Quercetin and luteolin are single-digit micromolar inhibitors of the SARS-CoV-2 RNA-dependent RNA polymerase

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has rapidly become a global health pandemic. Among the viral proteins, RNA-dependent RNA polymerase (RdRp) is responsible for viral genome replication and has emerged as one of the most promising targets for pharmacological intervention against SARS-CoV-2. To this end, we experimentally tested luteolin and quercetin for their ability to inhibit the RdRp enzyme. These two compounds are ancestors of flavonoid natural compounds known for a variety of basal pharmacological activities. Luteolin and quercetin returned a single-digit IC50 of 4.6 µM and 6.9 µM, respectively. Then, through dynamic docking simulations, we identified possible binding modes of these compounds to a recently published cryo-EM structure of RdRp. Collectively, these data indicate that these two compounds are a valid starting point for further optimization and development of a new class of RdRp inhibitors to treat SARS-CoV-2 and potentially other viral infections.

www.nature.com/scientificreports/ and, similarly to remdesivir, it has been repurposed against SARS-CoV-2 18 . However, the RdRp complex of coronavirus can excise erroneous mutagenic nucleotides incorporated into viral RNA, thus creating resistance to nucleotide analog drugs 19,20 . Consequently, non-nucleoside inhibitors could have more desirable mechanisms of action that would limit the development of resistance. In this context, natural products are another source of active compounds with promising antiviral activity. These compounds may serve as a starting point for the development of newer molecular entities with greater efficacy and affinity, and with fewer side effects 21 .
Recently some examples of natural products with appreciable inhibition potency against SARS-CoV-2 RdRp include lycorine and corilagin [22][23][24] . Also, luteolin and quercetin, which are two ancestors of flavonoid compounds, are known for having a range of basal pharmacological activities that, in addition to their well-established anti-inflammatory properties, have demonstrated antiviral properties against picornavirus (RNA virus) and DNA viruses, such as hepatitis B virus, herpes simplex, and adenovirus [25][26][27] .
As depicted in Fig. 1, luteolin and quercetin are based on a 15-carbon skeleton with a chromone core comprising bicyclic 1,4-benzopyrone (A-and C-rings) substituted on carbon 2 with a catechol moiety (B-ring). Ring A features a phloroglucinol substitution pattern with two free hydroxyl groups in position 5 and 7. Notably, quercetin differs from luteolin by only one additional hydroxyl group in 3 position.
Luteolin and quercetin have already been the subject of in silico and in vitro studies focused on the SARS-CoV-2 M pro and spike proteins 28,29 . Docking calculations followed by in vitro testing showed that luteolin and quercetin inhibit the viral protease 3CL pro with IC 50 values in the micromolar range (20 µM and 24 µM, respectively), and with K i ~ 7 μM in the case of quercetin [30][31][32] . In addition, quercetin was found to be active against two crucial targets of SARS-CoV, namely M pro and NTPase/helicase 33,34 .
Furthermore, molecular docking analysis of natural compounds in the active site of RdRp of SARS-CoV-2 suggest luteolin and quercetin as potential inhibitors of this crucial viral enzyme. [35]Nevertheless, to the best of our knowledge, they have never been experimentally tested on SARS-CoV-2 RdRp. Thus, we decided to evaluate their activity against this specific target. Here, we report their potency and computed binding mode at the viral RdRp target.

Results and discussion
First, luteolin and quercetin were tested at two fixed concentrations of 25 µM and 100 µM for their ability to block the viral RdRp target. Both compounds completely inhibited the enzyme at 100 µM, with an inhibition of more than 80% at 25 µM (Table 1). Prompted by these preliminary data, we measured dose-response curves to calculate the IC 50 values (Table 1) by determining each compound's inhibition activity at 10 different concentrations, ranging from 0.005 µM to 100 µM (Fig. 2). Luteolin returned an IC 50 of 4.6 ± 0.3 µM and quercetin an IC 50 of 6.9 ± 1.0 µM. Thus, both compounds displayed a greater potency against RdRp polymerase than those reported against SARS-CoV-2 M pro and spike proteins (see above).
Encouraged by such promising single-digit IC 50 values for these compounds, we decided to evaluate in vitro their drug-like properties, namely aqueous kinetic solubility, together with metabolic and plasma stabilities (Table 1). Luteolin and quercetin have a kinetic solubility of 21 ± 4 µM and 16 ± 5 µM in PBS neutral buffer (pH 7.4), respectively. In terms of metabolic stability, both compounds showed an optimal microsomal stability (t 1/2 > 60 min). In blood plasma, luteolin was stable in the measured time span (120 min), while quercetin was poorly stable (t 1/2 = 7 ± 2 min), probably due to the additional hydroxyl group in 3 position.
