key: cord-0285086-j65t0t1m
authors: Shoemark, Deborah K.; Oliveira, A. Sofia F.; Davidson, Andrew D.; Berger, Imre; Schaffitzel, Christiane; Mulholland, Adrian J.
title: Molecular dynamics of spike variants in the locked conformation: RBD interfaces, fatty acid binding and furin cleavage sites
date: 2022-05-06
journal: bioRxiv
DOI: 10.1101/2022.05.06.490927
sha: 6f17229eecce77062af1b960a27597c5e35b2f2e
doc_id: 285086
cord_uid: j65t0t1m

Since December 2019 the SARS-CoV-2 virus has infected billions of people around the world and caused millions of deaths. The ability for this RNA virus to mutate has produced variants that have been responsible for waves of infections across the globe. The spike protein on the surface of the SARS-CoV-2 virion is responsible for cell entry in the infection process. Here we have studied the spike proteins from the Original, Alpha (B.1.1.7), Delta (B1.617.2), Delta-plus (B1.617.2-AY1), Omicron BA.1 and Omicron BA.2 variants. Using models built from cryo-EM structures with linoleate bound (6BZ5.pdb) and the N-terminal domain from 7JJI.pdb, each is built from the first residue, with missing loops modelled and 45 disulphides per trimer. Each spike variant was modified from the same Original model framework to maximise comparability. Three replicate, 200 ns atomistic molecular dynamics simulations were performed for each case. (These data also provide the basis for further, non-equilibrium molecular dynamics simulations, published elsewhere.) The analysis of our equilibrium molecular dynamics reveals that sequence variation at the closed receptor binding domain interface particularly for Omicron BA.2 has implications for the avidity of the locked conformation, with potential effects on Omicron BA.1 and Delta-plus. Linoleate binding has a mildly stabilizing effect on furin cleavage site motions in the Original and Alpha variants, but has no effect in Delta, Delta-plus and slightly increases motions at this site for Omicron BA.1, but not BA.2, under these simulation conditions.

We focus on the sometimes subtle effects of changes between these variants around the fatty acid binding site in the closed spike trimer, the interface between the closed receptor domains in the trimer and the furin cleavage site. We discuss how these changes may have contributed to altered spectra of symptoms and modulation of infectivity, particularly in the case of the omicron variants.

The SARS-CoV-2 variants have not elicited the same range of symptoms or disease severity as compared to the original virus from Wuhan. In the unvaccinated, there were progressive increases in both R numbers and disease severity from the Original to Alpha, Delta and Deltaplus (2) In the UK vaccinated individuals with residual resistance have found that the Omicron variants, though more infectious (3) (4) (5) , have so far led to proportionally fewer hospitalizations and far fewer deaths. In the UK, which currently has around ~70% of the population vaccinated (2 vaccines, plus one booster) with Pfizer-BioNTech or Oxford-AZ vaccines, the ZOE COVID study (6) has revealed that symptoms, when present, are likely to be more cold-like than for previous variants. 82% of infected people experience runny nose, 70% fatigue, 69% sneezing, 69% headache, with only 53% of people reporting a persistent cough. On the 2 nd of April 2022, an estimated 4.5 million people had symptomatic COVID-19 in the UK (6) . Countries that have largely avoided COVID outbreaks in the past, such as Hong Kong are now experiencing a massively surging omicron outbreak that is claiming lives. In Hong Kong in March 2022, the death rate per 100,000 population (3.44) exceeded that of the Delta outbreak in London in 2020 (3.17) and 67% COVID fatalities in the 5th wave in Hong Kong were unvaccinated (Hong Kong Centre for Health Protection and the Department of Health, GovUK). Figure 1 . The Original ('Wuhan') uncleaved, locked, linoleate-bound spike trimer ectodomain showing the context for the furin cleavage site and changes at the RBD subunit interface occurring in the omicron variants. Glycosylation was removed for clarity. This structure is based on the cryoEM structures 6ZB5 and 7JJI with missing residues modelled from the first residue, and comprised of 45 disulphides per trimer, upon which all other variants in this paper are based. Spike chains are shown in cyan, wheat and magenta ribbon. The residues shown as spheres (circled) illustrate the position of the furin cleavage site. The inset panel is enlarged area showing residues as spheres where the RBD interface is changed in the omicron variants. Spike carbons are shown as spheres and coloured according to their chain, oxygens are in red and nitrogens in blue, linoleate (LA -) carbons are orange.

