Aminoglycoside binding and catalysis specificity of aminoglycoside 2″-phosphotransferase IVa: A thermodynamic, structural and kinetic study

2.1Chemicals

All aminoglycosides, ATP and other chemicals were obtained from Sigma-Aldrich at the highest purity grade. For control experiments, the sulfate ions contained in the commercialized aminoglycoside powders were removed by Ba(OH)₂ treatment as described elsewhere [8]. However, the presence of sulfate did not affect the thermodynamic and kinetic parameters. Therefore, sulfate was not removed for subsequent experiments. Equimolar concentrations of MgCl₂ were added to ATP stock solutions. In the text, unless otherwise stated, ATP refers to MgATP. Aminoglycoside and ATP stock solutions were prepared in 50 mM Tris–HCl, 40 mM KCl and 1 mM free MgCl₂ and were stored at − 20 °C after adjusting pH to 7.5.

2.2Protein purification

Recombinant APH(2″)-IVa from Enterococcus casseliflavus was produced in Escherichia coli BL21 (DE3) transformed with pET15b plasmid encoding for APH(2″)-IVa with a 6His-tag in N-terminal. aph(2″)-IVa gene was the generous gift of Professor Vakulenko, Notre Dame, USA.

Two liters of Luria Broth media, LB, supplemented with 0.1 mg/mL of Ampicillin, was inoculated at a final OD_{600 nm} of 0.1 with a 100 mL overnight starting culture. Cells were grown at 37 °C with an agitation of 180 rpm. Protein expression was induced at the late mid log phase (OD_{600 nm} = 0.9) with 1 mM final concentration of IPTG. After an overnight incubation at 20 °C, cells were harvested by centrifugation and the pellet was solubilized in 40 mL of lysis buffer (50 mM NaH₂PO₄ at pH 8.0, 300 mM NaCl, 10 mM imidazole, 1 mM DTT and 1 tablet of complete™ EDTA-free protease inhibitor). Bacteria were lysed on ice by sonication. Cell debris were removed by centrifugation and ultracentrifugation steps before loading the supernatant on a HisTrap FF 5 mL column in an Aktä Purifier system (GE Healthcare). The enzyme was eluted from the column with lysis buffer complemented with a linear gradient up to 500 mM of imidazole. Pure fractions were pooled and the protein was concentrated to 30–50 mg/mL on a 30 kDa Centricon® device (Millipore). The elution buffer was exchanged by a storage buffer containing 50 mM Hepes at pH 7.5, 50% glycerol and 1 mM DTT. The protein was stored at − 20 °C. Typical production yield was 80 mg/L of culture.

To improve protein purity for the ITC experiments and crystallogenesis, a supplementary gel filtration step was carried out on a HiLoad 16/60 Superdex 75 column (GE Healthcare) in either 50 mM Tris–HCl pH 7.5, 40 mM KCl, 1 mM MgCl₂ for ITC or in 50 mM Hepes pH 7.5, 10 mM MgCl₂ for crystallization assays.

The catalytic activity of APH(2″)-IVa was independent of the presence of the 6His-tag, as controlled by comparing the steady state rates obtained with kanamycin A before and after cleavage of the tagged protein with thrombin (data not shown). The 6His-tag has also no effect on the crystallogenesis of the protein. Therefore, in all experiments the 6His-tag was kept at the N-terminal part of the enzyme.

2.3Isothermal titration calorimetry experiments

The enzyme was extensively dialyzed against buffer containing 50 mM Tris–HCl at pH 7.5, 40 mM KCl, 1 mM MgCl₂ and ligand solutions were prepared at 5 mM in the final dialysate with readjusting pH to 7.5. The enzyme was used at a final concentration of 150 μM. ITC experiments were carried out at 25 °C in a VP-ITC calorimeter (MicroCal, GE Healthcare). Titrations consisted of 40 injections of 5 μL of ligand every 300 s. The cell stirring speed was 300 rpm. To obtain reliable dissociation constants, the c-value, a unitless parameter obtained by the multiplication of the association constant and the total concentration of ligand binding sites, was kept between 1.8 and 100 for all titrations. Control runs were performed by titrating ligands to buffer, and the resulting background signal was subtracted from the corresponding experimental data. Experiments were carried out at 1 mM free Mg^{2 +}.

