Novel Molecule Exhibiting Selective Affinity for GABAA Receptor Subtypes

Aminoquinoline derivatives were evaluated against a panel of receptors/channels/transporters in radioligand binding experiments. One of these derivatives (DCUK-OEt) displayed micromolar affinity for brain γ-aminobutyric acid type A (GABAA) receptors. DCUK-OEt was shown to be a positive allosteric modulator (PAM) of GABA currents with α1β2γ2, α1β3γ2, α5β3γ2 and α1β3δ GABAA receptors, while having no significant PAM effect on αβ receptors or α1β1γ2, α1β2γ1, α4β3γ2 or α4β3δ receptors. DCUK-OEt modulation of α1β2γ2 GABAA receptors was not blocked by flumazenil. The subunit requirements for DCUK-OEt actions distinguished DCUK-OEt from other currently known modulators of GABA function (e.g., anesthetics, neurosteroids or ethanol). Simulated docking of DCUK-OEt at the GABAA receptor suggested that its binding site may be at the α + β- subunit interface. In slices of the central amygdala, DCUK-OEt acted primarily on extrasynaptic GABAA receptors containing the α1 subunit and generated increases in extrasynaptic “tonic” current with no significant effect on phasic responses to GABA. DCUK-OEt is a novel chemical structure acting as a PAM at particular GABAA receptors. Given that neurons in the central amygdala responding to DCUK-OEt were recently identified as relevant for alcohol dependence, DCUK-OEt should be further evaluated for the treatment of alcoholism.

). The β1 subunit residue 265 seems to play an important role in determining the effect of certain GABA A receptor modulators: when S265 in β1 is mutated to N (homologous residue in β2 and β3) on the GABA A receptor complex, the modulators' potentiation is increased, and vice versa, when N265 in β2 or β3 is mutated to S, the potentiation is reduced [16][17][18] . When we tested DCUK-OEt on α1β2(N265S)γ2 compared to α1β2γ2, the effect of DCUK-OEt as a PAM was significantly reduced (Table 3), but not to the extent seen with drugs such as etomidate (no GABA potentiating effect of etomidate at concentrations up to 1 mM was evident with the α1β2(N265S)γ2 receptor combination) 16 .
To further investigate potential binding sites for DCUK-OEt on the GABA A receptor, we performed computationally-based small molecule docking studies to compare the potential interactions of DCUK-OEt with those of DCUKA, flunitrazepam, and etomidate, with either the classical benzodiazepine binding site (located at the α + γinterface of the pentameric receptor) or an alternative binding site (at the α + βinterface) (Fig. 3a). The corresponding binding energies are shown in Table 4. These studies indicated that DCUK-OEt exhibited the highest predicted affinity for an alternative binding site, while, as expected, flunitrazepam exhibited the highest predicted affinity for the benzodiazepine site.
The modeling studies predicted both the carboxylate of DCUKA and the ethyl ester moiety of DCUK-OEt to be oriented towards the α subunit in the region of α:Tyr160 in the alternative site ( Fig. 3b and c). The ethyl ester was predicted to participate in additional hydrophobic interactions with the residues of this region, and there exists a potential π-σ interaction with α:Tyr160. These additional interactions of the ethyl ester also appeared to  50 and Ki values were obtained by non-linear regression analysis of radioligand binding isotherms. Ki values are reported as estimates from the non-linear regressions and their associated standard errors (n = 10 points in the binding isotherms). optimize the positioning of the head group and amide linker within the binding pocket to allow for additional potential H-bond and π-π interactions with β:Asp43 and Tyr62, respectively, leading to the higher affinity for DCUK-OEt compared to DCUKA for the GABA A receptor. The predicted binding and interactions of flunitrazepam in the benzodiazepine site ( Supplementary Fig. S4) were consistent with previous studies [19][20][21] , and flunitrazepam made a number of favorable contacts, including H-bond interactions with α:Tyr160 and γ:Thr142, and π-π interactions with α:Tyr160 and Tyr210. DCUKA shared a number of these predicted contacts, while the ethyl ester of DCUK-OEt appeared to impair the optimal positioning of the head group in the benzodiazepine binding pocket (Fig. 3d and e). Flunitrazepam bound somewhat more deeply into the pocket, compared to the other tested compounds, with the fluorbenzene ring predicted to be locked in place by a three-way π-stacking interaction with α:His102 and γ:Tyr58, Phe77. An additional π-σ interaction with α:Phe100 and π-π stacking with γ:Phe77 not only distinguish the predicted binding of flunitrazepam from DCUKA, but also represent the crucial aspects of interaction of flunitrazepam with the receptor   Table 3. Comparison of DCUK-OEt induced changes in EC 10 GABA responses between receptors differing in a single subunit. These comparisons were executed with the linear mixed model using linear contrasts. Correction for multiple pairwise comparisons was by a Bonferroni adjustment.
