A conserved histidine in switch-II of EF-G moderates release of inorganic phosphate

Elongation factor G (EF-G), a translational GTPase responsible for tRNA-mRNA translocation possesses a conserved histidine (H91 in Escherichia coli) at the apex of switch-II, which has been implicated in GTPase activation and GTP hydrolysis. While H91A, H91R and H91E mutants showed different degrees of defect in ribosome associated GTP hydrolysis, H91Q behaved like the WT. However, all these mutants, including H91Q, are much more defective in inorganic phosphate (Pi) release, thereby suggesting that H91 facilitates Pi release. In crystal structures of the ribosome bound EF-G•GTP a tight coupling between H91 and the γ-phosphate of GTP can be seen. Following GTP hydrolysis, H91 flips ~140° in the opposite direction, probably with Pi still coupled to it. This, we suggest, promotes Pi to detach from GDP and reach the inter-domain space of EF-G, which constitutes an exit path for the Pi. Molecular dynamics simulations are consistent with this hypothesis and demonstrate a vital role of an Mg2+ ion in the process.

Intrinsic GTP hydrolysis by EF-G is unaffected by the H91 and F94 mutations. We studied the spontaneous GTPase activity of the EF-G variants by monitoring the time course of GTP hydrolysis in the absence of 70S ribosomes. For that, EF-G in excess was mixed with [ 3 H]GTP at 37 °C and the reaction was quenched with 25% formic acid at different time points. The amount of GDP formed was estimated from the ratio of the peak area of [ 3 H]GDP and [ 3 H]GTP as separated on a MonoQ column coupled to HPLC. The rate constant for the intrinsic GTPase activity of WT EF-G was estimated as 0.013 ± 0.003 s −1 , in close agreement with earlier reports 13,26 . Interestingly, none of the mutations affected the rate of intrinsic GTP hydrolysis. The rate constants for all EF-G variants were estimated as 0.014 ± 0.003 s −1 for F94L, 0.012 ± 0.001 s −1 for H91A, 0.014 ± 0.004 s −1 for H91E, 0.02 ± 0.006 s −1 for H91Q and 0.019 ± 0.004 s −1 for H91R (Fig. 2a, Table 1). Thus, in the absence of ribosome, GTP hydrolysis by EF-G does not involve H91. This is highly similar to a recent result with H84 mutations in EF-Tu, which did not show any defect in intrinsic GTPase activity 29 . H91 mutated EF-Gs show different degrees of defect in ribosome-stimulated GTP hydrolysis. GTP hydrolysis experiments in the presence of 70S ribosomes were performed in a quench-flow instrument. As shown in earlier reports, the GTPase activity of EF-G was significantly stimulated by the ribosome 13,26,30 . The WT EF-G hydrolyzed GTP on the ribosome with a single turnover rate constant of k GTP WT = 202 ± 29 s −1 , which is about 15 000 times larger than the rate constant of intrinsic GTP hydrolysis, as also reported earlier 31 . Mutants H91Q (k GTP H91Q = 174 ± 12 s −1 ) and the control mutation F94L (k GTP F94L = 170 ± 28 s −1 ) hydrolyzed GTP with rate constants comparable to that of WT EF-G (Fig. 2b, Table 1). In contrast, the H91A, H91E and H91R EF-Gs were defective in GTP hydrolysis, although the degree of defect varied. Mutants H91A and H91R showed a seven fold decrease in the rate constant compared to the WT (k GTP H91A = 28 ± 4.5 s −1 and k GTP H91R = 27 ± 2.5 s −1 ), while H91E showed a rate reduction by a factor of 100 (k GTP H91E = 1.9 ± 0.3 s −1 ) (Fig. 2b, Table 1). The mean time analysis (i.e. inverse of rate constants) shows that compared to 5 ± 0.7 ms for the WT, the H91A and H91R take 35 ± 5.6 ms and 37 ± 3.2 ms respectively, and H91E takes 520 ± 82 ms to hydrolyze one GTP on the ribosome (Fig. 2d). Since the H91 mutant EF-Gs show similar affinity to GTP as the WT, the defect in ribosome stimulated GTP hydrolysis must have originated from their interaction with the ribosome. H91 mutants show larger defects in Pi release than in GTP hydrolysis. Single turnover Pi release was studied using stopped-flow, with MDCC labelled phosphate binding protein (PBP-MDCC), the fluorescence of which increases instantaneously upon Pi binding 32,33 . In the presence of the ribosome, WT EF-G released Pi with a rate constant of k Pi WT = 28 ± 4.7 s −1 (Fig. 2c, Table 1), which is much slower than GTP hydrolysis. In comparison to the WT, except for the control mutation F94L, which did not show any defect in Pi release (k Pi F94L = 21 ± 3.5 s −1 ), all the H91 mutants were much slower in Pi release (Fig. 2c, Table 1). The H91Q mutant, which did not show any defect in GTP hydrolysis, was over 20fold slower than WT EF-G in releasing Pi (k Pi H91Q = 1.2 ± 0.16 s −1 ). The other two mutants, H91A and H91E released Pi with rate constants k Pi H91A = 0.32 ± 0.04 s −1 and k Pi H91E 0.25 ± 0.02 s −1 respectively. Pi release by H91R EF-G could not be measured. This is probably due to very slow release of Pi, which is either beyond the fluorescence life-time of PBP-MDCC or is readily removed by the phosphate mop (see Materials and Methods) present in the reaction. These results suggest that H91 plays an important role in Pi release from EF-G.

