Structure activity relationship studies on rhodanines and derived enethiol inhibitors of metallo-β-lactamases

Graphical abstract


Introduction
Following the clinical introduction of the penicillins in the 1940s, b-lactam antibiotics came to be, and remain, amongst the most important medicines in use. 1 The remarkable longevity and the widespread ability of b-lactams to act as antibiotics has been achieved in the face of multiple resistance mechanisms, 2-4 the most prevalent of which is mediated by b-lactamases which catalyse the hydrolysis of b-lactams. 2,[5][6][7][8][9][10] There are two mechanistic types of b-lactamase -the serine b-lactamases (SBLs), which employ a nucleophilic serine (Ambler classes A, C, D) and the metallo-b-lactamases (MBLs), which utilise a Zn(II) bound hydroxide in b-lactam hydrolysis (class B). 6,9 Inhibitors of the class A and C SBLs have been used successfully in combination with b-lactams. 11 More recently, a broad spectrum inhibitor of class A, C, and some D SBLs, 12 Avibactam, has been introduced for clinical use in combination with a cephalosporin. 13,14 However, no clinically useful MBL inhibitors are currently available, 15,16 and most SBL inhibitors are susceptible to MBL catalysed hydrolysis (Fig. 1). 17 The class B MBLs all utilise one (subclass B2 and some B3) or two (subclasses B1 and some B3) Zn(II) ions at their active site. 18 The B1 MBLs are the most important MBLs from a clinical perspective. 19 Developing MBL inhibitors with the breadth of activity required for clinical application is challenging, because of variations in the mobile regions surrounding the active sites of B1 MBLs. 20 Various types of MBL inhibitor have been developed, 21 most of which chelate to the active site Zn(II) ion(s); however, few if any, of the reported inhibitors have the required breadth of potency against the three major B1 MBL families that are clinically widespread (i.e. the New Delhi MBL (NDM), Verona integron-encoded MBL (VIM), and Imipenemase (IMP) MBLs). We have developed an assay platform for MBLs employing a fluorogenic cephalosporin substrate, 22 which we used to screen potential b-lactamase inhibitors. As part of this work we tested the potency of the rhodanine ML302 (Scheme 1), which was identified following a high-throughput screen, as inhibitor of VIM-2 and IMP-1. 23 Unexpectedly, we found that ML302 undergoes hydrolysis to give the enethiol fragment, ML302F (Scheme 1), which inhibits MBLs via active site Zn(II) ion chelation; in the case of VIM-2 we observed the unusual formation of a ternary complex between the enzyme and two different ligands, ML302 and ML302F. 24 We now describe structure activity relationship and biophysical studies on the rhodanine derived enethiol MBL inhibitors. 25

Experimental procedures for synthesis
The syntheses of 3a-s, ML302 analogues 5a-s, and ML302F analogues 6a-s were performed as previously described 24 . Following the procedure of Brem et al., 24 10 was prepared in two steps from the corresponding aldehyde via Knoevenagel condensation with rhodanine followed by amide coupling (Scheme 2).
Following the procedure of Shaffer et al., 27 a-mercaptocarboxylic acids 13a and 13b were prepared in two steps from the corresponding a-bromocarboxylic acids via nucleophilic substitution with potassium thioacetate followed by basic hydrolysis (Scheme 3).
Following the procedure of Braña et al., 28 a-hydroxycinnamic acid 7 was prepared in two steps from the corresponding aldehyde via Erlenmeyer azlactone synthesis of 6 followed by acid hydrolysis (Scheme 4).
a-Hydroxy phosphonic acid 22 and a-sulfanyl phosphonic acid 26 were synthesised according to the procedure of Bebrone et al. 29 (Scheme 5). Compounds were characterised as detailed in Supplemental information.

Inhibition analyses
Inhibition analyses against bacterial MBLs and SBLs were performed as described previously. 22,24,26 Residual enzyme activities were determined for a range of inhibitor concentrations. Non-linear regression fitting of IC 50 curves was carried out using a threeparameter dose-response curve in GraphPad Prism. Errors in IC 50 values are expressed as: Scheme 1. Synthesis of enethiol based b-lactamase inhibitors. (a) Route for preparation of ML302 5a-q analogues and ML302F 6a-q analogues.24 (b) R groups for 5a-q and 6a-q. MW: microwave irradiation.   Scheme 3. Synthesis of racemic a-mercaptocarboxylic acids 13a and 13b 26 .
Additional data are presented in Supplemental Information (Table S1 and Figs. S1-S6).

