BPS2026 – Recovering equilibrium statistics from vitrified cryo-EM samples

Cryogenic electron microscopy (cryo-EM) relies on the rapid cooling of biological samples to form water glasses, preserving molecular structures for high-resolution imaging. While plunge cooling of grids is rapid in a macroscopic sense, it is slow relative to most macromolecular motions. Because ice crystal formation is a common issue on grids, the cooling rate is close to the minimum needed to form a water glass for successful plunges. Experimental measurements estimate it to be approximately 6.4 million Kelvin per second, implying that cooling occurs across tens of microseconds. This non-equilibrium perturbation of the conformational ensemble has significant implications for interpreting cryo-EM data, particularly in time-resolved studies, structural heterogeneity analysis, and the direct recovery of populations from cryo-EM. Here, we discuss simulations of plunge-vitrification of conformational ensembles. We use simple potentials, as well as protein simulations based on the Trp-Cage, to make explicit comparisons between thermalized and vitrified ensembles. In our Trp-Cage simulations, we use temperature replica exchange to obtain equilibrium populations at 277 K, a common pre-plunge temperature, and 230K, below which we find little protein conformational state change. We built a Markov State Model of the 277 K ensemble using equilibrium simulations seeded from our temperature replica exchange to quantify equilibrium behavior. To simulate cooling, we ran thousands of temperature ramp simulations at different cooling rates seeded in the same way. We used the MSM to map the relationship between kinetics and thermodynamics at 277K and the perturbations induced by vitrification. Finally, we derive thermodynamic bounds on the possible distributions resulting from this cooling process enabling us to recover the original equilibrium ensemble from the quenched ensembles alone. We find that recovering the probabilities for the 277 K MSM are feasible providing a strategy for estimating state probabilities from heterogeneous Cryo-EM data.