Roles of variable linker length in dual acting virucidal entry inhibitors on HIV‐1 potency via on‐the‐fly free energy molecular simulations

Abstract

The Dual‐Acting Virolytic Entry Inhibitors, or DAVEI's, are a class of recombinant chimera fusion proteins consisting of a lectin, a flexible polypeptide linker, and a fragment of the membrane‐proximal external region (MPER) of HIV‐1 gp41. DAVEIs trigger virolysis of HIV‐1 virions through interactions with the trimeric envelope glycoprotein complex (Env), though the details of these interactions are not fully determined as yet. The purpose of this work was to use structural modeling to rationalize a dependence of DAVEI potency on the molecular length of the linker connecting the two components. We used temperature accelerated molecular dynamics and on‐the‐fly parameterization to compute free energy versus end‐to‐end distance for two different linker lengths, DAVEI L0 (His₆) and DAVEI L2 ([Gly₄Ser]₂His₆). Additionally, an envelope model was created based on a cryo‐electron microscopy‐derived structure of a cleaved, soluble Env construct, with high‐mannose glycans added which served as putative docking locations for the lectin, along with MPER added that served as a putative docking location for the MPER region of DAVEI (MPER_DAVEI). Using MD simulation, distances between the lectin C‐terminus and Env gp41 MPER were measured. We determined that none of the glycans were close enough to gp41 MPER to allow DAVEI L0 to function, while one, N448, will allow DAVEI L2 to function. These findings are consistent with the previously determined dependence of lytic function on DAVEI linker lengths. This supports the hypothesis that DAVEI's engage Env at both glycans and the Env MPER in causing membrane poration and lysis.

Abbreviations

CV

collective variable

CVN

cyanovirin‐N

DAVEI

dual acting virucidal entry inhibitor

MPER

membrane proximal external region

MVN

microvirin‐N

OTFP

on‐the‐fly parameterization

TAMD

temperature accelerated molecular dynamics

1. INTRODUCTION

The envelope glycoprotein (Env) is the only viral protein displayed on the surface of HIV‐1 virions. Env allows HIV‐1 to target CD4+ T‐cells and to mediate the membrane fusion and entry process. ¹ Env comprises the non‐covalently linked subunits gp120 and gp41 that form a heterotrimeric “spike”. ¹ , ² , ³ , ⁴ The gp120 subunits are highly glycosylated and are positioned completely outside the viral membrane. ¹ , ⁵ , ⁶ The gp120 subunit is the region that interacts with CD4 and coreceptors, leading to conformational changes allowing for host cell infection. ¹ , ² , ⁵ , ⁶ , ⁷ The gp41 subunit anchors the spike to the viral membrane and is thought to contain the machinery needed for fusion of viral and host cell membranes. ¹ , ⁵ , ⁷ Each gp41 subunit includes two helical segments and a membrane proximal external region (MPER) that is likely associated with the membrane of the virus in native Env. ³ , ⁸ After gp120 makes contact with CD4 and the coreceptor, gp41 undergoes a conformational change to form the six‐helix bundle. ⁷ , ⁸ , ⁹ , ¹⁰ This refolding is thought to bring the cell and virus membranes together while releasing energy to overcome the kinetic barrier to fusion. ⁸ , ⁹ , ¹⁰

The dual‐acting virucidal entry inhibitors (DAVEIs) are a class of molecules designed to bind to specific areas on Env to trigger this conformational cascade in the absence of an opposing membrane, causing the viral contents to be expelled into the extra‐cellular space. ³ , ⁴ , ¹¹ DAVEI is an engineered protein chimera designed to engage gp120 and gp41 simultaneously using a glycan binder linked to a short peptide derived from the HIV‐1 MPER sequence by a variable length flexible polypeptide linker [(Gly₄Ser)_xHis₆]. ³ , ⁴ , ¹¹ The original DAVEI molecules used cyanovirin‐N (CVN) as a glycan binder, while second‐generation DAVEI molecules use microvirin (MVN), which has more specific binding to mannose‐9 glycan structures like those on Env. ³ , ⁴ , ¹² , ¹³ , ¹⁴ The MPER‐like region of DAVEI is believed to bind to Env MPER determined by competition assays with MPER specific antibodies (4E10 and 10E8) and alanine mutation studies. ⁴ , ¹¹ The linker is used to provide space between the two functional ends of DAVEI to allow them access to their binding partners. ³ , ⁴ , ¹¹ , ¹⁴ As such, there is a minimal linker length that must be achieved to allow for the two active ends to function coordinately. Previous studies have shown that a repeat unit of zero (L0) is not functional while a repeat unit of two (L2) is functional with IC50's similar to repeat units of four and eight (L4 and L8); thus, L2 is the minimal effective linker length for DAVEI. ⁴

The purpose of this manuscript was to provide a structural rationale for this dependence on linker length using an all‐atom computational model. The model has two main components: (a) molecular dynamics simulation of an Env model derived from the cryo‐electron microscopy‐derived structure of a cleaved, soluble, clade‐B Env construct ¹⁵ that includes modeled‐in glycans and MPER segments that allows distance measurements between them, and (b) free‐energy calculations of the end‐to‐end distance of DAVEI linkers. By combining these measurements for five glycosylation sites and two different linker lengths, we can now propose both a particular glycan and particular linker length that minimally enable DAVEI function. These findings should guide design of future generations of DAVEI.

