Mechanism of peptide bond synthesis on the ribosome

Abstract

With the emergence of atomic-resolution crystal structures of bacterial ribosomal subunits, major advances in eliciting structure–function relationships of the translation process are underway. Nevertheless, the detailed mechanism of peptide bond synthesis that occurs on the large ribosomal subunit remains unknown. Separate x-ray structures of aminoacyl-tRNA and peptidyl-tRNA analogues bound to the ribosomal A- and P-sites, however, allow for structural modeling of the active complex in catalysis. Here, we combine available structural data to construct such a model of the peptidyl transfer reaction center with bound substrates. Molecular dynamics and free energy perturbation simulations then are used in combination with an empirical valence bond description of the reaction energy surface to examine possible catalytic mechanisms. Already, simulations of the reactant and tetrahedral intermediate states reveal a stable, preorganized H-bond network poised for catalysis. The most favorable mechanism is found not to involve any general acid–base catalysis by ribosomal groups but an intrareactant proton shuttling via the P-site adenine O2′ oxygen, which follows the attack of the A-site α-amino group on the P-site ester. The calculated rate enhancement for this mechanism is ≈10⁵, and the catalytic effect is found to be entirely of entropic origin, in accordance with recent experimental data, and is associated with the reduction of solvent reorganization energy rather than with substrate alignment or proximity. This mechanism also explains the inability of 2′-deoxyadenine P-site substrates to promote peptidyl transfer. The observed H-bond network suggests an important structural role of several universally conserved rRNA residues.

The ribosome translates genetic messages and catalyzes the synthesis of new proteins. The determination of highresolution x-ray structures of ribosomal subunits (1–3) has enabled considerable progress toward understanding the structural basis of the translation process (4–7). The peptidyl transferase center is located on the large 50S ribosomal subunit ≈20 Å below the subunit interface at the entry of the peptide tunnel that exits on the other side of the large subunit. High-resolution crystal structures of the 50S subunit have been reported together with different aminoacyl- and peptidyl-tRNA substrate analogs (4). These structures clearly locate the active site and verify the conserved structure and base pairing of the tRNA 3′ terminal CCAs with rRNA. Still, the actual mechanism of the peptidyl transfer reaction remains unclear. Although it is evident that the unprotonated form of the A-site α-amino nitrogen attacks the ester carbon of the donor peptidyl substrate, leading to a transient tetrahedral “intermediate,” a major question is whether or not this process is subject to general acid–base catalysis by ribosomal groups (Fig. 1). It is, in fact, not even known whether any rRNA groups are affecting the energetics of the chemical reaction or whether the observed catalytic effect is entirely due to positioning of the tRNA substrates as proposed by Moore, Steitz, and coworkers (4, 7). This suggestion would then imply that the chemical rate acceleration completely derives from reduced translational and/or rotational contributions to the activation entropy.

Fig. 1.

Possible mechanistic pathways for aminolysis of the ribosomal P-site peptidyl-tRNA ester bond by the A-site amino acid where the role of a putative rRNA base (B) is indicated.

It was proposed from structural data and dimethylsulfate reactivity experiments that the N3 nitrogen of the universally conserved A2451 (Escherichia coli numbering) nucleotide could act as a general acid/base in the catalytic process (8, 9). This hypothesis seemed to be supported by the fact that the A2451U mutation led to significantly reduced activity. However, a similar rate reduction effect (of ≈100) that titrates with a pK_a of ≈7.5 is observed in unmutated ribosomes (10), which indicates that either the N3 of A2451 must have a highly perturbed pK_a or that some pH-dependent conformational change can occur in the active site. Note that the unperturbed pK_a of the adenosine N3 atom is only ≈1 (9), which would require an extraordinary stabilization of its positively charged (protonated) form by >6 pK_a units, for which the structural data offer no explanation. As indicated in Fig. 1, it is of considerable importance whether the nucleophilic attack is accompanied by general base catalysis or not, because it affects the nature of the transient tetrahedral intermediate (TI), which can be either negatively charged or dipolar (formally zwitterionic). Judging from the available crystal structures, there is no obvious group that could provide significant stabilization of a negative intermediate, unless a hitherto-unobserved counterion were to be involved (4), which may speak in favor of a zwitterionic intermediate. However, what mediates proton transfer from the attacking amino group to the leaving ester oxygen is then difficult to say.