that lead to its pharmacological function. The presence of the phenyl (C ring) substitution at the 5 position of the benzodiazepine ring structure is necessary for the PAM actions of the benzodiazepine derivatives 22,23 . Therefore, even though DCUKA and DCUK-OEt may bind to the benzodiazepine binding site on the GABA A receptor (with lower affinity), the lack of the fluorbenzene ring on the DCUKA and DCUK-OEt structures would predict the lack of functional effect of DCUKA and DCUK-OEt via the benzodiazepine site. It is important to note that, due to the method by which binding energies are calculated, comparisons of relative binding affinity can only be reliably assessed between different molecules within the same binding site. The studies showing that potentiation of GABA responses by DCUK-OEt cannot be blocked by flumazenil do not preclude the possibility, suggested by the docking experiments, that DCUK-OEt could bind to the benzodiazepine site as an antagonist, while producing potentiation via binding to a different site (the alternative, extracellular site or a transmembrane one). We tested this hypothesis by co-applying DCUK-OEt (1 µM) and Dashed lines indicate predicted non-bond interactions (green = H-bonds, orange = electrostatic or π-cation/anion, magenta = π-π, purple = π-σ, pink = hydrophobic).
SCientifiC REPoRtS | 7: 6230 | DOI:10.1038/s41598-017-05966-x flunitrazepam (0.1 µM). The combined effect was larger than the sum of their individual effects ( Supplementary  Fig. S5), suggesting that the functional effects of the two drugs may be mediated by actions at two different sites.
The significant effects of DCUK-OEt on particular subunit combinations of the GABA A receptor led us to test the effects of this compound on neurons in the rat central amygdala (CeA). The CeA is primarily composed of GABAergic neurons and changes in CeA GABAergic neurotransmission have been implicated in the development and maintenance of alcohol dependence 24 . Focal application of DCUK-OEt (0.5 µM) significantly increased the holding current in medial CeA neurons ( Fig. 4a and b), while producing no significant effect on spontaneous inhibitory postsynaptic current (sIPSC) frequency, amplitude, rise or decay times (Fig. 4c).
To confirm that the changes in holding current that we observed were due to increases in tonic signaling at the GABA A receptor, the GABA A receptor antagonist gabazine (GBZ, 100 μM, Sigma Chemical Co., St. Louis, MO) was focally applied following DCUK-OEt application. GBZ produced a significant reduction in holding current when applied after DCUK-OEt, suggesting that the changes in holding current that were observed with DCUK-OEt, were due to DCUK-OEt-induced increases in tonic conductance via GABA A receptors on medial CeA neurons. In addition, we found that the increase in holding current with DCUK-OEt was positively correlated with the reduction in holding current seen with GBZ application (Pearson correlation coefficient = 0.838; p = 0.0094; n = 8; Fig. 4d).