2. RESULTS AND DISCUSSION

Free energy versus end‐to‐end distances for the L0 and L2 linker peptides, computed using OTFP in all‐atom, explicitly solvated systems, are shown in Figure 1. At physical temperatures, L0 linkers sampled mostly end‐to‐end distances ≈10 ≤ z_j(L0)≤ 16 Å, and for L2, the range was ≈10 ≤ z_j(L2) ≤ 30 Å. There was also a stable shelf for both L0 and L2 at ≈7 Å that represents a configuration where the linker is folded in half and the N‐termini and C‐termini are together. If we allow for a ≈1 kcal/mol fluctuation from the energy well, then the maximum linker length for L0 becomes ≈18 Å and for L2 becomes ≈37 Å.

FIGURE 1.

Free‐energy versus end‐to‐end distance for L = 0 and L = 2 DAVEI linkers from OTFP calculations

To determine the significance of these distances in the context of DAVEI binding to Env, we compared them to the distributions of distances measured between the C‐terminus of docked MVN molecules and the closest Env MPER. These distributions were built from sampling glycan conformations in all‐atom, explicit‐solvent MD simulations of glycosylated, solvated Env constructs based on the 5FUU cryo‐EM structure. Figure 3 shows the root‐mean squared deviation of the glycans in three replica systems. The fact that RMSD's plateaued indicates sufficient sampling of glycan conformations.

FIGURE 3.

Root‐mean squared deviation (RMSD) versus time of all Man9 glycans in three independent MD simulations of all‐atom, explicitly solvated trimeric HIV‐1 Env constructs based on the 5FUU cryo‐EM structure

For glycan‐MPER distance measurements, in all cases, the closest Env MPER segment to a particular gp120 glycan is presented by the gp41 of the protomer located one position counterclockwise (looking down at the trimer apex) from that gp120. Figure 2 illustrates this approach. We measured all 15 of these distances, for three terminal mannobioses over five N‐linked Man9s, over the three symmetry‐related protomers and 1,000 frames from 20 ns explicit‐solvent MD simulations of three replicas. For each glycan and terminal mannobiose, a density distribution function was computed from the raw distance measurements, and only simulation frames in which the docked MVN resulted in no steric clashes with the protein were permitted into the statistics. The resulting densities are plotted in Figure 4. These show that only one of the five glycosylation sites considered, namely N448, has a glycan that places a terminal mannobiose such that the C‐terminus of a docked MVN is within 25 Å of the closest MPER. Based on both the free‐energy calculations on the L0 linker end‐to‐end distance and the distances sampled in the Env MD simulation, there were no instances in which an L0 linker would be able to span the distance from the C‐terminus of a docked MVN to the closest MPER.

FIGURE 2.

Renderings of Envelope A) with glycans (green) added showing Asn (red) residues on a surface trace of Env (shades of gray representing each protomer) and B) showing a representative distance measurement from MVN (cyan) on one Man9 (green) attached to Asn448 (red) to MPER (blue) shown in yellow. The white arrow represents where MVN binds to glycans. The distance represented by the yellow arrow is 40.5 Å

FIGURE 4.

Probability distributions of distances defined by the C‐terminus of a docked MVN at each terminal mannobiose (red, green, and blue) on the Man9 at each N‐linked glycosylation site (N262, N332, N386, N392, and N448) to the center of mass of the modeled‐in MPER sequence from all‐atom MD simulation of a soluble trimeric JR‐FL Env construct based on the 5FUU cryo‐EM structure. Distributions are averages over three protomers and time‐averaged over the 20‐ns MD simulation

This finding provides a plausible structural rationale for the loss in virolytic activity of the lectin‐based DAVEI molecules when moving from the L2 to the L0 linkers. ¹⁴ We did not perform free‐energy calculations for the L4 linkers because of high computational expense. We speculate that an L4‐based DAVEI might be able to use more glycans than just the one at N448, which might explain why L4 has a lower EC₅₀ for virolysis. ⁴ These findings suggest that the model that was examined here could represent the interaction of DAVEI with Env that leads to virolysis. Further development of DAVEI is ongoing including a specific gp120 binder that will have a predictable binding location and building a less flexible linker that can direct the components of DAVEI to their respective active sites. In addition to the correspondence with experimental results mentioned above, it is possible to directly test our predictions using mutagenesis of Env. For example, an N262A mutated Env would remain susceptible to DAVEI binding while an N448A mutated Env would not.

