In silico analysis of the antidepressant fluoxetine and similar drugs as inhibitors of the human protein acid sphingomyelinase: a related SARS-CoV-2 inhibition pathway

Abstract Acid Sphingomyelinase (ASM) is a human phosphodiesterase that catalyzes the metabolism of sphingomyelin (SM) to ceramide and phosphocholine. ASM is involved in the plasma membrane cell repair and is associated with the lysosomal inner lipid membrane by nonbonding interactions. The disruption of those interaction would result in ASM release into the lysosomal lumen and consequent degradation of its structure. Furthermore, SARS-CoV-2 infection has been linked with ASM activation and with a ceramide domain formation in the outer leaflet of the plasma membrane that is thought to be crucial for the viral particles recognition by the host cells. In this study, we have explored in silico the behavior of fluoxetine and related drugs as potential inhibitors of ASM. Theoretically, these drugs would be able to overpass lysosomal membrane and reach the interactions that sustain ASM structure, breaking them and inhibiting the ASM. The analyses of docking data indicated that fluoxetine allocated mainly in the N-terminal saposin domain via nonbonding interactions, mostly of hydrophobic nature. Similar results were obtained for venlafaxine, citalopram, atomoxetine, nisoxetine and fluoxetine’s main metabolite norfluoxetine. In conclusion, it was observed that the saposin allocation may be a good indicative of the drugs inhibition mechanism, once this domain is responsible for the binding of ASM to lysosomal membrane and some of those drugs have previously been reported to inhibit the phosphodiesterase by releasing its structure in the lysosomal lumen. Our MD data also provides some insight about natural ligand C18 sphingomyelin conformations on saposin. Communicated by Ramaswamy H. Sarma


Introduction
The COVID-19 (Corona Virus Disease 2019) upcoming is caused by SARS-CoV-2 (Severe Acute Respiratory Syndrome Coronavirus 2).So far, more than 540 million people were infected and more than 6.5 million died (data from https:// coronavirus.jhu.edu/map.html,last access: 09 Oct 2022).Consequently, the global scenario requests urgent researching for alternatives on coronavirus treatment.Repurposing already licensed drugs is a short-term approach (Cavasotto & Di Filippo, 2021;Dechaumes et al., 2021;Koulgi et al., 2020), since these compounds are already available on market and most of the synthesis routes are currently known and optimized, enabling massive production if necessary (Pushpakom et al., 2019).
Recently, a clinical study with 7.345 hospitalized adults with moderate and severe COVID-19 indicated that some antidepressants use was associated with reduced inflammatory response in COVID-19 patients (although this trial is being contested, based on a larger database (Rauchman et al., 2021)).Specifically, fluoxetine, venlafaxine and citalopram consumption has been associated with a reduced risk of intubation or death in small retrospective studies (Hoertel, S� anchez-Rico, Gulbins, et al., 2021;Hoertel, S� anchez-Rico, Vernet, et al., 2021;Hoertel et al., 2020).Fluoxetine and citalopram are both antidepressant members of the Selective Serotonin Reuptake Inhibitors (SSRIs) class (DeVane, 1992;Kaserer et al., 2015), whereas venlafaxine is a Serotonin-Norepinephrine Reuptake Inhibitor (SNRI) (Joffe et al., 2014).Furthermore, another small study (165 patients from The Rockland Psychiatric Center) has suggested that the use of antidepressants on an ambient where the potential for exposure is likely uniform could drive a protective effect against SARS-CoV-2 contagion.Two drugs (fluoxetine and trazodone) have shown the greatest results (Clelland et al., 2022).
Lately, a mechanism for SARS-CoV-2 inhibition by fluoxetine, and other small compounds that functionally can inhibit the endolysosomal acid sphingomyelinase (ASM), has been hypothesized (Schloer et al., 2020).Basically, the disruption of the normal acid pH of the lysosomal environment by ASM inhibition could hamper the SARS-CoV-2 entry on host cells (Schloer et al., 2020).ASM is a lysosomal phosphodieserase that catalyzes the hydrolysis of sphingomyelin (SM) to produce ceramide and phosphocholine (Xiong et al., 2016).Recently, ASM has been proposed to play an important role as a modulator of receptor signaling and infection biology, including the COVID-19 disease (Carpinteiro et al., 2020).
Moreover, fluoxetine has been indicated as a potential negative modulator of the cytokine storm by decreasing interleukin 6 (IL6) signal transduction on COVID-19 cases (Creeden et al., 2021).Thus, the apparent beneficial effects of fluoxetine in the COVID-19 (Hoertel, S� anchez-Rico, Gulbins, et al., 2021) could be mediated by at least two different mechanisms, i.e. by inhibiting the entrance of the virus in the host cells and by lessening the inflammatory response.
ASM functional inhibitors (FIASMAs) have been considered as part of the drug repurposing strategy for either the SARS-CoV and SARS-CoV-2 viruses.FIASMAs are cationic amphiphilic molecules, usually polycyclic compounds and have at least one nitrogen atom.Thereby, fluoxetine and citalopram have already been recognized as FIASMAs (Carpinteiro et al., 2020;Kornhuber et al., 2010).In contrast, venlafaxine was not reported so far as a FIASMA, but it has all the described characteristics on its molecular structure.

