Bayesian inference of conformational state populations from computational models and sparse experimental observables

We present a Bayesian inference approach to estimating conformational state populations from a combination of molecular modeling and sparse experimental data. Unlike alternative approaches, our method is designed for use with small molecules and emphasizes high-resolution structural models, using inferential structure determination with reference potentials, and Markov Chain Monte Carlo to sample the posterior distribution of conformational states. As an application of the method, we determine solution-state conformational populations of the 14-membered macrocycle cineromycin B, using a combination of previously published sparse Nuclear Magnetic Resonance (NMR) observables and replica-exchange molecular dynamic/Quantum Mechanical (QM)-refined conformational ensembles. Our results agree better with experimental data compared to previous modeling efforts. Bayes factors are calculated to quantify the consistency of computational modeling with experiment, and the relative importance of reference potentials and other model parameters.

Computational screening and selection of cyclic peptide hairpin mimetics by molecular simulation and kinetic network models

Designing peptidomimetic compounds to have a preorganized structure in solution is highly nontrivial. To show how simulation-based approaches can help speed this process, we performed an extensive simulation study of designed cyclic peptide mimics of a β-hairpin from bacterial protein LapD involved in a protein-protein interaction (PPI) pertinent to bacterial biofilm formation. We used replica exchange molecular dynamics (REMD) simulation to screen 20 covalently cross-linked designs with varying stereochemistry and selected the most favorable of these for massively parallel simulation on Folding@home in explicit solvent. Markov state models (MSMs) built from the trajectory data reveal how subtle chemical modifications can have a significant effect on conformational populations, leading to the overall stabilization of the target structure. In particular, we identify a key steric interaction between a methyl substituent and a valine side chain that acts to allosterically shift population between native and near-native states, which could be exploited in future designs. Visualization of this mechanism is aided considerably by the tICA method, which identifies degrees of freedom most important in slow conformational transitions. The combination of quantitative detail and human comprehension provided by MSMs suggests such approaches will be increasingly useful for design.

Probing antibody internal dynamics with fluorescence anisotropy and molecular dynamics simulations

The solution dynamics of antibodies are critical to antibody function. We explore the internal solution dynamics of antibody molecules through the combination of time-resolved fluorescence anisotropy experiments on IgG1 with more than two microseconds of all-atom molecular dynamics (MD) simulations in explicit water, an order of magnitude more than in previous simulations. We analyze the correlated motions with a mutual information entropy quantity, and examine state transition rates in a Markov-state model, to give coarse-grained descriptors of the motions. Our MD simulations show that while there are many strongly correlated motions, antibodies are highly flexible, with F_ab and F_c domains constantly forming and breaking contacts, both polar and non-polar. We find that salt bridges break and reform, and not always with the same partners. While the MD simulations in explicit water give the right time scales for the motions, the simulated motions are about 3-fold faster than the experiments. Overall, the picture that emerges is that antibodies do not simply fluctuate around a single state of atomic contacts. Rather, in these large molecules, different atoms come in contact during different motions.

Slow unfolded-state structuring in Acyl-CoA binding protein folding revealed by simulation and experiment

Protein folding is a fundamental process in biology, key to understanding many human diseases. Experimentally, proteins often appear to fold via simple two- or three-state mechanisms involving mainly native-state interactions, yet recent network models built from atomistic simulations of small proteins suggest the existence of many possible metastable states and folding pathways. We reconcile these two pictures in a combined experimental and simulation study of acyl-coenzyme A binding protein (ACBP), a two-state folder (folding time ~10 ms) exhibiting residual unfolded-state structure, and a putative early folding intermediate. Using single-molecule FRET in conjunction with side-chain mutagenesis, we first demonstrate that the denatured state of ACBP at near-zero denaturant is unusually compact and enriched in long-range structure that can be perturbed by discrete hydrophobic core mutations. We then employ ultrafast laminar-flow mixing experiments to study the folding kinetics of ACBP on the microsecond time scale. These studies, along with Trp-Cys quenching measurements of unfolded-state dynamics, suggest that unfolded-state structure forms on a surprisingly slow (~100 μs) time scale, and that sequence mutations strikingly perturb both time-resolved and equilibrium smFRET measurements in a similar way. A Markov state model (MSM) of the ACBP folding reaction, constructed from over 30 ms of molecular dynamics trajectory data, predicts a complex network of metastable stables, residual unfolded-state structure, and kinetics consistent with experiment but no well-defined intermediate preceding the main folding barrier. Taken together, these experimental and simulation results suggest that the previously characterized fast kinetic phase is not due to formation of a barrier-limited intermediate but rather to a more heterogeneous and slow acquisition of unfolded-state structure.

