Conformational and chemical selection by a trans-acting editing domain

Significance

Charging of tRNAs with the wrong amino acid can result in mistranslation of protein-encoding genes. Proofreading by tRNA editing domains clears these mischarged tRNAs, ensuring fidelity; however, structures of complexes with mischarged substrates are not currently available, and many mechanistic questions remain unanswered. ProXp-ala, present in all domains of life, selectively removes Ala from tRNA^Pro, leaving the cognate Pro-tRNA^Pro intact. A chemically synthesized nonhydrolyzable substrate analog has allowed characterization of the enzyme-substrate complex using NMR spectroscopy. Based on NMR studies, mutagenesis, enzymatic assays, molecular dynamics, and X-ray crystallography, we conclude that ProXp-ala uses multiple strategies, including conformational selection by a dynamic helix, size exclusion, and chemical discrimination, to ensure editing solely of Ala-tRNA^Pro.

Abstract

Molecular sieves ensure proper pairing of tRNAs and amino acids during aminoacyl-tRNA biosynthesis, thereby avoiding detrimental effects of mistranslation on cell growth and viability. Mischarging errors are often corrected through the activity of specialized editing domains present in some aminoacyl-tRNA synthetases or via single-domain trans-editing proteins. ProXp-ala is a ubiquitous trans-editing enzyme that edits Ala-tRNA^Pro, the product of Ala mischarging by prolyl-tRNA synthetase, although the structural basis for discrimination between correctly charged Pro-tRNA^Pro and mischarged Ala-tRNA^Ala is unclear. Deacylation assays using substrate analogs reveal that size discrimination is only one component of selectivity. We used NMR spectroscopy and sequence conservation to guide extensive site-directed mutagenesis of Caulobacter crescentus ProXp-ala, along with binding and deacylation assays to map specificity determinants. Chemical shift perturbations induced by an uncharged tRNA^Pro acceptor stem mimic, microhelix^Pro, or a nonhydrolyzable mischarged Ala-microhelix^Pro substrate analog identified residues important for binding and deacylation. Backbone ¹⁵N NMR relaxation experiments revealed dynamics for a helix flanking the substrate binding site in free ProXp-ala, likely reflecting sampling of open and closed conformations. Dynamics persist on binding to the uncharged microhelix, but are attenuated when the stably mischarged analog is bound. Computational docking and molecular dynamics simulations provide structural context for these findings and predict a role for the substrate primary α-amine group in substrate recognition. Overall, our results illuminate strategies used by a trans-editing domain to ensure acceptance of only mischarged Ala-tRNA^Pro, including conformational selection by a dynamic helix, size-based exclusion, and optimal positioning of substrate chemical groups.

A fundamental function of the aminoacyl-tRNA synthetase (ARS) family of enzymes is to attach specific amino acids to cognate tRNA substrates for use in protein synthesis. Owing to the similar size of many amino acids, approximately one-half of the ARSs have evolved editing mechanisms that serve to clear misacylated tRNAs and maintain translational fidelity (1). Editing active sites, which are located in domains distinct from the amino acid activation domain, have been identified in both class I and class II synthetases. More recently, single-domain proteins that are homologous to class II synthetase editing domains also have been shown to clear misacylated tRNAs in trans (2–6).

Larger substrates are generally believed to be less problematic for ARSs than amino acids that are similar in size or smaller than the cognate substrate according to a proposed double-sieve mechanism (7, 8). Larger amino acids are rejected from the aminoacylation active site in the first sieve, whereas the second sieve accepts and clears the mischarged tRNA while rejecting the larger cognate substrate. In support of the sieve-based mechanism of discrimination, editing active sites present in both classes of ARSs have been shown to be tunable, with point mutations that alter the size of the amino acid binding pocket resulting in alternate substrate specificities (9–11). Paradoxically, Ser, which is larger than Ala, is misactivated by alanyl-tRNA synthetase (AlaRS) along with smaller Gly, and both Ser-tRNA^Ala and Gly-tRNA^Ala are edited by the AlaRS editing domain. In addition, a family of single-domain editing proteins known as AlaXps, which are homologs of the AlaRS editing domain, clear Ser-tRNA^Ala in trans (12). Rather than size-based exclusion, the AlaRS and AlaXp editing domains have been proposed to use chemical discrimination to reject Ala and accept only Ser via an active site Zn²⁺ ion (4, 13).

Single-domain editing enzymes in the INS superfamily are homologous to the prolyl-tRNA synthetase (ProRS) editing domain (INS). Some INS superfamily members function as a third sieve that deacylates mischarged tRNAs not cleared by ARS editing domains (3, 6, 14–16); for example, ProRS misactivates both Ala and Cys. Ala, which is smaller than cognate Pro, is hydrolyzed in the INS domain present in some ProRSs, whereas Cys, which is similar in size to cognate Pro, is cleared by a single-domain protein known as YbaK (3, 14, 15, 17). An alternative triple-sieve editing mechanism has been identified in some species that lack a ProRS editing domain but encode two free-standing editing domains, YbaK and ProXp-ala. The latter performs the same function as the INS domain, hydrolyzing Ala-tRNA^Pro, but solely in trans (6, 18). The mechanism of YbaK deacylation of Cys-tRNA^Pro has been well characterized and involves cyclization of the substrate sulfhydryl moiety (17); however, the structural and chemical basis of Ala recognition and Pro discrimination by ProXp-ala remain unknown.

We previously showed that Caulobacter crescentus (Cc) ProXp-ala achieves tRNA^Pro substrate specificity, in part through sequence determinants in the acceptor stem of tRNA^Pro, including recognition of the unique first base pair (C1:G72) (6, 18). Thus, Ala-tRNA^Ala, which lacks this element, is a poor substrate for ProXp-Ala, and binding by elongation factor-Tu (EF-Tu) further ensures that the cognate substrate is not deacylated (6, 18). How ProXp-ala avoids deacylation of Pro-tRNA^Pro is an open question. Ala has few distinguishing features for positive recognition, but several features of Pro may be used for negative discrimination, including its larger size, its unique ring structure, and the presence of a secondary α-amine group. To investigate the structural basis of substrate recognition by Cc ProXp-ala, we synthesized a nonhydrolyzable 3′-amide-linked mischarged acceptor stem mimic, nh-Ala-microhelix^Pro. Using NMR, we mapped the interactions between ProXp-ala and both uncharged microhelix^Pro and nh-Ala-microhelix^Pro. Furthermore, ¹⁵N-relaxation experiments identified a helix that exhibits fast ps-ns dynamics, as well as slower μs-ms motions, which are attenuated only when bound to the mischarged aminoacyl-tRNA mimic. These results, combined with extensive mutagenesis, deacylation assays, X-ray crystallography, and computational studies, reveal that a remarkable combination of substrate sampling, size-based selection, and specific chemical groups ensures rejection of cognate Pro-tRNA^Pro from the ProXp-ala active site.

Results

To test the role of size-based discrimination in the mechanism of substrate selection by Cc ProXp-ala, we assayed its ability to deacylate Escherichia coli (Ec) tRNA^Pro, a substrate for Cc ProXp-ala (6, 18), that was charged with amino acids of varying size, including Val, Pro, Cys, 2-aminobutyric acid (Abu), Ser, and azetidine-2-carboxylic acid (Aze), a smaller Pro analog with a four-membered ring (Fig. 1). With two exceptions, the measured k_obs values are directly correlated with substrate aminoacyl moiety size, with smaller substrates generally demonstrating faster rates of deacylation under single-turnover conditions. Ala-tRNA^Pro is the best substrate, followed by Ser-, Abu-, Cys-, and Pro-tRNA^Pro. Val is the largest amino acid tested, and Val-tRNA^Pro deacylation was not detectable. Aze-tRNA^Pro exhibited a deacylation rate between that of Abu- and Pro- or Cys-tRNA^Pro, despite the fact that Aze is more similar in size to Ser, which was deacylated almost as efficiently as Ala (Fig. 1). Thus, both Cys- and Aze-tRNA^Pro were deacylated somewhat more slowly than expected based on size alone, suggesting that the chemical nature of the side chain plays a role as well.

