Multidimensional Free Energy Surface Research Articles

Assessing the Intestinal Permeability of Small Molecule Drugs via Diffusion Motion on a Multidimensional Free Energy Surface.

A protocol that accurately assesses the intestinal permeability of small molecule compounds plays an essential role in decreasing the cost and time in inventing a new drug. This manuscript presents a novel computational method to study the passive permeation of small molecule drugs based on the inhomogeneous solubility-diffusion model. The multidimensional free energy surface of the drug transiting through a lipid bilayer is computed with transition-tempered metadynamics that accurately captures the mechanisms of passive permeation. The permeability is computed by following the diffusion motion of the drug molecules along the minimal free energy path found on the multidimensional free energy surface. This computational method is assessed by studying the permeability of five small molecule drugs (ketoprofen, naproxen, metoprolol, propranolol, and salicylic acid). The results demonstrate a remarkable agreement between the computed permeabilities and those measured with the intestinal assay. The in silico method reported in this manuscript also reproduces the permeability measured from the intestinal assay (in vivo) better than the cell-based assays (e.g., PAMPA and Caco-2) do. In addition, the multidimensional free energy surface reveals the interplay between the structure of the small molecule and its permeability, shedding light on strategies of drug optimization.

Force-Correction Analysis Method for Derivation of Multidimensional Free-Energy Landscapes from Adaptively Biased Replica Simulations.

A methodology is proposed for the calculation of multidimensional free-energy landscapes of molecular systems, based on analysis of multiple molecular dynamics trajectories wherein adaptive biases have been applied to enhance the sampling of different collective variables. In this approach, which we refer to as the Force-Correction Analysis Method (FCAM), local averages of the total and biasing forces are evaluated post hoc, and the latter are subtracted from the former to obtain unbiased estimates of the mean force across collective-variable space. Multidimensional free-energy surfaces and minimum free-energy pathways are then derived by integrating the mean-force landscape with a kinetic Monte Carlo algorithm. To evaluate the proposed method, a series of numerical tests and comparisons with existing approaches were carried out for small molecules, peptides, and proteins, based on all-atom trajectories generated with standard, concurrent, and replica-exchange metadynamics in collective-variable spaces ranging from one to six dimensional. The tests confirm the correctness of the FCAM formulation and demonstrate that calculated mean forces and free energies converge rapidly and accurately, outperforming other methods used to unbias this kind of simulation data.

Fast and Accurate Multidimensional Free Energy Integration.

Enhanced sampling and free energy calculation algorithms of the thermodynamic integration family (such as the adaptive biasing force (ABF) method) are not based on the direct computation of a free energy surface but rather of its gradient. Integrating the free energy surface is nontrivial in dimensions higher than one. Here, the author introduces a flexible, portable implementation of a Poisson equation formalism to integrate free energy surfaces from estimated gradients in dimensions 2 and 3 using any combination of periodic and nonperiodic (Neumann) boundary conditions. The algorithm is implemented in portable C++ and provided as a standalone tool that can be used to integrate multidimensional gradient fields estimated on a grid using any algorithm, such as umbrella integration as a post-treatment of umbrella sampling simulations. It is also included in the implementation of ABF (and its extended-system variant eABF) in the Collective Variables Module, enabling the seamless computation of multidimensional free energy surfaces within ABF and eABF simulations. A Python-based analysis toolchain is provided to easily plot and analyze multidimensional ABF simulation results, including metrics to assess their convergence. The Poisson integration algorithm can also be used to perform Helmholtz decomposition of noisy gradient estimates on the fly, resulting in an efficient implementation of the projected ABF (pABF) method proposed by Leliévre and co-workers. In numerical tests, pABF is found to lead to faster convergence with respect to ABF in simple cases of low intrinsic dimension but seems detrimental to convergence in a more realistic case involving degenerate coordinates and hidden barriers due to slower exploration. This suggests that variance reduction schemes do not always yield convergence improvements when applied to enhanced sampling methods.

Double Domain Swapping in Human γC and γD Crystallin Drives Early Stages of Aggregation.

