Transfer RNA in the hybrid P/E state: Correlating molecular dynamics simulations with cryo-EM data

Overall Conformation of the P/E-tRNA.

The isoleucyl P/E-tRNA in the cryo-EM map [supporting information (SI) Fig. 5 a and b] is structurally different from all x-ray structures of tRNA, including the P-tRNA in a complex with the 70S ribosome (ref. 2; from Thermus thermophilus; PDB ID code 2I1C; SI Fig. 5 c and d), an unbound form (ref. 17; from yeast; PDB ID code 1EVV; SI Fig. 5 e and f), and the tRNA in complex with isoleucyl-tRNA synthetase (IleRS) (ref. 18; from Escherichia coli; PDB ID code 1QU2; SI Fig. 5 g and h). For the purpose of the comparison, each of these x-ray structures was converted into an electron density map at the same resolution as the cryo-EM map, 11.7 Å, and then separately superimposed on the cryo-EM map of the P/E-tRNA. The superposition reveals differences in (i) the CCA stem, (ii) the T/D loops, and (iii) the ASL. Regardless of whether the x-ray structures of tRNA are from the free form, from a ribosome-bound complex, or from a synthetase-binding complex, their T/D loops are oriented differently from the twisted position observed in the P/E state, a position that affects the orientation and appearance of the CCA stem as well.

The anticodon stem in the P/E-tRNA is kinked at the places where the stem is connected to both the D stem and the ASL. The formation of this kinked geometry must be favored by the interactions of tRNA with the ribosome. Generally, the deformation of the anticodon stem must be induced by the interactions between the anticodon stem and its binding counterpart, such as in the complex of IleRS with tRNA^Ile. Because of the specific binding between tRNA^Ile and IleRS, the anticodon stem twist in that complex is unique among tRNAs in all x-ray forms studied thus far.

It is interesting to note that the conformation of the D stem in the unbound x-ray form is similar to that of the P/E-tRNA, as evident from a superposition of these two maps (SI Fig. 5). As will be shown later, the D stem in the P/E-tRNA is no longer bound with helix H69 of the 50S subunit so that it is indeed in an unbound form, just as in the free tRNA. In contrast, in both the binding complexes of the P/P-tRNA with the ribosome, where it interacts with helix H69, and the aa-tRNA with the aa-tRNA-synthetase, the conformation of the D stem is influenced by its binding position with respect to the binding counterpart.

Fitting of backbones of the various x-ray tRNAs provides a good indication as to what extent they match the P/E-tRNA density map. When the backbone of the P-tRNA (ref. 2; PDB ID code 2I1C) is superimposed onto the P/E-tRNA map, obvious mismatches occur in the T loop, the D loop, the anticodon stem around nucleotide 26, the ASL, and the CCA end (SI Fig. 6 a). Similar discrepancies are observed in the superposition of the synthetase-bound aa-tRNA (ref. 18; SI Fig. 6b; PDB ID code 1QU2) and the unbound form of tRNA (SI Fig. 6c; PDB ID code 1EVV). It is clear that these discrepancies would be even larger when all atoms of the tRNAs were included in the comparison. The P/E-tRNA shows a unique conformation in its ASL as well as in the twisted CCA arm, from the T loop to the CCA end.

MD Simulations: Searching for a Representative Structure for the P/E-tRNA.

In an attempt to characterize the structure of P/E-tRNA, we performed MD simulations on tRNA in the unbound form using the atomic coordinates of the tRNA from T. thermophilus (ref. 19; PDB ID code 1TTT) as the starting structure (see Experimental Procedures). Three trajectories were obtained: two covering a time period of 18 ns and one covering 44 ns.

The structural mobility of all nucleotides throughout the three MD simulations were characterized by (i) the size of the fluctuations of the phosphate atom in each nucleotide (in terms of the distances relative to the average coordinate, averaged among all snapshots in the three trajectories; Fig. 1a) and (ii) positional deviation of each nucleotide, in terms of its rmsd (the distances of each nucleotide relative to the simulation starting structure, averaged among all snapshots; Fig. 1b).

