In this work, we have performed both modeling and biochemical experiments in order to gain a better understanding of the mechanism of the lysyl-tRNA synthetase, LysU. An important aspect is that an initial modelling study provided a hypothesis of asymmetry that was subsequently confirmed by experiment. Further modelling studies have extended this hypothesis and provided important insights into the functioning of LysU.
The utility of the MM-PBSA method in predicting association energies and understanding binding processes in LysU has been demonstrated. Post-processing free energy methods have gained in popularity due to their speed advantage over the more rigorous free energy perturbation or thermodynamic integration approaches. In fact, the traditional techniques are not applicable to the present problem – the calculation of ATP binding free energies in a dimeric enzyme. Conventional methods rely on the calculation of relative free energies of two (or more) ligands binding at the same site, and cannot be applied to the same ligand binding at two different sites. Furthermore, it is not feasible to include the entire dimer in the more rigorous approaches, due to the large size of LysU. Previous thermodynamic integration calculations on asparagine and aspartic acid binding to the dimeric class II aspartyl-tRNA synthetase (AspRS)  have been limited to a single active site, with stochastic boundary conditions preventing movement in the rest of the enzyme, and hence any cooperative behaviour. In contrast, the simulations reported here are performed on the fully solvated dimer, with no restraints. This affords greater flexibility to the modelled enzyme and permits the observation of asymmetry and cooperative behaviour.
Asparagine binding by mutant AspRS has also been studied by approximate free energy calculations, using the Poisson-Boltzmann free energy method (PBFE) . Again the study focused upon binding at a single active site. The PBFE method is similar to MM-PBSA but neglects non-electrostatic interactions to the binding free energy. As has been noted for the FO model, the electrostatic terms are very similar, and binding free energy differences arise largely from the differences in the van der Waals contributions. The omission of this relatively inexpensive term may result in incorrect ranking of binding affinities in PBFE calculations.
At the outset of this study, no applications of the MM-PBSA method to a metalloprotein had been reported, however Donini and Kollman have reported studies on the zinc-dependent matrix metalloproteinases  in which the difficulties associated with the inclusion of metal ions are highlighted. In this work, binding energies were calculated in reasonable agreement with experiment, in the presence of three magnesium ions.
An ordered binding mechanism
A combination of isothermal titration calorimetry, fluorescence and circular dichroism has revealed the order and stoichiometries of the binding events and allowed the determination of associated thermodynamic parameters. In union with previous crystallographic data on the apo and lysine bound LysS structures  as well as the ATP, AMPPCP and adenylate bound LysU structures [13, 14, 16], this further work provides a comprehensive description of the substrate binding mechanism. Lysine appears to bind before the nucleotide in an ordered magnesium dependent process. The requirement for an ordered mechanism is suggested by the large conformational change that accompanies lysine binding, as measured by tryptophan fluorescence and CD. LysU required the presence of lysine to crystallize , however detailed changes brought about by lysine binding have been obtained by a comparison of the apo and lysine bound LysS crystal structures . As the overall structure of LysS is very similar to that of the highly homologous LysU, a common mechanism may be presumed. Onesti et al have observed that lysine binds in an induced fit mechanism, triggering a reorientation of the four helix bundle (H10 to H13) of the insertion domain. As W364 is a residue within the rotating insertion domain, it is likely to be responsible for the fluorescence changes observed during the binding of lysine to LysU. The structural studies also show that binding of lysine imposes significant order on the flipping loop (yellow in Figure 1) and the loop between two helices H14 and H15 of the insertion domain helix-turn-helix motif (shown in purple in Figure 1). The overall effect is to close up the active site around lysine . Similar observations have been reported for the Thermus thermophilus LysRS . These changes are likely responsible for the loss of entropy as measured by isothermal calorimetry, which shows the process to be enthalpically driven. A stoichiometry of two lysines per dimer is observed by ITC in agreement with the fluorescence data. Whilst the LysU and LysS structures do not reveal any magnesium ions in their lysine-bound states, our experiments have shown that magnesium is required for lysine binding.
Binding order appears to be enzyme specific, as it is not conserved even within the LysRS , however a recent series of papers has elucidated the mechanism of the Bacillus stearothermophilus lysyl-tRNA synthetase [29–31]. As in our experiments, it was observed that lysine binding causes a significant change in fluorescence whilst ATP only causes a small change that was not quantifiable. An ordered binding mechanism was found, with two moles of amino acid binding prior to ATP.