Human γD (HγD) and γC (HγC) are two-domain crystallin (Crys) proteins expressed in the nucleus of the eye lens. Structural perturbations in the protein often trigger aggregation, which eventually leads to cataract. To decipher the underlying molecular mechanism, it is important to characterize the partially unfolded conformations, which are aggregation-prone. Using a coarse grained protein model and molecular dynamics simulations, we studied the role of on-pathway folding intermediates in the early stages of aggregation. The multidimensional free energy surface revealed at least three different folding pathways with the population of partially structured intermediates. The two dominant pathways confirm sequential folding of the N-terminal [Ntd] and the C-terminal domains [Ctd], while the third, least favored, pathway involves intermediates where both the domains are partially folded. A native-like intermediate (I*), featuring the folded domains and disrupted interdomain contacts, gets populated in all three pathways. I* forms domain swapped dimers by swapping the entire Ntds and Ctds with other monomers. Population of such oligomers can explain the increased resistance to unfolding resulting in hysteresis observed in the folding experiments of HγD Crys. An ensemble of double domain swapped dimers are also formed during refolding, where intermediates consisting of partially folded Ntds and Ctds swap secondary structures with other monomers. The double domain swapping model presented in our study provides structural insights into the early events of aggregation in Crys proteins and identifies the key secondary structural swapping elements, where introducing mutations will aid in regulating the overall aggregation propensity.

Delineating the conformational dynamics of intermediate structures on the unfolding pathway of β-lactoglobulin in aqueous urea and dimethyl sulfoxide

The funnel shaped energy landscape model of the protein folding suggests that progression of folding proceeds through multiple pathways, having the multiple intermediates which leads to multidimensional free-energy surface. Herein, we applied all-atom MD simulation to conduct a comparative study on the structure of β-lactoglobulin (β-LgA) in aqueous mixture of 8 M urea and 8 M dimethyl sulfoxide (DMSO), at different temperatures. The cumulative results of multiple simulations suggest a common unfolding pathway of β-LgA, occurred through the stable and meta-stable intermediates (I), in both urea and DMSO. However, the free-energy landscape (FEL) analyses show that the structural transitions of I-states are energetically different. In urea, FEL shows distinct ensemble of intermediates, I1 and I2, separated by the energy barrier of ∼3.0 kcal mol−1. Similarly, we find the population of two distinct I1 and I2 states in DMSO, however, the I1 appeared transiently around ∼30–35 ns and is short-lived. But, the I2 ensemble is observed structurally compact and long-lived (∼50–150 ns) as compared to unfolding in urea. Furthermore, the I1 and I2 are separated through a high energy barrier of ∼6.0 kcal mol−1. Thus, our results provide the structural insights of intermediates which essentially bear the signature of a different unfolding pathway of β-LgA in urea and DMSO.Abbreviationsβ-LgAβ-lactoglobulinDMSOdimethyl sulfoxideFELfree-energy landscapeGdmClguanidinium chlorideIintermediate stateMGmolten globule statePMEparticle mesh EwaldQfraction of native contactsRMSDroot mean square deviationRMSFroot mean square fluctuationRgradius of gyrationSASAsolvent Accessible Surface AreascSASAthe side chain SASATrptryptophanCommunicated by Ramaswamy H. Sarma

Taming Rugged Free Energy Landscapes Using an Average Force.

The observation of complex structural transitions in biological and abiological molecular objects within time scales amenable to molecular dynamics (MD) simulations is often hampered by significant free energy barriers associated with entangled movements. Importance-sampling algorithms, a powerful class of numerical schemes for the investigation of rare events, have been widely used to extend simulations beyond the time scale common to MD. However, probing processes spanning milliseconds through microsecond molecular simulations still constitutes in practice a daunting challenge because of the difficulty of taming the ruggedness of multidimensional free energy surfaces by means of naive transition coordinates. To address this limitation, in recent years we have elaborated importance-sampling methods relying on an adaptive biasing force (ABF). In this Account, we review recent developments of algorithms aimed at mapping rugged free energy landscapes that correspond to complex processes of physical, chemical, and biological relevance. Through these developments, we have broadened the spectrum of applications of the popular ABF algorithm while improving its computational efficiency, notably for multidimensional free energy calculations. One major algorithmic advance, coined meta-eABF, merges the key features of metadynamics and an extended Lagrangian variant of ABF (eABF) by simultaneously shaving the barriers and flooding the valleys of the free energy landscape, and it possesses a convergence rate up to 5-fold greater than those of other importance-sampling algorithms. Through faster convergence and enhanced ergodic properties, meta-eABF represents a significant step forward in the simulation of millisecond-time-scale events. Here we introduce extensions of the algorithm, notably its well-tempered and replica-exchange variants, which further boost the sampling efficiency while gaining in numerical stability, thus allowing quantum-mechanical/molecular-mechanical free energy calculations to be performed at a lower cost. As a paradigm to bridge microsecond simulations to millisecond events by means of free energy calculations, we have applied the ABF family of algorithms to decompose complex movements in molecular objects of biological and abiological nature. We show here how water lubricates the shuttling of an amide-based rotaxane by altering the mechanism that underlies the concerted translation and isomerization of the macrocycle. Introducing novel collective variables in a computational workflow for the rigorous determination of standard binding free energies, we predict with utmost accuracy the thermodynamics of protein-ligand reversible association. Because of their simplicity, versatility, and robust mathematical foundations, the algorithms of the ABF family represent an appealing option for the theoretical investigation of a broad range of problems relevant to physics, chemistry, and biology.

