3.1. Secondary Structure Analysis
Figure 3 shows a summary of secondary structure adopted by quasi-equivalent A, B, C, and D chains of capsid-incorporated dimers in the apo-form and AT130-bound systems during MD simulations. Notably, the spikes of the apo-form AB dimer are well-folded in the crystal model (PDB 2G33, 3.96 Å [
33]), while those of the CD dimer are not (
Figure 3, left). Our previous work on the apo-form capsid highlighted the flexibility of the spikes, particularly those of the CD dimer, and increased mobility was shown to correlate with higher B-factors and lower local resolution in experimental data [
12]. Nevertheless, disordered spike conformations are in contrast with high resolution structures of the apo-form capsid (PDB 1QGT, 3.30 Å [
4] and 6HTX, 2.66 Å [
42]), which demonstrate that fully-folded CD spikes are possible.
During MD simulations of the apo-form capsid, the spikes of AB dimers were consistently ordered, with residues identified as helical in experimental structures exhibiting a high probability to remain so over microsecond timescales (
Figure 3a,b). Taken together with previous results characterizing flexibility [
12], these data indicate that spikes can be both well-folded and highly mobile, such that low resolution in experimental structures need not imply structural disruption. The spikes of CD dimers recovered some helicity relative to the crystal model, particularly in helix 3b, but were far more likely to be observed in partially-folded states (
Figure 3c,d). Specifically, a fully-folded spike tip was present in 11.5% and 8.2% of conformers in apo-form C and D chains, respectively, while a spike tip that was folded along helix 3b but partially-folded along helix 4a was present in 75.2% and 79.8% of conformers (
Figure S1a,b). Limited simulation timescales entail the consequence that biomolecules retain bias from their initial configurations. Given extended sampling, all constituent CD spikes may have eventually refolded. In previous MD simulations of the free dimer, we observed that a salt bridge between Glu77–Arg82 could stabilize partially-folded states of the spike [
16]. The salt bridge was identified in 22.7% and 27.5% of conformers in apo-form C and D chains, respectively, accounting in part for their reduced ability to refold (
Figure S1a,b).
In the crystal model of the AT130-bound capsid (PDB 4G93, 4.20 Å [
10]), the spikes of all four quasi-equivalent chains are disordered, with CD dimers exhibiting even less helical content than in apo-form (
Figure 3, right). This effect was attributed directly to the presence of the bound CpAM in the crystallographic study [
10]. However, during MD simulations of the AT130-bound capsid, the spikes of AB dimers rapidly recovered helicity, exhibiting nearly the same probability to be fully-folded as in the apo-form capsid (
Figure 3e,f). Interestingly, the majority of CD dimer spikes also refolded (
Figure 3g,h), with 77.7% and 64.0% of conformers in AT130-bound C and D chains, respectively, found to contain maximum helicity (
Figure S1c,d). Ultimately, following equilibration under physiological conditions, the AT130-bound capsid relaxes from the crystal model and adopts significantly more ordered spikes, even in the presence of the CpAM. This finding indicates that AT130 does not damage the helical nature of capsid spikes, but actually enhances the ability of spikes to refold relative to apo-form.
Beyond behavior of the capsid spikes, other differences between the crystal models and MD simulation ensembles are notable (
Figure 3). The 3
-helical element preceding helix 1, which was not consistently identified in crystal structures, was observed in around half of the ensembles; this element demonstrated a higher probability to exist as an
-helix in the AT130-bound capsid. Helix 1, which is not clearly defined in the A, B, and D chains of the highest resolution crystal (PDB 6HTX, 2.66 Å [
42]), was shown to exist as the expected
-helix the majority of the time in both apo-form and AT130-bound systems. Although the degree of disorder at the C-terminus of helix 2 varied, it occurred with the highest probability in chain A of the AT130-bound capsid. While a break in secondary structure associated with the dimer hinge between helix 4a and 4b was not emphasized in crystallographic data, it can be observed in MD simulation results in the 88–98 residue range. The length of the loop separating helices 4 and 5 exhibited some variability in length in the AT130-bound CD dimer, which is contacted by a CpAM on either side. The C-terminal tail exhibited some unexpected probability to contain 3
-helical regions, particularly around residue 140 in the B and C chains containing AT130, beyond the point at which the crystal model was well-resolved.
