The hepatitis B virus capsid represents a promising therapeutic target. Experiments suggest the capsid must be flexible to function; however, capsid structure and dynamics have not been thoroughly characterized in the absence of icosahedral symmetry constraints. Here, all-atom molecular dynamics simulations are leveraged to investigate the capsid without symmetry bias, enabling study of capsid flexibility and its implications for biological function and cryo-EM resolution limits. Simulation results confirm flexibility and reveal a propensity for asymmetric distortion. The capsid’s influence on ionic species suggests a mechanism for modulating the display of cellular signals and implicates the capsid’s triangular pores as the location of signal exposure. A theoretical image reconstruction performed using simulated conformations indicates how capsid flexibility may limit the resolution of cryo-EM. Overall, the present work provides functional insight beyond what is accessible to experimental methods and raises important considerations regarding asymmetry in structural studies of icosahedral virus capsids.

Defining mechanisms of direct-acting antivirals facilitates drug development and our understanding of virus function. Heteroaryldihydropyrimidines (HAPs) inappropriately activate assembly of hepatitis B virus (HBV) core protein (Cp), suppressing formation of virions. We examined a fluorophore-labeled HAP, HAP-TAMRA. HAP-TAMRA induced Cp assembly and also bound pre-assembled capsids. Kinetic and spectroscopic studies imply that HAP-binding sites are usually not available but are bound cooperatively. Using cryo-EM, we observed that HAP-TAMRA asymmetrically deformed capsids, creating a heterogeneous array of sharp angles, flat regions, and outright breaks. To achieve high resolution reconstruction (<4 Å), we introduced a disulfide crosslink that rescued particle symmetry. We deduced that HAP-TAMRA caused quasi-sixfold vertices to become flatter and fivefold more angular. This transition led to asymmetric faceting. That a disordered crosslink could rescue symmetry implies that capsids have tensegrity properties. Capsid distortion and disruption is a new mechanism by which molecules like the HAPs can block HBV infection.

For many viruses, capsids (biological nanoparticles) assemble to protect genetic material and dissociate to release their cargo. To understand these contradictory properties, we analyzed capsid assembly for Hepatitis B virus; an endemic pathogen with an icosahedral, 120-homodimer capsid. We used solution X-ray scattering to examine trapped and equilibrated assembly reactions. To fit experimental results, we generated a library of unique intermediates, selected by umbrella sampling of Monte Carlo simulations. The number of possible capsid intermediates is immense, ~ 10 30 , yet assembly reactions are rapid and completed with high fidelity. If the huge number of possible intermediates were actually present, maximum entropy analysis shows that assembly reactions would be blocked by an entropic barrier, resulting in incomplete nanoparticles. When an energetic term was applied to select the stable species that dominated the reaction mixture, we found that only a few hundred intermediates, mapping out a narrow path through the immense reaction landscape. This is a solution to a viral application of the Levinthal paradox. With the correct energetic term, the match between predicted intermediates and scattering data was striking. The grand canonical free energy landscape for assembly, calibrated by our experimental results, supports a detailed analysis of this complex reaction. There is a narrow range of energies that supports on-path assembly. If association energy is too weak or too strong progressively more intermediates will be entropically blocked, spilling into paths leading to dissociation or trapped incomplete nanoparticles, respectively. These results are relevant to many viruses, provide a basis for simplifying assembly models and identifying new targets for antiviral intervention. They provide a basis for understanding and designing biological and abiological self-assembly reactions.

There are ∼ 10 30 possible intermediates on the assembly path from hepatitis B capsid protein dimers to the 120-dimer capsid. If every intermediate was tested, assembly would often get stuck in an entropic trap and essentially every capsid would follow a unique assembly path. Yet, capsids assemble rapidly with minimal trapped intermediates, a realization of the Levinthal paradox. To understand the fundamental mechanisms of capsid assembly it is critical to resolve the early stages of the reaction. We have used Time-Resolved Small Angle X-ray Scattering, which is sensitive to solute size and shape and has millisecond temporal resolution. Scattering curves were fit to a thermodynamically curated library of assembly intermediates, using the principle of maximum entropy. Maximum entropy also provides a physical rationale for the selection of species. We found that the capsid assembly pathway was exquisitely sensitive to 1 initial assembly conditions. With the mildest conditions tested, the reaction appeared two-state from dimer to 120-dimer capsid with some dimers-of-dimers and trimers-ofdimers. In slightly more aggressive conditions, we observed transient accumulation of a decamer-of-dimers and appearance of 90-dimer capsids. In conditions where there is measurable kinetic trapping, we found that a highly diverse early intermediates accumulated within a fraction of a second and propagated into long-lived kinetically trapped states (≥ 90-mer). In all cases, intermediates between 35 and 90 subunits did not accumulate. These results are consistent with the presence of low barrier paths that connect early and late intermediates and direct the ultimate assembly path to late intermediates where assembly can be paused.

Machine-learning prediction algorithms such as AlphaFold and RoseTTAFold can create remarkably accurate protein models, but these models usually have some regions that are predicted with low confidence or poor accuracy. We hypothesized that by implicitly including new experimental information such as a density map, a greater portion of a model could be predicted accurately, and that this might synergistically improve parts of the model that were not fully addressed by either machine learning or experiment alone. An iterative procedure was developed in which AlphaFold models are automatically rebuilt on the basis of experimental density maps and the rebuilt models are used as templates in new AlphaFold predictions. We show that including experimental information improves prediction beyond the improvement obtained with simple rebuilding guided by the experimental data. This procedure for AlphaFold modeling with density has been incorporated into an automated procedure for interpretation of crystallographic and electron cryo-microscopy maps.