Structure-based simulations of the full Flock House virus capsid reveal pathways and energetics of an infection-critical peptide externalization event

Abstract

Viral capsids are metastable assemblies whose large-scale dynamics enable critical steps in infection, including the exposure of internal protein segments that interact with host membranes. Resolving the mechanisms underlying these transitions remains challenging because they involve rare, cooperative motions distributed across entire viral particles. Here, we develop a full-capsid structure-based molecular dynamics framework to investigate infection-relevant dynamics in the T=3 Flock House virus capsid, a model non-enveloped RNA virus. This native contact-centric coarse-grained representation enables systematic exploration of peptide externalization pathways while preserving the geometric and topological constraints of the intact capsid. To capture infection-relevant environmental changes, we introduce a low-pH capsid model derived from collective structural modes observed in full-capsid atomistic simulations. This approach provides a physically motivated representation of acid-induced capsid deformation while filtering high-frequency fluctuations. Simulations of γ peptide externalization reveal distinct symmetry-dependent escape mechanisms, with exit through the fivefold axis of symmetry being the most favored. The acidification-associated low-pH structural perturbation modestly reshapes the underlying energy landscape, reducing the energetic cost of peptide release, with the strongest favorability emerging when combined with internal mechanical stress that mimics genome confinement. These findings demonstrate how capsid geometry, symmetry, and environmental conditions couple to regulate peptide release in an RNA virus. More broadly, this work highlights how full-capsid structure-based models can provide an efficient framework for probing functional energy landscapes in large viral assemblies.

Introduction

Viral capsids are dynamic, metastable assemblies whose collective motions are tightly coupled to biological function.(1) Rather than serving as passive containers, capsids undergo coordinated structural rearrangements that mediate key events in the viral life cycle, including attachment, maturation, entry, and genome release.(2) These motions emerge from global constraints imposed by capsid geometry, symmetry, and connectivity, and often involve transitions that span large length scales and long timescales. Understanding how capsid-scale dynamics give rise to functionally relevant transitions remains a central challenge in viral biophysics.

A common manifestation of functional capsid dynamics is the externalization of internal capsid protein components. Many non-enveloped icosahedral viruses sequester protein components on the capsid interior and subsequently externalize them under specific conditions. In several virus families, these externalized protein components mediate membrane disruption and facilitate endosomal escape.(3–5) A well-established model system for studying such processes in non-enveloped icosahedral viruses is Flock House Virus (FHV), a T=3 nodavirus containing a positive-sense RNA genome (Fig. 1A-B).(6) FHV has been the subject of extensive investigation, with a wealth of structural, biophysical, biochemical, and computational studies now available.(7–21)

Figure 1: FHV Capsid and model preparation.

(A) FHV capsid exterior view: β_A (blue), β_B, (red), and β_C (green). (B) FHV capsid interior view : β proteins are colored in grey, while γ peptides are colored blue (γ_A), red (γ_B) and green (γ_C). (C) Distribution of the number of native contacts across 60 copies of γ peptides. (D) Sum of square RMSF errors as a function of reduced temperatures (points) and quadratic least-squares fits (dashed lines). Optimal temperature T = 0.566 shown with a solid vertical line.

During maturation, the FHV capsid undergoes an autoproteolytic cleavage in which the C-terminal 44 residues of capsid precursor protein α are cleaved, producing major capsid protein β and the γ peptide.(22) Formation of the γ peptides is essential for infectivity,(7) and multiple lines of evidence indicate that these peptides are externalized from the capsid under acidic conditions consistent with the endosomal environment.(14) Once externalized, the γ peptides interact with host membranes and are thought to play a direct role in membrane permeabilization, thereby enabling genome release into the cytoplasm. Despite this central role in infection, the molecular pathways by which γ peptides externalize from an intact capsid, the effect of large-scale perturbations on the process, and the role of distinct symmetry-related environments remain incompletely understood.

All-atom molecular dynamics (AA MD) simulations have provided unprecedented structural and dynamical detail for viral capsids, enabling atomistic characterization of stability, asymmetry, and local conformational fluctuations in intact particles.(23) Large-scale simulations of complete viral capsids have revealed particle-level heterogeneity, symmetry breaking and localized conformational changes in several systems including HIV-1(24), hepatitis B(25), phage HK97(26) and FHV(21) (among others), extending out to microsecond timescales in some studies. These studies demonstrate the power of AA MD to capture detailed molecular interactions within fully assembled virions. At the same time, the computational cost of atomistic representations inherently limits accessible timescales and the extent of pathway exploration. Functional transitions in capsids often involve rare, cooperative rearrangements distributed across multiple symmetry-related environments, making systematic sampling of alternative pathways and quantitative characterization of free-energy landscapes challenging within an unbiased atomistic framework. As a result, while AA MD excels at resolving local interactions and short-timescale responses, complementary approaches are needed to efficiently interrogate capsid-scale functional dynamics that emerge from global geometric and topological constraints.