A novel ring-shaped reaction pathway with interconvertible intermediates in chitinase A as revealed by QM/MM simulation combined with a one-dimensional projection technique.

Substrate-assisted catalysis (SAC), a mechanism of chitin hydrolysis by chitinases belonging to the glycoside hydrolase family 18 (GH18), has been studied experimentally and theoretically for several decades. However, the detailed reaction mechanism in chitinase A (ChiA) remains unclear at the atomic level. In this study, we investigated glycosylation, the first step of SAC, of ChiA obtained from Serratia marcescens (SmChiA), using QM/MM simulations combined with a one-dimensional projection (ODP) technique, which enabled us to explore the multi-dimensional free energy surface efficiently. The results showed that the reaction proceeds via a novel ring-shaped concerted reaction pathway with interconvertible intermediates, viz. oxazolinium ion and oxazoline, which have not been fully identified in previous studies. We also compared this chitin hydrolysis mechanism in SmChiA with that in SmChiB reported previously. The computational protocol developed in this study could also be applicable for elucidating complicated reaction mechanisms in other enzymes.

Tests of Kramers' Theory at the Single-Molecule Level: Evidence for Folding of an Isolated RNA Tertiary Interaction at the Viscous Speed Limit.

Dissipation and friction influence the conformational dynamics of biological polymers as they traverse barriers on rugged free energy surfaces. It is well established that the "speed limit" for macromolecular folding is dictated by a combination of (i) solvent friction, which depends on solvent viscosity, η, and (ii) internal friction, which is independent of solvent and depends solely on the molecular folding pathway. In this work, single-molecule Förster resonance energy transfer (FRET) confocal spectroscopy is used to study viscosity-dependent folding kinetics of an isolated RNA tertiary motif, that of the GAAA tetraloop receptor, allowing both solvent and internal frictional contributions to be investigated and extracted independently for both flexible PEG- and RNA-based (rU7, rA7) linkers in the unimolecular construct. Specifically, our single-molecule data reveal that (i) folding rate constants scale linearly with the inverse solvent viscosity (η), which supports Kramers'/Grote-Hynes' rate theory for η-dependent RNA folding and that (ii) they provide quantitative upper limits for the intrinsic viscosity, [ηint ≈ 0.1(2) cP], arising from internal friction associated with folding/unfolding of an isolated RNA tertiary interaction. Furthermore, in contrast to strong viscosity-induced shifts in the folding/unfolding rate constants, temperature-dependent studies demonstrate that the enthalpic, entropic, and free energy contributions to the transition state barrier are largely insensitive to the solvent viscosity. This supports a very simple picture for the conformational kinetics of isolated RNA tertiary interactions wherein rate constants for folding/unfolding are both inversely dependent on viscosity and limited by diffusional access to the transition state region on a multidimensional free energy surface. Particularly under cellular conditions, where ηsolv > 1 cp, this suggests that RNAs fold/unfold at a "speed limit" dictated by solvent viscosity and transition-state barrier thermodynamics rather than internal molecular friction.

Reaction Coordinate, Free Energy, and Rate of Intramolecular Proton Transfer in Human Carbonic Anhydrase II.

