Structure-Based Design of Reversible, Quinoline-2(1H)‑one Inhibitors of Serine- and Metallo-Carbapenemases

Abstract

Carbapenems are essential β-lactam antibiotics whose clinical utility is now threatened by emerging carbapenemases, including the Class A serine β-lactamase KPC-2 and the Class B metallo-β-lactamase NDM-1. Here, we describe a comprehensive, structure-based assessment of active-site binding determinants for KPC-2 and NDM-1 in the context of reversible, noncovalent inhibition by a quinoline-2­(1H)-one phosphonate scaffold. Systematic substitution of the scaffold core revealed the N1 and C7 positions as most tolerant of diverse substitution, while the less tolerant C3 and C6 positions nevertheless drive affinity and binding mode when modified with a C3-methyl or C6-fluoro substituent, respectively. We also describe biophysical and computational studies aimed at determining the pharmacological significance of the 2:1 ligand binding stoichiometry observed in several NDM-1 complex structures, concluding that only one of these binding events is relevant for potent NDM-1 inhibition. Although the current inhibitors lack significant whole-cell activity, the structural and biochemical findings described provide valuable information for the targeting of evolutionarily and mechanistically diverse carbapenemases.

Introduction

Characterized by a strained and reactive four-membered ring, the eponymous β-beta lactams are the most prescribed class of antibiotics and are subdivided by class: the penicillins, cephalosporins, carbapenems, and monobactams. The carbapenems (e.g., doripenem, Meropenem, and imipenem) have long been considered antibiotics of last resort due to their critical efficacy in treating serious, multidrug-resistant infections including those producing common β-lactamases. The β-lactamases can be subgrouped as class A, C, and D enzymes, which are all serine β-lactamases (SBLs), or class B enzymes, which are metallo-β-lactamases (MBLs). While the different classes of SBLs differ in the overall sequence homology, all contain an S-X-X-K motif, wherein the catalytic serine is the nucleophile that attacks the β-lactam carbonyl. Class B MBLs utilize one or two zinc ions within the enzyme active site to activate a water molecule for attack on the β-lactam ring.

Among β-lactamases, the so-called carbapenemases are of particular concern clinically due to their broad-spectrum activity against a range of β-lactams, including carbapenems. Carbapenemases are produced by the Gram-negative ESKAPE pathogens Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and the Enterobacteriaceae and include the class A SBL KPC-2, the class B MBLs NDM-1 and VIM-2, and the class D SBL OXA-48 , among others.

Combination of a β-lactamase inhibitor (BLI) with a β-lactam antibiotic is a clinically proven strategy to counter antibiotic resistance. BLIs function by targeting β-lactamases so the coadministered β-lactam can act upon the original PBP target, ultimately leading to cell death. First-generation β-lactam-based BLIs include clavulanic acid, tazobactam, and sulbactam, which inhibit SBLs but are not active against MBLs. Newer, non-β-lactam-derived BLIs such as avibactam, vaborbactam, and relebactam target a broad spectrum of class A, C, and D SBLs but lack activity against MBLs. There is thus an urgent need to identify broad-spectrum BLIs that retain activity against class A, B, C, and D enzymes since bacteria can produce β-lactamases of multiple classes. In this light, the cyclic boronate taniborbactam, currently under regulatory review, is notable for its broad-spectrum activity against both serine and metallo-β-lactamases. ,

Here, we describe an effort to define the binding determinants of reversible, nonelectrophilic BLIs based on a quinoline-2­(1H)-one methylphosphonate scaffold (Figure ). Initial studies of this novel scaffold largely using commercial analogs succeeded in identifying analogs with low-nM activity against KPC-2; unfortunately, only modest mid-μM activity was evident against NDM-1, a clinically relevant MBL. The objective of the current study was to broadly define the structural determinants of binding by this scaffold in exemplar SBL and MBL enzymes. A more versatile synthetic approach to the quinoline-2­(1H)-one core enabled exploration of nearly every position on the core heterocycle. The solution of multiple complex crystal structures identified several binding hotspots and key polar interactions, yielding improved analogs with similar, low-μM, or better activity against both SBL and MBL enzymes. The structural studies revealed two distinct binding modes for this class in KPC-2, which largely depend on the nature of the substitution at N1 and C3. In the MBL exemplar NDM-1, the structures revealed an unexpected 2:1 ligand binding stoichiometry for several analogs. We used isothermal titration calorimetry (ITC) and molecular dynamics simulations to show that only the first binding event is responsible for the low-μM inhibition observed biochemically. Finally, selected analogs were assessed for microbiological activities in combination with a carbapenem and for their ADME and PK properties, identifying the limitations of current leads while also setting the stage for their further optimization.

1.

Structure of the coumarin and quinoline-2-(1H)-one methylphosphonate BLI scaffold (at left) and the additional SAR and structure-based analog generation detailed herein (at right). Ring numbering as employed in the following discussion is indicated.

Results

Previous SAR studies of the quinoline-2-(1H)-one scaffold showed that C5/C7 substitution with halogens, as in 5,7-dibromo-quinolin-2­(1H)-one-4-methylphosphonate, afforded low-nM biochemical potency against KPC-2 but only mid-μM activity against the important MBL NDM-1 (Figure ). Although potent against the SBL exemplar, the presence of two bromine atoms led to suboptimal physiochemical properties. The current study thus aimed to identify more drug-like substituents at C7, using as a reference the C7-monobromo analog 1 (Table ). An improved synthetic route starting from commercially available isatins (vide infra) enabled exploration of both carbocyclic and heterocyclic ring systems at C7, including 3–5 membered cycloalkyl substituents (analogs 2–6), as well as pyrrolidine (7–8), piperidine (9), and morpholine congeners (10). These analogs were evaluated for inhibitory activities against KPC-2 and NDM-1 as exemplars of clinically relevant exemplar SBL and MBL enzymes, respectively (Table ).

