Identification of possible binding modes of SARS-CoV-2 spike N-terminal domain for ganglioside GM1

Graphical abstract

Abstract

Coarse-grained molecular dynamics simulations of the lipid bilayer mixture of POPC and cholesterol were carried out in the presence and absence of ganglioside monosialo 1 (GM1) with N - terminal domain (NTD) of SARS-CoV-2 spike glycoprotein. The interactions of GM1 with two different NTD orientations were compared. NTD orientation I compactly bind GM1 predominantly through the sialic acid and the external galactose moieties providing more restriction to GM1 mobility whereas orientation II is more distributed on the lipid surface and due to the relaxed mobility of GM1 there, presumably, the NTD receptor penetrates more through the membrane.

1. Introduction

The gangliosides, minor yet essential components of biological membranes, constitute approximately 2–5 % of total membrane lipids and are sequestered with other sphingolipids, cholesterol, saturated - chain phosphatidylcholines [1], [2]. Gangliosides may affect the lateral distributions of lipids in the membrane [3]. They participate in many physiological processes such as membrane fusion, viral budding, and protein sorting [4]. The most widely known ganglioside is the negatively charged ganglioside monosialo 1 (GM1) containing N-acetylneuraminic acid (Neu5Ac) or sialic acid, N-acetylgalactosamine (NAG), two galactoses, and one glucose moiety in its head group. An important structural feature of hydrogen bonding through the head groups provides extensive lipid-ganglioside and protein-ganglioside interaction [5].

Besides ganglioside being populated in the human plasma membrane of neurons [6], the sialic acid (Neu5Ac) linked to the head group of ganglioside is used by many viruses and bacteria as a receptor for cell entry [7], [8]. This phenomenon is well established for the influenza virus, polyomavirus, rotavirus, bovine coronavirus (BCoV), vibrio cholera, etc., but remains elusive for the severe acute respiratory syndrome coronavirus (SARS-CoVs), however, sialic acids are highly expressed on lung epithelial cells [9]. Our current motive is to understand the dynamics of the GM1-containing membrane in driving the spike glycoprotein of SARS-CoV-2 responsible for viral attachment and penetration into the mammalian plasma membrane. For the study, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), cholesterol (CHOL), and GM1 composing the ternary lipid bilayer mixture were selected to mimic the biological membrane.

The spike glycoprotein’s N - terminal domain (NTD) binds preferentially to gangliosides relative to other sphingolipids to penetrate the plasma membrane [10]. However, the approach of CoVs to utilize sialic acid linkages as sites of attachment and invasion in human host cells has not been sufficiently investigated. To our knowledge, only a few computer simulation studies support any direct or indirect interaction of SARS-CoV-2 with ganglioside GM1 in biological membranes [10], [11], [12]. Previous reports claimed that this NTD region is perceived by gangliosides on the lipid raft and must be considered the objective of neutralization of antibodies in vaccine strategies [10], [11] where a linear ganglioside binding domain (GBD) was identified that covers 52 residues of NTD stretching from D111–S162, is a characteristic combination of aromatic and basic residues. In particular, the QFN triad (Q-134, F-135, and N-137) residues at the tip of this domain interact with the sugar moiety of GM1 with CH-π stacking and electrostatic interactions [10], [11]. This indicates that NTD might be a potential receptor for gangliosides and its probable binding sites need to be further studied.

Thus the present study investigates the interaction of GM1 with two different orientations of the NTD sequence of the SARS-CoV-2 spike glycoprotein. Currently, intracellular processes including lipid bilayers such as particle insertion, and other dynamic phenomena occurring at large time and length scales are difficult to deal with at an atomic level in most existing resources so they require coarse-grained (CG) models [13]. Toward this goal, we ran a series of CG molecular dynamics simulations of NTD placed on a membrane bilayer containing 5 $\%$ GM1 for a long 10 μs. Indeed, to explore only the role of cholesterol in deriving the trans-membrane spike protein, simulations were also conducted in the absence of GM1 using a binary lipid bilayer mixture of POPC/CHOL. The CG simulation results show that GM1 acts as an antenna to capture NTD of SARS-CoV-2 whereas only the presence of cholesterol in the bilayer mixture was devoid of any interaction.