Fig. 1.

Cc ProXp-ala deacylation of aminoacyl-tRNA^Pro. (A) Representative time courses for deacylation of 0.1 μM Ala-tRNA^Pro (black), Ser-tRNA^Pro (magenta), Aze-tRNA^Pro (green), Cys-tRNA^Pro (violet), Abu-tRNA^Pro (blue), Pro-tRNA^Pro (red), and Val-tRNA^Pro (orange) by 0.75 μM Cc ProXp-ala. (B) Chemical structures and molecular volumes of the aminoacyl moieties used in A. Observed rate constants of deacylation (k_obs) are indicated below each structure. A representative time course is shown for each substrate, and k_obs values are the average of three trials with the SD indicated.

Cc ProXp-ala has previously been shown to deacylate an acceptor stem-derived Ala-microhelix^Pro substrate (Fig. 2), with only a twofold reduction in enzymatic activity (18). To investigate the mechanism of Ala-tRNA recognition by ProXp-Ala, we synthesized a nonhydrolyzable 3′-amide–linked nh-Ala-microhelix^Pro as a substrate analog, and used an uncharged microhelix^Pro to mimic the reaction product (Fig. 2). Apparent dissociation constants for ProXp-ala binding to microhelix^Pro and nh-Ala-microhelix^Pro, determined from sedimentation velocity analytical ultracentrifugation (AUC) experiments (19, 20), were 1.5 ± 0.3 μM and 0.52 ± 0.14 μM, respectively (Fig. S1A). Thus, the Ala moiety contributes modestly to enhancing the affinity of ProXp-ala for the misacylated substrate over the uncharged tRNA.

Fig. 2.

Ec tRNA^Pro-derived microhelix RNAs used in the study. Microhelix^Pro is derived from the acceptor stem of Ec tRNA^Pro and contains a stable UUCG tetraloop. Microhelix^Pro and nh-Ala-microhelix^Pro differ by the displayed “R” group; microhelix^Pro contains a free 3′ OH, whereas nh-Ala-microhelix^Pro contains an Ala moiety attached to the 3′ position via a nonhydrolyzable amide bond. The numbering of RNA bases refers to full-length Ec tRNA^Pro.

Fig. S1.

AUC data used for binding analyses in this study. (A) Isotherms of weight-average sedimentation coefficient as a function of ProXp-ala concentration. Analysis of the isotherm gives initial estimates of K_d values, which were refined by global data modeling using SEDPHAT. (B and C) Representative AUC sedimentation velocity data used to determine relative binding affinity of 1 μM microhelix^Pro to 1.2 μM (B) or 9.6 μM (C) WT or mutant ProXp-ala. Data are shown for microhelix^Pro alone (dashed line) and in the presence of WT (black), V29A (green), N46S (red), or K45A (blue) ProXp-ala. The c(s) distributions were integrated to find signal-weighted average sedimentation coefficients (s_w). The s_w values for each mutant were compared with that of 1.2 μM WT ProXp-ala (concentration at which ≈50% of microhelix^Pro is bound). The relative binding for each mutant was defined as follows: +++, equal or larger s_w values for the 2.4 μM mutant, indicating a less than twofold reduction in affinity (as for V29A in B); ++, s_w values smaller for the 2.4 μM mutant but larger for the 9.6 μM mutant, indicating a twofold to eightfold reduction in affinity (as for the N46S mutant in C); +, s_w values smaller for the 9.6 μM mutant, indicating a greater than eightfold reduction in affinity; −, no binary complex observed up to the 9.6 μM mutant (as for K45A in C).

We used NMR spectroscopy to map the microhelix^Pro and nh-Ala-microhelix^Pro binding surfaces of Cc ProXp-ala. We first obtained backbone resonance assignments in the absence of substrates, and then recorded 2D ¹H-¹⁵N HSQC spectra in the presence of varying concentrations of either microhelix^Pro or nh-Ala-microhelix^Pro (Fig. 3A). Resonance assignments were obtained for 156 of the 160 non-Pro residues (168 total) for free ProXp-ala, for 153 residues in the presence of microhelix^Pro, and for 140 residues in the presence of nh-Ala-microhelix^Pro; secondary chemical shifts (21) were generally consistent with the deposited crystal structure 1VJF (Fig. S2). Chemical shift perturbations (CSPs) were observed for a similar subset of residues of ProXp-ala with both microhelix^Pro and nh-Ala-microhelix^Pro, indicating a site-specific interaction with each RNA. Furthermore, the CSPs induced by Ala-microhelix^Pro are in the same direction as those for microhelix^Pro but are larger in magnitude (Fig. 3A).

Fig. 3.

¹⁵N-HSQC NMR spectra reveal similarities and differences in Cc ProXp-ala binding to microhelix^Pro or nh-Ala-microhelix^Pro. (A) 800 MHz ¹⁵N-HSQC spectrum of ProXp-ala alone (black) or in the presence of a 1:1 molar ratio of either microhelix^Pro (blue) or nh-Ala-microhelix^Pro (red). Blue arrows indicate signals with significant CSPs induced by microhelix^Pro, and red arrows indicate additional CSPs induced by nh-Ala-microhelix^Pro. Arrows are labeled with the corresponding residue of ProXp-ala. (B) Enlarged view of the regions of the ¹⁵N-HSQC spectrum boxed in A from the titrations of ProXp-ala with 0.2 (blue), 0.4 (green), 0.6 (orange), or 0.8 (red) equivalents of microhelix^Pro (Left) or nh-Ala-microhelix^Pro (Right). (C) Summary of per-residue CSPs [Δδ = (Δδ_H² + Δδ_N²/25)^1/2] on Cc ProXp-ala binding to microhelix^Pro (blue bars) or nh-Ala-microhelix^Pro (red bars). Secondary structure elements based on the Cc ProXp-ala crystal structure (PDB ID code 1VFJ) are displayed above the bar graph. The dashed line at 0.18 ppm represents the mean plus 1 SD of δΔ for microhelix^Pro. Residues whose amide resonances were broadened beyond detection are indicated by asterisks. (D) CSP values for nh-Ala-microhelix^Pro mapped onto the crystal structure as a linear gradient from white (≤0.1 ppm) to red (≥0.6 ppm). Prolines and other residues that could not be assigned in NMR spectra are in gray. Residues with peaks broadened beyond detection on the addition of nh-Ala-microhelix^Pro are colored cyan. Spheres are displayed for all residues with a CSP difference of at least 0.05 ppm, with relative sizes scaled to a range of 50–100% of the van der Waals radius.

Fig. S2.

Secondary structure predictions from NMR chemical shift assignments. The TALOS-N webserver (https://spin.niddk.nih.gov/bax/nmrserver/talosn) was used to predict the secondary structures of free ProXp-ala (A) and ProXp-ala bound to microhelix^Pro (B) using assigned HN, N, Cα, and Cβ chemical shifts. The red bars indicate probability for the α-helix, and the blue bars indicate probability for the β-strand. Unassigned residues are depicted as gray bars. The secondary structure of the ProXp-ala crystal structure (PDB ID code 1VJF) is displayed above each plot for comparison.

