Anesthetics can be classified into three types: general, local, and regional (Fig. 1). General anesthetics are mostly halogenated ethers, such as isoflurane, sevoflurane, desflurane, and enflurane. Inhaled anesthetics diffuse to the blood and influence the central nervous system by suppressing nicotinic acetylcholine (nACh) receptors and inhibiting spinal cord function, subsequently causing paralysis and immobility (Antognini and Schwartz, 1993; Rampil et al., 1993; Campagna et al., 2003; Grasshoff et al., 2005; Hentschke et al., 2005). General anesthetics are known to target inhibitory or excitatory ion channel receptors by prolonging the activation of inhibitory channel receptors, including glycine and γ-aminobutyric acid type A (GABAA) receptors (GABAARs), and depressing the activity of excitatory channel receptors, including serotonin, nACh, and glutamate receptors (Torri, 2010).

Although GABAARs, glycine receptors, nACh receptors, and type 3 serotonin (5-HT3) receptors are categorized as pentameric ligand-gated ion channels (pLGICs), general anesthetics generally conduct chloride ions into cells to reduce neuronal excitability in GABAA and glycine receptors but conduct cations in nACh and 5-HT3 receptors (Smart and Paoletti, 2012; Fourati et al., 2018). Moreover, in spinal motor neurons, although inhibitory glycine receptors are always stimulated by the general anesthetics, resulting in the suppression of both postsynaptic N-methyl-D-aspartate (NMDA) receptors and α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors, GABAAR responses can vary among specific channel subtypes or drugs (Cheng and Kendig, 2000). Accordingly, GABAARs, are differently modulated by general anesthetics as they are inhibited rather than potentiated by anesthetics (Howard et al., 2014). Meanwhile, nACh and 5-HT3 receptors that are conventionally inhibited are activated by some anesthetics, such as ethanol (Howard et al., 2014; Fourati et al., 2018). This is a complicated but important aspect of the structure-function relationship of general anesthetics in terms of their capacity to regulate receptors in contradictory ways. Clinically relevant concentrations of enflurane and isoflurane enhanced signaling by GABA and glycine receptors (Mihic et al., 1997). Two amino acid residues buried in the transmembrane domains (TMDs) were recently discovered to be critical for the allosteric modulation of both the glycine receptors and GABAARs triggered by enflurane and isoflurane (Mihic et al., 1997). The GABAAR has also been suggested as a binding target of propofol (Krasowski et al., 2001; Kuo et al., 2005; Chiara et al., 2014). The experiment using photolabeling of a propofol analog suggested that the binding site of propofol would be located in the β subunit of the GABAAR at the boundary between the TMDs and the extracellular domain (ECD) (Yip et al., 2013). The GABAAR is one of the targets for anesthetics, including volatile and intravenous types, or neurosteroids and drugs such as benzodiazepines (BZDs) (Sieghart, 1995; Lobo and Harris, 2008; Zhu et al., 2018). Some studies have suggested that increased agonist efficacy, even though the enhanced agonist binds at the orthosteric GABA binding site, causes propofol and barbiturates to potentiate the GABAAR responses to partial agonists (O’Shea et al., 2000; Steinbach and Akk, 2001). Therefore, the interaction of anesthetic ligands and GABAARs needs to be determined to explain the anesthetic mechanism. In humans, the GABAAR has a total of 19 subunits that are encoded by nineteen different genes: α1–6, β1–3, γ1–3, δ, ε, θ, π, and ρ1–ρ3. Most physiologically predominant heteromeric synaptic isoform formats are thought to be composed of two α1-subunits, two β2-subunits, and one γ2-subunit (Sigel and Steinmann, 2012). Common synaptic GABAARs are heteropentamers with an αβγ, 2:2:1 stoichiometry (Alkire et al., 1997; Eckenhoff, 1998; Cheng and Kendig, 2000). GABAAR exists as a homopentamer (Murlidaran et al., 2019; Eckenhoff et al., 2000), and 1:2:2 heteropentamer (Bhattacharya et al., 2000; Vemparala et al., 2010). Five subunits are assembled around a pivotal pore and each subunit is composed of an ECD, including ligand binding sites, and a TMD, including M1–M4 helixes (Murlidaran et al., 2019).

