N-methyl-d-Aspartate (NMDA) receptors are ligand-gated cation channels involved in excitatory synaptic transmission, nociceptive pathways, learning, memory, and the modulation of muscular activity at the spinal level.1 They have been implicated as a putative site of action for inhaled general anesthetics because clinically relevant concentrations of diverse inhaled anesthetics inhibit such receptors in vitro.2–5 However, recent studies suggest that the contribution of NMDA receptor blockade to the immobilizing activities of volatile anesthetics in vivo may be restricted to aromatic compounds that are potent NMDA receptor inhibitors.5–7
Aromatic anesthetics exhibit a wide range of NMDA receptor inhibitory potencies and immobilizing activities, but the molecular features that determine the activities of this structurally homologous group are not well characterized. One approach to characterization applies molecular modeling techniques such as comparative molecular field analysis (CoMFA).8 In CoMFA, the molecular structures are placed in a rectangular grid of regularly spaced lattice points. The steric and electrostatic interaction energies between the compounds and a charged probe are calculated at each grid point and correlated with potency to formulate an activity model. By measuring which grid points contribute most to the activity model, we can identify the key regions of steric and electrostatic interactions important to the activities of the compounds. These regions can be expressed as three-dimensional pharmacophoric maps. Using this method, we previously characterized the molecular bases for the immobilizing activity of IV and inhaled general anesthetics.9–11
The present study sought to identify the molecular basis for the interaction of aromatic anesthetics with NMDA receptors in vitro and to compare the resultant pharmacophoric maps with those from an equivalent model for immobilizing activity in vivo.
Sixteen halogenated and nonhalogenated aromatic anesthetics were considered. We supplemented literature potency data for 14 compounds7,12,13 with unpublished determinations for 2 additional agents, chlorobenzene [compound 3] and ααα-trifluorotoluene [compound 12]. The methods used to determine NMDA receptor inhibitory potency were as previously described12 and were approved by the Massachusetts General Hospital Animal Care Committee. Inhibitory potencies were expressed as IC50 values, the concentration of inhalant that reduced the peak current by one-half in human NR1/NR2B NMDA receptors expressed in Xenopus laevis oocytes. NR1/NR2B NMDA receptors were used because they are more sensitive to aromatic anesthetics, such as toluene, than other NMDA receptor subunit combinations.14 The methods used to determine in vivo immobilizing activity were as in previous studies13 and were approved by the Committee on Animal Research at the University of California, San Francisco. Immobilizing activity was expressed as MAC, the minimum alveolar anesthetic concentration of inhalant required to prevent movement in 50% of rats given a noxious stimulus.
The compounds were divided into a training set (n = 12), which was used to formulate the CoMFA activity models and a test set (n = 4), which was used to independently assess the models’ predictive capability. Preliminary comparison of NMDA-receptor IC50 and MAC data (Fig. 1) showed that the compounds could be divided into four activity clusters. One test set agent was randomly picked from each cluster.
Anesthetic structures were constructed using MacSpartan Pro (Wavefunction, CA) and geometry optimized using ab initio quantum mechanics (Hartree-Fock, 6-31G** basis set). Partial charges fitted to the electrostatic potential of the molecule were assigned to each atom. Separate CoMFA activity models were formulated for the two aspects of anesthetic activity using Sybyl 7.3 (Tripos, MO).
The anesthetics were aligned for CoMFA in a reiterative process designed to minimize the differences in the steric and electrostatic interaction energy fields of the agents with those of one or more lead structures.15,16 The lead structures were high activity (potency) compounds and were used as alignment templates for the remaining anesthetics. For this study, the lead structures were the compounds in activity cluster 1 (those agents with both a high NMDA receptor inhibitory potency and high immobilizing activity, Fig. 1). The alignment process was based on the method of Kroemer and Hecht17 and consisted of two stages.
Alignment Stage 1
Each alignment cycle started with a lead structure being placed at the center of a rectangular grid (dimensions of X: −8.5 to 8.5 Å, Y: −8.0 to + 8.0 Å, Z: −6.5 to + 6.5 Å) consisting of lattice points at 1 Å intervals. The remaining anesthetics were prealigned to the lead structure by superimposing the carbon atoms of their common aromatic rings. All possible starting orientations of this prealignment step (and hence all possible side-chain overlaps) were tested by rotating the anesthetics to be fitted through 60°. A rigid-body minimization15,16 was used to orient and translate the molecules so as to minimize the differences in their steric and electrostatic interaction energy fields from those of the lead structure. The alignments that produced the best-fit to the template molecule, assessed by the sum of the correlation coefficients for the lead structure-anesthetic steric and electrostatic fields,17 were retained and used to formulate a CoMFA model. In some cases, the symmetry present in the aromatic anesthetics led to multiple best-fit solutions. These were also retained and tested.
