Negative catalysis by the editing domain of class I aminoacyl-tRNA synthetases

Abstract Aminoacyl-tRNA synthetases (AARS) translate the genetic code by loading tRNAs with the cognate amino acids. The errors in amino acid recognition are cleared at the AARS editing domain through hydrolysis of misaminoacyl-tRNAs. This ensures faithful protein synthesis and cellular fitness. Using Escherichia coli isoleucyl-tRNA synthetase (IleRS) as a model enzyme, we demonstrated that the class I editing domain clears the non-cognate amino acids well-discriminated at the synthetic site with the same rates as the weakly-discriminated fidelity threats. This unveiled low selectivity suggests that evolutionary pressure to optimize the rates against the amino acids that jeopardize translational fidelity did not shape the editing site. Instead, we propose that editing was shaped to safeguard cognate aminoacyl-tRNAs against hydrolysis. Misediting is prevented by the residues that promote negative catalysis through destabilisation of the transition state comprising cognate amino acid. Such powerful design allows broad substrate acceptance of the editing domain along with its exquisite specificity in the cognate aminoacyl-tRNA rejection. Editing proceeds by direct substrate delivery to the editing domain (in cis pathway). However, we found that class I IleRS also releases misaminoacyl-tRNAIle and edits it in trans. This minor editing pathway was up to now recognized only for class II AARSs.


INTRODUCTION
Aminoacyl-tRNA synthetases (AARS) couple cognate amino acid and tRNA pairs for protein biosynthesis. They are divided into two, evolutionary distinct classes, class I and class II (1,2). In both classes, the pairing occurs at the synthetic active site by the same two-step mechanism bearing some class-dependent features (3). The first step, amino acid activation, comprises the formation of aminoacyl-AMP (AA-AMP) while the second step is the transfer of the aminoacyl moiety to the tRNA (formation of aminoacyl-tRNA, AA-tRNA) (Figure 1, paths 1 and 4). The coupling of non-cognate substrates leads to mistranslation, which can be toxic for the cell (4)(5)(6). Due to physicochemical similarities of cellular amino acids, around half of AARSs cannot achieve the tolerable level of fidelity (estimated to be 1 in 3300 (7)) in the synthetic reactions alone and thus have evolved editing (reviewed in (8,9)). The error can be corrected by hydrolysis of non-cognate AA-AMP within the confines of the synthetic site (pre-transfer editing, Figure 1, paths 2 and 3) (10,11) and/or by hydrolysis of Figure 1. IleRS pathways of aminoacylation (green arrows) and editing (red arrows). The synthetic pathway consists of amino acid activation (1) and the aminoacyl transfer step (4). The editing pathways include tRNA-independent (2) and tRNA-dependent (3) pre-transfer editing and post-transfer editing (6,9). Post-transfer editing can occur by translocation of AA-tRNA (5) to the editing domain for hydrolysis (6, in cis) or by AA-tRNA dissociation (7), its subsequent rebinding to the editing site (8) and hydrolysis (9, in trans). misaminoacyl-tRNA at the dedicated editing domain (posttransfer editing) (12,13). The latter appears to be the dominant pathway, operating by two possible routes--in cis (Figure 1, path 5 and 6) and in trans (Figure 1, paths 7-9) (9). Editing in trans, so far demonstrated only in class II AARS (14), entails dissociation of the AA-tRNA and its rebinding with the 3 -end facing the editing domain.
The interplay between the synthetic and editing sites was firstly addressed by Fersht's double-sieve hypothesis proposed originally for class I isoleucyl-(IleRS) and valyl-tRNA synthetases (ValRS) (15). It states that the synthetic site uses steric clash to discard larger than the cognate amino acids while the editing site clears smaller/isosteric non-cognate amino acids that were successfully aminoacylated to the tRNA. The steric clash was also proposed to prevent the binding of the cognate AA-tRNA to the editing domain. But, does the productive recognition at the editing site correlate well with the amino acid misrecognition at the synthetic site, and to what extent does the steric clash define the selectivity of the editing site? The former was anticipated but not experimentally addressed. The latter was tested to show that the selectivity against the cognate AA-tRNA arises from its imposed unproductive binding (16)(17)(18).
IleRS rapidly hydrolyses tRNA Ile s misaminoacylated with non-proteinogenic norvaline (Nva) and Val (6). This is expected as both Nva and Val are misactivated with a frequency that is 10-fold higher than the estimated tolerable error (7) and thus pose threats to the fidelity of Ile-tRNA Ile formation (6,15). Surprisingly, IleRS can also efficiently hydrolyse tRNA Ile misaminoacylated with a nonproteinogenic ␣-aminobutyrate (Abu) and its synthetic ␥fluorinated analogues (F 2 Abu and F 3 Abu), which are misactivated with up to a 20-fold lower frequency than the estimated tolerable error (19). This questions whether the editing site substrates need to be well misrecognized at the synthetic site, as anticipated.
Here, we set out to explore what shaped the selectivity of class I editing site and to unravel whether the same mechanisms and demands for selectivity are shared between the editing and the synthetic sites using IleRS as a model enzyme. We characterized amino acid activation and AA-tRNA Ile hydrolysis using a range of amino acids with different physicochemical properties (Ala, Ser, Thr, Met, Leu, Nle). We found that IleRS synthetic site discriminates with at least 20 000-fold against the tested non-cognate amino acids. Thus, these substrates should not pose a fidelity problem. Nevertheless, all misaminoacylated tRNA Ile s were rapidly hydrolysed (35-65 s -1 ) at the editing site. Only cognate Ile-tRNA Ile was weakly hydrolysed, demonstrating that evolution of the editing site was driven by negative catalysis (20,21), i.e. selection towards destabilisation of the transition state for the cognate AA-tRNA hydrolysis (misediting). We also found that negative determinants for misediting vary among the closely related class I editing domains. Finally, we discovered that in IleRS, delivery of the AA-tRNA to the editing domain entails the accumulation of free AA-tRNA in solution, reminiscent of class II AARSs editing in trans.