3. CONCLUSIONS

Using all‐atom molecular dynamics and free‐energy calculations via on‐the‐fly parameterization, we calculated the equilibrium end‐to‐end distance distributions for the L0 and L2 linkers used in lectin‐based DAVEIs. We compared this distance distributions to distances measured from MD simulations of glycosylated Env constructs that represent the best estimate of DAVEI interaction site placement. This comparison provided a structural rationale for the lack of activity of L0‐based DAVEIs and the realization of activity for DAVEIs with linkers L2 and larger and showed that L2 functions on N448 glycans. In summary, this modeling effort supports the idea that DAVEI dually engages Env to cause virolysis. Looking ahead, we plan to tailor modeling efforts to prediction of optimal linker chemistries to direct dual engagement and to match particular site identities in future generations of DAVEIs.

4. MATERIALS AND METHODS

4.1. System setup

The starting point of the on‐the‐fly parameterization (OTFP) simulations performed in this work was an all‐atom, explicit‐water system of the linker. Two different linker systems were created L0 [His₆] and L2 [(Gly₄Ser)₂His₆] as straight chains. The psfgen utility of VMD ¹⁶ was used to prepare each system with CHARMM22 parameters and TIP3P waters. ¹⁷ Each system was then minimized and equilibrated to allow the chain to relax for 1 ns using NAMD. ¹⁸ All MD simulations were run using a Langevin thermostat set at 310 K with a damping coefficient of 1 ps⁻¹, a Langevin piston barostat with a period of 100 fs and decay of 50 fs, RATTLE bond‐length constraints, particle‐mesh Ewald electrostatics with a 1 Å mesh spacing, a 12 Å nonbonded cutoff, and a 1‐fs time step. After equilibration, the relaxed linker systems were run using OTFP to generate free energy profiles.

4.2. On‐the‐fly parameterization calculations

OTFP is a method for extracting free‐energy profiles from temperature accelerated molecular dynamics (TAMD) simulations. ¹⁹ , ²⁰ Similar to methods such as metadynamics and adaptive biasing forces, ²¹ , ²² TAMD/OTFP is a collective‐variable based approach; unlike other methods, however, TAMD/OTFP does not rely on a history‐dependent biasing potential. TAMD accelerates sampling of the CV space by using a fictitious “particle” with coordinates z tethered to collective variables θ(x) by a harmonic potential with the spring constant, κ ¹⁹ , ²⁰ , ²¹ , ²² :

$$m_i{x}_i=-\frac{\partial V\left(x\right)}{\partial x_i}+\kappa \sum _j\left[z_j-{\theta }_j\left(x\right)\right]\frac{\partial {\theta }_j\left(x\right)}{\partial x_i}+\mathit{red}\sqrt{2M{\mathit{\gamma \beta }}^{-1}}{\eta }^x\left(t\right)$$ (1)

$$\overline{\gamma }{\dot{z}}_j=-\kappa \left(z_j-{\theta }_j\left(x\right)\right)+\mathit{red}\sqrt{2{\overline{\gamma }\overline{\beta }}^{-1}}{\eta }^x\left(t\right)$$ (2)

Here, V(x) is the interatomic potential energy, κ is a spring constant, η^{x, z}(t) are white noises, and $\overline{\gamma }$ and $\overline{\beta }$ are the fictitious friction and Boltzmann constant, respectively. Atomic coordinates x and fictitious particle coordinates z are updated in lockstep. $\overline{\gamma }$ is chosen large to enforce a time‐scale separation between z and x, where z will have slow motion that is uncoupled from the faster dynamics of x. If κ is taken to be large enough to sufficiently tether the collective variables, Θ_j(x), close to their corresponding z_j values then the forces acting on z_j will self‐average to the negative gradient of free energy:

$$\kappa \left[z_j-{\theta }_j\left(x\right)\right]\approx \frac{\partial F\left(z\right)}{\partial z_j}$$ (3)

Thus, the free energy of the system at the physical temperature T will become the potential energy surface of z moving at higher temperature $\overline{T}$. This allows for accelerated sampling of the CV. But instead of only using the spring forces to update z, OTFP additionally uses them to reconstruct the free energy. In order to obtain the free energy profile, F(z) from the TAMD output, OTFP first considers the decomposition of F(z) into an expansion in basis functions ϕ_m(z):

$$F\left(z\right)=\sum _m{\lambda }_m{\varphi }_m\left(z\right)$$ (4)