CoMFA activity models were formulated by placing a carbon sp3 probe atom with unitary positive charge at each of the lattice points in the grid and measuring the interaction energies between the probe atom and the aligned anesthetic molecules.8 Steric energies were calculated as a Lennard-Jones potential, which describes the attraction between molecules due to van der Waals forces (dispersion, dipole-induced dipole, and dipole-dipole interactions) and the repulsion due to steric clashes. Electrostatic interaction energies were calculated using Coulomb potentials with a distance-dependent dielectric function. Cutoffs were applied to both the steric and electrostatic interaction energies at 30 kcal/mol. The interaction energies at each lattice point were block-scaled to unit variance and correlated with either −log10(NMDA receptor IC50) or −log10(MAC) using partial least squares regression.18 The lead structure-aligned anesthetic combinations that gave CoMFA models with the greatest predictive capability, assessed by leave-one-out cross-validation,19 were retained for the second alignment stage.
Alignment Stage 2
In this stage, a second lead structure from the cluster 1 compounds was used as an alignment template. The alignment process and evaluation were the same as above, with the exception that the second lead structure was not at the center of the lattice grid, but was in its best field-fit orientation to the first lead structure. If the additional fitting to the second lead structure improved the cross-validated r2 (r2CV) of the resultant CoMFA model by >0.05, it was retained and the addition of a third lead structure tested. This stage was repeated (up to a maximum of four lead compounds, the total number of compounds in cluster 1 after removal of a test compound) if the r2CV continued to meet the improvement criterion.
Test Set Prediction
The final NMDA receptor inhibition and immobilizing activity CoMFA models, based on the alignments of the anesthetics to one or more lead structures, were used to predict the activities of the four randomly excluded test set anesthetics. Test set anesthetics were subjected to the same field-fit minimization to the lead structures as the training set compounds. If the symmetry of the anesthetics meant that two or more orientations represented the best fit, then the potencies were predicted for each orientation and a mean value calculated. The effectiveness of the CoMFA models for predicting the activities of the test set compounds was expressed as a predictive r2 (r2Pred), which is analogous to the r2CV of the training set.17
The ability of the NMDA receptor inhibition model to predict the activities of nonaromatic cyclic compounds was investigated using an additional test set of four agents. Immobilizing activity data were not available for these additional compounds.
NMDA Inhibitory Potency Model
The final CoMFA model for NMDA receptor inhibition was based on the alignment of the compounds to a single lead structure, p-xylene [compound 10]. Addition of a second lead structure was not justified as it did not significantly improve the cross-validated r2 for the model. Figure 2 shows the correlation between observed and predicted NMDA-receptor inhibitory potencies for the aromatic compounds in Table 1. The CoMFA model explained 99.3% of the variance in the observed activities of the 12 training set compounds (F2,9 = 661.47, P < 0.0001, mean residual ± sd 0.049 ± 0.033). The model has good intrinsic predictability assessed by leave-one-out cross-validation of the training set (cross-validated r2CV = 0.944) and was an effective predictor of inhibitory potencies for the 4 randomly excluded test set agents (predictive r2Pred = 0.966, mean residual ± sd 0.118 ± 0.067).
The individual partial least squares regression weightings for each lattice point in the CoMFA grid can be used to identify regions where steric and electrostatic interactions are important determinants of NMDA-receptor inhibitory activity. These regions can be visualized using isocontours to link together the lattice points where the standard deviation multiplied by the partial least squares coefficient (sd × COEFF) exceeds a threshold value. Hence, the pharmacophoric maps indicate areas where the differences in either steric or electrostatic interactions are strongly associated with changes in anesthetic activity. Figures 3A and B show the electrostatic and steric maps for NMDA-receptor inhibition. The relative contributions of the electrostatic and steric interactions to the activity model were 85.7% and 14.3%, respectively. Because of the greater contribution made by the electrostatic interactions, different isocontour thresholds were used for the electrostatic and steric maps. The electrostatic map (Fig. 3A) indicates regions where negative (red, sd × COEFF < −0.002) and positive (blue, sd × COEFF > +0.002) potential are favored for high NMDA-receptor inhibitory potency. There are two areas (regions A and B) above and below the plane of the aromatic ring where negative potential is favored for high activity (B is not visible in the view shown), and three zones (C, D, and E) where positive potential is favored. The steric map (Fig. 3B) indicates regions where molecular bulk is favored for high inhibitory potency (green, sd × COEFF > +0.001) and areas where molecular bulk is disfavored (magenta, sd × COEFF < −0.001). The dominant features of the steric map are the single zone (F) where molecular bulk is favored for high activity, and three regions (G, H, and I) where molecular bulk is disfavored (i.e., if the molecule extends into one of these regions, it will have a reduced NMDA-receptor inhibitory potency).