Materials and methods
AutoDock Vina was used for the docking simulations (Trott & Olson, 2012), with exhaustiveness of 50.The human ASM crystallographic structure was obtained from the Protein Data Bank (PDB) server with the code 5jg8 (Xiong et al., 2016; the 5jg8 structure is presented as a dimer, but we used a single monomer in the analyses).The residues protonation states were determined with the Hþþ server (pH ¼ 5.00) (Anandakrishnan et al., 2012;Gordon et al., 2005), because this application has already been used for structural exploration of ASM (Vadlamudi et al., 2016).Water, ions, ligands, and other small molecules were removed from the x-ray protein structures, while the hydrogen atoms were added using the CHIMERA program, followed by 100 steps of energy minimization (Pettersen et al., 2004).The zinc (II) partial charges (Zn1 ¼ 0.6632; Zn2 ¼ 0.2651) were obtained from a semi-empirical calculation (PM6) taking into account the residues at 5 Å around the two Zn(II) ions.Only the hydrogen atoms were optimized while the heavy atoms were considered fixed.
A chiral carbon is observed in the drugs structure.Consequently, the isomers R and S were included in the three-dimensional models of fluoxetine, venlafaxine, citalopram, atomoxetine, nisoxetine and norfluoxetine.For each molecule, the 20 best energy conformers (in terms of DG) were analysed in Discovery Studio Visualizer (including the formation of the non-bonding interactions) (BIOVIA, Dassault Syst� emes).The details of interaction of those drugs and metabolite were observed at all the conformers, but especially for those which displayed the best interaction with ASM residues and those that were considered the most representative.
To determine the most representative conformer, the heavy atoms RMSD (root mean square deviation) was calculated and the value of 2 Å was used to evaluate the most populous cluster (Chang et al., 2010).The conformer with the lowest energy from the most populous cluster was chosen.The most representative conformers were the 5 th for C18 SM, the 2 nd for (R)-fluoxetine, the 6 th for (S)-fluoxetine, the 6 th for (R)-venlafaxine, the 2 nd for (S)-venlafaxine, the 2 nd for (R)-citalopram, the 5 th for (S)-citalopram, the 1 st for (R)atomoxetine, the 8 th for (S)-atomoxetine, the 3 rd for (R)nisoxetine, the for 6 th (S)-nisoxetine, the for 6 th (R)-norfluoxetine and the 5 th for (S)-norfluoxetine.
Molecular dynamics simulations were run for a selection of compounds including (S)-fluoxetine, (S)-norfluoxetine and C18 SM (chosen based on docking results) employing AMBER2021 (Case et al., 2021).The complexes were treated using the AMBER ff14SB force field for the protein residues and the generalized AMBER force field (GAFF) to define the ligands' parameters.The structures were solvated in an octahedral box of TIP3P water molecules.Starting from the best poses obtained via the docking procedure and after energy minimization, heating to 310 K at constant volume and pressure was performed over 60 ps using the Langevin thermostat.Afterwards equilibration at constant temperature (310 K) and pressure (1 bar, Berendsen barostat) was conducted for 60 ps using weak restraints (2 kcal mol À 1 Å À 2 ) on the proteinligand complex and then for 2 ns without any restraint, followed by 200 ns production runs.500 frames from the MD trajectories taken at 0.4 ns intervals were extracted and employed in the calculation of binding energies (DG bind ) using the molecular Mechanic/Poisson-Boltzmann Surface Area (MM-PBSA) method.The MM-PBSA.py script implemented in AMBER20 was used to perform the binding energies calculations.Conformational entropy is computationally expensive and thus ignored (Hou et al., 2011;Oehme et al., 2012).Hou et al. (2011) showed that including the entropic contribution in the DG bind calculations does not guarantee the accuracy of the DG bind .

