Explicit Characterization of the Free-Energy Landscape of a Protein in the Space of All Its Cα Carbons

Giulia Sormani, Álex Rodríguez, Alessandro LaioJournal of Chemical Theory and Computation 2019 · DOI

Abstract

By using an approach that allows computing the free energy in high-dimensional spaces together with a clustering technique capable of identifying kinetic attractors stabilized by conformational disorder, we analyze a molecular dynamics trajectory of the Villin headpiece. We compute its free-energy landscape in the space of all its C<sub>α</sub> carbons. This landscape has the shape of a 12-dimensional funnel with the free energy decreasing monotonically as a function of the native contacts. There are no significant folding barriers. The funnel can be partitioned in five regions, three mainly folded and two unfolded, which behave as Markov states. The slowest relaxation time among these states corresponds to the folding transition. The second slowest time is only twice smaller and corresponds to a transition within the unfolded state. This indicates that the unfolded part of the funnel has a nontrivial shape, which induces a sizable kinetic barrier between disordered states.

Introduction

Protein folding is traditionally conceptualized through the "folding funnel" paradigm: a protein transitions from a high-entropy, disordered state to a single, stable native structure. To visualize this, scientists typically project the massive complexity of thousands of atomic coordinates onto one or two "collective variables" (CVs), such as the fraction of native contacts ($Q$).

However, these simplifications can be a double-edged sword. While they make data digestible, they often obscure true kinetics. A barrier that appears negligible in a 1D projection might be significant in the full configuration space. This study addresses the fundamental challenge of characterizing the free-energy landscape in its full dimensionality without the bias of pre-selected coordinates.

Methodology

The researchers developed a framework combining three advanced mathematical tools to map the folding landscape of the Villin headpiece, a small, fast-folding protein:

  1. Intrinsic Dimension (ID) Estimation: They determined that the protein's configuration space effectively moves on a manifold with an intrinsic dimension of approximately 12.
  2. PAk Free-Energy Estimator: This tool calculates the free energy at every point in the simulation trajectory based on the local density of configurations within that 12D space.
  3. $\hat{k}$-Peak Clustering: This novel approach identifies "kinetic attractors"—regions where the system lingers due to conformational disorder. Unlike standard clustering that seeks energy minima, this method can define states even in the relatively flat, unfolded regions of the landscape.

The study utilized a landmark 122 $\mu$s molecular dynamics trajectory of the Villin headpiece provided by the D.E. Shaw Research group.

The 12-Dimensional Funnel

The analysis confirms that the free-energy landscape of the Villin headpiece is indeed a funnel. As shown in the data, the free energy ($F$) decreases monotonically as the protein forms more native contacts ($Q$). This supports the "downhill folding" theory, suggesting an absence of singular, massive barriers once the folding process begins.

Free energy landscape (a) and probability distribution (b) as a function of native contacts and free energy.

The probability distribution of the free energy (Figure 5b) reveals distinct peaks, suggesting the presence of metastable states along the folding pathway that correspond to specific structural configurations.

Kinetic States and the "Texture" of Disorder

By applying the $\hat{k}$-Peak clustering algorithm, the researchers identified five distinct kinetic regions: three "folded" states (including the native state and two near-native intermediates) and two "unfolded" states.

Clustering results showing (d) free energy vs Q for three clusters and (e) the five identified kinetic states in the k-hat vs Q space.

A key insight is that the unfolded state is not a uniform "blob." Instead, it is partitioned into two distinct Markov states. The transition between these disordered states is the second-slowest process in the simulation, occurring at a rate only twice as fast as the folding transition itself. This indicates a significant kinetic bottleneck or "texture" within the unfolded part of the funnel, likely caused by a sharp drop in available configurations (entropy) at an intermediate point ($Q \approx 0.75$).

Conclusion

This unbiased mapping of the free-energy landscape provides a more accurate picture of protein kinetics than traditional low-dimensional projections. The calculated folding time ($t_f \approx 2.26\ \mu$s) aligns well with experimental data, proving that the high-dimensional approach captures the essential physics. While scaling this method to larger, multi-domain proteins remains a computational challenge, it offers a powerful tool for investigating misfolding pathways and toxic aggregates in more complex systems, such as those involved in Alzheimer's disease.

Summary updated

Abstract

By using an approach that allows computing the free energy in high-dimensional spaces together with a clustering technique capable of identifying kinetic attractors stabilized by conformational disorder, we analyze a molecular dynamics trajectory of the Villin headpiece from Lindorff-Larsen, K.; et al. How fast-folding proteins fold. <i>Science</i> <b>2011,</b> <i>334,</i> 517-520. We compute its free-energy landscape in the space of all its C<sub>α</sub> carbons. This landscape has the shape of a 12-dimensional funnel with the free energy decreasing monotonically as a function of the native contacts. There are no significant folding barriers. The funnel can be partitioned in five regions, three mainly folded and two unfolded, which behave as Markov states. The slowest relaxation time among these states corresponds to the folding transition. The second slowest time is only twice smaller and corresponds to a transition within the unfolded state. This indicates that the unfolded part of the funnel has a nontrivial shape, which induces a sizable kinetic barrier between disordered states.
BibTeX
@article{sormani2019explicit,
  author = {Giulia Sormani and Álex Rodríguez and Alessandro Laio},
  title = {{Explicit Characterization of the Free-Energy Landscape of a Protein in the Space of All Its C<sub>α</sub> Carbons}},
  journal = {Journal of Chemical Theory and Computation},
  year = {2019},
  doi = {10.1021/acs.jctc.9b00800},
}
Figure