Arginine Modulates Carbapenem Deactivation by OXA-24/40 in Acinetobacter baumannii

Abstract

The resistance of Gram-negative bacteria to β-lactam antibiotics stems mainly from β-lactamase proteins that hydrolytically deactivate the β-lactams. Of particular concern are the β-lactamases that can deactivate a class of β-lactams known as carbapenems. Carbapenems are among the few anti-infectives that can treat multi-drug resistant bacterial infections. Revealing the mechanisms of their deactivation by β-lactamases is a necessary step for preserving their therapeutic value. Here, we present NMR investigations of OXA-24/40, a carbapenem-hydrolyzing Class D β-lactamase (CHDL) expressed in the gram-negative pathogen, Acinetobacter baumannii. Using rapid data acquisition methods, we were able to study the “real-time” deactivation of the carbapenem known as doripenem by OXA-24/40. Our results indicate that OXA-24/40 has two deactivation mechanisms: canonical hydrolytic cleavage, and a distinct mechanism that produces a β-lactone product that has weak affinity for the OXA-24/40 active site. The mechanisms issue from distinct active site environments poised either for hydrolysis or β-lactone formation. Mutagenesis reveals that R261, a conserved active site arginine, stabilizes the active site environment enabling β-lactone formation. Our results have implications not only for OXA-24/40, but the larger family of CHDLs now challenging clinical settings on a global scale.

Introduction

β-lactam antibiotics reaching their bacterial targets become potential substrates of β-lactamases, bacterial enzymes that hydrolytically deactivate the drugs. The β-lactamases are a vast family of proteins consisting of four sequence-based classes, A, B, C, and D,¹ three of which (A, C, and D) collectively define the serine β-lactamases. The serine β-lactamases hydrolytically deactivate β-lactams via a shared mechanism that includes three well-known steps: i) entry of the β-lactam into the protein active site producing the non-covalent complex (E•S); ii) opening and covalent attachment to the β-lactam ring by an active site serine, forming the acyl-enzyme complex (E—S); iii) nucleophilic attack at the acyl carbonyl by an active site water molecule to release the cleaved and deactivated β-lactam. This hydrolytic mechanism, now over 30 years old,² guides contemporary research.

However, richer mechanistic behavior has become apparent for carbapenem-hydrolyzing Class D β-lactamases (CHDLs). Carbapenems are a subset of β-lactams with exceptional potency and broad-spectrum activity that have made them “drugs of last resort”.³ Their increasing susceptibility to hydrolytic destruction by CHDLs and other β-lactamases poses a serious challenge to public health. A recent study of CHDLs including OXA-48, OXA-10, and OXA-23⁴ revealed an additional deactivation mechanism involving conversion of the carbapenem substrate to an inactive, β-lactone product. The study also indicated micromolar binding affinity of the final β-lactone product for the active site pocket. Initially, the new “β-lactone mechanism” appeared specific for the later generation carbapenems such as ertapenem, meropenem, and doripenem that contain a 1β-methyl moiety in the pyrroline ring. More recently, Aertker et al.⁵ has shown how a single-site mutation in OXA-48 can enable β-lactone deactivation of imipenem, a non-1β-methylated carbapenem.

The CHDL features enabling the β-lactone mechanism remain largely unknown. Current hypotheses are based on modeling studies of extant crystal structures,⁴ rather than direct experimentation. In particular, the active site features supporting the β-lactone mechanism and their relationship to those supporting conventional hydrolysis remain obscure.

We have addressed this open issue via NMR studies of OXA-24/40, from the Gram-negative pathogen A. baumannii. OXA-24/40 is a well-documented CHDL,⁶ and is therefore related to those examined in the inaugural study revealing the β-lactone mechanism.⁴

Our NMR studies are part of our larger effort to understand the abilities of clinical point mutations to alter CHDL activity. This has involved comparing the activities of the parent OXA-24/40 sequence (designated WT hereafter) with those of variants toward carbapenem substrates. The versatility of NMR lets us study substrate activity from both the ligand and protein perspectives. Notably, the protein studies include rapid data collection techniques to record 2D ¹⁵N-¹H spectra of active OXA-24/40 during substrate turn-over. These spectra provide a site-resolved, “real time” view of OXA-24/40 activity that is, to our knowledge, the first for any serine β-lactamase.

Our main results are comparisons of two OXA-24/40 mutants versus WT in terms of their activities toward the same carbapenem substrate, doripenem.⁷ The two mutants, R261S and R261K, are substitutions of a conserved arginine that makes electrostatic contact with bound doripenem in crystal structures of deacylation-deficient OXA-24/40 variants.⁸

Our results have revealed new aspects of the WT active site that support the new β-lactone mechanism. Briefly summarized, we found that WT OXA-24/40 displays both conventional hydrolysis and β-lactone mechanisms, with the β-lactone mechanism dominating. We linked the coexistence of these two mechanisms to the WT active site being able to adopt two mechanism-specific states: one leading to β-lactam cleavage via conventional hydrolysis, another leading to β-lactone formation. The contribution of the β-lactone mechanism diminished for R261K and R261S. In fact, it nearly vanished for R261S, which deactivated doripenem mainly by conventional hydrolysis. The switch in mechanism caused by R261S indicated that the β-lactone mechanism dominating WT relies on R261-mediated interactions. In support of this, backbone ¹⁵N relaxation measurements and MD simulations of apo state revealed perturbed dynamics of residues known to interact with R261 and/or bound doripenem, as a result of the R261S substitution. These perturbations suggest changes to the range of bound state conformations accessible from the apo state, and thus, the range of mechanism-specific acyl-enzyme configurations.

Our findings for OXA-24/40 suggest a broader theme applicable to CHDL active sites: the range of conformations accessible to the acylated antibiotic bears directly on the diversity of possible deactivation mechanisms. Mutations enriching the former could also enrich the latter.

Results

Description of the OXA-24/40 active site and doripenem

Before presenting our results, we first describe the main structural features of OXA-24/40 and its carbapenem substrate, doripenem. Extant X-ray crystal structures include the apo enzyme,⁹ and two acyl-enzyme complexes of deacylation-deficient OXA-24/40 mutants (K84D, V130D) acylated by doripenem.⁸ Figure 1A shows the crystal structure of the V130D/doripenem complex. The active site is defined by the junction of three conserved motifs (magenta shading), 81-STFK-84, 128-SAV-130, and 218-KSG-220. These motifs, which are often referred to as the “active site elements”, are defining features of the Class D β-lactamases.^10,11 S81 is the active site serine acylated by the β-lactam via an ester linkage. The acyl carbonyl within the ester linkage can hydrogen bond to the oxyanion hole, defined by the backbone amides of S81 and W221.^8,12 The active site is partially shielded from solvent by three surface loops (cyan shading): the β5-β6 loop (residues G222-Q228), the Ω-loop (N153-T175) and the P-loop (K96-T120). These loop regions are conserved structural motifs among CHDLs.

Figure 1.

A: Structure of OXA-24/40 acylated with doripenem (pdb 3PAG) with α-helices labeled A – K and β-strands labeled 2 – 7. The following features are highlighted: active site residues (magenta spheres); P-loop, Ω-loop and, β5-β6 loops (cyan); R261 (magenta sticks); hydrolyzed doripenem (black sticks); interactions between the R261 side chain and doripenem (yellow dashed lines). B: Sequence alignment showing conservation of R261 among CHDLs (red), chromosomal OXA-66 (blue), a Class D non-carbapenemases (black), and the Class A TEM-1 (green). C: The structure of intact doripenem with carbons labeled.

To study OXA-24/40 activity, we used an established carbapenem substrate, doripenem (Figure 1C). Like the other advanced carbapenems (ertapenem, meropenem), the pyrroline ring of doripenem has a 1β-methyl group (C11) that protects it from degradation by human renal dehydropeptidase-I.¹³ The four-membered lactam ring opens upon covalent binding to S81. Bound (acylated) doripenem is stabilized by hydrophobic contacts between the 6α-hydroxyethyl moiety common to all carbapenems, and the methyls of L168 and V130. Further stabilization comes from a salt-bridge between its C3-carboxylate and the side chain of R261. R261 corresponds to an active site arginine that is highly conserved among CHDLs (Figure 1B) and some other non-carbapenemase SBLs, such as OXA-10 (Class D) and TEM-1 (Class-A).^3,8

OXA-24/40 reveals hydrolysis and β-lactone formation

Steady-state kinetics parameters for OXA-24/40 toward doripenem were determined previously via UV absorption by Leonard and co-workers.¹⁴ The results (K_M = 0.024 ± 0.001 μM, and k_cat = 0.074 ± 0.001 s⁻¹) were determined using a competition assay, with doripenem acting as an inhibitor of ampicillin hydrolysis. Here, we wanted estimates of catalytic activity under the same sample conditions used for the accompanying 2D ¹⁵N-¹H NMR studies of OXA-24/40. We therefore assayed OXA-24/40 activity toward doripenem by collecting ¹H NMR spectra of doripenem during turnover. An additional feature of the ¹H NMR assay turned out to be its ability to detect all molecular species affected by turnover, and not just those with a particular chromophore monitored by UV–vis (vide infra).