2. Method

2.1. System setup and simulation parameters

Molecular dynamics simulations of the N-terminal domain (NTD) of the SARS-CoV-2 spike glycoprotein (residues 13 – 305, retrieved from PDB ID: 6vsb using the UCSF Chimera supplemented with Modeller) [14] were performed in the bilayer consisting of a ternary mixture POPC/CHOL/GM1 molar ratio 7:2.5:0.5 and binary mixture POPC/CHOL 7:3 (Table S1) [2]. A coarse-grained representation of the NTD receptor was prepared using the martinize.py script [15] and a lipid system was built with the insane.py script [16]. The lipid bilayer was so structured that the GM1 cluster was present on the upper leaflet, with its head groups protruding onto the membrane. The Supplementary Fig. S1 of the modelled membrane depicts the alignment of the GM1 head groups with reference to CG Martini bead mapping of GM1 ganglioside [17] by positioning a GM1 molecule over the lipid bilayer along the z-axis. The protein was placed above the lipid bilayer along the z-axis so that the initial distance between the receptor and each GM1 was at least 10 nm. We used a CG water bead [18] model and the system was neutralized with 0.15 M NaCl. The simulation box size was 16 × 16 × 18 ${\mathit{nm}}^3$. All simulations were performed using GROMACS version 5.1.4 [19], with martini force-field version 2.0 [15], [17], [20]. Energy minimization of the system was done using the steepest descent (1000 steps) followed by 10 steps of polak - ribiere conjugate gradient algorithm. The temperature of the system was kept constant at 300 K using the v-rescale algorithm [21] with a coupling constant (${\tau }_t$) of 1.0 ps. The semi-isotropic pressure coupling was sustained at 1 bar independently in the bilayer plane and perpendicular to the bilayer using berendsen barostat [22] with ${\tau }_p=$ 4.0 ps and compressibility of 4.5×${10}^{-5}{\mathrm{b}\mathrm{a}\mathrm{r}}^{-1}$. The relative dielectric constant in water was set to 15 [23] because of the absence of partial charges in the standard martini water model [24]. The system was equilibrated for 100 ns followed by a production run in NPT ensemble using parrinello − rahman barostat [25] (${\mathrm{\tau }}_{\mathrm{p}}=12\mathrm{p}\mathrm{s}\mathrm{a}\mathrm{n}\mathrm{d}$ compressibility = 3 × ${10}^{-4}{\mathrm{b}\mathrm{a}\mathrm{r}}^{-1})$and v-rescale thermostat using 20 fs time step. Periodic boundary conditions were maintained in all directions. Two simulations of 10 μs each were carried out at different initial orientations of the receptor (Fig. S1). Also, an independent replica was generated each for the chosen orientations of the receptor on the ternary mixture with random velocity. The lipid bilayers POPC/CHOL/GM1 and POPC/CHOL without the receptor were simulated as controls and we analyzed a total of 80 μs trajectory in this study. The trajectory analysis was done with GROMACS, visually inspected using VMD 1.9.3 software [26] and the plots were created using Matlab R2019b.

3. Results and discussion

3.1. Effect of SARS-CoV-2 NTD on lipid bilayer properties

The POPC/CHOL/GM1 and POPC/CHOL lipid bilayers were constructed along the XY plane, so with the Z axis intersecting the membrane perpendicularly. The bilayer thickness was estimated from the distance between the two peaks of the POPC phosphate head group in the density profile. Without the NTD receptor, after 10 μs simulation the thickness of the POPC/CHOL/GM1 bilayer was observed to be 4.23 nm, whereas, upon interaction with NTD, the thickness of the bilayer was marginally shrunk to 4.18 nm and 4.06 nm for orientation I and II. The interaction of NTD did not alter the area per lipid (A_L) of the bilayers in any system as compared to the control. It was consistent with 0.51 nm² for both (with/without NTD) bilayer systems. Here, one of the trajectories for both orientations was reported independently. The data analysed from the replicas were consistent with the shown trajectories and not included in this manuscript. The 10 μs simulation trajectory for the 7:3 POPC/CHOL bilayer did not show any interaction with NTD positioned at either orientation (Fig. S2) and we did not further analyse the trajectories.

