As of June 29, 2020 more than 10 Million people have been confirmed to be infected with coronavirus disease 2019 (COVID-19) which is caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). This zoonotic pandemic has disrupted society worldwide on a peacetime-unprecedented scale. It also spurred a wide range of scientific endeavors to attack the various aspects of this disease. As the disease spreads there is a critical need for tools that enable the strategic design of biopharmaceutical countermeasures. We are here addressing computationally a molecular approach to aid in the design of a specific class of potential COVID-19 countermeasures.

The genomic sequence of the virus responsible for COVID-19, SARS-CoV-2, was made available in January 2020 [1], providing critical information on the primary amino acid sequences of potential targets. A particularly important target is the SARS-CoV-2 spike (S) protein that is responsible for the first step in the viral infection process, binding to human cells via the angiotensin converting enzyme 2 (hACE2) receptor. The conserved expression and interaction of ACE2 indicates a wide range of hosts (human and non-human) for SARS-CoV-2 [2]. The S protein contains two domains S1 and S2 on each monomer. It is a homotrimer with each monomer comprised of 1281 amino acids. The monomers are expected to be highly glycosylated with 22 N-linked glycosylation sequons and 4 O-linked predicted glycosylation sites [3], although only 16 N-linked glycosylation sites were observed in a cryo-EM map of S produced in HEK293F cells [4]. Very recently, Watanabe et al. performed site-specific glycoform analysis of full-length trimeric S protein made recombinantly in transfected HEK293F cells [5]. Their analysis showed high occupancy at all 22 sites, with about 14 sites classified as complex, 2 sites as oligomannose, and the remaining sites containing mixtures of oligomannose, hybrid and complex glycan structures. Seven of the sites with complex glycoforms, including the 2 sites on the RBD, also had a high degree (>95%) of core fucosylation. Viral coat proteins are often glycosylated which helps pathogens evade the host immune system, modulate access of host proteases, and can enhance cellular attachment through modification of protein structure and/or direct participation at the viral coat protein/cell receptor interface. These glycans are, however, only partially resolved in the experimental structure such that a computational approach is helpful to predict their behavior.

The human ACE2 protein is a 788 amino acid integral membrane protein with seven N-linked glycosylation sites in the extracellular domain. The binding kinetics between the SARS-CoV-2 spike protein and the hACE2 receptor will depend on the 3D structures of both molecules and their molecular interactions which may be impacted by glycosylation [6–8], as has been observed for other glycosylated viral spike proteins and their human receptors. Knowledge of the spike protein and ACE2 amino acid sequences have led to the commercial availability of the spike protein, ACE2, and various fragments of these, with and without purification/fusion tags, produced recombinantly in various expression hosts including Human embryonic kidney cells (HEK293), insect cells, Chinese Hamster Ovary cells (CHO), and E. coli. While the availability of recombinant sources for S and ACE2 glycoproteins have greatly contributed to our understanding of the structure and interactions between these proteins, it is important to recognize that glycosylation of recombinant S and ACE proteins will depend on the host cell [9], the recombinant protein, as well as production [10, 11] and purification methods [12]. As molecular models and molecular dynamics simulations can describe the interactions of proteins with glycans and the modulation of protein structure by glycans [13, 14] they are powerful tools to assess the significance of glycosylation on 3D structure and binding site interactions between the SARS-CoV-2 spike protein and the human ACE2 receptor, and to design novel biotherapeutics including optimizing glycosylation.

A promising strategy for the design of COVID-19 therapeutic proteins is a fusion of the extracellular domain of ACE2, the human receptor for SARS-CoV-2, with the Fc region of human immunoglobulin, IgG1, by a linker separating the two domains [15]. The neutralization strategy behind ACE2-Fc is shown in Fig 1 [15]. This therapeutic design is often called an immunoadhesin, a chimeric protein combining the ligand-binding region of the cell receptor with the constant domain of an immunoglobulin [16]. These chimeric molecules form dimers through disulfide bonds between Fc domains; this bivalency increases the affinity for the ligand. The human ACE2 receptor has been shown to be the primary receptor that SARS-CoV-2 uses for entry into and infection of human cells [17, 18], although the binding site is distinct from the catalytic domain of ACE2. With an ACE2-Fc immunoadhesin the ACE2 portion can act as a circulating “bait or decoy” to bind the SARS-CoV-2 spike protein preventing it from entering human cells while the Fc region confers longer circulatory half-life, provides effector functions of the immune system to clear the virus, and allows simple well-established purification using Protein A affinity chromatography. Immunoadhesins are a distinct class of antivirals that can be used prophylactically as well as post-infection and differ from both vaccines and antibodies. Unlike vaccines, they are not intended to elicit an immune response to the viral infection, and unlike antibody therapies, their design is greatly simplified since once the cellular receptor for viral entry is identified the immunoadhesin can be quickly designed and produced.