NMR spectra of ProXp-ala recorded over a range of concentrations of uncharged microhelix^Pro revealed most of the perturbed backbone resonances to be in the fast exchange regime, characterized by an incremental shift in peak position with increasing RNA concentration (Fig. 3B). In contrast, titration with nh-Ala-microhelix^Pro revealed most backbone resonances of ProXp-ala to be in the intermediate-slow exchange regime, characterized by the broadening or disappearance of the resonance corresponding to its “free” position and reappearance of the same resonance in the “bound” position (Fig. 3B). This is consistent with higher affinity and/or slower off-rates for the mischarged substrate analog (Fig. S1A). Lineshape analysis (Fig. S3) of the NMR titration data for the uncharged microhelix^Pro yielded a K_d of 0.97 ± 0.21 μM, similar to that obtained from the AUC experiments, and a k_off of 492 ± 28 s⁻¹. Lower signal-to-noise, line broadening, and the slow/intermediate exchange behavior of the titration data with nh-Ala-microhelix^Pro resulted in greater uncertainty in the fitted K_d value (0.037 ± 0.04 μM), although better determined was a fivefold lower k_off of 104 ± 8 s⁻¹. Lineshape simulations indicated that intermediate-to-slow exchange is obtained with a K_d value consistent with that measured by AUC as long as k_off is slower than ∼150 s⁻¹.

Fig. S3.

Representative results from 2D NMR lineshape analysis of Cc ProXp-ala bound to either microhelix^Pro (A) or nh-Ala-microhelix^Pro (B). Shown are experimental (blue) and fitted (red) contour plots and fitted free and bound peak positions (orange line) for each of the eight indicated titration points of residues S78 (Left) and E165 (Right). Lineshape analysis was conducted with TITAN (https://www.ucl.ac.uk/biosciences/departments/smb/christodoulou/titan) using a two-state binding model. Fitting of microhelix^Pro spectra (15 residues) yielded K_d = 0.97 ± 0.21 μM and k_off = 492 ± 28 s⁻¹, whereas fitting of nh-Ala-microhelix^Pro spectra (14 residues) yielded K_d = 0.037 ± 0.04 μM and k_off = 104 ± 8 s⁻¹.

Differences in CSPs between the substrate and product analogs pinpoint residues involved in recognition of the mischarged amino acid. In the presence of uncharged microhelix^Pro, significant CSPs (greater than the average CSP plus 1 SD) are observed for 20 residues of ProXp-ala (Fig. 3C, blue bars). Of the significant CSPs induced by nh-Ala-microhelix^Pro, 11 were similar to those induced by the uncharged RNA (e.g., K71, R80, L81, S82) (Fig. 3 A and C); however, some residues that showed weak to moderate CSPs in the presence of microhelix^Pro were significantly more perturbed by binding to nh-Ala-microhelix^Pro (F27, V29, E34, L47, E64, and S99) (Fig. 3 A and C). In addition to large perturbations throughout the enzyme, nine resonances assigned in the free protein were broadened beyond detection on the addition of nh-Ala-microhelix^Pro (H23, V26, K45, T96, V100, F129, H130, L132, and N134; Fig. 3C, asterisks). Overall, the greatest concentration of CSPs was seen in the region containing helix α2 and its preceding loop (residues 27–34). Significant perturbations are also observed for strand β2 (residues 46–51), the loop preceding helix α3 (residues 63–64 and 69–72), and strand β4 (residues 80–84). Residue S99 exhibits the largest overall perturbation and largest difference in CSP between the charged and uncharged substrates; assignments are not available for flanking residues (T96, P97, G98, and V100). Mapping the combined ¹H and ¹⁵N CSPs for nh-Ala-microhelix^Pro (Fig. 3C) onto the crystal structure of Cc ProXp-ala reveals that all significant CSPs are localized to one face of ProXp-ala defined by α2, β2, and β4 (Fig. 3D). The resonances that were broadened beyond detection are located within a pocket on this same face of ProXp-ala (Fig. 3 C and D). CSP differences >0.05 ppm (Fig. 3D) include residues F27, V29, and E34 within helix α2 and its preceding loop, and residue S99. In addition, the residues adjacent to peaks broadened beyond detection on addition of nh-Ala-microhelix^Pro in the β6-β7 loop (N128 and D135) are more perturbed in the presence of the Ala moiety than with the uncharged microhelix. The electrostatic potential map of ProXp-ala (Fig. S4A) shows that these regions surround a large electropositive surface to complement the electronegative phosphate groups of the CCA-3′ end of the tRNA.

Fig. S4.

Electrostatic potentials and conservation of Cc ProXp-ala. (A) Electrostatic potentials of Cc ProXp-ala plotted on the molecular surface (PDB ID code 1VJF) as a gradient from red (−3.0 kT) to white (0 kT) to blue (3.0 kT). Potentials were calculated using the adaptive Poisson–Boltzmann solver (APBS) plugin within PyMOL. (B) Normalized conservation score within all Bacteria of Cc ProXp-ala plotted on the structure (PDB ID code 1VJF) as a gradient from orange (−1) to white (0) to green (3). Scores were determined from the AL2CO webserver (prodata.swmed.edu/al2co/al2co.php). (C) Graph of normalized conservation scores used to generate the map in B. The highly conserved residues mentioned in this study are labeled in B and C.

Guided by the CSPs (Fig. 3) and sequence conservation (Fig. S4 B and C), we generated site-directed mutants (to Ala) to examine effects on substrate binding and catalysis (Table 1). We performed deacylation assays with full-length Ec Ala-tRNA^Pro as the substrate and AUC-binding assays using microhelix^Pro (Fig. S1 B and C). Mutations in helix α2 and the preceding β1-α2 loop resulted in only minor decreases in deacylation (≤3.5-fold) and binding (≤8-fold), with the exception of highly conserved H23, which showed a 12-fold decrease in activity. Mutation of H23 and α2 residue E34 resulted in significantly (>8-fold) reduced binding. No deacylation activity or binding was detected for the K45A variant, consistent with the critical role of this β2 residue in positioning the CCA-3′ end into the active site, as previously reported for the INS domain of ProRS and YbaK (9, 17, 22–24). Mutation of the β2 residues H43 to Ala and N46 to Ser (N46A could not be purified) also showed significantly reduced deacylation activity (by 8- and 14-fold, respectively) and 2- to 8-fold reduced binding. S99 displayed the largest CSP difference in the presence of nh-Ala-microhelix^Pro, but resulted in only minor (∼3-fold) defects in deacylation and binding on mutation. In contrast, mutation of G98 had a large effect on enzyme activity (95-fold), but only a modest effect on binding (Table 1). Although unassigned in our NMR experiments, G98 is part of a highly conserved GXXXP motif in INS-like domains and has been proposed to be involved in stabilization of the transition state during catalysis (25). The N128A, H130A, and N134A mutants demonstrated moderate decreases in deacylation (6.3- to 11-fold), whereas V127A ProXp-ala was only 2-fold less active than wild type (WT). V127A, N128A, and H130A resulted in weaker binding, but N134A had no observable binding defect, despite an 11-fold decreased rate of deacylation. Thus, mutagenesis experiments suggest that the side chains of conserved residues within β2, the α4-α5 loop, and the β6-β7 loop are generally more critical for both binding and catalysis than the residues within helix α2 and the β1-α2 loop.

Table 1.