Molecular dynamics simulations are consistent with the role of H91 in Pi release. Based on
the crystal structure of the ribosome bound to EF-G•GTP (PDB: 4CR1) 4 , we have built and equilibrated an initial model of EF-G bound to GDP and Pi using MacroMoleculeBuilder (MMB) 34 and with MD simulations 35 . This model essentially mimics the situation after GTP hydrolysis but before Pi release. In MD, the distance between Pi and the β -phosphate was 4.3 Å (Fig. 3a,c), while between H91 (N δ1 ) and Pi it was 3.6 Å (Fig. 3a,d); this did not differ significantly from what was observed with MMB. In addition, a strong coordination between Pi and Mg 2+ was observed in both MMB and MD runs, suggesting that the Mg 2+ stabilizes Pi even after GTP hydrolysis. Close comparison of the ribosome bound EF-G•GTP structures (PDB: 4CR1, 4JUW) 4,5 with the EF-G•GDP structures showed disappearance (PDB: 4KDA) 24 or major repositioning of the Mg 2+ with respect to β -phosphate and Pi (PDB: 2WRI) 22 (Fig. S2). This suggests that in order for the Pi to be released the Mg 2+ needs to be repositioned. To test this hypothesis, another model was built without the Mg 2+ and equilibrated in MMB. Then the model was relaxed with MD as the former one (Fig. 3b). We saw an immediate increase in distance between β -phosphate and Pi to ~5.9 Å (Fig. 3b,c). In contrast, the distance between Pi and H91 (either N δ1 or N ε2 ) remained unchanged at ~3.5 Å (Fig. 3b,d), suggesting that the Pi was strongly coupled with H91. Thus, H91 and Pi could together move towards the open GDP-bound configuration. This is consistent with the suggestion from the biochemical data that H91 facilitates Pi release.

Discussion
Our results show that GTP hydrolysis by EF-G gets strongly stimulated by the ribosome (Table 1) 26 , suggesting that the ribosomal components act as the GTPase activating factor for EF-G 30,36 . This is in agreement with the previous results and also similar to EF-Tu, where the rate of GTP hydrolysis has been reported to increase by a million fold upon binding to the ribosome 29 . In structural terms, the activation is probably achieved by proper positioning of the two switch regions of EF-G in relation to GTP and also by opening the so called 'hydrophobic gate' composed of the two hydrophobic amino acids Ile18 and Ile60 (E. coli numbers) (Fig. 4a). In crystal structures of EF-G with GTP analogues, H91 is directed towards the γ -phosphate with an average distance of 3.6 Å between them, which is typical for a salt bridge (Fig. 4a) 4,5 . Although the exact role of H91 in GTP hydrolysis cannot be deduced from these structures, a close coordination of H91 and the γ -phosphate is evident.