The unknown coefficients, λ_m, can be recovered from the negative gradient of free energy by minimizing an error defined by

$$E\left(\lambda \right)=〈\sum _j{\left[\kappa \left(z_j-{\theta }_j\left(x\right)\right)-\frac{\partial }{\partial z_j}\sum _m{\lambda }_m{\varphi }_m\left(z\right)\right]}^2〉$$ (5)

The minimization of this function is reduced to the linear set of equations Aλ = b where

$$A_{\mathit{mn}}=\frac{1}{2t}{\int }_0^t\sum _i\frac{\partial {\varphi }_m\left(z\right)}{\partial z_i}\frac{\partial {\varphi }_n\left(z\right)}{\partial z_i}\mathit{ds}$$ (6)

$$b_m=\frac{1}{t}{\int }_0^t\sum _i\frac{\partial {\varphi }_m\left(z\right)}{\partial z_i}\kappa \left[z_i\left(s\right)-{\theta }_i\left(x\left(s\right)\right)\right]\mathit{ds}$$ (7)

Here, the CV used is the end‐to‐end distance of the linker from the nitrogen on the first amino acid to the carboxylate carbon in the last amino acid. Since this is a one‐dimensional CV, we use one‐dimensional chapeaus as basis functions:

$${\varphi }_m\left(z\right)=\left\{\begin{array}{l}1-m+z/\mathrm{\Delta }z\text{if}m-1<z/\mathrm{\Delta }z<m\\ 1+m-z/\mathrm{\Delta }z\text{if}m<z/\mathrm{\Delta }z<m+1\\ 0\text{otherwise}\end{array}\right)$$ (8)

Each OTFP simulation was run with $\overline{T}$ of 7,000 K, a $\overline{\gamma }$ of 1,500 and a 2‐fs time step. Each L0 system was run for 360 ns of OTFP and each L2 system was run for 600 ns of OTFP in order for the free energy profiles to converge, and three independent replicas of each were run for reproducibility.

4.3. Envelope model setup

The starting point for our Env/DAVEI complex simulations was an all‐atom, vacuum system based on the all‐atom cryo‐EM model of the solubilized JR‐FL Env ectodomain trimer structure (PDB entry 5FUU ¹⁵ ) and the crystal structure of mannobiose‐bound microvirin (PDB entry 2YHH ²³ ). To generate an all‐atom Env model suitable for docking studies with the glycan‐binding MVN, we modeled Man9 glycans onto the asparagine residues 262, 332, 386, 392, and 448. To provide a reference location for the gp41 MPER segments which were not included in the ectodomain trimer, we extended the gp41 structure of each protomer out to residue 682 by modeling missing residues as an alpha‐helical extension of the C‐terminal heptad repeat domain of each gp41.

The psfgen utility of VMD ¹⁶ was then used to prepare the system with CHARMM36 parameters, ²⁴ and to embed the system in a solvent box of TIP3P waters and sufficient counterions to neutralize the system. After a short minimization, a 20‐ns MD simulation via NAMD ¹⁸ was performed. Three independent replicas of the system were created. All MD simulations were run using a Langevin thermostat set at 310 K with a dampening coefficient of 1 ps⁻¹, a dielectric constant of 80, a 10 Å nonbonded cutoff, and a 1‐fs time step. Configurations were saved every 1,000 timesteps. A complete set of scripts for generating the our glycosylated Env model are publicly accessible at github.com/cameronabrams/psfgen.

For each configuration from the MD simulations, a MVN molecule was docked sequentially onto each of the three terminal mannobiose units on each Man9 glycan for the five glycosylation sites, Asn262, Asn332, Asn386, Asn392, and Asn448, for each of the three protomers. The docking was performed by aligning the mannobiose atop the mannobiose in the glycan pocket of MVN. For each docking, the distance from the C‐terminal α‐carbon of MVN to the center of mass of the closest of the three Env MPERs was recorded, but only if the docked MVN did not sterically clash with any part of the protein.

AUTHOR CONTRIBUTIONS

Steven Gossert: Conceptualization; data curation; formal analysis; investigation; methodology; writing‐original draft; writing‐review and editing. Bibek Parajuli: Conceptualization; data curation; validation. Irwin Chaiken: Funding acquisition; project administration; writing‐review and editing. Cameron Abrams: Conceptualization; funding acquisition; investigation; methodology; project administration; resources; supervision; writing‐original draft; writing‐review and editing.

Gossert ST, Parajuli B, Chaiken I, Abrams CF. Roles of variable linker length in dual acting virucidal entry inhibitors on HIV‐1 potency via on‐the‐fly free energy molecular simulations. Protein Science. 2020;29:2304–2310. 10.1002/pro.3949