Calculation of rate spectra from noisy time series data

As the resolution of experiments to measure folding kinetics continues to improve, it has become imperative to avoid bias that may come with fitting data to a predetermined mechanistic model. Toward this end, we present a rate spectrum approach to analyze timescales present in kinetic data. Computing rate spectra of noisy time series data via numerical discrete inverse Laplace transform is an ill-conditioned inverse problem, so a regularization procedure must be used to perform the calculation. Here, we show the results of different regularization procedures applied to noisy multiexponential and stretched exponential time series, as well as data from time-resolved folding kinetics experiments. In each case, the rate spectrum method recapitulates the relevant distribution of timescales present in the data, with different priors on the rate amplitudes naturally corresponding to common biases toward simple phenomenological models. These results suggest an attractive alternative to the "Occam's razor" philosophy of simply choosing models with the fewest number of relaxation rates.

Peptoid conformational free energy landscapes from implicit-solvent molecular simulations in AMBER

To test the accuracy of existing AMBER force field models in predicting peptoid conformation and dynamics, we simulated a set of model peptoid molecules recently examined by Butterfoss et al. (JACS 2009, 131, 16798-16807) using QM methods as well as three peptoid sequences with experimentally determined structures. We found that AMBER force fields, when used with a Generalized Born/Surface Area (GBSA) implicit solvation model, could accurately reproduce the peptoid torsional landscape as well as the major conformers of known peptoid structures. Enhanced sampling by replica exchange molecular dynamics (REMD) using temperatures from 300 to 800 K was used to sample over cis-trans isomerization barriers. Compared to (Nrch)5 and cyclo-octasarcosyl, the free energy of N-(2-nitro-3-hydroxyl phenyl)glycine-N-(phenyl)glycine has the most "foldable" free energy landscape, due to deep trans-amide minima dictated by N-aryl sidechains. For peptoids with (S)-N (1-phenylethyl) (Nspe) side chains, we observe a discrepancy in backbone dihedral propensities between molecular simulations and QM calculations, which may be due to force field effects or the inability to capture n --> n* interactions. For these residues, an empirical phi-angle biasing potential can "rescue" the backbone propensities seen in QM. This approach can serve as a general strategy for addressing force fields without resorting to a complete reparameterization. Overall, this study demonstrates the utility of implicit-solvent REMD simulations for efficient sampling to predict peptoid conformational landscapes, providing a potential tool for first-principles design of sequences with specific folding properties.

Markov state model reveals folding and functional dynamics in ultra-long MD trajectories

Two strategies have been recently employed to push molecular simulation to long, biologically relevant time scales: projection-based analysis of results from specialized hardware producing a small number of ultralong trajectories and the statistical interpretation of massive parallel sampling performed with Markov state models (MSMs). Here, we assess the MSM as an analysis method by constructing a Markov model from ultralong trajectories, specifically two previously reported 100 μs trajectories of the FiP35 WW domain (Shaw, D. E. Science 2010, 330, 341-346). We find that the MSM approach yields novel insights. It discovers new statistically significant folding pathways, in which either beta-hairpin of the WW domain can form first. The rates of this process approach experimental values in a direct quantitative comparison (time scales of 5.0 μs and 100 ns), within a factor of ∼2. Finally, the hub-like topology of the MSM and identification of a holo conformation predicts how WW domains may function through a conformational selection mechanism.

Atomistic folding simulations of the five-helix bundle protein λ(6−85)

Protein folding is a classic grand challenge that is relevant to numerous human diseases, such as protein misfolding diseases like Alzheimer’s disease. Solving the folding problem will ultimately require a combination of theory, simulation, and experiment, with theory and simulation providing an atomically detailed picture of both the thermodynamics and kinetics of folding and experimental tests grounding these models in reality. However, theory and simulation generally fall orders of magnitude short of biologically relevant time scales. Here we report significant progress toward closing this gap: an atomistic model of the folding of an 80-residue fragment of the λ repressor protein with explicit solvent that captures dynamics on a 10 milliseconds time scale. In addition, we provide a number of predictions that warrant further experimental investigation. For example, our model’s native state is a kinetic hub, and biexponential kinetics arises from the presence of many free-energy basins separated by barriers of different heights rather than a single low barrier along one reaction coordinate (the previously proposed incipient downhill folding scenario).