Drugs protonation state at the intracellular and intralysosomal pH
The protonation state of fluoxetine, venlafaxine, citalopram, norfluoxetine, atomoxetine and nisoxetine were simulated at pH 5.0 (which is the acid pH of lysosome lumen) (Carpinteiro et al., 2020;Reijngoud & Tager, 1973) and 7.0-7.4(which is the intra and extracellular values found in the majority of mammalian cells) (Madshus, 1991).
The predominant structures at pH 5.0 to 7.4 are depicted in Figure 1.Details about the protonation states of all the tested compounds can be found in the Supplementary Figures 9-14.The structures presented in Figure 1 were geometrically optimized and used in the docking analyses.

ASM structure and docking results
Human ASM structure has been reported with resolutions of 2.8 Å (Xiong et al., 2016) and 2.5 Å (Zhou et al., 2016).Both crystals share several characteristics and the structure deposited by Xiong et al. (2016) is available as a dimer.Here we have chosen the monomer A for the analyses, since the monomeric form is predominant and docking analysis have previously been done with this structure (Xiong et al., 2016).There are three main regions in the enzyme: the N-terminal saposin domain, the proline-rich connector, and the catalytic domain (Zhou et al. divides the last in two: a catalytic metallophosphatase domain and a helical C-terminal domain (Zhou et al., 2016; Figure 2).
We have run docking simulations for the natural ASM ligand C18 sphingomyelin (C18 SM), because this is one of the main forms of sphingomyelin (SM) in biological membranes (Barenholz & Thompson, 1999; see Figure 3 for the structure).The results indicated that C18 SM interacted predominantly with the saposin domain (19 out of the 20 most thermodynamically favored conformers, that is, 19/20 or 95%).The presence of several hydrophobic interactions were evident (see Figure 4 for the most representative conformer of C18 SM interacting with ASM saposin domain and Table 1 for all the interactions).
The general pattern of interaction between (R)-fluoxetine/ (S)-fluoxetine with the saposin domain of human ASM is present in Figure 5.For venlafaxine and citalopram, the figures and tables are available on supplementary information (Supplementary Figures 1-4).
The main metabolite of fluoxetine (the N-demethylated form, called norfluoxetine) (Altamura et al., 1994;Flores et al., 2005) and drugs of similar chemical structure have been selected as potential ligands of ASM.The Drug Bank server was used to search for potential fluoxetine structurally related compounds (https://go.drugbank.com,last access: 13 Jul 2021).The search in the data base yielded 3 similar compounds: seproxetine, atomoxetine and nisoxetine.Since seproxetine is the (S)-isomer of norfluoxetine (Risley et al., 1996), the new obtained compounds were atomoxetine and nisoxetine.

