The diazepam binding inhibitor’s modulation of the GABA-A receptor is subunit-dependent

First synthesized in the 1950s, benzodiazepines are widely prescribed drugs that exert their anxiolytic, sedative and anticonvulsant actions by binding to GABA-A receptors, the main inhibitory ligand-gated ion channel in the brain. Scientists have long theorized that there exists an endogenous benzodiazepine, or endozepine, in the brain. While there is indirect evidence suggesting a peptide, the diazepam binding inhibitor, is capable of modulating the GABA-A receptor, direct evidence of the modulatory effects of the diazepam binding inhibitor is limited. Here we take a reductionist approach to understand how purified diazepam binding inhibitor interacts with and affects GABA-A receptor activity. We used two-electrode voltage clamp electrophysiology to study how the effects of diazepam binding inhibitor vary with GABA-A receptor subunit composition, and found that GABA-evoked currents from α3-containing GABA-A receptors are weakly inhibited by the diazepam binding inhibitor, while currents from α5-containing receptors are positively modulated. We also used in silico protein-protein docking to visualize potential diazepam binding inhibitor/GABA-A receptor interactions that revealed diazepam binding inhibitor bound at the benzodiazepine α/γ binding site interface, which provides a structural framework for understanding diazepam binding inhibitor effects on GABA-A receptors. Our results provide novel insights into mechanisms underlying how the diazepam binding inhibitor modulates GABA-mediated inhibition in the brain.


Abstract 22
First synthesized in the 1950s, benzodiazepines are widely prescribed drugs that exert 23 their anxiolytic, sedative and anticonvulsant actions by binding to GABA-A receptors, the main 24 inhibitory ligand-gated ion channel in the brain. Scientists have long theorized that there exists 25 an endogenous benzodiazepine, or endozepine, in the brain. While there is indirect evidence 26 suggesting a peptide, the diazepam binding inhibitor, is capable of modulating the GABA-A 27 receptor, direct evidence of the modulatory effects of the diazepam binding inhibitor is limited. 28 Here we take a reductionist approach to understand how purified diazepam binding 29 inhibitor interacts with and affects GABA-A receptor activity. We used two-electrode voltage 30 clamp electrophysiology to study how the effects of diazepam binding inhibitor vary with GABA-31 A receptor subunit composition, and found that GABA-evoked currents from α3-containing 32 GABA-A receptors are weakly inhibited by the diazepam binding inhibitor, while currents from 33 α5-containing receptors are positively modulated. We also used in silico protein-protein docking 34

Introduction 40
Since their discovery over fifty-five years ago, benzodiazepines (BZDs) have been some 41 of the most widely prescribed drugs in the world, and have remained on the World Health 42 Organization's list of essential drugs since the list's inception in 1977 [1]. BZDs are used to treat 43 a variety of conditions, ranging from epilepsy to insomnia to anxiety [2][3][4], and mediate their 44 effects by binding to the main inhibitory neurotransmitter receptor in the brain, the gamma-45 aminobutyric acid-A receptor (GABAR) [5]. BZDs modulate GABAR activity, and thus alter 46 neuronal signaling. 47 GABARs are heteropentamers comprised from nineteen possible subunits (α1-6, β1-3, 48 γ1-3, δ, ε, , π and 1-3) that are expressed in distinct brain regions, with distinct 49 pharmacological properties. There are two GABA binding sites in the extracellular interfaces 50 between β and α subunits, while BZDs bind in an extracellular pocket at the α/γ subunit 51 interface. BZDs, such as diazepam or flurazepam (FZM), can act as positive allosteric 52 modulators (PAMs), which increase GABA-mediated currents, or negative allosteric modulators 53 (NAMs) like methyl-6,7-dimethoxy-4-ethyl-beta-carboline-3-carboxylate (DMCM) which 54 decrease GABA-elicited currents. Zero-modulators, such as flumazenil (FLZ, Ro15-1788) when 55 acting on α1 and α5-containing GABARs [6], occupy the BZD site but do not alter GABA-elicited 56 currents. The effects of some BZDs like Ro15-4513 depend on GABAR subunit composition. 57 Ro15-4513 acts as PAM when interacting with α4β2γ2 and α6β2γ2 GABARs, but is a NAM when 58 bound to α1β2γ2 receptors [7]. 59 The presence of a binding site on the GABAR for synthetically manufactured BZDs has 60 long suggested the existence of an endogenous synthesized molecule that binds to the same 61 site. In 1983, a candidate peptide was isolated from rat brain homogenates, which displaced 62 tritiated diazepam in a radioligand binding assay [8]. This 10kD, 87 amino-acid protein was 63 named the diazepam binding inhibitor (DBI). Recent DBI knock-down and over-expression 64 studies in the thalamic reticular nucleus demonstrate that DBI works in vivo as a PAM where it 65 potentiates GABA-mediated currents and suppresses epileptic activity [9]. In contrast, DBI 66 knock-down in the subventricular zone of the lateral ventricles and in the hippocampal 67 subgranular zone demonstrate that DBI and one of its peptide fragments, ODN, inhibit GABA-68 induced currents [10,11], indicating that in these brain regions DBI and ODN work as NAMs of 69 Buffer A to a final concentration of 50μM. Aprotinin has a mass of 6.5kDa. 134