Discussion
DCUK-OEt acts as a subunit-selective PAM at the GABA A receptor, and our ligand binding studies produced no evidence of interaction of DCUK-OEt (<10 μM) with 32 other receptors/transporters/channel proteins. DCUK-OEt exhibited its most robust effects on submaximal GABA-induced currents when applied to the α1β2γ2 GABA A receptor, the subunit combination most highly expressed in mammalian brain 2 . Similar PAM activity of DCUK-OEt was exhibited with GABA A receptors composed of α1β3δ subunits. On the other hand, DCUKA, which lacks the ester moiety at the 2 position of the carboxyquinoline, and is the major metabolite of DCUK-OEt, was 10 times less potent than DCUK-OEt in acting as a PAM at the α1β2γ2-containing GABA A receptors.
The most studied PAMs at the GABA A receptor are benzodiazepine derivatives and other compounds (e.g., zolpidem) which act at the interface of extracellular domains of the α and γ subunits 4 . Our data produced no evidence for DCUK-OEt action at this site. DCUK-OEt did not displace flunitrazepam in ligand displacement experiments, and the electrophysiological effects of DCUK-OEt (and DCUKA) on GABA A receptors expressed in oocytes were not modified by the selective benzodiazepine antagonist, flumazenil. Additionally, the substitution of a δ subunit for a γ subunit in the GABA A receptor complex greatly diminishes the effects of benzodiazepines 25 but the effects of DCUK-OEt were similar in receptors containing either γ2 or δ subunits (compare α1β3γ2 and α1β3δ in Tables 2 and 3). Finally, when DCUK-OEt and flunitrazepam were applied together, the PAM effect was supra-additive.
Assessment of the possibility that DCUK-OEt acted at the "neurosteroid" site on the GABA A receptor produced somewhat equivocal results. 17PA, which has been reported 15,26 , and in our hands also shown, to be a weak partial antagonist at the "neurosteroid" site on the GABA A receptor, produced a statistically significant but modest (35%) inhibition of the PAM actions of DCUK-OEt, while inhibiting the effects of allopregnanolone by 56%. This difference in potency of 17PA could be due to differences in the affinity/efficacy of DCUK-OEt compared to allopregnanolone at the "neurosteroid" site(s). However, neurosteroid agonists acting at a "neurosteroid"

Compound
Binding energy H-bonds/Electrostatic π-π π-Anion/Cation π-σ Hydrophobic π-Amide Table 4. Docking binding energies and interactions at GABA A receptor sites. Summary of the binding energies and non-bond interactions of the top scoring predicted binding orientations for each compound docked into the homology model of the benzodiazepine binding site at the α + γsubunit interface or the "Alternative" binding site at the α + βsubunit interface of the human GABA A receptor shown in Fig. 2 and in Supplementary  Fig. S4. Binding orientations were predicted using the Discovery Studio flexible docking protocol and energies were calculated using the distance-dependent dielectric model, as outlined in the methods.
site 26 are particularly effective as PAMs, and are also direct agonists at GABA A receptors composed of the α4βxδ subunits, while DCUK-OEt had no significant effect on this subunit combination. Furthermore, the modulatory action of neurosteroids at low concentrations does not differ among β subunits 27 . On the other hand, both β2 and β3 subunits in combination with α1 and γ2 subunits responded to the addition of DCUK-OEt with a significant increase in the current induced by submaximal GABA, but the substitution of the β1 subunit for either β2 or β3 resulted in a notable decrease of the PAM activity of DCUK-OEt ( Table 2). The negative effect of the β1 subunit is reminiscent of the selectivity for β subunits shown by modulators such as loreclezole 18 and etomidate 28, 29 , among others. Three amino acids in the transmembrane domains of the β subunit, distinguish the sequence of β1 from β2/β3 30 , and mutation of the asparagine at position 265 in the β2 sequence, located at the interface of α/β