GABAA Receptor Modulators with a Pyrazolo[1,5-a]quinazoline Core: Synthesis, Molecular Modelling Studies and Electrophysiological Assays

As a continuation of our study in the GABAA receptor modulators field, we report the design and synthesis of new 8-chloropyrazolo[1,5-a]quinazoline derivatives. Molecular docking studies and the evaluation of the ‘Proximity Frequencies’ (exploiting our reported model) were performed on all the final compounds (3, 4, 6a–c, 7a,b, 8, 9, 12a–c, 13a,b, 14–19) to predict their profile on the α1β2γ2-GABAAR subtype. Furthermore, to verify whether the information coming from this virtual model was valid and, at the same time, to complete the study on this series, we evaluated the effects of compounds (1–100 µM) on the modulation of GABAA receptor function through electrophysiological techniques on recombinant α1β2γ2L-GABAA receptors expressed in Xenopus laevis oocytes. The matching between the virtual prediction and the electrophysiological tests makes our model a useful tool for the study of GABAA receptor modulators.


Introduction
The neurotransmitter α-aminobutyric acid, GABA, interacts with two different types of receptors, GABA A and GABA B receptors (GABA A Rs, GABA B R), which belong to the heteropentameric Ligand Gated Ion Channel (LGCI) superfamily with Cys-loop topology and are responsible for fast neuronal inhibition. The receptor is formed by five protein subunits arranged to form an anion channel permeable to chloride ions, which causes the postsynaptic hyperpolarization and the following inhibition of signal transmission after the binding of the neurotransmitter.
In recent years, cryogenic electron microscopy (cryo-EM) has made it possible to better elucidate the fine structure of synaptic or extrasynaptic GABA A Rs, as well as the mechanism and the possible binding modes of benzodiazepine ligands. These results also highlight the amino acid residues involved in the agonists (diazepam, alprazolam) or antagonists (flumazenil) binding and help to rationalize the design and synthesis of new benzodiazepine site ligands [3,5,6].
As a continuation of our research on 'benzodiazepine receptor ligands' with pyrazoloquinazoline (PQ) scaffold [7,8], we report here the synthesis of new 8-chloropyrazolo [1,5-a] quinazoline derivatives. Therefore, using the "Proximity Frequencies" [9], a molecular dynamic model (MD model) and statistical analysis developed in our laboratory, we tried to predict the pharmacological GABA A modulator profile (agonist or antagonist) for all new compounds. Finally, to test/verify the reliability of the MD results, all new compounds have been tested in electrophysiological studies on recombinant GABA A R (α1β2γ2-GABA A R), expressed in Xenopus Laevis oocytes by evaluating the variation of produced chlorine current.