Taming the complexity of protein folding

Protein folding is an important problem in structural biology with significant medical implications, particularly for misfolding disorders like Alzheimer's disease. Solving the folding problem will ultimately require a combination of theory and experiment, with theoretical models providing a comprehensive view of folding and experiments grounding these models in reality. Here we review progress towards this goal over the past decade, with an emphasis on recent theoretical advances that are empowering chemically detailed models of folding and the new results these technologies are providing. In particular, we discuss new insights made possible by Markov state models (MSMs), including the role of non-native contacts and the hub-like character of protein folded states.

Unfolded-state dynamics and structure of protein L characterized by simulation and experiment

While several experimental techniques now exist for characterizing protein unfolded states, all-atom simulation of unfolded states has been challenging due to the long time scales and conformational sampling required. We address this problem by using a combination of accelerated calculations on graphics processor units and distributed computing to simulate tens of thousands of molecular dynamics trajectories each up to approximately 10 mus (for a total aggregate simulation time of 127 ms). We used this approach in conjunction with Trp-Cys contact quenching experiments to characterize the unfolded structure and dynamics of protein L. We employed a polymer theory method to make quantitative comparisons between high-temperature simulated and chemically denatured experimental ensembles and find that reaction-limited quenching rates calculated from simulation agree remarkably well with experiment. In both experiment and simulation, we find that unfolded-state intramolecular diffusion rates are very slow compared to highly denatured chains and that a single-residue mutation can significantly alter unfolded-state dynamics and structure. This work suggests a view of the unfolded state in which surprisingly low diffusion rates could limit folding and opens the door for all-atom molecular simulation to be a useful predictive tool for characterizing protein unfolded states along with experiments that directly measure intramolecular diffusion.

Molecular simulation of ab initio protein folding for a millisecond folder NTL9(1-39)

To date, the slowest-folding proteins folded ab initio by all-atom molecular dynamics simulations have had folding times in the range of nanoseconds to microseconds. We report simulations of several folding trajectories of NTL9(1-39), a protein which has a folding time of approximately 1.5 ms. Distributed molecular dynamics simulations in implicit solvent on GPU processors were used to generate ensembles of trajectories out to approximately 40 micros for several temperatures and starting states. At a temperature less than the melting point of the force field, we observe a small number of productive folding events, consistent with predictions from a model of parallel uncoupled two-state simulations. The posterior distribution of the folding rate predicted from the data agrees well with the experimental folding rate (approximately 640/s). Markov State Models (MSMs) built from the data show a gap in the implied time scales indicative of two-state folding and heterogeneous pathways connecting diffuse mesoscopic substates. Structural analysis of the 14 out of 2000 macrostates transited by the top 10 folding pathways reveals that native-like pairing between strands 1 and 2 only occurs for macrostates with p(fold) > 0.5, suggesting beta(12) hairpin formation may be rate-limiting. We believe that using simulation data such as these to seed adaptive resampling simulations will be a promising new method for achieving statistically converged descriptions of folding landscapes at longer time scales than ever before.

Probing the nanosecond dynamics of a designed three-stranded beta-sheet with a massively parallel molecular dynamics simulation

Recently a temperature-jump FTIR study of a designed three-stranded sheet showing a fast relaxation time of approximately 140 +/- 20 ns was published. We performed massively parallel molecular dynamics simulations in explicit solvent to probe the structural events involved in this relaxation. While our simulations produce similar relaxation rates, the structural ensemble is broad. We observe the formation of turn structure, but only very weak interaction in the strand regions, which is consistent with the lack of strong backbone-backbone NOEs in previous structural NMR studies. These results suggest that either (D)P(D)P-II folds at time scales longer than 240 ns, or that (D)P(D)P-II is not a well-defined three-stranded beta-sheet. This work also provides an opportunity to compare the performance of several popular forcefield models against one another.