(S)-fluoxetine, (S)-norfluoxetine and C18 SM molecular dynamics (MD)
To provide better insight about molecular arrangement and interactions on ASM, three MD trajectories computed for (S)-fluoxetine, (S)-norfluoxetine and C18 SM complexes with ASM protein were analyzed on the basis of the root mean square deviation (RMSD) of the whole system during the simulations and of the RMSD of the ligand with respect to the docking pose (Figure 8   Table 5. Relevant energies results from blind docking (in kcal/mol) of C18 SM, atomoxetine, nisoxetine and norfluoxetine with ASM monomeric structure.
Docked drugs/molecules ASM most stable conformer (conformer 1) ASM least stable conformer (conformer 20) Most representative conformer energy C18 SM À 5.7 À 5.3 À 5.6 (R)-Atomoxetine À 6.2 À 5.4 À 6.2 (S)-Atomoxetine À 7.7 À 5.7 À 6.2 (R)-Nisoxetine À 7.6 À 5.6 À 6.0 (S)-Nisoxetine À 7.6 À 5.6 À 6.1 (R)-Norfluoxetine À 8.6 À 5.9 À 6.4 (S)-Norfluoxetine À 8.8 À 6.7 À 7.3 stable value is reached for the rest of the simulation.By inspecting the geometries along the trajectory, a definite change in ligand structure from the docking pose is seen in the MD simulation (Figure 8(d)).Specifically, if the alkyl chains of C18 SM are predicted to be folded to make the molecule more spherical on docking, after MD simulations these chains are found to be almost extended, with the ligand taking a more cylindrical shape, probably to favorably expose the hydrophobic portion of the molecule to interact with the inner hydrophobic portion of saposin domain.
Although fluoxetine and norfluoxetine are considered amphiphilic, they both do not present large alkyl chains like C18 SM; therefore, they cannot achieve this linear conformation.Furthermore, ligand binding energies for the three selected molecules were computed with the MM-PBSA method.Results (Figure 9 and Table 10) show that C18 SM is the ligand which displays the larger (most negative) binding energy with a value of À 7.80 kcal mol À 1 but at the same time also the highest energy fluctuations, followed by (S)-fluoxetine and lastly (S)-norfluoxetine which is computed to have the weakest interaction with the ASM protein.
To estimate the total binding free energy of (S)-fluoxetine, (S)-norfluoxetine and C18 SM, DG bind , is decomposed into DG solv and DG gas .E PB , E Polar and E disp terms are the polar and non-polar contributions to the solvation free energy (DG solv ).The terms are typically obtained by solving the PB equation, whereas the non-polar term is estimated from a linear relation to the solvent accessible surface area (SASA).
The E vdW and E elec terms dominate the net binding free energy (DG bind ) except in the case of (S)-norfluoxetine with DG bind À 0.84 kcal mol À 1 and DG solv of 105.30 kcal mol À 1 .The standard deviation of DG bind over the 500 snapshots in (S)fluoxetine, (S)-norfluoxetine and C18 SM is 4.95 kcal mol À 1 , 5.64 kcal mol À 1 and 8.13 kcal mol À 1 , respectively, giving a standard error of 0.22 kcal mol À 1 , 0.25 kcal mol À 1 and 0.36 kcal mol À 1 .
The per-residue energy decomposition analysis has been also carried out using two different methods: molecular   mechanics Poisson À Boltzmann surface area (MM/PBSA) and molecular mechanics generalized Born surface area (MM/ GBSA) (Wang et al., 2019) for (S)-fluoxetine, (S)-norfluoxetine and C18 SM.The graphics for both methods are shown in Figure 10.