Chemistry
The chemical steps to obtain the final compounds, the 3-ester (6a-c, 7a,b, 12a-c, 13a,b), the hydroxymethyl (14,15,17,18) and the 3-methoxymethyl derivatives (16 and 19), are depicted in Schemes 1-3 and the NMR spectra of some representative compounds are reported in Supplementary Materials. In recent years, cryogenic electron microscopy (cryo-EM) has made it possi better elucidate the fine structure of synaptic or extrasynaptic GABAARs, as well mechanism and the possible binding modes of benzodiazepine ligands. These resul highlight the amino acid residues involved in the agonists (diazepam, alprazola antagonists (flumazenil) binding and help to rationalize the design and synthesis o benzodiazepine site ligands [3,5,6].
As a continuation of our research on 'benzodiazepine receptor ligands' with zoloquinazoline (PQ) scaffold [7,8], we report here the synthesis of 8-chloropyrazolo [1,5-a]quinazoline derivatives. Therefore, using the "Proximity quencies" [9], a molecular dynamic model (MD model) and statistical analysis deve in our laboratory, we tried to predict the pharmacological GABAA modulator p (agonist or antagonist) for all new compounds. Finally, to test/verify the reliability MD results, all new compounds have been tested in electrophysiological studies combinant GABAAR (α1β2γ2-GABAAR), expressed in Xenopus Laevis oocytes by ating the variation of produced chlorine current.
The value of Proximity Frequencies (PFs), used in a linear discriminant function (LDA), was able to correctly collocate 70.6% of agonists and 72.7% of antagonists by combining a double PF (αVal203-γThr142) with a triple PF (αHis102-αTyr160-γTyr58). The predictive capacity was evaluated on an appropriate training set of molecules with a cross-validation 'leave one out' (LOO) procedure.
The agonist compounds, during a molecular dynamic simulation (60 ns), were simultaneously close to the αVal203 and γThr142 amino acids, with a frequency of 37% compared to the frequency of 16% found by the antagonist compounds, while the antagonist compounds were simultaneously close to the αHis102, αTyr160 and γTyr58 amino acids, with a frequency of 35% against a frequency of 13% for agonist compounds. Moreover, it highlighted that agonists' orientation in the binding site is significantly different from that assumed by the antagonists, as reported in the literature [11].
All the 3D structures of the molecules, as a training set and new final compounds, were designed (DS ViewerPro 6.0 Accelrys Software Inc., San Diego, CA, USA) and placed in the binding site of the BDZs with the AUTODOCK 4. it was acceptable to use one conformation, while two conformations for all other compounds (4, 6a-c, 7a-b, 8, 12a-c, 13a, 14, and 18) were required. The molecular dynamic simulations were performed on an isolated portion of the protein between the α and γ chains comprising all amino acids within a radius of 2 nm from the center of the benzodiazepine binding site. Applying the PF model [9] to the new synthesized compounds, we obtained the results reported in Table 1. From prediction based on the discriminant function calculated in the model [9], compounds 4, 6c, 7b and 12a gave uncertain results, since the prediction for the different conformations disagreed, while compounds 6b, 8, 13b and 14-16 showed borderline results, with about 50% of collocation for the two classes. Derivatives 6a, 7a, 12b, c, 13a, 17 and 18 are collocated in the agonist class, and among them, 6a, 13a and 18 reach a percentage of prediction of 93.1%, 72.4% and 64%, respectively. On the other hand, compounds 3, 9 and 19 are collocated in the antagonist class, with a percentage prediction range of 62-73%. From molecular dynamic studies, the poses of predicted antagonists (3, 9 and 19) and agonists (6a, 13a and 18) in the binding site emerged (Figure 1), but their overlapping seems to suggest that the two groups of compounds have the same orientation, in contrast with what we evidenced in the training set used to build the model. Therefore, the different predicted profiles could be justified by occupying other areas in the binding site (black circles in Figure 1). From prediction based on the discriminant function calculated in the model [9], compounds 4, 6c, 7b and 12a gave uncertain results, since the prediction for the different conformations disagreed, while compounds 6b, 8, 13b and 14-16 showed borderline results, with about 50% of collocation for the two classes. Derivatives 6a, 7a, 12b, c, 13a, 17 and 18 are collocated in the agonist class, and among them, 6a, 13a and 18 reach a percentage of prediction of 93.1%, 72.4% and 64%, respectively. On the other hand, compounds 3, 9 and 19 are collocated in the antagonist class, with a percentage prediction range of 62-73%. From molecular dynamic studies, the poses of predicted antagonists (3, 9 and 19) and agonists (6a, 13a and 18) in the binding site emerged (Figure 1), but their overlapping seems to suggest that the two groups of compounds have the same orientation, in contrast with what we evidenced in the training set used to build the model. Therefore, the different predicted profiles could be justified by occupying other areas in the binding site (black circles in Figure 1). For example, the 3-hydroxymethyl 18 and the 3-methoxymethyl derivative 19, predicted agonist and antagonist, respectively, interact with a different frequency and strength in the considered amino acid area during the molecular dynamic simulation. Figure 2 reports the overlapping of the two compounds 18 and 19 and the different orientations in the selected region. For example, the 3-hydroxymethyl 18 and the 3-methoxymethyl derivative 19, predicted agonist and antagonist, respectively, interact with a different frequency and strength in the considered amino acid area during the molecular dynamic simulation. Figure 2 reports the overlapping of the two compounds 18 and 19 and the different orientations in the selected region.
This different accommodation in the binding site of 18 (agonist) with respect to 19 (antagonist) could be due to hydrogen bond interactions with the amino acids in the region of αThr142 and αThr207.