Our NMR assay used a total doripenem (substrate) concentration, S_T = 2.5 mM, and a total OXA-24/40 (enzyme) concentration, E_T = 50 μM. The condition of S_T ≫ E_T made the early portion of the assay amenable to a steady-state analysis.¹⁵ S_T was also ≫ K_M, a condition promoting zero-order kinetics, manifesting as linear changes of the substrate and product concentrations.¹⁶

The ¹H NMR assay involved adding fresh doripenem to an NMR tube containing a pre-equilibrated solution of OXA-24/40, followed by uninterrupted collection of 1D ¹H NMR data. The 1D pulse scheme included a relaxation-filtering scheme for selective observation of free doripenem (substrate and product forms). Each 1D dataset required 0.65 minutes of the acquisition time. We collected a total of 400 1D datasets, with the n^th 1D representing t[n] = 4 + 0.65*[n] minutes after substrate addition (n going from 1 to 400). The starting four-minute offset accounts for inserting the sample into the magnet, locking, and shimming, before actual data collection. The end result was 400 1D spectra, displaying ¹H peaks of doripenem (substrate) and its products as function of time. Further assay details are in the Materials and Methods subsection.

Our analysis of the ¹H NMR assay data focused on the methyl proton peaks, which gave the clearest depiction of substrate turnover. These included the C9 methyl in the 6α-hydroxyethyl moiety, and the C11 1β-methyl the pyrroline ring (Figure 1C).

Originally, we intended these ¹H NMR assays of WT activity to serve as a baseline for subsequent ¹H NMR assays of clinical OXA-24/40 variants showing expanded hydrolytic activity. Guided by the conventional hydrolysis mechanism, we expected to see up to two ¹H peaks per methyl, corresponding to the free substrate and product forms of doripenem.

Instead, we observed four ¹H resonances per methyl, representing two doripenem products: i) the cleaved product expected from standard hydrolysis; ii) and a β-lactone product recently revealed for several CHDLs.⁴ Figure 2 shows these products (Figure 2A) and their ¹H methyl resonances (Figure 2B). For example, for the C9 methyl, the four resonances include the intact substrate at 1.28 ppm (blue label “I”), the hydrolyzed product at 1.25 ppm (black “II”), and two β-lactone resonances at 1.60 ppm (red “III”) and 1.63 ppm (green “IV”), corresponding to the 2R and 2S β-lactones, respectively. The 2R form appears first, subsequently converting to the more stable 2S-lactone; this conversion echoes that reported in the first study revealing the β-lactone mechanism in other CHDLs (OXA-48, OXA-10, and OXA-23).⁴ We note that the first ¹H spectrum showed a small amount of 2R and 2S product, with peak heights ~ 0.04 and 0.01 that of the substrate. These peaks can be explained by OXA-24/40 activity during the initial four-minute delay described above.

Figure 2.

A: Structures of (I) intact doripenem, (II) hydrolyzed doripenem (after tautomerization), (III) 2R-Lactone, and (IV) 2S-Lactone. B: ¹H NMR spectra following the time-dependence of free substrate and products in the presence of WT OXA-24/40. The minute values refer to the time elapsed from the start of NMR data acquisition, and do not include the initial 4 minutes for sample insertion, locking, and shimming. C: Peak integrals versus time for the C9-methyl protons of: doripenem substrate (blue), hydrolyzed doripenem (black), 2R β-lactone intermediate (red), and 2S β-lactone end product (green).

For quantitative analysis using peak integrals, we chose the C9 methyl peaks as they were essentially free from overlap with other proton resonances. We normalized the raw integral values by dividing them by the value measured for the substrate peak from the 1st 1D spectrum, resulting in the four data files of normalized peak integral versus time, plotted in the top panel of Figure 2C. The distinctive surgelike shape of the red 2R time trace reflects the 2R-to-2S conversion.

The results of the WT ¹H NMR assay made several points. First, WT OXA-24/40 represented another CHDL capable of both hydrolysis and the β-lactone mechanism, strengthening the notion that the capacity for the two mechanisms reflects features characteristic of CHDL active sites. Second, the serendipitous separation of methyl ¹H resonances from hydrolysis versus β-lactone formation gave us an opportunity to study the mechanism-specific effects of the R261K/S mutations. Third, the ability of NMR to track chemical shifts enabled it to detect product species that might otherwise be missed using methods tracking absorption of a specific chromophore.

Sensitivity of the deactivation mechanism to active site R261

To explore the origins of the β-lactone mechanism, we investigated its response to OXA-24/40 mutants. We focused on substitution mutations of an active site arginine, R261, based on the salt-bridge between the R261 side chain and the C3-carboxylate of doripenem depicted in X-ray crystal structures.⁸ The salt-bridge occurs away from the active site serine (S81) and doripenem, suggesting that an R261 substitution would more likely perturb, rather than destroy carbapenemase activity. Also, R261 corresponds to a highly conserved arginine in the active sites of CHDLs, and also narrow spectrum Class A β-lactamases such as SHV-1 and TEM-1 (Figure 1B). Thus, knowledge from the R261 mutants here could benefit functional studies of other SBLs.

We generated two substitution mutants, R261S and R261K, for NMR study alongside WT (Materials and Methods). Our S and K substitutions were inspired by early work on TEM-1 (Class-A SBL) that defined the mechanistic role of active site arginine R244 via S and K substitutions.^17,18 Here, the purpose of our primary substitution mutant, R261S, was to explore the consequences of removing the aforementioned salt-bridge to doripenem. Of course, the R261S substitution makes considerable changes in side chain bulk and charge. We therefore generated a secondary mutant, R261K, to define the effects of an ostensibly conservative point mutation. Both R261S and R261K preserved the overall fold of WT, as indicated by their 2D ¹⁵N-¹H TROSY HSQCs (Figures S1, S2, Supplementary Material). Their T_m values measured from Differential Scanning Fluorimetry (DSF) (Figure S3) indicated a ~ 2 °C increase compared to WT, suggesting increased thermal stability. This increase is consistent with previous OXA-24/40 studies, which describe stably-folded mutants of OXA-24/40 with both increases and decreases of T_m within a range of ±3 °C.¹⁹

We then measured the activities of the R261S/K mutants toward doripenem, using the same ¹H relaxation-filtered NMR assay applied to WT. The results for all three constructs, WT, R261K, and R261S, are shown in Figure 2C, which displays time traces of normalized integral values measured from the substrate and product C9 ¹H methyl peaks. The initial portions of the time-traces showed linear changes, consistent with the steady-state, zero-order kinetics promoted by our initial substrate concentration S_T satisfying S_T ≫ E_T, K_M (vide supra). We fit these linear regions (the first 15 data points) to linear functions, I(t) = m*t + b, thereby obtaining estimates of three catalytic rate constants, including: k_S (depletion of substrate “S”); k_H (formation of the conventional hydrolyzed product “P”); and k_L (formation of the initial β-lactone product, “2R”). The rate constants are collected in Table 1; further details concerning rate constant calculations are in the Materials and Methods.

Table 1.

Activity of three OXA-24/40 constructs toward doripenem, using the ¹H NMR assay, S₀ = 2.5 mM doripenem and 50 μM protein, T(nom) = 295 K, 16.4 T. The rate constants were determined from the first fifteen peak integrals of the ¹H methyl resonances of C9 in 6α-hydroxymethyl of doripenem (substrate and product). The rate constants include: k_S substrate consumption; k_H, formation of hydrolyzed product; k_L formation of the initial β-lactone product, 2R.

	kS × 10−2 s−1	kH × 10−2 s−1	kL × 10−2 s−1	kH/kL
WT	5.65 (±0.03)	1.07 (±0.47)	3.83 (±0.25)	0.28
R261K	1.22 (±0.07)	0.80 (±0.1)	0.40 (±0.01)	2.0
R261S	4.91 (±0.40)	3.29 (±0.3)	0.66 (±0.03)	4.98

The estimated rate constants and Figure 2C both show that the mutant activities differ significantly from WT. Remarkably, WT OXA-24/40 showed the largest k_L value, and was the most ardent practitioner of the β-lactone mechanism. By contrast, R261S showed the largest k_H value, and therefore behaved the most like a conventional SBL. R261K languished behind both, showing a greater loss of β-lactone production compared to WT.

To compare the mechanism preferences among OXA-24/40 constructs, we used the ratio of rate constants, k_H/k_L. Per Table 1, the k_H/k_L ratios followed the rank-ordering WT < R261K < R261S. Thus, WT had the lowest preference for hydrolysis and the highest preference for the β-lactone mechanism, while R261S showed the opposite preferences. In effect, the R261S substitution somehow altered the WT active site to establish a nearly complete preference for hydrolysis.