In this work, we have characterized four H91 mutant EF-Gs for GTP hydrolysis and Pi release and compared those with WT and F94L EF-Gs. All of these EF-G mutants have similar affinity to GTP in solution and the rate of intrinsic GTP hydrolysis is not affected by any of these mutations. This result, similar to EF-Tu suggests that the spontaneous GTP hydrolysis by EF-G does not depend on H91 29 . In a recent study, a monovalent cation far from H84 has been suggested as the catalytic cofactor for GTP hydrolysis in ribosome unbound EF-Tu 37 and likely to be the same in EF-G. It should be noted that in the free form of EF-G, H91 points away from the γ -phosphate of GTP (PDB: 2J7K) 38 , which explains why H91 is not involved in intrinsic GTPase activity of EF-G. However, on the ribosome, the mutated EF-Gs display distinct behavior. The largest defect in ribosome stimulated GTP hydrolysis was identified for the H91E mutant. This result is not unexpected, since in this case the bulky positive side chain of His is replaced by a negative side chain of Glu. This swap will not only destroy the coupling between the amino acid and the γ -phosphate, but may also alter the charge distribution and the intramolecular The initial models are built using MacroMoleculeBuilder (MMB) 34 , on the basis of the crystal structure of EF-G•GDPCP bound to the ribosome (PDB: 4CR1) 4 . Both models represent the scenario immediately after GTP hydrolysis but before Pi release. We did not observe any relevant structural differences in the replica MDs. (c,d) Time trace for the distances between Pi and β -phosphate (c) and Pi and H91 (either Nδ 1 or N ε2 ) (d) estimated from MD structures with (black trace) or without (red/green trace) Mg 2+ ion. The red and the green traces in (d) are distances involving N δ1 or N ε2 atoms of the H91 side chain, respectively. interactions in the GTP binding pocket. In contrast, no defect was observed for the H91Q mutant. This is also expected since Gln is present instead of His in all G-proteins except the translational GTPases. When side chains are compared, the NH 2 group of Gln can easily be superimposed on the NH group of His, which is actually involved in coupling with the γ -phosphate. Thus, it follows naturally that the H91Q mutant acts like the WT in GTP hydrolysis. Similar results have also been reported with the corresponding mutation in EF-Tu (H84Q), which showed only small to moderate reduction in the GTPase activity 9,10 . However, a recent report on H84Q mutation of E. coli EF-Tu shows about a 4000 fold decrease in the rate of GTP hydrolysis 29 , which contradicts the earlier results. Further investigation will be needed to solve the controversy in the rate of GTP hydrolysis by H84Q EF-Tu.
The H91A and H91R mutants are about seven fold slower than the WT in ribosome stimulated GTP hydrolysis. In these two cases, the positive side chain of His is replaced either with a small nonpolar side chain (A) or a longer basic side chain (R). In H91A, the lack of proper orientation of the β and γ -phosphate of GTP can be expected due to absence of the coupling between the γ -phosphate and the with GDPCP (PDBs: 4JUW 5 in orange and 4CR1 4 in sand) and with GDP and fusidic acid (PDBs: 4KDA 24 in green and 2WRI 22 in magenta). The "hydrophobic gate" formed by residues Ile20 and Ile63 are shown in blue and the bound GDP and GDPCP are shown in brown and red respectively. (b) Close-up view of the switch II loop with its conserved H91, color code as in (a) shows that the whole switch-II loop following the superimposed α helix changes its orientation in the GTP/GDP state. (c) The potential exit path of Pi (modelled in the crystal structure (PDB: 2WRI) 22 through the inter-domain space of EF-G•GDP (blue) on the ribosome (grey). It can be observed that the Pi exit path (marked with an arrow) in front of H91 (magenta) is not hindered by any ribosomal components.
Scientific RepoRts | 5:12970 | DOi: 10.1038/srep12970 nonpolar small side chain of Ala. The H91R mutant, in contrast, might have additional coupling with multiple phosphates of GTP due to the occurrence of three N atoms in its side chain. It should be mentioned that the H91A EF-G was previously reported as completely inactive in GTP hydrolysis 11 . This apparent discrepancy, we think, may arise from differences in the sample preparation and experimental conditions.