Considering only a hydrogen bond distance (D-H . . . A) < 3 Å, the predicted agonist 18 forms a hydrogen bond interaction with αThr207 at an average distance of 1.9 Å (sd 0.2), with a frequency of 52%. On the contrary, the predicted antagonist 19 interacts more weakly with the same residue, with a longer average distance of 2.4 Å (sd 0.2) and a frequency < 8% (see Figure 3). Regarding the Van der Waals interactions, considering only those with a distance < 3 Å, the interaction with γTyr58 is exclusive for compound 19 (predicted antagonist), while compound 18 interacts with a more significant frequency with γTyr142, as can be seen from the histogram in Figure 4. This different accommodation in the binding site of 18 (agonist) with respect to 19 (antagonist) could be due to hydrogen bond interactions with the amino acids in the region of αThr142 and αThr207.
Considering only a hydrogen bond distance (D-H … A) < 3 Å, the predicted agonist 18 forms a hydrogen bond interaction with αThr207 at an average distance of 1.9 Å (sd 0.2), with a frequency of 52%. On the contrary, the predicted antagonist 19 interacts more weakly with the same residue, with a longer average distance of 2.4 Å (sd 0.2) and a frequency < 8% (see Figure 3). Regarding the Van der Waals interactions, considering only those with a distance < 3 Å, the interaction with γTyr58 is exclusive for compound 19 (predicted antagonist), while compound 18 interacts with a more significant frequency with γTyr142, as can be seen from the histogram in Figure 4.   This different accommodation in the binding site of 18 (agonist) with respect to 19 (antagonist) could be due to hydrogen bond interactions with the amino acids in the region of αThr142 and αThr207.
Considering only a hydrogen bond distance (D-H … A) < 3 Å, the predicted agonist 18 forms a hydrogen bond interaction with αThr207 at an average distance of 1.9 Å (sd 0.2), with a frequency of 52%. On the contrary, the predicted antagonist 19 interacts more weakly with the same residue, with a longer average distance of 2.4 Å (sd 0.2) and a frequency < 8% (see Figure 3). Regarding the Van der Waals interactions, considering only those with a distance < 3 Å, the interaction with γTyr58 is exclusive for compound 19 (predicted antagonist), while compound 18 interacts with a more significant frequency with γTyr142, as can be seen from the histogram in Figure 4.  The molecular dynamic simulation of the protein-ligand complex was realized on a portion of protein identified by amino acids located at a distance < 2.0 nM from the center of the 'benzodiazepine binding site'. During the simulation, the selected fragment did not show excessive distortion, as evidenced by Ramachandran's plot [12] (Figure 5), which is relative to the complex conformation at the end of the 60 ns of simulation: only 9% of amino acid residues are located in a disallowed region (white). The molecular dynamic simulation of the protein-ligand complex was realized on a portion of protein identified by amino acids located at a distance < 2.0 nM from the center of the 'benzodiazepine binding site'. During the simulation, the selected fragment did not show excessive distortion, as evidenced by Ramachandran's plot [12] (Figure 5), which is relative to the complex conformation at the end of the 60 ns of simulation: only 9% of amino acid residues are located in a disallowed region (white).     The molecular dynamic simulation of the protein-ligand complex was realized on a portion of protein identified by amino acids located at a distance < 2.0 nM from the center of the 'benzodiazepine binding site'. During the simulation, the selected fragment did not show excessive distortion, as evidenced by Ramachandran's plot [12] (Figure 5), which is relative to the complex conformation at the end of the 60 ns of simulation: only 9% of amino acid residues are located in a disallowed region (white).
As evident in Figure 6a, all 3-ester derivatives were not able to modulate the GABA A function, and we can hypothesize that they act as null modulators. To confirm that these compounds act at the benzodiazepine binding site, the representative compound 3 was evaluated for its ability to antagonize the full agonist lorazepam (1 µM). Figure 7 clearly demonstrates that this compound can abolish the chlorine current evoked by the agonist used as a standard, thus confirming its antagonist profile. Figure 6a, all 3-ester derivatives were not able to modulate the GABAA function, and we can hypothesize that they act as null modulators. To confirm that these compounds act at the benzodiazepine binding site, the representative compound 3 was evaluated for its ability to antagonize the full agonist lorazepam (1 µM). Figure 7 clearly demonstrates that this compound can abolish the chlorine current evoked by the agonist used as a standard, thus confirming its antagonist profile. Very intriguing results emerged from derivatives bearing a hydroxymethyl or methoxymethyl group at position 3 of the scaffold (Figure 6b). In particular, the 8-chloro-3-hydroxymethyl-4-methylpyrazolo[1,5-a]quinazoline-5-one 18 behaves as an agonist, significantly enhancing the chlorine current at 100 µM (Emax + 100%). On the other hand, the 8-chloro-3-(methoxymethyl)-4-methylpyrazolo[1,5-a]quinazoline-5-one 19, which differs from 18 for the presence of the methoxymethyl chain at position 3, acts as a null modulator, thus indicating that the -OH group is important for agonist activity. These two compounds were tested in the presence of the antagonist flumazenil (compound 18) and of the agonist lorazepam (compound 19) (Figure 8a,b) to evaluate whether they act through the benzodiazepine binding site. Figure 8a reports flumazenil's ability to antagonize 18 (100 µM), and Figure 8b shows the capacity of 19 to revert the agonist effect of lorazepam, in both cases confirming that the two compounds bind at the benzodiazepine binding site. Very intriguing results emerged from derivatives bearing a hydroxymethyl or methoxymethyl group at position 3 of the scaffold (Figure 6b). In particular, the 8-chloro-3hydroxymethyl-4-methylpyrazolo[1,5-a]quinazoline-5-one 18 behaves as an agonist, significantly enhancing the chlorine current at 100 µM (E max + 100%). On the other hand, the 8-chloro-3-(methoxymethyl)-4-methylpyrazolo[1,5-a]quinazoline-5-one 19, which differs from 18 for the presence of the methoxymethyl chain at position 3, acts as a null modulator, thus indicating that the -OH group is important for agonist activity. These two compounds were tested in the presence of the antagonist flumazenil (compound 18) and of the agonist lorazepam (compound 19) (Figure 8a,b) to evaluate whether they act through the benzodiazepine binding site. Figure 8a reports flumazenil's ability to antagonize 18 (100 µM), and Figure 8b shows the capacity of 19 to revert the agonist effect of lorazepam, in both cases confirming that the two compounds bind at the benzodiazepine binding site.