Structure-based models (SBMs), also known as Gō-like models, construct the molecular topology based on contacts observed in an experimentally determined structure. In their simplest form, interactions corresponding to native contacts are energetically stabilized while non-native interactions are modeled with repulsive excluded-volume terms. Consequently, the native structure defines the global free-energy minimum and the resulting energy landscape is minimally frustrated, with dynamics governed primarily by the topology of the native contact network. SBMs have been extensively applied in protein folding studies, where their simplified energetics enable systematic exploration of pathways connecting unfolded and native states while preserving the underlying structural framework. (27)

The application of SBMs to large macromolecular assemblies has been comparatively limited, though notable studies have examined collective dynamics in systems such as the ribosome(28) and other multi-subunit complexes.(29) For viral capsids in particular, SBM-based studies are also quite limited but have been employed to investigate maturation pathways(30, 31) and mechanical deformation(32, 33), demonstrating that large-scale symmetry-coupled motions can be captured within a simplified SBM model. In the context of nanoindentation, structure-based simulations have shown reasonable quantitative agreement with experimentally measured mechanical responses, supporting the ability of these models to describe perturbations about a well-defined native state.(33)These studies suggest that when the dominant physics arises from progressive deformation and disruption of native contacts — rather than formation of highly non-native interactions — SBMs can provide meaningful energetic and mechanistic insight. However, despite this success, the use of SBMs to interrogate infection-relevant conformational transitions in intact capsids remains relatively unexplored. Processes such as peptide externalization originate from the assembled capsid and involve selective weakening of native contacts within a globally constrained architecture. Thus, native-centric models may provide a particularly useful approach for interrogating their underlying energetics and pathways.

FHV provides an ideal system for investigating these questions within a full-capsid, structure-based model. The capsid structure is well characterized, the functional importance of peptide externalization is firmly established, and prior atomistic studies provide valuable points of comparison. Moreover, the T=3 symmetry of the FHV capsid gives rise to three distinct peptide environments associated with the A, B, and C subunits, enabling direct examination of how peptide positioning, local geometry, and symmetry-related constraints bias externalization behavior within a single viral particle.

In this work, we use full-capsid coarse-grained SBM molecular dynamics simulations to probe infection-relevant capsid dynamics in FHV, using peptide externalization as a model process. We model peptide externalization from distinct subunit environments with steered MD simulations, enabling identification of dominant pathways and quantification of the energies required for release. These simulations define reaction coordinates for umbrella sampling calculations, from which free-energy profiles are constructed and compared across peptide types and externalization routes.

To further explore the influence of physical constraints present in real viral particles, we introduce a custom force to mimic the effects of genome confinement within the capsid interior. Internal pressure is a common feature of viral capsids and is expected to bias both peptide mobility and externalization pathways. By comparing free-energy landscapes obtained with and without this pressure-like term, as well as in modified models that destabilize peptide initial geometry, we examine how capsid confinement and peptide positioning reshape the energetic barriers governing peptide release. We also extend the SBM framework to develop a low-pH model of the FHV capsid based upon normal mode analysis of reference all-atom full capsid simulations.(21) We observe a reduced energetic cost of externalization in the low-pH model, observing an overall energetic favorability of externalization combined with the pressure-like term which was not captured in a partial capsid AA-MD study.(20)

Together, these results demonstrate that full-capsid structure-based models provide a flexible and computationally efficient approach for interrogating infection-relevant dynamics in large viral assemblies. While such models are not intended to reproduce atomistic kinetics or non-native interactions in quantitative detail, they capture the essential geometric and topological features that govern the dynamics at the capsid scale. More broadly, this work highlights the value of probing native-state energy landscapes to gain mechanistic insight into rare, functionally important transitions that remain challenging to access using higher-resolution approaches alone.

Methods

Structure-Based Model Construction and Calibration

The full-capsid Cα SBM of the FHV capsid was prepared using SMOG2 with default settings.(34) Input coordinates were taken from the crystal structure PDB ID: 4FTB and processed to fit SMOG2 formatting. Langevin dynamics simulations were performed in OpenMM 8.2.0(35) with default parameters from OpenSMOG(34): a time step of 0.5 fs, collision rate 1.0 ps⁻¹, and cutoff radius of 1.1 nm. Trajectory frames were output every 1,000 integration steps. During model construction, the number of native contacts assigned to individual γ peptides varied among symmetry-related copies (Fig. 1C). In an idealized empty icosahedral capsid, all copies of a given peptide type should be symmetry-equivalent; therefore, we do not interpret this copy-to-copy variation as evidence of intrinsic physical asymmetry. Rather, this variation arises during construction of the Cα SBM contact map in SMOG2, where contacts are assigned from the input structure through the shadow-contact/contact-mapping procedure rather than by enforcing strict icosahedral symmetry. To reduce sensitivity to any individual contact assignment, all 60 copies of each γ-peptide were externalized to analyze ensemble-based features. Full-capsid AA-MD simulations were performed by our lab previously(21) and data from that study was used as reference data for developing the SBM. The simulation temperature of the SBM was optimized to match the AA-MD results. In both models, the root mean square fluctuation (RMSF) was calculated for each residue and averaged across symmetry-related chains. For each component of the asymmetric unit (β_A, β_B, β_C, γ_A, γ_B, γ_C), the sum of square residuals between the AA and CG models were computed for each residue at a variety of temperatures (Fig. 1D). A quadratic least-squares fit was applied to estimate the temperature which minimized the error for the β chains, leading to a reduced temperature of 0.566, in the normal range for proteins of 0.5–0.6 in SMOG2. We omitted the fluctuations of the γ peptides in temperature selection as the objective of the model was to capture the structure and dynamics of the major capsid proteins.