1. Structures and KPC-2 and NDM-1 IC₅₀ Values of BLI Analogs Exploring SAR at C7, C6, and N1.

^aL.E. = ligand efficiency in units of kcal mol^–1(number of heavy atoms)⁻¹.

This first survey of the C7 substitution identified several notable trends. As compared to 1, we found that replacing C7 bromine with cyclopropyl (analog 2) produced similar potencies against both SBL and MBL enzymes, while cyclobutyl (5) and cyclopentyl (6) analogs had notably ∼5-fold better KPC-2 potency while retaining or slightly improving (for 5) NDM-1 activity. The analogous N-pyrrolidine (7), N-piperidine (9), and N-morpholine (10) comparators were somewhat weaker against KPC-2, but NDM-1 activity was quite dramatically improved for N-pyrrolidine 7 at IC₅₀ = 1.0 μM, a substantial improvement over the best NDM-1 activity observed previously from this scaffold. Overall, SAR at C7 suggested good tolerance for three- to six-membered cyclic substituents, with four- or five-membered rings affording the best combination of SBL and MBL activity.

Further exploring SAR in a background of C7 cyclopropyl or N-pyrrolidine substitutions revealed the favorable effect of N1 methylation on both KPC-2 and NDM-1 activity, as most dramatically revealed for the N-pyrrolidine analogs (cf. 7 vs 8). Next, a focused library of analogs with as yet unexplored C6 substitution was prepared to evaluate the steric and electronic preferences at this position (Table S1, Supporting Information). We found that more sterically demanding branched substituents (e.g., i-Pr and OCF₃) were poorly tolerated at C6, particularly with respect to NDM-1 activity (IC₅₀ >100 μM). By contrast, smaller electron-withdrawing substituents such as CN and F were beneficial and resulted in improved activities against both SBL and MBL enzymes. Next, the feasible C6 fluorine substitution was combined with a selection of more favored C7 substituents, including Br (11), cyclopropyl (12), N-pyrrolidine (13), and morpholine (14) (Table ). Gratifyingly, the C6 fluorine group produced a consistent ∼10-fold or greater boost in KPC-2 activity across the series (i.e., 11–14), when compared to their des-F comparators (1–2, 7, and 10, respectively). While the same trend unfortunately did not extend to NDM-1 activity, analogs 11–14 at least retained the NDM-1 activity of their des-F congeners, with slightly improved activities observed in the case of 12 and 14. To summarize, a focused survey of C6 and C7 substituents revealed the beneficial combination of C6 fluorination with small to medium-sized aliphatic or heteroaliphatic rings at C7 and provided promising new leads 12 and 13, which combine sub-μM activities against KPC-2 with single-digit μM activity against NDM-1.

To better understand the interplay of C6 and C7 substitution in SBL and MBL inhibition, we turned to X-ray crystallography, solving complex crystal structures of the fluorinated analogs 12 (C7 cyclopropyl), 13 (C7 pyrrolidine), and 14 (C7 morpholine). In their KPC-2 complex structures, all three analogs adopt the canonical binding mode for this scaffold, wherein the phosphonate moiety is anchored through multiple hydrogen bonds to polar residues of the β3 strand, while the C7 substituent contacts a hydrophobic shelf formed by Pro104 and Leu167 (Figure A–C). The solvent-exposed nature of this hydrophobic site likely explains why both smaller (cyclopropyl) and larger (morpholine) ring systems are well tolerated at C7 (Table ). Interestingly, the C6 substituent is directed into a narrow cleft of the active site, likely explaining the poor tolerance for branched substituents at this position. The broadly favorable effect of C6 fluorination in 12–14 was nicely revealed by the close contact of the C6 fluorine atom with the terminal carboxamide N–H bond of Asn132 (∼3.2 Å F–N distance), implying an energetically favorable C–F···H–N hydrogen bond (Figure A–C). Meanwhile, the N1 methyl group of these analogs is positioned 4 Å from Thr237 side chain Cγ2, an optimal distance for van der Waals interactions, and 4.8 Å from the Gly239 main chain Cα. The nonpolar interactions involving the N1 methyl group may therefore explain its contribution to KPC-2 binding. An aryl stacking interaction with Trp105 appears most optimal in the case of 12 (Figure A), which could explain the excellent KPC-2 potency of this analog despite a smaller cyclopropyl C7 substituent with less hydrophobic surface available to contact the Leu167 site. By contrast, in the structure of 13 (Figure B), Trp105 swings entirely away from the quinoline-2­(1H)-one core.

2.

Complex crystal structures of analogs 12 (panel A, PDB code 9D54), 13 (panel B, PDB code 9DTJ), and 14 (panel C, PDB code 9D2U) bound to KPC-2, illustrating the conserved binding modes and interaction of the C6 F atom with Asn 132 between 12 (measured at 2.9 Å) and 14 (measured at 3.1 Å). The distance between C6 F and Asn132 in the complex structure of 13 is 3.7 Å, being too far for a hydrogen bond.