A recently published cryoelectron microscopy (cryo-EM) structure of SARS-CoV-2 RdRp in complex with two molecules of suramin (PDB ID 7D4F) 36 , a century-old non-nucleotide analog drug, has revealed two new druggable pockets at the protein target. The binding of suramin to one pocket, B RNA (Fig. 3), prevents the binding of the RNA template strand, while the binding of suramin to the other pocket, B NTP (Fig. 3), prevents both the entry of the nucleotide triphosphate into the catalytic site and the binding of the RNA processed strand (Fig. 3). However, despite a promising IC 50 value of 0.26 µM 36 , suramin is associated with a high risk of off-target effects    www.nature.com/scientificreports/ on other enzymes in the cell, together with its highly negative charge, which may hinder its penetration into cells 36 . Nevertheless, these two newly identified binding sites at RdRp are suitable pockets to target with nonnucleotide analog drug hits. Based on these structural findings, we investigated the possible binding modes and the protein-ligand interactions for luteolin and quercetin at the B RNA and B NTP pockets in RdRp. Specifically, the B RNA cavity is formed by the conserved G motif and the N terminus of B motif of the enzyme, and the key residues interacting with suramin are Asn497, Lys500, Arg569, Gln573 and Lys577. In contrast, the B NTP cavity is located near the catalytic active site, which is formed by conserved A, C, E, and F motifs. Here, the key interactions are formed between suramin and Lys551, Arg553, Arg555, Arg836, and Asp865 residues 36 .
To explore the binding modes for luteolin and quercetin, we performed molecular docking of both molecules on the two binding sites (i.e. B RNA and B NTP ) after removing the suramin (Fig. 4). The Schrödinger's Protein Preparation Wizard tool was used to prepare the protein, with the addition of hydrogens and the prediction of pKa values for ionizable residues. Subsequently, an extensive visual inspection was carried to check the overall quality of the final structures. Then, for the molecular docking, luteolin and quercetin were processed with the LigPrep tool 37 to properly prepare the ligands (e.g. assigning atom charges, converting 2D to 3D structures, and generating tautomeric and ionization states-at pH = 7.0 ± 0.4).
LigPrep generated four structures (i.e. two for luteolin and two for quercetin), which differ in the protonation state of the OH group at position 7 ( Fig. S1), namely LutOH, LutO -, QueOH, and QueO -. These structures were used for protein−ligand docking with Glide. The docking grid was centered on the suramin's center of mass, either bound to the B RNA or the B NTP pocket. Extra-precision Glide (XP) 38 was used and a maximum of 20 poses for each molecule were generated (for a total of 24 and 39 poses for B NTP and B RNA pockets, respectively). The resulting docking scores for luteolin and quercetin are shown in the Supporting Information (Table S1). For the B NTP binding pocket, the best docking scores (in kcal mol -1 ) correspond to -7.62 and -5.23, for QueOand LutOH molecules, respectively. At the B NTP pocket, the main interactions are formed between QueO -/LutOH molecules and His439, Ser549, Lys551, Arg555, Ser814, His816, and Arg836 residues (Fig. 4). In contrast, at the B RNA binding pocket, the predicted higher docking scores corresponded to -7.69 and -6.18, for QueOand LutOmolecules, respectively. The main interactions identified by the molecular docking are between the ligands and Asn496, Asn497, Lys500, Arg 569, Gln573, Lys577, and Tyr689 (Fig. 4).
To further check the stability of the docked structures, we ran equilibrium force-field-based MD simulations (~ 480 ns in total) of the four selected XP poses for luteolin and quercetin at the B RNA and B NTP binding pockets. In both B NTP -LutOH/QueOsystems, the overall ligand-enzyme complex stably maintains the interactions of the starting docking structure. The RMSD mean values for the heavy atoms of the complexes are 3.2 ± 0.5 Å and 3.0 ± 0.4 Å, respectively (Fig. 4). During the MD simulations of B NTP -LutOH system, the ligand showed some flexibility over time, with the RMSD mean value for the heavy atoms of 7.5 ± 1.3 Å (Fig. S2). This reflects the reorientation and reorganization of the interactions established between LutOH and the enzyme. Specifically, after the first ~ 50 ns, the ligand moved closer to the side chain of Ser814, Arg836, and Asp865, forming a stable network of interactions that was maintained during the simulations (Fig. S2). Interestingly, these residues also interact through H-bond with suramin in the crystal structure. Additionally, other crystal structures of the SARS-CoV-2 RdRp-RNA complex showed that Ser814 and Arg836 interact with the RNA primer strand 3,44-50 , further supporting the relevance of these residues for ligand recognition and binding. In contrast, B NTP -QueOsystem showed a slightly more stable conformation of the ligand within the pocket, with an RMSD value for the heavy atoms of 6.3 ± 0.9 Å (Fig. S3). Here, the interactions formed by QueOwith the RdRp enzyme involved His439, Ser549, Ser814, Arg836 side chains (Fig. S3) and the Ile548 backbone. Although His439, Ile548, and Ser549 do not directly participate in RNA binding, they are positioned within ~ 10 Å from the double-strand RNA and from the entry path of the incoming nucleotide.
