Multi-Dimensional Theory of Protein Folding

Theory of multi-dimensional representation of free energy surface of protein folding is developed by adopting structural order parameters of multiple regions in protein as multiple coordinates. Various scenarios of folding are classified in terms of cooperativity within individual regions and interactions among multiple regions and thus obtained classification is used to analyze the folding process of several example proteins. Ribosomal protein S6, src-SH3 domain, CheY, barnase, and BBL domain are analyzed with the two-dimensional representation by using a structure-based Hamiltonian model. Extension to the higher dimensional representation leads to the finer description of the folding process. Barnase, NtrC, and an ankyrin repeat protein are examined with the three-dimensional representation. The multi-dimensional representation allows us to directly address questions on folding pathways, intermediates, and transition states.

💡 Research Summary

The paper introduces a novel theoretical framework for describing protein folding by representing the free‑energy landscape in multiple dimensions. Traditional one‑dimensional reaction coordinates, such as the total fraction of native contacts (Q), often fail to capture the complexity of multi‑domain proteins, intermediate states, and heterogeneous transition states. To overcome this limitation, the authors partition a protein into N structural regions (domains, secondary‑structure elements, or repeat units) and define an independent order parameter Q_i for each region, quantifying the degree of native contact formation within that region. The multidimensional free‑energy surface F(Q_1,…,Q_N) is then expressed as a sum of intrinsic contributions f_i(Q_i) that reflect the cooperativity of each region and pairwise interaction terms g_{ij}(Q_i,Q_j) that encode coupling between regions. This decomposition allows a clear classification of folding scenarios based on intra‑region cooperativity and inter‑region coupling.

The underlying physical model is a structure‑based Hamiltonian (Go‑type potential) in which only native contacts contribute attractive energy. This model is computationally tractable and reproduces the essential physics of folding funnels. Monte Carlo or molecular dynamics simulations are performed to generate trajectories of the Q_i variables. By accumulating a high‑dimensional histogram of the sampled configurations and applying kernel density estimation, the authors reconstruct a smooth free‑energy surface in the chosen dimensionality. Minimum‑energy paths (MEPs) are then extracted using the nudged elastic band or string method, providing explicit transition‑state ensembles.

The authors first apply a two‑dimensional representation (N=2) to five well‑studied proteins: ribosomal protein S6, src‑SH3, CheY, barnase, and the BBL domain. In S6 and src‑SH3, the two regions display strong positive coupling (g_12 > 0), leading to a cooperative “two‑step” folding pathway where both regions form native contacts almost simultaneously. The transition state in these cases is characterized by partial formation of both regions and a relatively narrow free‑energy barrier. CheY, by contrast, shows weak or negative coupling, resulting in a sequential, non‑cooperative pathway where one region folds first and the other follows. Barnase and BBL exhibit more complex behavior with detectable intermediates; barnase, for example, follows a hierarchical scheme where two regions fold cooperatively before the third region completes the process.

To demonstrate the added resolution of higher dimensionality, the authors extend the analysis to three dimensions (N=3) for barnase, the response regulator NtrC, and an ankyrin repeat protein. In these cases, the free‑energy landscape reveals multiple parallel pathways, several distinct intermediate basins, and transition‑state ensembles that involve partial folding of different subsets of regions. The ankyrin repeat protein, composed of tandem repeats, shows asymmetric coupling among repeats, producing a “sequential propagation” folding mechanism where each repeat folds after its predecessor, yet occasional long‑range interactions generate shortcuts. NtrC displays strong coupling between its receiver domain and output domain, resulting in a transition state that simultaneously contains partially folded structures of both domains, consistent with experimental φ‑value analyses.

A key advantage of the multidimensional approach is its ability to directly compare simulated transition‑state structures with experimental data such as φ‑values, Ψ‑analysis, and hydrogen‑exchange protection factors. Because each Q_i can be mapped onto specific residues or secondary‑structure elements, the model quantifies how perturbations (mutations, ligand binding, changes in solvent conditions) alter the cooperativity within a region or the strength of inter‑region coupling, thereby shifting the location and height of free‑energy barriers. This level of detail is inaccessible to one‑dimensional models, which often collapse distinct pathways into a single effective barrier.

The paper concludes that representing the folding free‑energy surface in multiple dimensions provides a powerful, physically grounded framework for dissecting the interplay of intra‑domain cooperativity and inter‑domain interactions. It enables a systematic classification of folding mechanisms (cooperative, hierarchical, sequential, or mixed) and offers quantitative predictions for how specific structural features govern the kinetics and thermodynamics of folding. The authors suggest that this methodology can be scaled to larger, more complex systems—such as multi‑domain enzymes, protein‑protein complexes, and intrinsically disordered proteins that acquire structure upon binding—thereby opening new avenues for rational protein design, interpretation of disease‑related mutations, and the development of folding‑based therapeutics.