Immobilizing Activity Model
Equivalent CoMFA models were derived for the in vivo immobilizing activities of the aromatic anesthetics. The best model obtained using a single lead structure was based on the alignments of the agents to p-xylene (the same lead structure used for the NMDA receptor inhibition model) and explained 94.7% of the variance in the observed activities of the training-set anesthetics (F2,9 = 80.89, P < 0.0001, r2CV = 0.779). Inclusion of o-xylene [compound 9] as an additional lead structure during the fitting process resulted in adjustments to the alignments for four of the training set anesthetics (chlorobenzene, p-difluorobenzene, ethylbenzene and pentafluorotoluene) and 2 of the test set agents (m-xylene and toluene). These alignment adjustments significantly improved the predictive capability of the model assessed by leave-one-out cross-validation of the training set (r2CV = 0.872). The final model (Table 2, Fig. 4) based on the alignments to 2 lead structures explained 98.0% of the variance in the observed activities of the training set anesthetics (F2,9 = 219.22, n = 12, P < 0.0001, mean residual ± sd 0.055 ± 0.035) and was a good predictor of activity for the test set agents (r2Pred = 0.926, mean residual ± sd = 0.091 ± 0.053).
Comparison of Pharmacophoric Maps
Several of the key regions identified in the NMDA-receptor inhibition model are also present in the pharmacophoric maps for immobilizing activity (Figs. 3C and D). Thus the electrostatic regions (A, B, C, D, and E) and the steric regions (F, G, and H) are common to both models. Two additional negative potential favored regions (J and K) and a steric favored region (L) are present in the immobilizing activity model but are not seen in the NMDA-receptor inhibition model. Furthermore, there are spatial conflicts with the overlap of the bulk favored region (L) of the immobilizing activity model and the bulk disfavored region (I) in the NMDA-receptor inhibition model, and the negative potential favored region (K) with the positive potential favored zone (D).
Despite a high level of similarity in the spatial distribution of the key regions, the relative contributions of the electrostatic and steric interactions differ between the two activity models. In the immobilizing activity model, the contributions of the total electrostatic and steric interactions are 61.4% and 38.6%, respectively, representing an increase in the importance of steric interactions compared with the NMDA-receptor inhibition model. The contributions made by each key region also reflect these differences (Fig. 5). The dominance of the electrostatic regions A, B, and D can be seen for the NMDA receptor inhibitory model, whereas the increased importance of steric interactions to immobilizing activity is reflected with the greater contribution of region F and the additional region L. The additional electrostatic region J also importantly contributes to the immobilizing activity model.
Nonaromatic Cyclic Compounds
The ability of the NMDA receptor inhibition model to predict the activities of four additional nonaromatic cyclic test agents is shown in Table 1. Good predictions were obtained for 1,4-cyclohexadiene [compound 19] and 1-methyl-1,4-cyclohexadiene [compound 20], with residuals comparable to those of the aromatic test set (absolute log residuals <0.1). However, weaker predictions were obtained for cyclohexene [compound 17] and 4-methyl-cyclohexene [compound 18], with residuals of 0.396 and 0.589, respectively. Further investigation showed that the interaction energy fields of the cyclohexenes differed from the other compounds, with >80% of the lattice points in the CoMFA grid having values outside the range shown by the aromatic training set agents. More accurate predictions can be obtained for cyclohexene (absolute log residual = 0.135) and 4-methyl-cyclohexene (absolute log residual = 0.113) by including these compounds in the training set and deriving a new CoMFA model (not shown).
The present article outlines the development of 2 CoMFA activity models for volatile aromatic anesthetics that describe the molecular bases of their NMDA-receptor inhibitory activities in vitro and their immobilizing activities in vivo. Both models are based on the spatial distribution of key regions where steric and electrostatic interactions are important determinants of the corresponding activities, and both models effectively predict the corresponding activities for aromatic training (n = 12) and test (n = 4) set agents. The use of a common lattice grid enables a direct comparison of the CoMFA models that show several important features for these two aspects of aromatic anesthetic activity.
