Chalcones from Angelica keiskei (ashitaba) inhibit key Zika virus replication proteins

Abstract

Zika virus (ZIKV) is a dangerous human pathogen and no antiviral drugs have been approved to date. The chalcones are a group of small molecules that are found in a number of different plants, including Angelica keiskei Koidzumi, also known as ashitaba. To examine chalcone anti-ZIKV activity, three chalcones, 4-hydroxyderricin (4HD), xanthoangelol (XA), and xanthoangelol-E (XA-E), were purified from a methanol-ethyl acetate extract from A. keiskei. Molecular and ensemble docking predicted that these chalcones would establish multiple interactions with residues in the catalytic and allosteric sites of ZIKV NS2B-NS3 protease, and in the allosteric site of the NS5 RNA-dependent RNA-polymerase (RdRp). Machine learning models also predicted 4HD, XA and XA-E as potential anti-ZIKV inhibitors. Enzymatic and kinetic assays confirmed chalcone inhibition of the ZIKV NS2B-NS3 protease allosteric site with IC₅₀s from 18 to 50 μM. Activity assays also revealed that XA, but not 4HD or XA-E, inhibited the allosteric site of the RdRp, with an IC₅₀ of 6.9 μM. Finally, we tested these chalcones for their anti-viral activity in vitro with Vero cells. 4HD and XA-E displayed anti-ZIKV activity with EC₅₀ values of 6.6 and 22.0 μM, respectively, while XA displayed relatively weak anti-ZIKV activity with whole cells. With their simple structures and relative ease of modification, the chalcones represent attractive candidates for hit-to-lead optimization in the search of new anti-ZIKV therapeutics.

Graphical Abstract

1. Introduction

Zika virus (ZIKV) is a pathogen neglected for more than 60 years because most infections were thought to be mild or asymptomatic [1,2]. However, the outbreaks in French Polynesia in 2014 and Brazil in 2015, which led to cases of Guillain-Barré syndrome and microcephaly [1], have brought to light the serious neuropathogenic action of the virus. The consequences of ZIKV infection have now been associated with severe neurological manifestations that can impact fetuses, infants and adults; including, microcephaly, Guillain-Barré syndrome, meningoencephalitis, fetal cerebral calcification, central nervous system alterations, meningoencephalitis, and myelitis [1–4]. Currently, neither antiviral agents nor a vaccine is available for treating or prevention of ZIKV infection.

Two essential ZIKV proteins have been targeted in the search for anti-ZIKV therapeutics; the ZIKV protease and the ZIKV RNA-polymerase (RdRp). The ZIKV protease cleaves the polyprotein into individual ZIKV proteins [2,5], and a number of natural compounds have been identified as ZIKV-protease inhibitors including myricetin [6,7], curcumin [6,8] and pedalitin [9] with IC₅₀ values ranging from 1.3 to 5 μM. The ZIKV NS5 RNA-dependent RNA-polymerase (RdRp), which generates double stranded RNA, has also been identified as a key target for therapeutic development. Several nucleotide-like inhibitors (NIs) of the ZIKV RdRp have been identified such as 2′-C-methylated nucleoside triphosphates [10], 2’-C-methyl- and 2’-C-ethynyl-substituted analog 5’-triphosphates [11], 10-undecenoic acid zinc salt [12] and sofosbuvir triphosphate [13,14]. Therapeutically, non-nucleotide inhibitors (NNIs), which block enzyme activity at the allosteric or “N-pocket”, are more desirable since they are more selective for viral over mammalian targets and display fewer side-effects [14,15]. Only one NNI has been described for the ZIKV RdRp, known as the TBP compound [16] with an IC₅₀ of 94 nM. A number of non-nucleotide RdRp inhibitors have been identified for dengue virus (DENV) [14] and ZIKV [10–13,16] but their activity towards the N pocket has not been confirmed.

The chalcones, also known as benzyl acetophenones, are open chain flavonoids structurally characterized by two aromatic rings interconnected by α,β-unsaturated carbonyl system. The medicinal plant Angelica keiskei Koidzumi also referred to as ashitaba produces over 20 chalcones [17,18]. Ashitaba-derived chalcones display anti-bacterial [18–21] and antiviral [17,18,22,23] activities and have been shown to act as inhibitors of both mammalian [24–26] and microbial proteins [21,23,27] including multiple viral proteins, including the influenza A neuraminidase [23] and the both proteases from SARS-CoV [22]. We hypothesized, therefore, that the chalcones might inhibit key ZIKV proteins and thereby block ZIKV replication. To test this hypothesis, we purified the three chalcones, 4-hydroxyderricin (4-HD), xanthoangelol (XA), and xanthoangelol-E (XA-E) from an ashitaba extract. Then, we performed molecular docking calculations with these chalcones within the OpenZika project [2,28], an open-science collaborative project, from IBM’s World Community Grid, using all ZIKV proteins and more than one binding site per protein, i.e., the catalytic and allosteric sites. From these results, we could indicate that chalcones could interact with ZIKV NS2B-NS3 protease, and the allosteric site of the NS5 RNA-dependent RNA-polymerase (RdRp). Also inside OpenZika project, machine learning models predicted 4HD, XA and XA-E as potential anti-ZIKV inhibitors. Accordingly, we tested the chalcones in enzymatic assays that confirmed these predictions and in vitro ZIKV replication assays revealed anti-ZIKV activity for 4HD and XA-E, demonstrating how integrated in silico and in vitro approaches can be used in the search for new anti-ZIKV agents.

2. Results and discussion

2.1. Chalcone isolation and characterization

Based on studies showing that ashitaba-derived chalcones exert anti-bacterial [18–21] and antiviral effects [17,18,22,23] we sought to purify several chalcones from a root extract of ashitaba. Organic extraction followed by chromatographic separations were used to purify the chalcones 4HD, XA and XA-E whose structures are shown in Fig. 1. All have a common dihydroxychalcone core scaffold (colored in black in Fig. 1). XA is the only one with a resorcinol ring (benzyl-1,3-diol), while 4HD and XA-E have a benzyl-1-ol-3-methoxy. 4-HD and XA have an aliphatic chain (prenyl and geranyl, respectively) (red moiety in Fig. 1) while XA-E has a polar peroxide moiety at the same position.

Figure 1.

Structures of 4-hydroxyderricin (4-HD), xanthoangelol (XA) and xanthoangelol-E (XA-E) isolated from Angelica keiskei Koidzumi.