Purification and activation of the EF-Tu
The elongation factor Tu (EF-Tu, EC 3.6.5.3) with the Cterminal His-tag was overexpressed in E. coli BL21(DE3). Cells were grown to OD 600 of 0.6-0.8 at 37 • C and expression was induced with 0.2 mM IPTG for 3 h. EF-Tu was purified on Ni-NTA resin (Cytiva) as described previously (23). EF-Tu was stored as the inactive GDP-bound form at −20 • C in a buffer containing 50 mM HEPES pH 7.5, 10 mM MgCl 2 , 50 mM KCl, 50% glycerol, 50 M GDP and 5 mM ␤-mercaptoethanol. Activation of the EF-Tu:GDP was performed in 70 mM HEPES pH 7.5, 50 mM ammonium acetate, 10 mM magnesium acetate, 30 mM KCl, 0.8 mM DTT, 10 mM phosphoenolpyruvate, 1 mM GTP and 0.08 U/l pyruvate kinase (Sigma) at 37 • C for 2 h. EF-Tu:GTP was used immediately after the activation. The activation is not efficient and results in about 10-15% of the total EF-Tu being capable of AA-tRNA binding (23)(24)(25).
Nucleic Acids Research, 2022, Vol. 50, No. 7 4031 Thus, the herein reported concentrations of EF-Tu present the form capable of binding AA-tRNA (10-15% of total concentration of EF-Tu).

Purification and labelling of tRNAs
Synthetic genes for tRNA Ile GAT (with G1-C72 instead of WT A1-U72 sequence), tRNA Leu TAA and tRNA Val TAC inserted into the pET3a plasmid were overexpressed in E. coli BL21(DE3) (11,22). Cells were grown to OD 600 of 0.5-0.6 at 37 • C and expression was induced with 1 mM IPTG overnight at 30 • C. Substitution of the first base pair enhances transcription and does not affect tRNA Ile participation in the IleRS synthetic and editing reactions (26,27). tRNAs were isolated and purified by phenol/chloroform extraction, PEG 8000 precipitation (removal of high molecular weight nucleic acids) and ethanol precipitation as described previously (11). Purified tRNA Leu and tRNA Val had acceptor activity >90%. The acceptor activity of tRNA Ile was around 50% so it was subjected to further purification by reverse phase chromatography on a semi-preparative Jupiter C4 column (Phenomenex), as described (11), which increased the acceptor activity to 80-90%. tRNAs were stored in 5 mM HEPES pH 7.5. Before further use tRNA Leu and tRNA Val were renaturated by heating at 85 • C for 3 min, adding an equal volume of pre-heated 20 mM MgCl 2 , and slow cooling to room temperature for about 1 h. Labelled [ 32 P]tRNAs were prepared as described (28,29). Briefly, 5 M tRNAs were incubated with 5 M tRNA nucleotidyltransferase at 37 • C in a buffer containing 1 M [␣-32 P] ATP (specific activity of 3000 Ci mmol -1 ), 20 mM MgCl 2 , 5 mM Na 4 P 2 O 7 , 200 mM Tris pH 8.0, and 0.5 mM DTT. After 1 min, 0.1 U/l of thermostable inorganic pyrophosphatase (TIPP, Sigma) was added to shift the equilibration towards the [␣-32 P]ATP incorporation and the mixture was incubated for 2 min at room temperature. tRNAs were purified by phenol/chloroform extraction followed by two consecutive chromatography steps on Bio-Spin P-30 columns (Bio-Rad) to remove the remaining [␣-32 P]ATP. The [ 32 P]tRNA samples were dialyzed against 10 mM HEPES pH 7.5. [ 32 P]tRNAs had acceptor activity 75-90%.

Preparation of misaminoacylated tRNAs
Misaminoacylated [ 32 P]tRNAs were prepared by mixing 25 M tRNA Ile with 5 M T243R/D342A IleRS (mutant inactive in post-transfer editing) and a particular amino acid at the following concentration (4 mM Ala or Met; 2 mM Val, Nle or Thr; 0.2 mM Leu; 10 mM Ser) in a buffer containing 20 mM HEPES pH 7.5, 10 mM MgCl 2 , 150 mM NH 4 Cl, 2 mM ATP, 0.008 U/l TIPP, 0.01 mg/ml BSA (New England Biolabs). The main purpose of the TIPP is shifting the equilibrium of aminoacylation reaction towards product formation by hydrolysis of pyrophosphate. All amino acids were purchased from Sigma. Amino acids were added in a moderate amount to prevent aminoacylation with possible Ile contaminations in the non-cognate amino acid samples. Reactions were quenched after 30 min at 37 • C by mixing with an equal amount of phenol/chloroform. AA-tRNA Ile s were purified by phenol/chloroform extraction followed by two consecutive steps on Bio-Spin P30 columns (Bio-Rad) and dialyzed against 10 mM NaOAc pH 4.5. Before further use AA-tRNA Ile s were renaturated by as described in Purification and labelling of tRNAs.