Materials and Methods
Reagents and starting materials were obtained from commercial sources. Extracts were dried over Na2SO4, and the solvents were removed under reduced pressure. All reactions were monitored by thin layer chromatography (TLC) using commercial plates pre-coated with Merck silica gel 60 F-254. Visualization was performed by UV fluores-

Materials and Methods
Reagents and starting materials were obtained from commercial sources. Extracts were dried over Na 2 SO 4 , and the solvents were removed under reduced pressure. All reactions were monitored by thin layer chromatography (TLC) using commercial plates pre-coated with Merck silica gel 60 F-254. Visualization was performed by UV fluorescence (λ max = 254 nm) or by staining with iodine or potassium permanganate. Chromatographic separations were performed on a silica gel column by gravity chromatography (Kieselgel 40, 0.063-0.200 mm; Merck, Kenilworth, NJ, USA), flash chromatography (Kieselgel 40, 0.040-0.063 mm; Merck) and silica gel preparative TLC (Kieselgel 60 F 254 , 20 × 20 cm, 2 mm). Yields refer to chromatographically and spectroscopically pure compounds, unless otherwise stated. Compounds were named following IUPAC rules, as applied by Beilstein-Institut AutoNom 2000 (4.01.305) or CA Index Name. All melting points were determined on a hot-stage Büchi microscope and are uncorrected. The identity and purity of intermediates and final compounds were ascertained through 1 H-NMR, 13 C-NMR and TLC chromatography. Monodimensional spectra 1 H-NMR and 13 C-NMR were registered by a 400 MHz field through Avance 400 apparatus (Bruker Biospin Version 002 with SGU). Chemical shifts (d) are in parts per million (ppm) approximated by the nearest 0.01 ppm, using the solvent as internal standard. Coupling constants (J) are in Hz; they were calculated by Top Spin 3.1 and approximated by 0.1 Hz. Data are reported as follows: chemical shift, multiplicity (exch, exchange; br, broad; s, singlet; d, doublet; t, triplet; q, quartet; m, multiplet; or a combination of those, e.g., dd), integral, assignments and coupling constant. Mass spectra (m/z) were recorded on an ESI-MS triple quadrupole (Varian 1200 L) system, in positive ion mode, by infusing a 10 mg/L solution of each analyte dissolved in a mixture of mQ H 2 O:acetonitrile 1:1 v/v. All new compounds possess a purity ≥95%; microanalyses indicated by the symbols of the elements were performed with a Perkine-Elmer 260 elemental analyzer for C, H and N, and they were within ±0.4% of the theoretical values.