The red and green traces of Figure 2C show “2R” to “2S” conversion of the β-lactone product, a phenomenon first described by Lohans et al.⁴ in their work on related CHDLs. To estimate a rate constant for 2R to 2S conversion, k_R, we used portions of the 2R (red) traces after the “cusp; the negative slopes of these portions indicated greater influence of 2R depletion (due to 2R→2S) over 2R formation from OXA-24/40. We fit the last 100 data points (200 to 260 minutes) of the 2R traces from WT and R261K to a mono-exponential decay, C₀*exp{−k_RS t}, and found k_RS = 0.010 ± 0.007 × 10⁻² s⁻¹. This is significantly smaller than the estimated k_L values for 2R formation (0.40 ± 0.01 for R261K, 4.84 ± 0.25 × 10⁻² s⁻¹ for WT), and suggests that the time scale for accruing substantial amounts of 2S is longer than that for substrate turnover.

2D NMR monitoring of active OXA-24/40 during turnover

To explore the basis for the R261S-induced change in mechanism preference, we turned to NMR studies of the protein. In particular, we wanted an experimental analog of the ¹H relaxation-filtered 1D NMR assays (Figure 2) for the protein, that would give a site-resolved description of active OXA-24/40 during turnover of substrate (doripenem).

We achieved this using the BEST-TROSY methods, developed by Brutscher and co-workers.²⁰ These methods allowed us to record multiple 2D ¹⁵N-¹H spectra of active OXA-24/40 before exhaustion of the substrate. Figure 3 shows a typical 2D BEST-TROSY spectrum for WT OXA-24/40. A single 2D BEST TROSY required 10 minutes and 40 seconds. Amide ¹H resonances < 8.0 ppm (upfield) were not excited due to the bandwidth of our frequency-selective ¹H pulses in the BEST-TROSY pulse-scheme.

Figure 3. Overlay of apo WT ¹⁵N-¹H BEST-TROSY on ¹⁵N-¹H TROSY.

Left: The ¹⁵N-¹H BEST-TROSY (red) compared to the ¹⁵N-¹H TROSY (black) of apo wild-type OXA-24/40. The 30 NH cross-peaks tracked during the BEST-TROSY series are labeled. Right: Locations of the 30 tracked NH cross-peaks on pdb 3PAG. Red spheres correspond to backbone amides. The two cyan spheres are the εNH of W167 and W221.

The 2D BEST TROSY spectra presented an opportunity to identify residues involved in turnover by chemical shift perturbations (CSPs) of the NH cross-peaks, relative to apo state. To our knowledge, this is the first such time-resolved observation of an active serine β-lactamase. The time-dependent view offered here complements the high-resolution structures from X-ray diffraction.

Continuous recording of 2D ¹⁵N-¹H BEST-TROSY spectra started immediately after adding fresh doripenem to a pre-equilibrated solution of protein. The total doripenem concentration, S_T = 14 mM, was 100 times that of the total OXA-24/40 concentration, E_T = 140 μM, and also much greater than the previously reported Michaelis constant (K_M = 0.024 ± 0.001 μM¹⁴). These concentrations served to promote substrate saturation of the OXA-24/40 active site, thereby ensuring that the observed NH cross-peaks in the 2D spectra represented the enzyme-substrate complex. BEST-TROSY data collection proceeded uninterrupted for 5.5 h, producing a series of 30 2D ¹⁵N-¹H spectra for each OXA-24/40 construct.

Doripenem-induced CSPs

We examined the BEST-TROSY series for cross-peaks showing perturbations relative to the apo state. After excluding cross-peaks suffering resonance overlap or poor signal-to-noise, we had a set of 30 NH cross-peaks for analysis. Figure 3 shows these cross-peaks in a typical WT BEST-TROSY spectrum, and their corresponding locations in the OXA-24/40 active site.

The distinct information from the BEST-TROSY series included time-dependent NH CSPs, or “CSP trajectories”, instigated by doripenem. Those trajectories showed revealing differences among the three constructs. Figure 4 shows a prominent example: the cross-peak of G220, which is part of the conserved “KSG” active site element. The crystal structures of doripenem-acylated OXA-24/40⁸ show the G220 NH close to the acyl carbonyl of the ester bond linking doripenem to S81. Figure 4 shows that G220 has distinct chemical shift “trajectories” for R261S versus WT. The R261S trajectory makes a cycle, consistent with standard expectations that the free (apo) enzyme should be regenerated after exhaustion of the intact substrate. By contrast, the WT trajectory fails to return to its origin, indicating some kind of irreversible effect (vide infra). In the case of R261K, the G220 cross-peak broadened out, and so did not reveal a trackable trajectory. These differing trajectories indicate significant differences in the G220 NH microenvironments among the WT versus mutant constructs. Such differences likely include mutation-induced perturbations of protein-substrate interactions, inter-residue interactions within the active site, or both.

Figure 4. Time-dependent CSP trajectories for G220.

Selected time points for G220 taken from the ¹⁵N-¹H BEST-TROSY series for real-time monitoring the active enzyme during doripenem turnover. G220 is part of the conserved “KSG” active site element. The CSP trajectories for WT (left) and R261S (right) are clearly distinct. Coloring indicates time points after addition of doripenem as follows: apo (black), 15.5 min (red), 110 min (orange), 215 min (green), 320 min (blue), 24 h (WT purple), and 7 h (R261S purple).

The presence of such perturbations is supported by Figure 5, which shows the most prominent CSPs from the very first 2D BEST-TROSY spectrum. Complete data collection for this first spectrum occurred ~ 15 minutes after adding doripenem; those 15 minutes included sample insertion, establishing the deuterium lock, coarse shimming, and then the actual 2D data collection (10 min 40 seconds). Figure 5 shows that the largest CSPs occur at active site residues proximal to acylated doripenem, for all three constructs. Figure 5 also reveals CSPs outside the active site (marked with red stars) that vary in magnitude for each construct. These remote CSPs are at residues corresponding to apparent “hinge” regions of Ω and P-loops. They may be some of the residues that provide non-equivalent protein-substrate interactions that distinguish the β-lactone mechanism from conventional hydrolysis. A notable difference among the constructs is the site of the largest CSP. In WT, the largest CSP is at G220, whereas in R261S, the largest CSP shifts two positions down to G222. This suggests non-equivalent modes of doripenem recognition in R261S versus WT, consistent with different mechanisms for doripenem deactivation.

Figure 5. NH CSPs induced by doripenem within the first 15 minutes.

Bar chart indicating NH CSPs induced by doripenem within the first 15 minutes for WT (purple), R261K (blue), and R261S (green). The locations of the WT CSPs (30 tracked residues) are the purple shaded spheres in the pdb structure. The shading intensity scales with the CSP magnitudes shown in the bar chart. White indicates an absence of data rather than an absence of CSP. Red stars mark residues outside the active site pocket at apparent “hinge” points for the P and Ω loops.

Distinct acyl-enzyme complexes

Some NH cross-peaks in the BEST-TROSY series not only shifted (showed CSPs), but also “split” in response to doripenem (examples in Figure S4). To be clear, these were not splittings due to scalar coupling. Rather, single cross-peaks observed for the apo state “split” into two cross-peaks upon addition of doripenem, indicating the onset of distinct NH microenvironments in slow exchange for the acyl-enzyme state. The “splitting” cross-peaks localized mostly to the active site, and included W221, G222 and G224 within the β5-β6 loop, L122 (SAV), W235 (β6), E254 and G258 (β7-h11 turn near R261). The active site residues showing splitting are known to be proximal to acylated doripenem.⁸ A few remote sites G67 (β2-h2 loop) and G91 (h3) also showed splitting.

The split peaks appeared to be spectral signatures of mechanism-specific substates of the active site. Evidence supporting this claim came from W221, an active site tryptophan next to the “KSG” motif that is also highly conserved among CHDLs. The backbone NH of W221, along with that of S81, forms the oxyanion hole stabilizing the acyl carbonyl.⁶ While the W221 backbone NH cross-peak was broadened out by exchange, its side chain indole εNH could be analyzed (Figure 6). W221 εNH shows only one cross-peak in the WT apo state. It then shifts and splits into two cross-peaks labeled “H” and “L”, upon addition of doripenem. Taking that the ratio of the “H” and “L” cross-peak intensities, I_H/I_L, as a proxy for the ratio of “H” and “L” substate populations, we tracked the I_H/I_L, time-dependence for all three constructs. For WT OXA-24/40, the “L” cross-peak dominates, as the I_H/I_L ratio stays < 1.0 over the 2D series, reaching a maximum of 0.9 at ~ 50 min. By contrast, for R261S, the “H” cross-peak dominates, as I_H/I_L reaches ~ 4 at 50 minutes, and a maximum of ~ 12 at 120 minutes. For R261K, we saw an intermediate response, with I_H/I_L increasing to a maximum of ~ 3 at 50 minutes. Remarkably, the maximum value of I_H/I_L across the three constructs increases as WT < R261K < R261S, the same rank ordering observed for the rate constants k_H/k_L from the ¹H CPMG NMR assay (Table 1). That is, the preference for the “H” cross-peak over the “L” cross-peak followed the mechanistic preference for hydrolysis over β-lactone formation, suggesting the mechanisms favored different subenvironments for W221 εNH, represented by the two “H” and “L” cross-peaks.