3.2. Dynamical Network Analysis
Figure 4 shows dynamical network models for the AB and CD capsid-incorporated dimers in the apo-form and AT130-bound systems, colored by communities and weighted by the extent of correlation in residue motions. Network models were calculated in triplicate, splitting each ensemble of three million frames into three sets of one million randomly selected frames. Notably, the three sub-datasets produced similar, yet inequivalent results (
Figure S2). The observation that 20 μs of conformational sampling was insufficient for precision in the analysis underscores the highly flexible nature of the Cp dimer. Nevertheless, the presented network models reveal important aspects of correlated motion and communication in the functional dynamics of Cp, and findings are consistent with experimental characterizations of dimer sub-domains based on concerted structural displacements observed in crystal structures [
11].
Dynamical network analysis is commonly applied to study functional dynamics in proteins and can shed light on mechanisms of allosteric communication and molecular signalling [
43]. Network models comprise nodes representing residues, and here edges between nodes are based on correlation measured in residue motions. Communities are partitions of the network, within which there are more and stronger connections between nodes, indicating concerted movement. They may be thought of as structural sub-domains, but are defined and grouped purely according to dynamics [
35]. Each Cp network model exhibited between seven and eleven communities, with the number of communities identified for a given dimer system varying over the three sub-datasets used for analysis (
Figure S2).
Consistently, each model included at least five communities in the body of the dimer, and as many as four in the C-terminal tails. One community typically encompassed helix 1 and 5 for each half-dimer (
Figure 4, purple and red). Importantly, helix 5 forms the interdimer interface or contact sub-domain. An additional community typically encompassed helix 2–3a for each half-dimer (
Figure 4, blue and green), which comprises a portion of the chassis sub-domain [
11]. The remainder of the Cp sub-domain commonly described as the fulcrum [
11] was alternately involved in the helix 1,5 or helix 2–3a community of its parent half-dimer, or split between these two communities depending on the sub-dataset. A fifth community typically encompassed helices 3b-4a of both half-dimers in the spike tip (
Figure 4, orange), which is also considered a distinct sub-domain [
11]. The spike tip is an important aspect of the intradimer interface, and the latter was the only community shown to significantly bridge the two half-dimers. Like the fulcrum, helix 4b could be associated with either the helix 1,5 or 2–3a community. Key communities and their associated Cp sub-domains are summarized in
Table 2.
Figure 1a,b describes the locations and compositions of the sub-domains. Significant differences between network models involved fragmentation of the five major communities.
The helix 1,5 communities are consistently prominent and highly correlated across all network models (
Figure 4, purple and red). This may arise in part from interdimer contact and constraints imposed by capsid quaternary structure enforcing shared motions. The concerted displacement of helix 5 during capsid incorporation has been proposed to modulate the assembly process [
44]. In one apo-form chain D model, and two AT130-bound chain B models, a small community within helix 1 developed, with minimal attachments to the chassis (
Figure 4c and
Figure S2, yellow). The loop connecting helix 1 and 2, part of the fulcrum, which has been suggested to mechanically coordinate the motions of the interdimer interface and the spike [
11], showed the ability to participate in the helix 1,5 community through weakly correlated motions (
Figure 4).
Distinct C-terminal tail communities found in the apo-form system were absorbed by helix 1,5 communities in 4 out of 6 models of the AT130-bound system (
Figure S2). That contact of the CpAM can establish communication between helix 5 and the C-terminals of Cp149 (which serve as flexible linkers to the RNA-binding CTDs in full-length Cp183 [
45]) raises the possibility of a secondary mechanism underlying AT130’s ability to disrupt the process of genome packaging, apart from mistiming. On the other hand, all network models of AT130-bound chain D show the C-terminal tail community encroaching into that of helix 5 (
Figure 4d, red), as far up the sequence as residue 126, likely resulting from direct interaction with the CpAM. Importantly, ordering of residues 126–136 has been correlated with capsid formation [
44,
46]. Secondary structure analysis showed that a 3
helix in this region becomes less likely in the A and D chains, yet more likely in the B chain in the presence of AT130 (
Figure 3).