3.2. Interactions between protein and lipid components

3.2.1. Number of contacts versus time

To investigate the interactions between NTD and the three lipid components POPC, CHOL, and GM1, we calculated the number of contacts using a cut-off $\le$ 0.6 nm throughout the 10 μs trajectory (Fig. 1 a & 1b). The contacts were defined with reference to every bead of NTD and GM1, PO₄ for POPC, and ROH for cholesterol. In the simulations, GM1 gathered around the NTD compared with POPC and cholesterol. After the systems attain the local equilibrium at 2 μs (NTD, orientation I) and 3 μs (NTD, $\mathit{orientation}\mathit{II})$ respectively, the average number of contacts between NTD and GM1 increased to 51.16 $\pm 5.63$ and 102.47 $\pm$ 5.44. The results do not reveal notable contacts with the other two lipid components.

Fig. 1.

Time evolution of the number of contacts between POPC, CHOL, and GM1 lipids and the spike NTD at (a) orientation I and (b) II with respect to the lipid bilayer membrane. Preferential partitioning matrix of the POPC/CHOL/GM1 bilayers: (c) without the receptor, (d) with the receptor (NTD orientation I), and (e) with the receptor (NTD orientation II). If any bead of lipid or receptor in the second group has contact with multiple beads in the first group then we counted it as only one contact.

3.2.2. Preferential partitioning of receptor and membrane lipids

The preferential partitioning of receptor and the lipids was computed as the relative number of contacts of an individual with each of the other components, normalized for the total number of components in the system [27], [28]

$$P_A=\frac{\frac{C_A}{n_A}}{{\sum }_x\frac{C_x}{n_x}}$$

where $P_A$ refers to the preferential partitioning with component A, $C_A$ is the number of contacts with component A, and $n_A$ is the number of molecules of component A. Two molecules within 0.6 nm were regarded to be in contact. We estimated the preferential partitioning from the last microsecond simulation time for both sets and averaged over it. From Fig. 1c, GM1 exhibited the highest preference for interaction with itself ($P_A=0.8791$) manifesting the formation of clusters. The preferential partitioning of GM1 with POPC was considerably lower ($P_A$ = 0.0661) and it marked the lowest for CHOL. Interestingly, CHOL shows more preference toward POPC ($P_A=0.5065)$ and vice-versa ($P_A=0.4595)$ than their self-association to rarely form clusters. Although the preferential partitioning of the membrane components prevailed finitely similar in the presence of the spike NTD, it is noteworthy that NTD accounts for the highest preference of interaction in the orientation I with the GM1 ganglioside$P_A$ = 0.9742 than orientation II with$P_A$ = $0.8907$ resulted due to the formation of large clusters of GM1 responsible for captivating the SARS-CoV-2 spike NTD (Fig. 1d & 1e) followed by POPC ($P_A$ = 0.0222 and $0.0682)$ and cholesterol ($P_A$ = 0.0036 and $0.0411).$ Slight changes in the structure of sphingolipids (including gangliosides) significantly modify their partitioning between lateral domains in the biological membrane [29], this is also evident here due to dominating NTD-GM1 interaction as compared to GM1 for interaction with itself ($P_A=0.8209\mathrm{a}\mathrm{n}\mathrm{d}0.7717).$

3.3. Influence of the ganglioside GM1 in spike NTD binding

Fig. 1 depicted GM1 to be the capture lipid for the coronavirus spike NTD receptor. Previous reports have also shown GM1 to be crucial in engaging simian virus 40 (SV40) [30], cholera toxin [8], [31], serotonin1A receptor [28], etc. Thus, in order to understand the lipid-receptor interaction stability we computed the center-of-mass (COM) distance between the local GM1 lipid (any GM1 bead within 0.5 nm of protein) and the binding residues of NTD (any residue within 0.5 nm of GM1). Fig. 2 a shows that GM1 achieves the local equilibrium around spike NTD after 2 μs simulation time and appeared to be stable around the receptor throughout the 10 μs simulation with an average COM distance of 0.74 $\pm$ 0.06 nm and 0.71 $\pm$ 0.09 nm for NTD orientation I and II respectively. Further, the influence of spike NTD binding on the lateral mobility of lipids POPC, cholesterol, and GM1 in the bilayer system can be quantitatively estimated from their translational diffusion constants, D, calculated from the dependence of mean-square displacement (MSD) on time according to Einstein’s equation [32]. The lateral diffusion coefficient was found to be maximum for CHOL followed by POPC in the control system as well as the systems with the NTD. However, the lateral mobility of GM1 was quite interesting to observe in each of the systems. We noticed that the lateral diffusion coefficient is significantly lower for the GM1 in the case of the NTD orientation I than II (Table 1 ), that is the orientation I of NTD on the bilayer surface provides more restriction to GM1 mobility after virus attachment to the host membrane. Additionally, our study shows that the binding of NTD through the other site in orientation II ameliorates the lateral motion of GM1 lipid components. This was further entrenched with the radial distribution function (RDF) calculations. The data were averaged over the simulation trajectory's initial and final 1 μs. In the case of the first NTD orientation, with the evolution of simulation time, it was observed that within the first shell the peak height for local GM1 lipids around NTD at 0.5 nm increased significantly from 4.8 to 31.0 (Fig. 2b), however, the same peak for another orientation increased from 1 to 17. Thus, the probability of finding local GM1 at 0.5 nm away from NTD is higher for orientation I than for orientation II.