2.2. Studies with the ZIKV protease

Firstly, we examined the possible binding and interactions of the chalcones with the Zika virus proteins performing docking calculations through the OpenZika (OZ) Project [2,28][28]. The OZ project, an open science collaboration with the IBM World Community Grid (https://www.worldcommunitygrid.org/research/zika/overview.s), uses Autodock Vina program [29] for computational docking calculations. Through this project, docking calculations of millions of compounds were performed against all ZIKV proteins such as NS1, NS2B-NS3 protease, NS3 helicase, NS5 RdRP, NS5 methyltransferase, as well as the envelope and capsid proteins, exploring proteins active sites and allosteric binding sites [2,28].

The NS2B-NS3 protease plays a key role in ZIKV replication, cleaving the major ZIKV polyprotein into individual non-structural and structural proteins. Since Park et al. [22] have reported that several alkylated chalcones can inhibit the activity of the SARS-CoV proteases including the chymotrypsin-like protease (3CL^pro) and the papain-like protease (PL^pro) [22], we examined the binding mode of the chalcones with the Zika virus NS2B-NS3 protease. From these calculations a score is generated which can be predictive of the level of protein-ligand affinity and guides the prioritization of ligands in the virtual screening of a dataset [30]. For visualization of the interaction between the ZIKV NS2B-NS3 protease and chalcones, the binding sites used to build the docking grids were based on both the catalytic (PDB ID 5YOD) [31] and the allosteric site (PDB ID 5GXJ) [32]. For the catalytic site, the protein was analyzed in the active or closed conformation, in which NS2B is wrapped around the NS3 active site, whereas for the allosteric site, the protein was analyzed in the catalytically inactive or open conformation, in which NS2B is dissociated from NS3 [33].

A summary of our docking calculations is shown in Table 1. The docking scores were generally favorable for docking to occur between the chalcones and the NS2B-NS3 protease allosteric and catalytic sites, with scores ranging from −6.5 to −8.0 Kcal·mol⁻¹.

Table 1.

Docking scores (Kcal·mol⁻¹) of the chalcones against the ZIKV NS2B-NS3 protease and the NS5 RdRp. Values were obtained using Autodock Vina [29] software.

	docking score (Kcal·mol−1)
compound	NS2B-NS3 Protease		NS5 RdRp†
	Allosteric pocket*	Catalytic site§	N-pocket
4-HD	−8.0	−6.5	−3.1
XA	−7.6	−7.0	−3.8
XA-E	−8.0	−7.2	−4.4

crystallographic structures: PDB ID

^*5GXJ;

^§5YOD;

^†5TFR

Next we sought to visualize the interaction of the chalcones with the ZIKV protease, which is a heterodimer formed by the N-terminal domain of NS3 and NS2B, with His51, Asp75, and Ser135 [31] of the NS3 domain forming the catalytic triad [31]. Additionally, four “pockets” which position the substrate for cleavage [34], lie near the catalytic triad. Each pocket is linked to a key amino acid; the S1 pocket and Asp129 of NS3, the S1’pocket and Ser135 of NS3, the S2 pocket and Ser81 of NS2B, and the S3 pocket and Tyr161 of NS3. Visualization of the docking pose at the active site suggested that 4HD can interact directly, via a hydrophobic bond, with His51 in the catalytic triad (Fig. 2A). 4HD also can form hydrophobic interactions with Tyr161 of the S3 pocket (S3) and Val155 and a hydrogen bond with nearby residue Gly159 (Fig. 2A). For XA, we found potential hydrophobic interactions with Tyr161 (S3) and Val155, and a hydrogen bond with Ala132 (Fig. 2B), while XA-E formed potential hydrophobic interactions with Val155 and Tyr161 (S3) and hydrogen bonds with Asp83 and Gly159 (Fig. 2C). Together, all the chalcones were predicted to form hydrophobic interactions with Tyr161 of the S3 pocket, but only 4HD displayed the potential for interaction with His151 of the active site. Structural studies of the NS2B-NS3 protease, co-crystallized with known inhibitors of the active site, such as the boronate [35] or a pyrazole ester derivative molecule [31], revealed interactions with His51 of the active site and with residues in two of the essential pockets, Ser135 of S1’, and Tyr161 of S3. Of the chalcones, only 4HD has the potential for interaction with the active site and one surrounding pocket (Tyr 161 of S3), while XA and XA-E, interacted only with Tyr161 of S3.

Figure 2. 3D intermolecular interactions of ZIKV NS2B-NS3 protease catalytic site in docking poses:

A) 4HD (colored in orange), B) XA (colored in pink) and C) XA-E (colored in mauve). Hydrophobic interactions are presented as white transparent surface and hydrogen bonds as green dotted lines. Hydrogen, nitrogen and oxygen atoms are colored in white, blue and red, respectively.

The NS2B-NS3 protease allosteric site is formed by Leu85, Glu86, Gly87, Val146, Gly148, Leu149, Tyr150, Gly151, and Asn152 [36]. Visualization of the docking poses of chalcones at the NS2B-NS3 protease allosteric site suggest that all three chalcones can interact with Asn152 through hydrogen bonding (Figs. 3A–C). 4HD interacted with Asn152 and with two other residues in the allosteric site (Leu85 and Leu149) and with Val155 in the same region. XA interacted with Asn152 and with Val155 as shown in Fig. 3B. Finally, XA-E interacted with Asn152 and with Val155, Lys73, Leu 76, and Trp83, four residues surrounding the allosteric site (Fig. 3C), which may stem from the inverted docking of this chalcone to the protease (Fig 3). Overall, XA-E and 4HD are predicted to establish more hydrogen bonds and hydrophobic interactions than did XA, which likely contributed to the more favorable docking scores for XA-E and 4HD as shown in Table 1.

Figure 3. 3D intermolecular interactions of ZIKV NS2B-NS3 protease allosteric pocket in docking poses:

A) 4HD (colored in orange), B) XA (colored in pink) and C) XA-E (colored in mauve). Hydrophobic interactions are presented as white transparent surface and hydrogen bonds as green dotted lines. Hydrogen, nitrogen and oxygen atoms are colored in white, blue and red, respectively.