WT and mutant Cc ProXp-ala deacylation rate constants and relative Ec microhelix^Pro binding affinity

Mutation	Location	Deacylation kobs, min−1*	Fold decrease	Relative binding†
WT		1.05 ± 0.07	1	+++
H23A	β1-α2	0.086 ± 0.019	12	+
P24A	β1-α2	1.07 ± 0.07	0	+++
P25A	β1-α2	0.75 ± 0.11	1.4	+++
F27A	β1-α2	0.35 ± 0.04	3.0	++
R28A	β1-α2	0.30 ± 0.04	3.5	++
V29A	α2	0.53 ± 0.11	2.0	+++
E30A	α2	2.2 ± 0.2	0	+++
E31A	α2	0.29 ± 0.04	3.6	+++
E34A	α2	0.54 ± 0.05	2.0	+
H43A	β2	0.13 ± 0.02	8.2	++
K45A	β2	ND	>100	−
N46S	β2	0.076 ± 0.007	14	++
G98A	α4-α5	0.011 ± 0.002	95	++
S99A	α4-α5	0.39 ± 0.10	2.7	++
V127A	β6	0.50 ± 0.06	2.1	++
N128A	β6	0.17 ± 0.01	6.3	+
H130A	β6-β7	0.11 ± 0.03	9.7	+
N134A	β6-β7	0.098 ± 0.02	11	+++

ND = not detected.

^*Deacylation rate constants (k_obs) were determined using 0.75 μM enzyme and 0.1 μM Ala-tRNA^Pro at 20 °C. All results are an average of three trials with the SD indicated.

^†AUC sedimentation velocity experiments were used to determine the relative binding affinity of microhelix^Pro to mutant ProXp-ala, as described in Fig. S1 B and C. Binding relative to WT ProXp-ala is indicated as follows: +++, less than twofold reduction; ++, twofold to eightfold reduction; +, greater than eightfold reduction; −, no observed binding.

Comparison of the crystal structures of ProXp-ala domains to the INS domain from Enterococcus faecalis ProRS [Protein Data Bank (PDB) ID code 2J3L; residues 233–402] (9, 22, 25) suggests that a local conformational change on substrate binding may be responsible for the CSPs observed in helix α2 (Fig. S5). The ProXp-ala domains from Cc and Agrobacterium tumefaciens (At; PDB ID code 1VKI) share 43% sequence identity and have nearly identical structures, with a backbone atom rmsd of 1.11 Å. INS shares 12% and 11% sequence identity with Cc and At ProXp-ala, respectively (22). In addition, we independently determined the crystal structure of Cc ProXp-ala at 1.69-Å resolution in a new space group (PDB ID code 5VXB; Table S1) and found its structure to be nearly identical to that of 1VJF (rmsd <0.5 Å), except for the β1-α2 loop and the beginning of helix α2 (rmsd 1–2.5 Å) (Fig. S5A). In a superposition of each of these structures, the most striking difference is in the orientation of helix α2 between the INS domain and the three trans-editing domains (Fig. S5B). Based on this observation, we hypothesize that helix α2 may dynamically sample “open” and “closed” conformations.

Fig. S5.

Comparison of the crystal structures of ProXp-ala and homologs. (A) Backbone RMSD values per residue between the two crystal structures of Cc ProXp-ala (PDB IDs 1VJF and 5VXB) and average backbone rmsd values per residue among the three monomers in the asymmetric unit of 5VXB. A secondary structure representation is displayed above the plot. (B) Superposition of Cα atoms of Cc ProXp-ala (red, PDB ID code 1VJF; blue, PDB ID code 5VXB), At ProXp-ala (cyan; PDB ID code 1VKI), and Ef ProRS INS (orange, PDB ID code 2J3L; residues 233–402). The alpha helix corresponding to Cc helix α2 is labeled.

Table S1.

X-ray crystallography data collection and refinement statistics for Cc ProXp-ala

Variable	Value*
Space group	P1211
Cell parameters
a, b, c, Å	62.04, 67.73, 62.93
α, β, γ, ◦	90.00, 118.62, 90.00
Resolution (Å)	19.80–1.69 (1.78–1.69)
Average redundancy	6.7 (6.4)
Average I/σ(I)	16.3 (3.3)
Rp.i.m.	0.026 (0.228)
Completeness, %	99.6 (99.0)
Refinement statistics
No. of reflections	51,071 (7,401)
R factor	0.1664
Free R factor	0.1896
Free R percentage	4.01
Average B factor, Å2	25
rmsd from ideality
Bond lengths, Å	0.37
Bond angles, ◦	0.57
Ramachandran favored, %	99
Ramachandran allowed, %	1
Ramachandran outliers, %	0

^*Values in parentheses are for the highest-resolution shell.

To test this conformational sampling hypothesis, we first used ¹⁵N NMR relaxation measurements to characterize the ps-ns dynamics of Cc ProXp-ala (26–28). Steady-state ¹H-¹⁵N heteronuclear NOE values could be measured for 136 of the 168 residues of ProXp-ala (Fig. 4A). ¹H-¹⁵N NOE values were relatively uniform, with an average of 0.81, except for the N and C termini and for residues 27–39, corresponding to helix α2 and the β1-α2 loop (Fig. 4A). The values observed in these residues indicate significant flexibility on the ps-ns timescale. These ps-ns dynamics persist for the same set of residues when ProXp-ala is bound to uncharged microhelix^Pro, as evidenced by steady-state ¹H-¹⁵N heteronuclear NOE values (Fig. 4B). Although the lower signal-to-noise ratios in the protein-RNA complexes limit quantitative analysis, on binding to nh-Ala-microhelix^Pro, ¹H-¹⁵N heteronuclear NOE values for the β1-α2 loop and helix α2 become uniform with the rest of the protein, indicating attenuation of ps-ns internal motions (Fig. 4C). Because the amide ¹H-¹⁵N heteronuclear NOE is dominated by motions on the time scales of the ¹H and ¹⁵N Larmor frequencies (29), these observations indicate fast ps-ns motions in the β1-α2 loop and helix α2.

Fig. 4.

Amide ¹⁵N-relaxation values indicate ps-ns dynamics in α2 of Cc ProXp-ala, which are attenuated on substrate recognition. (A) ¹H-¹⁵N heteronuclear NOE data for 0.5 mM free ProXp-ala. (B) 0.5 mM ProXp-ala bound to a molar equivalent of microhelix^Pro. (C) 0.2 mM ProXp-ala bound to a molar equivalent nh-Ala-microhelix^Pro. The red box in each indicates stretch of residues near helix α2 with values lower than the average (dashed line) in free ProXp-ala. Secondary structure elements of ProXp-ala are displayed above the plots.

Because concerted conformational changes in helix α2 imply motions on timescales slower than ps-ns, we performed ¹⁵N-single quantum Carr–Purcell–Meiboom–Gill (CPMG) relaxation dispersion NMR experiments (30) to characterize chemical exchange on the μs-ms time scale (Fig. 5A). Of the 119 residues for which relaxation curves could be determined and assigned, 14 exhibit significant exchange on this time scale (R_ex >2 s⁻¹). Nine of the 14 localize to helix α2 and the β1-α2 loop (D22, V26, R28, E31, G32, L33, I35, A38, and M39), and the other five are distributed throughout the enzyme (L56, D68, R80, S99, and V140) (Fig. 5B). The nine residues in helix α2 and the β1-α2 loop fit well to a two-state exchange model, yielding a global exchange rate (k_ex) of 954 ± 57 s⁻¹ and a major state population (P_A) of 98.3% ± 0.1%. These observations are consistent with a model in which conformational sampling by residues around helix α2 enables substrate binding and selectivity.

Fig. 5.