transmembrane domains, has been demonstrated to interfere with the potentiating action of etomidate and other anesthetics at GABA A receptors 16,17,30,31 . The introduction of a mutated β2 (N265S) into a complex containing α1 and γ2 subunits significantly reduced (Table 3) the PAM activity of DCUK-OEt. However, this mutation has been shown to eliminate etomidate's PAM action 28,32 . Mutation of β2N265 also decreases alcohol PAM activity on GABA A receptors 33,34 . However, ethanol potentiates GABA effects at receptors composed of dimeric αβ GABA A receptors, and does not discriminate between β1 versus β2 subunits 35 . Reports on the concentrations of ethanol necessary to potentiate the effects of GABA on α4β3δ GABA A receptors expressed in Xenopus oocytes have been contradictory [36][37][38] , but the ethanol effect on the α4β3δ subunit combination is always potentiation of the GABA actions, in contrast to the lack of any significant effect of DCUK-OEt. At the EC 10 concentration of GABA, DCUK-OEt exhibited PAM effects on α1β3δ GABA A receptors similar to effects seen with α1β2γ2. However, DCUK-OEt also enhanced the current produced by saturating concentrations of GABA with the α1β2/3δ subunit combination, but not with the α1β2/3γ2 combination (Fig. 2c). GABA has been shown to be a partial agonist at δ subunit-containing receptors 39 , and DCUK-OEt, and some other PAMs 40 , may allow for further activation of the GABA A receptor at concentrations seemingly maximal in the absence of PAMs. It also should be stressed that we detected no effect of DCUK-OEt at any concentration on any of the subunit combinations we tested in our paradigm, without the addition of GABA.
Overall, as noted above, there seems to be some overlap in the characteristics of DCUK-OEt with properties exhibited by allopregnanolone, CGS 9895, LAU-177 41,42 , loreclezole, etomidate and ethanol, but other characteristics regarding subunit selectivity of DCUK-OEt mitigate against assuming that DCUK-OEt binding/activity occurs specifically through the currently described site(s) for binding of these agents. Additionally, DCUK-OEt characteristics do not conform to what would be expected if DCUK-OEt were utilizing the canonical barbiturate, or intravenous or inhalation anesthetic sites to affect GABA action at the GABA A receptor 31,[43][44][45] .
Our models to ascertain the docking of DCUK-OEt to interfaces between the various subunits of the GABA A receptor (composed of α1β2γ2 subunits), indicated that a binding site for DCUK-OEt may exist between the α + βinterface in the pentameric receptor. The free energy (−ΔG) of binding at this site was highest for DCUK-OEt and lowest for etomidate and flunitrazepam. When examining the docking at the benzodiazepine site located between the α + γinterface, the order was reversed, with flunitrazepam showing the highest binding energy and DCUK-OEt and etomidate showing the lowest −ΔG. If the function of DCUK-OEt was dependent on binding at a single site at the α + βinterface, one would expect that GABA A receptors composed of only α and β subunits would respond as well as the receptors which also contain the γ or δ subunit. This was not the case, and the presence of the γ or δ subunit was necessary to exhibit the PAM action of DCUK-OEt. In fact, the type of γ subunit expressed with the α and β subunits was important, with the γ1 subunit being significantly less effective than the γ2 subunit. Because of the absence of the phenyl ring substituent (C ring) that generates functional (PAM) benzodiazepine derivatives, DCUK-OEt would not be expected to be an agonist at the benzodiazepine site, and our electrophysiologic experiments in the presence of flumazenil support this contention. It was, however, interesting that the combined effects of flunitrazepam and DCUK-OEt produced significantly more than an additive effect, possibly indicating an allosteric interaction between the benzodiazepine site and the site on the α + βinterface which binds DCUK-OEt with higher affinity.