To explore different possible mechanistic alternatives for the ribosomal peptidyl transfer reaction, we carried out extensive molecular dynamics (MD) free energy calculations by using the empirical valence bond (EVB) method (11, 12) for describing the reaction potential energy surface. These simulations were based on a model (Fig. 2) constructed by superimposing the large subunit structures with Protein Data Bank (PDB) ID code 1FG0 (8) (with 3′ terminal CC tRNA coupled to puromycin in the A-site) and PDB ID code 1M90 (7) (with CCA-Phe-caproic acid-biotin in the P-site and sparsomycin in A-site) in a similar manner to Hansen et al. (7). The crystal structures [including PDB ID code 1KQS (13)] show an excellent superposition of A- and P-site aminoacyl and peptidyl tRNA as well as the surrounding rRNA and indicate that nucleophilic attack will lead to a TI in the S-configuration.

Fig. 2.

Model of the reactive peptidyl transfer system with aminoacylated CCA fragments bound to the A- and P-sites. The model was constructed by combining the structures with PDB ID codes 1FG0 (8) and 1M90 (7). Water molecules (and counterions) present in the simulations have been removed for clarity.

Methods

MD calculations were carried out at 300 K with the program q (14) using the CHARMM22 force field (15) on a spherical system of 21-Å radius centered on the P-site carboxyl carbon from the structure with PDB ID code 1M90 (7). The substrate CC-puromycin from the structure with PDB ID code 1FG0 (8) was superimposed into the A-site. Atoms outside 18 Å from the sphere center were harmonically restrained to their initial positions with a 100-kcal/mol·Å² force constant and excluded from nonbonded interactions. The puromycin moiety was converted to 3′-linked adenine–tyrosine, and the P-site CCA-Phe-caproic acid was changed into CCA-Phe-Gly, where the terminal amino group was kept neutral in accordance both with experimental model assays containing Phe-caproic acid-biotin (16) and the native fMet residue. The 5′ end of the A- and P-site CCA sequences was terminated with hydroxyl groups as in the experimentally active assays (16). The 18-Å sphere was solvated, and waters close to the sphere surface were restrained to reproduce the correct density and polarization (14). No interaction cutoffs were applied to reacting fragments, whereas for other interactions a multipole expansion treatment (17) of long-range electrostatics (>10 Å) was used. To avoid artificial dielectric effects near the sphere boundary, metal ions were replaced by water molecules, and phosphate groups were neutralized, within 4 Å from the boundary. This method is an effective way to mimic their neutralization by a diffuse counterion distribution so that their long-range interactions with the reaction center become appropriately screened. Charged phosphate groups and counterions were retained within 14 Å from the center, where also five additional water molecules were converted to magnesium ions. These water molecules were chosen, from a set of 12 waters close to phosphate esters, on the basis of the interaction energy gained upon conversion. After initial energy minimization, the system was equilibrated in the TI state (S-enantiomer) by using a step-wise heating scheme from 0 to 300 K with a 1-fs time step and restrained rRNA and substrates. The TI then was further equilibrated for at least 200 ps before data collection. Standard CHARMM22 force field (15) parameters were used for the reacting groups together with Morse bond potentials from (18) and partial atomic charges derived from ab initio HF/6-31G* calculations (see Table 1, which is published as supporting information on the PNAS web site).