Figure 6. Time-dependence of tryptophan side chain εNH CSPs.

Overlay of W221 and W167 side-chain εNH cross-peaks for WT (left), R261K (middle), and R261S (right). Cross-peak coloring: apo (black), 10.5 minutes (red), 89 minutes (blue) after doripenem addition. Doripenem causes the W221 εNH cross-peak to split into two subpopulations labeled H and L; the time-dependence of the ratio of cross-peak heights, I_H/I_L graphed below. Other residues showing doripenem-induced splitting of NH peaks are indicated by the color spheres on pdb 3PAG: WT (red), R261K (blue), and R261S (green).

In summary, the spectra show that doripenem turnover by WT versus R261S amplifies different members of the “L/H” cross-peak pair. We also found that WT activity is dominated by the β-lactone mechanism, whereas R261S activity is almost entirely hydrolysis. Together, these observations suggest that the “L” and “H” cross-peaks represent non-equivalent active site environments: the “L” cross-peak representing an environment favoring the β-lactone formation, and the “H” cross-peak representing an environment favoring hydrolysis.

Binding of the end product from the β-lactone mechanism

After complete exhaustion of the doripenem substrate, we expected regeneration of the free OXA-24/40 enzyme. To verify this, we waited at least 7 h after cessation of enzyme activity, and then collected standard 2D ¹⁵N-¹H TROSY spectra of the very same enzyme/doripenem samples used to record the BEST-TROSY series. We expected the spectra to match the apo state spectra. But this was not so. The WT and R261K spectra, recorded 24 h after addition of doripenem, retained numerous doripenem-induced CSPs within the active site. A pointed example is V130 in the conserved ‘SAV’ motif contacting the 6α-hydroxyethyl moiety, shown in Figure 7 (full spectra shown in Figure S5).

Figure 7. Wild-type and R261K do not return to apo state.

Overlay of 2D backbone ¹⁵N-¹H HSQC spectra for WT, R261K, and R261S, comparing the apo state (black), with spectra collected long after addition of doripenem (red). The same two zoom regions are shown for each mutant (see Figure S5 for full spectra). R261S nearly returns its apo state 7.5 h after doripenem addition. By contrast, WT and R261K retain CSPs 24 h after adding doripenem.

By contrast, the R261S spectra showed a nearly complete return to the apo state only 7.5 h after adding doripenem. The few remaining CSPs belonged to solvent-exposed residues that have not been implicated in substrate recognition nor activity. Considering the vast excess of hydrolyzed doripenem (~14 mM), these CSPs are likely the results of non-specific binding.

To explain the numerous persistent CSPs for WT and R261K, but not R261S, we considered reversible (non-covalent) binding of the final β-lactone product, described in the first study revealing the β-lactone mechanism in other CHDLs.⁴

To check for binding, we conducted Differential Scanning Fluorimetry (DSF) measurements to check for doripenem-induced shifts in T_m (melting temperature), indicative of complex formation. We performed measurements for all three constructs, using a fixed protein concentration of 5 μM in the presence of variable doripenem concentrations at 0, 5, 10, and 50 μM. Figure 8 shows the DSF results, with T_m defined as the point of inflection. The WT T_m shows a clear dependence on the doripenem concentration (Table 2); the R261K and R261S T_m values do not. These WT and R261S results are consistent with WT producing an abundance of β-lactone end product, and R261S producing almost none. The R261K results are more perplexing; the T_m appears independent of the doripenem concentration, suggesting no binding, while also having persistent CSPs like WT (Figure 7), in favor of β-lactone binding. Based on our ¹H relaxation-filtered NMR assay (Figure 2C, Table 1), R261K produces less β-lactone and more slowly than WT. The β-lactone produced by R261K under the DSF conditions could be insufficient for an observable T_m shift.

Figure 8. Doripenem binding induces a T_m shift in wild-type.

DSF thermal shift curves for WT OXA-24/40, R261K, and R261S after addition of doripenem: apo (blue), 5 μM (red), 10 μM (green), and 50 μM (purple).

Table 2.

Differential scanning fluorimetry: WT T_m (°C) versus doripenem concentration.

	0 μM	5 μM	10 μM	50 μM
WT Tm (°C)	48.66 ± 0.05	48.74 ± 0.05	48.82 ± 0.04	53.85 ± 0.04
R261K Tm (°C)	50.75 ± 0.05	50.43 ± 0.05	50.58 ± 0.05	50.22 ± 0.06
R261S Tm (°C)	50.81 ± 0.02	50.79 ± 0.02	50.76 ± 0.02	50.51 ± 0.02

To corroborate our DSF results, we used the reported preference of the β-lactone mechanism for 1β-methylated carbapenems (e.g. doripenem).⁴ Imipenem, an early carbapenem, lacks this 1β-methyl. Accordingly, we conducted DSF experiments for all three constructs, using imipenem in place of doripenem. We expected and observed a lack of T_m shifts from imipenem (Figure S6). This null result supports the interpretation of the doripenem-induced T_m shifts as being caused by β-lactone binding. Furthermore, mass spectrometry results provide additional evidence that the persistent CSPs are caused by a non-covalent interaction and not the enzyme being trapped in the covalent acyl-enzyme state (Figure S7).

R261S perturbations of backbone dynamics

The behavior of R261S indicates mutation-induced perturbations of the WT active site needed for the β-lactone mechanism. The perturbed features could include the intermolecular salt-bridge between the R261 side chain and C3 carboxylic group of doripenem.⁸ They could also include inter-residue contacts in the active site needed to stabilize direct contacts to doripenem made by residues other than R261.

Changes in inter-residue contacts from R261S are plausible, given the large size of the WT arginine side-chain and opportunities for polar contacts via its side chain. Moreover, changes in inter-residue contacts can manifest as changes in local protein dynamics.²¹ We investigated this possibility by comparing the backbone flexibility profiles of apo R261S and WT, as determined from backbone ¹⁵N relaxation measurements. In particular, we measured ¹⁵N R₂-(R₁/2) values for the backbone amide NH bonds. The ¹⁵N R₂-(R₁/2) is proportional to an effective correlation time describing the rapidity and magnitude of NH bond motion.^22,23 We have demonstrated the effectiveness of these measurements on OXA-24/40 to explain expanded substrate activity by clinical mutants.²⁴

To highlight site-specific NH bond motion, we analyzed the mutation-induced differences in the ¹⁵N R₂-(R₁/2) values normalized by their trimmed averages, denoted Δρ = ρ_R261S – ρ_WT. Figure 9 shows the results for 162 cross-peaks common to R261S and WT. Significant changes were those Δρ values with magnitudes ∣Δρ∣ ≥ 0.08 (4 times the rmsd).

Figure 9. Changes in apo NH backbone mobility due to R261S.

Left: Barchart of Δρ = (ρ_R261S – ρ_WT) versus sequence for the apo state. Blue and red bars highlight the significant decreases and increases (magnitudes > 0.08, 4 × rmsd). Right: The locations of the residues with blue/red bars on the left, as depicted in the crystal structure of the deacylation-deficient mutant, K84D-OXA-24/40 acylated by doripenem (pdb 3PAE). The blue and red spheres correspond to the red and blue bars, with shading proportional to the magnitude of ∣Δρ∣ = ∣ρ_R261S – ρ_WT∣. Acylated doripenem shown as green sticks.

Figure 9 reveals three residue clusters with mutation-induced changes in local flexibility. The first cluster, N257 and M252, are near the mutation site (R261S), and their side chains are close to that of R261 in crystal structures.⁶ Unfortunately, the cross-peak of the mutated residue S261 was obscured by resonance overlap.

Residues of the second cluster, W167 and L168, are in the Ω-loop. The negative Δρ indicates larger amplitude re-orientational motions of the NH bonds on the subnanosecond time scale in R261S. Notably, the L168 side chain makes hydrophobic contacts with the 6α-hydroxyethyl group of acylated doripenem.⁸ Conceivably, these increases in backbone mobility for the apo state could lead to reduced contact with doripenem in the acylated state.

Finally, the third cluster includes the β5-β6 residues G222, M223, V225, and Q228. This cluster includes residues with cross-peaks that “split” upon addition of doripenem (e.g. residues 220–224). The positive Δρ suggests the onset of ¹⁵N exchange dynamics on the millisecond time scale. Dynamic responses in this region are of clinical interest, as we have shown previously that point mutations increasing β5-β6 flexibility coincide with expanded substrate activity.²⁴

To summarize the ¹⁵N relaxation studies, the R261S substitution perturbs the apo state backbone NH flexibility of residues with the following features: i) sensitivity to doripenem binding; ii) cross-peak splitting indicating mechanism-specific active sites; iii) known to contribute to substrate recognition or turnover from previous biochemical studies and X-ray structures.

