In SilicoFolding of a Three Helix Protein and Characterization of Its Free-Energy Landscape in an All-Atom Force Field

Abstract

We report the reproducible first-principles folding of the 40 amino acid, three-helix headpiece of the HIV accessory protein in a recently developed all-atom free-energy forcefield. Six of twenty simulations using an adapted basin-hopping method converged to better than 3Å backbone RMS deviation to the experimental structure. Using over 60,000 low-energy conformations of this protein, we constructed a decoy tree that completely characterizes its folding funnel.PACS numbers: 87.15. Cc,02.70.Ns,02.60.Pn Available genomic and sequence information for proteins contains a wealth of biomedical information that becomes accessible when translated into three-dimensional structure [1]. While theoretical models for protein structure prediction [2,3] that partially rely on experimental information have shown consistent progress [4], the assessment of de-novo strategies that rely on sequence information alone has been much less favorable [5]. The development of such techniques, in particular of transferable first-principle all-atom folding methods, would significantly benefit the understanding of protein families where little experimental information is available, the prediction of novel folds as well as the investigation of protein association and dynamics which are presently difficult to probe experimentally. Recent progress for small peptides [3,6,7,8] documents both the feasibility of this approach as well as its limitations [9,10], in particualr those associated with the direct simulation of the folding process through molecular dynamics [11].Overwhelming experimental evidence supports the thermodynamic hypothesis [12] that many proteins are in thermodynamic equilibrium with their environment: their native state thus corresponds to the global minimum of their free energy landscape [13]. The free energy of the system is accessible either indirectly by explicit ensemble averaging of the combined internal energy of protein and solvent, or directly in a free-energy forcefield where an implicit solvation model approximates direct interactions with the solvent as well as most of the entropic contributions. We developed an all-atom protein forcefield (PFF01) [8,14,15] with an area-based implicit solvent model that approximates the free energy of peptide conformations in the natural solvent. Using a rational decoy approach this forcefield was optimized to correctly predict the native structure of the 36-amino acid headgroup of villin [9,10,11]. Without further parameter adjustment we then simulated the structurally conserved 40 amino-acid headpiece of the autonomously folding HIV accessory protein (1F4I-40) [16] with a modified basin hopping technique [17,18]. Out of twenty simulations the five energetically lowest correctly reproduced the NMR structure of this three-helix protein with a backbone RMS deviation of less than 3Å. The combination of decoy based model development for the free energy with efficient stochastic optimization methods suggests a viable route for protein structure prediction at the all atom level with p...

Biomolecular Structure Prediction Stochastic Optimization Methods

“…Suitable optimization methods must therefore be able speed the simulation by avoiding high energy transition states, adapt large scale move or accept unphysical intermediates. Here we report on four different optimization methods, the stochastic tunneling method, [9] the basin hopping technique, [10,11] the parallel tempering method [12] and a recently employed evolutionary technique. [8] …”

Comparison of Stochastic Optimization Methods for All‐Atom Folding of the Trp‐Cage Protein

“…[4,5,13] The force field, which strictly speaking parameterizes the internal free energy of the protein excluding backbone entropy, is parameterized with the following nonbonded interactions [Eq. (1)]:…”

“…We developed an all-atom free-energy force field for proteins (PFF01), which is primarily based on physical interactions with important empirical, though sequence-independent, corrections. [5] We have already demonstrated the reproducible and predictive folding of three proteins: the 20 amino acid trp-cage protein (1L2Y), [3,6] the structurally conserved headpiece of the 40 amino acid HIV accessory protein (1F4I), [4,7] and the 60 amino acid bacterial ribosomal protein L20. [8] In addition, we could show that PFF01 stabilizes the native conformations of other proteins, for example, the 52 amino acid protein A [9,10] and the engrailed homeodomain (1ENH) from Drosophila melanogaster [11] as the global optimum of the free energy model.…”

“…We have recently demonstrated a feasible strategy for allatom protein structure prediction [3][4][5] in a minimal thermodynamic approach. We developed an all-atom free-energy force field for proteins (PFF01), which is primarily based on physical interactions with important empirical, though sequence-independent, corrections.…”

Efficient and reproducible folding simulations of the Trp-cage protein with multiscale molecular dynamics

“…The first-principle based ab initio folding simulations at atomistic resolution are currently being successfully applied to small proteins such as Trp-cage [1][2][3] , villin headpiece [4] and a threehelix headpiece of the HIV accessory protein [5] . However, for systems of protein folding which often involve large degrees of freedom occurring on the timescale of microseconds or milliseconds, the capacity of simulation is still limited.…”