The helix 2–3a communities and the remainder of the chassis represent the regions of Cp with the least degree of correlated motion (
Figure 4). Nevertheless, their dynamics are affected by AT130 in unexpected ways. Although it manifests as a single community in all models of apo-form chain A, the helix 2–3a community splits into separate helix 2 and helix 3a communities in all AT130-bound models (
Figure 4a,c and
Figure S2, light and dark green). This is an allosteric effect, possibly arising from changes in capsid morphology [
10], as chain A does not contact the CpAM. Separate helix 2 and helix 3a communities are observed in two out of three models of chain B, regardless of direct interaction with AT130 (
Figure 4a,c and
Figure S2, light and dark blue). Helix 2 and helix 3a of chain C are split into separate communities in two out of three models in the apo-form system, but only one out of three models in the AT130-bound system (
Figure 4b,d and
Figure S2, light and dark green), which likewise involves direct contact. A single helix 2–3a community is found in all models of chain D regardless of the CpAM (
Figure 4b,d and
Figure S2, light and dark blue). Interestingly, chains A and D, which consistently exhibit combined helix 2–3a communities in the absence of AT130, also represent the quasi-equivalent pockets that are not amenable to AT130 binding [
10].
The helix 3b-4a community is consistently prominent across all network models (
Figure 4, orange). The strength of dynamical correlation within the community is strong when the probability of helicity is high, and is clearly decreased for the case of the apo-form CD dimer where the majority of spike tips remain only partially-folded (
Figure 4b). Reduced helicity in the spike tips has been shown to disrupt the intradimer interface and negatively impact assembly [
16]. Remarkably, in one model of the AT130-bound CD dimer, the helix 3b-4a community splits into two separate communities, one for each half-dimer (
Figure 4d and
Figure S2, light and dark orange), indicating diminished communication across the interface. Correlation within the individual half-dimer spike tips is also weakened. The structural origin of this tertiary disruption in the presence of AT130 can be attributed to dramatic bending of the dimer hinge (discussed in the following section), which results in an open spike conformation and broken intradimer interface with an increased probability in chain D.
3.3. Hinge Curvature Analysis
Figure 5 shows distributions of curvature adopted by the dimer hinge for the AB and CD capsid-incorporated dimers in the apo-form and AT130-bound systems during MD simulations. Importantly, Cp contains a hinge that connects the spike tip and chassis sub-domains. The hinge is defined by conserved Gly63 and Gly94, which impart a bending compliance to helix 3 and 4, respectively, and partition them into helices 3a/b and 4a/b [
11]. We have previously used MD simulations to characterize hinge bending in free dimers, showing that fully-folded spike tips exhibited smaller angles of helix curvature and greater potential for a closed intradimer interface [
16]. Here, the analysis is repeated for capsid-incorporated dimers, measuring the angle at the point of highest curvature along helix 4 (the hinge vertex), which was consistently found to be Val93 in both free and capsid-incorporated dimers. Note that the reported angles capture helix 4’s deviation from linearity resulting from bending of the hinge; larger hinge angles indicate more pronounced bending along the helix.
All four quasi-equivalent chains exhibit significantly higher average hinge angles than those observed in experimental structures of the apo-form capsid [
4,
33,
42], more similar to those observed in crystals of assembly-incompetent dimers [
11] (
Figure 5). This result is noteworthy, as comparison of capsid and dimer structures has inspired hypotheses that relaxed hinges and closed intradimer interfaces are a feature of assembly-active Cp conformations [
16]. MD simulations indicate that capsid-incorporated and free dimers are more alike in spike conformation than experiments suggest. On the other hand, hinge angles measured for the AT130-bound crystal model [
10], particularly in the CD dimer, are quite extreme (
Figure 5d). Importantly, these values are artifacts arising from highly disordered secondary structure within the hinge region, which leads to inaccurate fitting of helix abstractions and assessment of curvature [
37].
The distributions of curvature measured for MD simulations of the apo-form capsid are Gaussian for all four quasi-equivalent chains (
Figure 5a,b). The A chain distribution is markedly flatter than the other three. As the A chain forms capsid pentamers, quasi-equivalence likely accounts for differences in hinge behavior relative to the B, C, and D chains that form capsid hexamers. The A chain distribution, centered at 40.4
, also exhibits the highest angle (most dramatically bent hinge), while the B chain distribution, centered at 37.0
, exhibits the lowest (
Figure 5a). Both A and B spike tips are well-folded during MD simulations, yet well-folded free dimers exhibited distributions centered at 37.0
(based on data from [
16]), indicating that preferred hinge states are influenced by quasi-equivalence and constraints imposed by capsid quaternary structure. It is likely that because the A chain hinge is more bent on average, the B chain must necessarily be less bent on average to maintain a closed intradimer interface. An example of an apo-form AB dimer sampled during MD simulations is shown in
Figure 6a.
