2-μs molecular dynamics simulation of the fully glycosylated HIV-1 SOSIP Env trimer
•
Env protomers undergo scissoring movements, which induce trimer asymmetry
•
Glycans form microdomains, which remained stable at the microsecond timescale
•
Neutralizing antibodies recognize interfaces between glycan microdomains
Summary
The trimeric HIV-1-envelope (Env) spike is one of the most glycosylated protein complexes known, with roughly half its mass comprising host-derivedN-linked glycan. Here we use molecular dynamics to provide insight into its structural dynamics and into how both protomer and glycan movements coordinate to shield the Env protein surface. A 2-μs molecular dynamics simulation of a fully glycosylated atomistic model of the HIV-1 SOSIP Env trimer revealed a spectrum of protomer-scissoring and trimer-opening movements. Network analysis showed that highly conserved glycans combined with protomer scissoring to restrict access to the binding site of the CD4 receptor. The network property of betweenness centrality appeared to identify whether glycans spread to restrict access or cluster to maintain the high-mannose character of the shield. We also observed stable microdomains comprising patches of glycan, with neutralizing antibodies generally binding at the interface between glycan microdomains. Overall, our results provide a microsecond-based understanding of the Env glycan shield.
Graphical Abstract
Keywords
betweenness centrality
broadly neutralizing antibody
CD4 receptor
fractional microsecond dynamics
glycan microdomain
glycan shield
HIV-1 viral spike
molecular dynamics simulation
network theory
scissoring movement
Introduction
The HIV type 1 (HIV-1) is an enveloped virus and the etiologic agent of acquired immunodeficiency syndrome (AIDS), which killed an estimated 1.1 million in 2015 (
). On the HIV-1 virion, the trimeric HIV-1 envelope (Env) spike is responsible for binding cellular receptors, CD4 and coreceptor (either CCR5 or CXCR4), and fusing viral and target cell membrane to facilitate entry. The spike is a heterodimeric trimer composed of two glycoproteins (gp): gp120 and gp41. It is one of the most highly glycosylated proteins, where host-derivedN-linked glycans account for about half of its total mass. Being the only viral component to protrude outside of the protective virion membrane, the trimeric HIV-1 envelope (Env) is the sole target of virus-directed neutralizing antibodies (
). Thus, trimeric Env must fulfill two functions, entry and evasion, and it uses both conformational change andN-linked glycosylation to do so.
On the surface of infectious virions, the HIV-1-Env trimer samples at least three conformations: a ground-state conformation, which is preferentially recognized by most broadly neutralizing antibodies; an obligate intermediate, which can be induced by the binding of a single CD4; and an activated state, which is competent to bind the coreceptor and can be induced by the binding of the multimeric or cell-surface CD4 (
). Additional conformational changes to facilitate entry involve the recognition of a coreceptor, the formation of a prehairpin intermediate, and the transition to postfusion states (reviewed in
). Changes between the prefusion conformations are observed on timescales of seconds, as measured by single-molecule fluorescence resonance energy transfer (
). However, little is known about the submicrosecond dynamics of the HIV-1 Env trimer.
The most characterized of the various prefusion Env conformations is a prefusion-closed conformation, for which structures of fully glycosylated HIV-1 Env trimer have been determined (
). In this prefusion-closed conformation, the trimeric Env is covered by a dense array ofN-linked glycans that all antibodies targeting this prefusion-closed conformation must accommodate (
). In an effort to gain insight into the submicrosecond dynamics of the glycan shielded prefusion closed conformation, we carried out a 2-μs all-atom molecular dynamics simulation of a fully glycosylated BG505 SOSIP.664 Env trimer and used network analysis to help understand the collective behavior of glycans. Our results showed how fractional microsecond dynamics of protomers andN-linked glycan could lead to structural asymmetry, modulate structural plasticity, and shield the protein surface of the Env trimer in its prefusion-closed conformation from both the CD4 receptor and broadly neutralizing antibody.
