Resolving Barrier Crossing in Protein Folding
Proteins are the active molecules of life. To carry out their functions, they adopt specific structures, or “folds.” Biophysicists have long been fascinated by the “protein-folding problem”: How does the sequence of amino-acid building blocks encode the protein’s ultimate fold, and how can folding occur so quickly and reliably? The folding process can be understood as a diffusive random walk through the large space of possible configurations, culminating in the crossing of an energy barrier to reach the folded state. The time spent exploring unfolded configurations can span many orders of magnitude and has been measured with various experimental techniques. By contrast, the comparatively short time to ultimately cross the energy barrier—known as the transition-path time—had never been measured in a naturally occurring protein under biologically relevant conditions. Now Hoi Sung Chung and collaborators at the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) in Maryland have used high-resolution single-molecule methods to measure transition-path times for eight small proteins, offering critical insight into how proteins fold [1].
Major progress in predicting the final folded structure of a protein has come from AlphaFold, a machine-learning approach that leverages experimentally determined protein structures and evolutionary constraints inferred from protein-sequence variations across species [2]. Yet fundamental questions remain about how an initially unfolded chain of amino acids coordinates itself to reach that structure.
A productive framework for understanding this problem has been to treat protein folding as diffusion on a free-energy landscape. In that picture, the vertical axis represents the protein’s free energy for a particular configuration, while the horizontal axes correspond to degrees of conformational freedom [3]. Because a typical protein comprises hundreds of amino acids, this landscape is enormously high dimensional. In practice, however, folding can be empirically modeled using an effective one-dimensional landscape, in which a single “reaction coordinate” tracks progress toward the folded state. Conceptually, that reduction casts folding as a thermally activated barrier crossing: The folded and unfolded states are minima of the energy landscape separated by a higher-energy transition state (Fig. 1, top). The transition state is high in energy because reaching it implies the formation of specific interactions in the protein at the cost of substantial conformational freedom. Crossing the barrier requires waiting for a sufficiently large random thermal fluctuation—which sets the overall folding time. Once this energy is supplied, the protein crosses the barrier, settling into its folded structure on comparatively short timescales.
Measuring the transition-path time—the brief interval during which the protein moves across the barrier—is experimentally challenging. The stochastic nature of the process means that different proteins cannot be synchronized to go over the barrier at the same time. This limitation precludes the possibility of experiments that average over an ensemble of molecules, motivating a single-molecule approach. However, resolving the transition requires submicrosecond temporal resolution, which is beyond the reach of most single-molecule techniques used to study protein folding.
In 2012, an NIDDK team led by William Eaton developed a promising method based on single-molecule Förster resonance energy transfer (FRET). In FRET, a protein labeled with two small-molecule dyes fluoresces primarily at one frequency in its expanded, unfolded configuration and at a different frequency in its compact, folded form [4, 5]. In those measurements, however, the transition-path time was comparable to the average interval between detected fluorescence photons, leading to a temporal resolution insufficient for natural proteins in a biologically relevant solution. (The resolution is constrained by how quickly one can accumulate a sufficient number of photons to determine the protein’s state through a maximum-likelihood statistical analysis [6].) This method could only measure the transition-path time of sufficiently slow proteins—such as engineered proteins and natural proteins slowed down in a viscous glycerol solution. As a result, molecular-dynamics simulations remained, despite their well-known limitations, the only way to access transition-path times for many proteins [7].
Chung and collaborators modified the single-molecule FRET setup to increase the photon detection rate. Their scheme involves nanopatterned substrates with surface features that act as waveguides (Fig. 1, bottom), facilitating the downward propagation of light toward the detector. As a result, the detected photon flux increases by a factor of 5–6 for proteins that have diffused into the waveguide, reducing the interval between detected photons. This reduction boosts the temporal resolution, enabling a more precise identification of the transition path.
The team applied this technique to eight proteins whose overall folding times range from 0.1 to 1000 ms. Despite that 4-order-of-magnitude spread, those proteins’ transition-path times clustered within a much narrower range of 0.7–4 µs. This result accords with one-dimensional energy-landscape theory, which predicts that the overall folding time depends exponentially on the barrier height, whereas the transition-path time varies only logarithmically with it. Interestingly, Chung and colleagues observed that the proteins with the fastest overall folding times had the slowest transition-path times. That inverse relationship is consistent with the difference between broad, low-energy barriers—favoring rapid folding but a slow passage across the barrier—and narrow, sharp barriers—prolonging the folding time but permitting faster crossings. The short transition-path times of natural proteins, compared with the 13-µs time observed for the aforementioned engineered protein, suggest that biological evolution may have optimized the folding energy landscape for fast transition-state crossing.
An analysis of how transition-path time scales with protein size revealed that larger proteins tend to cross the barrier more quickly. This scaling implies that larger proteins form native contacts—chemical interactions between amino acids present in the folded structure—faster than smaller proteins. This counterintuitive behavior is not expected from a random walk through conformational space. It instead suggests a more coordinated “zippering” mechanism in which interactions between distant segments of the amino-acid chain cooperatively promote the formation of native contacts.
Additional insight into folding mechanisms could come from measuring additional features of the transition, such as the distribution of transition-path times across individual molecules or the detailed dynamics of the protein during barrier crossing. Such characterization has been accomplished in DNA, whose longer transition-path times make it accessible to single-molecule force-spectroscopy methods, such as optical tweezers [8]. Making similar measurements in proteins will require further advances in single-molecule techniques. Transition-path measurements in membrane proteins would be especially informative, as all transition-path measurements to date have instead focused on soluble proteins. Because membrane proteins live within the lipid bilayers of cellular membranes, their folding is driven by different forces and may have significantly longer transition-path times than soluble proteins that fold in water [9, 10]. A fuller understanding of folding dynamics could enable more precise control of protein behavior—which could inform drug design or protein-engineering strategies that leverage evolutionarily optimized energy landscapes.
References
- C.-J. Feng et al., “Cooperative native contact formation facilitates free energy barrier crossing in protein folding,” Phys. Rev. Lett. 136, 108401 (2026).
- J. Jumper et al., “Highly accurate protein structure prediction with AlphaFold,” Nature 596, 583 (2021).
- J. N. Onuchic et al., “Theory of protein folding: The energy landscape perspective,” Annu. Rev. Phys. Chem. 48, 545 (1997).
- H. S. Chung et al., “Experimental determination of upper bound for transition path times in protein folding from single-molecule photon-by-photon trajectories,” Proc. Natl. Acad. Sci. U.S.A. 106, 11837 (2009).
- H. S. Chung et al., “Single-molecule fluorescence experiments determine protein folding transition path times,” Science 335, 981 (2012).
- I. V. Gopich and A. Szabo, “Decoding the pattern of photon colors in single-molecule FRET,” J. Phys. Chem. B 113, 10965 (2009).
- K. Lindorff-Larsen et al., “How fast-folding proteins fold,” Science 334, 517 (2011).
- N. Q. Hoffer and M. T. Woodside, “Probing microscopic conformational dynamics in folding reactions by measuring transition paths,” Curr. Opin. Chem. Biol. 53, 68 (2019).
- S. Kim et al., “Robust membrane protein tweezers reveal the folding speed limit of helical membrane proteins,” eLife 12, e85882 (2023) .
- D. R. Jacobson, “Single molecule force spectroscopy to probe intermediates and energetics of membrane protein folding,” Chem. Rev. 126, 2550 (2026).