Midazolam at Low Nanomolar Concentrations Affects Long-term Potentiation and Synaptic Transmission Predominantly via the α₁–γ-Aminobutyric Acid Type A Receptor Subunit in Mice

All procedures were approved by the animal care committee (either Eberhard-Karls-University, Tübingen, Germany, or Technical University Munich, Munich, Germany) and were conducted in accordance with German law on animal experimentation. All efforts were made to minimize animal suffering and the number of animals used. Up to six mice were housed in a cage with ad libitum intake of food and water in an environmentally controlled room (23 ± 0.5°C) with a 12-h light/12-h dark cycle.

For long-term potentiation experiments, 6- to 10-week-old mice of either sex were used. Due to limited availability, older mice (17 to 19 weeks old) were used for the α_1/3/5–knock-in and α_2/3/5–knock-in lines. The wild-type (C57BL/6) mice were obtained from Charles River (Italy), and the α₁–knock-in (line designation: 129X1.129P2Gabra1<tm1.1Uru/Uru>10Gabra1SvRR), α₅–knock-in (line designation: 129X1-Gabra5<tm1.1Uru/Uru>), and α_1/2/3–knock-in (line designation: 129X1.129P2/129P2/129T2-Gabra1<tm1.1Uru>Gabra2<tm1.1Uru>Gabra3<tm1.1Uru>GAB-Aa123SvJ) mice were from Charles River. Genotyping from these global knock-in lines was performed by Charles River. The α_1/3/5–knock-in (line designation:129X1.129P2/129T2/129X1-Gabra1<tm1.1Uru>Gabra3<tm1.1Uru>Gabra5<t1.1- Uru>) and α_2/3/5–knock-in (line designation: 129X1.129P2/129T2/129X1>Gabra2<tm1.1Uru> Gabra3<tm1.1Uru>Gabra5<tm1.1Uru>) lines were obtained from H.-U. Zeilhofer’s group at the University of Zurich (Zurich, Switzerland). The α₁–knock-in mice carry an H101R point mutation in the α₁-subunit of the GABA_A receptor, and the α₅–knock-in mice carry an H105R knock-in mutation in the α₅-subunit of the GABA_A receptor. Receptors containing the H-to-R mutations are fully functional, but benzodiazepines are not able to bind to the benzodiazepine-binding site formed in part by the α₁- and α₅-subunits of these receptors. The mutation does not alter the physiologic function of the receptor (the natural GABA ligand can still bind) but makes them resistant to modulation by allosteric modulators acting at the benzodiazepine-binding site (e.g., midazolam).⁹  In the triple transgenic α_1/2/3–knock-in mouse line with the mutations α₁(H101R), α₂(H101R), and α₃(H126R), benzodiazepines cannot bind to α₁-, α₂-, or α₃-subunits; It binds only to α₅-subunit. Similarly, in α_1/3/5–knock-in and α_2/3/5–knock-in, the only available α-subunits of the GABA_A receptor to which benzodiazepines can bind to are α₂ and α₁. For preparing cultured neocortical tissue slices, we used 2- to 5-day-old wild-type and α_2/3/5–knock-in mice of both sexes, obtained from the sources mentioned above.

The mice were deeply anaesthetized with isoflurane before decapitation. The brain was rapidly removed from the head and was immediately placed in ice-cold lactated Ringer’s solution containing 125 mM NaCl, 2.5 mM KCl, 25 mM NaHCO₃, 0.5 mM CaCl₂, 6 mM MgCl₂, 25 mM d-glucose, and 1.2 mM NaH₂PO₄ with a final pH of 7.3 and saturated with carbogen gas (95% O₂/5% CO₂). Sagittal hippocampal slices 350 μm thick were obtained using a microtome (HM 650 V; Microm International, Germany) at 4°C after cutting the whole brain into two hemispheres with a razor blade. Slices were allowed to recover at 34°C for 30 min in a chamber submerged with artificial cerebrospinal fluid containing 125 mM NaCl, 2.5 mM KCl, 25 mM NaHCO₃, 2 mM CaCl₂, 1 mM MgCl₂, 25 mM d-glucose, and 1.2 mM NaH₂PO₄, which was also bubbled with carbogen, and then for 1 h at room temperature before they were transferred to the recording chamber. A platinum ring with two nylon filaments was used to fix the slices in the recording chamber while carbonated fluid was continuously perfused at a flow rate of 5 ml/min. All experiments were performed at room temperature (20 to 21°C).