In general, all nucleotides in tRNA are mobile. The minimum fluctuation among phosphate atoms is larger than 1 Å, and the maximum is 3.2 Å, implying that there are significant changes in the backbone shape. When we disregard the notoriously large, expected fluctuations at the 3′ and 5′ ends, the anticodon is the most unstable element in the simulation of the free tRNA structure.

The simulation results also indicate that tRNA in the free form deviates extensively from the x-ray structure. Particularly nucleotides 4–5 and 33–43 vary strongly from their x-ray positions. In addition, the nucleotides in the D loop and the T stem loop have large rmsd values, implying that these nucleotides also vary strongly from their x-ray positions, even though their fluctuations are not the largest among the nucleotides. The phosphate atoms in these nucleotides deviate from their x-ray positions by 2 to 3.5 Å. These results are correlated with the sizes of B factors in the x-ray structure: large values for the D and T loops as well as in the variable loop. The largest rmsd of an individual nucleotide relative to its x-ray position is 4.5 Å, so some of the local deviations from the x-ray structure are quite substantial.

The large changes in the conformation of tRNA along the simulation trajectories raised the expectation that we might be able to find snapshots representative for the P/E-tRNA observed in the cryo-EM map. To search for such snapshots, we used the procedure described in Li et al. (20): selected snapshot structures were converted into density maps at the same resolution as the cryo-EM map. These snapshot-derived density maps then were aligned with the cryo-EM density of the P/E-tRNA by using a motif search procedure (21), each time yielding a cross-correlation coefficient (CCC) for the matching position. Because the 3D alignment of the density-converted snapshots is very time consuming, only a sparse subset was tested, selected from the full set of MD-simulated structures of one trajectory at time intervals of ≈1 ns, making a total of 17 snapshots. The no. 1 candidate gave the highest CCC of 0.86, compared with the x-ray structure (CCC = 0.76).

To find snapshots with similar structure as this no. 1 P/E-tRNA candidate structure, we performed an exhaustive rmsd calculation, extending over all three MD simulation trajectories, between the no. 1 P/E-tRNA candidate and all of the MD simulation snapshots (SI Fig. 7). In this search, we deemed an average rmsd <1.5 Å as indicating high structural similarity. We found only a few matches within the three trajectories, covering a total of 80 ns (two of the three are shown in Fig. 2 a and b), and the no. 1 candidate still remained one of two highest-ranking snapshots. Superposition of these matching structures on the cryo-EM map (Fig. 2c) confirms that all are sufficiently close to explain the appearance of the density map.

The candidate snapshot leads to an improved match of the tRNA with the cryo-EM map in three regions (Fig. 3 a and b): the D loop, the D stem, and the ASL. When this snapshot is superimposed exactly onto the x-ray structure at the ASL (where tRNAs in the P/P and P/E positions would coincide if the ratchet motion and other effects were ignored), the discrepancy evidently starts at the transition between the ASL and the D stem, at position 26 of the nucleotides. The D loop stem and the CCA arm are seen to be twisted toward the E site on the large subunit. The discrepancy increases as we move from the starting point, nucleotide 26, toward the CCA arm. By virtue of this twist, nucleotide 55 in the elbow region shifts by ≈15 Å, and nucleotide 19 shifts by ≈8 Å. Even though the change in tRNA conformation moves the CCA end toward the observed P/E position (Fig. 3c), the movement is not enough, by far, to reach the observed P/E position on the large subunit, at which the T loop is ≈40 Å away from its position in the P site (Fig. 3d). Thus, the additional shift is caused by the fact that the ASLs of the two tRNAs, in defiance of the crude assumption underlying the geometry shown in Fig. 3c, coincide neither in position nor in orientation (Fig. 3d). The positional difference (measured as 5 Å) must be attributed to the ratchet motion, whereas the difference in orientation must be on account of a change in mRNA configuration or mRNA–tRNA interaction.