Predicting peptide structures in native proteins from physical simulations of fragments

It has long been proposed that much of the information encoding how a protein folds is contained locally in the peptide chain. Here we present a large-scale simulation study designed to examine the extent to which conformations of peptide fragments in water predict native conformations in proteins. We perform replica exchange molecular dynamics (REMD) simulations of 872 8-mer, 12-mer, and 16-mer peptide fragments from 13 proteins using the AMBER 96 force field and the OBC implicit solvent model. To analyze the simulations, we compute various contact-based metrics, such as contact probability, and then apply Bayesian classifier methods to infer which metastable contacts are likely to be native vs. non-native. We find that a simple measure, the observed contact probability, is largely more predictive of a peptide's native structure in the protein than combinations of metrics or multi-body components. Our best classification model is a logistic regression model that can achieve up to 63% correct classifications for 8-mers, 71% for 12-mers, and 76% for 16-mers. We validate these results on fragments of a protein outside our training set. We conclude that local structure provides information to solve some but not all of the conformational search problem. These results help improve our understanding of folding mechanisms, and have implications for improving physics-based conformational sampling and structure prediction using all-atom molecular simulations.

Proteins must fold to unique native structures in order to perform their functions. To do this, proteins must solve a complicated conformational search problem, the details of which remain difficult to study experimentally. Predicting folding pathways and the mechanisms by which proteins fold is thus central to understanding how proteins work. One longstanding question is the extent to which proteins solve the search problem locally, by folding into sub-structures that are dictated primarily by local sequence. Here, we address this question by conducting a large-scale molecular dynamics simulation study of protein fragments in water. The simulation data was then used to optimize a statistical model that predicted native and non-native contacts. The performance of the resulting model suggests that local structuring provides some but not all of the information to solve the folding problem, and that molecular dynamics simulation of fragments can be useful for protein structure prediction and design.

The protein folding problem: when will it be solved?

The protein folding problem can be viewed as three different problems: defining the thermodynamic folding code; devising a good computational structure prediction algorithm; and answering Levinthal's question regarding the kinetic mechanism of how proteins can fold so quickly. Once regarded as a grand challenge, protein folding has seen much progress in recent years. Folding codes are now being used to successfully design proteins and non-biological foldable polymers; aided by the Critical Assessment of Techniques for Structure Prediction (CASP) competition, protein structure prediction has now become quite good. Even the once-challenging Levinthal puzzle now seems to have an answer--a protein can avoid searching irrelevant conformations and fold quickly by making local independent decisions first, followed by non-local global decisions later.

Exploring zipping and assembly as a protein folding principle

It has been proposed that proteins fold by a process called "Zipping and Assembly" (Z&A). Zipping refers to the growth of local substructures within the chain, and assembly refers to the coming together of already-formed pieces. Our interest here is in whether Z&A is a general method that can fold most of sequence space, to global minima, efficiently. Using the HP model, we can address this question by enumerating full conformation and sequence spaces. We find that Z&A reaches the global energy minimum native states, even though it searches only a very small fraction of conformational space, for most sequences in the full sequence space. We find that Z&A, a mechanism-based search, is more efficient in our tests than the replica exchange search method. Folding efficiency is increased for chains having: (a) small loop-closure steps, consistent with observations by Plaxco et al. 1998;277;985-994 that folding rates correlate with contact order, (b) neither too few nor too many nucleation sites per chain, and (c) assembly steps that do not occur too early in the folding process. We find that the efficiency increases with chain length, although our range of chain lengths is limited. We believe these insights may be useful for developing faster protein conformational search algorithms.

Information content of molecular structures

For a completely enumerated set of conformers of a macromolecule or for exhaustive lattice walks of model polymers it is straightforward to use Shannon information theory to deduce the information content of the ensemble. It is also practicable to develop numerical measures of the information content of sets of exact distance constraints applied to specific conformational ensembles. We examine the effects of experimental uncertainties by considering "noisy" constraints. The introduction of noise requires additional assumptions about noise distribution and conformational clustering protocols that make the problem of measuring information content more complex. We make use of a standard concept in communication theory, the "noise sphere," to link uncertainty in measurements to information loss. Most of our numerical results are derived from two-dimensional lattice ensembles. Expressing results in terms of information per degree of freedom removes almost all of the chain length dependence. We also explore off-lattice polyalanine chains that yield surprisingly similar results.