8-Chloropyrazolo[1,5-a]quinazoline-3-carboxylic acid (5).
The ester 4 (0.1 g, 0.36 mmol) was dissolved in a few drops of diglyme and then 10 mL of 15% sodium hydroxide solution was added and refluxed for 2.5 h. The final suspension was diluted with water, acidified with HCl conc. until pH = 1, and the precipitate recovered by filtration was pure enough for the next step. General procedure for obtaining compounds 6a-c. The carboxylic acid 5 (0.5 mmol) was transformed into the corresponding 3-carbonyl chloride by reaction with an excess of SOCl 2 in anhydrous conditions. After the standard work-up, the residue was suspended in dichloromethane (6 mL), and the suitable alcohol (excess, 0.15 mL) was added; TLC monitored the reaction until the disappearance of the starting material. Then the final solution was evaporated to dryness, and the residue recuperated with isopropyl ether and recrystallized.  13  General procedure for obtaining compounds 7a-b. The starting esters 6a and 6b (0.240 mmol) were solubilized in glacial acetic acid under nitrogen flow. Then, 0.972 mmol of sodium cyanoborohydride (NaBH 3 CN) was added, and the reaction was refluxed for 1 h and monitored by TLC. The final solution was cooled at room temperature and water added until a precipitate was formed; the raw product was purified by recrystallization by a suitable solvent. 3.86 (s, 5H, OCH 3 , CH 2 NH). 13 (C, H, N).   (C, H, N). General procedure for the synthesis of 10 and 11. A suspension of 8 or 9 (0.35 mmoles) in 10% sodium hydroxide solution (5 mL) was refluxed under stirring for 1 h, monitoring the reaction by TLC. After the starting material disappeared, water/ice and HCl 6N were added, and the precipitate was filtered under suction.  165.50, 144.44,  143.48, 141.13, 139.83, 127.38, 127.00, 125.93, 125.86, 125.29, 125.19 H, N).

8-Chloro-4-methyl-5-oxo-4,5-dihydropyrazolo[1,5-a]quinazoline-3-carboxylic acid (10).
General procedure for obtaining compounds 14, 15. The starting material 1 or 8 [10] (0.5 mmol) was dissolved in 10 mL of t-butanol abs. and 0.15 g (8.91 mmoles) of sodium borohydride was quickly added. The reaction was maintained at reflux temperature (36 h-days), then was worked up by adding water/HCl 1N (10/3 mL) and extracted with ethyl acetate. After it dried under anhydrous sodium sulphate, the evaporation of the organic layer gave a residue recovered with water and filtered. General procedure for obtaining compounds 17, 18. To a solution of starting material 1 or 8 [10] (0.38 mmol) dissolved in 5 mL of THF abs., 0.08 g (3.03 mmoles) of lithium borohydride and 1.5 mL of methanol were rapidly added. The reaction was maintained at reflux temperature, then was worked up by adding water/HCl 1N (10/3 mL) and extracted with ethyl acetate. After it dried under anhydrous sodium sulphate, the evaporation of the organic layer gave a residue recovered with water and filtered. Another procedure to obtain compound 18 with better yield and starting from 21 (achieved in turn from 20) is here reported: the starting material 8-chloro-4-methyl-5-oxo-4,5-dihydropyrazolo[1,5-a]quinazoline-3-carboxyaldeide 21 (100 mg, 0.38 mmoles) was dissolved in a mixture of THF abs./methanol (3 mL/2 mL). Sodium borohydride 15.8 mg (0.42 mmoles) solution was added, portion wise, to this, and the reaction was stirred for 30 . The final solution evaporated to dryness, gave the row derivative 17, which was recovered with water, filtered under suction and recrystallized by ethanol 96%.