The EVB reaction free energy surface was described in terms of three basic resonance structures (cf. Fig. 4a). The energetics of the uncatalyzed water reaction was first calibrated against available energetic data from high-level ab initio calculations [B3LYP/AUG-cc-pVDZ//HF/6–31G(d)] including solvation effects (19). This calibration gives an overall activation barrier of 24 kcal/mol for the ester aminolysis reaction, an exothermicity of 8 kcal/mol, and the TI 12 kcal/mol above the reactants. The barrier for reaching the TI is 15 kcal/mol, and it is thus its breakdown (second transition state) that limits the rate of the reaction in solution (19). The overall barrier and exothermicity predicted in ref. 19 is in perfect agreement with independent experimental results for the reverse peptide hydrolysis reaction (20) and also agrees with recent measurements of uncatalyzed ester aminolysis rates for different substrates (16). Calibration of the uncatalyzed reaction energetics in solution was done by simulation of the amino attack of 3′-Tyr-1,5-dideoxy-β-d-ribofuranose on a (neutral) 3′-(Gly-Phe)-1,5-dideoxy-β-d-ribofuranose molecule in a sphere of water by using the same procedures as in the ribosome calculations. Fitting of the EVB surface to the uncatalyzed solution energetics is achieved by parameterizing the relative free energies between different valence bond (VB) states and obtaining the intervening barriers by mixing these states by means of the off-diagonal VB Hamiltonian matrix elements (11, 12). The resulting parameterization then is used without change in the ribosome reaction, whereby the effects of the surrounding environment on the reaction energetics can be accurately evaluated.

Fig. 4.

Reaction energetics and product structure. (a) Reaction free energy diagram obtained from MD/EVB calculations for the uncatalyzed reference reaction in water (upper curve) and for the ribosome reaction (lower curve). Atoms included in the active EVB region (i.e., those with changing bonds, charges, etc.) (14), which define the VB states, are shown for the reactants, and the proton shuttle mechanism involved in decomposition of the transient intermediate is also indicated. (b) Calculated structure of the bound product (cyan) superimposed on the experimental structure of C-puromycin–caproic acid–biotin (C-pmn-pcb) in the ribosomal A-site (13).

Calculations on the A2451 base mechanism, where the N3 nitrogen accepts one of the amino protons to yield an anionic TI, requires an additional EVB resonance structure to be introduced. The free energy of this state in aqueous solution is obtained from the pK_a difference between the adenine N3 (pK_a ∼ 1) (9) and N-protonated intermediate (pK_a ∼ 8) with only a small intervening barrier (19). The possible uncertainties in these pK_a values have no effect on our conclusions because the simulations clearly show an increased, rather than decreased, pK_a difference between the two groups in the ribosome compared with water (see Results and Discussion).

The reaction free energy profiles in water and in the solvated ribosome model were calculated with the free energy perturbation/umbrella sampling method (11, 12). Each free energy perturbation calculation comprised 51 discrete steps with a 5- to 20-ps trajectory generated at each step. In practice, simulations were initiated from the TI and carried out toward reactants and products (the structural stability of the system can be seen in Fig. 5, which is published as supporting information on the PNAS web site, showing the root-mean-square atomic positional deviation with respect to the initial structure for the nonhydrogen EVB atoms in the TI state). A full reaction profile calculation thus comprised 0.5–2 ns, and four or five such simulations, with different initial conditions, were carried out both in water and in the ribosome to assess the accuracy of the calculated free energies (estimated error bars are less than ±1.5 kcal/mol). Detailed free energy profiles (complementary to Fig. 4a below) for both the water and ribosome reactions are shown in Fig. 6, which is published as supporting information on the PNAS web site, where the reduction of the free energy gap (directly reflecting reorganization energy) between diabatic EVB surfaces is clearly visible.