Asymmetry in nucleotide binding
Nucleotide binding elicited no change in intrinsic tryptophan fluorescence, and by inference no large structural changes. This is in agreement with the published crystal structures of the LysU ATP or AMPPCP bound states which show the only significant effect of nucleotide binding is to impart order on the motif 2 loop . These structures also show that there is little difference between the LysU-lysine-AMPPCP and the LysU-lysine-ATP states. This therefore supports our argument that AMPPCP is a good substitute for ATP in studying the binding reaction. The same analogue has been used in the ITC experiments described here, where it was not possible to use ATP due to hydrolysis and Ap4A synthesis under the same conditions. Calorimetry established the nucleotide binding process to be both enthalpically and entropically favourable, but perhaps of more interest is the stoichiometry of the nucleotide binding which was found to bind in the ratio of 1 per dimer. This important result was not anticipated from any previous biochemical data of LysU, but is compatible with the asymmetry observed in molecular dynamics simulations (Hughes et al, manuscript in preparation).
Other class II synthetases display asymmetric binding behaviour or half-of-the-sites activity. In the B. stearothermophilus LysRS only one mole of aminoacyl adenylate is formed per dimer, despite the presence of two moles of lysine. The functional asymmetry is therefore acquired after amino acid binding. Yeast LysRS, another homodimer, also exhibits half-of-the-sites binding with respect to aminoacyl adenylate as well as tRNALys . AspRS, from bakers yeast, shows cooperative behaviour, and has two different ATP binding constants that differ by several orders of magnitude (μM and mM) . No evidence for asymmetry or differential binding behaviour can be found from crystallographic studies of the lysyl-tRNA synthetase. This may be due to the effects of crystal packing, or simply that under conditions of high ATP concentrations, binding at the weaker ATP site forces symmetry on the molecule that is not relevant to its in vivo mechanism.
Molecular dynamics based free energy calculations also highlight the different binding characters of the two monomers. Structures from the previous simulations (the FO model) have been the subject of free energy calculations described here. The MM-PBSA method has been used to calculate ATP binding free energies along the trajectory.
A clear difference in calculated binding affinities has been observed with only one site binding strongly. The absolute free energies reported here are approximate due to the uncertainty in the solute entropy term. A further compounding feature is that the experimental value (-8.068 ± 0.321 kcal/mol) was measured for AMPPCP and not ATP. It has already been established that AMPPCP is a good analogue of the bound ATP hence binding affinities are expected to be in the same range. Supporting this assumption, a dissociation constant corresponding to a ΔGbindingof ca -7 kcal/mol for ATP association to B. stearothermophilus LysRS has been reported .
The large difference between the calculated binding affinities (ca 10 kcal/mol) has been examined by comparing ΔG*binding values which are independent of the solute entropy and hence more accurate. The difference in binding energies between the two sites of the FO model is statistically and biologically significant. Binding at the second site was found to be more favourable. In conjunction with the experimental study, these findings led us to propose that LysU may operate in a negatively cooperative manner.
Several mechanisms have been proposed to explain half-of-the-sites activity or strong negative cooperativity . The choice of functional active site can be random, or guided. In the flip-flop model, proposed by Lasdunski, binding at the first subunit reduces the affinity of the second subunit for the same ligand . Catalysis at the first site revives the binding affinity of the second site. The binding energy of the second ligand can be used to facilitate further catalysis at the first site, or help in product release. The "alternate site" model described by Boyers is a more general model with the same principle of ligand-induced subunit cooperation being an integral part of the catalytic process . Alternatively, a ligand-induced conformational change could lead to repeated utilisation of only one site .
The molecular dynamics studies have been carried out with ATP present in both active sites, and indicate only a preference for one site, furthermore the preferred binding region appears to be random. To further explore the possibility of cooperative behaviour, two half-of-the-sites models have been simulated, by removing ATP from a single monomer of the FO model. The binding energies of both models converge on the same mean despite the HS1 model being the initially weak site, as it was in the FO simulation. Similarity between this mean and the affinity of the strong binding site of the FO model indicates that the full binding potential can only be realised at one of the two active sites. This was seen as evidence for ligand induced anti-cooperative mechanism. Further evidence for negative cooperativity has been found in the time evolution of binding energies or gas phase interaction energies of the FO model. The mobile motif 2 loop residue Arg 269' was seen to enter the active site of the second monomer and interact with ATP for a period. During this interval, the interaction energy between ATP and Arg 269 is significantly enhanced in site 2 and weakened in site 1. Somehow, improved binding at one active site results in weaker binding at another active site 30 Å away. This behaviour is strongly reminiscent of an alternate sites model, especially in light of the role that Arg 269 is postulated to play in the removal of pyrophosphate from the active site once the first reaction has occurred.