Binding and thermodynamic parameters K_a (association constant), ΔH (enthalpy change) and stoichiometry were obtained by nonlinear least-squares fitting of experimental data using a single-site binding model of the Origin software package (version 5.0) provided with the instrument. The free energy of binding (ΔG) and entropy change at 25 °C (ΔS) were obtained using the following equations:(1)ΔG=–RTlnKa(2)ΔG=ΔH–TΔS

The experiments were performed in triplicate (except with G418 and kanamycin B, in duplicate) and the error values given in the tables are standard deviations.

2.4Electrostatic potential and free energy calculations

Electrostatic potential surfaces were computed with the APBS program [9]. The charges and radius of all atoms were calculated using PDB2PQR software [10]. The linearized traditional Poisson–Boltzmann equation was solved with APBS using a cubic spline charge discretization and with dielectric constants of 2.0 for the solute and 78 for the solvent at 300 K. Electrostatic potential is represented by a positive and a negative isosurface at ± 10 K_bT/e (K_b, Boltzmann's constant; T, temperature and e, charge of an electron).

All molecular dynamics simulations were performed using the NAMD 2.10 software [11] in the isobaric–isothermal ensemble. The pressure (1 atm) and temperature (300 K) were kept constant using Langevin dynamics and Nosé–Hoover Langevin piston [12], [13]. All protein atoms were described by the CHARMM27 force field [14] while the aminoglycosides were parameterized using the paratool plugin implemented in VMD [15] by assigning both the atomic partial charges computed with Gaussian (by fitting the electrostatic potential surface, ESP) and atom types from CHARMM27. The crystal structures containing either kanamycin A (PDB: 4DFB) or tobramycin (PDB: 3SG8) were immersed in a rectangular water box (TIP3 model) with a 12 Å thickness and neutralized with three or four Na⁺ ions. As the crystal structure of APH(2″)-IVa·sisomicin was not fully solved or unsatisfactory, the complex was obtained by docking (GOLD 5.2 program, CCDC Software Limited) into the chain B of the 4DFB crystal structure. Kanamycin was first removed and used as a center for the search of binding modes (docking poses) of sisomicin by applying 50 runs of genetic algorithms. Solutions were classified according to their scores using the goldscore function. This complex was immersed in a water box as the two other aminoglycosides. The solvated systems were replicated in each direction using periodic boundary conditions. The short-range Lennard–Jones potential was smoothly truncated from 10 to 12 Å and the PME (Particle Mesh Ewald) algorithm [16] was used to calculate long-range electrostatics with a grid spacing of 1 Å. The potential energy of the molecular systems was minimized for 50,000 steps of conjugate gradient (time step of 1 fs). After a gradual heating from 0 to 300 K, the two systems were further equilibrated for 100,000 steps. To explore the “unbinding” of the aminoglycoside from the APH binding site, a reaction coordinate was defined as the distance separating the center of mass of each entity (antibiotic and protein). A biased force was applied using Colvars module of NAMD in order to estimate the free energy change along the reaction coordinates using the ABF method [17]. The distance separating the two centers of mass was initially 7.5 Å and forced to reach 27.5 Å. A width of 0.1 Å was selected using four short window (5 Å) simulations (6 ns each) and the final potential of mean force (PMF) was reconstructed from these separate windows. Simulations were carried out on IBM blade cluster and trajectories were analyzed using VMD.

2.5Crystal structure determination

Crystals of binary complexes of APH(2″)-IVa were grown at 18 °C using similar conditions to those previously described by Toth et al.[18]. Co-crystallization of the enzyme with either sisomicin or G418 (Geneticin) was carried out by mixing 1 μL of protein (6 mg/mL^{− 1}) and aminoglycoside (2.5 mM) in 50 mM Hepes pH 7.5, 10 mM MgCl₂ with 1 μL of reservoir solution composed of 12% PEG3350 (w/v), 50–75 mM ammonium citrate at pH 7.4 to 7.9.