The three-dimensional (3D) binding characteristics of ligands within the GABAARs should be identified using structural data to understand the action mechanism of general anesthetics. The lack of structural information has prevented a complete understanding of how anesthetics modulate the mode of action of GABAergic neurotransmission (Forman and Miller, 2011; Wei et al., 2011). Before the crystal structure of the GABAAR was discovered, other pLGICs with homology to the GABAAR and resolved 3D structures were used to determine the binding site of anesthetics (Nury et al., 2011; Chiara et al., 2014). To specify these molecular characteristics, the structures of a protein bound to anesthetics require a resolution of 2 Å (Forman and Miller, 2011). Moreover, to elucidate the conformational changes, the redundant structures of the same part need to be determined in several modes (Forman and Miller, 2011). However, crystallographic data of mammalian pLGICs with 2 Å resolution were not available until 2011 (Forman and Miller, 2011). The cryo-electron microscopy (cryo-EM) structure of the Torpedo acetylcholine receptor muscle subtype constructed in the resting mode without agonist at 4 Å resolution was the only available structural data of a vertebrate (Unwin, 2005). However, this low resolution is incapable of elucidating the elaborate binding residues of anesthetics at side chains. The other structures determined for pentameric ion-channel receptors were the superfamily of bacterial members and included the TMD without an intracellular domain (Bocquet et al., 2009; Hilf and Dutzler, 2009). Since 2012, various crystal structures of the ligand-binding domain of cysteine-loop receptor homologs such as the acetylcholine-binding protein (AChBP), the nAChR, and bacterial homologs have been solved (Richter et al., 2012). GABA is suggested to be located between β–α-subunits and BZDs are located between the α–γ subunits (Sigel and Buhr, 1997).

Furthermore, functional studies using mutagenesis have suggested the binding sites for GABA/BZD ligands and many other compounds on GABAARs. However, information on the physiological GABAAR has been rarely reported (Whiting et al., 1995; Forman and Miller, 2011; Olsen, 2015). Recently, many studies have focused on the structural determination of GABAAR, and the interaction between GABAAR and anesthetic drugs have been elucidated. Most GABAARs of the human brain in the physiological state are heteromers of two α subunits (α1–6), two β subunits (β1–3), and one γ (γ1–3) or δ subunit, out of 19 possible types (Sigel and Steinmann, 2012; Dore et al., 2014). However, further studies are needed to elucidate the previous results on the specificity of GABAA ligands toward the distinct subtypes of the GABAAR of α1, α2, α3, α4, α5, and α6. Classical BZDs, including diazepam and alprazolam, act specifically on GABAAR subtypes containing α1, α2, α3 or α5, but not α4 or α6 (Tan et al., 2011). Therefore, various combinations of ligands on more GABAARs with various subtypes need to be analyzed to understand the elaborate action mechanism of GABAARs. This explains the necessity to identify the binding site of anesthetics to GABAARs using updated and elaborate 3D structural models with more types of anesthetics. The crystallographic data of GABAAR structure determined earlier was that of β3 homopentamer in an agonist-bound state, which provided the initial guide to explain the physiological receptor (Miller and Aricescu, 2014). Photoaffinity labeling to resolve the binding site for propofol in GABAARs was performed using partially expressed GABAARs consisting of both β3 homopentamers and α1β3 heteropentamers (Yip et al., 2013). Recently, the structures of apo- and neurosteroid-bound GABAAR constructs containing an α subunit TMDs were characterized and provided (Miller et al., 2017). Later, high-resolution cryo-EM structures of the human α1β2γ2 GABAAR were provided and applied to our analysis. The α1β2γ2 isoform is a predominant form in the adult brain, and the complex of this isoform with GABA and flumazenil, the BZD antagonist, has been reported (Zhu et al., 2018). Because this α1β2γ2 isoform is a physiological GABAAR, the current high-resolution structural information finely featured the association of the heteromeric subunit and the molecular interactions of the receptor with ligands such as GABA and flumazenil. In the most up-to-date GABAA construct of β2α1β2α1γ2 form, five subunits showed a pseudo-symmetrical pattern of assembly around an integral ion channel (Zhu et al., 2018). Viewing from the extracellular space, the ring of subunits is arranged as β2α1β2α1γ2 in a counter-clockwise direction (Tretter et al., 1997; Baumann et al., 2002; Baur et al., 2006; Zhu et al., 2018). There were two ECD interfaces in α1–β2 and γ2–β2, but no ligand was bound to these clefts, which are considered putative targets for designing novel drugs (Zhu et al., 2018). According to the crystallographic model comprising the α1, β2, and γ2-subunits with GABA and flumazenil, GABA is located in the conventional neurotransmitter ligand binding site of the pLGIC superfamily at the two β2α1 interfaces in the ECD (Zhu et al., 2018). Flumazenil was found to bind at a similar form of the cleft site at the α1γ2 interface of GABARs in this model.