At the allosteric site, a number of studies suggest that interaction at Asn152 is critical for inhibition [6,9,36–39]. The flavonoids myricetin [6] and pedalitin [9], for example, inhibitors of the ZIKV protease, and the NCI compound NSC135618 [36] which inhibits the DENV2 protease, all interact with Asn152. Similarly, Brecher et al. [36] identified a number of inhibitors of the allosteric site of DENV2, ZIKV, YFV and WNV proteases and found that an Asn152Ala mutation completely abolished the inhibitory activity of these compounds. In addition, Chen et al. [40,41] developed and validated a pharmacophoric model for the DENV protease allosteric pocket, identified several key features for chemical inhibitors including two hydrogen bond acceptors near Asn152 and Asn167. Building on these models, Mukhametov et al. [37] used in silico modeling to suggest a two-step process for inhibition of the allosteric site with ligands first hydrogen bonding to Asn152, which destabilizes the contact between the NS2B and the allosteric pocket of the NS3 protease domain. Next, the ligand binds NS3 Leu149 and NS2B Gly82 residues, reducing the mobility of the NS2B preventing its conformational change [37]. Our results support these investigations and suggest that all of the chalcones we studied have the potential to block the conformational change by forming hydrogen bonds with the key residue Asn152. The chalcones may also form interactions with additional residues near the allosteric site, interactions which may also contribute to inhibition of the conformational change [37].

Our modeling studies predicted that the chalcones would interact with the Zika virus protease, with impacts on possibly both the active and allosteric sites. These predictions were tested using a ZIKV protease activity assay [31]. First, we screened the chalcones for inhibitory activity at a single 200 μM concentration. We found that 4-HD, XA and XA-E inhibited protease activity (Fig. 4A), above our 80% threshold for biologically significant inhibition. Based on these results, the chalcones were tested over a broad range of concentrations with IC₅₀s for 4-HD, XA, and XA-E determined to be 47 ± 10 μM, 50 ± 5 μM, and 18 ± 5 μM, respectively (Fig. 4B–D). The results of these experiments confirmed the modeling predictions that these chalcones could interact with and inhibit the ZIKV protease, with XA-E the most promising inhibitor. Likely the potent inhibitory activity of this compound stems from the enhanced number of interactions this molecule forms in and near the protease allosteric site. A number of ZIKV protease inhibitors have already been reported [5,6,9,35] with IC₅₀ values ranging from 70 nM to 56.3 μM. The ashitaba-derived chalcones we examined all displayed IC₅₀s within this range. Chalcones have been recognized as valuable starting points for drug design, particularly for their ease of synthesis and breadth of possibilities for derivation that is bolstered by computational efforts, in addition to their broad bioactivity profiles [42].

Figure 4. ZIKV enzymatic protease assays.

A) Activity assays at 200 μM for 4-HD, XA, and XA-E. Concentration-response curves adjusted with Hill to determine IC₅₀ ± Δ IC₅₀ values for B) 4-HD, C) XA and D) XA-E. The positive control aprotinin had an IC₅₀ of 0.13 ± 0.02 μM.

Kinetic parameters of protease inhibition were also investigated to discriminate between active site and allosteric inhibitory effects. The measured fluorescence of 7-amine-4-metylcoumarin (AMC) was used to determine the velocity of the reactions, and to obtain Michaelis-Menten (Fig. S1A–C) and Lineweaver-Burk plots (Fig. S1D–F). We found V_max values were reduced with increasing chalcone concentration. The pattern of consistent K_m values (Table S1) with concentration-dependent decreases in the V_max is typical of allosteric inhibitors acting in a non-competitive mode. Apparently, the interactions of the chalcones at the allosteric site likely inhibiting the change in conformation of the protease is the dominant mode of inhibition for these compounds.

A similar mode of allosteric, non-competitive inhibition, was noted for several alkylated chalcones derived from 4-HD, XA, and XA-E when tested against the papain-like protease (PL^pro) [22] of severe acute respiratory syndrome coronavirus (SARS-CoV), in enzymatic and kinetic assays. Agreeing with our results, XA-E was characterized as the most potent SARS-CoV PL^pro inhibitor with an IC₅₀ of 1.2 μM, acting in a non-competitive mode. However, in the same study these chalcones were shown to be competitive inhibitors of chymotrypsin-like protease (3CL^pro). The chalcone derivatives 4-hydroxypanduratin A and panduratin A were tested against DENV2 NS2B-NS3 protease and displayed competitive inhibitory activity [39]. The structures of 4-hydroxypanduratin A and panduratin A are very different from XA, XA-E and 4HD and likely those differences change their mode of action.

2.3. Studies with the ZIKV NS5 RdRp

The ZIKV RdRp contains two important binding sites: the active site (formed by Asp535, Asp665 and Asp666) [43] where nucleoside triphosphates (NTPs) are incorporated, and the allosteric site or N-pocket [44], which includes the priming loop (Thr796, Trp797, Ile799, Lys802, Glu804 and Trp805) [14,45]. The priming loop is highly conserved among flaviviruses, responsible for the stabilization of the de novo initiation complex [46], which allows the release of newly synthetized dsRNA.

As shown in Table 1, in contrast to our results with the ZIKV protease, we found generally unfavorable docking scores of −3.1 and −4.4 Kcal·mol⁻¹ for the NS5 RdRp (PDB ID: 5TFR) [47].

Visual inspection of the docking poses of chalcones at the N-pocket revealed a limited number of potentially meaningful interactions. As shown in Fig. 5A, 4-HD formed hydrophobic interactions with Trp797 (priming loop) and hydrogen bonds with Lys462, Ile799 (priming loop) and Glu735. XA formed hydrophobic interactions with Glu460 and Trp797 (priming loop) (Fig. 5B). XA-E made hydrophobic interactions with Trp797 (priming loop) and hydrogen bond with Lys462, Cys711 and Ser798 (priming loop) (Fig. 5C).

Figure 5. 3D intermolecular interactions of ZIKV NS5 RdRp (N-pocket) docking poses:

A) 4HD (colored in orange); B) XA (colored in pink) and C) XA-E (colored in mauve). Hydrophobic interactions are presented as white transparent surface; hydrogen bonds, as green dotted lines and π-cation interaction, as red dotted lines. Hydrogen, nitrogen and oxygen atoms are colored in white, blue and red, respectively.

