12,13,16,17,18, and 20 (Table 1 and Fig. S7 †). Using the same assay conditions, we observed that L-captopril inhibits VIM-2 with an IC 50 value of 1.6 mM. As negative controls, we tested four compounds which did not pass the IFP score cut-off in the virtual screen at 2 mM against VIM-2 (44-47, Fig. S6, Table S4 †); these compounds displayed substantially weaker inhibition than those with IFP scores above the cut-off.
Since some of the compounds, e.g. 6,7,12,13,16,17,18, and 20, possess potential metal-chelating motifs, further VIM-2 inhibition assays at three different concentrations of Zn(II) (0 mM, 1 mM, and 100 mM) were employed to investigate the potential for metal chelation in solution. With the exception of compound 18, no obvious differences between the inhibitory activity with or without excess Zn(II) were observed, suggesting that most of these compounds were not strong Zn(II) chelators in solution (Fig. S8 †). L-Captopril was used as a positive control for active site/metal binding 18 and showed IC 50 values of 1.9 mM, 1.6 mM, and 1.4 mM at concentrations of 0 mM, 1 mM, and 100 mM Zn(II), respectively. Compound 18, however, likely causes inhibition, at least in part, by chelating Zn(II) in solution, so sequestering it from the enzyme.
With the aim of identifying compounds that bind (at least, substantially) without active site Zn(II) chelation, we used 1 H CPMG NMR (ESI Experimental section SE. 4 †) to investigate binding to catalytically active di-Zn(II) VIM-2 as well as to the apo-VIM-2 form for selected compounds from the virtual screen (including 6, 7, 12, 13, 16, 17, and 18). Notably, for 16 and 17, which displayed the most potent VIM-2 inhibition (IC 50 values of 10.6 mM and 32.8 mM, respectively, Table 1), we observed binding to both the di-Zn(II) VIM-2 and apo-VIM-2 protein by 1 H CPMG NMR analyses ( Fig. 2a and b). For comparison, we selected a known inhibitor of the VIM-2 MBL, L-captopril, which works via an established binding mode that involves direct zinc chelation as revealed by crystallographic analyses; 18,29 consistent with its reported mode of binding, we observed that L-captopril displays strong binding to di-Zn(II)-VIM-2, but only very weak binding to apo-VIM-2 ( Fig. S9 †). The reduction in signal intensity observations with 16 and 17 suggest that they bind more tightly to di-Zn(II) than apo-VIM-2 ( Fig. 2), possibly reecting the more ordered structure of the di-Zn(II) enzyme.
Overall, these results show that 16 or 17 can bind to VIM-2 in the absence of zinc ions. However, the presence of zinc indirectly participates in the binding of these inhibitors to VIM-2, possibly through protein stabilisation, bridging interactions or a combination thereof. We also observed that 6 (IC 50 ¼ 111 mM) displays strong binding to both di-Zn(II) VIM-2 and apo-VIM-2 ( Fig. S10 †), indicating that, like 16 and 17, 6 may bind without metal chelation. In contrast, 12 (IC 50 ¼ 87 mM) and 13 (IC 50 ¼ 135 mM) displayed strong binding to di-Zn(II) VIM-2, but weak binding to the apo-VIM-2 protein ( Fig. S11 and S12 †), suggesting that these compounds inhibit VIM-2 by a mode directly involving zinc chelation. For 7 and 18, for which assays showed VIM-2 inhibition (IC 50 values of 199 mM and 117 mM, respectively, Table 1), we observed binding to both di-Zn(II)-and apo-VIM-2 (Fig. S13 and S14 †) by NMR. In contrast to 16 and 17 (see below), 12 and 13, which may be metal-chelating VIM-2 inhibitors as observed in the inhibition assays and NMR studies ( Table 1, Fig. S11 and S12 †), displayed broad-spectrum inhibition against almost all the tested class B MBLs (Fig. S21 †).