The radioligand binding studies that led us to the electrophysiological examination of DCUK-OEt on the GABA A receptor, also produced some insight into the possible mechanism by which DCUK-OEt may generate its effects. DCUK-OEt produced a decrease in the affinity for muscimol at the GABA A receptor. Such action may be expected if DCUK-OEt is shifting the GABA A receptor into a state more likely to be in an open channel configuration. The GABA A receptor has been shown to display two affinity states for agonists such as muscimol 46,47 and the high affinity state of the GABA A receptor has been proposed to represent stabilization of the desensitized form of the receptor 48 . One can speculate that DCUK-OEt is increasing the proportion of receptors in a low affinity state at any particular concentration of agonist (muscimol). This speculation will require more investigation, but it is interesting that ethanol 49 and the anxiolytic/anticonvulsant etifoxine 50 , which both can act as PAMs at lower concentrations, reduce muscimol affinity at GABA A receptor in rat brain membrane preparations.
The α1β2γ2 combination of subunits is the primary combination of synaptically localized GABA A receptors in brain that mediate phasic inhibition, while α1/4/6βxδ receptors have been considered to be the primary type of extrasynaptic GABA A receptors that mediate tonic inhibition 51 . Given our results with GABA A receptors containing α4 and α1 subunits together with the γ2 or δ subunit, one could assume that DCUK-OEt would well affect the function of synaptically localized GABA A receptors as well as certain extrasynaptic GABA A receptors. We noted two characteristics of DCUK-OEt that suggest that its primary effect may be at extrasynaptic receptors containing either a γ2 or δ subunit together with an α1 and β3 subunit. These combinations of subunits (α1β3γ2 and α1β3δ) display a low EC 50 for GABA (see Supplementary Fig. S6) and DCUK-OEt can produce highly significant potentiation of α1β3γ2 and α1β3δ-mediated currents at the EC 10 concentration of GABA in our assays, and probably at concentrations of GABA consistent with those encountered in locations outside of the GABA synapse. This observation would be quite compatible with significant potentiation at extrasynaptic sites where concentrations of GABA have been considered to be in the high nM range, as opposed to the high concentrations (mM) of GABA that are present in the synapse 52 . We saw no measurable effect of DCUK-OEt on α1β2γ2 receptors at high concentrations of GABA (EC 60 and above), and non-significant effects on α1β1γ2 and α1β2γ1 GABA A receptors at low GABA concentrations (EC 10 ). Since αβγ is responsible for the major portion of the phasic actions of GABA, and relatively high amounts of β1 and γ1 were reported at synaptic sites in CeA [10][11][12]14 , it is plausible that phasic effects of GABA through these subunit combinations would not be modulated by DCUK-OEt. In fact, when we applied DCUK-OEt focally to CeA neurons, we found no change in sIPSC frequency, amplitude, rise or decay time, indicating no effects on phasic transmission (Fig. 4c).
There is strong evidence for the existence of α1βxδ receptors located extrasynaptically in particular areas of brain (i.e., the interneurons of the hippocampus and particularly those of the dentate gyrus) [53][54][55] . Tonic inhibition mediated by GABA A receptors containing the α1 subunit has also been noted in the CeA 56 . Our prior studies using slices of the CeA demonstrated that CRF1 receptor-positive (CRF1+) neurons express the α1 GABA A receptor subunit, and this subunit is integral for the GABA-mediated tonic conductance in these neurons as well as being involved in the phasic synaptic response to GABA 56 . When we measured tonic conductance in CeA neurons, focal application of DCUK-OEt produced an enhancement of the recorded tonic current, suggesting local effects of DCUK-OEt at extrasynaptic GABA A receptors. To further ascertain whether the effects of DCUK-OEt were mediated particularly by extrasynaptic GABA A receptors, we performed a comparison of the change (increase) in current produced by DCUK-OEt and the decrease generated by the subsequent co-application of 100 μM gabazine 57 . The strong correlation indicated that DCUK-OEt was indeed stimulating a tonic conductance in these neurons by actions at extrasynaptic GABA A receptors. Recently, de Guglielmo et al. 58 reported that inactivation of an ensemble of neurons in the CeA resulted in abrogation of excessive alcohol consumption by alcohol-dependent rats. The anatomical area of the CeA from which we obtained our electrophysiologic data coincides with the area containing the ensemble described by de Gugielmo et al. 58 . An increase in the tonic conductance through extrasynaptic GABA A receptors, mediated by DCUK-OEt, may engender reduced activity of the neurons identified by de Guglielmo et al. 58 and be an effective mode for reducing alcohol intake by dependent animals.
