Genes encoding for PenL-Cys69Tyr and PenL-Asn136Asp were isolated as described previously (Dobson et al., 2012) and sub-cloned into pET28a(+) expression vector. E. coli BL21(DE3)-competent cell strain was used to overexpress PenL-WT, PenL-Cys69Tyr, and PenL-Asn136Asp. Transformed cells were grown in Luria-Bertani (LB) media supplemented with 100 μg/mL of kanamycin at 37°C until the OD₆₀₀ reached ∼ 0.6 and induced by 0.5 mM of isopropyl-β-D thiogalactopyranoside (IPTG). After an additional 16 h of incubation at 18°C, cells were centrifuged at 5,000 × g and 4°C for 20 min. Cell pellets were resuspended in a buffer containing 50 mM Tris-HCl, pH 7.5, 500 mM NaCl and 10 mM imidazole, supplemented with 0.1 mM phenylmethane sulfonyl fluoride (PMSF), 1 mM dithiothreitol (DTT), and DNase I. The resuspension cocktails were disrupted by high-intensity sonication at 4°C, and the insoluble fractions were separated using high-speed centrifugation (20,000 × g at 4°C for 30 min). The soluble fractions, containing the desired proteins with N-terminal 6 × His-tag, were loaded onto Ni-agarose columns. Unbound proteins were washed out with excess buffer containing 50 mM Tris-HCl, pH 7.5, 500 mM NaCl, and 20 mM imidazole. PenLs were eluted using the above buffer containing 250 mM imidazole. To remove the N-terminal 6 × His-tag, the eluted fractions were pulled and dialyzed against the buffer containing 20 mM Tris-HCl, 150 mM NaCl, and 2 mM CaCl₂, pH 7.5, followed by treatment with 10 U of human α-thrombin (HTI, United States) per 1 mg/mL protein. PenLs were further purified via size-exclusion chromatography using a HiLoad 16/60 Superdex 200 pg column (GE Healthcare, United States) saturated with buffer (20 mM Tris-HCl, pH 7.5, and 50 mM NaCl).

The purified PenL-Cys69Tyr and PenL-Asn136Asp were concentrated to 10 μg/μl using a 10,000 Da cut-off Vivaspin centrifugal concentrator (Sartorius). Crystallization was carried out using the hanging drop vapor diffusion method by mixing a 1.2 μl protein sample with a 1.2 μl reservoir solution. Initial screenings were set up using commercial crystallization kits obtained from Hampton (USA) and Rigaku (Japan). Crystals of PenL-Cys69Tyr were grown in a solution containing 200 mM sodium acetate trihydrate and 20% (w/v) polyethylene glycol 3,350 at 4°C. By contrast, crystals of PenL-Asn136Asp were grown in a solution consisting of 100 mM sodium acetate at pH 5.0, 200 mM sodium chloride, and 25% (w/v) polyethylene glycol 3,350 at 20°C. Crystals were transferred to the cryo-solutions comprising a growth solution supplemented with 15–20% glycerol for 30 s, followed by flash-freezing via immersion in liquid nitrogen.

To determine the complex structure of PenL-Cys69Tyr and PenL-Asn136Asp with ceftazidime-like glycylboronate (CBA), the cryo-solutions were supplemented with 2 mM CBA, followed by soaking crystals overnight at 4°C before flash-freezing. All the crystals were diffracted at a maximum of ∼1.3 Å resolution. X-ray diffraction and data collection were performed at Pohang Light Source (PLS) beamline 5C (Pohang, South Korea) using the ADSC Q315r CCD detector. Collected data were indexed, integrated, and scaled using HKL2000 (HKL Research Inc.). The structure of the two PenL variants was determined via molecular replacement using MolRep (Abergel, 2013) and the structure of PenL-WT (PDB code 5GL9) (Yi et al., 2016) as a reference model. The refinement was carried out with Refmac5 (Murshudov et al., 1997) and phenix.refine (Afonine et al., 2012; Echols et al., 2014). The coordinates and restraints of CBA were generated by eLBOW, and manually fitted to mFo-Fc map employing Coot (Emsley and Cowtan, 2004). Details of data diffraction and structural refinement are shown in Supplementary Table S1.