Organotypic neocortical slice cultures were utilized for evaluating midazolam’s effects on GABA_A receptor–mediated synaptic currents. This preparation was used because neuronal population activity in this test system is sensitive to brain concentrations of midazolam in vivo, producing sedation and hypnosis in humans²²  (Supplemental Digital Content 1, table S1, https://links.lww.com/ALN/C845). Moreover, the study of Zeller et al.⁸  provided evidence that cortical glutamatergic neurons mediate the motor sedative actions of benzodiazepines in mice. Slice cultures were prepared from mouse pups housed together with their mothers in a breeding facility. The same numbers of male and female pups were separated from their mothers and transferred to our laboratory. The animals were rapidly anaesthetized with isoflurane and decapitated. In accordance with local regulations of animal experimentation, extensive efforts were undertaken to minimize the time window the pups stayed without their mother. The brains were immediately isolated, cooled down in ice-cold dissection medium composed of Grey’s balanced salt solution (Sigma-Aldrich, Germany) with 60 mM glucose (50%, Sigma-Aldrich) and 11 mM MgCl₂ (AppliChem, Germany), and fixed in the slicing chamber of a vibratome (NVSLMI motorized advance vibroslice, World Precision Instruments, Germany). Up to four brains of both sexes were mounted in the same chamber in a row. The dissection medium was added into the chamber until the brains were fully submerged. Tissue slices (thickness, 300 μm) were cut from all mounted brains simultaneously. After being separated, both female and male slices freely floated in the dissection medium, and the sex of individual slices was lost. From these tissue slices, the cortical hemispheres were manually excised with a scalpel, and each was divided into two or three slices. The neocortical slices were placed on a glass coverslip and fixed with a coagulate of chicken plasma and thrombin. The coverslips were transferred into plastic tubes containing 750 μl nutrient medium consisting of horse serum (25%), Hanks’ balanced salt solution (25%), and basal medium Eagle (50%), supplemented with glutamine and glucose. The organotypic tissue cultures were maintained using the roller tube technique.²³  One day after the preparation, the medium was exchanged, and antimycotics (10 μM 5-fluoro-2-deoxyuridine, 10 μM cytosine-1β-d-arabinofuranoside, 10 μM uridine) and neuronal growth factor (10 nM) were added to reduce glial proliferation. After preparation and after each medium renewal (two times per week), the cultures were incubated 1 to 2 h in an atmosphere of 5% CO₂ under room air, attaining a pH of 7.2 to 7.4. All chemicals were obtained from Sigma-Aldrich (Germany) except the horse serum, which was obtained from Invitrogen (Germany). Electrophysiologic recordings were carried out between days 14 and 45 ex vivo.

Extracellular field excitatory postsynaptic potentials were recorded in the hippocampal CA1 stratum radiatum, evoked by stimulation in the Schaffer collateral commissural pathway of the same region using borosilicate glass micropipettes (Hugo Sachs Elektronik-Harvard Apparatus, Germany) with an open tip resistance of 1 to 2 MΩ and filled with artificial cerebrospinal fluid. Potentials were evoked by alternately delivering a test stimulus (50 μs, 5 to 20 V) through one of two bipolar tungsten electrodes (Hugo Sachs Elektronik-Harvard Apparatus, insulated to the tip; 50-μm tip diameter), placed at either side of the recording pipette, hereby stimulating nonoverlapping populations of the Schaffer collateral-associational commissural pathway. For baseline recordings, stimulation intensity was adjusted to values evoking a potential slope of approximately 25 to 30% of the maximum response. After at least 20 min of stable baseline recordings, long-term potentiation was induced by delivering a high-frequency stimulation train (100 pulses delivered at 100 Hz for 1 s) via one of the stimulating electrodes. The use of both stimulating electrodes allowed the measurement of an internal control in the same slice. After high-frequency stimulation was delivered from one of the electrodes in the absence of any substance, potentiation of the responses was monitored for at least 60 min after the tetanic stimulus, maintaining the same settings used for the baseline. Then, midazolam was applied in the bath solution for 60 min before long-term potentiation induction in the second input after high-frequency stimulation delivered via the second electrode. Inhibition of long-term potentiation was defined when the potential’s slope after high-frequency stimulation was less than 120% of the prestimulation slope. Control experiments corroborate that the extent of long-term potentiation was independent during the time that slices were in the recording chamber, at least for the duration in the current studies (up to 5 h). The field potentials were amplified (BA-2S, npi electronic, Germany), filtered (3 kHz), and digitized (9 kHz) using a laboratory interface board (ITC-16, Instrutech Corp., USA) and the WinLTP program software,²⁴  available at http://www.winltp.com/Ltp24/indexLtp24.htm. Stimuli were administered in an alternating manner to each input every 15 s. Two signals from the respective input were averaged to one for analysis, representing one data point each minute. Offline reanalysis of the data was performed with the same WinLTP software. The field excitatory postsynaptic potential slope was measured between 20 and 80% of the peak amplitude and then normalized according to the 20-min baseline recording before tetanic stimulation. The slope is commonly measured (instead of the amplitude) because it is a less contaminated signal and therefore more reliable.