Crystals of binary complex APH(2″)-IVa·sisomicin were grown by the hanging drop vapor diffusion method over a reservoir of 500 μL. After 2–3 days, crystals with approximate dimensions of 300 × 100 × 50 μm were observed. They were briefly immersed in the reservoir solution supplemented with 15% of glycerol as cryoprotectant before being flash-cooled in liquid nitrogen. Data were collected under cryogenic conditions on a Pilatus 6M-F detector at ESRF (Grenoble, France) on beamline ID23-1. Crystals of binary complex APH(2″)-IVa·G418 were grown by the sitting drop vapor diffusion method over a reservoir of 40 μL. The approximate size of crystals was 100 × 50 × 50 μm. Data collection was carried out in situ as described by Gelin et al.[19] in a 96-well CrystalQuickTM X plate (Greiner BioOne) on beamline BM30A-FIP at the ESRF, equipped with an ADSC Q315r CCD detector. Only the density obtained in chain A was sufficient enough to build G418.

Both structures were refined by molecular replacement using chain A of the homologous APH(2″)-IVa·ADP complex (PDB: 4N57) after removal of the ligand as search model. Data were processed and scaled with XDS [20] and SCALA [21]. The atomic models were rebuilt in Coot [22] and refined using Refmac [23] with 10 sub-segments per protein chain of torsion–libration–screw from the TLSMD server [24]. G418 and sisomicin restraints were generated with the PRODRG2 server [25]. The 2F_o − F_c omit maps were calculated with AutoBuild from the PHENIX package [26] after omitting the aminoglycoside molecules present in the asymmetric unit. Data collection and refinement statistics are shown in Table 1 and the structures were deposited in the protein databank (PDB: 5C4K and PDB: 5C4L). The B-factor values for G418 were large which may be explained by the high flexibility of the aminoglycoside in the binding site.

The structures were analyzed using the PyMOL Molecular Graphics System (version 1.3, Schrödinger, LLC). The rmsd between non-hydrogen atoms of different ligands was calculated using the pair fit command.

2.6Kinetic experiments

Unless otherwise stated, kinetic experiments were carried out in a buffer that contained 50 mM Tris, 40 mM KCl, 1 mM free MgCl₂ adjusted to pH 7.5 with HCl at 25 °C. The initial concentrations of aminoglycoside and ATP are indicated in the figure legends. In the case of experiments with CaATP, the buffer was MgCl₂-free.

Steady state kinetics were carried out in thermostatically controlled beakers as already described [27]. Unless otherwise stated, the enzyme concentration was 0.5 μM in the presence of various concentrations of aminoglycoside (5 to 100 μM). With amikacin and paromomycin, the absence of activity was confirmed at 5 μM APH(2″)-IVa. Reactions were initiated by adding a saturating concentration of ATP (2.5 mM) in a thermostatically controlled beaker containing APH(2″)-IVa and aminoglycoside. At regular intervals, a volume of 80 μL of the reaction mixture was sampled from the beaker and mixed in a tube containing 40 μL of perchloric acid at 10%. The precipitated protein was then removed by centrifugation at 4 °C, 20 min at 19,000 g. ADP and ATP were separated by HPLC (Waters) on an anion-exchange column (SAX-Partisphere, AIT France). The mobile phase was 200 mM ammonium phosphate–HCl buffer at pH 5.5 and 10% acetonitrile. A 100 μL volume of the quenched reaction mixture was diluted in 900 μL of buffer containing the mobile phase supplemented with 5 N KOH to readjust the pH to 5.5. Quantification of ADP and ATP was obtained by integrating HPLC peaks at 259 nm and their concentrations were reported relative to the enzyme concentration. Concentrations (in the quenched reaction mixtures) between 0.5 and 500 μM of ADP or ATP gave a linear and equivalent OD₂₅₉ response. For each injected sample, the concentration of ADP was calculated using the following equation, where [ATP]_i is the initial ATP concentration used in the experiment:(3)ADP=ADPpeakarea×ATPi/ADPpeakarea+ATPpeakarea

Transient kinetic experiments were carried out at 25 °C in a QFM-400 quench–flow (Bio-Logic, France) or in a SF-61 DX2 stopped–flow (TgK Scientific, UK). In the quench–flow method, APH(2″)-IVa, pre-incubated with the aminoglycoside, was mixed with ATP in the apparatus. The mixtures were aged for specific times (from 8 ms to several seconds), quenched in 10% perchloric acid and the ADP and ATP measured by HPLC, as described above. Because perchloric acid dissociates all non-covalent complexes, by this method, concentrations in the reaction mixtures of total ADP, that is to say enzyme-bound ADP plus free ADP, are obtained [27]. In the stopped–flow experiments, a free ADP fluorescent biosensor, MDCC-ParM [28], was included in the ATP solution. By this method, concentrations in the reaction mixtures of free ADP only are obtained. Experiments were carried out as previously described for APH(3′)-IIIa [27].

