3  Results

3.1  Structural and conformational analysis of CLQ and CLQ-OH in water
The chemical structures of CLQ and CLQ-OH are shown in Fig. 1 (a,b). The only difference between the two molecules is the presence of a terminal hydroxyl group in CLQ-OH. This OH group has a marked influence on the conformation and water-solubilization properties of the drug. CLQ-OH may adopt a wide range of conformations, the most stable being the extended one shown in Fig. 1(c). When immersed in a periodic box of 31.5 Å2 with 1042 water molecules, the system reached, at equilibrium, an estimated energy of interaction of -92 kJ.mol−1, accounting for 56 water molecules solvating CLQ-OH Fig. 1(d). In contrast, due to an intramolecular hydrophobic effect, CLQ appeared to be more condensed than CLQ-OH Fig. 1(e). At equilibrium, CLQ was surrounded by 58 water molecules with an energy of interaction of -79 kJ.mol−1 Fig. 1(f).
Fig. 1 Chemical structure of chloroquine (CLQ) and hydroxychloroquine (CLQ-OH). (a) CLQ. (b) CLQ-OH. (c) CLQ-OH extended conformer. (d) CLQ-OH in water. (e) Typical condensed conformer of CLQ. (f) CLQ in water. The molecules in (c–f) are shown in either tube or sphere rendering (carbon, green; nitrogen, blue; oxygen, red; hydrogen, white). In (c) and (e), the chlorine atom of CLQ and CLQ-OH is indicated by an arrow.
These water-compatible conformations of CLQ and CLQ-OH were used as initial conditions for studying the interaction of these drugs with sialic acids and gangliosides.

3.2  Sialic acids as molecular targets of CLQ and CLQ-OH
Neu5Ac is the predominant sialic acid found in human glycoproteins and gangliosides. When CLQ was merged with Neu5Ac, a quasi-instantaneous fit occurred between the two molecules, whose global shapes in water are geometrically complementary Fig. 2 (a). This is particularly obvious in the views of the CLQ–Neu5Ac complex in mixed surface/balls and sticks rendition Fig. 2(a,b). The interaction was driven by the positioning of the negative charge of the carboxylate group of Neu5Ac and one of the two cationic charges of CLQ (pKa 10.2) Fig. 2(c). The energy of interaction of this complex was estimated to be -47 kJ.mol−1. As coronaviruses preferentially interact with 9-O-acetyl-N-acetylneuraminic acid (9-O-SIA) [10], this study used a similar molecular modelling approach to assess whether CLQ could also interact with this specific sialic acid. A good fit between CLQ and 9-O-SIA was obtained Fig. 2(d–f), with an energy of interaction of -45 kJ.mol−1. In this case, the carboxylate group of the sialic acid interacted with the cationic group of the nitrogen-containing ring of CLQ (pKa 8.1) Fig. 2(d). The complex was further stabilized by OH-π and van der Waals interactions.
Fig. 2 Molecular modelling of chloroquine (CLQ) interaction with sialic acids. (a,b) Surface representation of the CLQ–sialic acid (Neu5Ac) complex. Two opposite views of the complex are shown. Note the geometric complementarity between the L-shape conformer of CLQ dissolved in water (in blue) and Neu5Ac (in red). (c) Neu5Ac bound to CLQ via a combination of CH-π and electrostatic interactions with one of the cationic groups of CLQ (+). (d) Molecular modelling of CLQ bound to N-acetyl-9-O-acetylneuraminic acid (9-O-SIA). From right to left, the dashed lines indicate a series of van der Waals, OH-π and electrostatic contacts with both cationic groups of CLQ (+). (e,f) Surface representations of the CLQ–9-O-SIA complex.
Next, CLQ-OH was tested to assess whether it could, as CLQ, bind to 9-O-SIA (Fig. 3 ). The complex obtained with CLQ-OH was very similar to that obtained with CLQ [compare Fig. 3(a,b) with Fig. 2(e,f), although several conformational adjustments occurred during the simulations. Interestingly, the OH group of CLQ-OH reinforced the binding of CLQ to sialic acid through establishment of a hydrogen bond Fig. 3(c,d). Overall, this hydrogen bond compensated for the slight loss of energy caused by the conformational rearrangement, and the energy of interaction of the complex was estimated to be -46 kJ.mol−1, which is very close to the value obtained for CLQ (-45 kJ.mol−1).