The role of structure and dynamics of an enzyme has been investigated at three different stages of its function including the chemical event it catalyzes. A one-pot computational method has been designed for each of these stages on the basis of classical and/or quantum mechanical-molecular mechanical molecular dynamics and transition path sampling simulations. For a pair of initial and final states A and B separated by a high free-energy barrier, using a two-stage selection process, several collective variables (CVs) are identified that can delineate A and B. However, these CVs are found to exhibit strong cross-coupling over the transition paths. A set of mutually orthogonal order parameters is then derived from these CVs and an optimal reaction coordinate, r, determined applying half-trajectory likelihood maximization along with a Bayesian information criterion. The transition paths are also used to project the multidimensional free energy surface and barrier crossing dynamics along r. The proposed scheme has been applied to the rate-determining intramolecular proton transfer reaction of the well-known enzyme human carbonic anhydrase II. The potential of mean force, F( r), in the absence of the chemical step is found to reproduce earlier results on the equilibrium population of two side-chain orientations of key residue His-64. Estimation of rate constants, k, from mean first passage times for the three different stages of catalysis shows that the rate-determining step of intramolecular proton transfer occurs with k ≃ 1.0 × 106 s-1, in close agreement with known experimental results.

Effects of hydration on the protonation state of a lysine analog crossing a phospholipid bilayer - insights from molecular dynamics and free-energy calculations.

The low bioavailability of most therapeutic compounds is often counterbalanced by association with molecular vectors capable of crossing cell membranes. Previous studies demonstrated that for vectors bearing titratable chemical groups, the translocation process might be accompanied by a change in the protonation state. For simple compounds e.g. a lysine analog, free energy calculations, using a single collective variable, namely the insertion depth, suggest that such a transition could only take place if the amino acid diffuses deep enough into the hydrophobic core of the membrane, a situation thermodynamically unfavorable. Here, we determined the 2D potential of mean force associated with the translocation of lysine across a model membrane using as reaction coordinates not only its location in the bilayer but also its hydration. Our results cogently demonstrate that the change in protonation can result from a small fluctuation in the latter, even at low insertion depth.

Reaction coordinates and transition states in enzymatic catalysis

Enzymatic reactions are complex chemical processes taking place in complex dynamic environments. Theoretical characterization of these reactions requires the determination of the reaction coordinate and the transition state ensemble. This is not an easy task because many degrees of freedom may be involved in principle. We present recent efforts to find good enzymatic reaction coordinates and the implications of these findings in the interpretation of enzymatic efficiency. In particular, we analyze different strategies based on the use of minimum free energy paths and direct localization of the dividing surface on multidimensional free energy surfaces. Another strategy is based on the generation of reactive trajectories, using the transition path sampling method, from which transition state configurations can be harvested. Most of the applications carried out until now coincide to stress the change in the nature of the reaction coordinate, in terms of the participation of the chemical and environmental degrees of freedom, as the reaction advances. The degrees of freedom of the chemical system are dominant at the transition state while environmental participation can be more important at early or late stages of the process. WIREs Comput Mol Sci 2018, 8:e1329. doi: 10.1002/wcms.1329This article is categorized under: Structure and Mechanism > Computational Biochemistry and Biophysics Molecular and Statistical Mechanics > Free Energy Methods Software > Simulation Methods

Constructing a multidimensional free energy surface like a spider weaving a web.

Complete free energy surface in the collective variable space provides important information of the reaction mechanisms of the molecules. But, sufficient sampling in the collective variable space is not easy. The space expands quickly with the number of the collective variables. To solve the problem, many methods utilize artificial biasing potentials to flatten out the original free energy surface of the molecule in the simulation. Their performances are sensitive to the definitions of the biasing potentials. Fast-growing biasing potential accelerates the sampling speed but decreases the accuracy of the free energy result. Slow-growing biasing potential gives an optimized result but needs more simulation time. In this article, we propose an alternative method. It adds the biasing potential to a representative point of the molecule in the collective variable space to improve the conformational sampling. And the free energy surface is calculated from the free energy gradient in the constrained simulation, not given by the negative of the biasing potential as previous methods. So the presented method does not require the biasing potential to remove all the barriers and basins on the free energy surface exactly. Practical applications show that the method in this work is able to produce the accurate free energy surfaces for different molecules in a short time period. The free energy errors are small in the cases of various biasing potentials. © 2017 Wiley Periodicals, Inc.

Metadynamics for Perspective Drug Design: Computationally Driven Synthesis of New Protein-Protein Interaction Inhibitors Targeting the EphA2 Receptor.