135
GABA concentration-response curves were determined as described previously [21]. I is the peak current response to a given GABA concentration, Imax is the maximal amplitude of 140 GABA activated current, EC50 is the GABA concentration that produces the half-maximal 141 response, [A] is test GABA concentration, and nH is the Hill coefficient. 142

Drug modulation
143 Drug modulation was measured at GABA EC20 and evaluated as IGABA+Drug/IGABA, where 144 IGABA+Drug is the GABA-mediated current in the presence of drug and IGABA is the GABA-mediated 145 current in the absence of drug. To measure drug modulation, a 5s application of GABA EC20 146 was applied and followed immediately by a 5s application of GABA EC20+Drug. When multiple 147 drugs were applied to a single oocyte, GABA EC20 was applied following drug treatment until 148 current amplitudes returned to initial GABA EC20 level to ensure complete washout of drugs 149 between different drug treatments. 150 Protein-protein docking 151 We made α3β3γ2L and α5β3γ2L GABAR homology models based on the cryoEM structure 152 of the α1β3γ2L GABAR obtained in the presence of the BZD alprazolam (PDB 6HUO) [22]. The 153 homology models were built on Sybyl (Tripos Inc). The GABAR α3 or α5 sequences were 154 manually threaded on to the cryoEM model of the α1β3γ2L GABAR based on their sequence 155 alignment with the α1 subunit using the graphical interface of the Sybyl program. After resolving 156 steric clashes, either manually or by using focused minimization, the resulting structures were 157 subject to whole molecule energy minimization using the Tripos force field. Gasteiger-Huckel 158 charges were added to the atoms, and the dielectric function set at 1.0. After Simplex initial 159 minimization, the Powell function in Tripos was used to minimize to a termination gradient of 160 0.05 kcal/(mol x angstrom). The GABAR model images were developed using PyMOL 161 (Schrödinger, LLC, New York). BZD alprazolam was removed from models prior to docking. rather than an energy score. Docking clusters selected for analysis were based on their 168 proximity to the BZD binding site. We analyzed the hydrogen bonding between the GABAR 169 alpha subunits and DBI and the GABAR gamma subunit and DBI using Pymol's hydrogen 170 bonding feature. 171

Statistical analysis
172 All data were from at least three different oocytes from at least two different frogs. Data 173 are represented as mean ± SD. Significant differences in drug modulation between subunit 174 compositions were calculated via Kruskal-Wallis test with a Dunn's multiple comparisons (Prism 175 v9.1, GraphPad Software Inc, San Diego, CA). This test was selected due to the non-normal 176 distribution of data, as evaluated using a D'Agostino and Pearson test which evaluates 177 skewness and kurtosis and generates a P value based on how much these values differ from a 178 Gaussian distribution [27]. Normalized values are used to compare the effects of drug or DBI 179 between α3β3γ2L and α5β3γ2L GABARs. Normalized values were also compared to a hypothetical 180 null value of 1 using a one-sample t-test in Prism. 181 Statistical differences in GABA+drug versus GABA alone current amplitidues elicited 182 from the same oocyte were calculated using a ratio paired t-test of raw, non-normalized values 183 using Prism software rather than a standard paired t-test. For our data, differences between 184 control and treatment is not a consistent measure of effect. The differences are larger when 185 the control current amplitudes are larger. Thus, the ratio (treated/control) is a more 186 consistent way to quantify the effect of the treatment [28]. 187 The effects of DBI on α5β3γ2L GABA-elicited currents were more variable than effects 188 observed with FZM, a BZD (coefficient of variation for FZM α5β3γ2L is 17.2%, DBI is 36.0%). The 189 variation was not correlated with DBI purification batch nor oocyte/frog. We hypothesize that 190 flexibility and conformation dynamics of DBI (11kDa) may contribute to the increased variability 191 of its effects as compared to a small BZD drug.