It should be noted that the procedure used to find matching conformations through an rmsd comparison with a selected candidate structure is somewhat limited, as demonstrated by the case of the snapshot circled in Fig. 2b. Despite the fact that it gives high correlation with the P/E-tRNA density, of a value in the same range as candidate no. 1, its rmsd with respect to candidate no. 1 is larger than the 1.5-Å margin and so was rejected by the rule. It also is not strongly distinguished in rmsd value from its group of neighbors. This example illustrates the intuitively obvious fact that high correlation with a low-resolution density map still allows structures to differ significantly among themselves in rmsd. It indicates that an exhausting hunt for matching structures explaining the observation of the density can only be accomplished by alignment and cross-correlation of the density itself with all structures of the MD simulation trajectory.

Nevertheless, even without a more certain estimate of frequency, the result that the occurrence of the P/E conformation is a relatively infrequent event along the MD simulation trajectories is sensible. Clearly, this conformation is not a preference for the unbound tRNA, and the P/E-tRNA conformation will only be productive for proper binding when the ribosomal context (i.e., in terms of the intermolecular interactions it allows) is such that it stabilizes this particular conformation of tRNA.

Evidence for Conformational Heterogeneity in the P/E-tRNA Density Map.

Before describing the contacts of the P/E-site tRNA with the ribosome (see next section), we first will show some evidence for conformational variability. When the snapshot identified as representative for the P/E tRNA conformation is converted into a density map at the matching resolution, it is seen that the cryo-EM density has reduced dimensions and overall smoother contours (Fig. 4a versus c). The differences in size and appearance can be understood as a result of conformational averaging. Instead of representing one unique atomic structure, the cryo-EM map in fact represents an ensemble average of molecules assuming a large range of conformations. We made an attempt to simulate this effect by averaging over density maps generated from 21,500 aligned snapshots of trace no. 2 (44 ns) by using the same resolution cutoff as the EM map (Fig. 4 b versus c and d). Evidently, this simulated averaged density matches the cryo-EM density better, showing both the shrinkage in size and the smoothing of contours. (Note that the very CCA end is not included in the separated cryo-EM P/E-tRNA map because of the difficulty of segmenting the single-stranded structure from the ribosome density. In the averaged map from MD snapshots, the CCA end is missing for a different reason—because of its high instability in MD simulations of the unbound form of tRNA.)

Conformational averaging affects different parts of the structure differently. The effect of smearing out the local structure is a function of the structure itself. For instance, single-stranded regions and loops generally are more mobile than double helices, hence they have larger rmsd values from the average structure of the ensemble. These effects become quite obvious when looking at the density map with high threshold, as the cryo-EM map is seen to shrink nonuniformly and even break apart in some areas (Fig. 4 e versus g). Raising the threshold of the density-converted candidate P/E-tRNA MD simulation snapshot yields a skeleton structure whose features reflect the shape of the region where the low-resolution density is highest. Upon raising the threshold, the remaining core density in the averaged map is not completely congruent with the cryo-EM map, indicating that the two maps are not composed of identical structural mixes (Fig. 4 f–h). The cryo-EM density at the anticodon stem is much weaker than that in the averaged map, indicating that it may occur with greater conformational diversity in the experiment than among the snapshots along the MD simulation trajectories.

Another obvious difference between simulated and experimental density is in the ASL, which shows a breakup in the averaged map but a remaining density in the cryo-EM map. Based on the trajectory analysis (Fig. 3), the ASL has the largest fluctuations (again discounting the expected large fluctuations at the 3′ and 5′ ends), which is indeed where the largest density thinning and breakup effects occur in the simulated ensemble average map (Fig. 4g). The difference between the appearance of the ASL in the average and in the P/E-tRNA of the cryo-EM map is probably attributable to the fact that the ASL in the latter case is moderately stabilized by codon–anticodon interactions (see Discussion).

Interactions with the Ribosome.