In all, our characterization of DCUK-OEt indicates that this molecule has characteristics that resemble those of etomidate, other anesthetics, ethanol and neurosteroids, but the full profile of DCUK-OEt actions speaks to an interaction with a site or sites on the GABA A receptor that distinguish DCUK-OEt from currently known PAMs and direct agonists acting at GABA receptors.

Radioligand binding. [ 3 H]Flunitrazepam Binding and Displacement by DCUK-OEt. Membrane
Preparation. These experiments were performed at the University of Colorado Health Sciences Center, Denver, CO. Experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Colorado, Denver, and were performed in accordance with the NIH Guide for the Care and Use of Laboratory Animals. Male Sprague-Dawley rats (200-250 g) were maintained in an AAALAC-accredited facility and sacrificed by CO 2 exposure and decapitation. Brains were removed, and membranes were prepared from the forebrain as described previously 6 .
Ligand binding assay. The binding of [ 3 H]flunitrazepam was assayed in triplicate, using final incubation volumes of 0.55 ml consisting of protein (approx 200-300 mg/ml), [ 3 H]flunitrazepam (New England Nuclear) at a concentration of 1 nM, 10 µM GABA and DCUKA or DCUK-OEt at 0, 5, 10, 20, 50, 100 and 200 µM in DMSO solution (final DMSO concentration 0.2%). Nonspecific binding was measured in the presence of 10 µM diazepam. Binding was initiated by addition of protein, followed by incubation at 4 °C for 30 min. Bound and free ligand were separated by rapid filtration under vacuum over Whatman GF/B filters presoaked in buffer in a 24 port Brandel Cell Harvester. Filters were washed with 2 × 5 ml of ice-cold HEPES buffer and dried prior to measurement of bound radioactivity by scintillation counting (Beckman LS3800 scintillation counter) using Ultima Gold XR scintillation cocktail.

Displacement of [ 3 H]muscimol binding by DCUK-OEt or DCUKA.
The assays of [ 3 H]muscimol binding were performed by the Psychoactive Drug Screening Program/NIMH (PDSP). Rat brain membranes were prepared as described in the Protocol Manual on the PDSP website (https://pdspdb.unc.edu/pdspWeb/). DCUK-OEt or DCUKA were dissolved in 1.0% v/v DMSO and assayed at 11 concentrations ranging from 0.05 nM to 10 μM (final DMSO concentration, 0.2%). The final concentration of [ 3 H]muscimol in the assay mixture was 5 nM. Displacement of [ 3 H] muscimol by GABA at concentrations ranging from 10 nM to 10 μM was measured to provide a positive control. Table 1), were also performed by PDSP and in our laboratories (batrachotoxin binding) 6 . The experimental details for all of the PDSP binding studies can be obtained by connecting to the PDSP website (https://pdspdb.unc.edu/pdspWeb/) and clicking on "Assays" (binding or functional) on the menu bar. PDSP initially performed ligand displacement studies at a default concentration of 10 μM DCUK-OEt. For any receptor/ transporter at which the compound generated a 50% or greater displacement of the receptor/transporter-selective ligand, a secondary binding assay was performed to calculate K i values (see below).