Metadynamics (META-D) is emerging as a powerful method for the computation of the multidimensional free-energy surface (FES) describing the protein-ligand binding process. Herein, the FES of unbinding of the antagonist N-(3α-hydroxy-5β-cholan-24-oyl)-l-β-homotryptophan (UniPR129) from its EphA2 receptor was reconstructed by META-D simulations. The characterization of the free-energy minima identified on this FES proposes a binding mode fully consistent with previously reported and new structure-activity relationship data. To validate this binding mode, new N-(3α-hydroxy-5β-cholan-24-oyl)-l-β-homotryptophan derivatives were designed, synthesized, and tested for their ability to displace ephrin-A1 from the EphA2 receptor. Among them, two antagonists, namely compounds 21 and 22, displayed high affinity versus the EphA2 receptor and resulted endowed with better physicochemical and pharmacokinetic properties than the parent compound. These findings highlight the importance of free-energy calculations in drug design, confirming that META-D simulations can be used to successfully design novel bioactive compounds.

Unraveling the Reaction Mechanism of Enzymatic C5-Cytosine Methylation of DNA. A Combined Molecular Dynamics and QM/MM Study of Wild Type and Gln119 Variant

M.HhaI is a DNA methyltransferase from Haemophilus hemolyticus that catalyzes the transfer of a methyl group from S-adenosyl-l-methionine (SAM) to the C5 position of a cytosine. This enzyme is a paradigmatic model for C5 DNA methyltransferases due to its major homology to mammalian enzymes and to the availability of high-resolution structures of the DNA–enzyme complex. In spite of the number of experimental and theoretical analyses carried out for this system, many mechanistic details remain unraveled. We have used full atomistic classical molecular dynamics simulations to explore the protein–SAM–DNA ternary complex, where the target cytosine base is flipped out into the active site for both the wild type and Glu119Gln mutant. The relaxed structure was used for a combined quantum mechanics/molecular mechanics exploration of the reaction mechanism using the string method. Exploration of multidimensional free energy surfaces allowed determining a complete picture of the reaction mechanism, in agreement with...

Efficient Computational Research Protocol to Survey Free Energy Surface for Solution Chemical Reaction in the QM/MM Framework: The FEG-ER Methodology and Its Application to Isomerization Reaction of Glycine in Aqueous Solution.

In solution chemical reaction, we often need to consider a multidimensional free energy (FE) surface (FES) which is analogous to a Born-Oppenheimer potential energy surface. To survey the FES, an efficient computational research protocol is proposed within the QM/MM framework; (i) we first obtain some stable states (or transition states) involved by optimizing their structures on the FES, in a stepwise fashion, finally using the free energy gradient (FEG) method, and then (ii) we directly obtain the FE differences among any arbitrary states on the FES, efficiently by employing the QM/MM method with energy representation (ER), i.e., the QM/MM-ER method. To validate the calculation accuracy and efficiency, we applied the above FEG-ER methodology to a typical isomerization reaction of glycine in aqueous solution, and reproduced quite satisfactorily the experimental value of the reaction FE. Further, it was found that the structural relaxation of the solute in the QM/MM force field is not negligible to estimate correctly the FES. We believe that the present research protocol should become prevailing as one computational strategy and will play promising and important roles in solution chemistry toward solution reaction ergodography.

Sculpting bespoke mountains: Determining free energies with basis expansions.

The intriguing behavior of a wide variety of physical systems, ranging from amorphous solids or glasses to proteins, is a direct manifestation of underlying free energy landscapes riddled with local minima separated by large barriers. Exploring such landscapes has arguably become one of statistical physics's great challenges. A new method is proposed here for uniform sampling of rugged free energy surfaces. The method, which relies on special Green's functions to approximate the Dirac delta function, improves significantly on existing simulation techniques by providing a boundary-agnostic approach that is capable of mapping complex features in multidimensional free energy surfaces. The usefulness of the proposed approach is established in the context of a simple model glass former and model proteins, demonstrating improved convergence and accuracy over existing methods.

Multidimensional free energy surface of unfolding of HP-36: microscopic origin of ruggedness.