The P/E-tRNA forms specific interactions with the ribosome not seen for P/P and E/E. Based on the x-ray study (2, 4), the P-site anticodon stem interacts with the 30S subunit through A790 (to nucleotide 33), nucleotides 1229, 1230, and 1341 of 16S rRNA to the anticodon stem at nucleotides 30–31, and nucleotides 1338 and 1339 of 16S rRNA, as well as A1340-U1341, with base pairs 29–41 and 30–40. The x-ray structure shows that the A790 loop and G1338-A1339 form a barrier around the ASL that locks the codon–anticodon pairs into a stable position and prevents P-tRNA from moving toward the E site (SI Fig. 8). This barrier also is observed in a closed position in the accommodated state of aa-tRNA by cryo-EM (10) and in the x-ray structure of the ribosome in the absence of tRNA (1). The latter authors (1) predicted that the barrier must be opened in a translocating ribosome. The current study, as well as another cryo-EM study of the translocating 80S ribosome (22), provides evidence that this is indeed the case: the barrier appears opened by ≈8 Å (SI Fig. 8). The widening of the density gap is the result of changes on both sides of the gap. On one side, the density for the G1338–A1339 loop obviously is rotated because of the ratchet-like rotation of the 30S subunit. On the other side, density for the A790 loop has virtually disappeared, indicating that this loop is unstable.

The weakening of the density in the loop region implies that these loops are mobile structural elements during translocation. In addition, the conformation of helix h44 around A1492–A1493 and the connection of this helix to helix h45 are reshaped such that the space around the ASL is opened further (SI Fig. 9). As discussed above, a kink and twist are observed around base pair 26–41 of the P/E-tRNA, bending the D stem from the anticodon stem by ≈30° toward the E site. A slight kink, and a twist by 10°, also are observed in the x-ray structure at the same base pair (2). The distortion in the helical geometry of the anticodon stem in the P/E-tRNA is much more substantial compared with the P-tRNA. As a result, the ASL is brought completely out of the spatial barrier that confines the tRNA in its x-ray P/P conformation (2), making it possible for the tRNA to be translocated into the E site.

At the E site in the 50S subunit, the x-ray structure (2) shows a direct contact between base pair G19/C56 of E-tRNA and A2169/G2112 in helix H78, and E-tRNA at position 71 interacts with U1851 and C1892 in helix H68 (arrows in SI Fig. 10a). The T/D loops of the P/E-tRNA reach the position A2169/G2112 in helix H78, as in the x-ray structure, but the cryo-EM map reveals a major change in the E site, with both helix H68 and the L1 lobe assuming new conformations. Significantly, the P/E-tRNA is completely free from binding with helix H69 (SI Fig. 10b), unlike A-, P-, and A/T-tRNAs, whose binding interactions have been discussed previously (9–11). At the same time, helix H69 becomes highly mobile, as indicated by its shrunken density (SI Fig. 10b). The mobility of helix H69 goes hand in hand with an altered conformation of helix H68. The x-ray structure of helix H68, if taken as one rigid fragment, does not match the cryo-EM map in its expected location (SI Fig. 10 c and d). This discrepancy points to several major changes in the conformation of helix H68 (arrows in SI Fig. 10 c and d). First, the junction with helix H69 appears in a different orientation (SI Fig. 9d, arrow on left). Second, helix H76, instead of forming a direct contact with helix H68 as in the x-ray structure, forms a contact with the L1 protein following the motion of the L1 stalk (SI Fig. 10d, arrow in center). This shift releases nucleotides near the ending GCAA tetraloop of helix H68. Third, the tetraloop of helix H68 is bent away from the L1 lobe (SI Fig. 10d, arrow on right). Finally, the P/E position leaves the CCA arm of the tRNA at a distance from the helix H76–H77 stem, producing a density gap between the CCA arm and the helices (SI Fig. 10b).

Correlative Analysis of the P/E-tRNA.

In this study, we have expanded the approach of correlating cryo-EM map with MD simulation (20) and found structures resembling the observed density of P/E-tRNA within 80-ns MD simulation of free tRNA. The flexibility of the unpaired CCA end, showing large fluctuations among the MD structures (Fig. 1), might be critical for its ability to move from the P to the P/E state. There seems to be a similar requirement for the flexibility of the CCA end in the accommodation process [see Sanbonmatsu et al.'s (23) MD simulation of the aa-tRNA movement from A/T to A/A].