In our prior structures of coumarin and quinolin-2­(1H)-one analogs bound to MBL enzymes NDM-1 and VIM-2, the phosphonate function interacted with both Zn centers, with some differences in the exact location of the phosphonate group depending on the pH of the crystallization buffer. Thus, at low pH (pH 3.8) favoring partial protonation of the phosphonate function, this group was found to displace the nucleophilic hydroxide/water that bridges the two Zn centers in apo structures and is required for catalysis. By contrast, at a physiological pH (pH 7.5), the phosphonate group moved away from this location, and the catalytic water was retained in the complex structures. In either binding mode, the inhibitors occupied the substrate binding site and directly or indirectly contact the catalytic machinery, suggesting a clear hypothesis as to their mechanism of inhibition.

Next, complex structures of the new, more potent analogs 2, 3, and 13 bound to NDM-1 were solved at pH 7.5 (Figure ). The structure of 3 revealed an overall binding mode similar to that of the early inhibitors, with the phosphonate moiety near the catalytic Zn ions and retaining the bridging nucleophilic hydroxide/water molecule. The orientation of the quinoline-2­(1H)-one core of 3 was distinct from the earlier structures, however, lying sandwiched between Asn220 and Met67, with the C-7 cyclopropyl substituent making contacts with Phe70 (Figure A).

3.

Complex crystal structures of analogs 3 (panel A), 2 (panel B), and 13 (panel C) bound to NDM-1. In contrast to the structure of 3 (PDB 9O2W), the structures of analogs 2 (PDB code 9DEW) and 13 (PDB code 9DAG) include two ligands arranged in a head-to-tail stacking interaction. The 2Fo-Fc electron densities for analogs 3 and 2 are shown at 1.5σ.

While the binding mode of 3 echoed that of the earlier NDM-1 structures, the C-7 cyclopropyl (2) and pyrrolidine analogs (13) showed an unexpected 2:1 ligand:protein binding stoichiometry not observed for 3, nor for the early inhibitors (Figure B and 3C). The phosphonate of one ligand copy of 2 and 13 (denoted “copy 1”, Figure B,C) displaces the catalytic water, while the quinoline-2­(1H)-one ring contacts Val73, which along with Met67 forms the “floor” of the binding site. Another seemingly important contact in the new structures is a C–H···π interaction between the N–Me substituent of 2/13 (copy 1) and the aromatic surface of Phe70 (Figure B and 3C). This was interesting, as the early analogs lacked N-substitution, and in their complex structures, Phe70 was observed in multiple rotameric states, only some of which seemed able to form a possible edge–face interaction with the coumarin or quinoline-2­(1H)-one core. The importance of the C–H···π interaction with N–Me analogs is further supported by the detrimental effects of larger N1 substituents on NDM-1 activity, as discussed later.

The additional molecule (“copy 2”) of 2/13 observed in the complex structures is stacked above the first and in a longitudinally opposed, head-to-tail arrangement, of the quinoline-2­(1H)-one rings (Figure B and 3C). The phosphonate function of the second ligand makes direct contact with only one of the two Zn atoms while also forming a hydrogen-bonding contact with Lys221. The C6 fluorine atom in analogs 3 and 13 makes no direct contacts with NDM-1 residues (unlike in KPC-2), likely explaining why C6 fluorination is tolerated but does not improve NDM-1 activity (Table ).

2. Structures and KPC-2 and NDM-1 IC₅₀ Values of BLI Analogs Exploring SAR at C7, C6, C3, and N1.

^aL.E. = ligand efficiency in units of kcal mol^–1 (number of heavy atoms)⁻¹.

Intriguingly, MBL structures with 2:1 ligand binding stoichiometry were observed previously by us in a VIM-2 structure (unpublished) and by Yan et al., in multiple complex structures of reversible inhibitors bound to the MBL enzyme IMP-1, perhaps suggesting that this mode of binding is common in MBL enzymes. However, whether 2:1 ligand binding occurs in solution and is relevant for enzyme inhibition was not addressed in earlier studies. The fact that C7 cyclopropyl analogs 2 and 3 showed similar NDM-1 IC₅₀ values despite different binding stoichiometries in their respective complex structures suggests that the second binding event may be a lower-affinity one. On the other hand, the nicely self-complementary arrangement of the stacked ligands, combined with the additional protein contacts the stack affords, makes it difficult to rule 2:1 binding as contributing to enzyme inhibition in solution.

To provide additional insight, isothermal titration calorimetry (ITC) was employed to assess the binding stoichiometry of compound 13 in solution. This particular analogue was chosen because it exhibited a reasonable IC₅₀ of 3.50 ± 0.26 μM against NDM-1 and unambiguous density for two ligand copies in its complex structure (Table and Figure C). Thus, the heat released upon 13 binding to NDM-1 in solution was fit to a binding model, providing a measured K _d = 3.40 ± 0.51 μM, that was in good agreement with the biochemical IC₅₀ (Figure ). The ITC data was consistent with a binding stoichiometry near unity, which provides good evidence that inhibition of NDM-1 by 13 is driven by a single stoichiometric (1:1) binding event in the low μM regime, while binding of a second copy occurs in the context of crystallographic soaks where high concentrations of ligand (10 mM) are employed.

4.

(A) Thermogram of ITC runs performed at 200 μM 13 into 30 μM NDM-1 in 500 μM Zn²⁺ buffer (black line). A baseline heat release of 13 directly into the buffer with no protein present was also performed (orange line). Data is plotted with exothermic reactions down. (B) Heat release as a function of compound/protein mole ratio (exotherm up). Data points were then fit to a nonlinear regression (independent model of binding) to determine thermodynamic parameters. (C) Table of thermodynamic parameters measured from ITC curve.

