Several CB₁ receptor studies support the importance of E2 for cannabinoid ligand binding and efficacy [20-23]. For many rhodopsin family GPCRs, a disulfide linkage between Cys residues from TM3 and E2 is important for maintaining receptor structure and function [24,24-27], possibly stabilizing coupling between the extracellular loop and TM core domains [28]. Because the CB₁ receptor lacks such a disulfide linkage, evidence implicates the formation of an alternative disulfide linkage between the two E2 Cys residues (i.e., C257 and C264) in E2 [20,21]. Our model suggests that E2 could be further stabilized by a set of H-bonding networks and other non-bonding interactions with E1, E3 and TM domain residues. A recent mutation study [29] suggests that E1 is indirectly involved in modulating ligand binding through interaction with the neighboring E2, and that the negatively charged D3.20(184) would interact with the K259 residue in E2. In agreement, our receptor model shows that E1 and E2 interact closely through H-bonding and/or salt bridge formation (R182/Q261 and D184/K259) (Fig. 3). The possibility that K183 in E1 interacts with E258 would be less convincing due to the following reasons: (1) the K183A mutation was less detrimental than the D3.20(184)A mutation in CP55940 binding [29]; and (2) rat and mouse CB₁, in which the E258 residue in E2 is replaced by a Lys, is incapable of such ionic interaction but still is comparable to the human CB₁ receptor for CP55940 binding [30,31]. One of the H-bonding networks (i.e., W255(E2)-D266(E2)-K370 (TM6)), and an aromatic stacking interaction of W255 and F268, appear to function as a lock such that E2 is kept in a closed conformation (see Fig. 3). It can be postulated that as the receptor environment changes due to the approaching ligand, this lock might be disrupted, leading to an open conformation of E2 with enhanced mobility, together with a transiently reduced E2 interaction with the neighboring segments. This would create a space for the incoming ligand to enter the binding site from the extracellular side.

The effect of solvent was not rigorously considered under the assumption that E2 would exist as close as possible to the entrance of the TM core pocket as suggested from the X-ray structure of rhodopsin [32]. Our models favor the E2 conformers with increased interactions with the receptor residues for packing. Calculations of MD simulations in explicit water suggested that inclusion of solvent would not have much effect on the E2 conformation, inasmuch as the RMSD was 1.19 Å when compared with the present receptor model. However, the receptor/membrane environment, wherein the membrane polar head groups as well as the extracellular receptor residues greatly exert their influence, would be important for precisely determining the E2 conformation [33].

Assuming that the 1-heptenyl moiety of the structurally-restricted anandamide analogs would be equivalent to the 1',1'-dimethylheptyl of cannabinol analogs [19,34] and that the ethanolamide moiety is needed to form H-bonding to K3.28(192), we determined the docking poses of compounds 1 through 6. At first glance, compounds 1 through 6 look similar in structure in that they commonly contain a benzene ring, the 1-heptenyl group, and hydroxylamide (see Table 1). However, they significantly differ conformationally due to the difference in position of the substituent on the aromatic ring (i.e., ortho-versus meta-) and the difference in the number of carbon atoms linking the hydroxylamide moiety to the benzene ring (i.e., the linker size). Both Glide and Affinity/SA approaches, as any other docking program, attempt to sample as many ligand conformations as possible within the predefined binding pocket. However, due to the enormous conformational flexibility of the tested compounds, the currently applied docking methods appear to be limited in properly sampling the conformation space. In Glide, the definition of the computational search space, based upon defining a box around the ligand in its initial placement into the receptor complex, is a fairly conservative approach that did not generate as wide a sampling of poses as the Affinity/SA process. Glide does not move the protein side chains, possibly preventing the escape of the ligand from a local energy minimum. As a result, the poses from Glide are more predisposed to have a docking conformation similar to that of CP55244, which was used as a model for the initial point of the simulation. In an effort to increase the search space as well as to optimize the conformations, the poses were run through a sequence of Glide and Prime steps. The Affinity/SA method permits the opportunity for the system to overcome local energy traps, and appears to be better in conformation sampling. Each approach uses different scoring functions: the Glide score [35] in Glide versus binding interaction energy in Affinity/SA. One difficulty in scoring docking conformations of the tested compounds derives from the generation of many local energy minima conformations without substantial energy differences. Thus, we applied these scoring functions to screen a set of conformations for each compound for which the LIE method was applied to select the best docking conformations. The different positions of the K3.28(192) residue and the different hydrophobic pockets determined by Glide and Affinity/SA exemplifies and emphasizes the notion of multiple micro-conformational states for the receptor-ligand complex [36]. With the TM backbone fixed, the side chains of the binding pocket residues as well as the extracellular loop residues were allowed to move freely during the docking simulations. Because the binding pocket within the TM core is relatively large, it is possible for the pocket residues to contribute to the modification of the pocket shape. This is especially true for hydrophobic residues, which can only be held relatively fixed when they have close contact for favorable interactions with neighboring hydrophobic residues. In contrast, polar residues that form very specific H-bonds and/or salt bridges to other polar residues or the amide backbone appear to be important for helical packing to stabilize the receptor framework structure.