It is important to realize that numerous atomic snapshots that possess different conformations might fit a low-resolution cryo-EM map with equally high CCC. Therefore, when we search through the MD simulation trajectories for conformations close to the selected candidate structure by using an rmsd = 1.5 Å cut-off, this procedure has likely excluded many equally well matching snapshots outside the rmsd cut-off range.

Dynamics of P/E-tRNA Inferred from Density Analysis of Cryo-EM Map and Averaged MD Snapshots.

The analysis of the P/E-tRNA density at high thresholds indicates that the ensemble of structures in the EM specimen giving rise to the map is heterogeneous, and that the conformational heterogeneity is localized in the anticodon stem and loop, implying the existence of highly active motions of the tRNA in the hybrid state. We conclude that the ASL in the P/E-tRNA is destabilized during translocation, even though the tRNA stays bound with the codon and the ribosome. This behavior contrasts with the observed stability of the codon-bound anticodon in the P/P-tRNA in the x-ray structure (2) as indicated by small B factors at this location. Thus, the codon–anticodon pairing is not retained in the same stable form throughout the translocation process. This result agrees with a biochemical study (16) that indicated instability of codon–anticodon base-pairing during translocation. The mobility of the D loop of the P/E-tRNA observed in the current study is consistent with single-molecule FRET findings (14–15) showing that the D loop oscillates between the P- and the P/E-states.

The Kink and Twist in the Anticodon Stem of tRNA.

A deformed anticodon stem previously was observed in both the A/T-tRNA (9, 10) and the P-tRNA (2). In the current study, both the kink and turn in the P/E-tRNA are found to be as large as in the A/T-tRNA but with the turn going in the opposite direction (SI Fig. 11). The one commonality between the two kinked, turned tRNA conformations is the position at which the kink occurs, namely around nucleotides 26 and 44, where the D stem ends and the anticodon stem starts. However, the functional roles of the kink and turn in the ASL of both the A/T-tRNA and the P/E-tRNA are evidently quite different. In the A/T-tRNA, the deformed conformation ensures that the anticodon can reach the position of the codon. It also may be required for proofreading, as evidenced by mutation experiments (24, 25). In contrast, the deformed conformation in the P/E-tRNA enables the ASL of the P-tRNA to pass the spatial gate described by Schuwirth et al. (1).

The MD simulation results in this study indicate that an unbound tRNA may spontaneously form a kinked, turned conformation of the anticodon stem as shown in the atomic representation of the P/E-tRNA. However, interestingly, the turn in the direction observed in the A/T-tRNA is not found throughout our MD simulations.

Interaction Between P/E-tRNA and the Ribosome.

The P/E-tRNA directly contacts the ribosome at the following three sites: the anticodon with the spatial gate composed of the A790 loop and nucleotides 1338–1340 in the head of the 30S subunit, the T loop with helix H78, and the CCA stem with helix H68.

In turn, the interaction at the anticodon–codon differs from the P-tRNA by the adjusted orientation of the base pairs due to the changed geometry caused by both the ratcheting of the ribosome and the hybrid conformation of the tRNA. Simultaneously, the anticodon stem of the P/E-tRNA is kinked, and the density for the gate in the cryo-EM map is opened up. These structural features together create the spatial condition for the ASL of the P/E-tRNA to move into the E site from the P/E position. However, in the current intermediate state, the tRNA still is held at the P/E state when EF-G is stalled in the ratcheting ribosome. The advancement of the P/E-tRNA may have to be accompanied by the release of EF-G from the ribosome.

The interaction between the T/D loops and helix H78 differs from that in the x-ray ribosome with an E/E-tRNA mostly by the moved position of the entire L1 stalk. The elbow of the P/E-tRNA reaches helix H78, but the T stem is not as close to the L1 stalk as the E/E-tRNA is.