Two independent simulations of the first half of the ribosome reaction with dA2451 also were carried out to evaluate the effect of this chemical modification on the stability of the TI. In addition, 2-ns simulations were carried for the reactants (R) and TI states, in the ribosome and water, to estimate the separate enthalpy and entropy contributions following the straightforward procedures outlined in ref. 21. The entropies (–TΔS term) associated with moving from the R to the TI state, for the water and ribosome reactions, then were evaluated as the difference between the free energy and enthalpy (see Results and Discussion). The main difficulty with this type of calculation is that the averages of the total potential energy (〈U_pot〉_R and 〈U_pot〉_TI) converge slowly because of to the large absolute magnitude of the energy terms, on the order of 10,000 kcal/mol. Nevertheless, we can obtain sufficiently small error bars (±8 kcal/mol) to get a semiquantitative entropy estimate. The reason for this result mainly resides in the use of a finite spherical simulation system (≈5,000 atoms) that yields more rapid convergence than larger systems of the periodic boundary type. That this brute force approach gives reliable results also was verified for the simple test case of calculating the relative entropy of hydration between Na⁺ and Cs⁺ ions in a sphere of 460 water molecules. In that case, we obtained TΔΔS_sol = 3.38 ± 1.39 kcal/mol (experimental value 3.6 kcal/mol), by using 1-ns endpoint trajectories and 2 × 350 ps for the free energies, which illustrates the convergence of method.

Results and Discussion

Already MD equilibration simulations of the transient TI, as well as of the state with bound reactants, immediately suggest a mechanism for the peptidyl transfer reaction that does not require the counterintuitive general base action of A2451 N3. Instead, we observe a spontaneous and stable H-bond donated by the attacking amino group to the 2′-OH of A76 in the P-site substrate. Furthermore, this OH group is able to donate an H-bond to its 3′-OH neighbor, suggesting a proton shuttle mechanism where the 2′-hydroxyl acts as the elusive base by receiving a proton from the attacking amino group, while simultaneously protonating the leaving 3′-OH as the TI is decomposed into products (Fig. 3). An additional feature emerging from the simulations that supports this idea is that the 2′-OH group of A2451 can stabilize such a mechanism by donating a second H-bond to the A76 2′-OH. In fact, the simulations of both reactants and tetrahedral intermediate reveal an intricate network of H-bonds involving the substrates, the universally conserved C2063, C2064, A2451, U2584, and U2585, and several water molecules that act as bridges in these interactions (Fig. 3). This H-bond network seems extraordinarily well adapted for stabilizing the different configurations occurring on the reaction pathway.

Fig. 3.

Schematic (a) and detailed stereoview (b) of the structure of the bound (transient) TI from MD/EVB simulations. (a) H-bond interactions involving ribosomal groups (E. coli numbering) and water molecules are indicated. (b) The favorable stereochemistry for promoting the A76 2′-OH proton shuttle mechanism can be directly seen.

Standard MD simulations of (meta-) stable states, however, do not give information about their free energy differences or the heights of energy barriers separating the states. To evaluate detailed reaction free energy profiles, we carried out MD/EVB simulations by using the free energy perturbation/umbrella sampling approach (11, 12). Besides the mechanism proposed above, which involves a six-membered transition state in the ester cleavage reaction, we also examined the possibility of direct protonation of the leaving group by the attacking amino nitrogen leading to a four-membered transition structure. Furthermore, we attempted to enforce the A2451 general base mechanism proposed originally (8) by letting its N3 nitrogen abstract a proton form the TI. Both of the latter alternatives were, however, found to involve significant strain in the system and did not yield any reasonable free energy profiles. In the A2451 general base mechanism, there is clearly no stabilization of the protonated base available that could provide the large required upwards pK_a shift. Our calculations instead yield a destabilization of the state with A2451 protonated, and an anionic TI, by 5.5 ± 0.4 pK_a-units or ≈7.5 kcal/mol. Furthermore, this mechanism offers no explanation for how the P-site O3′ leaving group could be protonated because A2451 N3 is too far away (Fig. 3b). The alternative with a four-membered transition state for the second (bond-breaking) part of the peptidyl transfer reaction, conversely, is intrinsically associated with high strain, and, e.g., ab initio calculations (including solvation effects) on the uncatalyzed reaction of a model system yielded a 7-kcal/mol higher barrier than the six-membered case (19, 22).