To ensure proper representation after detachment from the β chains, the γ peptides of the SBM were independently simulated. While inter-subunit interactions already maintain the peptide geometry in the native position, the rigidity of the alpha helix is required to properly model pore translocation. In these trajectories, these residues displayed an unphysical degree of flexibility due to a low density of tertiary native contacts in the rod-shaped peptide.

To provide a reference for optimization, a 350 ns AA MD simulation of the γ_A peptide in explicit solvent was performed using GROMACS version 2025.3(36) with the CHARMM36m force field.(37) γ_A was chosen due to being the largest resolved γ peptide from the crystal structure. TIP3P water was used with a physiological ionic strength of 150 mM NaCl. The production simulation was performed in the NPT ensemble at 300 K and 1 atm with the V-rescale thermostat(38) and Parrinello-Rahman barostat(39) with isotropic pressure coupling. Periodic boundaries were applied in the x-, y-, and z- dimensions. Electrostatic and Lennard-Jones interactions were cut off at 1.2 nm and long-ranged electrostatics were calculated using Particle Mesh Ewald(40) a Fourier spacing of 0.16 nm. Bonds to hydrogen atoms were constrained using the LINCS algorithm.

A corresponding Cα SBM was prepared for an isolated γ_A peptide through the same process as the full capsid using OpenSMOG/SMOG2.(34) The distributions of torsion angles within the alpha-helical region was compared between models and found to be approximately normal in both cases, albeit with wider variances in the SBM. To match the AA distributions, the force constants for the CG torsion potentials were rescaled by the ratio of the sample variances (s² _AA / s²_CG). When tested, the new CG potentials reproduced the AA distributions. Since each γ peptide shares the same sequence, these new force constants were applied to all peptides (γ_A, γ_B, γ_C).

Low-pH SBM

For FHV and many other viruses, the capsid response to acidic conditions is a critical determinant of infection-related structural changes. While the neutral-pH capsid structure is experimentally resolved and can be used directly to construct a SBM in SMOG2, no corresponding reference structure exists for relevant low-pH intermediate states along the γ peptide externalization pathway. To approximate a low-pH capsid conformation suitable for coarse-grained structure-based modeling, we performed normal mode displacements to generate a low-pH structure. Low-frequency normal modes (200) were computed using the automated workflow implemented in pyCapsid.(41) These modes capture collective, symmetry-coupled motions that are consistent with large-scale capsid deformations.(30, 42, 43) Full-capsid AA-MD trajectories generated under low-pH conditions(20) were then analyzed as perturbations of the crystal structure projected onto the low-frequency normal mode basis. For each selected frame of the atomistic trajectory, mode amplitudes were determined via least-squares linear decomposition onto the low-frequency modes.

Rather than selecting a single atomistic snapshot, which would include high-frequency stochastic fluctuations, we reconstructed the low-pH conformation using only the dominant low-frequency normal modes. This procedure filters out short-timescale thermal noise and yields a smoothly deformed capsid structure that reflects collective, symmetry-coupled displacements. The selected low-frequency modes account for over 80% of the structural difference between the final frame of the AA trajectory and the crystal structure, suggesting that the remaining deviation primarily reflects high-frequency fluctuations filtered out in the reconstruction. As a result, the low-pH SBM represents an averaged perturbation of the native state driven by largescale modes, rather than a specific instantaneous configuration.

The final mode amplitudes obtained from the low-pH atomistic simulation were applied to the Cα coordinates of the CG neutral-pH SBM model to generate the low-pH conformation. Native contacts defined in the neutral-pH model were retained, but their equilibrium distances were updated to reflect the displaced geometry; contacts whose distances exceeded the defined cutoff were removed. No new contacts were introduced in the low-pH model.

Enhanced Sampling

Steered MD (SMD) simulations were implemented in OpenMM using a sliding harmonic potential with a force constant k = 240 kcal mol⁻¹ nm^-1. The reaction coordinate was defined by the distance between the centroid of all β chains (i.e. the centroid of the capsid) and the centroid of the first eight N-terminal residues of the selected γ peptide. This selection was applied to promote headfirst externalization, since (i) the >20 unresolved residues complicate tail-first mechanisms; (ii) observed fragments from transient surface exposure in limited proteolysis mass spectrometry are concentrated at the head region. (44) SMD simulations were run for 5×10⁶ integration steps with a pull velocity of 1.25 nm per 10⁶ timesteps. This pull velocity was selected was selected to produce steady externalization while limiting excessive orthogonal motion; the resulting SMD trajectories were used for pathway identification and umbrella-window initialization rather than kinetic interpretation. To reflect the asymmetry of peptides in this model (Fig. 1C), peptides of each type were treated as an ensemble. For each γ_A, γ_B, γ_C, the externalization of all 60 icosahedral copies was simulated with three replicates each (n_total = 540). The work of externalization was calculated by trapezoidal integration of the applied SMD force across the reaction coordinate.