In order to directly evaluate the ability of chalcones to inhibit ZIKV NS5 RdRp activity, we performed an enzymatic assay described by Sáez-Álvarez and coworkers [48]. All chalcones were tested against the ZIKV NS5 RdRp domain, but only XA inhibited RdRp activity above 80% threshold for biologically significant inhibition (Fig. 6A). In a 20 μM endpoint assay, XA inhibited 85% of polymerase activity. Concentration-response experiments showed that XA had an IC₅₀ value of 6.9 ± 0.9 μM (Fig. 6B). The significant inhibition of RdRp activity noted with XA seemed at odds with its low docking score (Table 1), its failure to inhibit virus growth significantly, and its relative lack of interactions with the RdRp in visualization studies (Fig. 5). Therefore, to confirm its ability to dock with the RdRp, we performed ensemble docking studies which allows the investigation of different binding modes of small molecules in multiple protein conformational states, such as the allosteric site of the RdRp.

Figure 6. ZIKV polymerase assays.

A) Activity assays at 20 μM for 4-HD, XA, and XA-E. B) Concentration-response curves adjusted with Hill to determine IC₅₀ ± Δ IC₅₀ values for XA.

In the ZIKV NS5 RdRp crystallographic structures available in the Protein Data Bank (PDB), the priming loop of the allosteric pocket exhibits a closed conformation which does not favor ligand binding. So, we performed an ensemble docking to study the relevance of the NS5 RdRp priming loop conformation on ligand binding. To obtain different protein conformations, we performed 500 ns of molecular dynamics (MD) simulations that showed the priming loop adopt several conformations during the trajectory (Fig. S2A). The MD conformations were grouped into 16 clusters (Fig. S2B) where the RMSD of the priming loop residues in clusters varied from 0.05 to 0.53 nm. We found that clusters 7 and 15 presented the priming loop in an open conformation, potentially favoring ligand binding at the allosteric pocket. Also, we noted that several DENV RdRp crystallographic structures available in the PDB (PDB ID: 5JJS, 5JJR, 5K5M, 5I3P and 5I3Q) [14] display the priming loop in an open conformation since they were co-crystallized with allosteric inhibitors. Therefore, we also constructed a refined ZIKV NS5 RdRp model based on DENV3 RdRp (PDB ID: 5JJS) [14].

The three structures of ZIKV NS5 RdRp with the priming loop in an open conformation (cluster 7, cluster 15 and the ZIKV NS5 RdRp model) were submitted for ensemble docking in Autodock Vina. The docking calculations were repeated 200 times and the five best poses were analyzed. In addition to the Vina score, we rescored the binding affinity of docking poses using a Random Forest score V3 (RFscore) [49]. All poses were clustered and analyzed through consensus score and the best consensus scores are shown in Fig. 7A.

Figure 7. Ensemble docking of XA against ZIKV NS5 RdRp.

A) Vina score versus RFscore of the XA - ZIKV NS5 RdRp clustered docking poses. The best docking pose is represented by cluster 1 (colored in blue) and pose 1 (circle) that had Vina score of −9.0 Kcal·mol⁻¹ and RFscore of −7.3 Kcal·mol⁻¹; B) XA-RdRp best docking pose. Hydrophobic interactions are presented as white transparent surface and hydrogen bonds, as green dotted lines. Hydrogen, nitrogen and oxygen atoms are colored in white, blue and red, respectively; C) Superposition of XA docking pose (colored in cyano) and DENV RdRp allosteric co-crystallized inhibitor (colored yellow) and D) 2D chemical structural comparison of XA and DENV RdRp allosteric inhibitor: common ligand moieties are highlighted: hydroxyls (colored in orange), delocalized π electrons (colored in green) and aromatic rings with hydroxyls/methoxyls (colored in blue).

Fig. 7B shows the best XA – ZIKV NS5 RdRp docking pose (docking score of −9.0 Kcal·mol⁻¹), demonstrating several interactions with residues in and near the priming loop, including hydrophobic interactions with Lys802 and Trp805 (priming loop), two additional residues of the N-pocket (Leu513 and Arg739), an anion - π interaction with Glu804 (priming loop) and hydrogen bonds with Thr796 (priming loop) and Tyr768 (N-pocket). It is likely that the aliphatic moiety of XA is making hydrophobic contacts with RdRp residues and which are likely required for its activity. These interactions were not observed in the docking calculations using the RdRp crystal structure (Fig. 7B), likely due to the closed conformation of the priming loop. A number of compounds have been identified against the RdRp flaviviruses, including several NIs [10,50,51] and a few NNIs [12]. A similar pattern of interactions with priming loop and N-pocket residues was noted with two DENV RdRp NNIs, compounds 27 and 29 [14], including water bridges with residue Arg737 (ZIKV Arg739), hydrogen bond with Thr794 (ZIKV 796), His800 (ZIKV Lys802) and Gln802 (ZIKV Gln804), and hydrophobic interactions with Leu511 (ZIKV Leu513) and Trp803 (ZIKV Trp805). Superimposition of the docking poses of XA and compound 27 (Fig. 7C) revealed that these compounds assumed the same orientation within their respective pockets. Also, XA and compound 27 utilize common ligand moieties to interact with similar regions of the RdRp allosteric pocket (Fig. 7D) including hydroxyls, delocalized π electrons, and aromatic rings with hydroxyls/methoxyls. Collectively our data reinforce the observation that forming multiple interactions in and around the priming loop creates a favorable mode for NNI’s that target the RdRp.

2.4. Machine learning predictions

A machine learning (ML) model developed with Assay Central^™ software [52] was used to predict the anti-ZIKV activity and cytotoxic potential of the chalcones. For each model, prediction (> 0.5 is active) and applicability domain (% of molecule present in the model) values are assigned. The ZIKV ML continuous model was built using datasets we previously curated from the literature [2]. All machine learning models 5-fold cross validation statistics were acceptable and shown in Fig. S8. As shown in Table S2, the Zika ML model predicted all compounds as active, with a stronger anti-ZIKV activity for XA than for 4-HD or XA-E. All scores were relatively similar and likely reflect a probability of activity.