The ectodomain of the spike protein (see Figure 1 ) is a trimer, each monomer consisting of two domains, S1 and S2. The S1 domain is largely responsible for host interactions and contains a head region: an N-terminal domain (NTD) and a receptor binding domain (RBD) containing the ACE II receptor binding motif (RBM), which is also a major antibody binding epitope (7) .

Our team identified a free fatty acid (FA) binding pocket at the RBD interfaces of the locked SARS-CoV-2 spike that was occupied by the essential fatty acid, linoleate (LA -) in 2020 (8).

Our previous simulations have shown that binding of LAin the FA site helps to stabilize a locked conformation (9) which appears more condensed than the LA-free closed form) and that the carboxylate head group of the FA makes consistent salt-bridge interactions with the R408, occasional interactions with K417 and persistent H-bonding interactions with Q409, across the locked subunit interfaces (10) . Fatty acids have since been retrospectively discovered in the FA sites of spike proteins from SARS viruses that infect other species (11, 12) . The S1 domain of the spike protein is responsible for attaching the SARS-CoV-2 virus to the host cell by means of interaction with host cell receptors. This is predominantly achieved through opening the conformation by raising one or more RBDs to expose the binding site for the human ACEII receptor (13) . Additional host cell interactions have been identified, e.g the exposed, positively charged region resulting from furin cleavage at the S1/S2 junction (see below) binds to the human NRP1 receptor (14) .

The S2 domain is required for membrane fusion events leading to cell invasion via the fusion peptide (FP), the fusion peptide proximal region (FPPR), two heptad repeats (HR1 and HR2) and the transmembrane domain (13, 15) . The S1/S2 domains of the SARS-CoV-2 spike are joined by the furin cleavage site, which is unique among the betacoronaviruses. Cleavage by the human endoprotease furin, facilitates cell infectivity in humans and separates the two domains, revealing a positively charged region at the C-terminus of S1 (16) . There is a further cleavage site in the S2 subunit (S2 ' ) that is a target for the human TMPRSS2 protease and cleavage there allows membrane fusion to occur (16) .

All structures used here are based on the LA-bound, locked, unglycosylated, cryo -EM structure (6BZ5.pdb), using the N-terminal domain (NTD) from the NOVAVAX structure 7JJI.pdb, as more of the loop structure of the NTD was resolved in the latter. All the variants were modelled from this Original structure to maximise comparability of the molecular dynamics simulations. Missing residues of the NTD are modelled from the first residue, missing residues in loops have been modelled throughout and each trimer has 45 disulphide bonds formed. These models supercede the locked spike models we have used previously, which were missing the first 24 residues and had 42 disulphides (9, 10, 17) . Model building and validation methods for models for this study are described in the supplementary material.

These spike systems are relatively large (~600,000 atoms) and here equilibrium simulations were performed over 200 ns for three replicates for each variant. The authors note that extensive and noteworthy simulations have been carried out with glycosylated spike proteins (18) (19) (20) however, for comparing these variants, it was deemed that resolution of protein perturbations within the RBD and throughout the protein may be better achieved in the absence of surfacebound and highly flexible glycans, which may obscure the underlying protein dynamics over this short time-period. All glycosylation sites in the Original spike sequence as described by Zhao et al., (21) | are retained in these variants. These equilibrium simulation data also provide the basis for dynamical non-equilibrium molecular dynamics simulations to map the allosteric pathways arising from the ligand annihilation from the fatty acid binding site in our accompanying paper (22) .

All models were energy minimised and underwent short position-restrained molecular dynamics simulations in order to settle waters and generate three different sets of velocities for the three replicate production dynamics runs of 200 ns each, using GROMACS. Each spike trimer was simulated in a box of explicit waters with 150 mM NaCl, under periodic boundary conditions as an NPT ensemble at 310 K and pH 7, as described previously (17) .