In order to investigate the inhibitory mechanism of 16 and 17, we then sought to obtain crystal structures for complexes of VIM-2 with them (ESI Experimental section SE. 5 and Table S1 † for details of crystallization and structure determinations). Cocrystallisation experiments yielded structures of VIM-2 in complex with 16 (1.40Å) and 17 (1.93Å) (Table S2 †). VIM-2:16 and VIM-2:17 crystallized in the P2 1 22 1 and P2 1 2 1 2 1 space groups (Table S2 †), respectively, with one molecule in the asymmetric unit (ASU); these space groups differ from those previously reported for VIM-2. 12,18,19 In both structures, there was clear F o À F c density in the VIM-2 active site, into which 16 and 17 could be condently modelled (Fig. S17 †). The crystal structures reveal that 16 and 17 both bind adjacent to the zinc ions of VIM-2, but in a mode which does not involve direct zinc chelation (Fig. 3a-c), consistent with the NMR results ( Fig. 2a  and b). All of the atoms of the inhibitor are >4Å from the zinc ions (Fig. 3c). The predicted binding modes of 16 and 17 are similar to those observed in the crystal structures (Fig. S18 †), indicating that the virtual screening method, which searched for compounds likely to interact with catalytically important active site features, is a useful strategy for identication of new MBL-fold enzyme inhibitors.
The crystal structures reveal that 16 and 17 have very similar binding modes, in which the 3-oxoisoindoline-4-carboxylate heterocyclic core is positioned to make hydrophobic and electrostatic interactions with the VIM-2 active site features,   including via p-p stacking interactions with Phe61 on the L3 loop, hydrogen-bond interactions with Arg228 and Asn233 on the L10 loop and water molecules (such as W3) ( Fig. 3a and b and S19 †). 16 (IC 50 ¼ 10.6 mM) appears positioned to make stronger hydrogen-bonding interactions with Arg228 (2.9Å), Asn233 (2.8Å), and a water molecule W2 (2.7Å) ( Fig. 3a and b and S19 †) than 17 (IC 50 ¼ 32.8 mM), which may, in part, explain why it inhibits more potently. As revealed by analysis of the Cambridge Structural Database, zinc-ligand coordination bond lengths are generally less than 2.5Å, 30  A) and appears to be important to the binding of the carboxylate groups of 16 and 17 (Fig. 3d).
We were attracted by the resemblance of the 3-oxoisoindoline-4-carboxylate scaffold to that of the bicyclic b-lactam MBL substrates, e.g. cephalosporins ( Fig. 1 and 4c). Using molecular docking analyses, we observed that the bicyclic core of a cephalosporin (a representative bicyclic b-lactam substrate) is likely to form hydrogen-bonding interactions with Asn233 and the water molecule W3 (Fig. S22a †), in a similar manner to the 3oxoisoindoline-4-carboxylate scaffold of 16 and 17, as observed by our crystallographic analyses (Fig. S22b †). As shown in Fig. 4c, 16 and 17 also have a similar binding mode to that observed in a complex of NDM-1 with a ring-opened cephalosporin intermediate; 33 the carboxylate groups of 16 and 17 are positioned to make electrostatic interactions with Arg228 in an analogous manner to that observed for the binding of the   (Fig. 4c). On superimposing the crystal structures of VIM-2:16, VIM-2:17, and di-Zn(II) VIM-2, we observe that the L3 loop of VIM-2 appears closer to the zinc ions when 16 and 17 are bound to VIM-2 (3.50Å and 3.81Å movements of the C-a atoms of Asp62, respectively, Fig. S23a †), whilst the L10 loop is positioned further away from the zinc ions (1.39Å and 1.53Å difference in the C-a atom of Asn233, respectively, Fig. S23 †). These results support the proposal that, at least in some cases, these loops contribute to the highly efficient nature of MBL catalysis by capturing potential substrates and delivering them to the zinc ion containing active site for hydrolysis.