The whole cell configuration of the patch clamp technique was used for analyzing midazolam’s effects on GABA_A receptor–mediated synaptic currents in cortical neurons. Cultured neocortical tissue slices showing signs of damage and degradation were excluded from electrophysiologic experiments. Intact neocortical slices were placed in a heated bath chamber (34°C) and continuously perfused with a modified artificial cerebrospinal fluid at a flow rate of 1 ml/min. The perfusion fluid consisted of 120 mM NaCl, 3.5 mM KCl, 1.1 mM NaH₂PO₄, 1 mM MgCl₂, 26 mM NaHCO₃, 1.2 mM CaCl₂, and 11 mM d-glucose and was bubbled with 95% O₂/5% CO₂ to adjust the pH to 7.4. Glass pipettes (2.5- to 4.5-MΩ tip resistance with recording solution) were pulled from borosilicate glass (World Precision Instruments, USA) with a puller (P-2000, Sutter Instrument Company, USA) and filled with high-chloride intracellular solution containing 121 mM CsCl, 24 mM CsOH, 1 mM MgCl₂, 5 mM EGTA, 10 mM HEPES, and 4 mM adenosine triphosphate. The recording electrode was positioned on a neuron with a micromanipulator, using infrared imaging. The cells were held at –70 mV in the whole cell voltage clamp configuration to record miniature synaptic events in the cultured neocortical slices. The N-methyl-d-aspartate and α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor antagonists d-2-amino-5-phosphonopentanoate (50 μM) and 6-cyano-7-nitroquinoxaline-2,3-dione (10 μM) or 2,3-dioxo-6-nitro-7-sulfamoyl-benzo[f]quinoxaline (5 μM; all Sigma-Aldrich) were added to the extracellular bath solution to block glutamatergic currents.

Quantification of the kinetic properties of synaptic currents requires events that do not overlap with others. This kind of event was rare when monitoring action potential–dependent and –independent GABAergic inhibitory postsynaptic currents together in cultured neocortical slices. Therefore, the sodium-channel blocker tetrodotoxin was added for blocking action potential dependent events. Currents were amplified with a MultiClamp 700A amplifier; sampled and digitized at a frequency of 20 kHz with a Digidata 1440A digitizer and Clampex 10.4 software (Molecular Devices, USA); and stored on a computer. Control recordings (duration, 180 s) were started after a 6-min wash-in phase with the blockers. Then the drug (50 or 500 nM midazolam) was added to the bath solution, and the drug recording (180 s) was started after 15 min. Offline processing of digitized data was performed using the MATLAB R2018b (The MathWorks Inc.‚ USA) program package. Digitized current recordings in abf format were imported, and synaptic events were detected by setting a threshold that was five times higher than the baseline noise (SD of the baseline current in the absence of events). Recordings with unstable baselines were excluded. Each experimental group contained recordings with cultures from at least two different preparations. For these experiments, the first n value represents the number of successful recordings, whereas the second one reports the number of animals.

In another set of experiments, we explored midazolam’s effect on GABAergic synaptic currents in acutely isolated hippocampal slices derived from wild-type animals. Because the frequency of spontaneously occurring synaptic currents was considerably smaller in this preparation than in cultured neocortical slices, action potential–dependent and –independent events were monitored together for evaluating midazolam’s effect on this type of neurons. Signals were amplified with the npi SEC-10L amplifier and digitized with HEKA LIH8 + 8 acquisition system and the PatchMaster v2x90.3 (HEKA Elektronik GmbH, Germany) software. For monitoring spontaneous synaptic events, control recordings were performed after assuring the health of the cell and then continued after 35 min of washing with midazolam 100 nM.

We opted for a longer drug exposure time in acutely isolated hippocampal slices than in cultured neocortical slices because the former were much thicker (acutely isolated hippocampal slices, 350 μm; in cultured neocortical slices, the initial thickness of about 300 μm is reduced to less than 30 μm after 2 weeks in culture), and the diffusion of drugs into the tissue is slower.²⁵  In all recording sessions, only a single concentration measurement was performed. After exposure to midazolam, the tissue slices were excluded from further experiments.

No explicit randomization or blinding methods were used for the assignment of individual animals to experimental conditions. All data values were normally distributed and are presented as the mean ± SD. For extracellular recordings, the n value is shown as x slices from y animals as, for example, n = 12/8, with the first number the number of slices and the second the total number of animals used in those conditions. The sample size was determined based on previous experience, and a maximum of two slices per animal were used, assuming that these slices are independent within animals. No statistical difference was found between male and female mice within genotype (Supplemental Digital Content 1, table S2, https://links.lww.com/ALN/C845). For all long-term potentiation experiments, comparisons of outcomes from three or more groups were performed using one-way repeated-measures ANOVA analysis with a 95% CI followed by Dunnett’s post hoc test to correct for multiple comparisons. We analyzed the long-term potentiation experiments containing two groups with the paired Student’s t test.

