Microcalorimetry reveals multi-state thermal denaturation of G. stearothermophilus tryptophanyl-tRNA synthetase

Mechanistic studies of B. stearothermophilus tryptophanyl-tRNA synthetase (TrpRS) identified two subtle, and apparently contradictory, ways in which differential conformational stability affects its catalytic cycle.^1–4 Catalytic assist by Mg²⁺ ion, on the one hand, depends entirely on its coupling to domain motions.^3,5,6 On the other hand, the sign of the free energy change for the catalytic conformational change is positive in the presence of the pyrophosphate product, PPi, and negative only in its absence.⁷ That deepened understanding, in turn, provides a novel perspective on how biology appears to circumvent the second law of thermodynamics.^4,8,9

The former effect couples assembly of the fully active catalytic apparatus tightly to domain motion; the latter ensures that domain motion requires PPi release, thereby coupling the conformational change reciprocally¹⁰ to the chemistry of ATP utilization. This novel escapement mechanism^3,4 ensures efficient vectorial utilization of ATP^11,12 and is thus a potentially general model for energetic coupling in a broader range of enzymes that transduce NTP hydrolysis free energy.^13,14 Neither front nor back-end gating has been analyzed in detail for any other transducing NTPases, although just as the TrpRS catalytic domain movement is triggered by PPi release, the myosin power stroke appears to be triggered by phosphate release.^13,14 The back-end gating may, thus, be common to many NTPases.

Key to the escapement mechanism are conformational stability changes induced, in turn, by the succession of bound ligands, as chemical events proceed.² Untangling different levels of energetic coupling that produce the contradictory gating effects necessarily entails measuring how these stability changes arise from differential ligand binding. That task is facilitated by a set of combinatorial mutants within a packing motif called the D1 switch that mediates shear developed during TrpRS domain motion.^1,15 The D1switch consists of 6–7 residues near the C-terminus of the Rossmann fold N-terminal α-helix that bind the preceding and following β-strands to that helix. It is one of the most widely conserved motifs in the proteome¹⁶ and central to this work.

We stumbled on the D1 switch motif when we attempted to identify residues whose nearest neighbors changed as TrpRS passed through its conformational trajectory during catalysis.¹⁷ Delaunay tessellation decomposes protein packing interactions uniquely into tetrahedra of nearest-neighbor side chains.^18,19 TrpRS crystal structures revealed that the vast majority of the ∼300 tetrahedra lie within domains that move as rigid bodies and are, thus, structurally invariant. Only about a dozen tetrahedra are structurally dynamic. A series of papers using the PATH algorithm to map the conformational changes during catalysis showed that repacking of aromatic residues in the D1 switch motif—F26, Y33, and F37—constitutes the rate-limiting step in assembling the active site for catalysis upon binding tryptophan and ATP.^1,7,20

Rosetta²¹ identified specific mutations of the three aromatic D1 residues plus I4 that should stabilize the excited Pre-TS state relative to the two ground states—Open and Products—on either side.¹⁷ Combinatorial mutagenesis of those residues showed that repacking is coupled by ∼–5.5 kcal/mol to the active-site Mg²⁺ ion. It accounts for the contribution of the metal to transition-state stabilization. A subsequent modular thermodynamic cycle²² implicated an untwisting domain motion to catalysis by the same coupling free energy. Those studies linked dynamic side chain re-packing in the D1 switch alternately to domain motion and to Mg²⁺-dependent rate acceleration of amino-acid activation. Recent studies of a related thermodynamic cycle measuring the energetic coupling between the HVGH and KMSKS catalytic signatures in the larger leucyl-tRNA synthetase²³ provide complementary evidence that such coupling is common to all class I aaRS. The ensemble of structural, kinetic, and computational studies on the D1 switch afforded a uniquely consistent set of cross-validating intercorrelations (see Fig. 7 in Ref. 20). Those correlations underlie the TrpRS escapement mechanism and so strongly motivate the present analysis of TrpRS melting behavior.