Umbrella sampling was implemented in OpenMM with a static harmonic potential. Umbrella windows were selected from the SMD trajectories with the lowest work value for each γ_A, γ_B, and γ_C. Due to a sharp increase in free energy upon the breakage of initial contacts, umbrella windows were initialized in two regimes, with 50 evenly spaced windows occupying the first 1.0 nm of the reaction coordinate to capture this transition and 75 windows spaced 0.05 nm apart beyond 1.0 nm for the remainder of the profile. Free energy profiles were calculated using the WHAM algorithm.(45) Temperature for the analysis was set as the simulation temperature converted from reduced units, corresponding to 68.08 K. Of note the temperature setting of 68.08 K does not reflect a physical temperature, but is a numerical value to achieve a temperature of 0.566 k_BT. To resolve difficulties from the bipartite window spacing and number of bins, the free energy curves were calculated independently and stitched together at a position of stable overlap.

In the indicated simulations, the internal pressure of the capsid resulting from genome confinement was modeled with a constant outward force in the outward radial direction of magnitude k. In these simulations, force was applied to the Cα beads of all γ peptides. The applied radial force was not intended to reproduce a true hydrostatic pressure acting uniformly on the virion interior. Rather, it was designed as an effective perturbation to isolate how confinement-derived radial bias influences γ-peptide mobility and externalization pathways within the intact capsid geometry. Applying equivalent forces to the entire capsid shell would primarily drive global deformation of the capsid architecture rather than selectively perturb peptide externalization, and pressure-related effects are already expected to impact the capsid geometry due to the presence of the genome in the original crystal structure. For computational efficiency, this force was formatted in terms of the Cartesian origin, requiring an additional harmonic potential to constrain the center of mass of the β chains to this point. To prevent numerical instability from γ peptides post-externalization, a half-open harmonic bounding force was applied to restrain free peptides within 50 nm of the simulation origin.

Analyses were conducted using in-house Python scripts (version 3.11.6) using the MDAnalysis package (version 2.6.1)(46). A coordinate system for describing positions relative to the icosahedral asymmetric unit of the capsid is described in (Fig. S1). Trajectories were visualized and snapshots were generated using VMD version 2.0.0a5.(47) Extended methods descriptions are provided in the Supplementary Materials.

Results and Discussion

Development of low-pH capsid SBM

To construct a Cα structure-based model of the low-pH FHV capsid, structural deviations from the crystal structure observed in full-capsid all-atom (AA) simulations were projected onto the normal mode basis of the capsid. This approach assumes that exposure to low pH induces collective structural rearrangements that can be described as combinations of low-frequency normal modes. By reconstructing the structure using mode amplitudes derived from the AA trajectories, we avoid biases associated with direct structure averaging or transient high-frequency fluctuations. While the AA trajectory did not capture the equilibrium structure, normal mode analysis can analyze progression towards this state, acting as an indicator for global rearrangements and changes in the dynamics facilitated by pH. The sensitivity to the number of modes used for the fit is detailed in (Fig. S2).

The largest-amplitude displacement in both neutral-pH and low-pH AA simulations occurred along mode 8, corresponding to global capsid breathing/expansion (Fig. 2A–B), which is a non-degenerate (i.e. icosahedrally symmetric) mode. In the neutral-pH trajectory, higher-frequency modes (180, 183, and 186) contributed to localized stretching at the fivefold axes, sharpening the icosahedral vertices. In contrast, the low-pH trajectory exhibited strong contributions from modes 1 and 2, which describe sliding and stretching motions at the twofold axes, as well as mode 16, which introduces asymmetric widening across threefold and fivefold regions. Only a subset of modes displayed a steady and persistent change in mode amplitude consistent with coherent capsid-scale displacement —Mode 8 in the neutral-pH trajectory and modes 8, 1, and 2 under low-pH conditions showed the clearest steady progression, suggesting that these collective modes describe the dominant capsid-scale deformation observed in the lowpH AA trajectory. Visualizations of the motion described by the normal modes are provided in the supporting information (Movie S1-S2).

Figure 2: Normal mode analysis and low-pH model preparation.

(A) Normal mode projections over time for neutral-pH and (B) low-pH all-atom trajectories. Modes with a final amplitude of less than 100 were omitted. (C) Radius of gyration over time for AA and CG models. (D) Principal component analysis of the normal modes of each model over their respective trajectories. The time domain of each simulation is represented with a color scalewith initial frames colored white. The input crystal structure is located at approximately (−350,125).

Expansion along the dominant collective coordinate is reflected in the radius of gyration (R_g), which increases substantially in the low-pH AA trajectory relative to the neutral-pH simulation (Fig. 2C). Principal component analysis (PCA) of the combined neutral-pH and low-pH normal mode amplitudes reveals clear separation between the two states along PC1, which accounts for 84.2% of the variance in the projected mode amplitudes (Fig. 2D). In contrast, PC2 captures only 4.2% of the variance and primarily reflects residual fluctuations in the neutral-pH trajectory, with contributions from modes 5, 180, and 183. The dominance of PC1 indicates that the structural divergence between neutral-pH and low-pH conditions is effectively described by a single collective coordinate within the normal mode representation. Both AA and CG models occupy distinct regions in principal component space; however, the shift from neutral pH to low pH occurs along the same PC1 direction in both representations. Although the absolute positions of the CG models differ from their AA counterparts, the parallel displacement along this dominant coordinate indicates that the projected low-pH CG structure captures the primary direction of acid-induced capsid deformation. Because the AA low-pH trajectory did not reach full equilibrium and the CG SBM does not explicitly encode protonation-dependent local chemistry, the low-pH model is best interpreted as a conservative structural perturbation along the observed acidification-associated deformation pathway.