Interestingly, the C and D chain distributions are balanced in average hinge angle, both centered at 39.3
(
Figure 5b). The C and D spike tips have a high probability to be only partially-folded during MD simulations, and partially-folded free dimers exhibited distributions also centered at 39.3
(based on data from [
16]). This result indicates that secondary structure is the primary determining factor for hinge behavior in CD dimers in the absence of AT130. The Glu77–Arg82 salt bridge, which was observed frequently in MD simulations of the apo-form CD dimer, can cause larger average hinge angles by stabilizing a loop at the apex of the partially-folded spike tip that has the potential to protrude into the interface, leading to steric clash and electrostatic repulsion that is alleviated by adopting increased curvature along helix 4 [
16]. An example of an apo-form CD dimer sampled during MD simulations is shown in
Figure 6b.
The distributions of curvature measured for MD simulations of the AT130-bound capsid are shifted and flattened for all four quasi-equivalent chains compared to apo-form (
Figure 5c,d). Further, the distributions are only strictly Gaussian for B and C chains, which contain the occupied CpAM binding pockets, while the distributions for A and D chains exhibit distortion. Higher populations of above- and below-average hinge angles in the AT130-bound capsid could partially account for lower resolution in the crystal model [
10], a consequence of averaging over more pronounced conformational variation in spikes. Importantly, the spike tips of all four quasi-equivalent chains are well-folded during MD simulations, so changes in hinge behavior are not a consequence of disordered secondary structure, but arise from the presence of AT130. The A and B chain distributions are shifted down by exactly 1.1
(
Figure 5c), while the C and D chain distributions are shifted down by exactly 4.5
(
Figure 5d), indicating relaxation of the hinge on average. An example of an AT130-bound AB and CD dimer sampled during MD simulations is shown in
Figure 6c,d and
Figure S4.
Notably, the D chain of AT130-bound dimers exhibits an unusual propensity for pronounced hinge bending during MD simulations, accounting for distortion on the right-side of the distribution (
Figure 5d). Upon examination, it was discovered that larger hinge angles in the D chain often lead to marked opening of the spike tip and breaking of contact across the intradimer interface. This behavior explains fragmentation of network communities in the spike tip in the presence of AT130. Further, this open conformation facilitates formation of a salt bridge between Arg28–Asp83, which introduces a direct contact between the spike tip and fulcrum and provides a mechanistic explanation for allosteric communication across these sub-domains (
Figure 6e–h).
Rarely, the open conformation, along with the salt bridge, was observed in apo-form CD dimers (
Figure 6e), but never in AB dimers with or without AT130. Interestingly, the presence of the salt bridge is not necessarily an indication of an open spike conformation, as disorder in helix 4a or displacement of helix 2 can bring Arg28 in proximity to interact with Asp83 without breaking the intradimer interface (
Figure 6f,g). Disruption in the chassis was possible, by prying away helix 4b, but network analysis did not indicate a role for the base of the four-helix bundle in intradimer communication. A partially-folded helix 3b appears to increase opportunity for the salt bridge. Overall, the Arg28–Asp83 interaction occurred with low probability, but was observed more often in the D chain with AT130 (
Figure 6h). Time series of hinge angles and Arg28–Asp83 contact for apo-form and AT130-bound CD dimers are provided in
Figure S3. Characterization of the typical separation distance between Arg28–Asp83 is provided in
Figure S4.
3.4. Principal Component Analysis
Figure 7 shows the top five modes of essential dynamics observed for the AB and CD capsid-incorporated dimers in the apo-form and AT130-bound systems based on principal component analysis of MD simulations. Our previous work on the apo-form capsid examined essential dynamics for the complete icosahedron, reporting asymmetric distortion and complex collective motions of dimers within the capsid surface [
12]. Overall, the intact capsid modes proved difficult to interpret due to limited conformational sampling. Here, each capsid-incorporated dimer system benefits from 60× more sampling, providing the advantage of reasonably converged datasets.
Twenty modes were calculated per dimer, and for each system, the top five modes together account for ∼40% of total variance. The first (lowest frequency) modes consistently contribute 10% or less to variance, ultimately indicating that there is no dominant fluctuation to be distilled from the dynamics. Cp remains a marvel of flexibility, even when constrained by the quaternary structure of the capsid. Due to the diversity of observed motions, it was not immediately clear, even given such a large conformational ensemble, whether the differences in modes across systems arose strictly from quasi-equivalence and the influence of AT130, or if the systems remain yet undersampled.
