We examined tRNA flexibility using a combination of steered and unbiased molecular dynamics simulations. Using Maxwell's demon algorithm, molecular dynamics was used to steer X-ray structure data toward that from an alternative state obtained from cryogenic-electron microscopy density maps. Thus, we were able to fit X-ray structures of tRNA onto cryogenic-electron microscopy density maps for hybrid states of tRNA. Additionally, we employed both Maxwell's demon molecular dynamics simulations and unbiased simulation methods to identify possible ribosome-tRNA contact areas where the ribosome may discriminate tRNAs during translation. Herein, we collected >500 ns of simulation data to assess the global range of motion for tRNAs. Biased simulations can be used to steer between known conformational stop points, while unbiased simulations allow for a general testing of conformational space previously unexplored. The unbiased molecular dynamics data describes the global conformational changes of tRNA on a sub-microsecond time scale for comparison with steered data. Additionally, the unbiased molecular dynamics data was used to identify putative contacts between tRNA and the ribosome during the accommodation step of translation. We found that the primary contact regions were H71 and H92 of the 50S subunit and ribosomal proteins L14 and L16.
tRNA is a key component for protein synthesis in the cell. tRNA delivers amino acids to the ribosome, where they are incorporated to the growing nascent polypeptide chains. The characteristic L-shaped tertiary structure of tRNA is intimately related to its function, and it has intrigued investigators for decades [
With the first high-resolution crystal structure for tRNA [
tRNA Conformations. Both kinked structures for tRNAPhe (PDB code: 1EHZ) and the hybrid state for cryo-EM densities of tRNA are shown. The different orientations of tRNA depict the unique differences caused from kinking or twisting (A/T and P/E states, resp.). The arrows depict the transition from A/T state to A-site (A/A state) to P-site and P/E hybrid state. (a) Secondary structure for Phe-tRNAPhe (1EHZ) [
The first concrete experimental evidence of tRNA flexibility came with determination of the tRNAAsp crystal structure for which an interarm bend angle of approximately 89° was observed, larger than that observed in tRNAPhe [
tRNA flexibility has been a motivating topic for early molecular modeling studies on tRNA [
MD simulations have been successfully characterized as structural dynamics, base pairing, specific hydration, cation binding, and electrostatic potential distributions, including mutagenesis [
In the present study, we report global range of motion for tRNA derived from molecular dynamics simulations. The simulation method was a novel application first reporting the conjunction of X-ray data driven toward that of the shape given from the cryo-EM density map (Biophysical Society Meeting 2006 and 2007) [
Multiple simulations of aa-tRNAs were completed using NAMD version 2.62 with the CHARMM27 force field, and then the simulation was conducted in duplicate using NAMD version 2.62 with the AMBER force field [
Several tRNA simulations were run only using neutralizing Na+ ions. These were initially placed using the Xleap module of AMBER9 at the positions of the lowest electrostatic potential. In one case, we neutralized with 76 Na+ counterions to the tRNA. In another case, we neutralized with counterions and then created a solvent with 150 mM Na+ Cl- to recreate physiological strength. We observed similar tRNA flexibility in both cases.
The AMBER force field parameters for the naturally occurring modified nucleosides of RNA were obtained from the web-server at the lab of SantaLucia and from the Sanbonmatsu lab (personal communication) [
Simulations were carried out using the particle mesh Ewald technique (PME) with repeating boundary conditions in a box approximately 70–105 Å3 with 9 Å nonbonded cutoff [
Translational and rotational center-of-mass motions (COMs) were initially removed. Periodically, simulations were interrupted to have the COM removed again by a subtraction of velocities to account for the “flying ice-cube” effect [
In the MdMD algorithm, we run a short “sprint” of MD (typically ranging from 50 fs to 5 ps) [
This application of MdMD allows the user to alter the shape of X-ray crystallographic structures to match the cryogenic-electron microscopic data, which may present an alternative conformation of the structure. In doing so, the cryo-EM density can drive the MD toward an unknown conformation. By automating this process, human biases and errors are minimized for the model making process. The first step is conversion of the MD all-atom structure into a representative low-resolution cryo-EM density, and the second step is the comparison of the MD-generated density to the target cryo-EM density. If the correlation between the MD density and the target density increases, then the iteration is saved and the process continues; otherwise the structure returns to the prior state and another MD sprint is carried out. We verified the correlation of the densities using both SPIDER and NMFF [
The CC represents the numerical correlation between the electron density of the target cryo-EM density map and the simulated density map from the current MD structure. The CC is defined as follows:
We compiled a set of tRNA crystal structures from structural databases to quantify any tRNA flexibility from experimental data. The Nucleic Acid Database and the Protein Data Bank RCSB [
tRNA Catalog. (a) Six crystal structures of unbound Phe-tRNAPhe. The ribbon in the middle is from 1EHZ. (b) Crystal structures (41) of tRNA free and in complexes, including synthetases and ribosomal (Table S1) aligned at the acceptor stems and the anticodon stems. The ribbon in the middle is from 1EHZ. (c) Phe-tRNAPhe crystal structures (6) aligned at the acceptor stem (shown in divergent stereo), thus illustrating the range of motion at the anticodon stem loops, and notice the closely aligned acceptor stem regions.