In this report, we use differential scanning calorimetry (DSC) and computational simulation to compare the melting behavior of fully liganded TrpRS bound to ATP and a non-reactive tryptophan analog stabilizing the crystallographic pre-transition conformational state (1MAU²⁴) with that of the unliganded apo form (1D2R²⁵). Given the differences between experimental (TrpRS dimer; Cp; heat capacity at constant pressure) and computational (TrpRS monomer; Cv; heat capacity at constant volume) methods, there is surprising consistency between the two methods. Moreover, the computational simulations also allow comparison of two structural metrics—the radius of gyration, R_g, and α-helical content—that facilitate interpretation. Our principal observations are as follows:

• Both liganded and apo TrpRS denature via a three-state process involving a molten globular intermediate that forms at temperatures ∼8° lower than those required to melt α-helical secondary structures.

• A broad, unidentified and substantial heat capacity change occurs at approximately the temperature at which the source organism grows optimally and precedes the formation of the molten globule.

• A computational free energy surface of a liganded TrpRS monomer exhibits similar multi-step experimental heat capacity changes. That simulation is reversible by definition, because the isolated monomer cannot aggregate. It also furnishes valuable structural metrics and cameo snapshots that aid interpretation of the melting behavior.

• The catalytic core known as the urzyme, an experimentally validated model for the ancestral form of class I aminoacyl-tRNA synthetases, containing only the 130 residues necessary to frame the active site, is the last remaining folded structure at high temperatures in the computed free energy surface. Preservation of the tertiary packing of D1 residues at high temperature appears to stabilize the N-terminal α-helix, which has only weak helical propensity.

An accompanying paper²⁶ describes melting curves obtained by differential scanning fluorometry and far UV circular dichroism for unliganded native and 15 combinatorial TrpRS D1 switch variants. Analysis of unliganded TrpRS described here provide a solid context in which to identify the intrinsic and higher-order structural interactions that allow the D1 switch residues to stabilize or destabilize different conformational states along the structural reaction profile, hence their functional behavior. Our goal is to establish a basis for pursuing two non-trivial longer-range questions: (i) Can we exploit the mutant proteins to correlate stability changes of all three conformations with the steady-state kinetic properties measured previously?³ (ii) How does coupling between the four mutated residues change along the conformational reaction path, allowing them to exert their highly cooperative contribution to the escapement mechanism?⁴ 

TrpRS solutions (2 μM) were dialyzed overnight against either buffer alone [20 mM HEPES, 50 mM NaCl₂ (pH 7.0)] of the same buffer plus tryptophanamide (50 μM) and ATP (10 mM). In each case, we reserved the dialysate for use as a blank in calorimetry. We loaded 500 μl of each of the four solutions (sample plus blank, liganded vs apo) into the chamber of a TA Instruments Nano DSC 250 microcalorimeter. Thermal scans were performed at 2.98 atm from 10 to 95 °C in 5 min intervals. Data from the thermal scans were processed using the TA Nanoanalyze software, v. 3.12 (https://www.tainstruments.com/itcrun-dscrun-nanoanalyze-software/) to subtract the blank profile, normalize the net heat capacity curves, select baselines, and identify peaks. We identified and integrated heat capacity peaks in the profiles using the Voigt Gaussian/Lorentzian hybrid given an onset temperature (Tonset), which gave superior local fit to all curves.

We make use here of an extended analysis of the free energy surface connecting the pre-transition state (PreTS) and products (Products) conformational states. The replica exchange algorithm efficiently samples the temperature-dependent conformational free energy landscape of a macromolecule by simulating replicas of the equilibrium state at different temperatures and, at predefined time points, exchanging the structures at different temperatures if the difference in their energies is within a threshold. It is reversible by definition because the monomer cannot aggregate.

Our original purpose was to find an independent estimate of the conformational transition-state structure that we could compare with that produced by a minimum action path algorithm.⁷ To that end, we projected the simulated trajectory onto the two dominant principal components of the motion to give a 3D function whose z-axis was conformational free energy. That surface was used to map the location of the conformational transition states separating the Pre-transition and Product state conformations to the stationary point of the conformational transition path computed using a minimum action path algorithm and which turned out in excellent agreement with the position of the global saddle point separating the PreTS and Products states in the replica exchange simulation.