A standard recording interval of 3 min delivered several hundred miniature spontaneous events in cultured neocortical slices. These events were individually inspected and eliminated from the analysis if they overlapped with others. Events sampled in the course of a single recording that fulfilled the latter condition were fitted with a biexponential equation in the form I(t) = A_f exp(–t/τ_fast) + A_s exp(–t/τ_slow) + c, where I(t) is the current amplitude at any given time t, c is the baseline current, τ_fast and τ_slow are the fast and slow time constants of current decay, and A_f and A_s are the estimated fast and slow intercepts of the components at time zero,^26–28  respectively. Weighted time constants were calculated using the following equation: τ_w = A_f/(A_f + A_s) * τ_fast + A_s /(A_f + A_s) * τ_slow. Midazolam’s effects on this parameter were visualized by plotting histograms, which were well fitted with a double Gaussian function (fig. 5B; Supplemental Digital Content 1, fig. 1B, https://links.lww.com/ALN/C845): $val(x)=a1*e−(x-b1)c12+a2*e−(x-b2)c22$ with a1, a2 = max1, max2; b1, b2 = mean1, mean2; and c1 · √2, c2 · √2 = SD.

For further statistical analysis of the kinetic properties of synaptic events, we used the weighted decay time, the amplitude (A_f + A_s), and the charge transferred per miniature event. The Lilliefors test indicated that these parameters were normally distributed. For curve fitting and further statistical analysis, the program package MATLAB R218b was used. From the fits of individual synaptic currents captured in the same recording, the mean amplitude, the mean of the weighted decay time, and the mean transferred charge per event were calculated. These means provided an estimation of the kinetic properties of synaptic currents obtained from one individual experiment. Results from different experiments were used for comparing different genotypes and different midazolam concentrations in a final step (fig. 6).

In a subset of experiments, the kinetic properties of synaptic currents were estimated by using a different analytical approach in addition to the method described above, which served as an internal control. In this case, synaptic events that were nonoverlapping and detected in the course of a single recording were aligned by the point in time the current crossed the detection threshold for the first time (fig. 4). Aligned synaptic currents were stored in a two-dimensional matrix with the dimension time on the x-axis and the running number of detected events on the y-axis. Next, the median current was derived from various events and was determined for all points in time, providing an estimation of the time course of the median synaptic event. We calculated medians instead of means because statistical distributions were a priori unknown. In a final step, the time course of this median synaptic event was fitted with a double-exponential function as detailed above (fig. 5A). Results obtained with this alternative way of quantifying kinetic properties of synaptic events confirmed the results provided by our standard method.

Lilliefors tests indicated a normal distribution of kinetic parameters that were estimated from various experiments. For the group comparison of the genotypes (wild-type vs. α_2/3/5–knock-in) at 50 or 500 nM midazolam, as well as the comparison of the concentrations (50 vs. 500 nM midazolam), we performed independent sample t tests, whereas for the comparison control vs. drug, a paired samples t test (due to the paired design of the experiments) was sufficient.

Individual spontaneous inhibitory postsynaptic currents in acutely isolated hippocampal slices were analyzed with mini analysis software (Synaptosoft, USA). The detection threshold was set five times higher than the SD of the baseline current in the absence of events, and data on decay time and amplitude were collected. Curve fitting and further statistical analysis were conducted with GraphPad Prism 6.01 (GraphPad Software, USA) applying equations and statistical tests as described above. Differences were considered significant when the two-tailed P < 0.05 and are indicated by asterisks.

First, we determined the effect of several concentrations of midazolam on the induction of long-term potentiation in the CA1 hippocampal region (fig. 1). In the control group, field excitatory postsynaptic potentials were potentiated to 156 ± 19% (n = 20/12, 8 male and 4 female mice). After 60 min of midazolam (3 nM) exposure, long-term potentiation was induced by administering a tetanic stimulation resulting in a potentiation of 150 ± 25% (n = 6/3, all males) that was not significantly depressed compared to that of the control group. In contrast, midazolam applied at 10 nM (fig. 1B) and 1 μM, but not at 3 nM, resulted in a significant blockage of long-term potentiation for 10 nM at 98 ± 11% (n = 14/8, 4 male and 4 female mice) and for 1 μM at 94 ± 17% (n = 6/5, all males), indicating a dose-dependent effect (fig. 1C; Supplemental Digital Content 1, table S3A, https://links.lww.com/ALN/C845).

After demonstrating that midazolam at 10 nM was blocking long-term potentiation, we were interested to know whether this effect is mediated via the classical benzodiazepine site. Accordingly, we applied the specific benzodiazepine-site antagonist flumazenil²⁸  in the presence of midazolam to prevent this action. After the previous application of midazolam, flumazenil was able to significantly block inhibition of long-term potentiation by midazolam (midazolam, 113 ± 8%, n = 10/5, all males vs. midazolam + flumazenil, 145 ± 17%, n = 10/5, all males; fig. 2; Supplemental Digital Content 1, table S3B, https://links.lww.com/ALN/C845) and did not display any intrinsic effect when applied alone at 30 nM (data not shown).

The next set of experiments was designed to explore the role of the different GABA_A receptor subunits in the effect of midazolam by using several transgenic mouse models, where one or three α-subunits were carrying an H-to-R knock-in mutation. We investigated midazolam’s actions on long-term potentiation evoked in hippocampal brain slices in which α₁-, α₂-, α₅-, α_1/2/3-, α_1/3/5-, and α_2/3/5-subunit–containing receptors are insensitive to midazolam. Because of its low expression in the hippocampus,²⁹  we did not focus on α₃.