2.5. Chalcone inhibition of activities ZIKV replication in vitro

These chalcones were then tested for their ability to inhibit the in vitro growth of ZIKV. As shown in Fig. 8A and C, 4-HD and XA-E exerted strong inhibitory effects on ZIKV replication with EC₅₀’s of 6.6 μM and 22 μM, respectively. Replication was blocked completely by 4-HD and XA-E at 45 and 75 μM, respectively. In contrast, XA displayed only weak inhibition of ZIKV growth, even at concentrations exceeding 50 μM and an EC₅₀ for XA could not be calculated (Fig. 8B). The cytotoxic activity of these compounds towards Vero cells was also tested as shown in Fig. 8D. We found that 4-HD and XA-E did exert cytotoxic effects, although at concentrations much higher than their anti-viral effects. For 4-HD and XA-E, CC₅₀ values were 103 ± 14 and 111 ± 10 μM, translating to selectivity indexes of 17 and 5, respectively. These experiments, for the first time, show that chalcones can effectively block the replication of ZIKV in vitro. XA, which did display both protease and polymerase inhibitory activities, did not display cytotoxic nor significant anti-viral effects, raises questions about the entry of XA into cells or its rapid degradation. Perhaps the aliphatic group on the right side of XA (see Fig. 1), which XA-E lacks, could causes XA to insert into cell membranes to a high degree, and it is unable to reach its necessary intracellular location at a sufficient concentration.

Figure 8. The inhibitory and cytotoxic effects of the chalcones on ZIKV.

Inhibition of ZIKV growth by 4HD (A), XA (B) and XA-E (C) and cytotoxicity of these compounds (D). Two-step assays were used to quantify ZIKV. First, Vero cells were infected with ZIKV 1h, compounds added, and supernatants harvested after 48 hrs. Virus titers in supernatants were then determined by plaque assay. Cytotoxicity was determined by measuring the release of lactase dehydrogenase from Vero cells. Statistical analysis was performed using a one-way ANOVA with Tukey’s post-test, *p < 0.05.

The summary of the computational and experimental results obtained in this project are presented in Table 2. Docking calculations predicted ZIKV NS3pro and NS5RdRp as promising targets. Enzymatic and kinetic assays highlighted XA and XA-E as potential inhibitors of ZIKV protease and polymerase and antiviral assays revealed 4HD and XA-E as ZIKKV antiviral candidates.

Table 2.

Summary of the docking, enzymatic and antiviral results for the studied chalcones.

Chalcone	NS3pro docking score (Kcal·mol−1)	NS5 RdRp docking score (Kcal·mol−1)	NS3pro IC50 (μM)	NS5 RdRp IC50 (μM)	Antiviral EC50 (μM)
4HD	−8.0	−3.1	47 ± 10	-	6.6
XA	−7.6	−3.8 (−9.0*)	50 ± 5	6.9 ± 0.9	-
XA-E	−8.0	−4.4	18 ± 5	-	22

^*most favorable ensemble docking score

3. Conclusions

Despite the significant health impact that ZIKV has on global health, antiviral drugs, whether natural or synthetic products, are not available for the treatment of ZIKV infection. The OpenZika project aimed to fill this gap in the search for new synthetic and natural compounds capable of inhibiting ZIKV proteins, using computational approaches and the World Community Grid volunteer’s computer network. Natural products extracted from A. keiskei (ashitaba) have the potential for use in the treatment of a variety of diseases and disorders [18,53,54]. 4-HD and XA-E displayed moderate anti-viral activity and low cytotoxicity, with inhibitory activity against the ZIKV protease. Moreover, XA-E displayed inhibitory activity against ZIKV NS3 protease acting in a noncompetitive mode of inhibition. In addition, XA displayed activity against the ZIKV protease and also might inhibit ZIKV RdRp at the allosteric site but did not display antiviral activity. The aliphatic chain of XA could be reduced sequentially to determine its optimum length for activity with whole cells. Further studies of the structure-activity relationships between chalcone uptake and stability, protease and polymerase inhibition, antiviral activity, and cytotoxicity could generate useful compounds for in vivo investigations, moving to hit-to-lead optimization.

4. Experimental

4.1. Plant Material and Extraction.

Fresh ashitaba roots were collected from Strictly Medicinal Seeds in Williams, Oregon (Sample #12421, N 42°12’17.211”, W 123°19’34.60) on November 14, 2015. Additional material (used to isolate XA-E) was collected on December 29, 2016 from the same source and at the same location (Sample #12444). Richard A. Cech from Strictly Medicinal Seeds confirmed the identity of plant material, and a voucher specimen was deposited at the herbarium at the University of North Carolina at Chapel Hill (Accession Number NCU627665). Ashitaba roots were dried at 40° C for 24 hours in a single-wall transite oven (Blue M Electric Company), yielding 138.90 g of dry mass. The dry mass was then ground using a Wiley Mill Standard Model No. 3 (Arthur Thomas Company) and submerged at 160 g/L in MeOH for 24 hours each day for three days. The resulting MeOH extract was concentrated in vacuo and subjected to partitioning. First, defatting was completed by partitioning between 10% aqueous MeOH and hexane (1:1). The aqueous MeOH layer was then partitioned between 4:5:1 EtOAc/MeOH/H₂O. The EtOAc layer was then washed with a 1% NaCl solution and dried under nitrogen. The 3,650.32 mg of dried extract was used to isolate 4-HD and XA. Scale up material (964 g) was prepared using the same methods outlined above, yielding 18.10 g of dried extract which was utilized to isolate XA-E.

4.2. Chromatographic Separation and Isolation of chalcones.

Isolation schemes are provided as Supplementary Material (Fig. S3). First, 4-HD and XA were isolated using 3,100 mg of the original EtOAc extract (AK). Normal-phase flash chromatography was completed with a 40 g silica gel column at a 40mL/min flow rate using a gradient of hexane/CHCl₃/MeOH over 35 min. The last two fractions (AK-3 and AK-4) were subjected to a second stage of normal-phase flash chromatography. AK-3 (1355 mg) was separated on a 40 g silica gel column with a 40 mL/min flow rate while AK-4 (536 mg) was separated with a 12 g column and a flow rate of 30 mL/min. This time, a 45 min hexane/EtOAc/MeOH gradient was used. The second fraction from each separation (fractions AK-3–2 and AK-4–2, respectively) were subjected to additional chromatographic separation. First, AK-3–2 (1000 mg) was separated using reversed-phase flash chromatography with a 130g C18 reversed-phase RediSep Rf column and a flow rate of 85 mL/min. A 25-min gradient of CH₃CN/H₂O was conducted, beginning at 40:60 and increasing to 85:15. It was then increased to 100:0 for 5 min, after which the original conditions were re-established. 4-HD (compound 1) eluted at 18 min (234.45 mg, 97% purity, 7.6% yield). Fraction AK-4–2 (364 mg) was also subjected to a final round of chromatography, this time using reversed-phase preparative HPLC. The sample was injected onto a Luna preparatory column (5 μm PFP, 250 × 21.20mm; Phenomenex) and separated over a 35 min gradient which began at 40:60 CH₃CN:H₂O and was increased to 100:0 over thirty minutes. XA (compound 2) eluted between 28–35 min (284.59 mg, 95% purity, 9.1% yield).