The protein folding funnel paradigm suggests that folding and unfolding proceed as directed diffusion in a multidimensional free energy surface where a multitude of pathways can be traversed during the protein's sojourn from initial to final state. However, finding even a single pathway, with the detail chronicling of intermediates, is an arduous task. In this work we explore the free energy surface of unfolding pathway through umbrella sampling, for a small globular α-helical protein chicken-villin headpiece (HP-36) when the melting of secondary structures is induced by adding DMSO in aqueous solution. We find that the unfolding proceeds through the initial separation or melting of aggregated hydrophobic core that comprises of three phenylalanine residues (Phe7, Phe11, and Phe18). This separation is accompanied by simultaneous melting of the second helix. Unfolding is found to be a multistage process involving crossing of three consecutive minima and two barriers at the initial stage. At a molecular level, Phe18 is observed to reorient itself towards other hydrophobic grooves to stabilize the intermediate states. We identify the configuration of the intermediates and correlate the intermediates with those obtained in our previous works. We also give an estimate of the barriers for different transition states and observe the softening of the barriers with increasing DMSO concentration. We show that higher concentration of DMSO tunes the unfolding pathway by destabilizing the third minimum and stabilizing the second one, indicating the development of a solvent modified, less rugged pathway. The prime outcome of this work is the demonstration that mixed solvents can profoundly transform the nature of the energy landscape and induce unfolding via a modified route. A successful application of Kramer's rate equation correlating the free energy simulation results shows faster rate of unfolding with increasing DMSO concentration. This work perhaps presents the first systematic theoretical study of the effect of a chemical denaturant on the microscopic free energy surface and rates of unfolding of HP-36.

Hierarchical Folding Free Energy Landscape of HP35 Revealed by Most Probable Path Clustering

Adopting extensive molecular dynamics simulations of villin headpiece protein (HP35) by Shaw and co-workers, a detailed theoretical analysis of the folding of HP35 is presented. The approach is based on the recently proposed most probable path algorithm which identifies the metastable states of the system, combined with dynamical coring of these states in order to obtain a consistent Markov state model. The method facilitates the construction of a dendrogram associated with the folding free-energy landscape of HP35, which reveals a hierarchical funnel structure and shows that the native state is rather a kinetic trap than a network hub. The energy landscape of HP35 consists of the entropic unfolded basin U, where the prestructuring of the protein takes place, the intermediate basin I, which is connected to U via the rate-limiting U → I transition state reflecting the formation of helix-1, and the native basin N, containing a state close to the NMR structure and a native-like state that exhibits enhanced fluctuations of helix-3. The model is in line with recent experimental observations that the intermediate and native states differ mostly in their dynamics (locked vs unlocked states). Employing dihedral angle principal component analysis, subdiffusive motion on a multidimensional free-energy surface is found.

Toward an Automatic Determination of Enzymatic Reaction Mechanisms and Their Activation Free Energies.

We present a combination of the string method and a path collective variable for the exploration of the free energy surface associated to a chemical reaction in condensed environments. The on-the-fly string method is employed to find the minimum free energy paths on a multidimensional free energy surface defined in terms of interatomic distances, which is a convenient selection to study bond forming/breaking processes. Once the paths have been determined, a reaction coordinate is defined as a measure of the advance of the system along these paths. This reaction coordinate can be then used to trace the reaction Potential of Mean Force from which the activation free energy can be obtained. This combination of methodologies has been here applied to the study, by means of Quantum Mechanics/Molecular Mechanics simulations, of the reaction catalyzed by guanidinoacetate methyltransferase. This enzyme catalyzes the methylation of guanidinoacetate by S-adenosyl-l-methionine, a reaction that involves a methyl transfer and a proton transfer and for which different reaction mechanisms have been proposed.

Electrostatically Guided Dynamics—The Root of Fidelity in a Promiscuous Terpene Synthase?

Terpene cyclases are responsible for the initial cyclization cascade in the multistep synthesis of more than 60,000 known natural products. This abundance of compounds is generated using a very limited pool of substrates based on linear isoprenoids. The astounding chemodiversity obtained by terpene cyclases suggests a tremendous catalytic challenge to these often promiscuous enzymes. In the current study we present a detailed mechanistic view of the biosynthesis of the monoterpene bornyl diphosphate (BPP) from geranyl diphosphate by BPP synthase using state of the art simulation methods. We identify the bornyl cation as an enzyme-induced bifurcation point on the multidimensional free energy surface, connecting between the product BPP and the side product camphene. Chemical dynamics simulations suggest that the active site diphosphate moiety steers reaction trajectories toward product formation. Nonetheless, chemical dynamics is not precise enough for exclusive product formation, providing a rationale for the lack of fidelity in this promiscuous terpene cyclase.