Fig. 3 Molecular modelling of hydroxychloroquine (CLQ-OH) interaction with sialic acids. (a,b) Surface representation of CLQ-OH bound to N-acetyl-9-O-acetylneuraminic acid (9-O-SIA). Two opposite views of the complex are shown. Note the geometric complementarity between CLQ-OH (in blue) and 9-O-SIA (in red). (c,d) Molecular mechanism of CLQ-OH binding to 9-O-SIA: combination of electrostatic interactions and hydrogen bonding.

3.3  Molecular recognition of gangliosides by CLQ and CLQ-OH
In the respiratory tract, sialic acids are usually part of glycoproteins and gangliosides. Molecular modelling approaches were used to assess whether CLQ and CLQ-OH can recognize sialic acid units in their natural molecular environment. In these simulations, ganglioside GM1 was chosen as a representative example of human plasma membrane gangliosides. A first series of simulations was performed with CLQ. When merged with the ganglioside, CLQ had two distinct binding sites, both located in the polar saccharide part of GM1. The first site was located at the tip of the saccharide moiety of the ganglioside Fig. 4 (a,b). The energy of interaction was estimated to be -47 kJ.mol−1. CLQ retained the typical L-shape structure of the water-soluble conformer bound to isolated sialic acids [compare Figs 2(c) and 4(a)]. From a mechanistic point of view, the carboxylate group of the sialic acid of GM1 was oriented towards the cationic groups of CLQ. The rings of CLQ faced the N-acetylgalactosamine (GalNAc) residue of GM1, establishing both OH-π interaction and hydrogen bonding Fig. 4(b). The second site was in a large area including both the ceramide–sugar junction and the saccharide moiety Fig. 4(c). The chlorine atom of CLQ was oriented towards the ceramide axis, allowing the nitrogen-containing ring of CLQ to stack on to the pyrane ring of the first sugar residue [i.e. glucose (Glc)]. The perfect geometric complementarity of the two partners Fig. 4(c,d) accounted for a particularly high energy of interaction in this case (-61 kJ.mol−1). Interestingly, there was no overlap between the two CLQ-binding sites on GM1, so the ganglioside could accommodate two CLQ molecules Fig. 4(e), reaching a global energy of interaction of -108 kJ.mol−1. A similar situation was observed with CLQ-OH, which occupies the same binding site as CLQ Fig. 4(f). In this case, the energy of interaction was further increased by stabilizing contacts established between the two CLQ-OH molecules, reaching -120 kJ.mol−1. Overall, these data showed that CLQ and CLQ-OH have a good fit for sialic acids, either isolated or bound to gangliosides.
Fig. 4 Molecular modelling simulations of chloroquine (CLQ) and hydroxychloroquine (CLQ-OH) binding to ganglioside GM1. The surface electrostatic potential of GM1 indicates a non-polar, membrane-embedded part corresponding to ceramide (white areas), and an acidic part protruding in the extracellular space corresponding to the sialic-acid-containing saccharide part (red areas). (a) CLQ bound to the tip of the carbohydrate moiety of GM1. (b) Molecular mechanism of CLQ–ganglioside interactions. (c) Molecular dynamics simulations revealed a second site of interaction. In this case, the aromatic cycles of CLQ are positioned at the ceramide–sugar junction, whereas the nitrogen atoms interact with the acidic part of the ganglioside (not illustrated). (d,e) Surface views of GM1 complexed with one (d) or two (e) CLQ molecules (both in blue), illustrating the geometric complementarity of GM1 and CLQ molecules. (f) One GM1 molecule can also accommodate two distinct CLQ-OH molecules simultaneously, after slight rearrangement allowing increased fit due to CLQ-OH/CLQ-OH interactions. To improve clarity, CLQ-OH molecules bound to GM1 are represented in two distinct colours (blue and green).