Amino acid activation
Amino acid activation was followed by an ATP-PP i exchange assay (30)(31)(32) which was performed at 37 • C in a buffer containing 50 mM HEPES pH 7.5, 20 mM MgCl 2 , 0.1 mg/ml BSA, 5 mM DTT, 4 mM ATP and 1 mM [ 32 P]PP i (Perkin-Elmer). Enzymes were present at 50-100 nM while amino acid concentrations were varied from 0.1 to 10 × K M . Reactions were quenched by mixing 1.5 l of the reaction mixture with 3 l of the quench solution (600 mM NaOAc pH 4.5 and 0.15% SDS). Formed [ 32 P]ATP was separated from the remaining [ 32 P]PPi by thin-layer chromatography (TLC) on polyethyleneimine plates (Macherey-Nagel) in 4 M urea and 750 mM KH 2 PO 4 pH 3.5. Signal visualization was performed on a Typhoon Phosphoimager (GE Healthcare) and quantified with ImageQuant software as described (29). Kinetic parameters (k cat and k sp ) were obtained by fitting the data to the modified Michalis-Menten equation: , where k cat is the turnover number, k sp is the specificity constant (k cat /K M ), [S] is the amino acid concentration and k obs is the observed rate constant. k obs is calculated as v 0 /[E] 0 , where v 0 is the initial reaction rate and [E] 0 is the enzyme concentration. The data were fitted using GraphPad Prism software. We opted for the modified Michaelis-Menten equation due to the benefits discussed recently by K. Johnson (33). Briefly, by fitting the data to the modified model, k sp is obtained directly, rather than being calculated from k cat and K M , which leads to smaller errors.

Two-step aminoacylation
Aminoacylation of tRNA Ile , comprising both activation and the transfer step, by various IleRS variants was followed at 37 • C in a buffer containing 20 mM HEPES pH 7.5, 10 mM MgCl 2 , 150 mM NH 4 Cl, 2 mM ATP, 0.008 U/l TIPP and 0.01 mg/ml BSA, 15 M tRNA Ile , 1 mM Ile and 20 nM WT, T246A, H333A, H333G, T246A/H333A or L247A/H333G IleRS. Misaminoacylation of tRNA Ile was followed under the same conditions except that the concentrations of enzymes were higher (1 M WT or T243R/D342A IleRS) and non-cognate amino acids were used at following concentrations: Leu 0.2 mM; Nle and Thr 2 mM; Ala and Met 4 mM; Ser 10 mM. The reactions were stopped by mixing 1.5 l of the reaction mixture with 3 l of the quench solution (600 mM NaOAc pH 4.5 and 0.15% SDS). 1.5 l of the quenched reaction mixture was mixed with 3 l of P1 nuclease (Sigma) (≥0.01 U/l in 300 mM NaOAc pH 5.0 and 0.15 mM ZnCl 2 ) and incubated for 1 hour at room temperature. The P1 nuclease treatment releases terminal adenine nucleotide, with or without amino acid attached as the result of tRNA aminoacylation. Free and aminoacylated [ 32 P]AMP were separated by TLC in 100 mM NaOAc and 5% HOAc. Signal visualization was performed on a Typhoon Phosphoimager (GE Healthcare) and quantified with ImageQuant software as described (29).

Parallel Formation of AMP and AA-tRNA
Formation of [ 32 P]AMP and AA-[ 32 P]tRNA were followed at 37 • C in parallel reactions, in a buffer containing 50 mM HEPES 7.5, 20 mM MgCl 2 , 0.1 mg/ml BSA, 2 mM DTT, 1 mM ATP, 0.004 U/l TIPP and 10-12 M tRNA. The reactions that monitored [ 32 P]AMP formation were supplemented with [␣-32 P]ATP (0.01-0.1 mCi/ml) (Perkin Elmer) while the reactions that monitored AA-[ 32 P]tRNA were supplemented with [ 32 P]tRNA (0.01-0.1 mCi/ml). The enzymes were 2 M and amino acids were used at the following concentrations: 2 mM Ile, 20 mM Val and 30 mM Nva or Thr. The concentration of the GTP-bound EF-Tu, when added, was estimated to 8-12 M (total concentration of added EF-Tu was 80 M). The reactions were stopped by mixing 1.5 l of the reaction mixture with 3 l of the quench solution (600 mM NaOAc pH 4.5 and 0.15% SDS). To measure AMP formation, [ 32 P]ATP and [ 32 P]AMP were separated by TLC in 100 mM NaOAc and 5% HOAc. When the formation of AA-tRNAs was followed, the quenched reaction mixtures were degraded by P1 nuclease and separated as described for the two-step aminoacylation (see above). Signal visualization was performed on a Typhoon Phosphoimager (GE Healthcare) and quantified with ImageQuant software as described (29).

Single-turnover hydrolysis
Single-turnover hydrolysis of AA-tRNAs was performed at 37 • C by mixing equal volumes of 20 M IleRS in a buffer containing 200 mM HEPES pH 7.5, 75 mM NH 4 Cl, 20 mM MgCl 2 , 5 mM DTT and 0.01 mg/ml BSA and freshly renaturated misaminoacylated [ 32 P]tRNA Ile (0.2-1 M) in 10 mM NaOAc pH 4.5 as described previously (19). Mixing of the enzyme and AA-[ 32 P]tRNAs was done using a rapid chemical quench instrument (RQF-3, KinTek Corp.). For the reactions having t 1/2 ≥ 5 s, manual mixing was performed. The reaction mixtures were quenched, treated with P1 nuclease, AA-[ 32 P]AMP and [ 32 P]AMP were separated and analysed as described for the aminoacylation (see above and (29)). The data were fitted to the single exponential equation Y = Y 0 + A × e −k hydrolysis ×t , where Y 0 is the y intercept, A is a scaling constant, k hydrolysis is the apparent hydrolytic rate constant, and t is time.