NMR time course experiments
NMR experiments were carried out using a Bruker Avance III 700 MHz machine equipped with a TCI inverse cryoprobe or a Bruker Avance III HD 600 MHz spectrometer equipped with a Prodigy cryoprobe at 298 K. Data were analysed using Bruker Topspin 3.5. Processing of spectra was done with a Lorentzian line broadening of 0.3 Hz. Chemical shifts (d) are given as parts per million (ppm) relative to residual HDO (dH 4.70 ppm for 1 H NMR).
For rhodanine stability studies solutions were buffered in either freshly prepared NH 4 HCO 3 (50 mM, pH 7.50) or Tris-d 11 (50 mM, pH 7.50) both with NaCl (100 mM) in D 2 O. ML302 stock solution (50 mM in DMSO d 6 ) was added to a sample to give a final concentration of 200 mM. When specified, NDM-1 was added to give a final concentration of 1 mM. The reaction was followed at 1 h intervals over 18 h (Figs. S7 and S8).

Crystallography
BcII crystals were prepared using the sitting drop vapour diffusion method (293 K, 200 mM ammonium sulfate, 100 mM bis-Tris buffer pH 5.5, 25% w / v polyethylene glycol 3350, and 5 mM inhibitor). The crystals were cryoprotected using well solution diluted to 25% v / v glycerol before being flash cooled in liquid nitrogen. All data sets were collected at 100 K. All data were autoprocessed at the beamline using xia2. 30 The structures were solved using molecular replacement (using PDB ID 4TYT as a search model 24 ) within PHASER. The structures were then fitted to the electron density and refined using COOT 31 and PHENIX 32 until R work and R free no longer converged.
For VIM-2, crystals were grown as reported 25 ; crystal soaking was performed by directly adding 32.5 nL of a 100 mM stock of ML302F in DMSO to 300 nL crystal drops using a Labcyte Echo 550 acoustic drop dispenser, which is part of the XChem pipeline at Diamond Light Source. 33 Crystals were soaked with the ligand for 135 min before the addition of 300 nL of 50% v / v glycerol and flash cooling. X-ray diffraction data were collected at Diamond Light Source beamline I04-1, and processed with Diamond's automated processing pipelines, using xia2 28 and XDS, 34 with XChemExplorer 35 and Dimple used for electron density generation. Initial ligand bound electron density was identified using PanDDA. 36 Grade 37 was used for ligand restraint generation. Final model preparation was performed by iterative cycles of refinement using REFMAC 38 and model building in Coot. 31 Data collection, PDB codes, and refinement statistics for all structures are given in Table S2.
Consistent with a literature report, 41 we observed decomposition of the ML302F analogues 6a-q in DMSO. Thus, biochemical assays were performed by using the corresponding sodium salts (conversion with 100 mM sodium bicarbonate immediately prior to assay), and characterizations were performed in MeOD (see Section 3 in Supplementary Information). The (ML302F) 6 analogues showed good stability as crystalline solids in their acid form, after purification by re-crystallisation from toluene.
For comparison with the enethiols, a-mercaptocarboxylic acids 13a and 13b were prepared in two steps from the corresponding a-bromocarboxylic acids (11a and 11b, respectively), following the procedure of Shaffer et al. 27 via nucleophilic substitution with potassium thioacetate followed by basic hydrolysis (Scheme 3).
The a-hydroxycinnamic acid enol analogue 17 of ML302F was prepared from the corresponding aldehyde via Erlenmeyer azlactone synthesis of 16 followed by acid mediated hydrolysis (Scheme 4). 28 To investigate the importance of the carboxylate 52
A number of trends for the rhodanine derived inhibitors (compound series, 3a-q, 5a-q, and 6a-q) are apparent from the results ( Table 1). In all cases the enethiols (6a-q) were the most potent inhibitors within a given set of rhodanine/enethiol derivatives, implying that the enethiols are the prime source of inhibition. With a few exceptions (mostly in the case of VIM-2 and the atypical subclass B1 MBL, SPM-1), the rhodanine-3-acetic acids (3a-q) were either inactive (at 50 mM) or only weakly active compared to the enethiols (6a-q). This is also the case for the amides (ML302, 5a-5q) against certain MBLs, though there were more exceptions (e.g. 5i-q). It is likely that, at least to some extent, the activities for the rhodanine-3-acetic acids (3a-q) and amides (ML302, 5a-5q) result from (partial) hydrolysis of the compounds to give their corresponding enethiols (6a-q). The differences in activities between rhodanine-3-acetic acids (3a-q) or amides (ML302, 5a-5q) may in part reflect the extent of hydrolysis. Whether or not such hydrolysis is enzyme catalysed is difficult to (partial) judge given the potency of enethiol (6a-q) inhibition. The proposal of enzyme mediated hydrolysis is supported by the different results observed for analogous amides and enethiols. Thus, amide ML302 manifests similar inhibition compared to the enethiol ML302F for two of the tested B1 subclass MBLs (IMP-1 and VIM-2), whereas for SPM-1, BcII and NDM-1, ML302 was $5, $4 and $15-fold less active than ML302F. For the subclass B2 MBL CphA, ML302 (IC 50 value > 50 mM) was also significantly less active than ML302F (IC 50 value 200 nM). Thus, ML302 may be hydrolysed at different rates by different MBLs.
Aside from the previously reported formation of ternary complexes 24 (Fig. 2), some of the results do, however, suggest the intact rhodanines may have inhibitory activity as precedented by work from Spicer et al. 53 Interestingly, although the 2,4-dioxo-1,3-thiazolidin analogue (10) of ML302 is less active than ML302 against all tested MBLs, it did manifest activity against SPM-1, BcII, and VIM-2, being much less active against IMP-1 and particularly, NDM-1. We did not observe hydrolysis of 10 by NMR on the timescale of the inhibition studies, 1 H NMR (700 MHz) in the presence or absence of NDM-1 MBL (Fig. S1). Although we cannot rule out the hydrolysis of 10 to form ML302F at low levels, these results suggest that further SAR studies on the intact rhodanine scaffold, or preferably more stable analogues of it, are of interest. One possibility is that the intact rhodanines can bind to the active site in a manner not involving metal chelation, as recently reported for another series of MBL inhibitors. 54 The results (Table 1) reveal some SAR trends -some of the di-/ tri-substituted amides and enethiols were clearly more active than the mono-substituted compounds. However, because the amides may be under-going hydrolysis/inhibiting via more than one mode of action, care must be taken in comparing the SAR for the two series. Thus, in most cases, the amides 5a-5e, showed almost no inhibition against all the subclass B1 MBLs (IC 50 values $50 or >50 mM), when the para-position of the phenyl ring was mono-functionalised with halogen or alkyl groups. By contrast, e.g., the monoortho-substituted 5f-5i (IC 50  We then synthesised and tested a set of analogues to investigate the importance of the different functional groups in the enethiols (6a-q, ML302F). The hydroxyl analogue of 6c, i.e. 17 (Table S1), was near inactive (at 100 mM), as was the 2-methyloxazol-5(4H)one (16), precursor of 17, supporting the importance of the sulphur atom for binding and inhibition. The phosphoric acids 22 and 26 were also inactive, implying the importance of the carboxylate in inhibition ( Table S1). The saturated analogues of 6a, i.e.  for 6a, 13a and 13b; 2.1 mM, 71.0 mM and 130.0 mM, respectively). 13a was more active than 13b against BcII and NDM-1. The hydroxyl analogues, i.e. mandelic acid and 3-phenylactic acid, were inactive (at 100 mM), supporting the importance of the thiol for potent MBL inhibition in this series ( Table 2). The observations support previous findings for the use of the a-mercaptocarboxylic acid motif for MBL binding/inhibition. 55 Although relative acidity of the functional groups may be a factor, the increased inhibition observed for the (racemic) saturated a-mercaptocarboxylic acids (13a, 13b) compared to the analogous enethiol (6a) could be a result of the different spatial relationship of the thiol and the carboxylate, enabling the saturated a-mercaptocarboxylic acids to bind better.