When high-frequency stimulation was administered in α₁–knock-in mice in the presence of 10 nM midazolam (fig. 3A; Supplemental Digital Content 1, table S3C, https://links.lww.com/ALN/C845), long-term potentiation was not significantly altered (136 ± 9% vs. control 141 ± 9%, n = 18/13, 8 male and 5 female mice), while a significant long-term potentiation blockage was observed for the α_2/3/5–knock-in (fig. 3B; Supplemental Digital Content 1, table S3D, https://links.lww.com/ALN/C845; 106 ± 12% vs. control 167 ± 27%, n = 12/6, 3 male and 3 female mice), α₅–knock-in (fig. 3C; Supplemental Digital Content 1, table S3E, https://links.lww.com/ALN/C845; 106 ± 17% vs. control 143 ± 12%, n = 20/15, 10 male and 5 female mice), and α_1/2/3–knock-in lines (fig. 3D; Supplemental Digital Content 1, table S3F, https://links.lww.com/ALN/C845; 101 ± 21% vs. control 148 ± 10%, n = 13/11, 9 male and 2 female mice). In contrast, long-term potentiation in slices of α_1/3/5–knock-in mice (fig. 3E; Supplemental Digital Content 1, table S3G, https://links.lww.com/ALN/C845; 141 ± 10% vs. control 139 ± 12%, n = 12/6, 6 male and 6 female mice) was not significantly altered in the presence of 10 nM midazolam. These results show that the effect of midazolam at small concentrations in wild-type mice is mainly mediated by the α₁-subunit (fig. 3, A and B). However, when this subunit is rendered insensitive to benzodiazepines by a knock-in point mutation, it is obvious that the combined action of midazolam on α₂- or α₅-subunits is unable to block long-term potentiation. Interestingly, long-term potentiation in the α_1/2/3–knock-in genotype (with an intact α₅) is midazolam-sensitive (fig. 3D).

In a second series of experiments, we investigated the effects of midazolam on GABAergic synaptic transmission, selectively mediated via α₁-GABA_A receptors. To this end, we recorded miniature synaptic currents in organotypic slice cultures prepared from the neocortex of wild-type and α_2/3/5–knock-in mice under drug-free conditions and in the presence of 50 or 500 nM midazolam. Spontaneous events recorded under drug-free conditions served as the control. Then midazolam (50 nM or 500 nM) was added, and spontaneous events were recorded under drug conditions. A representative experiment is displayed in figure 4, showing exemplary current traces (left) and the overlaid miniature synaptic events sampled from the same recording (right) under control conditions (fig. 4A) and after application of midazolam (fig. 4B) in slices of wild-type mice. In Supplemental Digital Content 1, figure S2 (https://links.lww.com/ALN/C845), the same kind of experiment using slices of α_2/3/5–knock-in mice is indicated. Figure 5 and Supplemental Digital Content 1, figure S1 (https://links.lww.com/ALN/C845) demonstrate the impact of midazolam on the kinetic properties of miniature synaptic events. Individual synaptic events monitored in the absence and presence of midazolam were fitted with biexponential functions. The median synaptic event obtained from synaptic events monitored in the presence of midazolam was normalized by the maximum current of the median synaptic event monitored under drug-free conditions. Midazolam clearly increased decay time in neocortical neurons of wild-type mice (fig. 5A) and α_2/3/5–knock-in mice (Supplemental Digital Content 1, fig. S1A [https://links.lww.com/ALN/C845]). Histograms in figure 5B and Supplemental Digital Content 1, figure S1B (https://links.lww.com/ALN/C845) display the weighted decay times of individual miniature events, sampled during the same recording interval. Distributions were best fitted by the sum of two Gaussian functions, suggesting the presence of at least two distinct populations of synaptic events before and during midazolam treatment.