Single peptide externalization pathways

To characterize externalization pathways, steered MD simulations were performed to pull individual γ peptides from the FHV capsid. The reaction coordinate was defined as the radial distance between the capsid center of geometry (COG) and the COG of the first eight N-terminal residues of the selected γ peptide, promoting head-first externalization while minimizing bias associated with the unresolved C-terminal region (Fig. 3A). Distinct symmetry-dependent pathways emerged for each peptide type (Fig. 3B). γ_A followed a relatively linear trajectory and externalized through the adjacent fivefold axis. In contrast, both γ_B and γ_C externalized through the threefold axis, remaining close to their initial positions for a greater portion of the trajectory before shifting toward the symmetry axis, leading to clustering about the threefold region and a lower density of intermediate positions. Peptides did not approach the twofold axis, consistent with its proposed role in genome release rather than γ externalization. (13, 16)

Figure 3: γ peptide externalization by steered MD.

(A) Visualization of steered MD simulation setup for γ_A externalization at the fivefold axis of symmetry. (B) Projection of γ peptide externalization pathways on the triangular asymmetric unit, showing minimum work pathways for γ_A (blue), γ_B (red) and γ_C (green) (see: Fig. S1; Triangular facet projection coordinates, Supporting Methods). The fivefold axis of symmetry is highlighted with a pentagon, threefold axis with triangles, and the twofold axis with an ellipsoid. (C) Ordered work of externalization for each γ by type in the neutral-pH CG model. (D) Ordered work of externalization for each γ by type in the low-pH CG model.

The non-equilibrium work of externalization was calculated for each trajectory by trapezoidal integration of the applied steering force. In both neutral-pH and low-pH CG models, γ_A exhibited consistently lower work values than γ_B or γ_C (Fig. 3C–D). In the low-pH model, γ_A produced the lowest individual work trajectory and maintained clear separation from γ_B and γ_C across the ordered distribution. In the neutral-pH model, the lowest-work γ_A and γ_B trajectories were comparable (~16 kcal/mol); however, across the full ensemble of 180 ejection events per peptide type, γ_A consistently required less work than γ_B or γ_C. Notably, γ_B and γ_C displayed significant high-work tails, indicating greater variability and generally higher energetic cost for externalization. The distribution of observed work values for γ_B and γ_C were largely consistent with the number of contacts defined by SMOG2 (Fig. 1C), with greater contact numbers predicting larger work of externalization. In contrast, the low end of γ_A had unexpectedly low work values in comparison; by contact numbers alone, the lowest work trajectory of γ_A should exceed the first decile of γ_B. While significant variation was seen between symmetry-related groups, pathways of externalization were highly similar among γ of the same type even across a wide range of work values (Fig. S3). This demonstrates that each peptide type (γ_A, γ_B, or γ_C) has a single, shared mechanism of externalization, while the variation in the work within symmetry groups likely results from local time-dependent effects. Together, these results indicate that γ_A preferentially externalizes through the fivefold axis along a more direct pathway and with lower energetic cost, whereas γ_B and γ_C would externalize through the threefold axis via less direct and energetically less-favorable routes. Visualizations of minimum work externalization trajectories are available in the supporting information (Movie S3-S5).

Minimum work path umbrella sampling

To quantify the energetics of γ peptide externalization, umbrella sampling simulations were performed using windows extracted from the minimum-work SMD trajectory for each peptide type. All three peptides exhibited qualitatively similar potentials of mean force (PMFs), characterized by an initial steep regime corresponding to native contact rupture followed by barrier regions associated with entry into and passage through the relevant symmetry pore (Fig. 4A, S4-S5). While the relative contributions of these barriers differed modestly among peptide types, the overall PMF shapes and magnitudes were comparable.

Figure 4: Free energy profiles of γ externalization.

(A) PMFs of γ externalization in the neutral-pH CG model and (B) low-pH CG model. (C) PMFs of γ_A externalization in an all-atom partial capsid model [Figure reproduced from (20)]. (D) Pathways of γ externalization in terms of the reaction coordinate (y-axis) and motions orthogonal to the reaction coordinate (x-axis). (E) Initial geometry of γ peptides and the interior capsid surface structure encountered during externalization. γ_A is colored blue, γ_B is red, γ_C is green, and contacting capsid (β) regions are colored in greyscale. β beads are colored by their accessibility on the reaction coordinate from black to white, with darker colors indicating earlier contacts in the externalization process. The solid lines approximate the geometry of the full β chain. Fivefold and threefold axes are denoted with a pentagon and triangle, respectively.