Screening for binding of DCUK-OEt to other receptors/transporters/enzymes. Additional ligand binding studies (Supplementary
Analysis Oocyte electrophysiology. Xenopus laevis frogs were obtained from Nasco (Fort Atkinson, WI, USA). All surgery was performed in accordance with a protocol approved by the University of Texas, Austin IACUC and the NIH Guide for the Care and Use of Laboratory Animals. The complementary DNAs encoding the GABA A subunits rat α1, β1, β3, γ2 s, δ, and human β2 were provided by Drs Myles H. Akabas, Paul J. Whiting and Richard W. Olsen. Human γ1 cDNA was synthesized de novo, optimized for Xenopus laevis oocyte expression and subcloned in pGEMHE by GenScript (Piscataway, NJ). The in vitro transcription of GABA A subunits was performed using mMessage mMachine (Life Technologies, Grand Island, NY). After isolation of Xenopus laevis oocytes, they were injected with capped complementary RNAs encoding wild-type or mutant subunits in different ratios, depending on the subunits: α1β2γ2 s, 2:2:20 ng; α1β2γ1, 2:2:6 ng; α1β2, 3:3 ng; α1β1γ2, 0.5:0.5:5 ng; α1β3γ2, 0.1:0.1:1 ng; α1β3, 0.5:0.5 ng; α1β3δ, α4β3γ2 and α4β3δ, 0.4:0.4:4 ng.
Electrophysiology. The injected oocytes were incubated at 15°C in sterilized Modified Barth's solution for 1-7 days before recording, and the responses of GABA A receptors expressed in oocytes were studied using two-electrode voltage clamp as described earlier 33,60 . Oocytes were discarded if the maximal current was over 20 µA or if the baseline was unstable or drifted to positive values.
Modulator application. DCUK-OEt and DCUKA stock solutions were prepared in DMSO weekly, then sonicated for 15 min, and stored at 4 °C, protected from light. On the day of the experiment, dilutions were prepared, sonicated for 5 min, and used immediately. The final DMSO concentration in the buffer bathing the oocyte was ≤0.1%. In order to test the effects of DCUKs, the agents were first pre-applied for 1 min and then co-applied with GABA. To verify the presence of a third subunit in expressed subunit combinations, the responses to GABA in the presence of Zn++ (1, 10 or 100 µM) were evaluated (Supplementary Table S2). The application sequence in each instance was as follows: Maximal GABA (20 s application, 15 min washout), EC 10 GABA (30 s application, 5 min washout), EC 10 GABA, pre-application of the drug followed by a co-application with EC 10 GABA, EC 10 GABA, pre-application of Zn++ immediately followed by a co-application with EC 10 GABA, EC 10 GABA. In most cases, we limited to one DCUK application per oocyte. Flumazenil and 17PA were pre-applied before their co-application with GABA. When co-applying with DCUK, the antagonist and DCUK were pre-applied together before their co-application with GABA. Flunitrazepam was not pre-applied before co-application with GABA.
Statistical Analysis. Responses to DCUK-OEt were quantified as the percent change in current between the response to the EC 10 concentration of GABA and the response to the EC 10 concentration of GABA in the presence of 0.3 μM concentration of DCUK-OEt. To control for batch effects a linear mixed model was implemented in SAS (version 9.4) to calculate the normalized percent change in current for each receptor subunit combination (each receptor combination was examined in two to nineteen separate experiments). Because each receptor was examined across several experiments, a random effect of batch was included in the model. For each receptor, the estimated percent change in the GABA EC 10 -induced current produced by addition of DCUK-OEt was compared to zero by ascertaining whether zero percent change was outside the confidence interval of the measured values. This was accomplished by using a single sample t-test in the MIXED procedure of SAS, and a Bonferroni adjustment to control for multiple comparisons across receptors. Comparisons between receptors with a single subunit difference were executed within the linear mixed model using linear contrasts. A Bonferroni adjustment was used to control for multiple pairwise comparisons. Significant effects are those with a Bonferroni adjusted p-value < 0.05 and marginal effects are those with a Bonferroni adjusted p-value < 0.2.
Other statistical tests (t-test and ANOVA) were applied as indicated in the corresponding table or figure legend.
The GABA concentration response curves (CRCs) were fitted to the following equation: Brain slice electrophysiology. Brain slice preparation. All procedures were approved by the Scripps Research IACUC and were consistent with the NIH Guide for the Care and Use of Laboratory Animals. Slices were prepared from brains of 5 adult male Wistar rats (250-350 g) as described by Herman et al. 56 . A single slice was transferred to a recording chamber mounted on the stage of an upright microscope (Olympus BX50WI).