Next, we analyzed midazolam effects on frequency, charge transfer, amplitude, and decay time of synaptic events and compared the effects of two different concentrations (50 and 500 nM) and genotypes (wild-type and α_2/3/5–knock-in mice; fig. 6; Supplemental Digital Content 1, fig. S3, https://links.lww.com/ALN/C845). We did not observe significantly differing effects on the frequency between genotypes and drug concentrations (data not shown). Midazolam clearly increased the charge transferred per synaptic event (fig. 6A; Supplemental Digital Content 1, table S6A1, https://links.lww.com/ALN/C845), and the effect was significant for both genotypes and drug concentrations. Midazolam (50 nM) increased charge transfer equally in wild-type and α_2/3/5–knock-in mice (fig. 6, A1 and A2): 50 nM versus control in wild-type: 172 ± 71% of control; 50 nM vs. control in α_2/3/5–knock-in: 171 ± 62% of control (Supplemental Digital Content 1, table S6, A2.1 and A2.2, https://links.lww.com/ALN/C845). Interestingly, increasing the concentration of midazolam further increased charge transfer in the wild-type but not in the α_2/3/5–knock-in cultures (fig. 6, A1 and A2; Supplemental Digital Content 1, table S6, A2.1 and A2.2, https://links.lww.com/ALN/C845): 500 nM midazolam in wild-type: 236 ± 54% of control, n = 6/6; 500 nM midazolam in α_2/3/5–knock-in: 175 ± 62% of control, n = 7/6. Since there is no difference in the effect size between wild-type and α_2/3/5–knock-in mice at 50 nM midazolam, the effect on charge transfer at this concentration seems to be largely mediated by α₁-GABA_A receptors. Additional experiments were added in response to peer review in which midazolam’s action on GABAergic currents was determined in the presence of the benzodiazepine site antagonist flumazenil. With flumazenil present, 500 nM midazolam did not cause a change in charge transfer that was statistically significant (107 ± 13% of control; n = 5/3; all wild-type).

The charge transfer per synaptic event is primarily determined by two components: amplitude and decay time of synaptic events. By analyzing the effect on these parameters, we can infer their relative contribution to the concentration-dependent effect on charge transfer in wild-type cultures.

Midazolam at concentrations of 50 and 500 nM significantly prolonged the decay of miniature synaptic events (fig. 6, B1 and B2) in wild-type and α_2/3/5–knock-in slice cultures (50 nM vs. control in wild-type: 172 ± 44% of control, n = 10/6; 500 nM vs. control in wild-type: 158 ± 21% of control, n = 7/6; 50 nM vs. control in α_2/3/5–knock-in: 153 ± 19% of control, n = 10/6; 500 nM vs. control in α_2/3/5–knock-in: 156 ± 37%, n = 7/6; Supplemental Digital Content 1, table S6, B1, B2.1, and B2.2, https://links.lww.com/ALN/C845). However, we did not find significant differences between the genotypes or a concentration dependency on the decay time of miniature synaptic events (fig. 6B2; Supplemental Digital Content 1, table S6B2.1, https://links.lww.com/ALN/C845).

The amplitude of synaptic events (Supplemental Digital Content 1, fig. S3A, https://links.lww.com/ALN/C845) was not significantly altered by 50 nM midazolam in wild-type and α_2/3/5–knock-in mice (Supplemental Digital Content 1, fig. S3, A and B, https://links.lww.com/ALN/C845; 50 nM vs. control in wild-type: 104 ± 27%, n = 10/6; 50 nM vs. control in α_2/3/5–knock-in: 120 ± 30%, n = 8/6). Midazolam (500 nM) did not change the amplitude of synaptic events in the α_2/3/5–knock-ins (123 ± 35%, n = 7/6; Supplemental Digital Content 1, fig. S3, A and B, https://links.lww.com/ALN/C845) but significantly increased the amplitude of synaptic events in the wild type (137 ± 36%, n = 7/6; Supplemental Digital Content 1, fig. S3, A and B, https://links.lww.com/ALN/C845). We did not observe a significant difference in the amplitude of synaptic events between the genotypes (Supplemental Digital Content 1, fig. 3B, https://links.lww.com/ALN/C845). Analogous to the effect on charge transfer, there was, however, a significant concentration dependency in the wild type (Supplemental Digital Content 1, fig. S3B, https://links.lww.com/ALN/C845). Therefore, the concentration-dependent increase of charge transferred per synaptic event observed in the wild type most likely results from an increase in the amplitude of synaptic events only occurring at high concentrations of midazolam. Since the effects of midazolam at low concentrations seem to be almost exclusively mediated via α₁-GABA_A receptors but did not further increase in the α_2/3/5–knock-ins at a higher concentration of midazolam, the additional effects of midazolam at higher concentrations observed in the wild type are possibly attained by modulating the amplitude of synaptic events via GABA_A receptor subpopulations without an α₁-subunit.

In CA1 hippocampal neurons from wild-type mice, spontaneous synaptic events followed the same pattern as in neocortical slices from wild-type animals. A significant increase in the decay time was seen after application of 100 nM midazolam (35 ± 1% vs. control 31 ± 3%, n = 10/6, all males; Supplemental Digital Content 1, fig. S4A, https://links.lww.com/ALN/C845). Similarly, the amplitude of the spontaneous synaptic events was significantly increased after application of 100 nM midazolam (Supplemental Digital Content 1, fig. S4B, https://links.lww.com/ALN/C845; 46 ± 15% vs. control 31 ± 12%, n = 10/6, all males).

Our results suggest that at amnestic concentrations (approximately 10 nM), midazolam blocks hippocampal long-term potentiation predominantly by potentiating α₁-GABA_A receptors. We hypothesize that midazolam targeting α₂-GABA_A receptors did not modulate long-term potentiation but dampened the activity of α₅-GABA_A receptors. At hypnotic concentrations (50 to 500 nM), midazolam’s enhancing effect on GABA_A receptor–mediated synaptic transmission was largely mediated by α₁-GABA_A receptors.