MD simulations of WT vs R261S

We also conducted all atom, explicit solvent molecular dynamics (MD) simulations to explore mutation-induced changes in protein flexibility. We used AMBER19²⁵ to simulate the apo states of WT and R261S at T = 300 K. The production runs were 100 ns, with snapshots saved every picosecond (Materials and Methods).

Our trajectory analysis focused on the side chain flexibilities of W221 and W167, the active site tryptophans highlighted by the NMR data. The large surface areas of the tryptophan indole rings suggest that changes in their mobility could require accommodating dynamic changes by neighboring amino acids. The net effect could alter the average size or shape of the active site pocket, and thus, its preferred mechanism for substrate turnover.

Figure 10 includes scatter plots comparing the side chain flexibilities of W221 and W167. Each dot is a pair of torsion angles χ₁ and χ₂ from a WT or R261S snapshot. The χ_1,2 dots for W167 suggest Gaussian fluctuations within one potential well, for both WT and R261S. R261S shows slightly broader sampling than WT, indicating larger amplitude fluctuations. The χ_1,2 dots for W221 show more complex motion, consisting of fluctuations within distinct potential wells, punctuated by rarer inter-well transitions. In particular, R261S expands the number of potential wells and the fluctuation amplitudes compared to WT. Thus, R261S enhances W221 side chain motion on both fast (intra-well) and slow (interwell) time scales.

Figure 10. Accessible side chain conformations of active site tryptophans.

Top: Overlay of three selected snapshots of apo WT (light gray) and apo R261S (light blue) from 100 ns MD simulations. The right zoom highlights conformations sampled by W221 in WT (yellow) and R261S (blue). Conformation 2 of W221 in R261S is not seen in the WT simulation, due to unfavorable proximity to the WT R261 side chain (red lines). The left zoom highlights the conformations sampled by W167 in WT (green) and W167 in R261S (magenta). Bottom: Scatter plots of the χ₁, χ₂ torsion angles from the MD trajectories. Left box (W167): WT (green) versus R261S (magenta). Right box (W221): WT (yellow) and R261S (blue). Also indicated are the χ₁, χ₂ angles of W167 and W221 for pdb 3pag (blue diamond) and 3PAE (cyan circle).

Figure 10 also suggests functional consequences of the increased W221 side chain mobility. Specifically, conformation #2 sampled by the R261S ensemble is absent in the WT ensemble. Its absence can be plausibly explained by the steric clash that would ensue between the WT W221 and R261 side chains. This suggests that the R261S substitution weakens the WT inter-residue interactions restraining W221 side chain mobility. This would be consistent with the ¹⁵N spin relaxation results. Further analysis of the MD ensembles shows that the R261S substitution deletes the WT R261 hydrogen bonds to K219 and G220 of the KSG motif, van der Waals contacts with neighboring residues W231, M252, and M223, and hydrogen bonds and a pi-cation interaction with W221.²⁶ Figure S8 also shows that the increased range of W221 side chain motion correlates with expansion of the active site cavity (measured from the Cβ of V130 to the Cβ of W221).

In summary, the MD simulations establish the plausibility of R261S enhancing local mobility within the active site. The affected residues include W221, a conserved residue stabilizing the acyl-carbonyl that also shows “split” cross-peak behavior during turnover, suggesting distinct active site substates.

Discussion

OXA-24/40 is an example of a carbapenem-hydrolyzing Class D β-lactamase (CHDL) expressed in the Gram-negative pathogen, A. baumannii.⁶ It can hydrolytically deactivate carbapenems, such as doripenem. The amino acid residues supporting hydrolysis have been identified by high-resolution X-ray crystal structures of the apo enzyme⁹ and doripenem-acylated OXA-24/40 variants with point mutations rendering them deacylation-deficient.⁸

Here, we assayed WT OXA-24/40 activity towards doripenem using ¹H CPMG NMR assays (Figure 2). The results revealed two parallel mechanisms deactivating doripenem: conventional hydrolytic cleavage, and non-hydrolytic formation of an inactive β-lactone product. The β-lactone mechanism was described only recently for other CHDLs by Schofield and co-workers.⁴ The fact that multiple CHDLs display this dual mechanism capability suggests the latter derives from conserved features of CHDL active sites that have remained hidden.

We have begun to reveal those features through studies of OXA-24/40 and two mutants R261K and R261S. We have identified features of the active site influencing the preferred mechanism (hydrolysis versus β-lactone formation) for deactivating doripenem, an established carbapenem substrate. Below, we discuss our main findings, and their broader implications for CHDL-mediated carbapenem resistance.

Distinct acyl-enzyme states for distinct deactivation mechanisms

WT OXA-24/40 activity actually favors the β-lactone mechanism, as indicated by the ¹H relaxation-filtered 1D NMR assays. The relative contribution of hydrolysis versus β-lactone formation follows the rank ordering WT < R261K < R261S. Notably, R261S deactivates doripenem almost completely via conventional hydrolysis (Table 1). This means the WT preference for the β-lactone mechanism relies on active-site interactions involving R261. The substitution mutations, R261K/S, somehow weaken these interactions in a manner that discourages β-lactone formation, promotes hydrolysis, or both.

To understand these observations, we investigated how active OXA-24/40 could support both mechanisms, using the rapid 2D data acquisition afforded by the ¹⁵N-¹H BEST TROSY pulse scheme.²⁰ The resulting series of 2D BEST TROSY spectra identified the residues perturbed during turnover of substrate (doripenem). To our knowledge, this is the first such characterization of an active serine β-lactamase.

The BEST-TROSY spectra indicated that acylation of OXA-24/40 by doripenem leads to at least two acyl-enzyme populations with distinct active site environments. In particular, for residues in and proximal to the active site (e.g. G220, W221, and G224), adding doripenem not only shifted the position of the apo state cross-peaks (CSP), but further “split” them into separate cross peaks corresponding to distinct active site microenvironments in slow exchange on the NMR chemical shift time scale. Comparisons of the time-dependence of the W221 side chain εNH across all three constructs (WT, R261K, R261S) suggested that the split peaks were spectral signatures of non-equivalent substates of the active site, one suited for hydrolysis and the other for β-lactone formation. That is, the dual presence of hydrolysis and β-lactone mechanisms could be explained by two acyl-enzyme complex subpopulations, with distinct active site environments promoting distinct mechanisms: hydrolysis or β-lactone formation.

The origins of the two complexes are not yet clear. We consider two possible mechanisms, labeled “A” and “B” in Scheme 1. Mechanism A stipulates a pre-existing equilibrium of two apo-state forms, with distinct active sites, acting in parallel. One form is better suited for β-lactone formation, the other for hydrolysis. Alternatively, mechanism B proposes that the apo-state is characterized by one active site. However, substrate recognition leads to two possible acyl-enzyme complexes, with distinct active sites poised for β-lactone formation or hydrolysis. As a result, the mechanism branches upon acylation, and then proceeds in parallel.

Scheme 1.

Two possible mechanisms of activity. A: OXA-24/40 has a pre-existing equilibrium of two apo-state active sites, one leading to hydrolysis, the other to β-lactone formation. The B: Substrate binding to OXA-24/40 results in distinct separate acyl-enzyme complexes that follow separate branches for hydrolysis or β-lactone formation. In both cases, the “H” and “L” subscripts refer to hydrolysis and β-lactone formation, respectively. The initial β-lactone product is L2R, which then coverts to L2S with rate constant k_RS.

Our ¹⁵N-¹H 2D spectra for apo WT OXA-24/40 lack obvious signs of population doubling stipulated in mechanism A. Nor have we detected evidence of transient minor states via ¹⁵N relaxation dispersion experiments. While these observations suggest mechanism A is unlikely, we cannot rule it out yet, as we have only the perspective from ¹⁵N-¹H spins. Other NMR spins, such as side chain ¹³C nuclei, may prove to be more sensitive detectors of multiple apo states.

Thus, our provisional interpretation favors mechanism B, in which an initial acyl-enzyme complex partitions quickly into distinct subpopulations with active site configurations poised for different mechanistic fates. These configurations are apparently separated by free energy barriers, such that they experience slow or no interconversion, giving rise to distinct bound state cross-peaks suggesting independent routes of doripenem deactivation.

R261S/K and decreased β-lactone formation

WT OXA-24/40 deactivates doripenem via both hydrolysis and β-lactone formation, with the latter mechanism dominating under our assay conditions. The R261S/K substitutions apparently destabilize features of the WT active site supporting the β-lactone mechanism. Moreover, for R261S, the loss of β-lactone formation coincides with a significant gain in hydrolytic activity (Table 1). Yet, the WT residue R261 is not part of the residue cohort directly involved in (de) acylation. For example, it is not part of the three active site elements (STFK, SAV, KSG) that define CHDLs.¹⁰ Nor does it make direct contact with the acyl carbonyl of the ester that covalently links doripenem to the active site serine, S81. Therefore, what kind of atomic-level perturbations issuing from the R261S/K substitutions could explain the substantial change in mechanism?

