Tricyclic antipsychotics and antidepressants can inhibit α5‐containing GABAA receptors by two distinct mechanisms

Background and Purpose Many psychotherapeutic drugs, including clozapine, display polypharmacology and act on GABAA receptors. Patients with schizophrenia show alterations in function, structure and molecular composition of the hippocampus, and a recent study demonstrated aberrant levels of hippocampal α5 subunit‐containing GABAA receptors. The purpose of this study is to investigate the effects of tricyclic compounds on α5 subunit‐containing receptor subtypes. Experimental Approach Functional studies of effects by seven antipsychotic and antidepressant medications were performed in several GABAA receptor subtypes by two‐electrode voltage‐clamp electrophysiology using Xenopus laevis oocytes. Computational structural analysis was employed to design mutated constructs of the α5 subunit, probing a novel binding site. Radioligand displacement data complemented the functional and mutational findings. Key Results The antipsychotic drugs clozapine and chlorpromazine exerted functional inhibition on multiple GABAA receptor subtypes, including those containing α5‐subunits. Based on a chlorpromazine binding site observed in a GABA‐gated bacterial homologue, we identified a novel site in α5 GABAA receptor subunits and demonstrate differential usage of this and the orthosteric sites by these ligands. Conclusion and Implications Despite high molecular and functional similarities among the tested ligands, they reduce GABA currents by differential usage of allosteric and orthosteric sites. The chlorpromazine site we describe here is a new potential target for optimizing antipsychotic medications with beneficial polypharmacology. Further studies in defined subtypes are needed to substantiate mechanistic links between the therapeutic effects of clozapine and its action on certain GABAA receptor subtypes.

(a) 100μM CLZ modulation of currents elicited by an EC5-10 GABA concentration in α1β2, α1β2γ2, α1β3, α1β3γ2, α2β3, α2β3γ1, α2β3γ2 and α5β3γ2 receptors. Precise n numbers are in Supplementary Table S1. Statistically significant differences were determined between the receptor subtypes by one-way ANOVA followed by Dunnett's multiple comparisons test, where *p<0.05. All α1-containing receptors were statistically different from α2-and α5-containing receptors. Additionally, one sample t test was performed to determine statistical significance of each mean response from control current and corrected for multiple comparisons using the false discovery rate method of Benjamini and Hochberg, with a discovery rate of 0.05 ( # p<0.05). All responses were significantly different from control. (b) 30, 100, 300μM CLZ modulation of currents elicited by an EC20-30 GABA concentration in α3β3 receptors (n=5). Data are depicted as mean ± SEM. The dotted line is used to visualize the baseline (100%) of control current. (c) Modulation of currents elicited by 30μM concentration of the neurosteroid THDOC in α1β3 receptors co-applied with 100μM CLZ, after pre-application of CLZ (n=5). (d) Modulation of currents elicited by an EC20-30 GABA concentration by 100μM CLZ in α5β2γ2 (n=5) and β2γ2 (n=5) receptors. Columns for each receptor subtype depict mean ± SEM. Statistically significant differences were determined by two-tailed student's t-test, where *p<0.05. One sample t test was performed to determine statistical significance of each mean response from control current, where # p<0.05. The response in β2γ2 receptors was not significantly different from control. Representative traces from electrophysiological recordings of CLZ co-applied with GABA in β2γ2 and α5β2γ2 receptors are also depicted. (e) Modulation of currents elicited by an EC40 GABA concentration by 100μM in β2γ2 (n=5) receptors. One sample t test was performed to determine statistical significance of each mean response from control current, where # p<0.05. The response in β2γ2 receptors was not significantly different from control.

Color Tanimoto
Tables providing the individual shape, color and combo scores. The shape score represents the volumetric overlap of the Van der Waals volumes of the atoms in a ligand pair. Color scores represent the overlap of chemical features. The scores were calculated using the software ROCS and have values ranging from 0 to 1 (2 for Tanimoto Combo) where 0 means no overlap at all and 1 (2 for Tanimoto Combo) means a perfect match. The compound groups that result from proximity in the 2D scatter-plots are emphasized in the combo score table by different background colors.
Hierarchical clustering dendrograms for the calculated combo (left), shape (middle) and color Tanimoto similarity scores. The y-axis represents a distance D which was calculated from the similarity scores S as D = 1 -S. The combo score, which ranges from 0 to 2, has been normalized before distance calculation by a factor of 0.5.

Supplementary Figure S4
2D scatter-plots reflecting compound similarity in terms of shape (left) and common chemical features (right). The euclidean distance between two compounds correlates with the deviation of their similarity score from 1.0 (= the maximum achievable similarity score). Points are color coded by their membership in the clusters identified from the combo score scatter-plot ( Figure 2b).
Pharmacophore features of the groups that emerged from the combo score scatter plot in Figure 2b, with CLZ added to each cluster. Features: 1 -aromatic (blue donuts), 2 -hydrophobic (yellow spheres), 3 -positive ionizable (blue stars/rays), 4 -hydrogen bond acceptor (red sphere), 5 -halogen bond (magenta arrow). CLZ can be seen to differ in shape from both groups.

Supplementary Figure S6
Sequence alignment and the bicuculline bound 6X3S ribbon structure of the ECD. The alignment and the α1 subunit are color coded to indicate diverging positions among α1, α2, α3, and α5 subunits (blue) in the context of the orthosteric site and the putative intrasubunit CPZ site. The segments (loops) D, E, F and G which contribute to the orthosteric site are labelled, the residues that form the CPZ site are shown in yellow. Strands 1 and 8 (located on segments G and F) contribute to the complementary face of the orthosteric site with sidechains facing towards the interface, and to the intrasubunit CPZ pocket with sidechains facing to the packing core as evidenced by the blue/ yellow alternating colors. Thus, these segments may contribute to potency and efficacy of events at both sites.