Compared to the mean time of GTP hydrolysis by WT EF-G (~5 ms) Pi release takes much longer time (~35 ms) (Fig. 2d), which suggests that Pi remains trapped in the nucleotide binding pocket even after GTP is hydrolyzed. Interestingly, the Pi retention time is significantly longer than the WT for all the four H91 mutants (lowest 4000 ms), while for the F94L mutant it is similar to the WT (Fig. 2d). Moreover, comparison of the mean times of GTP hydrolysis and Pi release clearly demonstrates a much bigger defect in Pi release for all the H91 mutants than in GTP hydrolysis. Thus, our results suggest a direct involvement of H91 in Pi release.
To understand the structural basis of Pi release we compared the crystal structures of EF-G bound to GTP analogues (PDB: 4CR1, 4JUW) 4,5 and GDP, both on the ribosome (PDB: 2WRI, 4KDA) 22,24 and in the free state (PDB: 1FNM, 2BM0, 4M1K, 4MYU) 39,40 . While the most obvious change can be seen in the orientation of the H91 side chain, which 'flipped' almost 130°-140° (Fig. 4a) and consequently altered the conformation of the entire switch II loop (Fig. 4b), several other changes were visible. Detailed inspection of the GTP binding site dragged our attention to a Mg 2+ ion coordinating with both the β -and the γ -phosphates of GTP. Interestingly, in the EF-G•GDP structures this Mg 2+ was either absent (PDB: 4KDA) 24 or its position varied a lot (PDB: 2WRI) 22 (Fig. S2). This analysis suggested that displacement of the Mg 2+ might be necessary for Pi release. Our MD simulations indeed support this hypothesis. When the EF-G•GDP•Pi model contained the Mg 2+ ion, no change in the interatomic distances was observed during the MD run (Fig. 3a). In contrast, an immediate increase in the distance between the β -phosphate and Pi was noticed when the model was built without the Mg 2+ (Fig. 3b). However, H91 remained constantly coupled with the Pi although the His side chain might undergo a rotation thereby switching the coupling from the N δ1 to the N ε2 atom. Thus we propose that the tight coupling of Pi to H91 and the 'flipping' movement of H91are the two key determinants of Pi release.
The lack of coupling with Pi in H91A and H91E EF-Gs renders them significantly defective in Pi release. On the contrary, perhaps a too strong coupling with the Arg side chain in H91R makes Pi release so slow that it could not be measured under our experimental conditions. Most interestingly, the H91Q mutant, which is not impaired in GTP hydrolysis showed a more than 20 fold defect in Pi release. It suggests that although EF-G can attain the optimal conformation for GTP hydrolysis with Gln instead of His, the bulky side chain of His is instrumental in changing the conformation of switch II required for Pi release (Fig. 4b). This view is further supported by the observation that in the flipped conformation (GDP state), the H91 side chain opens into a tunnel which runs through the inter domain space of EF-G 22,24 . Since Pi remains coupled with H91 after GTP hydrolysis, most likely it follows H91 and thereby reaches this tunnel, which constitutes the exit path for the Pi (Fig. 4c). This path is free from steric obstructions even on the ribosome (Fig. 4c). Thus, we infer that the tight coordination of Pi with H91 and the 'flipping' of the H91 side chain are the key determinants for efficient Pi release. This mechanism may have general significance as other translational GTPases (e.g. EF-Tu) possess a His corresponding to H91, which undergoes similar structural transitions.

Materials and Methods
Mutagenesis of EF-G and protein purification. The WT fusA gene form E. coli cloned into plasmid pET24b with a C-terminal hexa-histidine tag was a kind gift from Kevin S. Wilson, Oklahoma state University, USA 26 . Using this construct as a template and following standard protocols for site-directed mutagenesis, the H91 residue was mutated to alanine (A), Glutamine (Q), Arginine (R) and glutamic acid (E), resulting into four mutant EF-Gs, H91A, H91Q, H91R and H91E respectively. In a similar way, residue F94 was mutated to Leucine (L), creating F94L mutant. The mutations were confirmed by DNA sequencing. Over-expression and purification of the WT and mutant EF-Gs were carried out as described earlier 25 , by using Ni 2+ affinity chromatography followed by size exclusion. The homogeneity of the WT and mutant EF-Gs was confirmed by SDS-Page followed by mass spectrometric analysis.

Buffers and components.