Molecular Docking and Molecular Dynamic Simulation
The structure of the binding site was obtained from the human α1β2γ2-GABA A receptor subtype in complex with GABA and flumazenil, conformation B (PDB ID 6D6T) [3], considering all the amino acids within a distance of about 2 nm from the structure of the flumazenil. The ligands were placed at the binding site through AUTODOCK 4.2. [13] The molecular dynamics simulations of ligand-binding site complexes were performed on a minimum number of conformations (maximum 2), to cover at least 90% of the poses found by AUTODOCK.
A 60 ns MD simulation was performed for all complexes using the GROMACS v5.1 program, and it was conducted in vacuum [14]. The DS ViewerPro 6.0 program [15] was used to build the initial conformations of ligands. The partial atomic charge of the ligand structures was calculated with CHIMERA [16] using the AM1-BCC method, and the topology was created with ACPYPE [17] based on the routine Antechamber [18].
The OPLS-AA/L all-atom force field [19] parameters were applied to all the structures. To remove bad contacts, energy minimization was performed using the steepest descent algorithm, until convergence was achieved, or for 50,000 maximum steps. The next equilibration of the system was conducted in two phases: (1) canonical NVT ensemble, a 100 ps position restraint of molecules at 300 K was carried out using a temperature-coupling thermostat (velocity-rescaling, in stochastic terms) to ensure the proper stabilization of the temperature [20]; (2) isothermal isobaric NPT ensemble, a 100 ps position restraint of molecules at 300 K and 1 bar was carried out without using barostat pressure coupling to stabilize the system. These were then followed by a 60 ns MD run at 300 K with position restraints for all protein atoms. The Lincs algorithm [21] was used for bond constraints to maintain rigid bond lengths.
The initial velocity was randomly assigned taken from Maxwell-Boltzman distribution at 300 K and computed with a time step of 2 fs, and the coordinates were recorded every 0.6 ns for MD simulation of 60 ns. The conformations collected during the simulated trajectory were 100.
The 'Proximity Frequencies' (PFs) [9], with which the 100 conformations of each binding site-ligand complex intercepts two or more amino acids during the dynamic simulation, were calculated. The 'Proximity Frequency' (PF) is the frequency with which the ligand was, during the molecular dynamic simulation, at a distance of less than 0.25 nm from an amino acid of the binding site and also, simultaneously, from 2, 3 and 4 amino acids of the binding site.

Electrophysiology
Electrophysiological measurements were performed in oocytes 2 to 4 days after DNA injection. Oocytes were placed in a rectangular chamber (volume~100 µL) and perfused at a rate of 1.7 mL/min with MBS at room temperature with the use of a roller pump (Cole-Parmer, Chicago, IL, USA) and 18-gauge polyethylene tubing (Clay Adams, Parsippany, NJ, USA). Oocytes were impaled at the animal pole with two glass electrodes (0.5 to 10 MΩ) filled with 3 M KCl and were clamped at -70 mV with the use of an oocyte clamp (model OC725C; Warner Instruments, Hamden, CT). GABA-induced Cl-currents were measured and analyzed with the pClamp 9.2 software (Molecular Devices, Union City, CA, USA). GABA (Sigma, St. Louis, MO, USA) was dissolved in MBS and applied to the oocytes for 30 s. Oocytes were perfused with test drugs for 30 s either in the absence of the GABA or in its presence at the EC5-10 (the concentration of agonist that induces a peak current equal to 5 to 10% of the maximal current elicited by the maximal concentration of the agonist). The EC5 concentration was determined for each oocyte and was approximately 3-5 µM [23].
Experimental sequence was as follows: maximal GABA (1 mM GABA, 20 s application, 20 min. washout); EC5-10 GABA (30 s application, 10/15 min washout), pre-application of the drug (30 s); followed by a co-application with EC10 GABA, EC10 GABA (30 s). Test drugs were first dissolved in DMSO at a concentration of 10 mM and then diluted in MBS to the final concentrations. In each experiment, control responses were determined before and 10/15 min after application of the drug.

Statistic
Statistical analysis was performed on normalized data using the one-way ANOVA test followed by Dunn's post hoc test using Graph Pad Prism 9 (Graph Pad Software, Inc., San Diego, CA, USA).

Conclusions
The synthesis of 8-chloropyrazolo[1,5-a]quinazolines and 8-chloro-4,5-dihydropyrazolo[1,5-a]quinazolines was designed and realized and all new compounds (3, 4, 6a-c, 7a-b,  8, 9, 12a-c, 13a,b, 14-19) were subjected to a molecular dynamic study performed on an isolated portion of the GABA A receptor protein, between the α and γ chains, where the benzodiazepine binding site is recognized. Then, applying the 'Proximity Frequencies' model (PF) [9], we obtained a prediction that collocates 6a, 13a and 18 in the agonist class, reaching a percentage of prediction of 93.1%. Compounds 3, 9 and 19 are collocated in the antagonist class, with a percentage prediction range of 62-73%. Interestingly, these two classes of compounds occupy different areas in the binding site that might justify the different predicted profile. The virtual prediction for 18 and 19 as agonist and antagonist, respectively, was confirmed through electrophysiological assays: compound 18 significantly enhances the chlorine current at 100 µM (E max + 100%), and it was antagonized by flumazenil (100 µM), thus acting as an agonist, while compound 19 was able to antagonize the chlorine current produced by the standard agonist lorazepam, thus confirming its antagonist profile. In conclusion, our PF model can be a useful predictive model of the efficacy/profile of new benzodiazepine site ligands.