3.1Thermodynamic studies of aminoglycosides binding to APH(2″)-IVa

By isothermal titration calorimetry (ITC), we determined the thermodynamic parameters, and consequently the binding affinities, of aminoglycoside binding to APH(2″)-IVa. Although ITC has been used extensively to characterize aminoglycoside binding to APH(3′)-IIIa [29], [30], [31], [32] or to other aminoglycoside modifying enzymes [33], [34], [35], [36], [37], [38], [39], this work brings the first data for APH(2″)-IVa. Fig. S1 shows the binding isotherms and parameters obtained with APH(2″)-IVa and eight aminoglycosides, fitted with a single site model. Thermodynamic parameter values obtained with 4,6- or 4,5-disubstituted aminoglycosides are in Table 2. For all the aminoglycosides, the free energy changes, ΔG, were negative upon binding to APH(2″)-IVa, indicating that the formation of APH(2″)-IVa ∙ aminoglycoside complexes is favorable, even with 4,5-disubstituted aminoglycosides that are not substrates of this enzyme ([5] and confirmed below). Values of ΔG range from − 7.8 to − 5.6 kcal mol^{− 1}, corresponding to K_d of 2.3 to 82.2 μM.

Interestingly, there is a correlation between affinity and aminoglycoside molecular weight as shown in Fig. 2a. Excluding kanamycin A and amikacin, the two aminoglycosides having a hydroxyl instead of an amino group in C2′ position (discussed later) that showed the lowest affinities, the smallest aminoglycosides exhibited the tightest binding to APH(2″)-IVa. Another way to illustrate this observation is to exploit ligand efficiency, LE, a guide commonly used in drug discovery and lead optimization [40], [41]. Ligand efficiency is defined as the ratio of − ΔG to the number of non-hydrogen atoms of the ligand (N_CNO). As shown in Fig. 2b, ligand efficiency linearly depends on N_CNO with the largest aminoglycosides exhibiting the smallest ligand efficiencies. Consistently, there is a logarithmic dependence of K_d on ligand efficiency (Fig. 2c).

Does the presence of nucleotide affect APH(2″)-IVa affinity for aminoglycosides? With APH(3′)-IIIa, the formation of binary enzyme ∙ aminoglycoside and ternary enzyme ∙ ATP ∙ aminoglycoside complexes could be compared by the use of CaATP, because with it the enzyme is inactive [30]. It was not possible to assess this question with APH(2″)-IVa as CaATP is a substrate of this enzyme (Fig. 5a, see below). Instead, we used MgADP or MgAMP-PNP and showed that the pre-incubation of the enzyme with the nucleotide has no significant effect on its affinity for aminoglycoside or on binding thermodynamic properties. For example, with kanamycin A, the K_d values were 87 ± 8 μM in the presence of 870 μM MgADP, 103 ± 12 μM in the presence of 1 mM MgAMP-PNP and 80 ± 6 μM in the absence of nucleotide.

3.2Electrostatic potential and free energy change calculations

In order to determine the nature of the preponderant interactions governing the aminoglycoside binding to APH, the electrostatic potential surface of kanamycin A, tobramycin and sisomicin was computed (Fig. 3a to c) and showed obvious differences in their charge distribution. Indeed, for tobramycin and sisomicin, the supplementary positively charged amino group in 2′ position induces a very large positive surface compared to that of kanamycin A. This should be highly favorable to strong binding to APH(2″)-IVa, since the aminoglycoside binding site is fully negatively charged (Fig. 3d).

The binding free energy measured by ITC for the different aminoglycosides showed very small differences. Therefore, in order to explore thoroughly these small energy barriers, molecular dynamics simulations using the ABF method were carried out on several complexes. Free energy calculations and the resulting PMF (estimation of the energy required for ligand release) are shown in Fig. 3e.

As expected, the force required to unbind the aminoglycoside was similar for the three aminoglycosides tested. However, one can notice a higher PMF obtained in the case of tobramycin and sisomicin as compared to kanamycin A. These differences in terms of free energy are in good agreement with the values obtained by ITC. Indeed, the ΔG values of − 5.6, − 7.4 and − 7.8 kcal mol^{− 1} measured by ITC for kanamycin A, tobramycin and sisomicin were comparable to the PMF values which were 5.4, 8.2 and 8.1 kcal mol^{− 1}, respectively (values at the end of the simulation, with opposite sign to ΔG as unbinding is measured here). Comparison of such theoretical values should be taken with care, as the full binding process is not computed, but instead just the release of the aminoglycoside.