An alternating sites mechanism, and half-of-the-sites activity, requires a dimeric (or tetrameric) configuration. In general, the class I synthetases are monomeric while most class II enzymes are dimers, heterotetramers or tetramers . Two of the three conserved motifs in class II enzymes (motifs 1 and 2) are located at the dimer interface, and dimerisation was shown to be required for activity in ProRS and AspRS . Several dimeric class I synthetases also show evidence of half-of-the-sites activity in the adenylation reaction. Perhaps the earliest and best known example is the tyrosyl-tRNA synthetase (TyrRS) from B. stearothermophilus, which binds one mol of tyrosine and forms only one mol of adenylate per dimer . A second mol of tyrosine and ATP only binds after the tyrosyl adenylate has formed, hence an interacting sites mechanism was proposed for TyrRS activity . The crystal structure of TyrRS is symmetric  and tyrosine is present in both sites of the E:Tyr complex . To resolve this apparent discrepancy, Ward and Fersht created truncated mutant heterodimers, which showed that TyrRS is asymmetric in solution, and that this asymmetry is frozen-in and randomly distributed between active sites with no detectable interconversion over several minutes . It was concluded that the enzyme asymmetry, rather than negative cooperativity, was responsible for the observed half-of-the-sites activity . Given the nature of the mutations, which involved removing an entire anticodon binding domain of the dimer, it is plausible that the lack of interchange between sites is due to the removal of residues involved in a cooperativity pathway.
The closely related tryptophanyl-tRNA synthetase from beef pancreas is negatively cooperative in the formation of the enzyme-adenylate complex . The second mol of adenylate was shown to form two orders of magnitude more slowly. In the presence of pyrophosphatase (or tRNATrp), two mol of adenylate are formed without any apparent cooperativity. It was concluded that the anticooperativity was due to the presence of pyrophosphate in the first active site. This was an interesting finding, as it implied that asymmetry could be "trapped" in an alternating sites mechanism. Many kinetics experiments on aaRS are carried out in the presence of inorganic pyrophosphatase or tRNA, and would therefore not show negative cooperativity. In our ITC experiments on LysU a non-hydrolysable ATP analogue was used, allowing the hidden asymmetry present in the mechanism to be uncovered. Similarly in the molecular dynamics simulations, no reaction can occur and the pyrophosphate cannot diffuse from the active site. In a recent paper, Retailleau et al conclude that there is strong crystallographic evidence that the B.s. TrpRS is an asymmetric dimer under most circumstances and suggest that a second ATP binds with a much lower affinity .
Possible functional roles of negative cooperativity and a postulated alternating sites mechanism
Functional asymmetry is a fairly common feature of the dimeric aaRSs, therefore it is not unreasonable to assume that it might play an important role. Several reasons have been proposed for active sites to interact in a negatively cooperative manner. One advantage would be to increase reactivity at one subunit  or to regulate substrate binding in multimeric proteins . Another function proposed is more specific to aminoacyl-tRNA synthetases: interacting active sites might act as a safeguard against errors in genetic translation [24, 41]. Qiu et al have suggested that cooperativity may be necessary for efficient product dissociation and release in aaRS .
We propose that LysU might operate in the following alternating sites mechanism: initial ATP binding to the first site causes a closing up of that cleft and prevents strong binding at the second site. Once adenylate formation and pyrophosphate dissociation has occurred at site one, ATP binding can occur strongly at site 2. In this mechanism, a conformational change or release of strain caused by pyrophosphate dissociation from one monomer is transmitted across the dimer interface to allow strong binding at the other monomer. Reaction followed by release of pyrophosphate at this site in turn revives the ATP binding affinity of the first. Such a mechanism would account for the half-of-sites binding observed by experiment, and is also consistent with the negative cooperativity mechanism proposed for TrpRS . It is also relevant that in the molecular dynamics simulations Arg 269 only enters the strongly bound ATP site of the FO model. The simulated models were examined to see how this information could be transferred. The most direct pathway between the two active sites is through the motif 2 residues Arg 262 and Phe 261 of one monomer across the dimer interface to their symmetry equivalent Phe 261' and Arg 262' however given the nature of the conformational changes observed in the half-sites models a more distributed pathway can be proposed. The movement of the motif 2 loop, flipping loop and insertion domain elements in monomer 1 (illustrated in Figure 8) all serve to close up the active site slightly. This suggests some release of strain upon the removal of ATP from the second site. In closing up the reaction cleft of the first monomer, a shift of the motif 2 loop could be transmitted via the C2'-C3' loop, to strand C2'. From here, the signal could be transmitted across to strand C2, and hence to the C2-C3 loop above the motif 2 loop of the second monomer, as well as to the flipping loop of the second monomer, which is just downstream of strand C2'. The flipping loop is in close proximity to the twisted helix-loop-helix motif. This causes the helix bundle to rotate away from the active site in a concerted manner. In this way, the closing of one active site around ATP is coupled to the opening up of the site in the other subunit, and hence a reduced binding affinity.
Such a mechanism could improve catalytic efficiency in two ways. The same conformational changes that open and close the active sites could prime a particular monomer for tRNA binding and release. Concerted conformational changes also ensure the correct order of substrate binding and reaction. It is apparent that cooperative structural dynamics are an important feature of aminoacyl-tRNA synthetase activity.