198
In order to study the effects of DBI on the GABAR receptor, we needed milligram 199 quantities of pure, non-aggregated DBI folded predominantly in a single conformation. We used 200 a histidine-tagged human DBI construct (His-DBI) inserted into a pET28A vector, which is 201 optimized for inducible expression of protein in bacterial cells [29]. We expressed the His-DBI in 202 BL21 E. coli cells and purified the protein as described in Methods. DBI purity and size was 203 evaluated using 15% SDS-PAGE ( Fig 1A). Following purification, a single coomassie blue 204 values were 47±11.8 μM, n=5 for α3β3γ2L GABARs (Fig 2A) and 34±8.6 µM, n=3 for α5β3γ2L 227 GABARs (Fig 2B), which are consistent with previously published results [32,33]. We used the same methods to evaluate the effects of purified DBI (50μM) on GABA 261 EC20 currents from α3β3γ2L and α5β3γ2L GABARs. DBI weakly inhibited GABA currents from 262 α3β3γ2L GABARs (Fig. 3). Current amplitudes in the presence of DBI were significantly less than 263 from currents elicited by GABA alone (Fig. 3C, ratio paired t-test p=0.001). In contrast, DBI 264 significantly potentiated GABA currents from α5β3γ2L GABARs 1.3-fold (Fig 3, ratio paired t-test  265 p<0.0001). The data indicate that DBI acts as a very weak NAM for α3-containing receptors and 266 a PAM for α5-containing receptors, with a significant difference in the effects of DBI based on 267 subunit combination (Dunn's p<0.0001, Fig 3B). 268 Since the effect of DBI on α3β3γ2L GABARs was modest, we measured GABA currents 269 elicited from repeated pulses of EC20 GABA as a control. There were no significant differences 270 in the amplitudes elicited by two EC20 GABA pulses prior to GABA+DBI application (GABA1 271 and GABA2, ratio paired t-test α3β3γ2L p=0.08, α5β3γ2L p=0.08, S2A-B). Moreover, GABA-elicited 272 current amplitudes recovered to pre-DBI size after DBI was washed off. GABA currents 273 preceeding DBI (GABA2) and following GABA+DBI (GABA3) were not statistically different from 274 each other (ratio paired t-test α3β3γ2L p=0.36, α5β3γ2L p=0.33, S2C-D). Normalized GABA1 and 275 GABA3 (IGABA/IGABA2) were not significantly different than 1 (α3β3γ2L GABA1 p=0.22,GABA3 276 p=0.97;α5β3γ2L GABA1 p=0.24,GABA3 p=0.19; S2E-F) whereas normalized GABA+DBI were 277 significantly different (p<0.0001). Thus while the effect of DBI on α3β3γ2L GABARs was modest, 278 it is not due to a solution artifact. 279 In order to examine whether the effects of DBI were specific, we measured the effects of 280 aprotinin, a small soluble protein like DBI, on GABA currents elicited by EC20 GABA from 281 α5β3γ2L GABARs. Co-application of 50μM aprotinin did not affect GABA-elicied amplitudes (ratio 282 paired-test p=0.64, S3), indicating that the effects of DBI are specific and are not due to just 283 applying a small protein to the GABAR. 284 α3 and α5 are homologous GABAR subunit isoforms with many conserved residues 285 (73% identity) [37]. To explore potential mechanisms underlying DBI's actions on α3-containing 286 versus α5-containing GABARs, we constructed α3β3γ2L and α5β3γ2L GABAR homology models 287 and examined protein-protein interactions with DBI using ClusPro, a web-based program for For both α3β3γ2L and α5β3γ2L GABARs , ClusPro generated many highly populated 292 clusters with DBI bound at the BZD α/γ binding site interface (Fig.4)  Extracellular domains of α3 and α5 GABAR subunits with residues predicted to H-bond to DBI 309 shown in stick and colored red. α3 and α5 GABAR sequence alignments shown below, with H-310 bonding residues highlighted in red and non-identical residues underlined. C) DBI with residues 311 predicted to H-bond to α3 and α5 subunits shown in stick and colored in green. DBI sequences 312 shown below with H-bonding residues colored green. D) Extracellular domain of γ GABAR 313 subunit with residues predicted to H-bond to DBI shown in stick and colored blue. E) DBI with 314 residues predicted to H-bond to γ subunit shown in stick and colored grey. DBI sequences 315 shown below. For the highlighted hydrophobic residues such as DBI G46, M47, and L48, H-316 bonding is via the backbone, not the residue side-chain. 317 318 When comparing DBI bound to α3 and α5-containing receptors, several differences in 319 interactions between DBI and the GABAR were observed. The GABAR α5 subunit had more 320 residues predicted to H-bond to DBI than the α3 subunit. For example, the α3 histidine at 321 position 142 in Loop7 is not predicted to H-bond with DBI but the aligned α5 glutamine does. In 322 Loop C, the α3 Ile202 does not H-bond with DBI but the aligned α5 asparagine does. In 323 addition, DBI had more H-bond interactions with the γ subunit of α5-containing GABARs. 324 Overall, the computational dockings provide support for the idea that DBI can bind at the BZD 325 binding site interface of GABARs with its flexible ODN loop buried in the BZD binding pocket. 326 The docking is consistent with published work from the Huguenard group showing that adding 327 flumazenil or mutating the BZD site residue 3H126 reduces DBI effects on GABA currents 328 linking DBI's functional modulation to action at the BZD binding site [9]. The dockings provide a 329 framework for revealing structural mechanisms underlying DBI effects on GABARs. 330