We used these current architectures to analyze the putative binding sites of isoflurane, enflurane, halothane, and propofol. Docking analysis of these anesthetics was performed against the entire structure and not merely the three putative ligand binding sites of recent GABAA structures. We also used these architectures to analyze the putative binding sites of various types of BZD. Docking analysis of the aesthesia was performed against a receptor construct containing ECD, TMD, and TMD hole. Docking with volatile anesthetics incurred numerous false positives because the binding mode determined by docking was unstable or short-lived (Murlidaran et al., 2019). To reduce the rate of false positives, a receptor construct containing ECD, TMD, and TMD hole was used for docking analysis in this work. Rather than being defined to the existing ligand binding site, the docking algorithm was set to predict the probability of the preferable orientation of a ligand when it was bound to the protein and to rank it by scoring (Murlidaran and Brannigan, 2018). This approach can provide locations where multiple occupancies would be considered and suggest different putative candidates of binding locations. Recently, it has been suggested that diazepam binds at the extracellular “canonical” site and also at sites in the TMD (Iorio et al., 2020). Likewise, many ligands of volatile anesthetics and BZD drugs were predicted to bind at the extracellular site and also at sites in the TMD in this docking study (Bertaccini et al., 2013).

The binding sites occupied by each anesthetic ligand under general clinical concentrations were identified and ranked by affinity scores, computational analysis was used as described previously (Murlidaran and Brannigan, 2018). Briefly, the processing includes two phases: screening or finding possible candidate binding sites occupied by anesthetics ligands at clinical concentrations and measuring and quantifying binding affinities (Murlidaran and Brannigan, 2018). All atoms contained in proteins, lipids, salt, anesthetics ligands, and water were considered to be based on the thorough physics-based Molecular Dynamics Simulation and Alchemical Free Energy Perturbation (Murlidaran and Brannigan, 2018).

In this study, docking simulations of four volatile anesthetics ligands (isoflurane, sevoflurane, halothane, enflurane), propofol, and BZD/BZD antagonists (flumazenil, diazepam, alprazolam, midazolam, lormetazepam, nordazepam, oxazepam, temazepam, and lorazepam) were performed against the combined domain of TMD with ECD, TMD, ECD domains, and the TMD pore part, as shown in Fig. 2.

To identify binding sites and understand molecular mechanisms of anesthetics’ action on GABAARs, docking analysis has been used by employing structures of bound proteins and ligands. Since 2007, updated protein engineering techniques and crystallography analyses have provided a greatly increased number of solved protein structures (Venkatakrishnan et al., 2013). Although diverse pLGICs, including nAChR, GABAAR, glycine receptor, and NMDA receptor, are known to bind to a variety of anesthetics, docking studies to identify binding sites were not performed in many cases because of limited protein structures (Kim et al., 2018). Crystal structures of several LGICs have been recently solved. Crystal structures of a human GABAAR and a glycine receptor were obtained in 2014 and 2015, respectively (Miller and Aricescu, 2014; Huang et al., 2015; Kim et al., 2018). Docking analyses not previously performed because of unavailable structures are now possible, especially for those on the GABAAR with a variety of anesthetics. Crystalized structures of different kinds of GABAARs in complex with various ligands, including pharma-cores, drugs, peptides, antibodies, and G-proteins, have been reported (Sieghart, 1995; Bertaccini et al., 2013; Fourati et al., 2018; Murlidaran and Brannigan, 2018; Emsley et al., 2010). However, recent studies have revealed that when an anesthetic interacts with receptors to a multi-binding site, the influence on the GABAARs varies according to distinct subtypes. Therefore, more information needs to be obtained on the structure of GABAARs bound to anesthetic ligands.

Our docking studies revealed that isoflurane, halothane, and enflurane docked in a similar way to that observed for GABAARs by diazepam, except for an absence of interaction in the same residues in the diazepam binding site on GABAARs. Furthermore, these three anesthetics were majorly bound to an ECD of GABAARs, in contrast to sevoflurane, distributed in the TMD at the nine most highly ranked affinities. This differentiation in the docking mechanism might be related to sevoflurane’s different clinical effects, such as recovery from anesthesia, preconditioning, liver, and renal toxicity. In clinical practice, sevoflurane is more frequently used in pediatric patients for induction of anesthesia than other anesthetics due to its mild airway irritation (TerRiet et al., 2000). Moreover, recovery from anesthesia with sevoflurane is more rapid than with isoflurane (Arain et al., 2005). Although solubility influences the duration of anesthesia, the approach to different binding sites with membrane receptors also seems to affect anesthesia duration because these inhalation anesthetics have poor aqueous solubility. Within the grid box of the TMD pore, sevoflurane and propofol were located in the pore face of ß2. This result agrees with a previous report showing that all sevoflurane molecules are observed to be partitioned into membrane or protein sites, especially to the intrasubunit cavity of both ß subunits (but not α or γ subunits) in the flooding of ligands rarely soluble in the aqueous phase (sevoflurane and propofol) (Murlidaran et al., 2019). In this flooding simulation, amounts of sevoflurane partitioning into the water fraction, the membrane, and protein cavities were measured, and a high portion of molecules was introduced into the cytosolic phase and spontaneously partitioned into the lipid bilayer and/or bound to receptor protein cavities (Murlidaran et al., 2019). In this flooding model, sevoflurane was reported to bind spontaneously to pore sites and inter-subunit/intra-subunit sites, in agreement with our docking results. The report suggests that sevoflurane can penetrate the pore through the ECD vestibule over 1.3 µs and transfer from the pore to the α+-β− inter-subunit site (Murlidaran et al., 2019). In our docking results, interacting residues for sevoflurane on the GABAAR existed between β2 and γ2 at the best pose (1/9) or between α1 and β2 (2/9) at the second-best pose. Contacted residues for propofol on GABAARs were between α1β2 (1/9) or between β2 and γ2 (3/9). Although inhalation anesthetics and propofol had low solubility, our results showed that they were distributed in the pores.