XA-E was isolated from 17.5 g of scale up material. The EtOAc extract (AK) was separated on a 120 g silica column and a flow rate of 85 mL/min using the same hexane/CHCl₃/MeOH gradient as described for the first fractionation of the original extract. The second fraction (AK-6, 5.3 g) was separated again with normal-phase flash chromatography on a 120 g silica column and 85 mL/min flow rate with a 55-minute gradient of hexane/EtOAc/MeOH. XA-E (compound 3) eluted at 31 minutes (150 mg, 95% purity, 0.85% yield). Compound purity for all three compounds was assessed using photo-diode array (PDA) analysis (Supplementary Material, Fig. S4).

4-hydroxyderricin (4-HD, 1):

yellow crystalline solid; HRESIMS m/z 337.1438 [M-H]⁻ (calc. for C₂₁H₂₁O₄⁻, 337.1440, −0.59 ppm); ¹H NMR (500 MHz, CDCl₃) and ¹³C NMR (125 MHz, CDCl₃) data were reported in a previous publication (Caesar et al. 2018), and chemical shifts matched literature values (Kawabata et al. 2011). Additional ¹H NMR spectra were collected to ensure compound stability (Supplementary Material, Fig. S5). Chemical shifts were consistent with previous reports (Caesar et al. 2018, Kawabata et al. 2011).

Xanthoangelol (XA, 2):

yellow crystalline solid; HRESIMS m/z 391.1907 (calc. for C₂₅H₂₇O₄⁻, 391.1909, −0.51 ppm); ¹H NMR (500 MHz, CDCl₃) and ¹³C NMR (125 MHz, CDCl₃) data were acquired for a previous publication (Caesar et al. 2018) where chemical shifts matched literature values (Kawabata et al. 2011). Again, ¹H NMR spectra were collected to ensure stability (Supplementary Material, Fig. S6), and chemical shifts remained consistent with previous reports (Caesar et al. 2018, Kawabata et al. 2011).

Xanthoangelol E (XA-E, 3):

yellow, amorphous powder; HRESIMS m/z 351. 1231 [M-H]⁻ (calculated for C₂₁H₁₉O₅⁻, 351.1232, −0.28 ppm); ¹H NMR (500 MHz, DMSO-d₆) and ¹³C NMR (125 MHz, DMSO-d₆) data were acquired in a previous study (Caesar et al. 2018) where chemical shifts matched literature values (Baba et al.1990). ¹H NMR spectra were collected under the same conditions to ensure that no compound degradation had occurred, and chemical shifts remained consistent with existing literature (Caesar et al. 2018, Baba et al. 1990) (Supplementary Material, Fig. S7).

4.3. Complementary analysis of purified chalcones.

Nuclear magnetic resonance (NMR) spectra for purified ashitaba chalcones were collected with a JEOL ECA-500 HMz spectrometer. Ultra-High Performance liquid chromatography-mass spectrometry analysis was completed in negative mode using a Thermo Scientific LTQ Orbitrap XL Mass Spectrometer connected to a Waters Corporation Acquity UPLC system. UPLC-MS analysis was conducted using a 3 μL injection of 1 mg/mL samples suspended in MeOH and a flow rate of 0.3 mL/min injected onto a BEH C18 column (1.7 μm, 2.1 × 50 mm, Waters Corporation). The gradient consisted of water (solvent A) and acetonitrile (solvent B), each with 0.1% formic acid. The gradient began at 90:10 (A:B) from 0–0.5 min, and increased from 0:100 (A:B) from 0.5–8.0 min. It was then held at 100% B for 0.5 min before returning to starting conditions from 9.0–10.0 min. Mass analysis was conducted in positive mode with a scan range from 150–2000 with a capillary voltage of −21.00 V, a capillary temperature of 275.00 °C, a tube lens offset of −95.00 V, a spray voltage of 3.50 kV, a sheath gas flow of 30.00, and an auxiliary gas flow at 15.00. Flash chromatography was completed with a Teledyne-Isco CombiFlash RF system attached to a PDA detector and evaporative light scattering detector (ELSD). Reversed-phase preparative HPLC was done with a Varian HPLC system (Agilent Technologies) and assessed using Galaxie Chromatography Workstation Software (1.9.3.2, Agilent Technologies). All solvents were acquired through Sigma-Aldrich and were spectroscopic grade.

4.4. Molecular docking studies.

Molecular docking calculations were performed using the Autodock Vina software [29], considering the ligand flexible and the protein rigid. The 3D protein structures of NS2B-NS3 protease protein (closed conformation) (PDB ID: 5YOD) [31] and protein (open conformation) (PDB 5GXJ) [32] and NS5 polymerase (5TFR) [47] were obtained from the RCSB protein data bank (https://www.rcsb.org). The proteins structures were prepared on AutoDockTools 1.5.6 [55], using the standard AutoDockTools preparation protocol for proteins[55]. The ligands were prepared in the Avogadro program 1.2.0 [56], adding hydrogens (at pH 7.4) and minimizing the molecules geometry, using force field MMFF94s. The minimized molecules were then prepared in AutoDockTools 1.5.6 [55], following the standard preparation protocol for ligands. The protein grid coordinates for the binding sites of NS2B-NS3 protease (catalytic and allosteric site) [31,36] and RdRp (active, RNA and allosteric site) [47,57] were built based on literature. These docking calculations were part of our OpenZika project [28]. Detection and visual analysis of the non-covalent interactions between chalcones and ZIKV proteins were performed using Medicinal Chemistry-based visual inspection and the PLIP web server [58] to investigate protein-ligand patterns of interaction, such as hydrogen bonds, hydrophobic contacts, π- π stacking, water bridges and salt bridge interactions. To identify protein-ligand interactions, PLIP algorithm matches interacting atoms, using rule-based systems of geometric constraints, and measures distances and angles among the atoms [58]. These analyses were then combined with the docking scores to identify ZIKV proteins, and their binding sites.