ASM as a target for SARS-CoV-2 inhibition
Recently, a mechanism of inhibition was reported to explain the potential role of ASM in SARS-CoV-2 infection.The authors have proposed that the presence of the spike protein can trigger the activation of ASM (the interactions between ACE2 and spike seen to be profoundly related to ASM activation).Then, the activated enzyme over synthesizes ceramide, which forms a ceramide-rich domain in the outer leaflet of the lysosomal membrane and moves to the plasma membrane.This formation was proposed to facilitate SARS-CoV-2 infection once it could be used to signalize host cells for the virus.In accordance, the pharmacological inhibition of ASM decreased the infection of human cells with SARS-CoV-2 (Carpinteiro et al., 2020).Alternatively, another mechanism has been suggested.The down regulation of ASM has been demonstrated to cause a cholesterol accumulation in the lysosomal membrane (Schloer et al., 2020).The elevated cholesterol levels decreased the activity of the vacuolar-type membrane ATPase (v-ATPase), increasing the pH values in the lumen of lysosomes (Lafourcade et al., 2008;Marshansky & Futai, 2008).Taking into consideration that SARS-CoV-2 is an enveloped virus and uses the endocytic pathway uptake to enter the host cells (Ou et al., 2020), the pH drop is vital for the endocytosed virus leave the late endosome/lysosome at the right time (White & Whittaker, 2016).Thus, the inactivation of ASM by FIASMAs could interfere on the virus replication mechanism by inhibiting the lysosomal proton pump.
Functional inhibitors of acid sphingomyelinase (FIASMAs) disrupt the electrostatic interactions that maintain ASM bound to the inner part of lysosomal membrane.The enzyme is then released to the lysosomal lumen, and its proteolytic degradation follows (Le Corre Pharm & Loas, 2021).FIASMAs are small polycyclic cationic amphiphilic molecules that present at least one nitrogen (Le Corre Pharm & Loas, 2021) that can be protonated at cellular pH.Fluoxetine and citalopram compounds are well-known examples of FIASMAs (Carpinteiro et al., 2020), but venlafaxine, atomoxetine and nisoxetine have not yet been formally reported as a FIASMA, though possessing all the features of the drug class.In accordance, our docking results indicated that these compounds interacted favorably with human ASM and in a similar way to fluoxetine and the natural ligand C18 SM.
MD simulations on a charged lipid bilayer and ASM structure have been performed and indicated that ASM interacts with the lysosomal lipid bilayer via its saposin domain (Xiong et al., 2016).Additionally, the interactions partially orientate the catalytic site of ASM (a zinc dyad) to the membrane lipids, facilitating the SM extraction from bilayer (Xiong et al., 2016).Therefore, FIASMAs are supposed to disrupt the ASMmembrane lipid interactions near the saposin domain.Furthermore, saposin helix a3 (residues 128-149) mediates catalytic site-saposin domain association, which is responsible for transporting the extracted SM to the catalytic zinc centers (Xiong et al., 2016;Zhou et al., 2016).Thus, drugs that interact with helix a3 may not only release ASM to lumen, but also interfere on saposin proper SM transport function.
Our docking data suggests that fluoxetine and the majority of its analogs interact in a stable and preferential way with the saposin domain or on the interface with active site of human ASM (see section 3.2 and 3.3).Of note, one of the endogenous substrates of ASM, C18 SM also interact with a clear preference for the saposin domain (19/20 conformers, 95%).It is also interesting to observe that fluoxetine and analogs interact in a more thermodynamically favored way with the ASM than C18 SM (see Supplementary Table 1), so we hypothesize that they might have a preference for interacting with ASM over the endogenous substrate of the enzyme.On the other hand, the MD data suggest that the SM molecule could suffer some drastic conformational changes and have a better DG than (S)-fluoxetine and (S)norfluoxetine.
Saposin domain helices adopt a V-shape conformation formed by the four helices.This open conformation exposes an inner hydrophobic concave surface (Xiong et al., 2016), and C18 SM molecule presents amphiphilic properties.So, the formation of several hydrophobic interactions between the enzyme and the ligand was expected, which is confirmed by dockings results (Figure 4, Table 1).The same chemical behavior is shared by the antidepressants studied here (Table 3, Figure 5, Supplementary Figures 1-6, Supplementary Tables 2-7), a good indicator that the allocation of these drugs on saposin domain can contribute to ASM inactivation.Furthermore, the presence of a significant number of non-bonding interactions between fluoxetine and analogs with ASM can explain the stable proteinligands formation.