From the results above, we propose that the 3-oxoisoindoline-4-carboxylate scaffold of 16 and 17 is an important factor contributing to the inhibitory potency and the selectivity for VIM-2. We carried out structure-activity relationship (SAR) studies using commercially available and synthesised 3-oxoisoindoline-4-carboxylate derivatives with different indole-Nsubstituents (R 1 ) and/or C-1 substituents (R 2 ) (compounds 21-43 in Table 2). We synthesized 34 and 35 via the routes in ESI methods SE. 2 and Scheme S1. † Compared to the hit compound 16 identied in initial screening (IC 50 ¼ 10.6 mM), which has a 2-chloro-6-uorobenzyl moiety at the R 1 position, compounds 21-29 with different substituted-benzyl moieties at the R 1 position exhibited lower inhibitory activities against VIM-2. The results indicated that chloro-and uoro-substituents at the ortho position of the N-benzyl group are important in binding. The VIM-2:16 complex structure implies that the uorine and chlorine atoms of 16 are likely positioned to form halogen-bonding interactions with Asn233 (2.88Å), and Trp87 (3.41Å), and Asp120 (3.34Å), respectively ( Fig. 3a and S19a †). Compounds bearing substituted phenyl moieties at the R 1 position (30-35) displayed better VIM-2 inhibitory activities than 21-29, some having comparable activity to 16 and 17. Compound 30 is more potent than 17 (IC 50 ¼ 32.8 mM) with an IC 50 value of 17.8 mM, although its N-phenyl group is unsubstituted. As revealed by subsequent crystallographic analyses of VIM-2 complexes, 30 and 17 have similar binding modes ( Fig. 5a and b and 3b). However, compared with 17, 30 is apparently better positioned to interact with the Asn233 side chain (the triuoromethyl group of 17 hinders hydrogen bonding interactions between the Asn233 side chain and the 3oxoisoindoline-4-carboxylate, Fig. 5b and S24 †). Compound 35 (IC 50 ¼ 7.7 mM), which contains a 3-uoro-4-hydroxyphenyl motif at its R 1 position, was more potent against VIM-2 than compounds 16 and 17 ( Table 2). As observed by crystallography, 35 has the same overall binding mode as 30 ( Fig. 5c and  d); compared with 30, 35 is positioned to make additional hydrogen-bonding interactions with Asp119 (3.03Å), as well as  Table 2). As observed in a VIM-2:42 complex crystal structure, the allyl group of 42 is positioned to make hydrophobic interactions with Phe61 and Tyr67 ( Fig. 5e and S26 †).
Notably, comparison of structures of VIM-2:30 and VIM-2:42 reveals evidence for exibility in the conformation of the L3 loop, suggesting that Tyr67 and Phe61 may be important in substrate/inhibitor capture (Fig. 5f). 43 showed much lower inhibitory potency, supporting the proposed role of the 3-oxoisoindoline-4-carboxylate motif in binding/inhibition. For VIM-5 and VIM-1, almost all the tested 3-oxoisoindoline-4carboxylate derivatives (except 16, 24, and 36) displayed low inhibitory activities, indicating that the 3-oxoisoindoline-4carboxylate scaffold is an important factor (along with its precise substitution pattern) in determining selectivity for VIM-2. Together, these results led us to conclude that the 3-oxoisoindoline-4-carboxylate derivatives selectively inhibit the VIM-2 MBL via a mode which does not involve direct chelation with zinc ions.

Discussion
The overall results clearly reveal the potential of a customized virtual screening approach targeting specic active site features for the identication of hit compounds for MBL/MBL-fold/ metallo-enzymes. We successfully used an NMR-based approach employing di-Zn(II)and apo-VIM-2 to identify hits from virtual screening that bind to both the apo-and the di-Zn(II) containing enzymes, hence are less likely to inhibit solely via direct zinc ion coordination. Indeed, the approach led to some compounds which bind via direct zinc ion chelation and some which do not, e.g. 16 and 17, as supported by subsequent crystallographic analyses.
The combined biophysical analyses reveal 3-oxoisoindoline-4-carboxylate derivatives as a new class of MBL inhibitor, which do not, at least with the analysed compounds, strongly chelate the active site zinc ions. MBL selectivity proling analyses showed 3-oxoisoindoline-4-carboxylate derivatives can be highly selective for VIM-2, even with respect to very closely related variants, e.g. VIM-5. The apparently narrow selectivity of the 3-oxoisoindoline-4-carboxylate derivatives may limit their use for potentiating b-lactam antibacterials, since clinically useful MBL inhibitors should be broad spectrum targeting multiple MBL types, at least the B1 subclass. However, the limited SAR results presented here suggest that increasing the potency and broadening the spectrum of the 3-oxoisoindoline-4-carboxylate inhibitors towards clinically relevant MBLs should be possible. Most importantly, the results reveal that non-metal-chelating inhibitors present a route to new types of selective MBL-fold enzyme inhibitors. This may be particularly useful in targeting human MBL-fold enzymes involved in cancer drug resistance, 8,9 e.g. where highly selective inhibition is desirable. When coupled with appropriate assays, such selective inhibitors may be useful in proling clinically observed MBL variants.
From a mechanistic perspective, the biophysical analyses are of interest because the crystallographically observed binding modes for the 3-oxoisoindoline-4-carboxylate inhibitors, to some extent, mimic those of intact b-lactam substrates to MBLs. In this regard, the interactions of the 3-oxoisoindoline-4carboxylate inhibitors with the L3/L10 loops are of particular interest, since it is possible, at least in some cases, that these loops may contribute to the highly efficient nature of MBL catalysis by capturing potential substrates and delivering them to the zinc ion containing active site for hydrolysis.