In contrast, the mechanism involving the A76 2′-OH as a proton shuttle shows favorable energetics and is associated with a significant catalytic effect. The calculated reaction free energy profile is shown in Fig. 4a together with the corresponding energy profile for the uncatalyzed ester aminolysis reaction in water (see also Fig. 6). The product structure resulting from these MD/EVB simulations also agrees remarkably well with the independent structure with PDB ID code 1KQS (Fig. 4b). Note that no information from the latter structure was used to construct the model used in the simulations. It can be seen that the major effect of the ribosomal reaction is a stabilization of the TI together with its flanking transition states, where the rate-limiting one is lowered by ≈7 kcal/mol. This profile yields an overall rate acceleration by a factor of 100,000, and the calculated catalytic effect is thus in very good agreement with the value of k_cat/k_non recently provided by Sievers et al. (16). Furthermore, this mechanism immediately explains why peptidyl tRNAs that terminate with a 2′-deoxyadenine or a 2′-F substituted ribose are inactive as P-site substrate and will not donate their peptide to the A-site substrate (23, 24). Substitution of the 2′-OH with a group that retains both proton donor and acceptor capabilities, such an amine or thiol (25), thus would be predicted to be less severe.

As mentioned above, we find no support for a significantly perturbed pK_a of the A2451 N3 atom, but judging from our simulation structures it seems very likely that protonation of one of the rRNA adenine N1 or cytosine N3 atoms close to the reaction center (there are several such potential protonation sites, including A2451) could cause conformational effects detrimental to catalysis. This explanation for the observed catalytic dependence on an ionizing group with pK_a of ≈7.5 is more attractive because it requires a much smaller pK_a shift. To obtain a preliminary estimate of the effect of the dA2451 modification, we also evaluated the relative free energy of the TI for this case, which shows a destabilization of ≈3 kcal/mol. The corresponding effect on the transition states presumably would be similar, and this calculation demonstrates the importance of the H-bonds involving A2451 2′-OH, in agreement with recent experiments (26).

What, then, is the origin of ribosomal catalysis of the peptidyl transfer reaction? It was suggested by Moore and Steitz (4) that the main catalytic effect of the ribosome is to use binding energy to overcome the entropic cost of aligning the substrates for reaction, following the pioneering ideas by Page and Jencks (27). This type of catalytic hypothesis, however, has been subject to major controversies over the years (28–30), and the actual effects of “alignment” and “proximity” still have not been quantified in a convincing way for intermolecular (e.g., enzyme) reactions. It is relatively clear, however, that the concept mainly refers to a reduced loss of (reactant) translational and rotational entropy associated with reaching the transition state of the reaction. Sievers et al. (16) recently reported detailed kinetics of uncatalyzed and ribosomal peptide bond formation, which showed that the activation entropy term, –TΔS^‡, was reduced from 13.1 kcal/mol in the solution reaction to only –0.7 kcal/mol for the ribosome bound reaction. The activation enthalpy, conversely, was actually larger by ≈8 kcal/mol on the ribosome. This finding thus would seem consistent with the alignment idea. However, the fact that approximately the same reduction of –TΔS^‡ that was found for the first-order rate constant k_cat also was found for the second-order k_cat/k_M rate indicates that this result may not be an effect associated with substrate binding. That is, the temperature dependence of k_cat/k_M in this case reflects the activation entropy measured from the reactant state with only P-site substrate bound and the A-site substrate (puromycin) free in solution. It thus seems more likely that the observed reduction of –TΔS^‡ is a transition-state effect because it is manifested in the reactions rates for both bound and unbound A-site substrate.

Activation entropy is a macroscopic (thermodynamic) quantity that actually encompasses a number of microscopic effects deriving from solvation, molecular conformations, and phase-space configurational volumes. To what extent activation entropies really are correlated with reduced configurational volumes in the reactant state has been debated for several classes of organic reactions (28–31). It is, however, clear that both activation and binding entropies are in many cases entirely dominated by solvation effects [a pertinent example is the binding of largely nonpolar drugs to an enzyme (32)].