Because of its poor spectroscopic property, ceftazidime (CAZ) can only be measured at a maximum of ∼100 μM. Therefore, appropriate amounts of purified enzymes were mixed with various CAZ concentrations ranging from 5 to 100 μM in a reaction buffer comprising 50 mM potassium phosphate at pH 7.0 supplemented with 20 μg/mL of bovine serum albumin. The absorbance at 260 nm was immediately monitored at 25°C using a cuvette holder temperature controller. The initial velocity of PenL-WT was measured during the first 10 s using a standard procedure, and the velocities (v) were fitted to the Michaelis-Menten equation (Supplementary Information). The first-order persistence of reaction by PenL-WT prevented the calculation of K_M and k_cat within the experimental setup range; however, the catalytic efficiency can be estimated by reciprocal plotting, in which the slope of the regression curve is k_cat/K_M. Data fittings were carried out using in-house Python scripts with the power of SciPy (Oliphant, 2007) library for parameter estimation, and Matplotlib (Hunter, 2007) module for data visualization.

PenL-WT, PenL-Cys69Tyr, and PenL-Asn136Asp were subjected to all-atom molecular dynamics (MD) simulations in explicit solvent using Gromacs 5.0.7 suite (Lemkul et al., 2015) and Gromos96-43a1 force field (Ramos et al., 2019). The coordinate of PenL-WT was obtained from the RCSB Protein Data Bank by fetching the ID 5GL9 (Yi et al., 2016). Molecules were solvated using the TIP3P water model and neutralized with a cubic boundary of 0.10 M of NaCl. Particle mesh Ewald method (Abraham and Gready, 2011) was used to determine the electrostatic interaction of systems, with real space and a van der Waals distance cut-off of 10 Å. After energy minimization by 500-step in steepest descent calculation, systems were heated to 300 K and simulated for 50 ns. The MD trajectory analysis was performed directly in the Gromacs package, and the RMSD and radius of gyration plotted with 100 frame intervals (Supplementary Figure S8).

To elucidate the mechanism associated with CAZ hydrolysis by the two PenL NCR-ESBLs (PenL-Cys69Tyr and PenL-Asn136Asp), we first determined their kinetic parameters (Table 1). The two PenL NCR-ESBLs and wild-type PenL (PenL-WT) hydrolyzed CAZ over time in which the CAZ decay was linearly related to time for 10 s (Supplementary Figure S2). Figure 1B illustrates the comparative kinetics of CAZ hydrolysis by the two ESBLs and WT of PenL. The kinetics of CAZ hydrolysis were similar between the two ESBLs, with V_max and K_M values of 156.4 nmol/min per mg and 22.3 μM, respectively, for PenL-Cys69Tyr, and 155.8 nmol/min per mg and 12.5 μM, respectively, for PenL-Asn136Asp. The mutant’s K_M values are 30–60 times lower than that of WT (Table 1), indicating that a stable acyl-enzyme complex may be formed after binding in the mutant ESBLs. Although the k_cat values of the two PenL-ESBLs were lower than that of PenL-WT, their catalytic efficiency (k_cat/K_M) was enhanced 3–5-fold compared with that of PenL-WT (Table 1, Figure 1B, and see Supplementary Information). Those biochemical results are likely to correspond to the higher MIC values of CAZ than PenL-WT for B. thailandensis (Dobson et al., 2012). Overall, the changes in kinetic property, primarily with the decreased K_M, in the two PenL ESBL variants indicate that the two single substitutions led to alternative substrate recognition for the enzyme. Interestingly, the single mutations at non-canonical regions are involved neither in substrate binding or catalysis.

Next, to examine whether the enhanced CAZ affinity and catalytic efficiency were related to changes in the conformation of PenL-Cys69Tyr and PenL-Asn136Asp, Circular dichroism (CD) spectroscopy was used for comparison with that of wild-type (Supplementary Information). The CD spectra of their respective apo-forms (PenL-Cys69Tyr-apo and PenL-Asn136Asp-apo) showed no significant differences compared with the PenL-WT (Supplementary Figure S3A). However, the CD spectra of the two NCR-ESBLs in the presence of a non-hydrolyzable CAZ analog glycylboronate (CBA) slightly deviated from their apo-forms as well as the PenL-WT-apo form. The degrees of the CD spectral change of the mutants appeared similar to that of WT from CAZ presence. Moreover, the deviations in a spectral difference between the two ESBLs seemed even less than those with CBA, which might result from the rapid hydrolysis of CAZ accompanied by the slight conformational changes (Supplementary Figure S3B). These observations suggested that the conformation of the two NCR-ESBLs was similar to that of PenL-WT; however, the NCR-ESBLs may recognize the substrate CAZ (or CBA) by marginally changing their conformation of the side chain of residues in the active site or recognition ensemble.