Results
Principal Component Analysis Revealed Four Conformations of Env Trimer
To characterize the motions of the HIV-1 SOSIP Env trimer on the microsecond timescale, we used the trajectory from a 2.0-μs all-atom molecular dynamics simulation of the BG505 SOSIP.664 Env trimer (Figure 1A), for which we modeled mannose5-N-acetyl glucosamine2(Man-5) glycans, as the glycan unit present at each sequon. Man-5 represents a prototypical glycan species that would be present at a given sequon and, with fewer degrees of freedom than Man-9, would guarantee adequate sampling during the simulation.
Figure 1Principal Component Analysis Reveals Four Distinct Conformations of the Prefusion HIV-1 SOSIP Env Trimer
The prefusion closed trimer remained stable during the simulation with an average root-mean-square deviation (RMSD) of 3.0 ± 0.5 Å when averaged over 500-ps snapshots and with only marginal fluctuations of the secondary structure. To extract the fundamental motions of the trimer, we carried out a principal component analysis (PCA) on the structured region of the trimer (
). The first two principal components accounted for 46% of the total variance (Figure S1) and corresponded collectively to scissoring movements between individual protomers (Figure 1B). The projection of all snapshots into this eigenspace revealed that, during the course of the simulation, the HIV-1 Env trimer sampled several conformations. Using a mean shift clustering algorithm, we identified four distinct clusters with RMSDs between clusters of 3–4 Å (as measured between the centroid of each cluster) (Figure 1B). Transitions between the different clusters occurred on the 0.5- to 1-μs timescale.
To quantify the differences between clusters and corresponding scissoring motions, we computed the distance distributions measured between the center of mass of the α2 helix (residues A335 to F352) of gp120 for each pair of protomers (Figure 1C). We refer to the distance between protomers, x and y, as dx-y. The average distance between all pairs of protomer atoms was slightly greater than the one measured for starting X-ray structure, 91.99 ± 2.97 Å and 88 Å, respectively. None of the models from the molecular dynamics simulation were perfectly symmetrical, and the d1-3distance was always the largest. The conformations sampled in the first and fourth clusters were more symmetrical. In the second and third clusters, the center of the distribution d1-3shifted up to ∼97 Å, while d2-3remained closer together (∼88 Å), leading to even more asymmetric structures.
To gain further insight into these conformational changes, we performed a second PCA on the gp120 region of known experimentally defined trimer structures as well as of the molecular dynamic centroids (Figure 2A). In this second analysis, the first principal component captured the scissoring motion of the trimer and the second principal component captured a symmetric opening motion of the trimer. These two principal components accounted for 52% of the total variance (Figure S2andMovie S1). It should be noted that many of the experimental structures have been determined by X-ray crystallography, with a single protomer in the asymmetric unit, and are thus perfectly symmetrical (principal component 1 close to 0). The PGT151-bound PDB:5FUUstructure (
) shows a degree of scissoring similar to that observed with clusters 1 and 4, whereas clusters 2 and 3 displayed considerably more scissoring than that induced by PGT151. At the same time an opening of the trimer, most apparent with the single-CD4-bound PDB:5U1Fstructure (
), was also apparent with clusters 3 and 4 (Figure 2A). A comparison of the RMSDs for the experimental structures (Figure 2B) and the molecular dynamics-generated models (Figure 2C) indicated the occurrence of both scissoring and opening motions.