Interestingly, the dissociation of the aminoglycoside seems to be governed by the NH₃⁺ groups that link successively numerous Asp or Glu residues present at the periphery of the aminoglycoside binding site. This is shown by the different phases punctuating the PMF curves (Fig. 3e) and this may reflect the entry or the exit pathway of the aminoglycoside.

3.3Structure determination of APH(2″)-IVa ∙ aminoglycoside binary complexes

In order to characterize interactions between enzyme and ligands, we determined the crystal structures of APH(2″)-IVa bound to aminoglycosides with different affinities.

A crystal structure of APH(2″)-IVa in complex with sisomicin, that has the lowest K_d of the eight aminoglycosides tested here, has been solved at a resolution of 2.35 Å (Fig. S2). However, the binding position of the aminoglycoside in chain B is probably induced by the crystal packing. Indeed, sisomicin is stabilized by an interaction with the backbone of Ser136, which belongs to a symmetric protein of the unit cell. This feature shifts the position of the aminoglycoside from its presumed physiologically relevant orientation. In chain A, the electron density obtained permits only to build the ring A of sisomicin. The rest of the molecule is probably too flexible to be visualized. In both chains, sisomicin is stabilized by salt bridges occurring between the side chains of Glu235 and Glu268 and the 6′-NH₃⁺ of the aminoglycoside and by hydrophobic interaction between Trp271 and ring A.

We solved the structure at a resolution of 3.05 Å of APH(2″)-IVa in complex with G418 (geneticin), that has an intermediate affinity for APH(2″)-IVa. Comparison of the crystal structures of binary complexes with tobramycin, kanamycin A and G418 (PDB: 3SG8, 4DFB and 5C4K, respectively) shows that the position of the three aminoglycosides is highly similar with an rmsd (between ligands non-hydrogen atoms) of 0.59 Å between kanamycin A and G418 and of 0.36 Å between tobramycin and G418 (Fig. 4).

As already observed in the previous published structures, Glu235 stabilizes rings A and B of the aminoglycoside (Fig. 4a, d, g). Interestingly, the interaction between Glu238 and the ring A of aminoglycoside is maintained even if the substituent in the 6′-position of G418 is different (Fig. 4b, e, h). However, its interaction with ring B is, in turn, abolished. Similar to kanamycin A and tobramycin structures, ring C is maintained by an interaction with Asp197. In general, the APH(2″)-IVa ∙ G418 complex is characterized by less interactions than the binary complex with tobramycin, which may be at the origin of the weaker affinity of G418 compared to tobramycin.

3.4Steady state and transient kinetics by direct assessment of ADP production

In order to see if the differences in the binding affinity of aminoglycosides affect the enzymatic reaction rate constants, we carried out steady state and transient kinetic experiments. Here, we used a direct method for separating and quantifying ADP and ATP by HPLC in APH reaction mixtures that had been quenched at different times in acid. With this method, concentrations of ADP down to 0.5 μM can be measured.

In addition to its phosphotransferase activity, in the absence of aminoglycoside, APH(2″)-IVa has a low ATPase activity of 0.014 s^{− 1} with 2 mM MgATP (Fig. 5a). Its phosphotransferase activity in the presence of 2 mM CaATP as phospho-donor and kanamycin A as acceptor was low compared to that with MgATP, but not negligible (0.092 and 0.84 s^{− 1}, respectively). Typical time courses of the phosphotransferase activity at two different initial concentrations of MgATP are illustrated in Fig. 5b. It is noteworthy that in each case the reaction went to near completion, which is to say that ATP was completely converted into ADP. Steady state rates were obtained from the initial linear portions (Fig. 5c).

The steady state parameters, k_cat and K_m, obtained with 8 aminoglycosides and with MgATP as phosphate donor, are summarized in Table 3. As can be seen, with the exception of tobramycin, our values are in reasonable agreement with those of Toth et al. using the coupled enzyme system [5]. With tobramycin, Toth et al. reported a K_m of 1.5 μM whereas we obtained 13.2 μM. We are unable to explain this difference, but this could potentially be a consequence from using a coupled assay versus a direct method such as the quench–flow assay.