First, optimum alignments of the anesthetics were obtained for the NMDA receptor inhibitory potency model using a single lead structure (p-xylene). In contrast, two lead structures (p-xylene and o-xylene) were required to derive a model with comparable performance for immobilizing activity. This suggests that no single cluster 1 compound has all the steric and electrostatic molecular features that characterize the immobilizing activity of aromatic anesthetics. Hence, more than one lead structure is required to align the compounds for maximal model performance. This may simply reflect the limited range of compounds considered and studies with additional aromatic anesthetics (particularly trimethylbenzene compounds) would test this possibility. Alternatively, it may indicate that immobility is mediated by more than one site of action, with each site having a distinct alignment requirement.
Second, the relative contributions of the electrostatic and steric interactions to the 2 CoMFA activity models differ. For NMDA receptor inhibition, electrostatic interactions make a much greater contribution (electrostatic: steric contribution ratio of 6.0:1). This agrees with the work of Raines et al.12 who demonstrated that NMDA-receptor inhibition potency correlates with the cation-pi energy of the molecule (represented by the negative potential favored regions A and B in our pharmacophoric maps). Although electrostatic interactions are also most important to the production of immobility, steric interactions gain in significance (electrostatic: steric ratio of 1.6:1).
Third, many key regions identified in the pharmacophoric maps are common to both activity models. These include the negative potential favored regions above and below the plane of the aromatic ring (A and B), the positive potential favored regions (C, D, and E), the molecular bulk favored region (F) and the bulk disfavored regions (G and H). However, the relative importance of each key region to the two aspects of anesthesia clearly differs (Fig. 5). There are also some regional differences between the maps, with additional negative potential (J) and molecular bulk (L) favored regions appearing in the immobilizing activity model alone, and two spatial conflicts have been identified (where a favored zone in one map is associated with a disfavored zone in the other).
Do aromatic agents act at NMDA receptors to produce immobility in response to noxious stimuli? The commonalities between the CoMFA maps are consistent with NR1/NR2B NMDA receptors contributing to overall immobilizing activity. However, the combination of a low correlation between NMDA-receptor IC50 and MAC (r2 = 0.399), the different lead structures required for the optimal alignment of the anesthetics, the regional differences identified and the different contributions made by the common key regions suggest that NMDA receptors alone do not mediate immobilizing activity. Recent studies by Kelly et al.20 indicate a reciprocal relationship between the percentage inhibition of NMDA receptors at one MAC and the percentage enhancement of γ-aminobutyric acid, type A receptors at the same concentration. Whether the γ-aminobutyric acid, type A receptor is the additional site of action for aromatic anesthetics remains to be determined. The focus on aromatic anesthetics precludes the possibility of drawing any conclusions about the role of NMDA receptors in mediating the immobilizing activities of other nonaromatic volatile anesthetics.
Although primarily designed for aromatic anesthetics, the NMDA receptor inhibition model can predict potencies for some nonaromatic cyclic compounds if they have similar electrostatic and steric profiles to the agents used to formulate the model. Weaker predictions are obtained if the molecular fields of the test compounds differ substantially from those of the training set. This is a known limitation of CoMFA, which can be overcome by the inclusion of more structurally diverse anesthetics in the training set, enabling activity predictions by interpolation rather than extrapolation.21 The nonaromatic cyclic compounds were not included in the training set for this study due to the lack of immobilizing activity data. The inclusion of additional compounds in one model, but not the other, would have prevented the direct comparison of the pharmacophoric maps for the two aspects of anesthetic activity.
The pharmacophoric maps indicate the regions where steric and electrostatic interactions are important for high activity. In the case of the NMDA receptor inhibition model, the maps represent the key regions that determine inhibitory potency. These include the steric and electrostatic features that determine the interaction of the aromatic agents with the binding site of the NR1/NR2B receptors. However, the pharmacophoric maps should not be interpreted as a direct corollary of the binding site, since steric and electrostatic interactions also influence binding enthalpy through other effects, such as changes in solvation and desolvation energies. The key regions that determine the specificity of aromatic binding to NR1/NR2B receptors could be established by comparing pharmacophoric maps derived from the interactions of the anesthetics with other receptor systems. Such data are only available for a subset of the compounds considered in this study.
Our study characterized the molecular bases of NMDA receptor inhibition and immobilizing activity of aromatic anesthetics in terms of the spatial distribution of key steric and electrostatic regions. Studies making use of the differences in the relative contributions of the key steric and electrostatic regions identified may inform the design of aromatic anesthetics that favor one aspect of anesthetic activity over the other.