For γ_A externalization at neutral pH, the CG PMF qualitatively matches the features of previously reported AA umbrella sampling results obtained from a pentameric subsection of FHV capsid (Fig. 4A,C). (20) In both models, contact rupture is followed by two barrier regions peaking at ~3 nm displacement. Quantitatively, the AA barrier is higher and the AA end state increases in free energy, whereas the CG profile plateaus. These differences arise from the simplifying assumptions of the Cα SBM. Coarse graining smooths the free-energy landscape and lowers effective barriers; the absence of explicit solvent removes the hydrophobic penalty associated with peptide exposure. Although absolute free energy values are not directly comparable between models, the CG SBM reproduces the overall topology and sequence of mechanistic steps observed in the AA simulations.

The neutral-pH and low-pH CG PMFs were broadly similar (Fig. 4A–B), with low-pH reducing the CG barrier by ~1 kcal/mol for γ_A and ~2 kcal/mol for γ_B and γ_C. In contrast, the AA calculations showed a more pronounced reduction in free energy across the externalization profile (Fig. 4C). In the AA model, low pH was implemented through fixed protonation of titratable residues (D75, H215, H334), altering the net capsid charge and directly modifying electrostatic interactions during externalization.(20) Moreover, AA SMD and umbrella sampling simulations were initiated from equilibrated low-pH trajectories. By comparison, the CG low-pH model represents acidification implicitly via a mode-based structural perturbation without explicit changes to residue charges or solvent-mediated effects. These differences in electrostatic representation and peptide flexibility likely account for the stronger pH-dependent shift observed in the AA PMFs.

Notably, although the CG SMD work distributions suggested that γ_A externalization is energetically most favorable (Fig. 3C–D), the minimum-work PMFs for the three peptides are similar in magnitude (Fig. 4A–B). This suggests that the lower work observed for γ_A arises from differences in pathway accessibility rather than from a substantially lower intrinsic free-energy barrier. Decomposition of umbrella trajectories into radial and orthogonal motion (Fig. 4D, S6) reveals that γ_A follows a predominantly radial pathway after initial contact rupture, whereas γ_B and γ_C require substantial lateral displacement along the capsid interior before reaching the threefold axis. Visualization of peptide–capsid contacts across umbrella windows (Fig. 4E) confirms that γ_A rapidly accesses a narrow cavity (“bottle neck”) at the fivefold axis, while γ_B and γ_C traverse a broader interior basin prior to externalization. Thus, γ_A appears favored not because of a dramatically lower intrinsic energy barrier, but because it accesses a more direct escape route.

Importantly, when considering these results, the radial reaction coordinate should be interpreted as an order parameter for externalization progress rather than as a complete description of the multidimensional pathway. Although γ_B and γ_C undergo substantial orthogonal displacement before reaching the threefold pore, externalization necessarily involves an increase in radial distance from the capsid center, making this coordinate a physically meaningful measure of peptide release. In the umbrella sampling calculations, restraints were applied only along the radial coordinate; orthogonal degrees of freedom were not directly restrained and therefore contribute to the PMF through sampling within each window. In an intact virion, however, the capsid interior is densely packed with genomic RNA, which is expected to restrict lateral peptide displacement and bias motion toward radial escape. These considerations motivate explicit examination of confinement effects using a pressure-like radial force applied to the γ peptides, presented in the subsequent section.

Genome Confinement and radial pressure effects on peptide externalization

To examine how genome confinement may bias γ peptide dynamics, we introduced a phenomenological pressure-like radial force applied to the Cα beads of γ peptides. Unlike the SMD and umbrella sampling simulations, this potential was applied to all γ peptides in a single simulation, allowing the full-capsid representation to capture collective responses to confinement (Fig. 5A). Simulations were performed for 1.2×10⁷ integration steps at varying force magnitudes, and the number of externalized peptides was recorded.

Figure 5: Externalization by genome-like radial force.

(A) Schematic of the simulation setup. A constant force directed radially away from the center of the capsid is applied to all γ peptide beads to mimic internal pressure. (B) Projection of peptide positions on the asymmetric unit from the neutral-pH CG model with a force of 0.72 kcal mol⁻¹ nm^-1. (C) End-state number of peptides ejected in varying pressure-like force simulations in the neutral-pH CG and (D) low-pH CG models. (E) End-state fraction of modified peptides ejected when initial contacts are removed entirely or (F) weakened in varying pressure-like force simulations.

In the neutral-pH CG model, the first γ_A externalizations were observed at forces of 0.69 kcal mol⁻¹ nm⁻¹ (Fig. 5C, Movie S6). Up to forces of 0.86 kcal mol⁻¹ nm⁻¹, no γ_B or γ_C externalizations were observed, and these events remained occasional at the upper limits of force. The projection of peptide positions onto the triangular asymmetric unit (Fig. 5B) shows that γ_B and γ_C remain restricted to their initial binding sites while γ_A peptides move radially toward the fivefold axis. At excessive forces (≥ 1.07 kcal mol⁻¹ nm⁻¹), the applied pressure was sufficient to fragment the capsid itself while γ_B and γ_C largely remained bound to their associated β chains as they dissociated (Movie S7). Even at a limiting level of force, the γ_B and γ_C peptides are entirely lodged within their local geometric environment. In the low-pH CG model, the same effect was observed with enhanced selectivity. γ_A externalization occurred at lower force (0.67 kcal mol⁻¹ nm⁻¹), whereas γ_B and γ_C externalization was not observed until a force of 0.93 kcal mol⁻¹ nm⁻¹ (Fig. 5D). Thus, under conditions mimicking genome-induced confinement, externalization is strongly dominated by γ_A peptides, with the bias further amplified under low-pH conditions.