We implemented the replica exchange algorithm using the REX/DMD suite^27–30 that uses Discrete Molecular Dynamics^28,30–32 to simulate replicas at 24 temperatures ranging from $∼$ 175 to $∼$ 405 K separated by a constant interval. DMD approximates atomic interactions by multistep square well potentials^27,33 and uses an Andersen Thermostat to regulate system temperature, while the Lazaridis–Karplus implicit solvation model³⁴ accounts for the solvation energy. Simulations were set up with the Product state structure, 1I6L, with the PreTS state structure, 1MAU which appears to be the least stable of the crystal structures.¹⁷ 

We excised the terminal amino acid (R328) from the structures as it does not appear in most crystal structures. Also, since DMD force field does not include parameters for Mg²⁺, we replaced it with Zn²⁺. The two metals have almost exactly the same ionic radii and hydration energies, and they behave almost identically in molecular simulations with empirical force fields.³⁵ Moreover, Zn²⁺ ion parametrizations prove to be the most similar to those of Mg²⁺ in potential of mean force profiles used to model ionic interactions with nucleic acids.³⁶ For these reasons, we do not expect these changes to affect the DMD of the system in a significant manner. We applied a harmonic constraint (well width = ±1 Å) between ligand atoms and the binding pocket within 3.5 Å of each other to retain ligands in their binding pockets. We applied another weak harmonic constraint (well width = ±2 Å) to atoms which uniquely form native contacts (8 Å), in either 1I6L or 1MAU, in order to sample conformational spaces adjacent to the two boundary states. Simulations at each temperature ran for 3 × 10⁶ steps (⁠ $∼$ 150 ns). Omitting the time to equilibrate (initial 500 000 steps), snapshots were generated every 1000 steps for a total of 2500 snapshots which were then used in all the analyses.

Coordinates from the trajectory furnished a variety of useful metrics to aide interpretation of the melting behavior. The WHAM—weighted histogram analysis method—algorithm enable computation of the specific heat capacity, $C v$⁠,³⁷ from the energies of structures generated at different temperatures. The atomic coordinates of the TrpRS monomer enable computation of the average radius of gyration, R_g from the mean structures at each temperature. Similarly, we used the DSSP algorithm^38,39 as implemented in Pymol⁴⁰ with default options to compute the fraction of α-helical residues for each averaged structure.

Experimental heat capacity changes during TrpRS thermal melting exhibit complex behavior with clear differences between apo [Fig. 1(a)] and liganded [Fig. 1(b)] forms. Binding of tryptophanamide and ATP lead to an inhibited state we have termed the pre-transition-state (Pre-TS) conformation because the tryptophanamide and ATP ligands trap it just prior to the chemical step in tryptophan activation. Crystal structures of the apo and Pre-TS states differ substantially^24,25,41 as the anticodon-binding domain rotates toward the catalytic domain by ∼8° and twists by ∼3.5° in the PreTS state. The different melting curves must reflect these structural differences.

Melting occurs in three distinguishable stages: initial (T_onset = 56, 60 °C), intermediate (T_onset = 62,69 C) with a major change in heat capacity, and terminating (T_onset = 75, 73 °C). Pre-TS TrpRS melting is also significantly more cooperative with all changes occurring between 62 and 73 °C], compared to 56 and 82 °C.