In the first study documenting CHDL β-lactone activity,⁴ β-lactone formation was proposed as the likely outcome of internal cyclization resulting from nucleophilic attack on the acyl carbonyl by the hydroxyl of the 6α-hydroxyethyl moiety. The 6α-hydroxyethyl group is common to all carbapenems, and in doripenem, it corresponds to the OH attached to C8, and the acyl carbonyl is carbon C7 (Figure 2A II). Nucleophilic attack by the C8 hydroxyl could then lead to β-lactone formation, instead of hydrolysis. In this scenario, the intramolecular nucleophilic attack by 6α-hydroxyethyl hydroxyl requires the bound carbapenem to adopt a conformation positioning the 6α-hydroxyethyl hydroxyl close to the acyl carbonyl, and the side chain of K84, the active site lysine of the STFK element. Such opportune positioning could result from rotation about the C6-C8 single bond; this would place the C8 hydroxyl close to the acyl carbonyl, while simultaneously blocking nucleophilic attack by a water molecule that would otherwise lead to hydrolysis.

Our findings suggest that the ability of the bound carbapenem to adopt and maintain this critical conformation depends significantly on interactions between other carbapenem functional groups and the side chains of active site residues. The documented example is the salt-bridge between the doripenem carboxylate group at C10 to the guanidinium side chain of R261 in WT OXA-24/40.⁸ This WT interaction could help stabilize the orientation of the other parts of bound doripenem to allow for the C6-C8 rotations needed to place the 6α-hydroxyethyl hydroxyl close to the acyl carbonyl. On the other hand, the destabilization or loss of the salt bridge could enable greater rotational mobility about single bonds such as C5-C6 or the C-S bonds, thereby decreasing the likelihood for proper placement of the 6α-hydroxyethyl hydroxyl. We hypothesize that this accounts for the behavior of R261S/K. Specifically, the R261S substitution would weaken the WT salt-bridge, as intended. Replacing the large R261 side chain with the small S261 side chain could enable greater mobility of the bound carbapenem and other neighboring active site residues. Such changes would be consistent with the R261S-induced changes in active site mobility highlighted by both NMR spin relaxation and MD simulations. At the same time, the smaller size of S261 and its own hydroxyl would simultaneously increase access by water molecules to the acyl carbonyl. This could explain the substantial gain in hydrolytic activity coinciding with the loss of β-lactone activity we observed for R261S.

The R261K mutant suggests a different means of perturbing the WT salt-bridge interaction. The gross similarity in size and charge between lysine and arginine, suggests that the K261 side chain could retain some electrostatic interaction with the carboxylate group of doripenem. However, lysine and arginine differ in their side-chain branching and the distribution of their side chain charges. These differences could cause K261 to contact doripenem differently from R261; these alternative contacts could actually hinder, rather than promote proximity of 6α-hydroxyethyl hydroxyl to the acyl carbonyl. The end result would again be decreased β-lactone activity, but for reasons distinct from R261S. Moreover, if K261 does make significant but “incorrect” contacts with doripenem, rather than deleting them as is likely for S261, then water access to the acyl carbonyl could be as difficult or worse than for WT. When considering this possibility, we would not expect R261K to open up new crevices to the active site, as is likely for R261S, and so we would not expect increased hydrolytic activity for R261K. This is consistent with our R261K results showing decreases in both β-lactone formation and hydrolysis compared to WT (Table 1).

The two paragraphs above discuss plausible mutation-induced perturbations of direct interactions between WT R261 and bound doripenem. But another possibility would be mutations influencing the active site mechanism via indirect effects. More specifically, they could perturb inter-residue interactions between R261 and other active sites residues making direct contacts with doripenem. Examples include interactions between the W221 amide NH to the acyl carbonyl, and hydrophobic contacts from L168 to the methyl of the 6α-hydroxyethyl group. Conceivably, the stability of these contacts could rely on particular side chain conformations for W221 and/or L168, stabilized by interactions with R261. The R261K/S substitutions could compromise such stabilization, causing W221 and L168 to lose their WT contacts with doripenem. Our NMR and MD results both indicate that the R261S substitution perturbs multiple active site residues interacting with doripenem. For example, R261S causes different members of the split peaks to dominate during the “real-time” BEST-TROSY than observed for WT. The cross-peaks that “split” upon addition of doripenem turn out to overlap well with residues showing changes in backbone NH bond mobility, as revealed by ¹⁵N spin relaxation. The MD simulations show that the R261S substitution deletes the WT R261 hydrogen bonds to K219 and G220 of the KSG motif, hydrogen bonds and a pi-cation interaction with W221 of the oxyanion hole, and neighboring long-chain residues W231, M252, and M223. These changes coincide with larger amplitude fluctuations of side chains, such as those of W221, that affect the flexibilities of other active site residues, such as L168 and W167.

Sticky loop contacts for communication around the bound carbapenem

The results of the NMR relaxation measurements showed that the R261S substitution changes the apo-state flexibility of β5-β6 residues G222-V225, and the more distal residues in the Ω-loop, L168 and W167. The β5-β6 changes appear reasonable, as these residues are proximal to the R261 side chain, particularly W221. In particular, the WT crystal structure shows evidence of a pi-cation interaction^27,28 between R261 and W221 in the beginning of the β5-β6 turn. The R261S mutation would likely weaken this interaction, opening the way for greater mobility of W221 and its sequential neighbors in the β5-β6 motif.

On the other hand, the concomitant dynamic changes observed for L168 and W167 are more difficult to explain. There appears to be a mechanism whereby dynamic perturbations in one surface loop (e.g. β5-β6 loop) can propagate to other loops (e.g the P and Ω loops).

We propose that propagated effects can be explained, in part, by transient contacts between the residues in different loops, which effectively make the loop interfaces “sticky”. For example, extant crystal structures^8,9 show that G222, G224, and V225 in the β5-β6 loop lie within contact distance of L168 and V169 in the Ω-loop. The side chains of these residues could provide “bridge” contacts between the β5-β6 and Ω-loops. This “bridge” would be in addition to the original “hydrophobic bridge” between M223 (β5-β6) and Y112 (P-loop) described in the first structures of OXA-24/40.^8,9 Together, these bridges enable transient interactions – “stickiness” -- between the semi-flexible β5-β6, Ω, and P loops. The sticky loop contacts could enable R261-mediated interactions to affect more distal residues around (i.e. on the far side of) the acylated doripenem, such as L168 and W167. Figure 11A sketches the basis for these “sticky loop contacts”. Such contacts could mutually reinforce those loop conformations that stabilize doripenem for either β-lactone formation or hydrolysis.

Figure 11. “Sticky Loop” Model.

A: Diagram of the proposed “Stick Loop Contacts” model, in which a point mutation at position R261 (yellow sticks) could further affect the multiplicity of conformations sampled by the β5-β6 (green), Ω-loop (cyan), and P-loop (magenta). Interactions between loop residues provide “sticky” contacts. These transient contacts cause fluctuations in the volume and accessibility of the active site cavity. B, C: Scatter plots from the 100,00 snap shots of the WT and R261S MD trajectories, indicating correlations between fluctuations of the side chain W221 χ1 angle, and loop contact distances including B: β5-β6 G222 and Ω-loop L168, and C: β5-β6 M223 and P-loop Y112. Black/Red dots are snapshots from the WT/R261S trajectories, respectively.

The notion of sticky loop contacts could also explain how a mutation in one loop could affect the conformational flexibility of other loops. For example, the loss of the pi-cation interaction between R261 and W221 (β5-β6) could explain the flexibility changes observed for both the β5-β6 and Ω-loop residues. Increased mobility of the W167 and L168 side chains would reduce those contacts with doripenem supporting β-lactone formation. At the same time, reduced loop interactions could enable larger fluctuations of the active site pocket, and thus, easier access of a deacylating water for hydrolysis. Indirect evidence for such fluctuations are the CSPs induced by doripenem outside the active site, at apparent “hinge” regions of the Ω-loop and P-loops during the BEST-TROSY series.

Support for the sticky loop model comes from correlated motions revealed by scatter plots generated from the MD trajectory snapshots. In particular, Figure 11B and C includes scatter plots of the side chain χ1 fluctuations of W221 on the horizontal axis, versus inter-residue distance fluctuations including G222(β5-β6 loop)-L168(Ω loop), and M223(β5-β6 loop)-Y112(P loop) on the vertical axis. G222-L168 is a bridge contact between the β5-β6 and Ω loops, and M223-Y112 is the hydrophobic bridge contact between the β5-β6 and Ω loops. The black and red dots represent WT and R261S conformations, respectively. For both WT and R261S, the shapes of the distance distributions vary with χ1, indicating correlations between distance and χ1 fluctuations. Moreover, R261S shows a different variation from WT, indicating a mutation-induced perturbation of WT correlated motion. The distances sampled by R261S are generally longer, suggesting a decrease in “sticky” contacts; we noted this for other inter-loop distances (Figures S9 and S10). The physical basis of the correlated behavior appears to be the local contacts between residues in or near the active site.