To determine whether this selectivity arises from peptide geometry or from capsid structural constraints, modified CG models at neutral-pH were constructed by altering the β–γ structure-based contacts. In one model, these contacts were removed entirely (U(β−γ)^off ), allowing peptides to dissociate freely. In another model, the contact strengths were reduced to the minimum required to maintain native positioning at the simulation temperature (U(β−γ)^weak ). To avoid steric clogging effects during externalization, these modifications were applied to only 12 peptides of each type. In the model where β−γ contacts energies are turned off, the strong preference for γ_A externalization was largely lost (Fig. 5E). γ_B and γ_C externalized at rates comparable to γ_A. Similarly, weakening the β–γ interactions eliminated the bias toward γ_A externalization (Fig. 5F). These results demonstrate that the selectivity depends on the peptide initial geometry and not other symmetry-related effects (e.g. ease of translocation through the fivefold versus threefold pores). The pathway of γ_A externalization is more direct and aligned with radial stress, allowing it to externalize in the context of a genome-mimicking force while γ_B and γ_C remain constrained by their local structural environment. Together, these results support the hypothesis that γ_A peptides are major contributors to pH-mediated membrane disruption, while γ_B and γ_C may be less accessible because of their local geometry and interactions within the capsid interior and/or genome (Fig. S1A).

Free-energy landscape under radial stress

To determine how genome-derived mechanical stress influences the energetics of peptide externalization, umbrella sampling simulations were repeated with an applied radial force of about 0.24 kcal mol⁻¹ nm⁻¹ (Fig. S7). For a Cα coarse-grained bead representing the excluded volume of an entire residue, this force corresponds to ~1.7 pN per residue. Because the force is applied locally to peptide beads rather than uniformly over the capsid interior, we do not interpret it as a quantitative hydrostatic pressure. Using a residue-scale cross-sectional area of ~1 nm² provides only an order-of-magnitude estimate of an effective mechanical stress of approximately 16 atm. This value is below the ~20–60 atm internal pressures estimated for dsDNA bacteriophages, which arise from dense packaging of stiff DNA genomes. (48, 49) In contrast, ssRNA viruses assemble cooperatively around their genomes rather than packaging them into preformed capsids(50, 51) and are therefore not expected to maintain the large stored pressures characteristic of dsDNA bacteriophages. Nevertheless, confinement of the highly charged RNA genome within the capsid can generate outward osmotic pressures due to electrostatic repulsion and polymer confinement. Theoretical estimates suggest that this polyelectrolyte contribution alone can produce internal stresses on the order of ~10 atm for typical ssRNA viruses. (52) The applied radial force therefore provides a physically reasonable perturbation for probing how genome-derived internal stresses may bias γ-peptide externalization in FHV.

Steered MD externalizations were repeated under the same protocol with the added pressure force, and minimum-work trajectories were similarly used to initialize the umbrella windows. When peptide motion is unhindered, the applied force contributes a linear mechanical work term to the free energy (G_appF) (Fig. S7), enabling reconstruction of the intrinsic free-energy profile (G_int) by subtracting the pressure contribution from the observed PMF (G_obs = G_int + G_appF). Sampling distributions are provided in (Fig. S8).

In the neutral-pH CG model, application of radial pressure shifts the dominant barrier for γ_A externalization from pore entry to rupture of the initial peptide–capsid contacts (Fig. 6, top row; Fig. S9). Once these contacts are disrupted, the following translocation through the fivefold pore is energetically favored. In contrast, γ_B exhibits a substantial increase in the free-energy cost associated with reorientation toward the threefold axis prior to insertion (Fig. S9). This likely reflects the broader free-energy basin sampled near the threefold axis compared to the narrow fivefold channel, which geometrically constrains γ_A to remain aligned along the externalization pathway. For γ_C, the barriers associated with contact rupture and pore entry are comparable in shape but occur at higher free-energy values (~10 kcal/mol), indicating a less favorable externalization pathway compared to γ_A.

Figure 6: Free energy curves in the context of a genome-like radial force.

Free energy curves of the neutral-pH CG model (top), low-pH CG model (middle), and a comparison between the models (bottom). Standard free energy curves are from Fig. 4, “Pressure” free energy curves were performed with a force of 0.24 kcal mol⁻¹ nm⁻¹ applied to γ peptide beads, and “reconstructed” curves (i.e. intrinsic) were generated by subtracting the free energy of the applied force from the “pressure” PMFs (Fig. S7). The blue zones represent the positioning of bulk solution (i.e. γ egress from the capsid) on the reaction coordinate with the cutoff calculated from the “pressure” trajectories.