To assess convergence of simulation data, we used the root mean square inner product (RMSIP) [
47] to compare the sub-spaces described by the modes. RMSIP scores range from zero to one, indicating the degree of similarity, where a value of one means the sub-spaces are identical or overlap completely. A score of 0.7 is considered excellent, while a score of 0.5 is considered fair [
40,
48]. Results of RMSIP analysis using twenty and five modes are given in
Figure S5. Scores comparing the first and second halves of the simulation data are 0.9 for all systems, imparting confidence that their essential dynamics are well-sampled. Comparisons between each of the four simulations, AB and CD dimers in the apo-form and AT130-bound capsids, range from 0.5–0.7 for twenty modes and 0.3–0.5 for the top five. These results suggest that the motions described below can indeed be attributed to the effects of quasi-equivalence or AT130.
All calculated modes necessarily exhibit motion relative to the chassis, which has been described as the underlying constant substructure around which other dimer sub-domains pivot [
11], and was used here as the standard for alignment. Indeed, the majority of mode movements can be linked to distinct sub-domains, including the fulcrum (helix 1–2), spike tip (helix 3b-4a), and interdimer contact sub-domain (helix 5) [
11]. Overall, most motions capture flexing of the spike and radial movements of the fulcrum and helix 5. The latter may be related to the ability of the fivefold vertices of the capsid to protrude and sixfold/threefold vertices to flatten, a known morphological adjustment that can accommodate the presence of CpAMs [
10,
15,
33,
49].
For the apo-form AB system (
Figure 7a), Mode 1 is characterized by spike waving, primarily on the side of chain A, consistent with the observation that its average dimer hinge angle is larger than that of chain B. Helix 3 of chain A concomitantly sinks into the dimer core. Mode 2 is characterized by a slight twisting of the half-dimers against each other at the chassis, along with upward motion of the fulcrum and helix 5 of chain B. Mode 3 is the converse of this movement, characterized by downward motion of the fulcrum and helix 5 of chain A, with corresponding motion of the spike. Mode 4 shares similarities with Mode 2, but the movement of the fulcrum of chain B is skewed toward chain A and carries the spike with it. Mode 5 involves a lateral stretching of helix 5 of chain A and opposite waving of the spike.
For the apo-form CD system (
Figure 7b), Mode 1 is characterized by concerted motion of chain C, along with the spike tip of chain D, consistent with the observation from network models that helix 3b-4a is the key location of intradimer communication. Helix 5 of chain C undergoes an upward motion, perhaps shifting along with chain B where the two share an interface. Mode 2 is characterized by a rotary motion at the chassis, which sends the fulcrum and helix 5 of chain D upward. Mode 3 shares similarities with Mode 2 of the AB dimer, with slight twisting of the half-dimers against each other that sends the fulcrum and helix 5 of chain B downward. Mode 4 involves a lateral compression of the dimer, particularly along helix 4b-5 of chain D. Mode 5 pushes the fulcrum and helix 5 of chain C upward.
For the AT130-bound AB system (
Figure 7c), Mode 1 is characterized by a slight twisting of the half-dimers against each other at the chassis, and a downward motion of the fulcrum and helix 5 of both chains. Mode 2 is characterized by spike movement, particularly in chain A. Interestingly the spike tips do not travel in the same direction. Mode 3 involves rotation of the half-dimers toward each other at the chassis, pressing upward the spike tip of chain B. Mode 4 shares similarities with Mode 1, entailing a dimer expansion that pushes the fulcrum and helix 5 of both chains upward, and that of chain B to the side. Mode 5 is a rotation of the dimer in the direction of chain B, which causes a seesawing movement along the helix 5 axes.
For the AT130-bound CD system (
Figure 7d), Mode 1 is characterized by spike movement, particularly in chain C, similar to Mode 2 of the AB dimer. Again the spike tips do not travel in the same direction. Mode 2 is characterized by a familiar downward motion of helix 5 of chain D. Mode 3 involves concerted downward motion of the entire dimer, except the spike tip of chain D, which may arise from the conformations that exhibit a broken intradimer interface with disrupted cross-communication. Mode 3 entails a pressing of helix 5 of chain C downward. Mode 5 includes a contraction of helix 5 of both chains inward toward the chassis.