Sequence alignments are shown in Figure S1 This close-up shows the LA -(depicted as stick, orange carbons) with its hydrophobic tail in the fatty acid pocket of RBD subunit A (green ribbon) and its carboxylate atoms interacting with residues (notably R408 and Q409) across the RBD interface on chain C (cyan ribbon). K417 and E406 are also shown for context. K417 makes occasional salt-bridging interactions with the LAcarboxylates, but also interacts with E406 on the same subunit. Mutating K417 to asparagine (variant Delta-plus and Omicrons BA.1 and BA.2) reduces its ability to shield the repulsive negative charge of E406 (and D405) from the carboxylates of the LAwhose hydrophobic tail is bound in the FA site of the opposite subunit. Of all the major variants of concern to date, the Omicron lineages have diverged most from preceding ones with many more mutations throughout the spike protein (3). The FA pocket, in Omicrons BA.1 and BA.2, accommodating the hydrophobic tail of the FA is slightly modified by the substitution of (WT) S373 for proline. However, residues flanking this proline at the RBD subunit interface (see Figure 1 ) may affect the ease with which the RBD can be raised (9) , as discussed later. In Omicron BA.1 the residues that interact with the LAcarboxylate are equivalent to the Original residues R408 and Q409 (numbered R405 and Q406 in Omicron 

The residues in positions equivalent to the Original K417 are N415 (Delta-plus) and N414 Figure S1 ). This stretch lies on the interface between two subunits required to release an RBD to be raised for binding to the ACE2 receptor (9, 23 Figure 3 ). Interaction distances between S373 on one RBD subunit and R403 and Y505 of the neighbouring subunits for the Original spike are shown in Figure 3 . The substitution of these Original serines (S371, S373 and S375) by hydrophobic residues in the Omicron variants precludes hydrogen-bonding interactions from contributing to keeping the RBDs locked at this point. Indeed, under these simulation conditions, this loop can become deflected away from the interface by around 5 Å (Figure 4 ). 

The Original spike furin cleavage site spans residues PRRAR (Original P681-R685). The

Original proline residue is mutated in all the variants discussed here. The furin active site has four negatively charged residues so it is likely that the addition of extra positive charge around the substrate cleavage site may increase binding affinity and/or cleavage rate by furin. The

Original SARS-CoV-2 furin cleavage sequence was not ideal, but the addition of flanking positively charged residues may be making cleavage more efficient (24) . In the alpha variant, the proline is mutated to a histidine, as is the case for Omicron. Histidine on an external loop may also become protonated at low pH. The loss of the proline alters O-glycosylation in this region which has been postulated to affect loop exposure to increase cleavage (25) . Figure S5) . Ttests performed to compare the RMSFs for each furin-cleavage site between their respective apo and LAbound forms can be found in table S1. Comparisons of furin-cleavage-site RMSF data for the Original apo with each of the variant apo forms can be found in Table S2 . Finally furin-cleavage site RMSF data comparisons of the Original LAbound form with the LAbound forms of the variants can be found in Table S3 . The caveat here is that the effects of glycans on these sites should also be considered. Our equilibrium simulations support our previous data (Original spike starting from residue 25) (10) suggesting that LAmay have a weakly stabilising effect on fluctuations at the furin site in the full-length, Original spike. These data suggest this stabilising effect may also occur in the Alpha variant in the presence of LAin the FA site, but not for the Delta, Delta plus or Omicron variants. Indeed, for Omicron BA.1, bound LAmay even increase fluctuations, which may influence the rate of furin cleavage. (Figure S7 and Tables S1 and S3).

The sequence variations of the spike proteins of different variants may affect their respective infectivity, and perhaps disease severity. The initial encounter with the host cell in the oronasal cavity may prove a crucial step in deciding the fate of the infection. The ease with which the spike RBD can open in the absence or presence of a fatty acid in the FA site is probably influenced by differences in directly interacting residues and the intersubunit interactions that hold the trimer closed. Our previous work (9) has shown that greater force is required to open an RBD in the Original spike when LA is bound. This was more pronounced for the BrisΔ (furin-cleavage site deletion). We surmise then that bound LAhelps to reduce the frequency of spike RBD opening and consequentially, once bound and with the RBDs closed, LA is more likely to be trapped within and effectively sequestered. The ability of LAto limit RBD opening when occupying the fatty acid site would at first seem counter-productive to the virus. However if that fatty acid is LAsequestered from membranes of cells lining the oronasal cavity, which provide the first line of defence, this may benefit the virus by helping to subdue an initial, robust response as LApromotes an inflammatory response in the upper airways (26) .

It is perhaps worth noting that linoleic acid cannot be produced by the body, therefore availability within membranes relies on dietary intake and probably varies between individuals.