With ITC data indicating stoichiometric ligand binding, it was of interest to determine which copy of the ligand in the crystal structures of 2 and 13 represents the high-affinity, biochemically relevant binding event. As noted above, the phosphonate function of copy 1 (in 2 and 13) displaces the Zn-bound nucleophilic water/hydroxide, mirroring the binding mode of single-copy binder 3 (Figure A). This suggested copy 1 as the high affinity, biochemically relevant binding event. Moreover, we note that the N1 methyl substituent of copy 1 in both 2 and 13 forms an apparent C···H···π interaction with Phe70, which places its C7 ring within van der Waals contact of His 250 (4.2 Å). We propose that this arrangement of copy 1, together with His250 and Phe70, serves to template a binding surface for the second, low-affinity event. This notion appears consistent with the observation of single-copy binding in the structures of N1 des-Me analog 3 as well as several other des-Me quinoline-2­(1H)-one and coumarin analog structures described previously.

The proximity of Phe70 to N1 of the quinoline-2­(1H)-one position motivated the exploration of larger side chains in this position, which in KPC-2 is mostly solvent exposed but close to a pocket formed by His274 and the main chain atoms surrounding Gly239. Thus, a library of N1-substituted analogs was prepared in the background of C7 bromo substitution, including the N1 des-methyl analogue 15 as a reference compound (Table ). Both des-Me analog 15 and N-cyclopentylmethyl analog 16 exhibited slightly enhanced KPC-2 activity over comparator 1 (N–Me), while NDM-1 activity was more substantially improved for 15. The greater ligand efficiency of 15 (L.E. = 0.42 and 0.43 for KPC-2 and NDM-1) as compared to 16 (0.33 and 0.28), indicated that larger N1 substituents as in 16, while tolerated by both enzymes, are likely solvent exposed and not engaged in productive contacts with active-site residues. Additional N1-alkyl analogs closely related to 16, but with cyclohexyl (17), pyran (18), piperidine (19), difluorocyclohexyl (20), or cinnamyl (21) termini showed similar low-μM KPC-2 activity and ligand efficiencies (0.30–0.33), with the exception of 19, which was ∼10-fold less active (L.E. 0.24). The NDM-1 activities of analogs 17–21 were comparable or weaker than 16 with ligand efficiencies falling between 0.22 and 0.28, as compared to L.E. = 0.41 for unsubstituted comparator 15.

Overall, the SAR for N1-substitution suggested that N–H or N–Me substitution was preferred for NDM-1 binding, perhaps due to the productive interaction with Phe70, as hypothesized. Moreover, the lack of improvement in KPC-2 potency indicated a failure of these analogs to effectively bind the crevice site between His274 and Gly239. On the other hand, the larger N1 side chains, when combined with a C3 methyl substituent as in analog 22, gratifyingly produced a >10-fold improvement in KPC-2 activity (IC₅₀ = 180 nM for 22 vs comparator 16), with ligand efficiency now equivalent to N–Me progenitor 1 (L.E. = 0.39). Moreover, NDM-1 activity was also improved with the C3 methyl modification by ∼5-fold for 22 (NDM-1 = 3.55 μM) over comparator 16.

To better understand the synergistic effect of large N1 substituents with C3 methylation, we again turned to crystallography. Interestingly, in the complex structure of 16 with KPC-2, an entirely new binding mode was observed in which the phosphonate function forms the expected polar interactions with β3-strand residues, but the quinolin-2­(1H)-one core was oriented such that its N1 side chain interacts with Leu167, the same site that contacts the C7 substituent in the normal binding mode. The opposing binding orientation is clearly revealed by superposition of the complex structures of 16 and 13 (Figure A). Analog 21 with its more extended cinnamyl side chain also binds in this alternate orientation, with its terminal phenyl ring stacked on Leu167 (Figure B). These structures provide additional evidence for Leu167 as a key binding hot spot in KPC-2. Strikingly, this alternate binding mode of 16/21 leaves a small void near C3 of the quinolin-2­(1H)-one ring, suggesting a rationale for the favorable effect of the C3 methyl group on KPC-2 potency. Indeed, the complex structure of C3 methyl analogue 22 confirmed binding of the methyl group in this very pocket (Figure C).

5.

Superposition (panel A) of the KPC-2 complex crystal structures of 16 (green, PDB code 9D8T) with 13 (wheat) illustrating the opposed binding orientations, in which the N1 side chain of 16 forms hydrophobic contact with Leu167, the site of C7 pyrrolidine binding in 13 and related analogs. Complex structure (panel B) of analog 21 (PDB code 9D5Q), showing its extended N1 side chain and terminal styrene moiety contacting Leu167. Superposition of 16 (wheat) and its C3 methyl congener 22 (panel C, PDB code 9DB7) illustrating how the C3 methyl projects deeper into the KPC-2 binding site, likely explaining its improved KPC-2 activity.

Moving to NDM-1 crystallography, analog 22 is observed to exhibit two-copy binding and the overall similar binding site interactions as we observed for 13, with some important differences. Thus, in the complex structure of 22, Phe70 shifts away from the large N1 substituent and closer to the C7 bromo and C8 position of the quinolin-2­(1H)-one ring (Figure A). At the same time, the two copies of 22 are more substantially offset, and more dramatically inclined, when compared to ligand stacking in 13 (Figures B and S1). It appears that the favorable effect of C3 methylation in 22 (vs 16) derives from contact of copy 1 (the presumed high-affinity interaction) with the backbone and side-chain methylene of His122. Overall then, the structures of 13 and 22 revealed how very similar ligands can be recognized in different ways by exemplar SBL and MBL enzymes. Indeed, these structures exemplify the active-site malleability of carbapenemases in binding diverse β-lactam substrates (and inhibitors) as noted previously.

6.