Supplementary Figure S8
Modulation of currents elicited by an EC15-30 GABA concentration by 10 & 30μM LOX (a) and 30μM CLOT (b) in α5β3γ2 (n=5 for 30μM LOX and n=6 for 10μM LOX, n=10 for 30μM CLOT) and α5F53W;L222Wβ3γ2 (n=5 for 30μM LOX and n=7 for 10μM LOX, n=5 for 30μM CLOT) receptors. Columns for each receptor subtype depict mean ± SEM. Statistically significant differences were determined by two-tailed students t-test, where *p<0.05. All responses were found to not be significantly different between wild-type and mutated receptors. The dotted line is used to visualize the baseline (100%) of control current.
(a, b) CLZ (a) and CPZ (d) dose response curves in α5β3γ2 and α5F53Wβ3γ2 receptors. Data were normalized and fitted to the Hill equation using non-linear regression (fixed bottom of 0) and points are depicted as mean ± SEM. The precise n numbers, as well as the IC50, logIC50, Hill slope and maximum efficacy values can be found in Supplementary  Tables S6 and S7. Dose response curves in α5β3γ2 receptors are represented with dotted lines, as they are reproduced here for easier comparisons.  Figure 5a and in Supplementary Figure S9. Control current = 100% (GABA EC20-30). GABA concentrations can be found in Supplementary Table S10.  Figure 5d and in Supplementary Figure S9. Control current = 100% (GABA EC20-30). GABA concentrations can be found in Supplementary Table S10. Best CPZ binding modes. CPZ docking into 6A96 WT (white) and the selected mutants F53W (grey), L222W (pink), F53W;L222W (red), showing the three different candidate binding modes (BM1-3) which are consistent with a small right shift (loss of apparent affinity) in L222W and F53W;L222W. Binding mode 3 (panels i-l) shows similarity to the CPZ in the ELIC bound structure. Panels d, h, l display the interacting features as identified by Ligand Scout for the three binding modes. Of note, binding mode 1 is in best agreement with the ligand based analysis (Figure 2c).

Supplementary
(a) 2D representations of BIC, CPZ, CLZ and LOX with the matched features identified in the screen (yellow sphere: hydrophobic, blue donut: aromatic, blue rays: cationic). The cationic feature forms cation-pi interactions with Y157, the hydrophobic/aromatic feature is in close contact with multiple aromatic and hydrophobic sidechains. (b) Inhibition of binding of [3H]muscimol to rat hippocampal membrane GABAA receptors (n=3-5). Membranes were incubated with 10nM [3H]muscimol in the presence of various concentrations of the displacing ligand. 100% is the amount of radioligand bound in the presence of 1% DMSO. Data shown are mean ± SEM of three independent experiments performed in duplicates each (for the concentrations <1mM) and five independent experiments performed in duplicates each (for 1mM). Visual inspection and sigmoid fitting indicated that the displacement points are not described by a single sigmoid function, as would be expected due to different affinities for the diversity of subtypes that are present in hippocampal membranes. Therefore, the individual points are displayed without fitting. (c) Inhibition of binding of [3H]muscimol to rat hippocampal membrane GABAA receptors at 1mM BIC and LOX. Hippocampal membranes from five independent membrane preparations were incubated with 10nM [3H]muscimol in the presence of 1mM of displacing ligand in five independent experiments performed in duplicates each. One-way ANOVA followed by Tukey's multiple comparisons test was performed to determine statistically significant differences between BIC (n=5), LOX (n=5), CLZ (n=5) and CPZ (n=5), where *p<0.05 (see Figure 6f).

Supplementary Figure S12
The pKa-function of MarvinSketch was used to select the protonation states of CLZ for the docking based on pKapredictions using pKa values and microspecies distribution of the nitrogens in CLZ which are susceptible to protonation/deprotonation. The orange curve (1) corresponds to the completely unprotonated species (1), the green curve (2) to the species only protonated at the methyl-substituted piperazine nitrogen, and the blue curve (3) to the species additionally protonated at the amidine nitrogen. Thus, species 1 and 2 were docked as described in the methods. The tabular inset provides the scores from the top 10 solutions of the best candidate binding mode that is displayed in Figure 6g. The protonated species 2 assumes a binding mode which is top ranked with both scoring functions (chemPLP and chemsocore), and a very similar solution is found in two more top ten positions. Interestingly, the unprotonated species 1 can also assume this binding mode which is found a total of five times in the top ten results.
Chord diagram representing the LOX, CLOT, CLZ, CPZ, LEVO, IMI and NOR (left/bottom) and their mammalian drug targets (right/top) grouped according to the IUPHAR recommended categories. "Other targets" are described in more detail in Supplementary Table S8. All drugs and target full names, as well as their sources are listed in Supplementary  Table S9. For the chord diagram, DrugCentral (https://drugcentral.org/ accessed on 06.01.2020) was used and only mammalian drug targets were taken into account. Additionally, for GABAA and GABAB as well as for nAChRs, literature findings were added (Supplementary Table S9) [1][2][3][4][5][6][7][8][9][10] . Terminology was unified across all drug targets (since differences exist, e.g. Serotonin (5-HT3) receptor 3 and 5-hydroxytryptamine receptor 3). Targets were grouped according to the IUPHAR recommended categories. The chord diagram was created with python 3.8 and the python package chord (https://pypi.org/project/chord/).

Numbering Name
Gene name   All drugs and target full names, as well as their sources, as depicted in the chord diagrams in Supplementary Figure  S13. All references are listed in the manuscript. Asterisk (*) refers to unspecified subtypes/isoforms.