Structural studies
Previously, we have reported crystallographic studies of VIM-2 and the BcII MBLs in complex with ML302F and in the case of VIM-2, ML302. 24 In the case of BcII, a single ML302F molecule was apparent at the active site. 24 Unexpectedly, when ML302 was co-crystallised with VIM-2, each of the two molecules in the asymmetric unit had ML302F chelating Zn(II) at the active site. However, an additional molecule of ML302 that was apparent only near the active site of chain A (and not chain B), was positioned to interact with ML302F, via staggered p-stacking between the rhodanine ring of ML302 and the 2,3,6-trichlorophenyl ring of ML302F (Fig. 2). 24 In order to further investigate the mode of enethiol binding to MBLs in relation to our SAR results, we obtained four additional high resolution crystal structures of BcII co-crystallised with enethiols 6c (Fig. S9, PDB ID: 5JMX), 6k (Fig. S10, PDB ID: 6EUM), 6L (Fig. S11, PDB ID: 6EWE), and 6s (Fig. 3, PDB ID: 6F2N). We also obtained a new structure of VIM-2 in complex with ML302F (Fig. S12, PDB ID: 6EW3) using a low volume soaking method. 56 (Note: The geometric restraints generated by GRADE for the C@C double bond length of the enethiols reported here are 1.4 Å, whereas that in our previously reported VIM-2:ML302F structure (PDB: 4PVO 24 ) was slightly longer (1.5 Å) due to the geometric restraints output by ELBOW. 24 ) Comparison of the BcII structures in complex with the different enethiols reveal that the core enethiols have a remarkably similar binding mode to ML302F, with the thiol(ate) displacing the bridging water molecule normally located between the two Zn(II) ions (Zn1 and Zn2) and the inhibitor carboxylate ligating to Zn2. The enethiol linked phenyl rings of the inhibitors all occupy the same region of the active site (Fig. 3). As observed for binding of the products of MBL-catalysed b-lactam hydrolysis, in the structure Table 3 Observed inhibition of MBLs by a 1:1 mixture of rhodanine amides (5) and enethiols (6)  Note: that in most cases the mixture is of similar potency to the enethiol alone, but that in a few cases (notably ML302/ML302F) the mixture is more potent. of NDM-1 (PDB ID: 4EYF) 56 (Fig. S13), one of the enethiol carboxylate oxygens is positioned to interact with Zn2. The other oxygen is positioned to interact with the N e amino group of Lys-224 (Fig. 4).
As observed in the VIM-2:ML302F complex, the plane of the phenyl side chain on all of the enethiol inhibitors is rotated about the C3-C4 bond such that it is not co-planar with the plane of the enethiol alkene, likely hindering conjugation. For ML302F the skewed arrangement was thought to be, at least partially, caused by steric hindrance due to the ortho di-chloro substituents on the phenyl ring as proposed previously, 24 Because enethiols without ortho-substituents are also observed to retain similar conformations (as evidenced by a crystal structure of BcII in complex with 6c, Fig. S9), the ortho-substitution may not be an essential factor in obtaining potent inhibition by the enethiols.
Superimposing the structure of BcII in complex with 6s and VIM-2 in complex with ML302F, implies that there may be a steric clash between the naphthalene side chain of 6s and Tyr67 on the L3 loop of VIM-2, suggesting unfavourable binding. However, the BcII and VIM-2 IC 50 values for 6s are comparable (IC 50 = 0.2 mM and 0.1 mM, respectively), indicating 6s might adopt a different conformation when binding to VIM-2 and/or that it induces a conformational change of the VIM-2 L3 loop.