Curiously, although ligand binding normally stabilizes protein structures against thermal melting, the overall melting of the apo form occurs at higher temperature (79 vs 74 °C). That observation confirms previous estimates, derived from nucleotide-ligand binding affinity changes between different conformations and from comparing statistical potentials derived from side chain packing analysis which led us to conclude that the Pre-TS state was significantly destabilized by the conformation induced prior to catalysis.¹⁷ 

To validate the thermodynamic interpretations of TrpRS melting behavior, we take advantage here of detailed correspondences—and differences—between the experimental melting curves (Fig. 1) and those derived from a computational temperature-dependent conformational free energy surface of the liganded, wild-type TrpRS monomer PreTS complex. That surface (Fig. 2) is reversible by definition,⁴² because the isolated monomer cannot aggregate. It was constructed previously using Replica Exchange Discrete Molecular Dynamics (REX/DMD^29,42,43) simulations to confirm the computational transition states identified by a minimum action path analysis.^1,7,20

Figure 2 shows heat capacity changes, $C v$⁠, as a function of temperature. That plot shares important characteristics with the experimental melting curves. All three show three main transitions; they have the same relative peak heights, integrated areas, and relative widths. They differ only in the temperatures at which they occur. We discuss possible reasons for this discrepancy further in the subsequent section discussing influences on the melting.

Thus, the correspondence between experimental Thermofluor and CD melting curves (Fig. 1) and computational occurrences of these two transition events (Fig. 2) is significantly detailed.

Structural metrics—radius of gyration, R_g, and percentage helix—computed from the mean coordinates of the DMD trajectories accompany the simulated heat capacity changes obtained during the DMD simulation. Temperature-dependences of both metrics are consistent with the maxima in the $C v plot.$ Heat capacity maxima are stationary points where the protein absorbs and loses excess heat from and to the system in equal amounts, as it undergoes conformation changes, so heat capacity changes transiently remain constant with respect to system temperature. In keeping with stationary behavior, local maxima in the $C v$ traces for the two major transitions, coincide with discontinuities in the slopes of both structural metrics, each changing faster before the $C v$ maximum and more slowly afterward (black tracings, Fig. 2).

The average R_g increases by 14% during the second transition, corresponding to the increase expected for a molten globule transition (15%).⁴⁴ R_g increases again by 20% at 87 °C. At the 67 °C transition temperature for the second peak of the specific heat capacity curve, 70% of the helices present in the native protein remain intact. At 87 °C, the third transition temperature, only 28% of the native helices remain intact. Each heat capacity peak, therefore, corresponds to a structurally different transition in both metrics and thus to a different unfolding process.

The initial peak in the $C v$ trace is broader, has a smaller amplitude, and involves minimal changes in either structural metric. It is evident mainly because it initiates the increasing variance of R_g, which occurs almost exclusively with the formation of the molten globule. As 60 °C coincides with the optimal growth temperature for Geobacillus stearothermophilus, it is possible that this local maximum in $C v$ has biological significance. However, temperature-dependent activity curves for native TrpRS⁴⁵ are not informative on this point.

Figure 3 compares transitions in experimental and simulated melting curves. The computational melting transitions distribute substantially more widely than the experimental ones in Fig. 1. The dimer interface buries ∼2160 Ų of surface area in the PreTS state. That value suggests a rather low dimer dissociation constant 10⁻⁸ < K_D < 10⁻⁶.⁴⁶ We previously used scanning force microscopy to count monomers and dimers at successively higher dilutions to measure the strength of the dimer interface.⁴⁷ Those estimates, 6.6 ± 4  × 10⁻⁹ for unliganded and 1.1 ± 1 × 10⁻⁸ M for a complex with a non-hydrolysable aminoacyl-5′-adenylate product analog, are at the lower end of that range. Thus, considerable binding free energy is bound into the dimer dissociation constant, which probably induces a considerable difference between the melting behavior of monomer and dimer. Although we have no ready means to verify the assumption, we reason that much of the difference between the three melting temperatures in Fig. 3 may arise because we simulated the TrpRS monomer, not the dimer, and that the dimer interface also compresses the separation between transitions.

Consistent with that assumption, Pre-TS ligands compress the experimental traces still further to a narrower temperature range over which the three melting events occur (dashed lines above the scale vs solid lines above the temperature scale). Simulation of the monomer shows the same three melting events seen in the DSC of the dimer, but they are separated much more widely in temperature. Structural cartoons through the middle and high temperature thermal events both entail significant volume increases, supporting identification of the middle transition as formation of a molten globule.