In our previous analyses of the steric trigger mechanism of CB₁ receptor conformational changes induced by agonists [15], the efficacy of non-classical cannabinoid CP55244 was explored by conformational analysis, rotational barrier calculations, and molecular dynamics (MD) simulations. It was demonstrated that the torsion angles of the C3 side chain showed the most dramatic change when compared with the ground-state receptor-bound conformation, indicating that rotation around these torsion angles could be responsible for releasing the ligand strain energy. We also demonstrated that multiple stages were involved in the ligand conformational change. We proposed that the C3 side chain serves as the steric trigger, while the ACD-ring moiety of CP55244 serves as the plug, and that steric clash with helices within the binding pocket could induce microconformational adaptation within the protein. This mechanism would suggest that rotational flexibility in a ligand may be as important a determinant of agonist activity as the pharmacophoric elements that can be identified. Compound 1 exhibits the best activity of the compounds tested, showing greater than two-fold increase over basal in [³⁵S]GTPγS binding to G-proteins and maximal efficacy in the Gi-mediated inhibition of adenylyl cyclase. The anchoring of the amide moiety by multiple H-bonds as well as the stable positioning of the aromatic ring by interactions with hydrophobic residues constitutes the plug, consistent with a steric trigger mechanism of activation focusing on conformational mobility of the alkyl chain. This is similar to the C/D-ring region of CP55244 being anchored, and the dimethylheptyl chain being capable of free rotation around its single bonds, possibly inducing conformational changes in the receptor upon rotation [15]. The importance of the ligand interaction with the receptor through the hydrophobic tail is exemplified in the receptor binding properties of CP55940 [37] which exhibited a Ki = 0.137 nM. The analog having no A-ring OH to H-bond with K3.28(192) exhibited a Ki = 40.2 nM, whereas the analog having no C3 tail, exhibited a Ki = 441 nM, representing a ten-fold greater loss of affinity.

Among tested anandamide analogs, compound 2 shows quite similar binding affinity as anandamide (see Table 1). However, it is striking that compound 2 fails to stimulate GTPγS binding or to inhibit adenylyl cyclase activity (see Table 1). This would be expected of a compound that either could not form a "plug" or could not contribute a steric trigger interaction to promote a conformational change associated with receptor activation. As suggested from our MD simulation (see Fig. 9B), compound 2 might not be efficient at occupying both the hydrophobic pocket and maintaining an interaction with K3.28(192) simultaneously. This would reduce the probability of conformational changes associated with efficacy, rendering the molecule less likely to trigger a conformational change in the hydrophobic pocket. In agreement with this observation, the least potent of the series, compound 6, shows the least fit to the hydrophobic pocket (Figs. 5 &6). As shown in the compound 6 movie (Additional File 3), compound 6 does not conform well to the space occupied by CP55244 and compound 1. It appears that the very poor binding affinity of compound 6 is due to the lack of interaction with the key binding site amino acid residues that are involved in binding to CP55940 or compound 1. In agreement, our MD simulation suggested that compound 6 (see Fig. 9C) would fail to exhibit geometry required to satisfy the two important binding interactions with residues in the hydrophobic pocket as well as the interaction with K3.28(192). In addition, an increased conformational rigidity of the tail of this compound suggests its reduced efficacy as the steric trigger.