3.4  Structural analysis of the NTD of SARS-CoV-2 S protein
The next step of this study was to determine how SARS-CoV-2 could interact with plasma membrane gangliosides, and whether such interaction could be affected by CLQ and CLQ-OH. The global structure of the SARS-CoV-2 S protein [16] is shown in Fig. 5 (a–d). It consists of a trimer of S proteins, each harbouring two distinct domains distant from the viral envelope: the receptor-binding region (RBD) and the NTD.
Fig. 5 Structural features of the SARS-CoV-2 spike (S) protein. (a) Trimeric structure (each S protein has a distinct surface colour, ‘blue’, ‘yellow’ and ‘purple’). (b) Ribbon representation of ‘blue’ S protein in the trimer (α-helix, red; β-strand, blue; coil, grey). (c) Surface structure of the ‘blue’ S protein isolated from the trimer. (d) Ribbon structure of the ‘blue’ S protein. (e) Zoom on the N-terminal domain (NTD) of the ‘blue’ S protein. (f,g) Molecular model of a minimal NTD obtained with Hyperchem [ribbon in representation in (f), surface rendering in (g)]. (h) Highlighting of the amino acid residues of the NTD that could belong to a potential ganglioside-binding domain.
It was reasoned that if the RBD is engaged in functional interactions with the ACE-2 receptor, it would be interesting to search for potential ganglioside-binding sites on the other cell-accessible domain of the S glycoprotein (i.e. the NTD). The NTD contains approximately 290 amino acid residues. The tip of the NTD was of particular interest, as it displays a flat interface Fig. 5(f) ideally positioned for targeting a ganglioside-rich plasma membrane microdomain, such as a lipid raft.
The amino acid sequence of the planar interfacial surface located at the tip of the NTD was analysed for the presence of consensus ganglioside-binding domains [20]. These motifs are constituted by a triad of mandatory amino acid residues such as (K,R)-Xn-(F,Y,W)-Xn-(K,R). The Xn intercalating segments, usually four to five residues, may contain any amino acid, but often Gly, Pro and/or Ser residues. The strict application of this algorithm did not allow the detection of any potential ganglioside-binding domain in this region of the NTD. However, an intriguing over-representation of aromatic and basic residues was found in the 129–158 segment: 129-KVCEFQFCNDPFLGVYYHKNNKSWMESEFR-158. This 30-amino acid stretch also contains Gly, Pro and/or Ser residues that are often found in ganglioside-binding motifs. These observations supported the notion that the tip of the NTD could display a large ganglioside-attachment interface.

3.5  Molecular interactions between gangliosides and the NTD of SARS-CoV-2 S protein
Molecular dynamic simulations of a structural motif encompassing amino acid residues 100–175 of the NTD Fig. 5(f–h) merged with ganglioside GM1 further supported this concept. As shown in Fig. 6 , the large flat area of this structural domain fitted very well with the protruding oligosaccharide part of the ganglioside. Several amino acid residues appear to be critical for this interaction, especially Phe-135, Asn-137 and Arg-158 (Table 1 ). Overall, the complex involved 10 amino acid residues for a total energy of interaction of -100 kJ.mol−1. At this stage, it was observed that approximately 50% of the interface was involved in the complex, leaving the remaining 50% available for interaction with a second GM1 molecule. As expected, merging a second GM1 molecule with the preformed GM1–NTD complex led to a trimolecular complex consisting of two gangliosides in a typical symmetrical chalice-like structure into which the NTD could insert its interfacial ganglioside-binding domain (Fig. 7 ). The formation of this trimolecular complex was progressive, starting with a conformational rearrangement of the first ganglioside–NTD complex triggered by the second GM1 molecule. The energy of interaction of the new complex was consistently increased by 37%, reaching an estimated value of -137 kJ.mol−1. At this stage, attachment of the NTD to the ganglioside-rich microdomain involved the whole interface (i.e. 15 surface-accessible residues from Asp-111 to Ser-162). The critical residues were Asp-111, Gln-134, Phe-135, Arg-158 and Ser-161 (Table 1).
Fig. 6 Molecular complex between the N-terminal domain (NTD) of SARS-CoV-2 spike protein and a single GM1 ganglioside. The NTD is represented in ribbons superposed with a transparent surface rendering (light green). Two symmetric views of the complex are shown (a,b). The amino acid residues Q-134 to D-138 located in the centre of the ganglioside-binding domain are represented as green spheres. The saccharide part of the ganglioside forms a landing surface for the tip of the NTD.