To elucidate midazolam’s actions on hippocampal neurons, we utilized mouse lines carrying knock-in mutations in the α-subunit of the GABA_A receptor, which cause a dramatic decrease in benzodiazepine binding.^9,18  The results summarized in figure 3 suggest that midazolam, which at low nanomolar concentrations causes moderate sedation and amnesia in humans, blocks hippocampal long-term potentiation predominantly via α₁-GABA_A receptors. This conclusion is based on two complementary findings. First, in slices derived from α_2/3/5–triple–knock-in mice, the drug was effective in blocking long-term potentiation. Second, midazolam failed to block long-term potentiation in slices derived from α₁–single–knock-in mice. Thus, modulation of α₁-receptors by midazolam seems to be sufficient to depress long-term potentiation. At first glance, these observations are surprising since previous studies showed that hippocampal-dependent long-term potentiation and learning are tightly controlled by α₅-receptors.^30,31  These receptors are densely expressed extrasynaptically on the dendrites of pyramidal cells, and their kinetic properties match functional properties of N-methyl-d-aspartate receptors.³²  In the current study, we also observed that in α_1/2/3–triple–knock-in mice, midazolam was effective in depressing long-term potentiation, and MRK-016 (Supplemental Digital Content 1, fig. S5, https://links.lww.com/ALN/C845), a selective, negative modulator of α₅-receptors, enhanced long-term potentiation-induction in wild-type mice. Thus, selective modulation of α₅-receptors by midazolam and MRK-016 is effective in altering long-term potentiation. However, why does midazolam fail to depress long-term potentiation if the drug is a positive allosteric modulator of α₅- and α₂-receptors at the same time? This condition applies to slices derived from α₁–single–knock-in mice, since α₁-receptors are resistant to the drug in this genotype.²⁹  We hypothesize that the pool of α₂-receptors exerts inhibitory control over α₅-receptors. A possible scenario is that GABAergic interneurons, innervating the dendrites of hippocampal pyramidal cells, express α₂-receptors. As a consequence, concomitant modulation of α₅- and α₂-receptors reduces α₅-receptor function by ultimately attenuating total GABA release onto pyramidal neurons (see fig. 7 for explanation). This hypothesis arises from comparing figure 3A (α₁–knock-in renders long-term potentiation midazolam-insensitive with α₂, α₃, and α₅ intact) with figure 3D (α_1/2/3–knock-in renders long-term potentiation midazolam-sensitive with only α₅ intact). It is important to note that hypothesizing that the activity of α₅-receptors is dampened by α₂-receptors is compatible with midazolam’s qualitative actions on all six genotypes investigated in the current study (fig. 7). Even though Crestani et al.²  reported reduced α₅-receptor expression in the hippocampus of α₅–knock-in mice, they clearly showed that baseline functions (e.g., long-term potentiation in α₅–knock-in) are unaffected by the knock-in mutation; this lets us conclude that these changes in α₅ receptor expression in knock-in mice may not interfere with our hypothesis. Of course, the idea of α₂-dependent α₅-receptor activity needs further corroboration, e.g., monitoring neuronal activity within hippocampal inhibitory microcircuits of α_1/3/5– and α₁–knock-in mice. In addition, the importance of α₁- and α₅-receptors involved in learning has been found also in rhesus monkeys.³³  Consistent with our hypothesis, in this study, α₁-receptors appear to be sufficient to induce learning impairments and thus to be dominant over α₅, as shown with the specific α₁ enhancers zolpidem and zaleplon.

Benzodiazepines induce multiple behavioral endpoints such as amnesia, anxiolysis, sedation, and hypnosis by acting on different sets of GABA_A receptor subtypes that display diverging expression patterns in the central nervous system.³⁴  The drug’s effective concentration is one crucial factor that defines these behavioral endpoints. Fortunately, in the case of midazolam, active metabolites are of minor importance,^22,35,36  and according to the calculation provided in the first supplementary table (Supplemental Digital Content 1, table S1, https://links.lww.com/ALN/C845) and legend, a free concentration of midazolam of approximately 10 to 23 nM can be estimated. This concentration range, which is assumed to cause amnesia and moderate sedation in human subjects, is in good accordance with our observation that midazolam was effective in depressing long-term potentiation in hippocampal slices when applied at a concentration of 10 nM.