The interaction between the CCA stem and helix H68 is affected by the release of the very CCA end from the P site in the 50S subunit so that the CCA stem is exclusively attached to helix H68. As described in Results, the unique conformation of helix H68 goes hand in hand with the relaxed state of helix H69 in the ratcheted ribosome and with the CCA arm being twisted toward its position in the P/E-tRNA. It has been recognized that interaction with helix H69 of the 50S subunit plays a crucial role in the formation and stabilization of the A/T-, the A-, and the P-site tRNAs (3, 10). In contrast, our study shows that helix H69 is no longer bound to the P/E-tRNA in the translocating ribosome. This release apparently destabilizes helix H69, causing it to be seen in the map as a much reduced density (SI Fig. 10d). At the same time, the conformation of helix H68 is relaxed and different from the helix H69-bound state in the x-ray ribosome.

Thus, we observe several interrelated conformational changes in the transition to the intermediate translocation state: release of P-tRNA from binding with helix H69, liberation of the D stem of the tRNA from its contact with helix H69, and destabilization of helix H69. The mobility of helix H69 reduces the structural constraints on both helices H68 on the 50S subunit side and h44 on the 30S subunit side, so that these two helices are reformed in relaxed conformations, compared with their appearances in the x-ray structure. The liberation of the D stem of the tRNA provides the structural flexibility required when the tRNA is being twisted in the ratcheting ribosome.

The structural roles of helix H69 of the 23S rRNA in translation have been discussed extensively in numerous studies (5, 26). Directly relevant for this question is recent biochemical study by Ali et al. (27), who demonstrated that deletion of helix H69 does not reduce normal translational accuracy, although it disrupts the association between the 50S and the 30S subunits. Our current finding shows that helix H69 is in fact disconnected from the P/E-tRNA in the ribosome and that its deletion cannot disrupt the formation of the intermediate translocating state.

MD Simulations.

The MD simulations were performed by using the Sander module in the molecular simulation software package AMBER7 (28) and the Cornell force field (29). The x-ray atomic coordinates of the tRNA (PDB ID code 1TTT; from T. thermophilus; ref. 19) are used as the starting structure of the simulations. The particle-mesh-Ewald method is used for the treatment of the electrostatic interaction energies.

Net charges on the tRNA are neutralized by adding sodium ions. Additional 20 Na⁺ and 20 Cl⁻ are added in the simulation system to create the ionic conditions. The system is placed in a water box with a water shell that is at least 10 Å away from the solute to the outer edge of the shell, by using the TIP3P water model. The water box has a size of 98 × 59 × 96 ų and contains 14,063 water molecules.

The simulation system was subjected to an energy minimization by using the steepest descent minimization algorithm with 200 steps, followed by 50 ps of dynamics to heat the system from 100°K to 300°K with a 20 kcal/mol per Ų constraint on the tRNA. This procedure then was followed by 50 ps of equilibration with the same constraint on the solute. Next, the system was energy-minimized with gradually decreasing constraints on the tRNA starting from 20 kcal/mol per Ų, 15 kcal/mol per Ų, then 10 kcal/mol per Ų, 5 kcal/mol per Ų, 1 kcal/mol per Ų, and finally no constraints with 200 steps for each minimization. Then a reduced water box size was used to replace the initial size, and the previous heating dynamics simulation was repeated. The size of the water box was reduced to 91 × 55 × 89 ų after the preparation steps. After the constraint was released, three independent production runs were performed by using changed velocity distributions (random seeds), lasting 44 ns, 18 ns, and 18 ns, respectively. The constant-pressure condition was used. SHAKE (30) was applied to the bonds involving hydrogen atoms. The time-step was 2 fs, and the cutoff distance for nonbonded interactions was 9 Å. AMBER7/CARNAL was used for the trajectory analysis.

Docking of Atomic Structures into Cryo-EM Maps.

The Motif search procedure in SPIDER was used to locate atomic structures within the cryo-EM map; see the details elsewhere (20). The graphic illustrations were produced by using the programs IRIS Explorer (Numerical Algorithms Group, Downers Grove, IL), Ribbons (31), and INSIGHT II (Accelrys, San Diego, CA).