The variation of the ratio K_d/K_m between aminoglycosides is striking — from 20 with kanamycin A to 0.3 with tobramycin. Are these different ratios explained by differences in the kinetics of the transient formation of enzyme ∙ ADP complex? Using the same quenching/sampling method and a rapid quench–flow apparatus, we could measure transient kinetics of total ADP production catalyzed by APH(2″)-IVa with kanamycin A, tobramycin, G418 or sisomicin and ATP, as illustrated in Fig. 6 (circles).

Except the nature of the antibiotic, the four experiments were carried out under identical conditions. Each time course for the production of total ADP (that is to say, free ADP plus enzyme-bound ADP which dissociates with the acid quenching) consists of two phases: a burst of ADP followed by the steady state. These traces suggest that an ADP-containing complex or free ADP is rapidly formed compared to the steady state rate of the reaction.

In parallel, and in the same experimental conditions, we used fluorescence stopped–flow with an ADP biosensor in order to measure the kinetics of free ADP production. The reaction was followed by measuring the ADP-induced fluorescence increase of MDCC-ParM as a function of time. Results are shown in Fig. 6 as continuous lines. Interestingly, the time courses of free ADP production showed a lag preceding the steady state, the amplitude and rate constant of which are similar to those determined for the burst of total phosphate. This suggests that free ADP is not formed rapidly compared to the steady state rate of the reaction, but instead an ADP-containing complex — the lag is a reflection of the time needed to build up this intermediate. Values for the amplitudes and rate constants of the burst/lag phases and the steady state rate constants are summarized in Table 4.

The transient kinetic parameters determined here did not show large differences between the aminoglycosides, although there may be a correlation between k_cat and the transient rate constants (k_lag and k_burst). Also, transient rate constants were not very large compared to the steady state rate constants. This is especially true for tobramycin with only a 2 to 3-fold difference. Altogether, these results suggest that with all aminoglycosides, even those with a low affinity for the enzyme, the product release is the rate limiting step of the reaction, but the phosphotransfer step, that leads to the production of ADP-containing complex, may also contribute.

4.1Comparison of thermodynamic parameters of different aminoglycoside phosphotransferases

APH(2″)-IVa is specific for 4,6-disubstituted aminoglycosides as phosphate acceptors. In comparison, APH(3′)-IIIa has a broader specificity for antibiotics — it can use 4,5- as well as 4,6-disubstituted aminoglycosides. It is therefore interesting to compare the aminoglycoside affinities of APH(3′)-IIIa and APH(2″)-IVa, two enzymes that catalyze the same reaction but on different positions on the aminoglycoside and with a different substrate specificity. The K_d values determined here for 8 aminoglycosides with APH(2″)-IVa are comparable to those measured by Özen and Serpersu [30] with APH(3′)-IIIa (K_d between 0.3 and 92.6 μM). With both phosphotransferases, amikacin shows a much lower affinity relative to other aminoglycosides, probably due to its bulky substituent at position N1.

Another difference between these two phosphotransferases is that the affinity for aminoglycosides depends on the antibiotic size with APH(2″)-IVa, but not with APH(3′)-IIIa. With APH(2″)-IVa, this correlation may be the effect of entropic contributions where longer substrates are restricted in degrees of freedom in the binding site, without proper compensation with enthalpic terms, since the additional residues of the substrate are not interacting strongly with the enzyme. This suggests that the binding site on APH(2″)-IVa is less extensible and dynamic than that of APH(3′)-IIIa. Indeed, molecular dynamics simulations [42] showed that a highly flexible loop of APH(3′)-IIIa (V154-K166), positioned over the antibiotic binding pocket and colored in red in Fig. 7a, becomes more constrained in the presence of kanamycin A, with which it interacts. The flexibility of this loop may explain the ability of the APH(3′)-IIIa to accommodate aminoglycosides with various sizes, contrary to the APH(2″)-IVa. Indeed, this loop is not present in the latter (Fig. 7b), which probably limits the binding of 4,5-disubstituted aminoglycosides or, at least, dramatically decreases their affinity for the protein.