Figure 2Protomer-Scissoring and Trimer-Opening Movements Observed in Molecular Dynamics Are Consistent with Experimentally Determined Structures
Glycan Dynamics Alter Access to the CD4-Binding Site
To quantify the effect of the protomer conformational changes on the glycan shielding of the CD4-binding site, we extracted and analyzed structures associated with each of the four clusters. We computed the average occupancy of glycans around the HIV-1 SOSIP Env trimer for each cluster using VMD's volmap plug-in (Figure 3A). This showedN-linked glycans in the molecular dynamics simulation to be highly mobile, to extend up to 25 Å from the surface of the trimer, and to form a protective layer around the entire trimer (Figure 3B). The glycan density around structures from cluster 1 was only partially defined (Figure 3B). This was most likely due to a combined effect of a small cluster (representing ∼10% of all structures) and the movement of the glycans still equilibrating at the beginning of the simulation. Several glycans close to the N332 super site occupied a well-defined volume and could be clearly identified, in particular N301, N331, N362, and N392.
Figure 3One Out of Three CD4-Binding Sites on the Dominant Conformation of the HIV-1 SOSIP Env Trimer Is Substantially Free from Glycan Shielding
An absence of glycan occupancy was observed around one CD4-binding site in cluster 4, creating a hole in the shielding. To quantify how this hole might modulate access to the CD4-binding site, we estimated accessibility to the CD4-binding site based on the maximum radius of a sphere that did not intersect with any atom of the HIV-1 SOSIP Env trimer (protein and glycans) (Figure 3C). We used probe radii ranging from 1 to 10 Å, the latter corresponding approximately to the radius of a CD4 footprint (Figure 3C). The maximum probe size at each position was averaged for all structures within a cluster. The measured accessibility was different between clusters. In cluster 4, only one CD4-binding site was accessible to a spherical probe with a 10-Å radius (Figure 3D). The increased accessibility of this CD4-binding site appeared to be due to the absence of neighboring glycans close to the CD4-binding site. In clusters 2 and 3, none of the CD4-binding sites were freely accessible (Figure 3D). The small scissoring motion observed as the principal movement in the simulation was reminiscent of PGT151-induced recognition that led to a substantial change in glycan shielding of the CD4-binding site. For more symmetrical conformations (cluster 4), at most only one CD4-binding site was accessible to a 10-Å probe radius at any given time, whereas no CD4-binding site for any of the asymmetric conformations in clusters 2 and 3 were accessible when using the same 10-Å probe radius. Thus protomer scissoring appeared to combine with glycan movements to shield the CD4-binding site, with only a single binding site on the HIV-1 SOSIP Env trimer being sterically available to bind CD4 at any given time. Such asymmetric accessibility may play a role in the observed propensity of HIV-1 Env to bind a single CD4 as the first step for the HIV-1 entry pathway (
Betweenness Centrality of Highly Conserved Glycans and CD4-Binding Site Accessibility
We carried out a network analysis on the dense array of glycan interactions on the surface of the trimer. Each glycan was represented as a node in the graph with the edge connecting two nodes weighted by the inverse of the average non-bonded energy (van der Waals and electrostatic) between the two glycans. One network was built for each cluster and a separate network was built to represent the average of all four clusters.
Betweenness centrality represents the number of shortest paths that pass through a node and is an important indicator of the influence of a node within the network. We used changes in betweenness centrality as a means to identify changes in the interactions between glycans. We observed a cluster of highly conserved glycans (N156, N295, N301, and N332) around the CD4-binding site to become less central when shielding of the CD4-binding site was increased (Figure 4). These observations were consistent when comparing the accessible CD4-binding site within the same trimer or from trimer models residing in different clusters. When the cluster of highly conserved glycans became less central, they interacted less with their neighbors, extended toward a neighboring protomer, and reduced accessibility to the CD4-binding site.
Figure 4Betweenness Centrality of Highly Conserved Glycans and Their Impact on the Accessibility of the CD4-Binding Site
Glycan Betweenness Centrality and High-Mannose Character of the Glycan Shield
In addition to analyzing changes in betweenness centrality, we also computed the median betweenness centrality for each glycan over all three protomers and over the entire course of the simulation (Figure 5A). Notably, glycans with high median centrality were almost all high mannose. Furthermore, when the median betweenness centrality was compared with the impact of the removal of that glycan on the resultant oligomannose abundance of the glycan shield (
), a negative trend was observed for all glycans with the exception of N262 (Figure 5B). Glycan N262 is known to be different, interacting substantially with the surface of the Env but not with other glycans (
). These results suggest that glycans with high median betweenness centrality display a high degree of interaction with other glycans and are less accessible to glycan-processing enzymes; these highly central glycans appear to be essential in maintaining the mannose character of the glycan shield.