4.5. Ensemble docking of NS5 RdRp.

We performed 500 ns molecular dynamics (MD) simulations of ZIKV NS5 RdRp using GROMACS [59,60] to obtain different protein conformations for the ensemble docking of NS5 RdRP. The MD trajectory conformations were clustered, in relation to the priming loop residues, through gmx cluster [61] tool, using the gromos method [61] and cut-off 0.2. We also built a ZIKV NS5 RdRp model using the MODELLER [62] program to generate an open conformation of the ZIKV NS5 RdRp priming loop, using DENV RdRp PDB ID 5JJS [14] as template. 20 models were generated using MODELLER and refined twice, applying 300 steps of the “slow” optimization protocol. During the modelling procedure, the models were assessed using DOPE [63] and GA341 [64] scores. The model presenting the best DOPE score was used as receptor for docking. Autodock Vina [29] program was used for docking calculations at the allosteric pocket of RdRp. We generated up to 5 docking poses (num_modes=5) within a maximum energy range of 10 Kcal·mol⁻¹ (energy_range=10) and the exhaustiveness was set to 40. The protein was prepared using MGLTools v1.5.6 [65] and XA structure was prepared using prepare_ligands4.py included in the same MGLTools distribution. The Open Drug Discovery Toolkit (ODDT) [66] was used to re-score the docked poses, using the RFScore_V3 [49] function, trained on the PDBBind2016 dataset [67,68]. To increase the robustness of the results, the above-described procedure was repeated 200 times and the results were clustered using the “gmx cluster” program, part of the GROMACS [59,60] package. The clustering procedure was carried out with a RMSD cut-off of 0.15 nm and the docking poses were not fitted prior to the clustering to capture translational and rotational differences. The results were visually inspected using PyMol [69] and plotted using DataWarrior [70].

4.6. Protein cloning, expression and purification.

The gene encoding NS2B-NS3^Pro (GenBank accession no KU321639.1) based on the sequence gZIPro reported by [71] and synthesized by Eurofins Genomics. The gene was cloned into the expression plasmid pETSUMO-1a/LIC and the plasmid was transformed into Rosetta (DE3) E. coli (Novagen). The transformed cell was grown in Luria Broth (LB) medium, supplemented with 50 μg.mL⁻¹ kanamycin and 34 μg.mL⁻¹ chloramphenicol at 37 °C overnight and then inoculated 1L of LB, supplemented with the same antibiotics. Cells were grown until Optical Density at 600 nm (OD₆₀₀) reached 0.6 and protein expression was induced with 0.1 mM Isopropyl β-D-1-thiogalactopyranoside for LB medium for 16 h at 18 °C and for ZYM medium temperature was reduced to 18 °C and the expression performed for 24 h.. Cells were harvested by centrifugation at 4000g at 8 °C for 40 min and suspended in 50 mM Tris pH 8.0, 300 mM NaCl, 30 mM Imidazole, 10% Glycerol (Buffer A). Cells were lysed by sonication and cell debris was separated by centrifugation at 10000g at 8 °C for 30 min. The supernatant was loaded on a HisTrap HP 5.0 mL with a Ni Sepharose resin (GE Healthcare) equilibrated with Buffer A or A’ and eluted by Buffer A or A’ with 500 mM Imidazole. The buffer was changed to 50 mM Tris pH 8.0, 300 mM NaCl (Buffer B) with dialysis concomitantly with TEV protease cleavage from 6His-SUMO-tag. Another HisTrap HP 5.0 mL with a Ni Sepharose resin (GE Healthcare) equilibrated with Buffer B was performed to detach NS2B-NS3^Pro. Finally, a size-exclusion chromatography on a XK 16/60 Superdex 75 column (GE Healthcare) equilibrated in buffer 20 mM Hepes, pH 7.0, 500 mM NaCl, 5% Glycerol. The gene encoding NS5 RdRp (GenBank accession no KU321639.1) and based on the sequence gZIPro reported by [11] was codon optimized for Escherichia coli (E. coli) heterologous expression and synthesized by Eurofins Genomics. The gene was cloned into the expression plasmid pETTrx-1a/LIC and this plasmid was transformed into Rosetta (DE3) E. coli (Novagen). The transformed cells were grown in Luria Broth (LB) medium, supplemented with 50 μg.mL⁻¹ kanamycin and 34 μg.mL⁻¹ chloramphenicol at 37 °C overnight and then inoculated 1L of ZYM-5052 medium supplemented with the same antibiotics. Cells were grown until Optical Density at 600 nm (OD₆₀₀) reached 0.6 and protein expression medium temperature was reduced to 18 °C and the expression performed for 24 h. Cells were harvested by centrifugation at 4000g at 8 °C for 40 min and suspended in 50 mM Tris pH 9.0, 200 mM NaCl, 20 mM Imidazole, 10% Glycerol (Buffer A). Cells were lysed by sonication and cell debris was separated by centrifugation at 10000g at 4 °C for 30 min. Buffer A loaded the supernatant on a HisTrap HP 5.0 mL with a Ni Sepharose resin (GE Healthcare) equilibrated with Buffer A and eluted with 200 mM Imidazole. The buffer was changed to Buffer A with dialysis concomitantly with TEV protease cleavage from 6His-Trx-tag. Another HisTrap HP 5.0 mL with a Ni Sepharose resin (GE Healthcare) equilibrated with Buffer A was performed to detach NS5 RdRp from the 6His-Trx-tag. Finally, a size-exclusion chromatography on a XK 16/60 Superdex 75 column (GE Healthcare) equilibrated in buffer 50 mM Tris, pH 7.5, 200 mM NaCl, O.5 mM TCEP, 5% Glycerol. The protein purity and molecular weight were verified by SDS-Page.

4.7. NS2B-NS3 protease activity assay.

The protease enzymatic assay was performed according to protocol described by Li and coworkers [31]. First of all, single time point assays were performed. 4 nM of enzyme and 200 μM of each compound was incubated for 30 minutes in 20 mM Tris pH 8.5, 10% Glycerol and 0.01% Triton X-100. DMSO (1% vol/vol) was used as reference. The reactions were started with 30 μM Bz-nKRR-AMC substrate. The reaction was maintained at 37 °C and fluorescence measurements (λ_excitation = 380 nm and λ_emission = 460 nm) were performed for 30 minutes. Assays were performed in duplicate on 96- white flat plate and in a SpectraMax Gemini EM Microplate Reader (Molecular Devices Co., Sunnyvale, CA). Aprotinin (10 μM) was used as positive control as described in Fernandes et al [72]. It is a small protein that has already been described as an inhibitor of flavivirus proteases [73,74]. To kinetic assays, Fluorescence conversion to product concentration was performed using a standard curve of AMC (7-amine-4-metylcoumarin). For concentration-responses 4 nM of protein and varied concentration from 200 μM to 0.195 μM of each individual compound at 37 °C and for determination of the kinetics parameters (Km and Vmax) and the mechanism of action, the substrate was diluted to a range of concentrations from 300 μM to 0.59 μM. Concentration-responses were performed in duplicate and analyzed on GraphPad Prism using a Hill fitting to obtain IC₅₀ values and kinetics parameters were obtained using the Michaelis-Menten fitting on Origin version 2018. The Lineweaver-Burk plot was used to determine the mechanism of action. All data was measured in triplicates.