Conclusion
Structurally, the docking and MD results here reported suggest that the selected antidepressants are able to interact with the saposin domain of human ASM.In view of the inhibitory effects of fluoxetine against SARS-CoV-2 in vitro, the results presented for the fluoxetine analogs (atomoxetine, nisoxetine), venlafaxine and metabolite (norfluoxetine) indicate that they should be tested in vitro as inhibitors of

Figure 4 .
Figure 4. C18 sphingomyelin (5 th most stable conformer and the most representative) interacting with saposin domain.Light purple dashed lines indicate the formation of hydrophobic interactions, whereas green represents hydrogen bonds.The hydrogen bond angle was measured to 3.53 � .
at intralysosomal pH (check section 3.1), they are polycyclic compounds and have one nitrogen atom (Le Corre Pharm & Loas, 2021).Docking results indicated that both drugs and metabolite interacted with the ASM monomer in a similar way to fluoxetine, venlafaxine and citalopram.In fact, the majority of the 20 conformers was distributed in the saposin domain ((R)atomoxetine: 18/20; (S)-atomoxetine: 14/20; (R)-nisoxetine: 14/20; (S)-nisoxetine: 09/20; (R)-norfluoxetine: 13/20; (S)-norfluoxetine: 13/20).As observed for fluoxetine, venlafaxine and citalopram, the preponderance of hydrophobic interactions was also observed for atomoxetine, nisoxetine and norfluoxetine (Figures (a)).Notably, RMSD stability is achieved faster in the C18 SM complex (at around 100 ns) than with either (S)-fluoxetine or (S)-norfluoxetine in whose cases a RMSD increase is seen until 150 ns followed by fluctuations around a mean value of 4.1 Å in case of (S)-fluoxetine and of 3.3 Å for (S)-norfluoxetine.Root mean square fluctuation (RMSF) show that for all complexes the highest fluctuations, found in the initial part of the chain, do not differ significantly from one another meaning that the different structure of C18 SM and the fluoxetine derivatives does not change the overall structure of the protein.Focusing on the ligand dynamics, (S)-fluoxetine and (S)norfluoxetine behave very similarly displaying an RMSD with respect to the docking structure that, after a significant change during energy minimization, heating, and equilibration, remains almost constant throughout the simulations.On the other hand, C18 SM shows a large increase in RMSD (occurring between around 20 and 40 ns) after which a

Figure 5 Figure 6
Figure 5. a) (R)-fluoxetine (2 nd most stable and most representative conformer) and b) (S)-fluoxetine (6 th most stable most representative conformer) molecules interacting with saposin domain.Light purple dashed lines indicate the formation of hydrophobic interactions, cyan dashed lines fluorine interactions and yellow dashed lines sulfur interactions.

Figure 7 .
Figure 7. a) (R)-norfluoxetine (6 th most stable and most representative conformer) and b) (S)-norfluoxetine (6 th most stable and most representative conformer) conformers interacting with ASM Saposin.The light purple dashed lines indicate hydrophobic bonds, the green dashed lines hydrogen bonds and the cyan dashed lines fluorine interactions.The hydrogens bonds angle from (R)-norfluoxetine were measured and were 100.60 � and 112.30� .distances were 2.89 Å and 2.70 Å, respectively.

Figure 8 .
Figure 8. RMSDs (a), RMSFs (b), superposition of the crystal structure with the docked ligand (blue) and the one at the end of the molecular dynamics (yellow) (c) and comparison of the ligands positions between the docking (blue) and the last frame of the molecular dynamics (yellow) (d) for the three investigated ligands.

Figure 9 .
Figure 9. Binding energy along the molecular dynamics trajectory.

Table 1 .
The main types of interactions of C18 sphingomyelin with ASM saposin domain.

Table 2 .
Relevant energies results from the blind docking (in kcal/mol) of C18 SM, fluoxetine, venlafaxine and citalopram with ASM monomeric structure.

Table 10 .
Average binding energy calculated (DG bind ) with the MM-PBSA method and its standard deviation.Energy term DG/kcal mol -1 SD DG/kcal mol -1 SD DG/kcal mol -1 SD