The correlated motion from MD suggests how the R261S substitution could influence sticky loop interactions. The substitution disrupts local interactions, such as the pi-cation interaction from R261 to the W221 side chain. This enhances the mobility of the W221 side chain; by itself, this is unsurprising as it represents dynamic change at the mutation site. However, the central location of W221 in the active site provides for numerous local contacts to the side chains of loop residues. The local contacts create correlations between the fluctuations of neighboring residues that relay the W221 perturbations to other residues, including those mediating loop “stickiness”. The end result may include changes in loop proximity that influence mechanism of substrate turnover.

What about bound doripenem itself?

As stated, we provisionally favor mechanism B in Scheme I (vide supra), in which an initial acyl-enzyme complex partitions quickly into two subpopulations with distinct active site configurations: E-S_HYD poised for hydrolysis, and E-S_LAC poised for β-lactone formation. Our experimental support for these distinct configurations include the active site residues whose NH cross-peaks “split” into two upon addition of doripenem (Figure 6). In particular, for W221 εNH, one member of the peak pair (“L” cross-peak) corresponded to an active site favoring β-lactone formation, while the other, for hydrolysis (“H” cross-peak) corresponded to that favoring hydrolysis. More detailed statements about the E-S_HYD versus E-S_LAC configurations remain speculative. We note that previous studies of carbapenem binding by Class A^17,18 and Class D^8,29 β-lactamases describe how the initial acylation event can produce distinct active site configurations corresponding to the Δ² versus Δ¹ pyrroline tautomers of the acylated open-ring carbapenem. The Δ² tautomer (proton on pyrroline N) readily hydrolyzes while the Δ¹ tautomer (proton on pyrroline carbon 2) does not. These features would seem appropriate for the E-S_HYD and E-S_LAC configurations in mechanism B. However, this is unlikely, as the crystal structures of deacylation-deficient variants of OXA-24/40 variants (V130D pdb 3PAG and K84E pdb 3PAE) show just the Δ². To go beyond speculation, further studies into E-S_HYD and E-S_LAC could measure parameters sensitive to the bound carbapenem conformation(s), using active OXA-24/40 during turnover. We are investigating the potential of isotope-filtered NMR experiments for such studies.

Our proposed explanation for the R261S/K mutation effects stipulates that mutations destabilize the doripenem conformations that allow for the internal nucleophilic attack by the hydroxyl of the 6α-hydroxyethyl needed for β-lactone formation.⁴ Destabilization could include the loss or corruption of WT R261-mediated contacts, thereby changing the range of doripenem flexibility, and thus, the likelihood that its own 6α-hydroxyl or a water molecule is appropriately poised for nucleophilic attack on the acyl carbonyl. In the context of W221 εNH in the BEST-TROSY series, the cross-peak “H” would represent the subset of active site conformers granting a deacylating water molecule easier access to the acyl carbonyl. Alternatively, the bound antibiotic flexibility could remain essentially unchanged, however, the mutations change its overall orientation within the active site pocket that decreases the probability of β-lactone formation. Investigating these possibilities requires experiments interrogating the internal dynamics of the acylated antibiotic. A promising approach are the off-resonance ¹H R_1ρ relaxation measurements we applied to the Gram-positive S. Aureus sensor domain, BlaR^S, acylated by CBAP^.30 Equally informative would be protein experiments comparing the side chain dynamics of active site residues for apo versus acylated OXA-24/40. Such experiments are in progress.

Concluding remarks

Our summary model for our OXA-24/40 results is as follows: The R261S substitution removes the intermolecular salt-bridge, and van der Waals contact, perturbing the extent and lifetime of non-covalent interactions with other active site residues. These contact changes lead to larger amplitude conformational fluctuations of the active site, as revealed by the ¹⁵N spin relaxation measurements and MD simulations for apo WT and R261S. The increased active site flexibility destabilizes the bound doripenem conformation characteristic of WT; the active site now samples a broader range of conformations. Formation of β-lactone suffers as the conformations supporting it become less frequent. Hydrolysis becomes more efficient due to easier access to the acyl carbonyl by a deacylating water molecule.

Our results go beyond OXA-24/40 to include the other CHDLs sharing the β-lactone mechanism. They show that different deactivation mechanisms require binding “poses” which include the possible conformations of doripenem, and those of active site residues that either provide stabilizing interactions (e.g. R261) or that need to move out of the way. This could mean local conformational adjustments by those active site residues, which in turn requires that they have some local flexibility. Mutations could remove the required flexibility, or introduce too much, making certain binding poses e.g. those required for β-lactone formation, less probable.

In the context of carbapenem resistance, such mutations could also expand opportunities for alternative mechanisms of substrate activity. This perspective views the active site as harboring “latent” mechanisms that can take hold upon suitable mutations.^31,32 To anticipate new deactivation mechanisms, one should consider the effects of clinical point mutations on the conformational ensemble of the active site. Mapping the scope of inter-residue interactions affecting the bound antibiotic conformation(s) could help reduce the number of surprising clinical mutations at unexpected residue positions.

Materials and methods

Protein preparation and purification

The WT OXA-24/40 construct was previously cloned into a pET24a vector without an affinity tag. Mutants were generated using site-directed mutagenesis via the QuikChange XL kit (Agilent). Mutagenesis was confirmed through sequencing by Functional Genomics. Plasmids were transformed into BL21(DE3) E. coli competent cells and plated onto LB agar plates containing 50 μg/mL kanamycin. Cell culture preparation, protein purification, and isotope labeling was done following the previously described protocols.²⁴

The final OXA-24/40 NMR samples (WT and mutants) were prepared through buffer exchange into 20 mM NaH₂PO₄, 50 mM NaCl, pH 7.0 containing 10% D₂O and 0.03% NaN₃, using centrifugal filters with a 10 K Da molecular weight cutoff (Amicon). Sample concentrations ranged from 50 to 500 μM, according to the application, as given in the Results subsection.

Differential Scanning Fluorimetry (DSF)

Master mix stocks were prepared containing final concentrations of 5 μM protein and 10X SYPRO Orange (Pierce Thermo) in NMR buffer and contained either 0 μM, 5 μM, 10 μM, or 50 μM doripenem or imipenem (Sigma-Aldrich) and were incubated overnight to ensure reactions were complete. Each master mix was dispensed in 50 μL triplicate aliquots into a 96-well skirted PCR plate (Bio-Rad) and subsequently sealed with a clear thermal-seal film to prevent evaporation. The sample plate was spun down at 2,000 RPM for 2 minutes in an Eppendorf 5430 R centrifuge equipped with a A-2-MTP rotor. DSF experiments were conducted on a CFX96 Touch Real-Time PCR system (Bio-Rad) using the HEX channel for both excitation and emission. The experimental program collected fluorescence intensity at 0.5 °C increments with a 5 second equilibration step between each temperature. The fluorescence intensity (RFU) data points were normalized by the highest value, and then fit to Boltzmann Sigmoidal curve using Prism GraphPad, where the inflection point indicated the melted temperature (T_m). Statistical uncertainties were estimated as the standard deviation of the inflection point for each triplicate data set.

NMR spectroscopy: General aspects

NMR spectra were collected on a 16.4 T (700 MHz ¹H Larmor frequency) Bruker Avance I system, equipped with a TCI cryogenically-cooled probe. Data acquisition and processing used Topspin 2.1 (Bruker Biospin, Inc.). The nominal sample temperature was 295 K.

Peak picking and sequential resonance assignments were carried out using Sparky 3.³³ The chemical shifts for OXA-24/40 backbone atoms were previously assigned²⁴ and are available in the BMRB (accession number 50777). Mutant resonance assignments were confirmed by recording standard 3-D triple-resonance experiments including ¹⁵N-edited NOESY-HSQC,^34,35 TROSY³⁶ versions of 3D-HNCACB,^37,38 and 3D-HNCO spectra^38,39 and comparing them with those used to assign WT.²⁴

¹H NMR of doripenem

The ¹H resonances of doripenem were assigned using standard 2D ¹H–¹H spectra (ROESY,⁴⁰ dipsi2-TOCSY⁴¹), and natural abundance 2D ¹H–¹³C (HSQC-dipsi2-TOCSY) spectra. The spectra were collected for fresh samples of intact doripenem dissolved in OXA-24/40 buffer at 2.5 mM. We used the same experiments for hydrolyzed doripenem. The ¹H assignments were confirmed via comparisons with those in the literature.⁴

The pulse scheme for the ¹H 1D NMR activity assay consisted of a π/2 pulse followed by ¹H CPMG ^42,43 spin-lock for R₂ relaxation filtering of protein signals, and then a SOGGY gradient echo⁴⁴ for water suppression prior to ¹H signal-detection. The ¹H CPMG spin-lock was a train of 50 microsecond ¹H π-pulses, separated by 2 milliseconds, applied for 100 ms.