4.8. NS5 RNA-dependent RNA-polymerase activity assay.

For the detection of RNA synthesis by ZIKV NS5 RdRp, we established a real-time assay based on the fluorescent dye SYBR Green I, analogous to the methodology previously described for ZIKV NS5 RdRp [48]. The reaction was performed using a buffer containing 50 mM Tris pH 7.0, 1.0 mM MnCl2 and 0.01% Triton X-100. A serial dilution of each compound (80 – 0.039 μM) was incubated with 250 nM RdRp for 15 minutes at room temperature. The assay was initiated by the addition of 150 nM RNA poly-U, 0.005% (v/v) of the fluorescent dye SYBR Green I and 500 μM ATP. Fluorescence was recorded in real time over 60 min at 30 °C, with a Stratagene MX3005P Agilent Technologies® QPCR system. The excitation and emission filters were set at 494 and 521 nm, respectively. The endpoint assays were performed at 20 μM, compound that inhibited more than 80% of activity in the 20 μM assay, was submitted to a concentration-response test. The percentage inhibition values were calculated based on a control reaction, containing only DMSO in the same concentrations used for the compounds. The assays were carried out in duplicates in each plate, and the whole experiment is repeated twice. The results were analyzed and plotted using the OriginPro 9.0 program.

4.9. Virus infections.

Vero cells (ATCC CCL-81) were cultured in Minimal Essential Media (MEM) with Earle’s salts and L-glutamine (Genese Scientific, El Cahon, CA) with 6% fetal bovine serum (Gemini Bio-products, West Sacremento, CA) at 37°C and 5% CO₂. 1% antibiotic/antimycotic mix (Gibco, Gaithersburg, MD) was added in assays. For treatments with extracts and pure compounds, 30,000 Vero cells per well were added to flat bottom 96 well plates (Genesee Scientific, El Cahon, CA) for 24 h. Media was removed and fresh media added with Zika virus (ATCC strain MR 766) at a MOI of 1 for 1 h. Media was then aspirated and the cells washed 1X with PBS. Fresh media with extracts or pure compounds, at the indicated concentrations, was then added. Finally, supernatants were collected and stored at −80°C following a 48h incubation period.

5.0. Plaque assays.

Virus titers in supernatants were determined using a plaque assay with Vero cells. Initially, 40,000 cells per well in 500 μl of media were added to 48 well plates for 48 h. The media was then removed and 50 μl of 10-fold supernatant dilutions were added to triplicate wells and incubated for 1 h. A 1.2% tragacanth (Alfa Aesar, Tewksbury, MA) overlay in media was then added to each well and the plates incubated for 96 h. Cells were stained using 200 μl 0.5% crystal violet (Millipore Sigma, Merck, KGaA, Darmstadt, DE) for 1 h at room temperature. The stain was removed, cells washed to remove excess stain, and plaques counted using a transilluminator (UVP, Upland, CA).

5.1. Cytotoxicity assay.

The cytotoxicity of extracts and pure compounds was determined according to the manufacturer’s instructions using the CytoScan LDH cytotoxicity assay (G‐Biosciences, St. Louis, MO). Briefly, 30,000 cells per well were added to 96 well plates and incubated for 24 h at 37°C and 5% CO₂. Extracts and pure compounds were added with fresh media at the indicated concentrations to triplicate wells for 48 h. Following the incubation, the plates were centrifuged at 250 × g for 5 m and 50 μl of supernatant from each well was transferred to a new plate. An equal volume of substrate mix was added to each well and the plates incubated at room temperature for 30 m. Then the stop solution was added and the absorbance measured at 490 nm using a plate reader (Synergy HT, BioTek, Winooski, VT). Per cent cytotoxicity was determined using the following formula; (Experimental-Spontaneous absorbance/Maximum-spontaneous absorbance) × 100.

5.2. Assay Central machine learning predictions.

The Assay Central^® software [75–79] uses the source code management system Git to gather and store molecular datasets from diverse sources, in addition to scripts for curating well-defined structure-activity datasets. These scripts employ a series of rules for the detection of problem data that is corrected by a combination of automated structure standardization (i.e. removing salts, neutralizing unbalanced charges, merging duplicate structures with finite activities) and identifying advanced problems to be resolved by human re-curation. The output is a high-quality dataset and a Bayesian model which can be conveniently used to predict activities for proposed compounds. Assay Central^® prediction workflows assign a probability-like score [80,81] and applicability domain (which assesses the portion of fragments overlapping with the training set molecules) to the input compounds according to a user-specified model. Active classifications are assigned to compounds with a predictions score > 0.5. Bayesian machine learning models using the ECFP6 descriptor [80,81] were generated using Assay Central^® software for various Zika virus datasets [2] as well as cytotoxicity and used to predict the three natural compounds from ashibata. Molecular Notebook [82] was used to collate datasets. Two Zika virus models were utilized for predictions, one of binary data from literature in Huh7 cells [13], and another from continuous data curated from public databases like PubChem and ChEMBL, with an activity cutoff of 10 μM [2]. A model of cytotoxicity in Vero cells at 39 μM was also used in predictions, curated from published datasets [83].

ABBREVIATIONS

ZIKV

Zika virus

DENV

Dengue virus

WNV

West Nile Virus

YFV

Yellow Fever Virus

XA

Xanthoangelol

XA-E

Xanthoangelol-E

4HD

4-hydroxyderricin

C

capsid

M

membrane

E

envelope

NS

nonstructural proteins

NMR

Nuclear magnetic resonance

RdRp

RNA-dependent RNA-polymerase

SARS-CoV

Severe Acute Respiratory Syndrome Coronavirus