1.Loftis JM, Janowsky A. The N
-methyl-d-aspartate receptor subunit NR2B: localization, functional properties, regulation and clinical implications. Pharmacol Ther 2003;97:55–85
2.Hollmann MW, Liu HT, Hoenemann CW, Liu WH, Durieux ME. Modulation of NMDA receptor function by ketamine and magnesium. Part II: interactions with volatile anesthetics. Anesth Analg 2001;92:1182–91
3.Yamakura T, Harris RA. Effects of gaseous anesthetics nitrous oxide and xenon on ligand-gated ion channels: comparison with isoflurane and ethanol. Anesthesiology 2000;93:1095–101
4.de Sousa SLM, Dickinson R, Lieb WR, Franks NP. Contrasting synaptic actions of the inhalational general anesthetics isoflurane and xenon. Anesthesiology 2000;92:1055–66
5.Solt K, Eger EI II, Raines DE. Differential modulation of human N
-methyl-d-aspartate receptors by structurally diverse general anesthetics. Anesth Analg 2006;102:1407–11
6.Dutton RC, Laster MJ, Xing YL, Sonner JM, Raines DE, Solt K, Eger EI II. Do N
-methyl-d-aspartate receptors mediate the capacity of inhaled anesthetics to suppress the temporal summation that contributes to minimum alveolar concentration? Anesth Analg 2006;102:1412–18
7.Eger EI II, Liao M, Laster MJ, Won A, Popovich J, Raines DE, Solt K, Dutton RC, Cobos FV, Sonner JM. Contrasting roles of the N
-methyl-d-aspartate receptor in the production of immobilization by conventional and aromatic anesthetics. Anesth Analg 2006;102:1397–406
8.Kubinyi H. QSAR and 3D QSAR in drug design. I. Methodology. Drug Discov Today 1997;2:457–67
9.Sewell JC, Sear JW. Derivation of preliminary three-dimensional pharmacophoric maps for chemically diverse intravenous general anesthetics. Br J Anaesth 2004;92:45–53
10.Sewell JC, Sear JW. Derivation of preliminary three-dimensional pharmacophores for nonhalogenated volatile anesthetics. Anesth Analg 2004;99:744–51
11.Sewell JC, Sear JW. Determinants of volatile general anesthetic potency: a preliminary three-dimensional pharmacophore for halogenated anesthetics. Anesth Analg 2006;102:764–71
12.Raines DE, Gioia F, Claycomb RJ, Stevens RJ. The N
-methyl-d-aspartate receptor inhibitory potencies of aromatic inhaled drugs of abuse: evidence for modulation by cation-π interactions. J Pharmacol Exp Ther 2004;311:14–21
13.Fang Z, Sonner J, Laster MJ, Ionescu P, Kandel L, Koblin DD, Eger EI II, Halsey MJ. Anesthetic and convulsant properties of aromatic compounds and cycloalkanes: implications for mechanisms of narcosis. Anesth Analg 1996;83:1097–104
14.Cruz SL, Mirshahi T, Thomas B, Balster RL, Woodward JJ. Effects of the abused solvent toluene on recombinant N
-methyl-d-aspartate and non-N
-methyl-d-aspartate receptors expressed in Xenopus
oocytes. J Pharmacol Exp Ther 1998;286:334–40
15.Cramer RD III, Patterson DE, Bunce JD. Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J Am Chem Soc 1988;110:5959–67
16.Clark M, Cramer RD III, Jones DM, Patterson DE, Simeroth PE. Comparative molecular field analysis (CoMFA). Part 2. Toward its use with 3D-Structural databases. Tetrahedron Comput Methodol 1990;3:47–59
17.Kroemer RT, Hecht P. A new procedure for improving the predictiveness of CoMFA models and its application to a set of dihydrofolate reductase inhibitors. J Comput Aided Mol Des 1995;9:396–406
18.Dunn WJ III, Wold S, Edlund U, Hellberg S, Gasteiger J. Multivariate structure-activity relationships between data from a battery of biological tests and an ensemble of structure descriptors: the PLS method. Quant Struct Act Relat 1984;3:131–7
19.Cramer RD III, Bunce JD, Patterson DE, Frank IE. Crossvalidation, boot strapping and partial least squares compared with multiple regression in conventional QSAR studies. Quant Struct Act Relat 1988;7:18–25
20.Kelly EW, Solt K, Raines DE. Volatile aromatic anesthetics variably impact human g-Aminobutyric acid type A receptor function. Anesth Analg 2007;105:1287–92
21.Kroemer RT, Hecht P, Guessregen S, Liedl KR. Improving the predictive quality of CoMFA models. Perspect Drug Discov Design 1998;12:41–56