Effect of the expression of IleRS variants on the growth of E. coli
E. coli BL21(DE3) strain was transformed with pET28 plasmids carrying genes for either WT IleRS, H333A IleRS or T246A/H333A IleRS. The empty plasmid was used as a control. The cultures were grown in 100 ml culture flasks in the M9 media with the addition of 0.4% glucose and 30 g/ml kanamycin. Overnight cultures were diluted to OD 600 of 0.04 and supplemented with 100 M IPTG to induce the protein expression. The growth at 37 • C and 250 rpm was monitored using UV-Vis spectrophotometer Evolution 60S (Thermo Scientific). The expression profile was followed by SDS-PAGE. The data were fitted to the reparametrised Gompertz growth model ln OD 600 (t) , where A is the maximal cell growth, λ is the lag phase and k is the maximal growth rate (34). The doubling time was calculated as t = ln 2 k .

Molecular dynamics simulations
The structure of EcIleRS was obtained by AlphaFold deposited under identifier AF-P00956-F1 in the AlphaFold Protein Structure Database (35). The T246A/H333A variant was made by deleting the side chains of Thr246 and His333 residues except for the beta carbon. The ligands were 2 -L-isoleucyladenosine (Ile2A), and 2 -Lvalyladenosine (Val2A) (Supplementary Figure S1A). EcIl-eRS in a complex with Val2A was prepared by overlapping EcIleRS (AF entry: AF-P00956-F1) with the crystal structure of the editing domain of T. thermophilus IleRS (PDB entry: 1WNZ) bound to 2 -L-valy-2 -aminoadenosine (Val2AA) (36). The Val2AA coordinates were then transferred to the EcIleRS editing domain. We exchange 2 amido (Val2AA) to 2 ester (Val2A) linkage to fit better the native substrate. Ile2A was prepared by adding the methyl group to the valyl side chain of Val2A. The AM-BER FF14SB force field (37) was used for protein structure parametrization, whereas parameters for Ile2A and Val2A (Supplementary Figure S1B and S1C) were prepared de novo by using the antechamber module and the general Amber (GAFF) force field (38). Each complex was solvated in a truncated octahedral box of TIP3P water molecules (39) such that no solute atom was within 10Å of any box edge. Sodium ions were placed randomly within the solvent box to keep the system neutral. Bonds with hydrogen atoms were constrained using the SHAKE algorithm (40) which allowed us to use the time step of 2 fs. The Particle Mesh Ewald method was applied to calculate long-range electrostatic interactions. vdW cut off of 10.0Å was used. Minimization and equilibration procedures were performed using the PMEMD module whereas classical all atom molecular dynamics simulations were performed using the PMEMD.cuda MD engine (41) from the AMBER v16 software. Protein geometry optimization was conducted in four cycles with different constraints to remove steric clashes caused by hydrogenation, solvation and substrates introduction. Minimized systems were gradually heated from 0 to 300 K in the NVT ensemble for 500 ps and equilibrated during 5 ns in the NPT ensemble, after which the systems were subjected to productive, unconstrained MD simulations at constant temperature (300 K) and pressure (1 atm). The temperature was held constant using a Langevin dynamics with a collision frequency of 1 ps −1 , whereas pressure was held constant by using Berendsen barostat. Trajectory for each system was 300 ns (Nstlim = 150 000 000 steps; dt = 2 fs) and every 1000 step was sampled to the output file (NTPR) and trajectory file (NTWX). Trajectories analyses were conducted using the CPPTRAJ program (42) from the AmberTools v18 software suite. Clustering was performed by using K-means clustering algorithm available in CPPTRAJ to obtain the representative structures of the most populated clusters. Clustering was finished when number of clusters was 10. We used RMSD of editing domain backbone heavy atoms as distance metric. When it comes to the critical hydrogen bond discussed in the main text, this interaction was described as a hydro- gen bond if the donor-acceptor distance was ≤3.0Å and the donor-hydrogen-acceptor angle was within the range 180 • ± 40 • .