Stacked binding of the 22 (PDB code 9DB7) dimer in NDM-1 (panel A). Overlay of 22 with structure of 13 (wheat) illustrating shift in the binding of 22 within the active site (panel B). The greater steric demands of the larger N1 substituent in 22 may force the alternate binding mode, compared to the planar π–π stacking seen with 13.

Next, N1 substitution was re-examined in the context of C3 methyl substitution, and either with or without C6 fluoro substitution (analogues 23–30, Table ). Encouragingly, N1-methyl, C3 methyl analogue 23 was ∼4-fold more potent than its C3-des-Me comparator 1 against both KPC-2 and NDM-1, with IC₅₀ values of ∼1.5 and ∼7 μM, respectively. The introduction of C6-fluoro in 24 produced a further ∼4-fold improvement in KPC-2 activity (IC₅₀ = 0.37 μM; L.E. = 0.44), while NDM-1 activity was unchanged. This SAR suggests that analogs 23 and 24 bind KPC-2 in the canonical fashion with the C–F function in contact with Asn132. Moreover, the result reveals that a methyl group at C3 is favored both in the case of small N1 substituents (canonical binding mode as in 13) and with larger N1 side chains (alternate binding mode as in 22).

Moving on to C3 methyl analogs bearing larger N1 substituents (25–28, Table ), we gratifyingly observed KPC-2 activity that was consistently ∼10-fold improved over des-methyl comparators. In these cases, the introduction of C6 fluoro afforded no further improvement in KPC-2 activity (cf. analogue pairs 25/26 and 22/28). Taken together, the C3 and C6 SAR in the context of larger N1 substituents suggests the alternate 22-like binding mode, with the C3 methyl buried and the C6 position solvent exposed. Interestingly, C3 methylation was also generally favorable for NDM-1 activity in analogs with larger N1 substitutions (25–30, Table ). Quite possibly, this reflects the proximity of the C3 methyl to His122 as observed for the complex structure of 22 bound to NDM-1 (Figure ). A final pair of analogs (29 and 30) were prepared in which C3 methyl was combined with a large N1 side chain and also a large C7 moiety. In these two analogs, either binding mode would allow favorable contact with the Leu167 site. However, the 4-fold greater KPC-2 potency of C6 F analogue 30 clearly suggests that 29/30 adopts the canonical binding mode in KPC-2.

In contrast to KPC-2 where structure and SAR provide a clear picture of ligand binding principles, the same could not be said for binding to NDM-1. To provide additional insight, molecular dynamics (MD) simulations were carried out for a representative analog lacking N1 substitution (3), one with small, N–Me, substituent (2), and one with a larger N1 substituent (22). Simulations of 300 ns were performed, starting from the respective crystallographic structures as the initial starting frame. The metal binding sites were parametrized using the MCPB.py script made available through AmberTools23. − Zinc chelating residues and small molecule parameters were determined by density functional theory (DFT) calculations performed at the w-B97XD|def2SVP level of theory. , System stability over the course of the simulations was determined retrospectively via RMSD calculations, and protein–ligand interactions were evaluated using the MD Analysis package in MOE2024.0601. ,

In the simulations, single-copy binder 3 as well as copy 1 of 2 and 22 maintained key protein–ligand interactions observed in the starting frames, most notably with metal-coordinating NDM-1 residues including Cys208, His189, His120, and His122 (Figures S1–S3, Supporting Information). This observation was consistent with the hypothesis that copy 1 represents the high-affinity binding event, given its chelation of both Zn²⁺ ions and its direct interaction with several of the Zn-chelating side chain residues. Other key interactions include contacts with the hairpin loop L3 (residues 63–73) and importantly with Phe70, as already noted from the complex structures. In fact, it is known that NDM-1 L3 plays an important role in enzymatic activity and substrate specificity due to the hydrophobic nature of the L3 side chains. , Interestingly, N–Me analog 2 as well as N1-cyclopentylmethyl analog 22 remained in close proximity to L3 over the entire course of the simulations, a finding consistent with favorable hydrophobic contacts between ligand and the L3 loop. By contrast, the proximity of L3 to compound 3 in the starting frame was not retained over the course of the simulation, suggesting a weaker interaction of 3 with L3 residues due to lack of N1 aliphatic substitution (Figure S3). In light of this, the superior NDM-1 potencies of 2 and 22, which were, respectively, 2-fold and 6-fold improved relative to 3, are consistent with the importance of engaging the L3 residues with lipophilic ring substituents, and in particular at N1. The results of the MD simulations thus provide good evidence for the importance of binding to L3 loop residues and inform a strategic focus for future SAR studies targeting NDM-1.

Next, selected BLIs (6, 7, 13, and 29) were evaluated in combination with imipenem against carbapenemase expressing Gram-negative bacteria using a checkerboard assay. In total, four clinical isolates were employed, comprising P. aeruginosa and K. pneumoniae strains producing either KPC-2 or NDM-1 carbapenemases (Table ). The minimum inhibitory concentration (MIC) of imipenem against K. pneumoniae producing KPC-2 was 256 μg/mL and >256 μg/mL for the NDM-1 producer. The MIC of imipenem was ≥128 μg/mL in all four strains, confirming their carbapenem-resistant phenotype. Avibactam (8 μg/mL) was used as a positive control and reduced the imipenem MIC to 1 μg/mL in both KPC-2 producing strains, consistent with potent KPC-2 inhibition. The four quinoline-2­(1H)-one analogs were tested at 128 μg/mL in combination with imipenem. Unfortunately, none of the compounds lowered the imipenem MIC against either K. pneumoniae or P. aeruginosa strains. Compound 7 showed a modest 2-fold reduction in the MIC of imipenem in NDM-1 expressing P. aeruginosa, which could reflect weak on-target activity (Table ). The disconnect between biochemical potency and cellular activity likely reflects poor uptake in these Gram-negative strains and/or efflux of the inhibitors. Previously, two quinoline-2­(1H)-one analogs lacking N1 substitution showed more promising cellular activity (32–64-fold shifts in imipenem MIC) that was specific to KPC-2 expressing strains. Though the specific strains employed in that study were different from those employed here, these results may suggest an effect of N1 alkylation on bacterial uptake and/or efflux. Whatever the case, more extensive SAR and cellular profiling are required to identify structural determinants of cellular uptake for this scaffold. Along these lines, the chemical modifications recommended in the “eNTRy rules” for uptake by Gram-negative bacteria would appear worthy of investigation.