331
Despite the fact that DBI was identified as a putative endozepine in the mid-1980s [8], 332 the mechanisms underlying DBI's actions are still unclear. Due to DBI's intracellular roles in 333 long-chain fatty acid metabolism, and its multiple biologically active cleavage products [39][40][41][42], 334 direct effects of DBI on GABAR function have not been extensively studied. DBI is highly 335 expressed across many tissue types, with high expression in liver, breast, blood, and the brain. 336 In the brain, mRNA DBI expression patterns obtained from the Human Protein Atlas show DBI 337 expression is high in the hippocampus, amygdala, thalamus and midbrain [43,44]. 338 The use of flumazenil, Ro15-1788 (FLZ, a BZD binding site antagonist for some GABAR 339 subtypes) provides indirect evidence for the existence of an endogenous GABAR modulator. zone and hippocampal subgranular cells suggest DBI is a NAM [10,11]. In this study, using 346 purified DBI and heterologous expression of GABARs, we found that DBI is a PAM for α5-347 containing GABARs and a weak NAM for α3-containing receptors demonstrating that effects of 348 DBI on GABAR activity are regulated by receptor subunit composition. 349 Recordings from cells in the hippocampal subgranular zone suggest that DBI or its 350 peptide cleavage product, ODN, acts as a NAM [11]. Given α5 subunits are highly expressed in 351 the hippocampus [48], we expected that heterologously expressed α5-containing receptors 352 would be negatively modulated by DBI versus positively modulated, which we observed (Fig. 3). 353 Similarly, patch clamp recordings from the thalamic reticular nucleus where α3 is highly 354 expressed [49] suggested that these receptors are positively modulated by DBI [9], while our 355 experiments with heterologously expressed α3β3γ2L GABARs demonstrated that DBI acts as 356 weak NAM (Fig. 3). 357 The differences between data reported here and previously reports are likely due to 358 experimental systems used. In neurons, GABARs are likely associated with accessory subunits 359 such as GARLH or Shisa7 [50,51], which may alter DBI actions. Furthermore, α3 and α5 are 360 not the only α subunits expressed in the thalamic reticular nucleus and hippocampus, 361 respectively. In native cells, the presence of GABARs comprised of other subunits that may be 362 modulated by DBI make it difficult to assign DBI's effects to one specific GABAR subtype. Here, 363 using a reductionist approach, we demonstrate that DBI is a PAM of α5β3γ2L GABARs and a 364 lay an important foundation for understanding how inhibition in the brain is regulated. 379  Overall, we observed the purification of a single ~11 kDa protein, which is the expected size of 388

S3 -Aprotinin does not affect GABA-elicited currents. A) Sample GABA current traces from 401
α5β3γ2L GABARs elicited from two sequential applications of EC20 GABA (left) or EC20 GABA 402 immediately followed by EC20 GABA + 50μM aprotinin, a small soluble protein like DBI (right). 403 Co-application of aprotinin had no effect on GABA-elicited current amplitude. B) Plotted are 404 currents from GABA EC20 alone or GABA+Aprotinin normalized to initial GABA EC20 current 405 (I/IGABA). Data from individual oocytes are shown as filled circles, bar graphs represent mean +/-406 SD. Dotted line at 1 represents no effect of drug. (C)