However, this alone does not explain the lack of catalytic activity of APH(2″)-IVa with the 4,5-disubstituted aminoglycosides. Instead, unfavorable positioning of the 2″-site may be the reason. This is similar to what is observed with the aminoglycoside nucleotidyltransferase(2″)-Ia, to which neomycins bind tighter but are not substrates of the enzyme. In this case, the 2″-site becomes too distant to the α-phosphate group for a direct nucleophilic attack [35]. Thus, unfavorable orientation of the 2″-site in the more constrained binding site on APH(2″)-IVa may be correlated with its inability to catalyze phosphotransfer with 4,5-disubstituted aminoglycosides.

4.2Comparison of the interactions between APH(2″)-IVa and aminoglycoside in the crystal structures

Five APH(2″)-IVa ∙ aminoglycoside binary complexes have been solved by X-ray crystallography: two in the presence of kanamycin A (PDB: 4DFB[43] and PDB: 3SG9[6]), one with tobramycin (PDB: 3SG8[6]), one with G418 (PDB: 5C4K, here) and one with a partially solved sisomicin (PDB: 5C4L, here).

In the structures determined by Shi et al.[6], the superposition of the two different aminoglycosides reveals similar binding positions, except for ring C, coplanar to ring A for kanamycin A and lightly tilted for tobramycin (Fig. 8a). High resolution electron densities were visible for the aminoglycoside in the tobramycin structure (3SG8), see Fig. 8b. In contrast, we found that the density obtained for both chains of APH(2″)-IVa ∙ kanamycin A complex (3SG9) was not sufficient to completely build the ligand (Fig. 8c), resulting in much larger B-factor values than the surrounded residues. Nevertheless, the superposition of the APH(2″)-IVa ∙ tobramycin structure (3SG8) and that with kanamycin A solved by Shakya et al. (4DFB) shows identical positioning of aminoglycosides in the binding site (Fig. 8d) and very slight modifications of interacting amino acid side chains. The electron density obtained for 4DFB fully covers the aminoglycoside substrate (Fig. 8e). We consequently decided to use the 4DFB structure for following discussion.

Comparing the APH(2″)-IVa ∙ tobramycin (Fig. 8b) and the APH(2″)-IVa ∙ kanamycin A (Fig. 8e) binary structures, most interactions between the aminoglycoside and protein residues are conserved. Nevertheless, there may be a supplementary H-bond between Glu268 and the 3′-OH of kanamycin A which is not present in the tobramycin structure, not substituted in 3′ (Fig. 8b and e). However, this interaction may be weak because the distance measured between the 3′-O of kanamycin and the δ-O of Glu268 is rather long for a H-bond (3.57 Å) in chain B and it does not exist in the chain A, where Glu268 occupies an alternative position as shown in Fig. 4a. More importantly, the H-bond formed between Glu235 and 2′-OH of kanamycin A is replaced by a salt bridge with 2′-NH₃⁺ of tobramycin or G418. As a result, the affinity of the apoprotein is 20 times greater for tobramycin than for kanamycin A and for amikacin which also lacks a 2′-NH₃⁺.

4.3Effect of different substituents of aminoglycosides on APH(2″)-IVa affinity and catalytic activity

In comparison with other aminoglycosides, the different substituents in positions C3″, C4″ and C5″ of gentamicin, G418 and sisomicin do not alter significantly the affinity for APH(2″)-IVa. We note that these three aminoglycosides have the lowest k_cat and K_m values of the eight tested. Since they also present modifications on ring A, it is difficult to draw clear conclusions concerning the effect of ring C substituents. However, the crystal structure of the APH(2″)-IVa ∙ G418 complex shows that the hydroxyl and methyl groups on the C6′ of G418 do not create novel interactions with the protein (Fig. 4h and i). The absence of a primary amine on C6′ of ring A seems to affect the binding affinity of aminoglycosides for APH. Indeed, G418 has a weaker affinity for the apoprotein than gentamicin, with a secondary amine at C6′, and even weaker than sisomicin with a primary amine. The crystallographic structure obtained in the presence of sisomicin shows that the aminoglycoside is stabilized by salt bridges between 6′-NH₃⁺ and Glu235 and Glu268 (Fig. S2). An interaction of 2′-NH₃⁺ of sisomicin with Asn228 may also play a role. Even if these data are incomplete or biased by the crystal packing, these results highlight the importance of 2′- and 6′-NH₃⁺ groups in the binding of aminoglycosides, explaining the very high affinities obtained by ITC.