Conclusions
The overall results reveal that rhodanine derived species have potential as broad spectrum MBL inhibitors, which might be in part due to the proposal that enethiol carboxylate binding mimics that of b-lactam hydrolysis product (Fig. S13). Their capacity to inhibit SBLs and penicillin binding proteins appears more limited, at least among those compounds tested in this study. 24,26 Although the enethiols (6a-6q), which are derived by rhodanine hydrolysis, are the most potent of the series identified, the SAR on compounds with intact rhodanine ring structures suggests that rhodanine related heterocycles that do not chelate via a thiol/sulphur may also have potential as MBL inhibitors. Recent work on another series suggests that such compounds have potential to inhibit without active site metal chelation. 57 The proposal of different binding modes for the enethiols (6a-61) and rhodanine amides (5a-51) is supported by the observation of only partially overlapping SAR trends for the two series.
We have previously reported structural evidence that ML302/ ML302F can form a ternary complex with VIM-2. 24 The SAR results presented here suggest that formation of such a ternary complex is not a general feature of rhodanine derived MBL inhibition and hence, although interesting, is unlikely to be a productive path for the development of broad spectrum clinically useful MBL inhibitors.
Rhodanines are often characterised as 'difficult to progress' and 'promiscuous' compounds. [57][58][59] Our work reveals further complexities involved in interpreting assay results involving rhodanines. Despite their complex nature, one rhodanine is clinically approved for use in nerve damage due to diabetes mellitus (Epalrestat Ò , an aldose reductase inhibitor) 60 and other rhodanine-related heterocycles are in development. 61 The results presented here support the proposal that rhodanines (at least) have potential as promiscuous enzyme inhibitors/protein binders, in part owing to their tendency to undergo hydrolysis to products, including enethiols, which have potential to inhibit the multiple metallo-enzymes present in cells, including related MBL fold enzymes, which have important biological roles beyond antibiotic resistance including in nucleic acid repair and metabolism. 62 Our results also imply that any medicinal chemistry studies employing rhodanine inhibitors should as a matter of course include testing of their hydrolysis products.