The editing site clears a broad range of misaminoacylated tRNA Ile s
Misaminoacylation of tRNA Ile with the non-cognate amino acids that are significantly different from Ile (Supplementary Figure S2) is challenging. However, this is a prerequisite for addressing the selectivity of the editing site as post-transfer editing is tested by following hydrolysis of preformed misaminoacylated tRNAs (29). In doing so, we found that all tested non-cognate amino acids were activated ( Figure 1, path 1) albeit with high discrimination factors (D > 20 000,  Figure S4). Interestingly, we did not observe aminoacylation with WT IleRS (Supplementary Figure S3, inset). Thus, the editing deficient AARSs can provide an alternative to the ribozyme approach (48) for tRNA misaminoacylation. This analysis is further complicated by artefacts that may arise from contamination of non-cognate amino acid samples with trace amounts of the cognate amino acids (5,15,49). For that reason, we estimated the purity of the used amino acids (Supplementary Figure S5) and found that Leu, and possibly also Met and Nle, may contain trace amounts of Ile.
Next, we isolated the post-transfer editing step by mixing preformed misaminoacylated tRNA Ile with a surplus of WT IleRS, using a rapid chemical quench instrument. The hydrolysis of misaminoacylated tRNA Ile was followed in time to calculate the first-order rate constant (Supplementary Figure S4). The single-turnover conditions ensure that product dissociation does not limit the observed rate (22). A 2-fold higher concentration of IleRS or AA-tRNA Ile returned the same hydrolysis rate confirming that binding is not rate-limiting. Thus, the observed rate constants (Figure 2, Supplementary Figure S4) represent the catalytic step (hydrolysis of misaminocylated tRNA Ile ) within the editing site.
The single-turnover analysis revealed, in agreement with the incapacity of the WT IleRS to accumulate misaminoacylated tRNAs Ile (Supplementary Figure S3, inset), that all misaminoacylated-tRNA Ile s were rapidly hydrolysed with similar rates ranging from 35 to 65 s -1 (Figure 2, please note that a possible Leu, Met and Nle contaminations did not compromise the editing analysis). This is surprising as these amino acids (except Val and Nva) are efficiently discriminated at the synthetic site and as such cannot pose a threat to IleRS aminoacylation fidelity. Finding that amino acids are rapidly cleared at the editing domain irrespectively of the requirement for their editing, lends a new paradigm about the editing selectivity principles. Moreover, the editing site shows no clear preference towards physicochemical features of the editing substrates, like size (Met and Nle, both with longer unbranched side chain, are eliminated), hydrophobicity (polar Ser and Thr are efficiently cleared at the editing site) or branching (Leu-tRNA Ile is also rapidly hydrolysed). Cognate Ile-tRNA Ile was the only exemption, suggesting, that prevention of cognate AA-tRNA misediting was a major constraint during the evolution of the editing site. Thus, we set to explore how the editing site excludes the cognate Ile-tRNA Ile  Thr-tRNA Ile 43 ± 6 Figure 2. Single-turnover hydrolysis of misaminoacylated tRNAs by WT IleRS. tRNAs misaminoacylated with amino acids that are well discriminated at the IleRS synthetic site (D > 3300) are presented in the striped area. Value 3300 is taken as the tolerable error of protein synthesis was estimated around 1 in 3300 (7). Rapid hydrolysis of Leu-, Met-and Nle-tRNA Ile confirmed that possible traces of cognate Ile in the Leu, Met or Nle samples did not compromise the editing analysis. Time courses from which the first-order rate constants (k hydrolysis ) were calculated are presented in Supplementary Figure S4. k hydroysis for Val-, Nva-, Abu-, F 2 Abuand F 3 Abu-tRNAs Ile were taken from (6,19). and in parallel promotes editing of misaminoacylated tRNAs.

Negative determinants for Ile-tRNA Ile misediting
Structural data showing the binding of Ile-tRNA Ile to the editing domain are missing. Therefore, we used the crystal structure of the T. thermophilus IleRS editing domain bound to 2 -L-valyl-2 -aminoadenosine (Val2AA, PDB code: 1WNZ, (42)) to guide the positioning of Ile-tRNA Ile analogue, 2 -L-isoleucyladenosine (Ile2A; Supplementary Figure S1), to the editing domain of EcIleRS (AF-P00956-F1). The complex was then subjected to 300 ns of the productive MD simulations. The representative structure of the WT editing domain ( Figure 3A) unravelled that the Ile2A 3 OH group, which is essential for IleRS editing (27), does not H-bond to the conserved Thr241. In T.thermophilus, interaction with analogous Thr228 productively positions the 3 OH group of Val2AA (1WNZ). Hence, the observed mode of Ile2A binding can be taken as non-productive in agreement with kinetic data showing a residual level of Ile-tRNA Ile hydrolysis by WT IleRS (Table 2, (27)). Two conserved residues of the editing domain, Thr246 and His333, seem to clash with the isoleucine side chain ( Figure 3A) and may influence the 3 OH displacement (36,50). We produced H333A and T246A mutants along with the double mutant and tested their activity under single-turnover conditions (Table 2). Indeed, T246A substitution, increased the rate of Ile-tRNA Ile misediting by 2fold (0.126 ± 0.006 s -1 ), while the H333A mutant showed a 20-fold increase (1.04 ± 0.06 s -1 ). The T246A/H333A mutant exhibited a 7-fold higher effect than the cumulative effects of the independent mutations and hydrolysed Ile-tRNA Ile with a rate (14 ± 1 s -1 ) approaching the WT rate with Val-tRNA Ile . Thus, kinetic data demon-strated the crucial role of His333 and Thr246 in preventing misediting. How do these two residues operate? Using MD simulations, we analysed Ile2A binding to the in silico built T246A/H333A variant. In this case, the representative structure ( Figure 3B) showed that the Ile2A 3 OH group H-bonds to Thr241, indicating, in agreement with the kinetic data (Table 2), the productive mode of binding. Evidently, the enlargement of the binding pocket allowed accommodation of the substrate's isoleucine side chain without mispositioning of the 3 OH group. Accordingly, the distance between the 3 OH and Thr241 longer than 3Å (nonproductive mode) is measured in 68% of the simulated time for the WT enzyme and 37% for the T246A/H333A variant (Supplementary Figure S6C). Next, we tested editing of non-cognate Val-tRNA Ile . All mutants displayed only 2-fold slower rates of editing relative to the WT ( Table 2) Figure S6B). In summary, we found that the main negative determinant of the IleRS editing site is His333 whose role is synergistically supported by Thr246 (the coupling energy around −1 kcal/mol, calculated as in (51)). This contrasts closely related LeuRS which utilizes Thr252 (analogous to Thr246 in EcIleRS) as a sole negative determinant (16) and suggests idiosyncratic evolution of the mechanisms governing rejection of the cognate product in class I editing domains.