In the low-pH CG model, the pressure-like force further enhances the thermodynamic drive for externalization (Fig. 6, middle row; Fig. S10). Following disruption of initial contacts, γ_A experiences a strongly downhill free-energy profile, indicating that radial work is efficiently converted into peptide displacement. For γ_B, the secondary barrier associated with pore entry is largely eliminated under pressure, whereas γ_C retains a higher entry barrier but becomes increasingly stabilized once it reaches the threefold pore. Reconstruction of the intrinsic PMFs reveals that pressure primarily alters the dynamics beyond the contact-rupture transition state, transforming previously flat regions of the landscape into downhill escape pathways.

Direct comparison of neutral-pH and low-pH profiles (Fig. 6, bottom row) demonstrates that γ_A is uniquely stabilized by the combined effects of capsid expansion and internal pressure. Although all peptides show increased thermodynamic favorability after contact rupture under pressure, γ_A is the only peptide whose entire free-energy profile shifts downward in the low-pH model. These results indicate that γ_A externalization is both structurally predisposed and selectively amplified by the coupled effects of genome confinement and pH-induced capsid dynamics.

Integration with experimental observations

These findings can be integrated with recent experimental observations of FHV disassembly and related non-enveloped RNA viruses. Structural analysis of an early FHV disassembly intermediate generated under entry-mimicking conditions revealed asymmetric capsid expansion, disordering of flexible membrane-penetrating peptides, openings near the twofold axis, and substantial reorganization of the genomic RNA.(16)These observations support the view that FHV entry involves localized, symmetry-dependent rearrangements rather than uniform capsid expansion. They also suggest that the genome may actively contribute to disassembly dynamics, consistent with our finding that genome-like radial stress can reshape the energetic landscape for γ peptide externalization. Importantly, the twofold-associated openings observed experimentally likely correspond to genome release, whereas our simulations suggest that γ peptide externalization is preferentially biased toward pentamer-associated γ_A peptides that access the fivefold axis. Thus, peptide exposure and genome release may proceed through distinct but coupled symmetry-dependent pathways during entry.

Our results also align with experimental studies of the non-enveloped RNA virus Nudaurelia capensis omega virus (NωV). Although NωV has a T=4 capsid, its capsid subunits adopt a fold similar to FHV and undergo a related maturation cleavage that releases a membrane lytic peptide. Time-resolved cryo-EM studies showed that maturation-associated structural changes and cleavage occur at position-dependent rates in NωV, with quasi-equivalent subunits following distinct maturation pathways.(53) Subsequent experiments demonstrated that membrane disruption is driven primarily by lytic peptides released from pentamer-associated, rather than hexamer-associated, subunits.(54)These observations support our prediction that pentamer-associated lytic peptides are the most likely contributors to membrane disruption. In FHV, this corresponds to γ_A peptides, which our simulations predict externalize most readily through the fivefold axis.

This comparison suggests a direct experimental test of the model. Selective stabilization or restriction of the FHV fivefold environment, for example through engineered crosslinks or mutations near fivefold-associated interfaces, should suppress γ_A externalization and reduce low-pH peptide exposure, membrane disruption, and/or infectivity. Such experiments would provide a direct means of evaluating whether the symmetry-dependent externalization pathways identified here contribute to productive viral entry.

Conclusion

In this work, we used full-capsid structure-based molecular dynamics simulations to investigate infection-relevant dynamics in the T=3 Flock House virus capsid. These simulations reveal that γ peptide externalization follows symmetry-dependent pathways determined by the geometric organization of peptides within the capsid. Peptides associated with the pentameric environment preferentially externalize through the fivefold axis along relatively direct radial routes, whereas peptides located in other symmetry environments require substantial lateral rearrangement before reaching externalization sites. Although acidification alone produces only modest changes in the intrinsic free-energy barriers for peptide release, it alters the structural state of the capsid and modifies the accessibility of these pathways.

When combined with internal mechanical stress that mimics genome confinement effects, the low-pH capsid strongly favors externalization of pentamer-associated peptides and converts otherwise flat regions of the energy landscape into downhill escape pathways following rupture of peptide–capsid contacts. These results highlight how capsid geometry, environmental conditions, and internal mechanical forces cooperate to regulate peptide release during viral entry. More broadly, this work demonstrates that full-capsid native-centric models provide a computationally efficient approach for probing functional energy landscapes and collective transitions in large viral assemblies, complementing atomistic simulations and experimental studies of viral infection mechanisms.

Statement of Significance.

Many non-enveloped viruses contain internal protein elements that become exposed during infection to disrupt host membranes and enable genome delivery. Mechanisms of externalization for these components remain poorly understood because they involve rare, cooperative motions across the entire particle. Using full-capsid structure-based simulations of a model RNA virus, we show that the internal geometry of T=3 icosahedral symmetry itself biases peptide release towards the fivefold axis of symmetry, despite near-identical chemical constraints. While the low-pH structural perturbation produces only modest changes in the energetic barriers for peptide externalization, internal mechanical stress that mimics genome confinement strongly and synergistically promotes release along these pathways. These results reveal how capsid architecture and environmental conditions together regulate peptide release during viral entry.

Data and code availability

Models and trajectories are available from the corresponding author upon reasonable request.