QM/MM Calculations on Protein–RNA Complexes: Understanding Limitations of Classical MD Simulations and Search for Reliable Cost-Effective QM Methods

Abstract

Although atomistic explicit-solvent Molecular Dynamics (MD) is a popular tool to study protein-RNA recognition, satisfactory MD description of protein-RNA complexes is not always achieved. Unfortunately, it is often difficult to separate MD simulation instabilities primarily caused by the simple point-charge molecular mechanics (MM) force fields from problems related to the notorious uncertainties in the starting structures. Herein, we report a series of large-scale QM/MM calculations on the U1A protein-RNA complex. This experimentally well-characterized system has an intricate protein-RNA interface, which is very unstable in MD simulations. The QM/MM calculations identify several H-bonds poorly described by the MM method and thus indicate the sources of instabilities of the U1A interface in MD simulations. The results suggest that advanced QM/MM computations could be used to indirectly rationalize problems seen in MM-based MD simulations of protein-RNA complexes. As the most accurate QM method, we employ the computationally demanding meta-GGA density functional TPSS-D3(BJ)/def2-TZVP level of theory. Because considerably faster methods would be needed to extend sampling and to study even larger protein-RNA interfaces, a set of low-cost QM/MM methods is compared to the TPSS-D3(BJ)/def2-TZVP data. The PBEh-3c and B97-3c density functional composite methods appear to be suitable for protein-RNA interfaces. In contrast, HF-3c and the tight-binding Hamiltonians DFTB3-D3 and GFN-xTB perform unsatisfactorily and do not provide any advantage over the MM description. These conclusions are supported also by similar analysis of a simple HutP protein-RNA interface, which is well-described by MD with the exception of just one H-bond. Some other methodological aspects of QM/MM calculations on protein-RNA interfaces are discussed.

Improving The Performance Of The Amber Rna Force Field By Tuning The Hydrogen-Bonding Interactions

“…Finally, also high-quality QM calculations suggest that relying only on the common Lennard-Jones sphere and point-charge model in description of H-bonding has inevitable accuracy limitations. [75][76] In this work, we introduce a generalized formulation of the HBfix potential, in order to tune all interactions of the same kind, henceforth labelled as gHBfix. Most importantly, the gHBfix correction can be applied without previous knowledge of the native structure.…”

UNCG RNA tetraloop as a formidable force-field challenge for MD simulations

“…The resulting 25 starting structures are denoted as 5a-e, 6a-e, 7a-e, 8a-e and 9a-e. Water distribution may influence the relaxed solute structures as shown in ref. 26. For the sake of simplicity, only results for structures after the 100 ps-long solvent equilibration (i.e., structures 5-9e) are documented in the main text while the full set of data can be found in Supporting Information.…”

“…The additive QM/MM scheme with point-charge approximation for electrostatic embedding was used for all QM/MM calculations. The QM/MM module of AMBER 14 48,49 was coupled with TURBOMOLE V7.3 50,51 using an in-house modified version of Sander 26,52,53 from the AMBER 14 program package. For geometry optimization the limited-memory Broyden-Fletcher-Goldfarb-Shanno quasi-Newton algorithm (L-BFGS) 54 was used with a convergence threshold of 1 e −4 kcal⋅mol −1 ⋅ Å −1 for the gradient norm.…”

“…Similarly to the work by Pokorna et al, 26 QM/MM optimizations were mirrored by equivalent MM optimizations from identical starting structures. For the sake of completeness we note that the ff99bsc0χ OL3 was used for structures 1-5 and 9 and χ OL3CP for structures 6-8; the ff difference is, however, marginal for optimizations (see Figure S1 in the Supporting Information).…”

“…Then, we try to unravel specific ff deficiencies using massive quantum mechanical/molecular mechanical (QM/MM) 25 and QM calculations. Obviously, as discussed elsewhere, [26][27][28] QM and QM/MM calculations cannot replace MD simulations due to the lack of Boltzmann sampling. There is generally a limited transferability between calculations of potential-energy surfaces (gradient optimizations or energy scans) as done by QM methods and free energies that determine populations in MD simulations.…”