IleRS deprived of the negative determinants misedits Ile-tRNA Ile in trans
During steady-state, (mis)aminoacylated AA-tRNA partitions between hydrolysis (editing; Figure 1 path 6) and dissociation ( Figure 1, path 7) (product release). To reach the editing site, AA-tRNA translocates (Figure 1, path 5) its single-stranded 3 end while the tRNA body remains bound to the enzyme (editing in cis). In case when the whole AA-tRNA dissociates from the enzyme prior to reaching the editing site, it can re-bind from the solution with the 3 end facing the editing domain (Figure 1, paths 8 and 9, editing in trans). Editing in cis depletes the product and thus compromises steady-state aminoacylation. In contrast, editing in trans may not affect the aminoacylation rate, because rebinding of AA-tRNA for hydrolysis is not favoured at low steady-state enzyme concentration. Therefore, the finding that both H333A and T246A/H333A IleRSs exhibit little to no change in steady-state aminoacylation rates relative to the WT enzyme (k aminoacylation , Table 2), despite rapid Ile-tRNA Ile hydrolysis at their editing sites (k hydrolysis , Table 2), implies that these mutants misedit Ile-tRNA Ile in trans. This is unexpected as editing in trans was not yet demonstrated for class I AARSs.
Non-stoichiometric ATP consumption is diagnostic of active editing as futile aminoacylation/editing cycles consume ATP without accumulating AA-tRNA. To address whether the IleRS mutants misedit Ile-tRNA Ile in trans, two   (52). ATP consumption (AMP formation) and Ile-tRNA Ile formation were followed in parallel reactions that differ only in the labelled components -[ 32 P]ATP was used for the former and [ 32 P]tRNA Ile for the latter (Supplementary Figure S8). The ratio of consumed ATP per Ile-tRNA Ile accumulated in solution (AMP/Ile-tRNA Ile ) was calculated for the reactions without and with EF-Tu  ( Figure 4). In the absence of EF-Tu, both mutants consume 18-(H333A) to 1100-fold (T246A/H333A) higher than the stoichiometric amount of ATP per released Ile-tRNA Ile . Thus, multi-turnover conditions at high concentrations of the mutants support misediting. That misediting takes place in trans, is further supported by 9-(H333A) to 18-fold (T246A/H333A) drop in AMP/Ile-tRNA Ile ratio in the presence of EF-Tu. The WT enzyme, exhibit- ing marginal Ile-tRNA Ile misediting, used a stoichiometric amount of ATP per Ile-tRNA Ile , independently on the presence/absence of EF-Tu. Interestingly, the significant energetic cost of misediting was exhibited mainly with T246A/H333A, raising an intriguing question--how detrimental is hydrolysis of Ile-tRNA Ile ?

Ile-tRNA Ile misediting impairs cell growth
To investigate to what extent misediting of Ile-tRNA Ile affects cell viability, we followed the growth of E. coli BL21(DE3) strain transformed with the plasmids encoding WT IleRS or its Ile-tRNA Ile misediting active variants (H333A and T246A/H333A). A moderate expression (Supplementary Figure S9) of the WT enzyme did not show any growth defects demonstrating that expression per se is not a burden for the cell ( Figure 5). Interestingly, the H333A mutant did not influence the growth rate suggesting that Ile-tRNA Ile misediting of 1 s -1 could be physiologically tolerated. In contrast, the T246A/H333A mutant (hydrolytic rate of 14 s -1 ) showed a noticeable growth defect ( Figure  5), in agreement with the negative selection against this activity. Further, it is worth noticing that the magnitude by which the above mutations impair cell growth is likely underestimated due to the expression of WT IleRS from the genome.

IleRS is unique among class Ia AARSs in exhibiting a high level of editing in trans
To address whether IleRS edits in trans its biological threat Val-tRNA Ile , we followed the accumulation of AMP and Val-tRNA Ile by the WT enzyme as described above (note that EF-Tu may bind tRNAs misaminoacylated with nearcognate amino acids (14,23,53)). In the absence of EF-Tu, the analysis returned the AMP/Val-tRNA Ile ratio of 1330 and a minor accumulation of Val-tRNA Ile (Figure 6, Supplementary Figure S10), both in agreement with the efficient Val-tRNA Ile editing (27). The addition of EF-Tu increased the accumulation of Val-tRNA Ile by 10-fold and decreased the amount of consumed ATP by >3-fold, leading to a significant drop (58-fold) in AMP/Val-tRNA Ile ratio (23 versus 1330). This indicates that IleRS edits Val-tRNA Ile in trans, providing to the best of our knowledge the first demonstration of editing in trans for a class I AARS.
Next, we tested whether LeuRS and ValRS also use editing in trans with their main biological threats Nva and Thr, respectively (5,54). Both LeuRS and ValRS consumed a highly non-stoichiometric amount of ATP per accumulated misaminoacylated tRNA (14 500 and 13 800, Figure 6, Supplementary Figure S10), in agreement with established editing of Nva-tRNA Leu and Thr-tRNA Val (6,22,54). The addition of EF-Tu dropped the ATP/Thr-tRNA Val ratio by <2-fold (13 800 versus 8600). The lack of EF-Tu effect indicates that ValRS edits Thr-tRNA Val in cis. The picture is more complicated for LeuRS, where the presence of EF-Tu promotes a 12-fold drop in ATP/AA-tRNA ratio (14 500 vs 1250) that may indicate the participation of editing in trans. However, the drop does not stem from a decrease in ATP consumption, which is only 1.04-fold lower in the presence of EF-Tu. Thus, the cycles of Nva-tRNA Leu hydrolysis and the subsequent tRNA Leu misaminoacylation which consumes ATP are not influenced by EF-Tu. This strongly suggests that LeuRS mainly operates in cis in agreement with the previous data (14). The observed EF-Tudependent accumulation of AA-tRNA Leu is puzzling and likely originates from trace contaminations of Leu in the Nva sample (Supplementary Figure S11). Indeed, the accumulation of Leu-tRNA Leu , like of Nva-tRNA Leu can be diminished by rebinding to LeuRS and (mis)editing. Yet, hydrolysis of Leu-tRNA Leu is 3 × 10 3 -times slower than Nva-tRNA Leu (22), contributing minimally to the ATP consumption. EF-Tu may bind Leu-tRNA Leu and thus affects its accumulation but without a noticeable effect on ATP consumption. To conclude, our data show that IleRS is distinct from closely related LeuRS and ValRS in a fraction of post-transfer editing that operates in trans.