The competition between flumazenil and isoflurane for the binding site might be related to the clinical role of flumazenil. Flumazenil is a competitive inhibitor of isoflurane or sevoflurane. The BZD antagonist flumazenil can restore the effects of BZD drugs bound to GABARs. It could also restore the clinical effect of isoflurane (Griffiths et al., 1984a). In clinical anesthesia, extubation and early anesthetic recovery are important to decrease morbidity and incidence of complications (Tvete et al., 2016).

BZD pharmacophore models based on SAR data have predicted a lipophilic pharmacophoric feature to be an essential part of BZDs (Rikke et al., 2013).

Considering that diazepam had a similar molecular weight, oxazepam showing the lowest docking score, might have the best affinity to the receptor at the best docking mode. At the best docking mode, flumazenil and oxazepam are bound to ECD, while diazepam is bound to TMD. These differences are related to lipophilicity in that flumazenil and oxazepam are less lipophilic than diazepam. In clinical reports, diazepam is considered to be less additive to dependency than oxazepam (Griffiths et al., 1984b). Further, diazepam is more rapidly absorbed, followed by a fast distribution phase (distribution half-life of about 1 h) with a half-life of 20–200 h, whereas oxazepam is absorbed more slowly, with a half-life of 4–15 h (Tvete et al., 2016). Classified in terms of BZD elimination half-life, diazepam is a long-acting BZD, whereas alprazolam, oxazepam, lorazepam, and temazepam are short to intermediate-acting BZDs (Fox et al., 2011). However, the clinical difference between diazepam and oxazepam is likely due to these pharmacokinetic differences and also the binding site and mode to the receptor. These differences suggest that relative potency is another way to characterize BZDs. Initially used BZDs were defined as having low to medium potency. However, alprazolam, lorazepam, and clonazepam were categorized as high-potency BZDs (Griffin et al., 2013). Of three BZD receptor agonists, namely diazepam, midazolam, and lorazepam, commonly used in anesthesia in clinical treatment, midazolam is the most lipid-soluble (Griffin et al., 2013). This agrees with our result that midazolam has the highest logP value among BZD agonists. It is distributed mainly in TMD. However, overall, our results are in line with a previous report proposing at least two different “common” binding modes rather than a single one for BZDs in general (Elgarf et al., 2018). Not all BZDs bind with the same type of BZD receptor or with equivalent binding affinity to the same receptor (Griffin et al., 2013).

Each anesthetic interacts with receptors to a multi-binding site, and its influence on the GABAARs varies according to the distinct subtypes. This necessitates a better understanding of the structure of GABAARs bound to anesthetic ligands. Our study results for the computational docking of volatile anesthetics such as isoflurane, sevoflurane, halothane, and enflurane into GABAARs suggested that sevoflurane binds with a mode that is distinct from that of isoflurane, halothane, and enflurane. Comparing the inhalation anesthetics, sevoflurane was mostly distributed in TMD, whereas isoflurane, enflurane, and halothane were bound to ECD.

In the best docking pose acquired for isoflurane, halothane, and enflurane, the ligand molecule binds to γ2Pro175 through the N terminal amine group and binds to γ2Glu178 through the side carboxyl group, conforming to the previously known experimental information. Because inhalation anesthetics and propofol are low solubility molecules, these anesthetics are unlikely to be distributed in receptor pores or intersubunit and intrasubunit spaces. However, converging evidence and our docking results demonstrated the binding of anesthetics to the pores or intersubunit and intrasubunit spaces. In the best docking pose obtained for flumazenil from our docking result, the ligand was observed in a well-known BZD binding site at both the ECD α1+/γ2− interface and the TMD α1/ß2 interface. From our docking result, diazepam was mostly distributed in TMD. The docking results of these ligands showed a broad diversity of high-score poses. Overall, our results reveal at least two different “common” binding modes, rather than a single one, for BZDs in general.