Fig. 2.

(a) The center-of-mass (COM) distance between spike NTD receptor and GM1 ganglioside all through the 10 μs MD simulation and (b) radial distribution function g(r) of local GM1 around the NTD receptor orientation I & II in POPC/CHOL/GM1 bilayer. The pairs of black & red lines and blue & green lines indicate the initial and the final 1 μs simulation time. The peaks are smoothed using MATLAB R2019b.

Table 1.

Lateral diffusion coefficient ($\times {10}^{-7}{\mathit{c}\mathit{m}}^2/\mathit{s}$) of different lipid components calculated from the last 1 μs trajectory.

System	Lipid components
	POPC	CHOL	GM1
Without NTD	2.37 $\pm$ 1.20	3.17 $\pm$ 1.53	0.23 $\pm$ 0.89
With NTD (orientation I$)$	2.22 $\pm$ 0.06	2.41 $\pm$ 0.18	0.61 $\pm$ 0.80
With NTD ($\mathit{orientation}\mathit{II})$	2.71 $\pm$ 0.21	3.15 $\pm$ 0.13	1.60 $\pm$ 0.31

3.4. Residue-wise fractional occupancy of NTD around GM1

To analyze the interacting sites of spike NTD, we calculated the fractional occupancy in terms of a fraction of contacts with respect to each amino acid residue around GM1 over 10 μs of simulation (Fig. 3 ). In the first orientation of NTD, the highest occupancy was observed mainly at the part of the NTD tip $\beta 9-\beta 10$ loop region consisting of V143, Y145, K147, N148, N149, K150, S151, W152, and S155 followed by other $\beta 9$ residues E132, Q134, F135, N137 (residues of QFN triad), and $\beta 13$ P230, I231, and G232 around GM1. This binding mode is specific to the GBD stretch, and the QFN triad is highly conserved among SARS-CoV-2 variants, which has been suggested as a therapeutic and vaccine target as previously mentioned in studies by Fantini et al. [10], [11], [33]. However, in the second case, relatively lower occupancy was noticed for the interacting residues i.e., R21 to T33, F59, $\beta 2-\beta 3$ loop region residues H69, S71, G72, and N74, residues of $\mathrm{\beta }6$ loop K97 and $\beta 12$ strand N211, L212, V213, and R214. Klinakis et al. [34] identified residues of $\beta 9-\beta 10$ and $\beta 14-\beta 15$ (L230-A251) loops as an epitope for convalescent plasma containing neutralizing antibodies. They also revealed that $\beta 3-\beta 4$ is not directly engaged with antibody residues but their members A67, H69, and D80 stabilize the epitope $\beta 14-\beta 15$ loop through interloop interactions.

Fig. 3.

Fractional occupancy of amino acid residues from orientation I and II of NTD around ganglioside GM1 over 10 μs simulation of protein-lipid bilayer complexes.

Chi et al. [35] characterized the 4A8 antibody in convalescent Covid-19 patients that recognize a discontinuous epitope of the spike receptor NTD residues 144 to 158. Concerning this, our findings contribute to predicting the efficacy of ganglioside binding in the pocket observed by the NTD orientation I system, which is consistent with an earlier report [33] on the inhibition activity of the 4A8 antibody in the spike protein interaction with membrane gangliosides.