This strategy also precludes the coronavirus mutating to escape binding with the ACE2-Fc protein, as it would also lose binding affinity to the native ACE2 human cell receptor resulting in a less pathogenic virus. The re-emergent SARS-CoV-1 virus in 2003–2004 had a lower affinity for human ACE2 resulting in less severe infection and no secondary transmissions [19]. In this strategy the exogenous ACE2 would compensate for decreased ACE2 levels in the lungs during infection, contributing to the treatment of acute respiratory distress, and potentially reduce inflammation and reactive oxygen species in the lung [20]. Most importantly, recombinant ACE2-Fc fusion proteins, with native ACE2 catalytic activity as well as a mutant version with lower ACE2 catalytic activity, produced using transfection of HEK293, have shown high affinity binding to the SARS-CoV-2 spike protein and to potently neutralize SARS-CoV-2 in vitro [21]. Simulations are an ideal tool to optimize such a construct and guide the experimental production of ACE2-Fc.

Glycans are branched, flexible chains of carbohydrates that explore a much wider range of conformations at equilibrium conditions than the protein chain itself as the latter is typically not dynamically changing strongly from its folded form as that would affect its functionality. The faster dynamics of glycans complicates the structural and conformational characterization of glycans in laboratory experiments [22]. In atomistic molecular dynamics (MD) simulations, glycan conformations can be straight-forwardly analyzed to obtain structural information, as glycan dynamics are much closer to the computational timescale than the protein dynamics. However, neighboring glycans can interact with each other and essentially lock each other in which can lead to very slow equilibration into the correct conformation [13]. Therefore, algorithms are needed to generate realistic glycan configurations as glycans are regularly not fully resolved experimentally. Consequently, only a few simulations of related fully glycosylated proteins available [23–26] among them recently a proposed glycosylated model of the Spike protein [27]. Very recently a short simulation of the Spike protein with glycosylation has been published which is enabling longer studies [28]. Our group has made significant progress in the field of glycan modeling in recent years [13, 14, 29].

N-glycan structure is highly heterogeneous, and the relative abundance of glycans depends on the expression system for glycoprotein production. Plant-based transient expression systems are well-suited to produce recombinant ACE2-Fc under the current COVID-19 pandemic given high production speeds. Two glycovariants of ACE2-Fc are simulated in this work: one is targeted for ER retention with high mannose glycoforms, and the second is targeted for secretion with plant complex glycoforms. These glycovariants are currently being expressed and purified at UC Davis.

In order to properly understand the interaction between the spike protein and the variant ACE2 receptors bound to its fusion partner the glycosylation of both entities needs to be taken into account. The few existing computational studies of ACE2 interaction with the spike protein we are aware of are using aglycosylated proteins [30–32]. Also, molecular docking studies have been performed with the older related SARS-CoV-1 virus protein implicated in the SARS epidemic in the early 2000s [33]. We develop in silico models to predict the 3D structure of two glycosylated ACE2-Fc variants. Additionally, we evaluate the interactions between these two ACE2-Fc variants and a glycosylated spike protein fragment (SpFr) which contains the receptor binding domain of the SARS-CoV-2 spike protein.

Fig 4 shows the starting configurations generated as described above (left) and the configurations after MD for 75 ns (right) of all simulated systems. Systems of this size will not fully equilibrate in 75 ns, but evidence of structural stability and concerted motion can still be observed. This is in agreement with a recent equilibration study of a fully glycosylated Spike protein [28]. All systems exhibit varying length of the flexible linker domain between ACE2 and Fc during simulation. The domain separation can be quantified by analyzing the center of mass distance between the ACE2 and Fc ordered domains, shown in S2 Fig in S1 File. AF^M8 and AF^M8/SpFr exhibit clearly more shortening of the linker domain than AF^GG and AF^GG/SpFr, possibly due to the difference in glycosylation in the Fc domain, which is closest to the flexible linker region. AF^M8/SpFr has the shortest distance between the ACE2 and Fc domains after 75 ns, which is consistent with its final configuration shown in Fig 4. In the AF^M8/SpFr and AF^GG/SpFr systems, ACE2 glycans near the ACE2/SpFr interface form contacts across the binding interface, indicating that glycosylation may significantly affect binding kinetics. Additionally, glycans on SpFr that were initially oriented away from the protein are reoriented towards ACE2 after 75 ns. The structure of the ordered domains of ACE2-Fc and SpFr appear to retain structural stability. As expected, the glycans, on the other hand, show significant reorientation, as the configurational dynamics of glycans is faster than proteins [13].