3. In Vitro Activity (MIC) of Imipenem against Bacterial Strains Expressing KPC-2 and NDM-1 in the Presence of 128 μg/mL of Compounds 6, 7, 13, and 29 .

bacterial strain	β-lactamase	imipenem (μg/mL)	imipenem/avibactam (μg/mL)	imipenem/6 (μg/mL)	imipenem/7 (μg/mL)	imipenem/13 (μg/mL)	imipenem/29 (μg/mL)
K. pneumoniae	KPC-2	256	1	>256	>256	>256	>256
K. pneumoniae	NDM-1	>256	–	>256	>256	>256	>256
P. aeruginosa	KPC-2	128	1	256	256	128	256
P. aeruginosa	NDM-1	>256	–	256	128	256	256

^aAvibactam at 8 μM.

To identify potential ADME/PK liabilities, representative analogs 13, 22, 29, and 30 were evaluated in a panel of standard in vitro ADME assays (Table ). Encouragingly, all four compounds showed excellent stability in the presence of mouse liver microsomes and good kinetic solubility in PBS at pH 7.4 (Table ). Perhaps unsurprisingly, the permeability of exemplar 13 in a Caco-2 monolayer assay was low in both directions, suggesting poor prospects for oral bioavailability. The in vivo pharmacokinetics of 13 were assessed in male CD-1 mice (N = 9) following a single IV dose. Clearance was low to moderate (∼20% of hepatic blood flow), while the plasma half-life was short at <1 h, and the total plasma exposure dropped below the compound’s KPC-2 IC₅₀ by two hours, and below the NDM-1 IC₅₀ by the 0.5 h time point (Tables and S2 and Figure S4). The relatively higher in vivo clearance compared with in vitro microsome stability studies may suggest a role for renal clearance. Overall, the microbiological assessments and ADME/PK studies revealed key areas for improvement in future lead optimization of the series.

4. Selected In Vitro ADME Profiling of Quinoline-2­(1H)-one Analogs.

compound	MLM t 1/2 (minutes)	MLM CLint (μL/min/mg)	kinetic solubility (μM)	Caco-2 A to B (cm/s)	Caco-2 B to A (cm/s)
13	>120	<11.6	506.3	0.1 × 10–6	0.2 × 10–6
22	>120	<11.6	515.9	–	–
29	>120	<11.6	478.8	–	–
30	>120	<11.6	489.3	–	–
verapamil	5.4 ± 0.1	255.0	–	–	–
amiodarone	–	–	<3	–	–
testosterone	–	–	335.7	–	–

5. Pharmacokinetic Parameters for 13 Following a Single IV Dose of 5 mg/kg in Male CD-1 Mice.

PK parameter	value
CL	1.41 L/h/kg
V ss	0.63 L/kg
T 1/2	0.81 h
AUClast	3542 h·ng/mL
AUCINF	3552 h·ng/mL
MRTINF	0.448 h

^aLiver blood flow (mouse) = 7.2 L/h/kg.

Synthesis

All of the final analogs described herein were prepared from one of four key building blocks bearing C7 bromo substitution and with C6 = H or F and/or C3 = H or Me substituents as detailed in the schemes below. First, commercially available isatin building blocks 31 and 32 (Scheme ) were subjected to the Pfitzinger reaction to afford the quinoline-2­(1H)-one heterocyclic core bearing C7 bromo and C4 carboxylate substitution (33 and 34). The carboxylic acid intermediates were reacted with CDI prior to sodium borohydride reduction to afford benzylic alcohols 35 and 36. Conversion of the primary alcohols to bromide 37 and 38 was followed by Arbuzov reaction with triethyl phosphite to afford 39 and 40, which were useful intermediates to prepare final analogs, as further detailed below. Heating 39 or 40 with 4 M HCl in dioxane served to afford the free phosphonic acids 15 and 41, respectively (Scheme ).

1. Synthesis of Key Quinolin-2­(1H)-one Building Blocks. Conditions: (a) Malonic Acid, AcOH, Reflux; (b) (i) CDI, THF, (ii) NaBH₄, THF/MeOH; (c) HBr, AcOH, Reflux; (d) P­(OEt)₃, Dioxane, 100 °C; (e) 4M HCl/Dioxane, 100 °C.

An analogous synthetic approach was employed to produce building blocks bearing a C3 methyl substitution. Thus, starting from isatins 31 or 32, Pfitzinger reaction with propionic anhydride followed by treatment with refluxing aqueous sodium hydroxide afforded the C3 methyl intermediates 42 and 43 (Scheme ). Formation of the mixed anhydride with isobutyl chloroformate, followed by reduction with sodium borohydride, afforded alcohols 44 and 45, which were converted via the bromides 46/47 to key building blocks 48 and 49. Deprotection of 48 and 49 as before afforded phosphonic acids 50 and 51.