The ¹H 1D NMR activity assay consisted of a π/2 pulse followed by 100 ms ¹H CPMG^42,43 spin-lock (50 microsecond π-pulses, 2 millisecond interpulse delay) as a relaxation filter to suppress protein signals, and then a SOGGY gradient echo⁴⁴ for water suppression prior to detection. The assay steps included adding doripenem to a protein solution pre-equilibrated in an NMR tube, then a four-minute delay for inserting the sample into the magnet, locking, and shimming, and then continuous collection of 400 1D ¹H datasets. Fourier transformation produced 400 ¹H 1D spectra with peaks depicting substrate and product populations at time points t[n] = 4 + n*0.65 minutes after substrate addition (n = 1 to 400).

For quantitative analysis, we integrated the C9 methyl peaks from the substrate S (intact doripenem), the hydrolyzed product P, the initial β-lactone product 2R, and the final β-lactone product 2S), producing the time traces plotted in Figure 2C. The early portions of the S, P, and 2R time traces showed linear behavior (Figure S11), consistent with zero-order kinetics expected under steady-state conditions and [S] ≫ K_M,¹⁶. We therefore fit the first 9.75 minutes (first 15 time points) to the linear steady-state expressions:

$$I_S(t)=-m_{S}t+b_S,I_P(t)=+m_{H}t,I_{2R}(t)=+m_{L}t$$ (1)

I_S(t), I_P(t), and I_2R(t) denote the normalized peak integrals for substrate ‘S’, hydrolyzed product ‘P’, and the initial β-lactone product, ‘2R’. The slopes m_S, m_H, and m_2R, and the intercept term b_S were the adjustable parameters, with subscripts “S”, “H”, and “L” indicating substrate, hydrolysis, and β-lactone formation, respectively.

For each assay, we first determined the substrate intercept parameter, b_S, which corresponds to the ratio S_T/S₀, where S_T is the total substrate concentration, 2.5 mM, and S₀ is the substrate concentration indicated by the 1st 1D spectrum, S₀. We could then calculate rate constants k_S (substrate depletion), k_H (formation of hydrolyzed product), and k_L (formation of 2R β-lactone product), using

$$k_S=\left(\frac{m_S}{b_S}\right)\left(\frac{S_T}{E_T}\right),k_{H,L}=\left(\frac{m_{H,L}}{b_S}\right)\left(\frac{S_T}{E_T}\right),b_S=\frac{S_T}{S_0}$$ (2)

The ratio of total substrate to total enzyme, S_T/E_T, was 50 for all cases. Statistical uncertainties were estimated using Monte Carlo simulations based on estimates of 1D peak-height uncertainties.

Time-dependent OXA-24/40 CSPs

We generated U-¹⁵N WT, R261, and R261S to compare their responses to the addition of fresh doripenem. For each construct, we added a small volume of concentrated doripenem to a protein solution already in an NMR tube. The sample (140 μM protein and 14 mM doripenem) was briefly and gently vortexed, and then placed in the NMR spectrometer for continuous recording of 2D ¹⁵N-¹H spectra, using the ¹⁵N-¹H BEST-TROSY²⁰ pulse scheme. Each 2D dataset consisted of 64 complex data points (128 FIDs) in the ¹⁵N dimension, with 16 scans per FID (1024 points). The total time required per 2D dataset was 10.68 minutes. Data collection continued without pause for ~ 5.5 h, producing 30 2D spectra, the first starting < 5 minutes after the addition of 14 mM doripenem to 140 μM OXA-24/40. Subsequently, we recorded standard 2D NH TROSY spectra, separated by longer time intervals (i.e. 24 h and 1 month later) to monitor the return of the enzyme to its apo state.

The BEST-TROSY NH chemical shift perturbations (CSPs) were measured relative to apo protein (zero-time point), using

$$\begin{gathered} {\Delta }_{NH}^{Apo\to Complex}=\sqrt{{\left(\Delta {\delta }_H^{Apo\to Complex}\right)}^2+{\left(\alpha \Delta {\delta }_N^{Apo\to Complex}\right)}^2} \\ \Delta {\delta }_H^{Apo\to Complex}={\delta }_{H,Apo}-{\delta }_{H,Complex} \\ \Delta {\delta }_N^{Apo\to Complex}={\delta }_{N,Apo}-{\delta }_{N,Complex} \end{gathered}$$ (3)

The constant α = 0.154 compensates for the larger shift range of ¹⁵N versus ¹H^N, and is the average ratio Δδ_H/Δδ_N in the BMRB chemical shift database.⁴⁵

¹⁵N relaxation measurements

We measured backbone amide ¹⁵N spin relaxation rate constants on U-¹⁵N OXA-24/40 samples using 2D ¹⁵N-¹H methods described previously.²⁴ Each 2-D spectrum consisted of 1024 points in t₂ (¹H^N) and 150 total points (75 unique time points) in t₁ (¹⁵N), covering an ¹⁵N sweep-width of 2500 Hz. We collected a series of 2D spectra with relaxation delay consisting of an ¹⁵N CPMG spin lock with duration T_LOCK, followed by a longitudinal relaxation time, T_Z = T₀ – (T_LOCK/2), with T₀ was fixed at 100 milliseconds. The CPMG train consisted of 130 microsecond π-pulses separated by 1 millisecond. We collected a series of 2D spectra with T_LOCK = 16.5 (x2), 33.0, 24.7, 41.2, 49.4 (x2) milliseconds, from which we generated residue-specific data files of cross-peak integrals versus T_LOCK. The data files were fit single-exponential decay functions to determine the R₂ – R₁/2 values, using standard methods described previously.²⁴ The R261S and WT plots of ¹⁵N R₂-(R₁/2) versus sequence were quite similar, indicating preservation of overall tumbling motion (Figure S12). Their trimmed averages were identical within the estimated errors: WT < R₂ – R₁/2> = 26.1 ± 2.6 s⁻¹, while R261S < R₂ – R₁/2> = 26.3 ± 2.4 s⁻¹.

For larger globular proteins (>20,000 Da) at higher magnetic fields, the value of R₂ – (R₁/2) is essentially proportional to J_eff (0), a mobility parameter of a given NH bond.^22,23 Specifically,

$$\frac{R_2-(R_1∕2)}{2D\{1+(C∕3D)\}}=J_{eff}(0)=J_{NH}(0)+\lambda R_{ex}$$ (4)

The J_NH(0) term reflects re-orientational motion of the NH bond due to overall protein rotational diffusion, plus internal dynamics on the nanosecond-subnanosecond time scale. The R_ex term accounts for microsecond-millisecond exchange processes that broaden the ¹⁵N linewidth, thereby increasing J_eff(0). The pre-factor 2D*(1 + C/3D) contains constants relevant for the ¹⁵N-¹H dipole–dipole and ¹⁵N chemical shift anisotropy (CSA) relaxation mechanisms, respectively. At B₀ = 16.4 T, 2D(1 + C/3D) ~ 3.9 × 10⁹ r/s. Figure 9 plots the sequence-specific differences in J_eff(0) values, normalized by the trimmed averages for apo WT and apo R261S,

$$\Delta \rho =\frac{J_{eff}^{R261S}(0)}{\langle J_{eff}^{R261S}(0)\rangle }-\frac{J_{eff}^{WT}(0)}{\langle J_{eff}^{WT}(0)\rangle }$$ (5)

Explicit solvent MD simulations

MD simulations of apo WT and R261S OXA-24/40 were performed at 300 K using the AMBER 19 software package,²⁵ and the ff14SB force field.⁴⁶ The initial structures were constructed from pdb 3PAE, the crystal structure of the deacylation deficient K84D variant of OXA-24/40 in complex with doripenem.⁸ The D/K and R/S mutations and removal of doripenem were carried out using Pymol Version 2.0.⁴⁷ The resulting apo WT and R261S constructs were simulated separately, using essentially the same protocols for initial energy minimization, equilibration, and the production runs. All covalent bond lengths were restrained using the SHAKE⁴⁸ algorithm, with a relative tolerance of 0.001. The protein was first immersed in an octahedral box filled with OPC water molecules.⁴⁹ Energy minimization began with just the proton coordinates for 10,000 steps, followed by the solvent for 20,000 steps, and then the whole system for 25,000 steps. After minimization, the systems underwent a short equilibration simulation of 40 ps at a constant pressure for 100 ps, and at constant energy for 100 ps. The initial atomic velocities were generated from a Maxwellian distribution. An additional 600 ps equilibration simulation was carried out to establish steady values for temperature (300 K), pressure (1 bar),⁵⁰ and potential energy. Thereafter, production run simulations were carried out using a time step of 2 femtoseconds for a total trajectory time of 100 nanoseconds. Coordinate snapshots were written every picosecond (every 500 steps), producing a total of 100,000 snapshots. The trajectories were analyzed using CPPTRAJ.⁵¹