Brain slice electrophysiological recording. Neurons were visualized and whole cell patch clamp recordings were made as previously described 56 . Series resistance was typically <15 MΩ and was continuously monitored with a hyperpolarizing 10 mV pulse. Electrophysiological properties of cells were determined by pClamp 10 Clampex software online during voltage-clamp recording using a 5 mV pulse delivered after breaking into the cell. The resting membrane potential was determined online after breaking into the cell using the zero current (I = 0) recording configuration and the liquid junction potential was included in the determination.
DCUKA and DCUK-OEt were prepared as described for the experiments on oocyte electrophysiology. Other drugs were dissolved in aCSF, and all drugs were applied by Y-tubing application for local perfusion primarily on the neuron of interest. To isolate the inhibitory currents mediated by GABA A receptors, all recordings were performed in the presence of glutamate and GABA B receptor blockers 56 . All voltage clamp recordings were performed in a gap-free acquisition mode with a sampling rate per signal of 10 kHz or a total data throughput equal to 20 kHz (2.29 MB/min) as defined by pClamp 10 Clampex software.
Data Analysis. Frequency, amplitude and decay of spontaneous inhibitory postsynaptic currents (sIPSCs) were analyzed and visually confirmed using a semi-automated threshold based mini detection software (Mini Analysis, Synaptosoft Inc.). Averages of sIPSC characteristics were determined from baseline and experimental drug conditions containing a minimum of 60 events (time period of analysis varied as a product of individual event frequency) and decay kinetics were determined using exponential curve fittings and reported as decay time (ms). All detected events were used for event frequency analysis, but superimposed events were eliminated for amplitude and decay kinetic analysis. In voltage clamp recordings, tonic currents were determined using Clampfit 10.2 (Molecular Devices) and a previously-described method 61 . Responses were quantified as the difference in holding current between baseline and experimental conditions. Events were analyzed for independent significance using a one-sample t-test and compared using a two-tailed t-test for independent samples, a paired two-tailed t-test for comparisons made within the same recording, and a one-way ANOVA with a Bonferroni post hoc analysis for comparisons made between 3 or more groups. All statistical analysis was performed using Prism 5.02 (GraphPad, San Diego, CA). Data are presented as mean ± SEM. In all cases, p < 0.05 was the criterion for statistical significance.

Molecular modeling.
All molecular modeling studies were conducted using Biovia Discovery Studio 2016 (Biovia Inc., San Diego, CA) and all crystal structure coordinates were downloaded from the protein data bank (www.pdb.org). The homology model of the human GABA A receptor pentamer was generated with the MODELLER protocol 62 utilizing the crystal structures of the human GABA A receptor β3 homopentamer as a template (PDB ID: 4COF 63 , Uniprot accession: P28742). Homology models of the human α1 (Uniprot accession: P14867) and γ2 (Uniprot accession: P18507) subunits were superimposed over the template, with the crystal structure of two β3 subunits, so that the final pentameric model consisted of two α1, two β3, and one γ2 subunits, arranged in an γβαβα pattern (counterclockwise, as seen from above). The resulting final structures were subjected to energy minimization utilizing the conjugate gradient minimization protocol with a CHARMm forcefield and the Generalized Born implicit solvent model with simple switching (GBSW) 64 . The minimization calculations converged to an RMS gradient of <0.01 kcal/mol. The Flexible Docking protocol 65 , which allows flexibility in both the protein and the ligand during the docking calculations, was used to predict the binding orientations of both known and candidate GABA A PAMs in the binding site located at either the classical α-γ benzodiazepine site (α + γinterface) or the alternative α-β site (α + βinterface). Predicted binding poses were energy-minimized in situ using the CDOCKER protocol 66 prior to calculation of binding energies using the distance-dependent dielectric model. All numeration refers to the corresponding mature protein.