Amide ¹⁵N CPMG relaxation dispersion experiments reveal μs-ms exchange dynamics involving helix α2 of Cc ProXp-ala. (A) Relaxation dispersion curves for ProXp-ala residues R28 and E31 at 600 and 800 MHz globally fit to a two-state exchange model with a k_ex of 950 s⁻¹. Residuals of the fit are plotted below each graph. (B) Exchange contributions to transverse relaxation, R_ex, mapped to the structure as a linear gradient from white (≤2 s⁻¹) to red (≥10 s⁻¹).

To further explore the dynamic sampling hypothesis, we used molecular dynamics (MD) simulations to investigate the conformational flexibility of helix α2 of Cc ProXp-ala in its free and ligand-bound states. To simulate the closed conformation, we built a homology model using the INS domain from Ef ProRS (PDB ID code 2J3L; residues 233–402; Fig. S5B) as a template (9, 22, 25). With this model, we performed docking simulations using the terminal portion of the Ala-tRNA^Pro ligand (CCA-Ala), such that the ligand was flexible and the Cc ProXp-ala active site side chains were allowed to sample alternative rotamer conformations. Out of 2,000 successful poses, the largest ligand cluster (pairwise rmsd of Ala heavy atoms <1 Å) featured Ala moiety binding in the active site pocket, with the positioning of the remaining RNA nucleotides highly variable. From the cluster, a representative pose was used as the starting point for a 15-ns explicit solvent MD simulation to allow the ligand to sample a small range of conformational space. The resulting structure was used as the starting model for subsequent MD simulations in which the aminoacyl moiety was modified to Pro to mimic properly charged tRNA (ProXp-ala + CCA-Pro), or the aminoacyl moiety was deleted to mimic the product (ProXp-ala + CCA). All analyses were performed on 40 ns of production simulation after equilibration. To simulate the dynamics of the unbound protein, we extended a trajectory from a ProXp-ala + CCA MD simulation in which the CCA ligand dissociated after ∼42 ns. In these simulations, the coordinate root mean square fluctuations (RMSFs) showed helix α2 to be significantly more flexible in the free protein but with reduced dynamics when bound to the CCA-aminoacyl ligand. Interestingly, the presence of either aminoacyl group (Ala or Pro) reduced the dynamics similarly (Fig. 6A).

Fig. 6.

RMSFs and active site snapshots from MD simulations. (A) RMSF in backbone nitrogen atom coordinates for each residue of ProXp-ala either alone (black) or in the presence of a CCA trinucleotide (blue), CCA-Ala (red), or CCA-Pro (purple) over a 40-ns MD simulation trajectory. Secondary structure elements of ProXp-ala are displayed above the plot. (B and C) Substrate conformations in the Cc ProXp-ala active site after flexible docking and MD simulations for “mischarged” CCA-Ala (B, green) and cognate CCA-Pro (C, blue). A76-aa is shown in each case, and only ProXp-ala residues within at least 5 Å of the aminoacyl moiety are shown. Potential hydrogen bond distances are indicated in Å.

Investigation of the ligand-bound active site after 30 ns of MD simulation revealed a significant difference between the CCA-Ala and CCA-Pro substrates (Fig. 6 B and C). The CCA-Ala ligand is stabilized by the potential for eight hydrogen bonds, five with the Ala moiety and an additional three with the backbone of A76 (Fig. 6B). In the presence of CCA-Pro, the ligand is shifted slightly out of the active site, correlating with a shift in the position of H130 toward the GXXXP loop (Fig. 6C). The GXXXP loop is also significantly perturbed when CCA-Pro is bound. The altered conformation of H130, the GXXXP loop, and the ligand cause a rotation in the A76 backbone, resulting in only transient interactions between the phosphate of A76 and K45, as well as the potential for only two intermolecular hydrogen bonds to stabilize the interaction.

Our MD results are consistent with stabilization of the closed conformation of helix α2 in the presence of CCA-Ala, but we found a similar result for CCA-Pro (Fig. 6A). However, the significant distortion of the active site suggests that CCA-Pro is unlikely to bind efficiently or in a catalytically productive manner (Fig. 6C). Therefore, we tested whether cognate Pro-tRNA^Pro could inhibit Ala-tRNA^Pro deacylation and thus compete for binding to ProXp-ala. Our results show that Pro-tRNA^Pro is a very poor inhibitor of deacylation, behaving just like uncharged tRNA, demonstrating that Pro does not make any significant contribution to overall Pro-tRNA^Pro binding (Fig. S6). In contrast, nh-Ala-microhelix^Pro is a potent competitor, substantially reducing the rate of deacylation of Ala-tRNA^Pro when present at a fourfold excess over the cognate substrate, further validating its use in this study (Fig. S6).

Fig. S6.

Inhibition of Cc ProXp-ala deacylation of Ala-tRNA^Pro. Shown is the fitting of time courses for deacylation of 250 nM Ala-tRNA^Pro by 20 nM Cc ProXp-ala in the absence of an inhibitor (black circles) or in the presence of 1 µM Pro-tRNA^Pro (blue diamonds), 1 µM tRNA^Pro (red squares), or 1 µM nh-Ala-microhelix^Pro (green triangles). All results are the average of three trials with SDs indicated.

Discussion

Trans-editing domains are believed to be ancient proteins that preceded the addition of homologous domains into tRNA synthetases (31, 32). Here we show that the bacterial trans-editing enzyme ProXp-ala discriminates Ala-tRNA^Pro from Pro-tRNA^Pro through various strategies, including conformational selection in the substrate binding site, size-based exclusion of nonsubstrates, and chemical discrimination. Conformational differences in the position of active site-flanking helix α2 between crystal structures of ProXp-ala and INS suggest that it might sample both open and closed conformations (Fig. S5). Consistent with this hypothesis, our NMR relaxation experiments showed that in the free protein, helix α2 is dynamic across multiple timescales (ps-ns and μs-ms) (Figs. 4A and 5). NMR resonances for residues proximal to this region showed greater CSPs in the presence of the Ala moiety relative to the uncharged microhelix^Pro, suggesting stabilization of a closed conformation on binding of the aminoacylated substrate. Likewise, ¹H-¹⁵N heteronuclear NOE experiments showed attenuation of fast-timescale dynamics only when the tRNA mimic was charged with the Ala moiety. These observations support a model in which ProXp-ala uses conformational selection as a mechanism of ligand sampling and discrimination.

ProXp-ala catalyzes the same reaction as the homologous INS editing domain present in many bacterial ProRSs. In the Ef ProRS INS domain, the backbone carbonyl of G261 (V26 in the β1-α2 loop of Cc ProXp-ala) and the 2′OH of the substrate A76 are proposed to hydrogen bond with a water, activating it for nucleophilic attack on the scissile ester carbon. Proton transfer from the water to the O3′ of A76 results in cleavage of the 3′O-linked ester bond (25). In the case of ProXp-ala, a mechanism involving backbone and tRNA functional groups is supported by the fact that, with the exception of K45, which has a proposed role in CCA-3′ end substrate positioning (9, 17, 22–24), the activity of the trans-editing domain is not dramatically affected on mutation of any other active site residues (Table 1). In addition, V26 is broadened beyond detection in the presence of nh-Ala-microhelix^Pro (Fig. 3C), supporting its proximity to the substrate.