Figure 5High Median Betweenness Centrality for Key Glycans Preserving the High-Mannose Character of the Glycan Shield
). The glycans composing each microdomain did not appear to depend on protein conformation, but instead remained similar between all four clusters as well as for the average network. In total, four microdomains were identified for each HIV-1-Env protomer (Figure 6A). The high-mannose patch close to the N332 super site was divided into two microdomains (gp120 outer domains I and II). The gp120/gp41 interface was composed of one glycan microdomain. The apex also formed a single glycan microdomain, shared by all protomers. Experimentally defined glycan-processing profiles were generally similar within a microdomain (
), suggesting that the accessibility to cellular processing was somewhat homogeneous across microdomains, with microdomains on the gp120-outer domain being the most protected.
Figure 6Glycans Form Four Stable Microdomains, and Broadly Neutralizing Antibodies Often Target the Interfaces between Microdomains
Although the dense packing of glycans at the surface of the gp120-outer domain can preventN-glycan processing, broadly neutralizing antibodies can generally pierce this dense array and bind to glycopeptide epitopes on the Env trimer surface (
). To characterize how the recognition of broadly neutralizing antibodies related to the identified glycan microdomains, we used antibody overlap analysis (
) to quantify the glycan microdomain (or domains) that interacted with each broadly neutralizing antibody (Figure 6B).
Apex-interacting antibodies generally interacted with apex glycans spanning multiple protomers, with protruding loops reaching the Env protein epitope by penetrating through the glycan-free region at the very tip of the trimer apex. The gp120-outer domain-interacting antibodies all showed overlap with at least two glycan microdomains (Figure 6B). These broadly neutralizing antibodies preferentially recognized the interface between glycan microdomains, where they generally recognized the Env protein surface (examples are shown inFigure 6C, with a more complete set of antibodies given inFigure S3). The gp120/gp41-interacting antibodies overlapped with glycans from a single microdomain; they appeared to have sufficient accessibility to the Env protein surface to allow direct antibody-protein interaction, while still binding to the generally dispersed complex glycans that cover the gp120/gp41 interface.
Thus HIV-1 Env glycans formed stable microdomains, with broadly neutralizing antibody preferentially targeting the interfaces between the apex, gp120 outer domain I, and gp120 outer domain II microdomains. When antibodies did not recognize the interface between domains, they generally recognized holes at the trimer apex or next to dispersed glycans. Meanwhile, antibody 2G12, which only recognizes glycan, not Env protein surface (
), recognized glycans that had a high median betweenness centrality and were evenly distributed between the two gp120 outer domain microdomains (N295, N332, N339, N386, and N392) (Figure S4). Thus, while the interface recognition appeared to be driven in part by the recognition of the protein surface between glycan microdomains, the recognition by 2G12 suggested that the cross-microdomain binding might have other advantages.
Discussion
As is the case for other type 1 fusion machines, the HIV-1-Env trimer is a conformational engine that uses energy from structural rearrangements and receptor binding to power the fusion of viral and target membranes (reviewed in
), movement on the microsecond or fractional microsecond level might also affect Env mechanisms of entry and evasion. In this study, we investigated these faster timescales with a 2-μs molecular dynamics simulation of the fully glycosylated HIV-1 SOSIP Env trimer.