The MD simulations give an immediate answer as to the origin of the catalytic effect on the ribosome in terms of the stable H-bond network observed along the reaction. Unlike the uncatalyzed water reaction, which requires a significant reorganization of water molecules due to the charge separation involved in forming a zwitterionic (dipolar) transient TI, the ribosomal reaction takes place within a more or less preorganized H-bond network. This situation provides a direct explanation for why the TI with its flanking transition states are not associated with a large loss in entropy. In fact, analysis of the energetics behind the free energy profiles of Fig. 4 shows that it is indeed the preorganization of the reaction site, reflected in a reduced free energy gap between the reactant and product states for both reaction steps, that is responsible for stabilization of the high-energy structures (see Fig. 6). This type of reorganization energy reduction effect has been observed in many enzymes and seems to be a general feature of biological catalysis (11, 12, 18, 33, 34). In the cases of regular enzymes, reorganization energy reduction often is combined with a direct enthalpic stabilization of high-energy intermediates occurring in the reaction (18).

Remarkably, our calculations show that there is no enthalpic stabilization of the transition states and TI for the ribosome compared with the solution reaction but that the free energy stabilization on the ribosome is instead due to a much smaller entropy loss upon climbing the reaction hills. Although it is fundamentally more difficult to get precise values of ΔH and ΔS from MD simulations than of ΔG (21, 30), we carried out long simulations at the R and TI states to estimate the different contributions to the free energy from

[1]

where U_pot denotes the potential energy of the system. From these calculations, we find that the total TΔS term is significantly more positive (19 ± 8 kcal/mol) on the ribosome than in water, i.e., ΔΔS(ribosome – water) is positive for the R → TI process. Although this estimate does not refer to the actual activation entropy but to the intermediate, it provides a reliable semiquantitative energetic decomposition despite the large error bar. This entropy effect is further found to be directly associated with the interaction terms involving the surroundings of the reacting groups and dominates over the proximity effect, which is estimated to be <1 kcal/mol from available configurational volumes derived from the simulations. Hence, our MD/EVB simulations not only yield correct overall rate constants for the new mechanism but also are in accord with the thermodynamic picture provided in ref. 16. The conclusion is therefore that it is not the alignment of substrates with reduced translational and rotational entropies that is the origin of the catalytic effect but the preorganized reaction site that, unlike bulk water, does not require major reorganization along the reaction.

The overall elongation process on the ribosome is very complex one, but, nevertheless, computer simulation approaches can be very useful for elucidating the mechanisms of specific events as shown here. In fact, such computational analysis may provide the missing links between crystal structures and functional experimental data. The present simulations have allowed us to predict the identity of the groups involved in the peptidyl transfer reactions, and out of three different possible mechanisms examined only one turns out to be energetically compatible with experimental kinetics.

It is interesting to note that the principles of ribosomal catalysis seem to differ somewhat from those of ordinary enzymes, as already suggested by the ribosome crystal structures, and a possible evolutionary rationale for substrate-assisted catalysis may be hypothesized (24). Although reorganization energy plays a role also in many enzymic reactions, it is normally accompanied by some type of enthalpic stabilization of transition states and intermediates. Enzymes also have a much more versatile collection of functional groups (small, big, hydrophobic, polar, positively and negatively charged, etc.) to “choose” from to achieve a desired effect on their substrates than ribozymes. The nucleobases of RNA and DNA have a more limited repertoire of chemical functionalities and are presumably chosen partly because they do not have ionizing groups within the normal pH ranges, which would have been hazardous to the genetic machinery. Hence, the ribosome is thus not only an ancient catalyst, but it also has to play according to different chemical rules than the enzymes. The importance of structural preorganization is, however, evident in both cases.