To further investigate the conformational changes based on substrate recognition, we delineated four crystal structures of the two PenL NCR-ESBLs: two involving the apo-form and the other two associated with the CBA-bound form (Supplementary Table S1 and Supplementary Figure S4). Consistent with CD spectra, PenL-Cys69Tyr-apo and PenL-Asn136Asp-apo were structurally similar to PenL-WT (PDB ID:5GL9) (Yi et al., 2016) with root-mean-square deviations (RMSDs) of 0.221 and 0.186 Å at their Cα atoms relative to the wild-type, respectively (Figure 2A). In particular, the configurations of the conserved catalytic residues and the catalytic water (W4) associated with hydrolysis coincided with those of PenL-WT (Figures 3B, 2B and Supplementary Figure S5). These similarities suggest that the respective substitution does not structurally alter the active site involved in acylation/diacylation of the β-lactam backbone during hydrolysis (Lamotte-Brasseur et al., 1991; Page, 2008; Brown et al., 2009; Drawz and Bonomo, 2010; Papp-Wallace et al., 2013).

Interestingly, the PenL-Cys69Tyr-apo showed structural variation in the β3-β4 strands compared with PenL-WT-apo (Figure 2C). Notably, the C-terminus of β3 and the N-terminus of β4 in PenL-Cys69Tyr-apo were disrupted in three (Thr237 to Asp240) and two (Thr243 to Gly244) residues, respectively. The disruption in the β-strands was apparently induced by the steric hindrance of bulky Tyr69 against the β3-β4 sheet, resulting in enhanced flexibility and dislocation of the loop by 7.4 Å at Tyr241 (Figure 2C). The increased flexibility of the β4 N-terminus in PenL-Cys69Tyr-apo induced a loss of hydrogen bonds with adjacent C-terminus of β5 and subsequently dislocated the β5-α8 loop away from the corresponding position of Pen-WT-apo by 10.8 Å at Arg270 (Figure 2C and Supplementary Movie S1). Since the β3-β4 segment is one of the recognition ensembles, these changes likely expand the active site of the enzyme to accommodate large substrates such as CAZ. The structure of PenL-Cys69Tyr-CBA shows that the residues ranging from Thr237 to Asp240 of β3 mediated substrate recognition, and the loop interacts with the aminothiazole ring and acetamido backbone, rather than the oxyimino group of CBA (Figure 2D). These results suggest that the NCR mutation induced similar effects as in the known canonical ESBLs. For instance, the dislocation of the loop between β3-β4 strands in TEM-52 ESBL (Gly238Ser on β3) enlarged the active site to bind CAZ without further loop extension and with minimal dislocation (∼2.9 Å) (Papp-Wallace et al., 2012). In addition to structural analysis, we calculated the electrostatic distribution on the surface of proteins (Figure 3). Surprisingly, we found a significant change in electrostatic property from negative to positive in the active site of PenL-Cys69Tyr-apo, which has never been reported in ESBLs. Upon distortion of strands β3-β4, the helix α8 was extended by four residues (Ala272-Arg275) (Figure 2C and Supplementary Movie S1). Accordingly, Arg275 rotated 180° to relocate closer to the Thr237 with a distance of ∼6.7 Å (between N atom of Arg275 and backbone N of Thr237) in the Arg220-Asp276-Thr237 cluster, which is critical for substrate binding and hydrolysis (Papp-Wallace et al., 2010, 2012; Figures 2C,D, 3C). The Thr237 residue in the cluster mainly contributed to the oxyanion hole, stabilizing the tetrahedral intermediate formed via nucleophilic attack by catalytic Ser70 (Figures 2C, 3 and Supplementary Figure S5). Note that, based on the established mechanism of β-lactam hydrolysis (Drawz and Bonomo, 2010; Papp-Wallace et al., 2013), the N atom of Ambler residue 237 (i.e., Thr237 in this case) participates in the formation of the oxyanion hole, stabilizing the tetrahedral intermediate of a substrate with class A β-lactamase. Besides the residue 237 in the cluster, Arg220, whose configuration was not changed between PenL-WT and PenL-Cys69Tyr, is directly in close contact with Thr237 (∼4.4 Å) (Figures 2C,D). Remarkably, the Arg275 in the Cys69Tyr mutant positions adjacent to Arg220 within ∼3.6 Å, which may strengthen the positive charge around the oxyanion hole (Supplementary Figure S9C). Consequently, the accumulation of additional basic amino acid (i.e., Arg275) increased the positive electrostatic potential around the oxyanion hole (Figure 3 and Supplementary Movie S2). Notably, due to the presence of two negatively charged groups on CAZ (Figure 1C), the increased net positive charge in the PenL-Cys69Tyr active site may enhance the binding affinity with CAZ, which is consistent with the kinetic data (Table 1). Taken together, the findings indicate that the Cys69Tyr mutation in class A β-lactamase affects the structural interaction between β3–β4 strands and substrate by enlarging the substrate-binding cleft of the enzyme and altering electrostatic property around the oxyanion hole in the active site. These changes in the NCR-ESBL may result in enhanced recognition of large CAZ without affecting the conserved catalytic residues of class A β-lactamases located in the active site for hydrolysis.