For the B RNA -LutO -/QueOsystems, the overall ligand/enzyme complex showed no major differences in the overall stability, as reported by an RMSD mean value for the heavy atoms of 2.8 ± 0.3 Å and 2.6 ± 0.3 Å, respectively (Fig. 4). Here, both ligands showed a reduced flexibility compared to the B NTP -LutOH/QueOsystems, with a remarkable stability of the Que-Oligand at the pocket. Indeed, the RMSD mean values for the heavy atoms of LutO -/QueOare 6.4 ± 2.1 Å and 2.0 ± 0.8 Å, respectively (Figs. S4 and S5). This also reflects the stable H-bond interactions formed by QueOand RdRp enzyme. In detail, interactions are established between the side chains of Arg569, Gln573, and both LutOand QueOligands (Figs. S4 and S5), while only the latter interacts with the side chain of Asn497, Tyr689, and Ser759 (Fig. S5). Notably, considerable structural evidence (e.g. PDB 6XEZ, 7B3B, 7B3C, 7B3D) 49,50 shows that these three residues interact with the template strand, thus stabilizing the RNA substrate binding 3,44-50 . Moreover, Ser759 side chain is located close to the incoming nucleotide binding side 51 . These residues are therefore an optimal anchor point for inhibitors of the catalytic activity of the RdRp enzyme.
Overall, the results of our MD simulations indicate that both binding pockets may properly bind and stably host luteolin and quercetin. Nevertheless, the increased stability and higher number of contacts between the B RNA pocket and these ligands suggests that this binding site may be more suitable for ligand binding and structurebased drug design.
These computational insights will serve to start future campaigns for hit-to-lead design, as witnessed recently in computational studies used to guide experiments for drug design targeting viral proteins 52,53 . Notably, Jorgensen et al. 54 recently performed a virtual screening of ∼2000 approved drugs with a consensus virtual screening protocol used together with MD simulations and biochemical assay. This indicated 14 known drugs active in the micromolar range against 3CL pro . Starting from this evidence, the group subsequently applied free-energy perturbation (FEP) calculations to finetune the drug-target interaction of the initial hit, perampanel. This led to the design of a new set of compounds with IC 50 values in the low nanomolar range, whose binding poses have been corroborated by co-crystal structures 55 . With this successful example in mind, our results now form the basis for a hit-to-lead campaign targeting the SARS-CoV-2 RdRp enzyme.

Conclusions
In summary, starting from the pharmacological properties of flavonoids, we experimentally tested luteolin and quercetin against SARS-CoV-2 RdRp, a crucial target of the virus responsible for the COVID-19 pandemic. The IC 50 value in biochemical enzymatic assay is 4.6 ± 0.3 µM for luteolin and 6.9 ± 1.0 µM for quercetin. To the best of our knowledge, this is the first study that quantifies the inhibitory potency of luteolin and quercetin against RdRp, with the evidence of a one-digit micromolar range. Notably, this inhibitory activity is better than previous IC 50 values reported for these two compounds against other viral proteins of SARS-CoV-2, as M pro and 3CL pro . We also investigated and proposed potential binding modes of these compounds to the target protein. Thus, our experimental and computational results complete previous computational investigations that proposed these two known natural products against COVID, providing experimental values for activity and new mechanistic insights 35 . Taken together, our results endorse a further exploration of a new chromone class of RdRp polymerase inhibitors to treat Sars-CoV-2 and potentially other viral infections.