The deacylation assays show a strong inverse correlation between hydrolysis rate and volume of the amino acid. This correlation is imperfect, however, because Aze and Ser have similar volumes but Ser is hydrolyzed 20 times faster, and because Cys and Abu are of similar size but Cys hydrolysis is only 1/10th as efficient as that of Abu. The cyclic rings of Pro and Aze both feature secondary amines, whereas the thiol group on Cys distinguishes it structurally and electronically from Ser or Abu. These observations reveal likely contributions from conformational and electronic features of the potential substrates. Interestingly, although extant ProRS is not known to misactivate Ser, we found that Ser-tRNA^Pro is hydrolyzed as efficiently as Ala-tRNA^Pro, suggesting that this activity of Cc ProXp-ala may be an evolutionary remnant.

Computational docking experiments provided useful insight into the structural basis for discrimination. The CCA-Ala-docked active site model contains numerous potential hydrogen bonds between active site residues and the primary α-amine group of Ala (Fig. 6B). Side chains of N46 and H130 are each within hydrogen-bonding distance, consistent with significantly decreased activity upon mutation (Table 1). Backbone carbonyls of T44 and V100 also are within hydrogen-bonding distance, consistent with CSPs induced by nh-Ala-microhelix^Pro. Residue K45 is universally conserved among INS superfamily members with one exception, and has been shown to play a role in substrate positioning (9, 17, 22–24). In our active site model, the side chain of K45 is within hydrogen-bonding distance of the A76 phosphate group, in agreement with those findings. In contrast to CCA-Ala, low-energy models of CCA-Pro docked into the active site feature a Pro ligand rotated around the ester bond (A76O2′-C) and around the C-Cα bond, such that the secondary α-amine group is not positioned for favorable hydrogen bonds (Fig. 6C). The lack of favorable contacts is consistent with the poor Pro-tRNA^Pro inhibition of Ala-tRNA^Pro deacylation observed experimentally (Fig. S6).

These studies were enabled by the availability of a stable aminoacyl-tRNA acceptor stem substrate mimic. A significant obstacle to the structural determination of enzyme-aminoacyl-tRNA complexes is the difficulty in preparing sufficient substrate in a form that is nonhydrolyzable under physiological conditions. Thus, to date, no structures of synthetases or editing domains bound to aminoacyl-tRNAs have been determined. Two crystal structures of EF-Tu in complex with GTP and two different aminoacyl-tRNAs have been elucidated because the ternary complex protects the aminoacyl-ester bond from hydrolysis (33, 34). The ribosome is a ribonucleoprotein (RNP) complex that also binds aminoacyl-tRNAs. Nonhydrolyzable ACCA-3′-NH-Pro conjugates have been used to probe the unique properties of the cyclic structure of Pro as a ribosome substrate (35). Future studies using nonhydrolyzable aminoacyl-tRNA acceptor stem analogs should provide additional insight into substrate selection by editing domains, as well as other enzymes and RNPs that recognize aminoacyl-tRNA substrates.

Another family of trans-editing enzymes that act on misacylated tRNAs is the d-aminoacyl-tRNA deacylases (DTD), which remove d-amino acids mischarged onto a variety of tRNAs (36–38). Recent NMR studies using the stable posttransfer substrate analogs d- and l-alanyl-3′-aminoadenosine indicated binding of the d-analog but not of the l-analog (32). The achiral glycyl-3′-aminoadenosine substrate analog also bound DTD. These data are consistent with the efficient deacylation of Gly-tRNA^Gly by all DTDs tested. Interestingly, these deacylases appear to reject substrates with l-chirality rather than specifically selecting substrates with d-chirality. Overexpression of DTD is toxic to Ec cells, and EF-Tu likely protects Gly-tRNA^Gly from normally low levels of DTD, as has been demonstrated in vitro (32). Similarly, we hypothesize that in the cell, EF-Tu protects Pro-tRNA^Pro from the low levels of ProXp-ala deacylation observed in our in vitro assays (Fig. 1), as we have previously shown for Ala-tRNA^Ala (6). DTDs are structurally homologous to the N-terminal editing domain of archaeal ThrRS, which deacylates (l-Ser-tRNA^Thr), and in both cases the 2′-OH of A76 has been shown to play a catalytic role, activating a water molecule for the deacylation reaction (32, 39, 40). Thus, as for the ProRS editing domain homologs, the role of the RNA and the lack of critical amino acid side chains within the active site suggest a primordial origin for these free-standing editing domains (9, 17).

Given the highly congruous nature of amino acids, efficient deacylation of mischarged RNAs has likely been a hallmark of life since the genesis of RNA adaptor-based aminoacylation (40). Many extant organisms use specialized trans-editing enzymes, which supplement the activity of the cis-editing domains present in nearly one-half of ARSs. Double- and triple-sieve size-based substrate discrimination are the primary biological mechanisms adopted by editing proteins. Such mechanisms are evolutionarily advantageous; subtle changes in the identity of active site residues can result in altered substrate specificity, allowing for rapid genetic adaptation to environmental stresses. Using a unique nonhydrolyzable substrate analog, we show that Cc ProXp-ala deviates from the canonical sieve-based paradigm, instead using a strategy that includes size-based exclusion and conformational and chemical selection. This multipronged approach takes advantage of the unique stereochemical features of Pro to ensure robust discrimination against the cognate amino acid while retaining efficient deacylation of Ala-tRNA^Pro.

Methods

Amino Acid Volume Calculations.

Each standard free amino acid (Ala, Cys, Gly, Pro, Ser, and Val) was obtained from the PyMOL library and used with no adjustments. Aze was built in PyMOL starting from proline; the Cδ carbon and its two hydrogens were removed, and a new bond was built between the Cγ carbon and the nitrogen. Abu was built in a similar manner from leucine; both Cδ carbons and their hydrogens were removed, and two additional hydrogens were added to the Cγ carbon. The newly built Aze and Abu were subjected to two-step energy minimization using the MMFF94 force field. Molecular volumes for each amino acid were calculated by loading each PDB file into the Volume Assessor module of the 3V webserver (3vee.molmovdb.org/volumeCalc.php) using a 1.5-Å probe radius and high grid resolution (0.5 Å voxel size) (41).

Preparation of Cc ProXp-ala.

All Cc ProXp-ala mutants were generated using the QuikChange Site-Directed Mutagenesis Kit (Agilent). WT and mutant Cc ProXp-ala variants were prepared essentially as described previously (6). In brief, the gene encoding Cc ProXp-ala was cloned into pET15b (Novagen). Protein expression from Ec BL-21-CodonPlus(DE3)-RIL (Agilent) was carried out by induction with 0.1 mM isopropyl β-d-1-thiogalactopyranoside (Gold Biotechnology) for 16–20 h at room temperature. Uniformly singly labeled ([U-¹⁵N]) and doubly labeled ([U-¹³C, ¹⁵N]) ProXp-ala samples were prepared in M9 minimal medium supplemented with 100 µM FeCl₃, 1 µg/mL thiamine HCl, and 1 µg/mL biotin and containing 1 g/L [¹⁵N]-ammonium chloride (Cambridge Isotopes) as the sole nitrogen source or 3 g/L [¹³C]-glucose (Cambridge Isotopes) as the sole carbon source in addition to [¹⁵N]-ammonium chloride. His₆-tagged ProXp-ala was purified on His-Select Nickel Affinity Gel (Sigma-Aldrich) using a 5–250 mM imidazole gradient. The His-tag was cleaved using the Thrombin Cleavage Capture Kit (Novagen) following the manufacturer’s suggested protocol. Cleaved His-tag and any remaining His-tagged proteins were captured by His-Select Nickel Affinity Gel (Sigma-Aldrich). Protein samples were further purified by size-exclusion chromatography (SEC) using a Superdex 200 Increase 10/300 column (GE Healthcare) and either NMR buffer (50 mM sodium phosphate pH 7.5 and 10 mM NaCl) or AUC buffer (50 mM Hepes pH 6.8, 30 mM KCl, and 1 mM MgCl₂). Enzyme concentrations were determined using the Bio-Rad Protein Assay Kit.