In clinical anesthesia, midazolam is used because of its profound sedative and hypnotic properties. Plasma concentrations of midazolam causing hypnosis in human subjects are in the range of 300 to 400 ng/ml.^37,38  Persson et al.³⁷  reported that immediately after patients regain consciousness, this concentration is about 166 ng/ml. Based on these data, we assume that brain concentrations of midazolam producing deep sedation should range between 50 and greater than 150 nM (Supplemental Digital Content 1, table S1, https://links.lww.com/ALN/C845). For elucidating the potential contribution of α₁-GABA_A receptors, we quantified the effects of 50 and 500 nM midazolam on GABA_A receptor–mediated synaptic currents in neocortical neurons. For evaluating the specific role of α₁-GABA_A receptors, these studies were conducted in tissue slices derived from wild-type and α_2/3/5–triple–knock-in mice (only α₁-receptors are sensitive to midazolam). In the wild type, 50 and 500 nM midazolam increased the charge transfer of inhibitory postsynaptic currents to 175 and 240% of control values, respectively. It is interesting to relate these numbers to the effects of etomidate and propofol, two hypnotic drugs that act predominantly via GABA_A receptors and produce loss of consciousness at brain concentrations close to 1 μM.³⁹  In a previous study, we observed that these agents increased charge transfer by 210% (1 μM etomidate) and 220% (1 μM propofol),²⁶  which is well within the effects produced by 50 and 500 nM midazolam. As a rule of thumb, it seems that loss of consciousness is induced by GABAergic drugs at concentrations that double charge transfer of fast synaptic inhibition.

Zeller et al.⁸  showed that the sedative properties of benzodiazepines are mediated by α₁-containing GABA_A receptors expressed in forebrain glutamatergic neurons. These receptors constitute the most prominent GABA_A receptor subtype in the cerebral cortex. It has been estimated that about 60% of all GABA_A receptors harbor α₁-subunits.¹  Because of their frequent occurrence, α₁-GABA_A receptors are a prime candidate in mediating the amnestic, sedative, and hypnotic properties of midazolam. We found that clinically relevant concentrations of midazolam significantly enhanced GABAergic synaptic currents in neocortical slices derived from α_2/3/5–triple–knock-in mice, in which the drug exclusively modulates α₁-GABA_A receptors. At a concentration of 50 nM, midazolam enhanced the charge transferred per average synaptic event in slices from wild-type and α_2/3/5–knock-in mice by about the same amount (fig. 6A1), suggesting that midazolam’s action is predominantly mediated by α₁-receptors at this concentration. However, increasing the drug’s concentration from 50 to 500 nM further enhanced the charge transferred per average synaptic event in tissue slices derived from wild-type mice but not in tissue slices derived from α_2/3/5–knock-in mice. This result suggests that 50 nM is a saturating concentration on α₁-GABA_A receptors but not on non–α₁-GABA_A receptors. It is important to note that this conclusion is based on several assumptions, including similar expression levels of GABA_A receptor subunits in wild-type and knock-in animals, similar expression patterns in different types of neurons, and a selective molecular action of midazolam on classical benzodiazepine-sensitive receptors. Nevertheless, these observations compare well to previous findings. At a concentration of 5 nM, midazolam significantly attenuated high-frequency action potential firing in neocortical slices derived from wild-type, but not in slices prepared from α₁–single–knock-in mice (in this genotype, the drug acts via α₂-, α₃-, and α₅-receptors), confirming our previous studies, showing that high-frequency firing is under the control of α₁-receptors.^19,22,40  However, after increasing the concentration of midazolam from 5 to 100 nM, the drug significantly reduced high-frequency firing also in slices derived from α₁–single–knock-in mice, indicating the involvement of non–α₁-receptors at higher (greater than or equal to 100 nM) midazolam concentrations.

Taken together, our findings suggest that an already sedative, low nanomolar concentration of midazolam blocks long-term potentiation via α₁-GABA_A receptors. The drug’s high potency and efficacy in blocking long-term potentiation prompts the question of how long these effects persist in patients after the treatment. Two studies investigated midazolam’s hangover effect about 12 h after subjects received an oral dose of 15 mg.^41,42  The authors report that at a free plasma concentration of about 20 nM, corresponding to a free concentration on brain GABA_A receptors of about 3 nM, midazolam caused low performance in visual memory and in telephone testing tasks. Interestingly, impaired memory was midazolam’s only residual effect 10 h after drug intake. If midazolam is used as premedication and for inducing anesthesia and if a plasma half-time of about 2 h is assumed,⁴³  it seems possible if not likely that the drug is still effective in enhancing the function of α₁-receptors and decreasing cognitive performance, even more than 12 h after terminating midazolam treatment. Moreover, modulation of α₁-receptors may contribute to the occasionally reported phenomenon of transient global postoperative amnesia after midazolam administration.^44–46  The possible involvement of α₁-receptors is consistent with the observation that this complication can be antagonized by flumazenil.

Supported by grant No. AN 321/5-1, RA 689/14-1 from the Deutsche Forschungsgemeinschaft (Bonn, Germany).

Dr. Zeilhofer has a financial relationship with Engrail Therapeutics (San Diego‚ California), Eliem Therapeutics (Redmond‚ Washington), and the Swiss National Science Foundation (Bern, Switzerland). Dr. Rudolph has a financial relationship with Elsevier (New York, New York), Concert Pharmaceuticals, Inc. (Lexington, Massachusetts), and the National Institutes of Health (Bethesda, Maryland). The other authors declare no competing interests.