All experiments were performed in HEPES polymix buffer (pH 7.5) con- The GTPase activity of the EF-Gs on the ribosome was measured in a similar way by mixing vacant 70S ribosomes (3 μ M) and [ 3 H] GTP (15 μ M) (in one mix) with EF-G (15 μ M) in a quench-flow device (RQF-3 KinTek Corp.) at 37 °C. In order to estimate the single turnover GTP hydrolysis rate, data points only from the initial fast phase were used. The rate constants of GTP hydrolysis (k GTP ) was obtained by fitting the data with a single exponential function using ORIGIN 8.0 (Originlab Corporation). The mean time τ GTP was estimated as (1/k GTP ).
Single-round Pi release. The release of inorganic phosphate from EF-G after GTP hydrolysis was measured in a stopped flow instrument (Applied Photophysics SX20) by monitoring the fluorescence of MDCC-PBP that shows an immediate increase in fluorescence upon binding to Pi 32,33 . Mix A containing 70S ribosomes (2 μ M) and GTP (100 μ M) was rapidly mixed with mix B containing EF-G (20 μ M) and MDCC-PBP (1 μ M) after preincubating both mixes for 10 minutes at 37 °C. Both mixes also contained a phosphate mop comprising 7-methylguanosine (300 μ M) and purine nucleoside phosphorylase (0.1 unit/μ l). The rate constants of Pi release (k Pi ) was estimated by fitting the fluorescence time traces with a single exponential function using Origin 8.0 (Originlab Corp.). The mean time for Pi release (τ Pi ) was estimated as (1/k Pi ).
Measurement of affinity for GTP. The equilibrium dissociation constant (K D ) for GTP to EF-G was estimated using a fluorescent analog of GTP called mant-GTP ((2′ , 3′ )-O-N-methylanthraniloyl-GTP). When exited at 290 nm, fluorescence resonance energy transfer (FRET) occurs between the Trp residues present in the proximity of the nucleotide binding pocket in EF-G and the mant group of mant-GTP, thereby leading to increase in its fluorescence 28 . EF-G (2 μ M) was mixed with increasing amounts of mant-GTP (5-60 μ M) at 37 °C and the fluorescence change at 445 nm was recorded in a fluorimeter (HITACHI F-7000). The difference in the FRET signal with and without EF-G was plotted against the concentration of mant-GTP and the K D was estimated by fitting of a hyperbolic function using ORIGIN 8.0 (Originlab Corporation).

Modelling of EF-G•GDP -Pi and Molecular Dynamics simulations. We built and equilibrated
an initial model of EF-G bound to GDP and Pi using MMB 34 based on the crystal structure of the ribosome bound to EF-G•GTP (PDB: 4CR1) 4 , as a reduced-dimensionality prototype simulation prior to full MD. During equilibration, EF-G residues 83-91, Mg 2+ , β -phosphate of GDP and Pi were flexible. We placed a small water droplet restrained to a ~1.6 nm radius about H91. The PARM99 force field 35 was applied to the water and all atoms within 1.2 nm of the flexible regions. A strong coordination between Pi and Mg 2+ was observed, in addition to coordination between H91 and Pi. To understand the role of the Mg 2+ , another model without Mg 2+ was built and similarly equilibrated in MMB. For MD simulation, GDP parameters (charges, dihedrals, angles, bonds, and van der Waals parameters) were based on Meagher et al. 42 . For Pi, the atomic point charges were calculated with the GAUSSIAN 09 43 at the HF/6-31G* level, while the other parameters were obtained with general Amber force field (GAFF) 35 . The model with Mg 2+ was solvated with 45347 single point charge (SPC) waters while the model without Mg 2+ was solvated with 46144 SPC waters. Both systems were submitted to 100 steps of steep descent energy minimization to remove bad contacts between the solvent and the protein, followed by 200 ps of restrained MD simulation to relax the water molecules. Finally, for both models, two random initial velocity 50 ns MD runs were performed using a time step of 0.002 ps. Periodic boundary conditions were used in a NPT ensemble (v-rescale thermostat 44 and Parrinello-Rahman barostat 45 τ T = 0.1 ps, T ref = 300 K, P ref = 1 bar). The Particle Mesh Ewald (PME) 46 method was applied with a cut-off of 1.0 nm. A twin range cut-off with neighbor list cut-off 1.0 was used for the van der Waals interactions. All simulations and consequent analyses were carried out using the Gromacs 4.6.5 software package conjugated with the PARM99 force field 35,47 .