Preparation of RNA.

Amino acids were obtained from Sigma-Aldrich, and [α-³²P]-ATP was obtained from PerkinElmer Life Sciences. Ec tRNA^Pro was prepared by in vitro transcription using T7 RNA polymerase as described previously (42) and 3′-[³²P]–labeled using tRNA nucleotidyltransferase as described previously (43). Preparation of aminoacylated Ec tRNA^Pro was carried out in 50 mM Hepes pH 7.5, 20 mM KCl, 4 mM ATP, 10 mM MgCl₂, 0.1 mg/mL BSA, and 20 mM β-mercaptoethanol by incubating 10 µM editing-deficient K279A Ec ProRS, 10 µM tRNA^Pro, and 0.03 mg/mL pyrophosphatase (Roche) with 900 mM Ala, 30 mM Pro, 300 mM Aze, or 600 mM Abu for 10 min at 37 °C. Aminoacyl-tRNAs were phenol chloroform-extracted, followed by ethanol precipitation. Ser- and Cys-tRNA^Pro were prepared using dinitro-flexizyme and Ser-3,5-dinitrobenzyl ester (Ser-DBE) or Cys-DBE as described previously (44). Val-tRNA^Pro was prepared using enhanced flexizyme and Val-4-chlorobenzyl thioester (Val-CBT) (44). Substrates for deacylation assays were dissolved in 3 mM sodium acetate at pH 5.2 and stored at −80 °C before use.

Synthetic microhelix^Pro RNA used in NMR titrations and AUC experiments was purchased from GE Healthcare. The nh-Ala-microhelix^Pro containing a hydrolysis-resistant 3′-amide-linked Ala moiety (Fig. 2) was synthesized as described previously (45) using a solid-phase synthesis approach with 3′-aminoacylamino-3′-deoxyadenosine–modified functionalized solid supports (46) and 2′-O-TOM nucleoside phosphoramidites (47). Before the experiments, RNAs were refolded by heating at 80 °C for 2 min then 60 °C for 2 min, followed by the addition of MgCl₂ to 10 mM and then slow cooling at room temperature and equilibration with either NMR or AUC buffer using SEC, as described above for protein preparation.

NMR Backbone Resonance Assignments and Chemical Shift Perturbation Analysis.

For NMR experiments, ProXp-ala samples in NMR buffer were supplemented with 10% D₂O (vol/vol) as a lock signal and 0.001% 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS) (wt/vol) as an internal chemical shift reference. All spectra were recorded at 25 °C. Titration experiments were performed by recording HSQC spectra on [U-¹⁵N]-ProXp-ala on a Bruker DRX-800 spectrometer equipped with a TXI cryoprobe. Either microhelix^Pro or nh-Ala-microhelix^Pro was added to 0.1 mM ProXp-ala at 0.2, 0.4, 0.6, 0.8, 1, 1.5, and 2 molar equivalents. Sequential backbone assignments of the free proteins were obtained from triple-resonance NMR spectra: HNCO, HNCA, HNCOCA, HNCACB, and CBCA(CO)NH (29). NMR spectra were recorded on [U-¹³C/¹⁵N]-ProXp-ala (0.75 mM) on a Bruker DRX-600 spectrometer, also containing a TXI cryoprobe. Signals corresponding to backbone HN, N, CA, and CB atoms for 156 of the 160 (98%) non-Pro residues could be assigned. Backbone assignments of ProXp-ala in complex with microhelix^Pro were determined by the same method using 0.63 mM each of ProXp-ala and microhelix^Pro, resulting in assignment of 153 (96%) of the non-Pro residues. Backbone assignments of ProXp-ala in complex with nh-Ala-microhelix^Pro were inferred by monitoring CSPs during the titrations. Signals corresponding to 140 (88%) of the non-Pro residues could be assigned in this manner. Data were processed with NMRPipe (48) and analyzed with NMRViewJ (49). Chemical shifts were referenced directly (¹H, ¹³C) or indirectly (¹⁵N) to DSS. Composite amide ¹H and ¹⁵N CSPs were quantified as Δδ (ppm) = (Δδ_H² + Δδ_N²/25)^1/2. NMR resonance assignments have been deposited in the Biological Magnetic Resonance Data Bank, www.bmrb.wisc.edu (accession nos. 27185 and 27191).

Lineshape Analysis.

Dissociation constants and off-rates for NMR titrations were determined from 2D NMR lineshape analysis using TITAN software (https://www.ucl.ac.uk/biosciences/departments/smb/christodoulou/titan) (50). From spectra of the titration series described previously (eight points with 0–2 molar equivalents), 15 residues from Cc ProXp-ala titrations with microhelix^Pro and 14 residues from titrations with nh-Ala-microhelix^Pro were selected for fitting to a two-state ligand binding model, followed by bootstrap error analysis.

X-Ray Crystallography.

Cc ProXp-ala was crystallized via hanging drop vapor diffusion in 0.005 M trisodium citrate pH 5.6, 15% glycerol, and 25.5% wt/vol PEG4000 (Hampton Research). Diffraction datasets (collected at the NE-CAT beamlines, Advance Photon Source) were indexed, integrated, and scaled with the NE-CAT RAPD automated protocol (https://rapd.nec.aps.anl.gov/rapd). The phase was obtained by molecular replacement with PHASER, using the structure of Cc ProXp-ala (PDB ID code 1VJF) as a search model. Three molecules were found in the asymmetric unit and improved with iterative cycles of refinement with PHENIX (51). Crystallization and refinement statistics are detailed in Table S1, and the coordinates have been deposited into the PDB (PDB ID code 5VXB).

¹⁵N Relaxation Measurements.

Steady-state ¹H-¹⁵N heteronuclear NOEs were obtained at 600 MHz at 25 °C using published pulse sequences (26–28). Samples contained 0.5 mM ProXp-ala, 0.5 mM ProXp-ala and microhelix^Pro, or 0.2 mM ProXp-ala and nh-Ala-microhelix^Pro in NMR buffer. The steady-state heteronuclear ¹H-¹⁵N NOE values were obtained by recording spectra with and without a ¹H presaturation period (9 s), in which all proton signals were saturated by a train of 90° pulses, applied before the start of the experiment. Intensity ratios were calculated using the hetNOE analysis tool within NMRViewJ (49). Owing to spectral overlap and/or lack of assignments, values could be determined for 136 of the 160 non-Pro residues of ProXp-ala, for 132 of 160 residues for ProXp-ala bound to microhelix^Pro, and for 130 of 160 residues for ProXp-ala bound to nh-Ala-microhelix^Pro.

¹⁵N CPMG Relaxation Dispersion.

Single quantum ¹⁵N CPMG relaxation dispersion experiments were carried out at 600 and 800 MHz at 25 °C using published pulse sequences (30) on a sample containing 0.7 mM ProXp-ala in NMR buffer. Spectra were recorded using a constant relaxation delay (Τ_relax) of 40 ms, with CPMG frequencies (ν_CPMG) ranging from 0 to 1,500 Hz (25 points) at 800 MHz and from 0 to 1,250 Hz (23 points) at 600 MHz. Three repeated frequencies at both 800 MHz (50, 250, and 1,250 Hz) and 600 MHz (50, 300, and 900 Hz) were included for error analysis. Spectra were processed using NMRPipe (48), and peak intensities were extracted from the pseudo-3D datasets by fitting resonances to a Gaussian model using the nlinLS function. Dispersion curves were obtained for 119 residues, 14 of which exhibited significant exchange (R_ex >2 s⁻¹). GUARDD software was used to fit the data to the Carver–Richards equation describing two-state exchange (52). The nine residues located in helix α2 and in the β1-α2 loop of ProXp-ala were globally fit to obtain a single exchange rate (k_ex) and major state population (P_A). The errors reported are estimated from 100 Monte Carlo simulations.