Table 1 Energy of interaction of each amino acid residue of SARS‐CoV‐2 spike protein in contact with GM1 molecules.
Amino acid residues Energy of interaction (kJ.mol‐1)
First step: one GM1 molecule
 Asp‐111 −5.6
 Lys‐113 −8.2
 Gln‐134 −8.6
 Phe‐135 −20.1
 Cys‐136 −7.0
 Asn‐137 −15.2
 Asp‐138 −6.4
 Arg‐158 −17.4
 Ser‐161 −9.7
 Ser‐162 −2.0
 Total −100.2
Second step: two GM1 molecules
 Asp‐111 −15.8
 Ser‐112 −10.7
 Lys‐113 −9.2
 Gln‐134 −11.2
 Phe‐135 −10.5
 Cys‐136 −6.2
 Asn‐137 −4.7
 Phe‐140 −5.2
 Gly‐142 −5.6
 Glu‐156 −9.0
 Phe‐157 −13.8
 Arg‐158 −19.8
 Tyr‐160 −3.2
 Ser‐161 −9.7
 Ser‐162 −2.0
 Total −136.6
Fig. 7 Molecular complex between the N-terminal domain (NTD) of SARS-CoV-2 spike protein and a dimer of GM1. In (a), (c) and (d), the NTD is represented as in Fig. 4. In (b), the surface of the NTD is shown without any transparency. The amino acid residues Q-134 to S-162 belonging to the ganglioside-binding domain (GBD) are represented as green spheres. Compared with a single GM1 molecule, the dimer of gangliosides forms a larger attractive surface for the NTD. In the above view of (d), the anchorage of the NTD to the gangliosides is particularly obvious. As chloroquine also interacts with the saccharide part of GM1, its presence would clearly mask most of the landing surface available for the NTD, preventing attachment of the virus to the plasma membrane of host cells.

3.6  Potential coordinated interactions between SARS-CoV-2 and the plasma membrane of a host cell: key role of gangliosides in lipid rafts
Taken together, these data strongly support the concept of a dual receptor/attachment model for SARS-CoV-2, with the RBD domain being involved in ACE-2 receptor recognition, and the NTD interface responsible for finding a ganglioside-rich landing area (lipid raft) at the cell surface.
Such a dual receptor model, consistent with the topology of the SARS-CoV-2 S protein, is proposed in Fig. 8 . With this model in mind, the potential effects of CLQ and CLQ-OH were studied, both of which, according to the molecular modelling data, have a high affinity for sialic acids and gangliosides.
Fig. 8 Dual recognition of gangliosides and angiotensin-converting enzyme-2 (ACE-2) by SARS-CoV-2 spike (S) protein. The viral protein displays two distinct domains, the tips of which are available for distinct types of interactions. The receptor-binding domain binds to the ACE-2 receptor, and the N-terminal domain (NTD) binds to the ganglioside-rich domain of the plasma membrane. Lipid rafts, which are membrane domains enriched in gangliosides (in yellow) and cholesterol (in blue), provide a perfect attractive interface for adequately positioning the viral S protein at the first step of the infection process. These structural and molecular modelling studies suggest that amino acid residues 111–162 of the NTD form a functional ganglioside-binding domain, the interaction of which with lipid rafts can be efficiently prevented by chloroquine and hydroxychloroquine.