Docking of sisomicin predicts a similar position of binding compared to the crystal structure (Fig. S2e and h). Indeed, the 2′- and 6′-amines are still involved in crucial interactions with Glu235, Glu239 and Glu268. However, the predicted binding mode shows that the 2′-NH₃⁺ group of sisomicin is lightly shifted towards Glu235, creating more interactions with the glutamate residues. The presence of a double bond which makes the A ring more rigid should drastically reduce the conformational freedom of the ring and therefore may stabilize the binding of the aminoglycoside by favoring a particular ring orientation.

The difference in position 1 of amikacin compared with the other aminoglycosides results in a dramatic lowering of its catalytic efficiency. However, this different substituent does not affect much its binding to the apoenzyme as its K_d is very similar to that of kanamycin A (82.2 μM and 79.9 μM, respectively). The bulky group in the C1 position of amikacin may induce steric hindrance which could be deleterious for the phosphotransfer on the 2″-OH.

4.4Interpretation of APH(2″)-IVa transient kinetic data

As a first approximation, we interpret our data via Scheme 1 in which APH represents APH(2″)-IVa, AMG aminoglycoside and AMG-P the corresponding 2″-phosphorylated aminoglycoside.

The steps 1 to 5 are described by the forward rate constants k_{+ i} and the backward rate constants k_{− i}. For simplicity, in Scheme 1 substrate binding and product release steps are shown as single steps, but they are most probably composed of two steps: a rapid equilibrium leading to the formation of the collision complex, followed by a conformational change. Toth et al.[5] proposed that APH(2″)-IVa catalyzes the phosphotransfer by a random mechanism, that cannot exclude the possibility that the binding of one substrate affects the binding of the other (that is to say K₁ ≠ K₂′ and K₁′ ≠ K₂). Isothermal titration calorimetry experiments carried out here gave information on step 1. The K_d for aminoglycoside represents the equilibrium constant of aminoglycoside binding step, that is K₁ = k_{− 1}/k_{+ 1}. In our experimental conditions, we measured a 20-fold larger K_d for kanamycin A compared to tobramycin. Now, unexpectedly, under catalytic conditions, that is in the presence of ATP, the enzyme had a 3.3-fold lower K_m for kanamycin A compared with tobramycin. Segel proposed that, if the ratio K_m/K_d for a given substrate is < 1, there is substrate synergy by which the binding of one substrate increased the affinity for the second [44]. On the other hand, if the ratio is > 1, there is substrate antagonism (for a recent discussion, see [45]). With the caveat that a K_m is not a simple dissociation constant for the formation of a ternary enzyme complex, this suggests that the presence of ATP may have different effects on the interactions of kanamycin A and tobramycin with the enzyme, although we showed that ADP or AMP-PNP did not.

Alternatively, it could be that for efficient catalysis, the aminoglycoside must not be too strongly bound to the phosphotransferase to allow the rapid release of the phosphorylated antibiotic. Thus, if the phosphorylated aminoglycoside interacts tightly with the enzyme, it could slow down its release kinetics. A slow release would manifest itself by the accumulation of intermediates that involve the products. We confirmed this by transient kinetics: starting with APH ∙ AMG plus ATP, the time course of total ADP formation was biphasic with a rapid burst of ADP preceding the steady state rate whereas that of free ADP was biphasic with a lag instead of the burst. These observations suggest that intermediates containing ADP (APH ∙ AMG-P ∙ ADP and/or APH ∙ ADP) accumulate in the steady state. Therefore, we can conclude that with APH(2″)-IVa, steps 2 and 3 are fast compared to step 4 or 5′. It implies that ADP release, step 4 or 5′, is rate-limiting. We note that with three different APH(3′), a product release step is also rate limiting [27], [46], [47].

However, the observation that the transient rate constants measured here were not much larger than steady state rate constants, suggests that step 3 may not be much faster than the product release steps. To evaluate the relative contribution of each step on the transient and steady state kinetics, we intend to study the dependence of the transient amplitudes and rate constants as a function of the aminoglycoside or ATP concentrations, and starting from the apoenzyme or binary complexes.

Shi and Berghuis published crystal structures of APH(2″)-IVa in complex with adenosine and guanosine that provide explanation for the peculiar dual nucleotide specificity of this enzyme [48]. Therefore, it would be interesting to compare the transient kinetic rate constants obtained using MgATP or MgGTP as phosphate donor, to determine if the phosphotransfer step is similarly fast with the two nucleotides.