Substantial protomer movements were observed, averaging ∼3 Å (Figure 1). These were dominated by a scissoring motion, which resulted in trimer asymmetry, with trimer-opening movements also observed (Figure 2). Similar scissoring was induced by the binding of antibody PGT151 (
) (Movie S1), thus indicating a dominant aspect of the molecular dynamics-observed movement to be reflected in the experimental data. During the molecular dynamics, the glycans formed a protective coat extending up to 25 Å from the protein surface. Previously, specific and stable glycan interactions representing possible low-energy states of the glycan shield have been reported (
); while preserving the nature of these interactions, the glycans also explored a number of different conformations during the molecular dynamics.
In terms of entry, the fractional microsecond movements appeared to restrict the accessibility of the CD4-binding site through a combination of trimer scissoring and glycan occupancy (Figure 3). Such alteration of the receptor-binding site accessibility may play a role in defining the initial contact of CD4 with the Env trimer, which occurs through two Env protomers extending outward from the trimer three-fold axis to engage a single CD4 molecule (
). Indeed, our results suggested that, at any given time, the protein surface associated with only one of the three CD4-binding sites of the Env trimer would be sterically accessible to CD4.
In terms of evasion, flexing around the trimer axis may diminish immune response to potential inter-protomer (or inter-domain) epitopes, in the same way that flexing of the capsid with flaviviruses has been observed to reduce antibody recognition of inter-protomer and inter-domain epitopes (
). Thus flexing may be a significant means of immune evasion at the less glycosylated central regions of the trimer, whereas steric occlusion by glycan may be dominant at the highly glycosylated peripheral regions of the trimer.
The plethora of interactions generated by all-atom molecular dynamics simulations may confound the identification of biologically relevant features. Here we used network analysis to identify a number of salient features regarding the glycan shield. The substantial changes in betweenness centrality over the course of the simulation allowed us to identify glycans that hindered access to the Env protein surface. We also found glycans of high median betweenness centrality to be important for the maintenance of the high-mannose character of the glycan shield. Furthermore, network analysis permitted us to identify four glycan microdomains, whose organization appeared to be stable over the time course of the simulation.
Broadly neutralizing antibodies seemed to recognize preferentially the interface between glycan microdomains. The interface-preferred targeting of glycan microdomains by broadly neutralizing antibodies may reflect requirements needed by antibodies to recognize the protein surface, and highlights the superior shielding of glycans within a stable microdomain. Interestingly, the CD4 receptor appeared to represent an extreme example of this, binding at the interface of multiple microdomains and using the decreased bulk of its single-headed immunoglobulin recognition to avoid almost completely interaction with glycan (Figure 6B). The collective behavior of glycans at the surface of the HIV trimer is an important property of the glycan shield that needs to be considered in the context of both CD4 binding and antibody engagement.
Further information and requests for data related to this work should be directed to and will be fulfilled by the Lead Contact, Peter D. Kwong (pdkwong@nih.gov)
Method Details
Molecular Dynamics Simulations
A molecular dynamics simulation was carried out in order to characterize the fully glycosylated BG505–SOSIP Env trimer under physiological conditions. The X-ray structure 4TVP (
).N-linked Man-5 were modeled withGlycosylator, an in-house program. The program first identified the glycan species that were crystallographically resolved at each sequon. Mannose moieties were then added or removed from these glycans to create a Man-5. Afterwards, the trimer was solvated in a 17Å padding water box and neutralized by the addition of NaCl at a concentration of 150 mM. The final system was composed of about half a million atoms and measured 172 x 165 x 170 Å3.
The molecular dynamics simulation was performed using NAMD2.9 engine (
). The periodic electrostatic interactions were computed using particle-mesh Ewald (PME) summation with a grid spacing smaller than 1 Å. Constant temperature was imposed by using Langevin dynamics with a damping coefficient of 1.0 ps. Constant pressure of 1 atm was maintained with Langevin piston dynamics, 200 fs decay period and 50 fs time constant. During equilibration, the trimer backbone atoms were restrained with harmonic restraints (force constant: 1 kCal/mol/Å2). The system was first minimized by 5000 conjugate gradient steps and then equilibrated by using a linear temperature gradient, which heated up the system from 0 to 310 K in 5 ns. An additional 10 ns were done before removing all restraints. The length of all bonds involving hydrogen atoms was constrained with the RATTLE algorithm, thus allowing a time step of 2 fs. Unrestrained molecular dynamics were performed up to 2.0 μs.