3.5. Lipid head groups distribution

Since the interactions of sialic acid (Neu5Ac) with viruses are predominant among all the head groups of GM1 [36], the distribution of individual lipid head groups of GM1 around the NTD receptor must be identified in unprecedented detail. We evaluated the probability density function (PDF) of the standard normal distribution at the values in the number of contacts possessed by each of the GM1 lipid head groups explicitly with the receptor over the 10 μs simulation time (Fig. 4 a & 4c). A broader distribution was obtained for the external Gal and Neu5Ac with $\mu \pm \sigma$ of 13.83 $\pm$ 6.79 and 11.02 $\pm$ 5.42 in the case of NTD orientation I followed by the GalNAc. The other two sugar moieties internal Gal and Glc show narrow distributions with higher probabilities of lesser contacts. However, all the head groups of GM1 show broader distributions around the orientation II of NTD with external Gal 18.49 $\pm$ 9.72 and Neu5Ac 20.33 $\pm$ 10.95 topping the list. Particularly, Q134, F135, N137, V143, Y145, and G232 in orientation I of NTD and T33, F59, G72, K97, N211, L212, V213, and R214 in orientation II were observed to be the key interacting residues with sialic acid. The mean (µ) and standard deviation ($\sigma$) of the normal distribution PDF for the obtained number of contacts between the different GM1 head groups and spike receptor NTD is shown in Supplementary Table S2. The dominating accumulation of external Gal and Neu5Ac could be clearly seen in the snapshot taken at the ultimate frame of the 10 μs trajectory (Fig. 4b & 4d). The higher density of GM1 sugars is probably due to their more compactness around the orientation I of NTD.

Fig. 4.

The normalized probability density function (PDF) of the contact number between individual GM1 lipid head groups and spike receptor NTD (a) orientation I and (c) orientation II over the entire 10 μs simulation. (b & d) The snapshots taken at the ultimate time frame of 10 μs highlight the accumulation of GM1 head groups around NTD orientation I and II respectively. The backbone and the side chain of the receptor are shown in mauve and yellow beads and the GM1 head groups as glucose (Glc) in pink, internal galactose (internal Gal) in ochre, N-acetyl-galactosamine (GalNAc) in blue, external galactose (external Gal) in red, and sialic acid (Neu5Ac) in green beads.

4. Conclusions

Gangliosides residing on the outer leaflet of the plasma membrane play a vital role in captivating protein virus receptors. Their close association with cholesterol help in controlling the fusion process [37]. We show here that GM1 binds to the SARS-CoV-2 spike N-terminal domain (NTD) by performing coarse-grain molecular dynamics simulations of the receptor in membrane bilayers with varying orientations of the receptor. The gangliosides’ interaction with the NTD receptor results in a binding site predominating with the β9 − β10 loop region (V143 – S155) and the QFN triad observed from orientation I, fits well on the GM1 head group sialic acid and external galactose. The involvement of this QFN triad was previously identified by Fantini et al. [10], [11] in GM1 binding. Our CG-MD study shows maximum occupancy of $\beta 9-\beta 10$ region around GM1 than the binding site observed with the NTD orientation II (residues R21 to T33, $\beta 2-\beta 3$ loop region, and $\beta 12$ strand). NTD becomes more compact around GM1 with a higher preferential partitioning value in the orientation I system. However, orientation II is more distributed on the lipid surface and due to the relaxed mobility of GM1 there compared to orientation I, we speculate that the spike NTD receptor is likely to penetrate through the membrane with this orientation. Thus, employing two different orientations of spike protein NTD receptor, we propose two binding sites, one with the tip containing the QFN triad and the other with the widely distributed mode of orientation II containing sugar receptor-interacting residues. Monitoring the binding orientations of the SARS-CoV-2 spike protein, which is responsible for protein-lipid interaction, we believe, will shed more light on receptor binding abilities in the cell membrane and provide molecular targets for a therapeutic strategy.

CRediT authorship contribution statement

Tanushree Das: Conceptualization, Methodology, Software, Validation, Investigation, Data curation, Visualization, Writing – original draft. Chaitali Mukhopadhyay: Conceptualization, Methodology, Software, Validation, Supervision, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Data availability

All the data are available in the manuscript and the associated supplementary file.