Similar to Cys69, the residue Asn136 is positioned outside the essential catalytic segments and not involved in the catalysis of antibiotics. However, a novel ESBL variant carrying the substitution of Asn136Asp was found only in PenL (Dobson et al., 2012) and also exhibited an increase in CAZ hydrolysis activity (Table 1 and Figure 1B). Asn136 is located proximally to active site cleft so that its side chain forms a hydrogen bond to the backbone of Glu166 on Ω-loop and stabilizes the energetically unfavorable non-proline cis-peptide (Berg et al., 2012) between Glu166-Thr167 (Figure 4A). Notice that Glu166 is a critical catalytic residue that involved in both the activation of the Ser70 and deacylation step in β-lactam hydrolysis (Lamotte-Brasseur et al., 1991; Guillaume et al., 1997; Meroueh et al., 2005; Page, 2008; Drawz and Bonomo, 2010; Papp-Wallace et al., 2013). Asn136 thus appears to be responsible for the proper orientation of Glu166 and Ω-loop. Evidently, the loss of the stabilization by Asn136 (i.e., by mutation Asn136Ala) resulted in the functional deficiency of a class A β-lactamase (Banerjee et al., 1997, 1998). The replacement of Asn136 to aspartate abolishes a hydrogen bond formed with Glu166 in WT (Figure 4A), which probably leads to an increase in Ω-loop flexibility. To our surprise, the structure of PenL-Asn136Asp demonstrates no significant difference in comparison with PenL-WT (Figure 4A). Nonetheless, PenL-Asn136Asp can accommodate the CBA into its active site wherein the conformation does not change substantially (Figure 4C and Supplementary Figure S6). Residue Arg275 also adopts the same configuration as that of the wild-type enzyme, implying no such alteration of electrostatic distribution around active site cleft, as seen in PenL-Cys69Tyr (Figure 4B). However, the large-sized CAZ may bind to the active site with the help of the change in lid segment, although no such considerable distinction between PenL-Asn136Asp and PenL-WT, as well as between PenL-Asn136Asp-apo and PenL-Asn136Asp-CBA, is observed. Two residues, including Arg104 and Tyr105, in the lid of PenL-Asn136Asp-apo, appeared to move away from the active site than WT (Figures 4B,C). In PenL-Asn136Asp-apo, residue Arg104 swung away by ∼4.0 Å from the corresponding position of WT, whereas Tyr105moved upward by ∼1.3 Å. Then, Arg104 returned toward the active site when the CBA bound. Instead, the Tyr105 moved further upward by another 2.0 Å in the CBA bound form (Figure 4, Supplementary Figure S7, and Supplementary Movie S3). These observations indicate that Asn136Asp mutation may affect the conformation of the lid segment to expand the size of the substrate-binding cleft and receive large size CAZ. Interestingly, the changes in the lid resemble that the effects of mutation on Ω-loop propagated into those residues to enlarge and accommodate CAZ (Yi et al., 2016). Therefore, the mutation at 136 may cause the instability of the Ω-loop.