2. Synthesis of Quinolin-2­(1H)-one Building Blocks with C3 Methyl Substitution. Conditions: (a) (i) Propionic Anhydride, Reflux, (ii) aq. NaOH, Reflux; (b) (i) Isobutyl Chloroformate, Et₃N, THF, (ii) NaBH₄, THF-MeOH or BH₃-THF; (c) HBr, AcOH, Reflux; (d) P­(OEt)₃, Dioxane, 100 °C; (e) 4M HCl/Dioxane, 100 °C.

The key building blocks 39, 40, 48, and 49 were further modified starting at N1 by reaction with alkyl halides to introduce the R⁴ substituent (Scheme ). Although both the O- and N-alkyl products were formed in these reactions, the desired N-alkylated products 52–64 could be separated by column chromatography in pure form. Next, the C7 substituent R¹ was introduced by Buchwald–Hartwig or Suzuki cross coupling reactions to afford intermediates 65–77. Final deprotection of the phosphonate function afforded the final analogs 1–30, which were typically purified by HPLC prior to biochemical or microbiological assessments.

3. Synthesis of Final Analogs 1–30 from Building Blocks 39–40 and 48–49. Conditions: (a) R⁴–X, K₂CO₃, DMF or R⁴–X, NaH, THF. (b) R¹–B­(OH)₂, PdppfCl₂, Cs₂CO₃, Dioxane-H₂O, Reflux or R₂NH (R¹), Pd₂dba₃, BINAP, Cs₂CO₃, Toluene, Reflux. (c) 4M HCl/Dioxane, 100 °C.

Conclusions

Here, we describe a structure-based effort to define the determinants of binding for quinoline-2­(1H)-one phosphonates to representative serine and metallo-carbapenemases. Every ring position of the heterocyclic core was explored, with early SAR and synthetic accessibility leading to a focus on substitution at C3, C6, C7, and N1 of the quinoline-2­(1H)-one core. The solution of KPC-2 complex crystal structures of several analogs identified two distinct binding modes that could be explained (and predicted) based on ring substituents. The potential for this scaffold to exhibit degenerate binding modes could be advantageous in the context of a resistance mutation that might affect one but not both possible binding modes. Inhibitor binding to the metallo enzyme NDM-1 was driven largely by interactions of the phosphonate with the metal centers and by contact with hydrophobic residues of the L3 loop, especially Phe70. Finally, we used ITC and MD simulations to define the biochemically and pharmacologically relevant binding event and identify ligand stacking in the active site as a likely artifact of ligand soaking conditions. Overall, this study provides new structural insight into the targeting of mechanistically distinct carbapenemases while also identifying limitations of the current molecules as drug leads for drug-resistant Gram-negative infections.

Experimental Section

Protein Expression and Purification

KPC-2 and NDM-1 were both cloned, expressed, and purified as previously described. Briefly, the KPC-2 gene was cloned into a pET-GST vector containing a His-tag that was transformed into Escherichia coli BL21 cells. Transformed cells were grown overnight at 37 °C in LB medium supplemented with 50 μg/mL kanamycin. A 10 mL aliquot was used to inoculate 1L of LB with kanamycin, and cells were grown at 37 °C to an OD600 of 0.6, followed by induction with 0.25 mM IPTG. Cells were harvested by centrifugation, resuspended in lysis buffer (20 mM Tris–HCl pH 8.0, 300 mM NaCl, and 20 mM imidazole), and then lysed by sonication. The lysate was clarified by centrifugation at 45,000 rpm, and the supernatant was loaded into HisTrap HP affinity column (Cytiva). The protein was eluted using a 20–500 mM imidazole gradient, pooled, and subjected to size-exclusion chromatography on a Superdex75 16/60 column (Cytiva) in 20 mM Tris–HCl (pH 8.0) and 300 mM NaCl. Purified protein was concentrated to 15 mg/mL, flash-frozen in liquid nitrogen, and stored at −80 °C.

For NDM-1, overnight cultures grown in LB media with 50 μg/mL kanamycin were diluted 1:100 into 2X YT medium containing kanamycin. Cells were grown at 37 °C to an OD600 of 0.6–0.8, and protein expression was induced with 0.25 mM IPTG and 100 μM ZnSO₄, followed by overnight incubation at 20 °C. Cells were pelleted, resuspended in lysis buffer (20 mM HEPES pH 7.4, 500 mM NaCl, 10% glycerol, 20 mM imidazole, 10 mM β-mercaptoethanol, 50 μM ZnSO₄), sonicated, and clarified by centrifugation. The supernatant was loaded onto a HisTrap HP column (Cytiva) and eluted with a 20 mM-500 mM imidazole gradient. Eluted protein was pooled, concentrated, and cleaved by thrombin overnight. The cleaved mixture was run through the HisTrap column, and the flow-through containing cleaved protein was collected and concentrated. Protein was then loaded onto HiLoad 16/600 Superdex 75 pg column (Cytiva) equilibrated in 20 mM HEPES (pH 8.0), 100 mM NaCl, and 50 μM ZnSO₄. Protein was concentrated to 22 mg/mL for crystallography and assay studies.

Enzyme Kinetics

KPC-2 assays were performed in 100 mM Tris–HCl (pH 7.0), 200 mM NaCl, and 0.001% (v/v) Triton X-100, using 0.5 nM enzyme and 20 μM nitrocefin as the chromogenic substrate. NDM-1 assays were conducted in 50 mM HEPES (pH 7.2), 50 μM ZnSO₄, 1 μg/mL BSA, and 0.001% (v/v) Triton X-100. Reactions included 3 nM NDM-1 enzyme and 10 μM nitrocefin, with compounds preincubated in reaction buffer for 10 min prior to enzyme addition. DMSO concentrations were kept consistent across all of the samples.