Principal Component Analysis
The principal component analysis was carried out using the GROMACS analysis toolkit (
). Each glycan corresponded to one node in the graph. It has previously been shown that a correlation exists between the number of crystallographically defined glycan units and the number of sequons within a 50 Å sphere (
). Therefore, any nodes, which sequons were within 50 Å of each other, were connected in the graph. The average non-bonded energy (van der Waals and electrostatic) was measured with VMD. The edges between glycans that interacted with less than 0.1 kcal/mol were removed. Finally, each edge was weighted with the inverse of the interaction energy.
Glycan-Antibody Overlap Analysis
We sought to determine the number of glycan atoms from each microdomain that occupied the same volume that would be potentially occupied by an antibody. We considered 15 broadly neutralizing antibodies targeting the HIV-1 Env trimer:3U2S(PG9)(
). Each co-crystallized structure was aligned to the MD trajectory. After superposition, all glycan atoms from the MD trajectory within 3.0 Å of the antibody structure were assigned to their microdomain.
Molecular Representations
All molecular representations were rendered with VMD and UCSF Chimera.
Author Contributions
Conceptualization, T.L., C.S., and P.D.K.; Resources, T.L., C.S., and P.D.K; Investigation, T.L. and C.S.; Formal Analysis, T.L. and C.S.; Writing – Original Draft, T.L., C.S., J.S., and P.D.K.; Writing – Review & Editing, T.L., C.S., J.S., and P.D.K.
Acknowledgments
We thank William F. DeGrado (UCSF) and members of the Structural Biology Section and Structural Bioinformatics Core Section for discussion and comments on the manuscript. Molecular dynamics simulations were carried out on Extreme Science and Engineering Discovery Environment (XSEDE) (TACC's Stampede system: MCB080011) and NIH HPC center (Biowulf2). T.L. acknowledges the support of NIH ( GM54616 ) and SNSF ( PA164691 ) grants. Support for this study was provided by the Intramural Research Program of the Vaccine Research Center , NIAID, NIH.
Movie S1. Trimer Transformations, Related to Figure 1
Tube representation for the gp120 domains of the symmetric BG505 SOSIP trimer (PDB:4TVP) are shown with the antibody PGT151-bound asymmetric trimer (PDB:5FUU) and the single-CD4 bound PDB:5U1F. Also shown are the similarity between cryo-EM-determined5FUUstructure and centroid 1 of the molecular dynamics simulation.
Optimization of the additive CHARMM all-atom protein force field targeting improved sampling of the backbone phi, psi and side-chain chi(1) and chi(2) dihedral angles.
Hagberg, A.A., Schult, D.A., and Swart, P.J. (2008). Exploring network structure, dynamics, and function using NetworkX. In Proceedings of the 7th Python in Science Conference (SciPy2008). 11–15.
Complete epitopes for vaccine design derived from a crystal structure of the broadly neutralizing antibodies PGT128 and 8ANC195 in complex with an HIV-1 Env trimer.
Acta Crystallogr. D Biol. Crystallogr.2015;71:2099-2108
The broadly neutralizing anti-human immunodeficiency virus type 1 antibody 2G12 recognizes a cluster of alpha1-->2 mannose residues on the outer face of gp120.
30252-6&id=gr1.jpg){kind=link}
30252-6&id=gr2.jpg){kind=link}
30252-6&id=gr3.jpg){kind=link}
30252-6&id=gr4.jpg){kind=link}
30252-6&id=gr5.jpg){kind=link}
30252-6&id=gr6.jpg){kind=link}