An interesting question can then be raised here on how the relatively negative CAZ can be attracted to an electrostatically negative substrate-binding cleft of the PenL-Asn136Asp (Supplementary Figure S6) like that of PenL-WT. Taken together with the CD spectra data and kinetic analysis, these observations indicate that the binding of CAZ into the binding site might be governed by latent factors other than the electrostatic attraction, which was not be revealed in the static crystal structure. We thereby suggest two hypotheses. First, the higher degree of freedom would be induced to the Ω-loop of PenL-Asn136Asp due to the loss of one hydrogen bond by the mutation that occurred outside the Ω-loop. Indeed, this effect resembles cases in other class A β-lactamase ESBLs where the stable network on Ω-loop per se was devastated by the substitutions at Ambler position Arg164 or Asp179 located on Ω-loop (Orencia et al., 2001; Wang et al., 2002a). In this case, the resulting enhanced flexible motion of Ω-loop may induce the transient enlargement of active site cleft and facilitate the accommodation of third-generation cephalosporins like CAZ or CTX. Second, the mutation Asn136Asp may create intrinsic dynamic conformers that would efficiently accommodate CAZ. The viewpoint of functional promiscuity in protein conformation has been suggested formerly (Tokuriki and Tawfik, 2009), whereby poorly packed or disordered conformation by an accumulation of single mutations in proteins evolves conformationally diverse structures to adapting novel substrates.

For the reasons, the molecular dynamics (MD) simulation was therefore conducted to compare the dynamic property of PenL-Asn136Asp with PenL-WT and PenL-Cys69Tyr on the CAZ binding (Figure 5 and Supplementary Figure S8). The MD simulation indicates that the overall structure of PenL-Asn136Asp was stable through a 50 ns trajectory and roughly similar to PenL-WT. However, RMSD at three critical segments of PenL-Asn136Asp, similar to PenL-Cys69Tyr, was higher than PenL-WT and appeared to fluctuate, especially at the lid region (Figure 5 and Supplementary Figure 8). Furthermore, the radius of gyration (Rgyr) plots also demonstrate a potential unfolding of two PenL-ESBLs, in contrast to the sustainable motion of PenL-WT (Figure 5). These observations strongly suggest that PenL-Asn136Asp (and PenL-Cys69Tyr) has a higher tendency for disorder than PenL-WT, which is correlated with the higher adaptability toward CAZ in term of protein dynamism (Tokuriki and Tawfik, 2009). In the molecular view, two regions appeared to involve in the accommodation of the large-size antibiotics followed by the transient disorder. First, the MD results showed the serial propagation of the mutation effect from Ω-loop into the β3-β4 element. The conformations of primary residues, including Asn136, Glu166, reactive Ser170, and the three essential catalytic segments involved in binding, remained unchanged in PenL-WT through MD trajectory (Figure 5A). By contrast, the mutant Asp136 side chain in PenL-Asn136Asp-apo shifted away from the Glu166 backbone of Ω-loop, leading to a large fluctuation of Glu166 (approximately 5.2 Å) and Ser70 (Figures 5B,C, Supplementary Figures S8B,C and Supplementary Movie S4). Surprisingly, the high degree of fluctuation involving the Ω-loop and Ser70 pushed the β3–β4 loop away via steric hindrance with a concurrent breakdown of the hydrogen bond between Asn170 and Asp240, located on the Ω-loop and β3, respectively (Figures 4B, 5C and Supplementary Movie S4). The changes in the regions, which are intensively involved in substrate recognition, may result in the momentary expansion of the active site space and the improvement of CAZ binding, likewise observed from PenL-Cys69Tyr and Ω-loop tandem repeat ESBL of PenL (Yi et al., 2016). Second, the Arg104 residue in PenL-Asn136Asp-apo penetrated the substrate-binding region compared with the Pen-WT, to facilitate the recognition and interaction with acetamido backbone of CBA through hydrogen bonding (Supplementary Figure S6B). Interestingly, the conformation of Arg104 in PenL-Asn136Asp-CBA was similar to those of PenL-Cys69Tyr-apo and -CBA, but further away than PenL-Asn136Asp-apo, suggesting that the entry of Arg104 into the substrate-binding cleft stabilized the acyl-form of PenL-Cys69Tyr via interaction with W5 in CBA (Figure 5C and Supplementary Figure S6). The increased affinity of PenL-Asn136Asp to CAZ also supported this result compared with PenL-WT (Table 1). Besides, Tyr105 in PenL-Asn136Asp-CBA, a well-conserved residue for ligand interaction (Papp-Wallace et al., 2013), was dislocated further way than the PenL-WT and PenL-Asn136Asp-apo (Figures 4C, 5C). Indeed, the side-chain conformation of two residues in PenL-Asn136Asp dislocated further away from the substrate-binding site in crystal structures with respect to those of PenL-Cys69-Tyr and WT (Figure 4 and Supplementary Figure S7). Therefore, the changes involving the lid region may induce momentary expansion of the active site and improve CAZ binding in the PenL-Asn136Asp and PenL-Cys69Tyr as in the canonical ESBLs (Patel et al., 2017).