Inhibitor potency was assessed using a serial dilution of compounds starting at 200 μM. All reactions were initiated by the addition of the enzyme and monitored at 486 nM using a Biotek Cytation multimode plate reader at 37 °C. Initial rates of nitrocefin hydrolysis were determined from duplicate measurements. Kinetic parameters V _max and K _m of nitrocefin were determined via nonlinear regression fitting to the Michaelis–Menten equation using SigmaPlot 12.5. IC₅₀ values were derived from sigmoidal concentration dependence curves and K _i values were calculated using the Cheng–Prusoff equation: K _i = IC₅₀/(1 + [S]/K _m), where [S] is the concentration of nitrocefin, and K_m is the Michaelis constant of each enzyme.

Crystallization

KPC-2 crystallization was performed as previously described. KPC-2 crystals were soaked in 1.44 M sodium citrate and 10 mM inhibitor for 1 h and then cryoprotected with the same solution supplemented with 20% glycerol before flash-cooling in liquid nitrogen. NDM-1 (10–20 mg/mL) was crystallized using a hanging-drop vapor diffusion method at 20 °C, mixing protein 1:1 with well solution (0.1 M Bis-Tris, pH 7.5, and 25% PEG 3350). Crystals formed within 1 week and were further optimized by microseeding (2:2:0.5 protein: well: seeds), yielding crystals within 2 days. These crystals were then soaked with well solution containing 10 mM inhibitor for 30 min to 1 h. These crystals were cryoprotected with a well solution, 25% glycerol, and 10 mM inhibitor before flash-freezing.

Data Collection and Structure Determination

X-ray diffraction data (Table S2) were collected at the Southeast Regional Collaborative Access Team (SER-CAT) 22-ID and 22-BM beamline and at the Structural Biology Center (SBC) 19-ID beamline at the Advanced Photon Source (APS), Argonne National Laboratory, IL. Data were processed using HKL2000 and the CCP4 suite. Molecular replacement was performed with PHASER using previously solved structures of KPC-2 (PDB ID: 6D19) and NDM-1 (PDB ID: 6D1A) as search models. Model building and refinement were conducted by using the CCP4 suite, Coot, and the PDB REDO server (pdb-redo.eu). The PyMOL Molecular Graphics System (Schrödinger, LLC) was used to generate all images used.

Isothermal Titration Calorimetry

All ITC titrations were performed on an Affinity ITC Low Volume instrument (TA Instruments, WatersTM). Reactions were performed at 25 °C in 50 mM HEPES (pH 7.2), 0.001% v/v Triton X-100, 1 μg/mL BSA, and 500 μM of ZnSO₄. Titrations were performed with 30 μM of NDM-1 and 200 μM compound 13 in the injection syringe using 30 injections of 3 μL each at 200 s intervals with a 300 s initial baseline and 200 s postinjection baseline. Data were analyzed using the NanoAnalyze software v4.1.0 (TA Instruments, WatersTM). Thermograms were integrated and normalized binding enthalpies fitted to an independent binding model (4-variable nonlinear least-squares fit).

Checkerboard Assays

Checkerboard assays were carried out as previously described. Briefly, the clinical K. pneumoniae KPC-2 strain was provided by Dr. Yohei Doi (University of Pittsburgh) and the K. pneumoniae NDM-1 strain was purchased from ATCC (BAA-2146). The P. aeruginosa KPC-2 (PAM4135 pUCP24:KPC) and NDM-1 (PAM4179 pUCP24:NDM1) strains were kind gifts from Dr. Olga Lomovskaya of QPex Biopharma Inc. All bacterial strains were stored in glycerol stocks at −80 °C and struck onto tryptic soy agar or luria broth agar plates supplemented with 15 μg/mL gentamicin for the K. pneumoniae and P. aeruginosa strains, respectively, and grown for 18 h at 37 °C. Single colonies were then picked from each plate and used to inoculate 5 mL of MHBII and grown overnight (18 h, 250 rpm, at 37 °C). Overnight cultures were then diluted 1:1000 in MHBII and used for all assays in a 96-well plate format. For each assay, imipenem was tested at a concentration from 256 to 1 μg/mL (using 2-fold dilutions) and avibactam was tested at a concentration from 128 to 8 μg/mL (using 2-fold dilutions). Initially, each BLI compound was tested from a concentration of 256–8 μg/mL also using 2-fold dilutions; however, the BLI compounds were retested at 128 μg/mL. All 96-well plates were incubated for 20 h at 37 °C with shaking (250 rpm). Growth was quantified by reading the optical density (OD600) of each well by using a Cytation5 plate reader (BioTek). Wells with the same optical density as wells filled with medium only were used to determine growth inhibition.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c12676.

• Supplementary tables of SAR, individual plasma sample data for PK study of 13, table of crystallographic data collection and refinement statistics, supplementary figures, including MD simulation data and plasma exposure profile of 13, supplementary experimental methods and synthetic procedures, synthetic procedures and characterization data, and crystallographic data collection and refinement statistics (PDF)

L.M.C.J. and P.J. contributed equally. The manuscript was written through contributions of all authors.

This work was supported by National Institutes of Health Grant R01AI161762 to Y.C. and A.R.R.

The authors declare the following competing financial interest(s): P.J., Y.C., and A.R.R. are listed as inventors on a patent application describing compounds studied herein.