We used the interarm angle between the anticodon stem and the acceptor stem as a measure of flexibility. There were eight species of tRNA for which multiple X-ray crystal structures were available: Asp, Glu, Gln, Met, Phe, Trp, Tyr, and Val. All eight of these had similar ranges of interarm angles, somewhere between 70° and 100°, which suggests that interarm flexibility is an intrinsic feature of the topology of tRNA and does not depend strongly on the amino acid bound to the acceptor stem (Supplementary Table S1).
In addition, we computed the RMSDs of the crystal structures, relative to 1EHZ tRNAPhe, and found a range from 0.5 Å to approximately 6 Å (Supplementary Table S1) for superposition of the main stems. Consequently, on a residue-by-residue basis, the greatest regions of flexibility were the 3′-CCA end of tRNA and the anticodon stem-loop (ASL) (Figure
Jittergram of kinked A/T-tRNA from molecular dynamics simulations. (a) Using multiple unbiased molecular dynamics for kinked A/T-tRNA, we generated a set of structures. These are superposed based on all P-atoms found in the backbone of the stem regions. This orientation optimally shows the alignment of the stems and the variation at the acceptor stem loop and anticodon stem loop. The different tRNA colors help distinguish the different conformations. (b) Illustrated in cyan as a series of snapshots is a simulation for kinked A/T-tRNA using free molecular dynamics with tethering constraints at the anticodon atoms (H-bond atoms of nucleotides 34, 35, 36), which was run for >10 ns. We used tethering (harmonic constraints) at the anticodon atoms to mimic base pairing at the codon. This data was used for the contact map (Figure
Cryo-EM reconstructions of tRNA bound to the 70S
We examined four distinct tRNA conformations derived from cryo-EM density maps. These tRNA conformations included the A, P, A/T, and P/E states. Using Maxwell’s demon Molecular Dynamics (MdMD), the starting structure of the native tRNA structures (1EHZ) is driven into conformations that match the cryo-EM data based upon a cross-correlation calculation between a theoretical density for the modeled structure and the experimental cryo-EM density. MdMD is derived from existing methods in the literature for biasing simulations [
The steps are archived into a single trajectory based upon the outcome of the progress variable. During the MD sprint the variable is free to fluctuate into space that would otherwise be forbidden by the Maxwell demon.
In the case of cryoEM-fitting with MdMD, our progress variable is the cross-correlation between the electron density of the simulation structure and the experimental density from cryo-EM, where we rejected MD sprints that decrease the cross-correlation.
Using MdMD, we obtained pathway transitions from the crystal structure state (1EHZ) to the cryo-EM states A, P, A/T, and P/E using less than 10 nanoseconds of molecular dynamic simulation (Supplementary Movies S4, S6). From these simulations, the ensembles of structures are consistent with the density distributions from the cryo-EM density maps. Figure
Cryogenic-electron microscopy fitting using Maxwell’s demon Molecular Dynamics for tRNAs A/T, A, P, and P/E is shown. All tRNA atomic models are shown in ribbons/CPK, while densities are as solid. (a) A/T-site structure for tRNA fit to density, cross-correlation coefficient (CCC) of 95% between the structures theoretical density and the cryo-EM experimental density (Divergent stereo). (b) A-site structure for tRNA fit to density with a CCC of 83% (Divergent stereo). (c) P-site structure for tRNA with a CCC of 86% (Divergent stereo). (d) P/E-site structure with a CCC >90% (Divergent stereo).