Class I AARS synthetic and editing sites act in an opposing manner
AARSs are textbook examples of how high selectivity emerged under strong evolutionary pressure to evade deleterious errors (47). Their synthetic sites adopt numerous strategies to enforce recognition of the cognate and rejection of the non-cognate amino acids (55)(56)(57)(58). If noncognate amino acid, however, gets coupled to the tRNA, post-transfer editing resolves the problem. The editing site evolved to clear amino acids that jeopardize the accuracy of translation arguing that amino acids well discriminated at the synthetic site (large discrimination factor, D) will be Nucleic Acids Research, 2022, Vol. 50, No. 7  poorly edited. But is it so? What did shape the selectivity of the editing domain? We used IleRS as a model enzyme and a series of amino acids of distinct physicochemical features and evolutionary origin to address these questions.
We surprisingly found that tRNAs misaminoacylated with non-cognate amino acids that are efficiently discriminated at the activation step (Table 1) were all edited with the same rates as the biological threats Nva-and Val-tRNA Ile (Figure 2). This demonstrates that recognition at the editing site is not influenced by how well the non-cognate amino acid is discriminated at the synthetic site. A similar was found for ValRS (6) and LeuRS (5,22). It appears that the editing site is non-selective (except for the cognate AA-tRNA) and hydrolyses tRNAs misaminoacylated with amino acids spanning a broad range of physicochemical properties. What could be the evolutionary driving force for such broad selectivity? Several amino acids may compromise the fidelity of a particular AARS providing the evolutionary force for selection of the editing domain lacking the amino acid side chain recognition. Yet, as we discuss below, to prevent deleterious cognate hydrolysis, high specificity against the cognate amino acids was shaped by negative selection. How is it possible to achieve efficient catalysis without recognition of the amino acid substrate side chain? The substrate recognition and catalysis at the class I editing site relies on the common parts of all AA-tRNAs; the terminal adenosine of the tRNA (A76) and ␣-NH 3 + group of the amino acid attached to the tRNA both anchor the substrates (36) while the 2 OH or 3 OH group of the A76 acts as a general base and promotes catalysis (shown for class I (16,27) and class II (59,60)). Changes of the terminal adenosine (61), lack of 2 OH or 3 OH (16,27,59,60) or loss of the ␣-NH 3 + anchoring interactions deprived editing (22,62). Thus, it is plausible to assume that a common tRNA carrier and preselection of the editing substrates by the aminoacylation sieve, could have driven the evolution of the editing domain towards a broad substrate acceptance and the lack of recognition of the non-cognate amino acid's side chain. In contrast, the synthetic site, acting as the first sieve, recognizes standalone amino acid and uses most of its side chain (Table 1) to minimize the error and ATP consumption (editing) (47). Thus, the synthetic and editing sites act in an opposing manner; while the former is highly selective to prevent errors, the latter exhibits low selectivity to clear each non-cognate amino acid that comes loaded to the tRNA.
Finding that class II PheRS, which recognizes the functional group of Tyr at the editing site (63) edits Ile-tRNA Phe (64), suggests that broad selectivity may not be necessarily confined only to the class I editing domain. Indeed, Daminoacyl-tRNA deacylase, which bears a structural resemblance to the archaeal class II threonyl-tRNA synthetase editing domain (65), edits all D-amino acids at similar rates while efficiently rejecting L-amino acids. (66). In contrast, the editing domain (INS) of class II prolyl-tRNA synthetase (ProRS), as well as the free-standing bacterial ProRS INS domain homologs, have well-defined noncognate amino acid specificity (67). Similarly, alanyl-tRNA synthetase (AlaRS) uses Zn 2+ at the editing domain to select for serine, being the main threat of AlaRS fidelity (68). Interestingly, the free-standing trans-editing proteins also evolved a broad specificity--in this case regarding the tRNA substrate (69). Thus, across the editing systems, a similar concept emerged independently arguing for the benefits of broad substrates acceptance in the design of the efficient error correction mechanisms.