Methods
Biochemical assay. The natural flavonoids luteolin and quercetin were tested against SARS-CoV-2 RdRp with an in vitro enzymatic inhibition assay in collaboration with BPS Bioscience. The assay was performed with compounds obtained from commercial sources (luteolin from Fluorochem, quercetin from Sigma-Aldrich). Compounds purity is > 99% based on our HPLC analysis (see SI). The RdRp reactions were conducted in duplicate at 37 °C for 60 min in a 10 µl mixture containing assay buffer (20 mM Tris pH8.0 and 0.01% Triton X100), RNA duplex, ATP substrate and enzyme, and the test compound. The enzyme was produced by BPS Bioscience, and was formulated as 45 mM Tris-HCl pH 8.0, 124 mM NaCl, 2.4 mM KCl, 4 mM MgCl 2 , 1 mM TCEP, 10% glycerol. Typical purity was 95-97%, and typical concentration was 1 mg/ml. These 10 µl reactions were carried out in wells of 384-well Optiplate (PerkinElmer). A 10 mM stock solution of test compound in DMSO was prepared. Dilutions of this stock solution were prepared in assay buffer (5% DMSO concentration) and 2 µl www.nature.com/scientificreports/ of the dilution was added to a 6 µl of RdRp (final concentration 0.08 mg/mL) containing RNAse inhibitor for preincubation (30 min at room temperature with slow shaking). Reaction was started by addition of 2 µl of the substrate mix containing RNA duplex (40 nM) and ATP substrate (3 µM). Final concentration of DMSO was 1% in all reactions (reference compound-0% DMSO). After enzymatic reactions, 10 µl of anti-Dig Acceptor beads (PerkinElmer, diluted 1:500 with 1 × detection buffer) were added to the reaction mix. After brief shaking, plate was incubated for 30 min. Finally, 10 µl of AlphaScreen Streptavidin-conjugated donor beads (Perkin, diluted 1:125 with 1 × detection buffer) were added. In 30 min, the samples were measured in AlphaScreen microplate reader (EnSpire Alpha 2390 Multilabel Reader, PerkinElmer). In the absence of the compound, the intensity (C e ) in each data set was defined as 100% of activity. In the absence of the enzyme, the intensity (C 0 ) in each data set was defined as 0% of activity. The percent activity in the presence of each compound was calculated according to the following equation: % activity = (C-C 0 )/(C e -C 0 ), where C is the intensity in the presence of the compound. As a positive control, the reference compound 6-chloropurine-ribose TP was tested at three different concentrations (0.02 µM, 0.2 µM, and 2 µM).

Molecular docking of luteolin and quercetin with SARS-CoV-2 RdRp. First, the SARS-CoV-2
RdRp was retrieved from PDB database (PDB ID 7D4F 1 ) and subsequently prepared for docking. The preparation was carried by Schrödinger's Protein Preparation Wizard tool and included: (i) addition of hydrogen atoms, (ii) elimination of water molecules not involved in ligand-binding interaction, (iii) assignment of atomic charges. Subsequently, energy minimized 3D molecular structures of luteolin and quercetin were generated and prepared for docking using LigPrep tool 37 . Additionally, possible ionization states were generated using LigPrep tool 37 , thus resulting in tow possible states for each molecules (see Fig. S1). Eventually, the SARS-CoV-2 RdRp structure (PDB ID 7D4F 36 ) was used for docking luteolin and quercetin. The grid was centered on the suramin's center of mass, and the docking was performed using Glide XP methodology 38,56 . Structural models for molecular dynamics simulations. We used four different systems for the MD simulations: (i) B RNA -LutO -, (ii) B RNA -QueO -, (iii) B NTP -LutOH and (iv) B NTP -QueO -. Each system was solvated with a 12-Å layer of TIP3P water molecules, and Na + ions were added to neutralize the net charge of the systems. The final models include a total of ~ 183,000 atoms. www.nature.com/scientificreports/ Molecular dynamics simulations set-up. We used force-field-based MD simulations to check the stability of the docked structures. Here, the AMBER-ff14SB 57 force field was used to treat the RdRp enzyme. All four ligands were parametrized with the general Amber force field (GAFF) 58 , and the atomic charges were derived using the RESP procedure, according to the Merz − Singh − Kollman scheme 59 . Na + metal ions were treated using the Joung-Cheatham parameters 60 . A time integration step of 2 fs was used and the lengths of all bonds involving hydrogen atoms were constrained using the P-LINCS algorithm 61 . A velocity-rescaling thermostat was used to set a system temperature of 310 K 62 , while the Parrinello-Rahman barostat maintained a constant pressure of 1 bar 63 . Long-range electrostatic interactions were calculated with the particle mesh Ewald (PME) method using a Fourier grid spacing of 1.6 Å. Periodic boundary conditions in the three directions of Cartesian space were applied. All MD simulations were performed with Amber2020. The systems were all subject to the same MD simulations procedure. First, we carried out energy minimization to relax the water molecule and the ions. Here, both the ligand and the RdRp backbone were kept fixed with harmonic positional restraints of 300 kcal mol -1 Å 2 . Then, the systems were heated up from 0 to 310 K with NVT simulations for a total of 1 ns with 300 kcal mol -1 Å 2 restraints on the ligand. Additionally, 1 ns of simulations in NPT ensemble was performed with the same positional restraints used in the NVT simulations. Three additional NPT simulations of 1 ns each were performed gradually removing the restraints on the ligand. Finally, a production run were performed in the NPT ensemble for each system. We collected overall ~ 480 ns of MD trajectories, specifically ~ 120 ns for each system.