The cross correlations between the fit structures and the maps give a good indication that the fit structures are not deformed. Specifically, the pathways between the fit structures reveal the transitional motion, dynamically, between modes of tRNA: from A/T to A, from A to P, and from P to P/E, respectively, exploring stochastically reversible excursions along a tRNA pathway between experimentally verified states [
We carried out two sets of unbiased MD simulations on tRNA: (1) We ran ~0.5
We repeated this simulation with a tethered anticodon to mimic the hydrogen bonding with mRNA. We imposed restraints on anticodon nucleotides 34, 35, and 36 (1 to 10 kcal/(mol*Å2) per hydrogen bonding atom). One would expect that this tether would result in dampening of tRNA motion, because the D-loop, T-loop, and acceptor-stem must rotate in order to relieve the twist in the anticodon stem. This simulates the effect of being bound to the ribosome (Supplementary Movie S2). This situation is similar to the simulation of Sanbonmatsu et al., except that there is no ribosome present to hinder swinging, and no biasing forces were used [
From our unbiased MD simulations, we found that tRNAPhe unkinked in the 2 ns following equilibration, when the anticodon atoms were unrestrained. However, when the anticodon was restrained, unkinking occurred between 6 and 10 ns. We also found that simulations of the A/T to the A/A state provide an approximate range for tRNA swinging-type motion that agrees with the structural data in our database of crystal structures.
Our results suggest that tRNA deforms quite readily at the anticodon stem from an initial free tRNA state. We found that (1) the kinked A/T structure’s trajectory converged toward the crystal structure of tRNAPhe (Figure
Figures
Various snapshots from multiple simulations of A/T-tRNA and native tRNA are shown, where the time sampled over is >200 ns (Supplementary Movie S1). (a) A/T-site tRNA (shown in purple) moves toward the A-site tRNA (red) which is shown in reverse view of the ribosome density, to illustrate the >70 Å motion covered by the 3′-CCA acceptor end. The tRNA shows an oscillation about the A-site tRNA structure (red) (Divergent stereo). (b)
RMSD graph for three types of tRNA simulations presented in this study. This graph represents the root mean square deviation between the simulation of tRNAPhe with that of the crystal structure for tRNAPhe (1EHZ) [
The RMSD values from multiple simulations of kinked and unkinked tRNA, as well as native tRNA, show that these different conformations often converge within 10 ns, indicating that the favored conformation in all cases occurs in a region of conformational space similar to the relaxed native structure (Figure
We docked the unbiased MD simulation of tethered tRNA into the ribosome in order to study contacts between tRNA and the 70 S
Recent data presented in the literature demonstrates a low-resolution structure for the 80S P-site tRNA that suggests a state in which the anticodon stem-loop is significantly bent [
Our simulations of tRNA support a complex flexible hinge motion between the anticodon stem and acceptor stem with a characteristic relaxation time on the nanosecond timescale. The hinge is centered on nucleotides 26, 44, and 45, and additionally there are complex stem deformations visible in the dynamics movies. In simulations where the anticodon was tethered, which mimic hydrogen bonding to the mRNA codon during translation, we found that the correlation time for rotational diffusion of tRNA about the fixed codon-anticodon duplex was approximately 10 ns. We found that tethering the anticodon bases in space resulted in diffusion being driven by an unkinking force, which was able to swing the entire tRNA 3′-terminus along a trajectory mimicking tRNA accommodation in the ribosome during translation. As the A/T-tRNA moved away from the kinked state, it progressed into a range of conformations similar to those found in other catalogued crystal structures.
Our results confirm that tRNA has a stochastic molecular spring-like motion from the biasing method [
Here, we report that a Maxwell demon-type algorithm that acts externally to the potential one may utilize experimental cryo-EM densities to “drive” molecular dynamics simulations into preferred conformations. This result is promising for future work to obtain transition pathways. We found that our biased MdMD method can sample multiple hybrid conformations of tRNA that have been experimentally observed from cryo-EM and X-ray crystallography, thus bridging between two experimental methods.
Finally, we constructed a putative ribosomal contact map from the microsecond dynamics of tRNA bound to the ribosome. We observed contacts with H71 and H92 of the 50S subunit and ribosomal proteins L14 and L16. It is possible that these areas of the ribosome act to filter near-cognate tRNA in a test of anticodon base pairing during accommodation.
This work was supported by a GAANN and Sam Nunn-MacArthur Foundation doctoral fellowship to T. R. Caulfield., and by the College of Computing Computer Resources Allocation at the Georgia Institute of Technology, and by the Bluegene/L computing facility at the UAB Center for Computational and Structural Biology at the University of Alabama at Birmingham. The authors thank M. Wolf of the College of Computing at Georgia Institute of Technology for resource administration, M. Hanby, M. Perez, and J. Segrest at U.A.B. for being computational liaisons for the Bluegene/L, and Drs. S. C. Harvey and A. Petrov for insightful contributions to computer simulations, ideation, and collaboration.