We assembled structural snapshots over the course of the exchange trajectory from each replica at given temperatures, noted in Fig. 3. The average structures at each temperature were analyzed for native helical content using DSSP as indicated in Fig. 1. Notably, melting is localized to both the connecting peptide 1 (CP1) and anticodon-binding (ABD) domains (deep purple surfaces). The transition at 65 °C can reasonably be assigned to molten globule formation because tertiary structures of those two domains are expanded, but not unfolded.

Ding et al.⁴⁸ suggested the persistence of a core 3D structure in unfolded forms of many proteins. Remarkably, for TrpRS, that core, Fig. 4, coincides closely with its urzyme (ur = primordial, authentic, original + enzyme⁴⁹), a model for the ancestral enzyme that itself appears to be a catalytically active molten globule.⁵⁰ The TrpRS urzyme was constructed by fusing two disjoint segments with similar secondary structures—N- and C-terminal β-α-β crossover connections.⁵¹ One of the core helices that remains intact at 94 °C in the computational profile is the long helix (154–165) from the C-terminal segment (salmon). The sequence of that segment has an unusually high fundamental helix propensity as determined by the physics-based and unbiased database-independent Agadir algorithm.⁵² In contrast, the helix in the N-terminal segment, has a far lower propensity, but its first eight residues nevertheless remain helical at 368 K, suggesting that extrinsic forces sustain it. Further discussion in the accompanying paper²⁶ considers what these extrinsic forces might be.

Hilvert's work with chorismate mutase^53,54 identified an efficient, potentially widespread coupling between folding and catalysis—i.e., high transition-state affinity. In light of those findings, it is notable that the last part of the TrpRS monomer to melt in the REX/DMD free energy simulations closely approximates the TrpRS urzyme, a construct studied as a model for the evolution of aminoacyl-tRNA synthetases (Fig. 4). At 94 °C, the CP1 and ABD domains, as well as the long-C-terminal alpha helix are all fully unfolded. Thus, the modules that remain intact at the highest temperatures are also likely the most ancient part of the protein which, when excerpted, retains a full range of catalytic activities.^49,51,55 The structural integrity of the TrpRS urzyme at high temperature is consistent with the experimental observation that the urzyme itself seems to function as a molten globule,⁵⁰ whose only well-folded state is its complex with the transition state for amino acid activation.

G. stearothermophilus tryptophanyl-tRNA synthetase is notable because multidisciplinary mechanistic studies have worked out reciprocally coupled gates making catalysis conditional to domain motion and, paradoxically, making domain motion conditional on catalysis. Maximizing the efficiency of converting the free energy of ATP hydrolysis to useful forms of work and information requires both gating functions.^4,10 This paper describes thermal melting studies pursuant to attributing thermodynamic aspects of gating to specific amino acid residues in the D1 switch, a molecular switching motif that imposes multi-state behavior. The accompanying paper²⁶ outlines a paradigm for using high throughput differential scanning fluorimetry to derive thermodynamic estimates for inter-residue energetic coupling, potentially in a ligand-dependent framework.

This work was supported by NIGMS research Grant Nos. R01–40906 and R01-78227 as well as the Alfred P. Slan Foundation Matter-to-Life program, Grant No. G-2021-16944 to C.W.C., Jr. and by R35 GM134864 and the Passan Foundation; National Science Foundation, Award No. 2210963 to N.V.D. Michael Dzuricky (Chilkoti Laboratory, Biomedical Engineering Department, Duke University) provided a microcalorimeter and advised on its use.

The authors have no conflicts to disclose.

Srinivas Niranj Chandrasekaran: Conceptualization (equal); Data curation (equal); Investigation (equal); Methodology (equal); Project administration (equal); Writing – review & editing (equal). Jhuma Das: Methodology (equal); Writing – review & editing (equal). Nikolay V. Dokholyan: Methodology (equal); Writing – review & editing (equal). Charles W. Carter: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.