Deacylation Assays.

Aminoacyl-tRNA^Pro deacylation reactions were performed under single-turnover conditions at room temperature. Typical reactions contained 0.1 μM [³²P]-labeled aminoacyl-tRNA^Pro and 0.75 μM Cc ProXp-ala in deacylation buffer (150 mM potassium phosphate pH 7.0, 5 mM MgCl₂, and 0.1 mg/mL BSA). Reactions were initiated by mixing equal volumes of aminoacyl-tRNA^Pro and enzyme. At the indicated time points, 2-µL aliquots were quenched into a solution containing 2 units of P1 nuclease (Sigma-Aldrich) in 200 mM sodium acetate, pH 5.2. Product formation was monitored by separating aminoacyl-A76 from A76 on PEI-cellulose plates (EMD Millipore) using a mobile phase of 0.05% ammonium chloride/5% acetic acid. Radioactive products were detected by autoradiography using a Typhoon FLA 9500 (GE Healthcare) and quantified using ImageQuant TL 8 software (GE Healthcare). Observed rate constants (k_obs) were obtained by fitting the time course for aminoacyl-tRNA^Pro deacylation with a single-exponential equation using SigmaPlot (Systat Software). Each reported rate constant is an average of three independent assays.

Inhibition assays were performed under steady-state conditions at room temperature in 50 mM Hepes pH 6.8, 30 mM KCl, 5 mM MgCl₂, and 0.1 mg/mL BSA. Typical reactions contained 250 nM [³²P]-labeled Ala-tRNA^Pro, 500 nM or 1 µM inhibitor (nh-Ala-microhelix^Pro, Pro-tRNA^Pro, or tRNA^Pro), and 20 µM ProXp-ala. At the indicated time points, reaction aliquots were quenched and analyzed as described above, except that the rate of deacylation was determined by fitting the data points with a linear equation.

Analytical Ultracentrifugation.

Analytical ultracentrifugation was performed using an Optima XL-I ultracentrifuge (Beckman Coulter) with an An50 Ti rotor and standard double-sector Epon centerpieces equipped with sapphire windows. Sedimentation velocity experiments were performed at 50,000 rpm at 20 °C in AUC buffer, monitoring absorbance at 260 nm. AUC data analyses were carried out using SEDFIT and SEDPHAT, available at http://sedfitsedphat.nibib.nih.gov (19, 20, 53).

For K_d determination, samples contained 1 μM microhelix^Pro or nh-Ala-microhelix^Pro and Cc ProXp-ala at concentrations ranging from 0.2 to 10 μM. For each sample, the sedimentation profiles were first fit to the Lamm equation to obtain the sedimentation coefficient distribution, c(s) (53). The c(s) distributions were integrated to create an isotherm of signal-weighted average sedimentation coefficients, s_w, as a function of ProXp-ala concentration. Data were fit with mass action law models in SEDPHAT to obtain an estimated K_d, which served as a starting point for the global fit. The set of sedimentation profiles obtained for various ProXp-ala concentrations was then globally fit using the hetero-association model in SEDPHAT to obtain the K_d and sedimentation coefficient of the binary complexes. The sedimentation and extinction coefficients for unbound ProXp-ala and each RNA species were obtained from independent experiments and remained fixed during the global fit.

For analysis of ProXp-ala mutants, relative binding affinities were determined using a semiquantitative AUC-based assay. Sedimentation velocity experiments were performed as described above using 1.2 μM, 2.4 μM, 4.8 μM and 9.6 μM Cc ProXp-ala. The c(s) and s_w values were obtained for each mutant, and the s_w values were compared with those of 1.2 μM WT ProXp-ala to estimate relative binding affinities, as described in Fig. S1.

Computational Modeling.

The starting structure for docking simulations was prepared using the Ef ProRS crystal structure (PDB ID code 2J3L; residues 233–402) (22) as a template for homology modeling using SWISS-MODEL (54–56). This structure was used rather than the known crystal structure of Cc ProXp-ala (PBD ID code 1VJF) because the latter was crystallized with the α2 helix in the open rather than closed conformation. Semiflexible docking was performed using Autodock Vina (57) with the CCA-Ala ligand built in Avogadro (58). All rotatable bonds between nucleotides and within the amino acid were left unrestrained to allow for unbiased docking geometries. Based on homology with the INS domain of Ef ProRS and other INS-like domains (6), as well as CSP data, the side chains of the following active site residues were set to be flexible: Arg28, Val29, Lys45, Asn46, Leu47, Phe83, Gln86, Met89, Val100, and His130. Docking into a 25-Å cubic volume centered on the active site resulted in 2,000 successful docking poses, with ∼8% (162 poses) oriented such that the Ala moiety was in the active site and all atoms from the RNA were outside of the pocket as determined by visual inspection. Of these, a large cluster of 78 had a nearly identical position of the Ala atoms (heavy atom rmsd <1 Å), with the three best-scored docking poses also having nearly identical nucleotide positioning (heavy atom rmsd <1 Å). The top scoring pose was chosen as the representative for further analysis. Standard explicit solvent MD simulations were carried out using the AmberTools14 and AMBER14 program suites (59). The AMBER ff14SB force field was used for all nucleosides (60, 61). Constructs were first neutralized with sodium ions and solvated in an octahedral box of TIP3P water (62) with a 10.0-Å cutoff.

Minimization and equilibration were each conducted in two steps. During the first minimization, 2,000 total steps were performed, with 1,000 steps of steepest descent and a particle mesh Ewald (PME) implementation of constant volume periodic boundaries (63). The solute was held fixed using a positional restraint of 500 kcal/mol⋅Ų. In the second minimization step, 5,000 steps were performed, with 1,000 steps of steepest descent, and the solute was left unrestrained. The system was gradually heated from 0 K to 300 K in the first step of equilibration for a total of 100 ps at a time step of 2 fs while holding the solute fixed with a weak positional restraint of 10 kcal/mol⋅Ų. The PME implementation for a constant pressure periodic boundary was used for the second equilibration and production simulations. The SHAKE algorithm was used to constrain all bonds involving hydrogen (64), and Langevin dynamics was applied at a collision frequency of 1.0 ps⁻¹ to control the temperature of the system. A second equilibration was used to relax the system for 900 ps under production conditions. Production simulations were performed in steps of 10 ns until the rmsd vs. time plot from the initial structure leveled off, indicating equilibrium. A further 40 ns of production simulation in steps of 10 ns was performed after equilibrium. All data analysis and statistical calculations were performed in the final 40 ns of each simulation. A random seed was used for all equilibration and production simulations to alleviate the possibility of synchronization artifacts of the simulations, which may result from the use of Langevin dynamics (65, 66). For all steps of simulation, a nonbonded cutoff of 8.0 Å was used. All equilibration and production simulations were performed using the GPU accelerated implementation of the PME MD in the sander module of Amber14 (67). PyMol (68) and VMD (69) were used for structure and trajectory visualization. Investigation of the active site conformations was done on snapshots taken at 30 ns, to ensure that the structures were stable for at least 10 ns before and after the selected timepoint.