Negative catalysis ensures high specificity and broad selectivity of class I editing domain
Enzymes significantly differ in their physiological requirements for high selectivity (47). In some cases, low selectivity is beneficial allowing a broad substrate scope as in cytochrome P450 (70). The same applies to the editing domain. Yet, a unique feature of the editing domain, in which it mirrors highly selective enzymes (71), is its exquisite specificity in rejection of the cognate AA-tRNA. In general, specificity may evolve by positive and negative selection (21). While the former is a consequence of a selection for the enzyme's high catalytic efficiency towards the cognate substrate, the latter is an explicitly evolved trait against a particular non-cognate substrate to avoid deleterious errors. Here, we propose that specificity of the editing domain evolved through negative selection against the cognate AA-tRNA. We and others have previously shown that the cognate AA-tRNA is rejected from the editing site not by mitigating the binding, but by diminishing the catalysis (14,(16)(17)(18). Destabilization of the transition state solely for cognate AA-tRNA hydrolysis can be viewed as an example of negative catalysis (20). This concept was introduced to explain that alongside promoting a wanted reaction by lowering the energy of the transition state for the desired product (positive catalysis), enzymes may also increase the energy barrier of the competing transition state preventing the side reaction (negative catalysis). Herein, we broaden this concept to compare transition states for the competing substrates. Thus, the residue conferring negative catalysis should not influence the rate of native (wanted) reaction but should diminish the reaction rate with the prohibited substrate. Visualisation of our data by activity-specificity graph revealed that the His333 and Thr246 IleRS residues confer negative catalysis (Figure 7). Their substitutions do not influence editing of Val-tRNA Ile (k mut /k WT for editing is close to one) but promote misediting of Ile-tRNA Ile resulting in the variants with decreased specificity (drop in the ratio of the k mut /k WT values for editing over misediting). In contrast, D342A mutation in IleRS promoted a decrease in both activity and specificity conferring the positive role for the Asp342 residue in catalysis (promoting both wanted and unwanted reaction by anchoring ␣-NH 3 of the cognate and non-cognate amino acid (27,36)). Similarly, LeuRS Thr252, which imposes unproductive positioning of Leu-tRNA Leu (16), confers negative catalysis while the Asp345 residue (22), analogous to IleRS Asp342, promotes positive catalysis. In conclusion, negative selection/catalysis appears as a powerful mechanism to ensure low selectivity of the editing domain while keeping, at the same time, exquisite specificity in the rejection of the cognate AA-tRNA. The former was driven by relying on the common parts of all AA-tRNAs and the latter by the evolution of a specific kinetic rejection . Activity-specificity relationship for editing of IleRS and LeuRS mutants. Activity (editing) was calculated as the ratio of the rate constants for hydrolysis of misaminoacylated tRNAs (NCAA-tRNA) by mutant AARS and WT, respectively. Specificity was calculated as the ratio of editing (wanted) over misediting (unwanted), where the misediting is the ratio of the rate constants for hydrolysis of cognate AA-tRNAs by mutant AARS and WT, respectively. His333 and Thr246 establish a hydrogen bond, while Asp342 is on the other side of the pocket ( Figure 3A). mechanism based on the cognate amino acid side chain (16)(17)(18).

IleRS -class Ia enzyme with unique editing features
In class I AARSs, among which only Ile/Leu/ValRSs have the editing domain, delivery of the amino acid from the synthetic to the editing site occurs through fast translocation (estimated as faster than 80 s -1 in LeuRS (22)) of the 3end of the (mis)aminoacylated tRNA (72) (Figure 1). The fast translocation allows that the 3 -end reaches the editing site on a shorter time scale relative to the rate-limiting dissociation of the (mis)aminoacylated tRNA. Therefore, editing occurs in cis without the release of misaminoacylated tRNA. Release and rebinding of the misaminoacylated tRNAs to AARS was shown so far only for class II AARS (14), and was attributed to fast dissociation of (mis)aminoacylated tRNA (3). It, therefore, came as a surprise that IleRS edits Val-tRNA Ile with a significant contribution of the in trans pathway. At the same time, both LeuRS and ValRS predominantly operate in cis ( Figure  6). Thus, IleRS appears unique among closely related class Ia Ile/Leu/ValRSs in post-transfer editing. Although the mechanistic basis is still missing, it is worth commenting here that only IleRS, among Ile/Leu/ValRSs, uses tRNA to both optimize the affinity for the amino acid substrate and to stimulate pre-transfer editing (Figure 1, path 3). Indeed, E. coli IleRS showed substantial tRNA-dependent pre-transfer editing, which does not seem to strongly depend on the non-cognate amino acid, that comprises about 30% of total editing (11,22,46). Thus, an idiosyncratic util-isation of the tRNA as a co-factor may influence the rates of the tRNA-dependent steps, for example, translocation. It has been shown that PheRS competes with EF-Tu for binding to Tyr-tRNA Phe (14) providing a basis for editing in trans. In our experimental setup, designed to promote the EF-Tu bidning, EF-Tu efficiently competes with IleRS for binding to Val-tRNA Ile . This in vitro design is distinct from in vivo environment where other AA-tRNAs compete for EF-Tu. Nevertheless, it is plausible to assume that editing in trans is generally of lower proficiency than editing in cis, because EF-Tu may bind tRNAs misaminoacylated with amino acids similar to the cognate one (14,23,53) and redirect them to ribosomal translation. That said, IleRS capacity to edit errors prior misaminoacylation could be relevant, opening a provocative question to what extent are editing in trans and pre-transfer editing intertwined.

DATA AVAILABILITY
All data are available from the authors upon request.

SUPPLEMENTARY DATA
Supplementary Data are available at NAR Online.