3.7  Molecular mechanism of CLQ and CLQ-OH antiviral effect: preventing SARS-CoV-2 S protein access to cell surface gangliosides
With the aim of establishing whether CLQ and CLQ-OH could prevent the attachment of SARS-CoV-2 to plasma membrane gangliosides, the initial NTD–GM1 complex was superposed with a drug–GM1 complex (Fig. 9 ). To improve clarity, the ganglioside is not presented in Fig. 9. This superposition shows that the NTD and the drug (CLQ-OH in this case) share the same spatial position when bound to GM1, so GM1 cannot bind the viral protein and the drug simultaneously. This is due to the fact that the NTD and the drugs (CLQ and CLQ-OH) bind to GM1 with a similar mechanism controlled by a dyad of functional interactions: a hydrogen bond and a geometrically perfect CH-π stacking interaction. In the case of the NTD, the hydrogen bond involves Asn-167, whereas CH-π stacking is mediated by the aromatic ring of Phe-135 Fig. 6(b). On one hand, Asn-167 establishes a network of hydrogen bonds with the GalNAc residue of GM1. On the other hand, the flat aromatic ring of Phe-135 stacks on to the cycle of the Glc residue of GM1. In the case of CLQ and CLQ-OH, it is the nitrogen-containing ring of the drug that stacks on to the Glc ring Fig. 4(c). Note that both the Phe-135 (in red) and CLQ-OH (in green) rings are located in the same position (Fig. 9). The other CLQ-OH molecule, which covers the tip of the sugar part of the ganglioside, interacts with the GalNAc ganglioside Fig. 4(b). When the NTD is bound to the ganglioside, the side chain of Asn-137 is found in this exact position (Fig. 9). Thus, once two CLQ-OH (or two CLQ) molecules are bound to a ganglioside Fig. 4(e,f), any binding of a SARS-Cov-2 S protein to the same ganglioside is totally prevented. The energy required to overcome this steric incompatibility is estimated to be several hundred kJ.mol−1, which is far too high to occur.
Fig. 9 Mechanism of action of hydroxychloroquine (CLQ-OH). The N-terminal domain (NTD) bound to GM1 was superposed onto GM1 in interaction with two CLQ-OH molecules. The models only show the NTD and both CLQ-OH molecules (not GM1, to improve clarity). (a,b) The aromatic ring of F-135 (in red), which stacks onto the glucose cycle of GM1, overlaps the aromatic CLQ-OH rings (in green) which also bind to GM1 via a CH-π stacking mechanism. The N-137 residue of the NTD interacts with the N-acetylgalactosamine residue of GM1, but this interaction cannot occur in the presence of CLQ-OH as this part of GM1 is totally masked by the drug. (c,d) Superposition of the NTD surface (in purple) with the two CLQ-OH molecules bound to GM1, illustrating the steric impossibility that prevents NTD binding to GM1 when both CLQ-OH molecules are already interacting with the ganglioside.

3.8  Sequence alignment analysis of SARS-CoV-2 and related coronavirus: evolution of the ganglioside-binding domain at critical amino acid residues
As CLQ and CLQ-OH are potential therapies for SARS-CoV-2 infection, it is important to check whether the amino acid residues identified as critical for ganglioside binding are conserved among clinical isolates. The alignment of the 111–162 domain of 11 clinical isolates of SARS-CoV-2 from various geographic origins (including Asia and USA) is shown in Fig. 10 . In this region, which contains the ganglioside-binding domain identified in the present report, all amino acids are fully conserved. Interestingly, the motif is built like a giant consensus ganglioside-binding domain: a central region displaying the critical aromatic residue (Phe-135) and a basic residue at each end (Lys-113 and Arg-158). In the middle of each stretch separating this typical triad, there is a N-glycosylation site (Asn-122 and Asn-149). These last regions are not directly involved in ganglioside binding, so the oligosaccharide linked to these asparagine residues could be perfectly intercalated between the sugar head group of gangliosides.
Fig. 10 Amino acid sequence alignments of the ganglioside-binding domain (GBD) of the SARS-CoV-2 spike protein. (a) Clinical SARS-CoV-2 isolates aligned with the reference sequence (6VSB_A, fragment 111–162). The amino acid residues involved in GM1 binding are indicated in red. Two asparagine residues acting as glycosylation sites are highlighted in yellow. (b) Alignments of human and animal viruses compared with SARS-CoV-2 (6VSB_A, fragment 111–162). Deletions are highlighted in green, amino acid changes in residues involved in ganglioside binding are highlighted in blue, conserved residues of the GBD are highlighted in red, and asparagine residues acting as glycosylation sites are highlighted in yellow.
It was also noted that the ganglioside-binding domain of the NTD is fully conserved in bat RaTG13, which indicates a close relationship between the bat coronavirus and the human isolates that are currently circulating around the world. However, the motif is slightly different in other bat- and human-related coronaviruses (Fig. 10), suggesting a recent evolution which could explain, at least in part, why SARS-CoV-2 is more contagious than previously characterized human coronaviruses.