The relatively small loss of hydrogen-bonding across the RBD subunit interfaces for the Omicron spikes (LYNLAP/LYNFAP for LYNSAS) in conjunction with N414 (Originally K417), may subtly ease RBD opening. We see from these equilibrium simulations that the closed RBD spike interface is less tightly packed in the Omicron variants than for the Original closed spike, suggesting that this subunit interface more readily dissociates to open. If the Omicron variants exhibit a higher frequency of RBD opening due to mutations at the closed RBD interface, then more frequent openings would lead to more frequent, productive, ACEII interactions, hence increased infectivity and the corollary is that being more often open, they may also be less efficient at trapping or sequestering LA from cell membranes of the oronasal cavity. A more sparing effect on LAwould seem to be supported by the observation that Omicron infections elicit more cold-like symptoms in the host (27) than previous variants. This could allow for a more immediate, robust and better contained inflammatory response within the upper airways, especially in those with some pre-existing immunity due to infection by previous variants, vaccination, or both. In Omicron BA.2, LAsequestration may be further hampered by the substitution of the dominating carboxylate interacting residue R408 (WT) for a serine (S405). S405 fails to come within 5 Å of the LAcarboxylates in any of the omicron BA.2 holo simulations and our simulations reveal enhanced LAmotility within the FA binding site, compared to all the other variants in this study. Reinfections have been problematic recently (28) . Serine at the equivalent position to 408 may also potentially compromise the binding of some neutralising antibodies that would otherwise interact with R408 (29) and may facilitate re-infection with BA.2 following infection with BA.1, which is deemed possible but rare, according to a recent Danish study (30) . Perhaps in the case of Omicron BA.2 key mutations resulting in a reduced susceptibility to antibodies may also reduce the need to sequester LA -. A combination of reduced LAsequestration, slightly reduced RBD interface affinity to ease opening, coupled with an escape-enhancing mutation at (WT) 408 could contribute to the increased infectivity, altered symptom spectra and enhanced immune evasion seen with Omicron BA.2.

Delta Variant: What we know about the Science

Mutational and phylogenetic analyses of the two lineages of the Omicron variant

Omicron SARS-CoV-2 variant: a new chapter in the COVID-19 pandemic

Omicron emerges

Epitope Classification and RBD Binding Properties of Neutralizing Antibodies Against SARS-CoV-2 Variants of

Free fatty acid binding pocket in the locked structure of SARS-CoV-2 spike protein

Structural insights in cell-type specific evolution of intra-host diversity by SARS-CoV-2

The fatty acid site is coupled to functional motifs in the SARS-CoV-2 spike protein and modulates spike allosteric behaviour

CoV-2 and bat RaTG13 spike glycoprotein structures inform on virus evolution and furincleavage effects

Bat and pangolin coronavirus spike glycoprotein structures provide insights into SARS-CoV-2 evolution

Structural and Functional Basis of SARS-CoV-2 Entry by Using Human ACE2

Neuropilin-1 is a host factor for SARS-CoV-2 infection

Cryo-EM structure of the 2019-nCoV spike in the prefusion conformation

Proteolytic activation of SARS-CoV-2 spike protein

Molecular Simulations suggest Vitamins, Retinoids and Steroids as Ligands of the Free Fatty Acid Pocket of the SARS-CoV-2 Spike Protein*

Development and simulation of fully glycosylated molecular models of ACE2-Fc fusion proteins and their interaction with the SARS-CoV-2 spike protein binding domain

Dynamic Interactions of Fully Glycosylated SARS-CoV-2 Spike Protein with Various Antibodies

Beyond Shielding: The Roles of Glycans in the SARS-CoV-2 Spike Protein

Glycans of SARS-CoV-2 Spike Protein in Virus Infection and Antibody Production

SARS-CoV-2 spike variants differ in their allosteric response to linoleic acid

Free fatty acid binding pocket in the locked structure of SARS-CoV-2 spike protein

SARS-CoV-2 spike and its adaptable furin cleavage site

Furin cleavage of the SARS-CoV-2 spike is modulated by O-glycosylation

Too much linoleic acid promotes inflammation-doesn't it?

Covid-19: Runny nose, headache, and fatigue are commonest symptoms of omicron, early data show

Rapid emergence of SARS-CoV-2 Omicron variant is associated with an infection advantage over Delta in vaccinated persons

Cross-Neutralization of a SARS-CoV-2 Antibody to a Functionally Conserved Site Is Mediated by Avidity

Occurrence and significance of Omicron BA.1 infection followed by BA.2 reinfection

The data is openly available from MolSSI/BioExcel COVID-19 public data repository for biomolecular simulations (https://covid.molssi.org/simulations/ ).