Introduction

The GABAA receptor chloride channel is the predominant inhibitory neurotransmitter receptor in the mammalian central nervous system and has been widely used as a target for neuromodulatory drugs. Many compounds in clinical use as anxiolytics, sedatives, hypnotics, or antiepileptics increase GABAA receptor activation via the allosteric benzodiazepine (BZD) binding site. Such compounds have been termed “BZD receptor agonists,” whereas BZD-binding site ligands producing the opposite effect, i.e., decreasing receptor activation, are called “inverse agonists.” BZD “antagonists” are ligands which bind to the receptor without modulating its function (Haefely 1989). BZD receptor inverse agonists have so far only been tested in animal behavior experiments and in a very few exploratory human studies. The results showed beneficial activity against impaired cognition; however, further clinical development of these compounds was prevented by anxiogenic or proconvulsive side effects (Dorow et al. 1983; Duka et al. 1996; Jensen et al. 1987; Little et al. 1984; McNamara and Skelton 1993; Petersen et al. 1983; Venault et al. 1986).

There is a large unmet medical need for treatment of cognitive deficits in aging western societies with millions of patients suffering from Alzheimer’s disease and other types of dementias. Currently available therapies are based either on cholinesterase inhibition (e.g., donepezil) or on NMDA receptor antagonism (memantine). However, the effectiveness of current therapies is small, and cholinesterase inhibitors suffer from mechanism-related side effects (Birks and Harvey 2006; McShane et al. 2006). There is a clear need for a novel therapy with improved efficacy and better tolerability. Therefore, inhibition of GABAA receptor function remains an attractive alternative, provided compounds can be made selective for the receptor subtype mainly involved in memory formation (Maubach 2003).

GABAA receptors are pentamers mostly consisting of two α, two β, and a γ subunit. Several gene products are available for each of the subunits giving rise to a large number of receptor variants. The importance of different α subunit subtypes has been elucidated by the generation of transgenic mice lacking the normal diazepam sensitivity of the α1, α2, α3, or α5 subunit (α4 and α6 are diazepam insensitive). The results suggest that α1 is responsible for the sedative effects and α2 and perhaps α3 for the anxiolytic effects of agonistic BZD receptor ligands (Löw et al. 2000; McKernan et al. 2000; Möhler 2006; Rudolph et al. 1999). The consequences of a modified pharmacology of the α5 subunit are less evident, but reduced expression of the subunit could be associated with facilitated cognition in hippocampal-dependent tasks (Collinson et al. 2006; Crestani et al. 2002). This is in line with the preferential localization of α5 subunits in the hippocampus (Fritschy and Möhler 1995; Pirker et al. 2000). It was therefore hypothesized that a BZD site ligand with inverse agonism selective for α5 containing GABAA receptors should enhance cognitive function without anxiogenic and proconvulsant side effects.

Selectivity of a BZD site ligand can be achieved by different affinities to GABAA receptor subtypes (“binding selectivity”). Alternatively, in the case of similar subtype affinities, different degrees of receptor modulation (“functional selectivity”) can be attempted, i.e., inverse agonism at α5, but (ant)agonism at other subtypes. A compound may also have a combination of both binding and functional selectivity although, so far, this is rare. A number of compounds with binding and/or functional selectivity for α5 containing GABAA receptors have recently been synthesized and tested on animal behavior (Liu et al. 1996; Quirk et al. 1996; Savic et al. 2008; Sternfeld et al. 2004). Most results confirm the hypothesis that compounds with such a pharmacological profile can improve cognitive function without CNS-mediated adverse effects (Atack et al. 2006b; Collinson et al. 2006; Dawson et al. 2006). However, an optimal combination of potency, efficacy, binding and functional selectivity, oral bioavailability and brain penetration has not yet been achieved.

In this paper, we will describe the properties of RO4938581, an imidazo-triazolo-benzodiazepine combining both binding and functional selectivity at the α5 containing GABAA receptor and will show that the compound improves cognition in several paradigms of learning and memory in rats and monkeys, with no anxiogenic or proconvulsant effects.

Material and methods

Plasmids and recombinant cell expression

cDNAs encoding the rat GABAA α1, α2, α3. and α5 receptor subunits were subcloned into the pIRESpuro vector, and the cDNAs encoding the rat GABAA β2 and β3 and γ2 (short) receptor subunits were subcloned into the pIRESneo and pIREShygro vector, respectively (Clontech, Mountain View, CA, USA). Stable cell lines expressing rat GABAA α1β2γ2, α2β3γ2, α3β3γ2, and α5β3γ2 receptor subtypes were derived by transfection of pIRES plasmids containing the desired subunit cDNAs (α, β, γ, 1:1:2 ratio) into HEK 293 cells, using the lipofectamine 2000 kit according to the manufacturer’s instructions (Invitrogen, Carlsbad, CA, USA). The cells were grown in minimal essential medium (Invitrogen) supplemented with 10% fetal calf serum (Invitrogen), 20 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES; Invitrogen), and 100 U/ml penicillin/100 μg/ml streptomycin (Invitrogen). The culture medium for the stable cell lines contained, in addition, 0.3 μg/ml puromycin (Clontech, Mountain View, CA, USA), 300 μg/ml hygromycin B (Roche Diagnostics, Mannheim, Germany), and 200 μg/ml G418 (Invitrogen). For electrophysiological experiments, the cells were plated onto glass coverslips coated with poly-d-lysin (Sigma-Aldrich, St. Louis, MO, USA).

Membrane preparation and [3H] flumazenil-binding assays

Cells expressing the different GABAA receptor subtypes were harvested and washed three times with cold phosphate-buffered saline and frozen at −80°C. The pellet was suspended in cold Krebs–Tris buffer pH 7.5 containing (in mM) Tris 15, NaCl 120, KCl 100, CaCl2 25, and MgCl2 25 and homogenized with a Polytron homogenizer (IKA Labortechnik AG, Staufen, Germany) for 20 s at 10,000 rpm. After centrifugation at 50,000×g for 60 min at 4°C, the pellet was resuspended in cold Krebs–Tris buffer pH 7.5, and the protein content was measured using the Bradford method (Bio-Rad Laboratories, Reinach, Switzerland) with bovine serum albumin as the standard. The membrane homogenate was frozen at −80°C before use. The inhibition of 1 nM [3H]flumazenil ([3H]RO0151788) binding by RO4938581 was measured in rat GABAA α1β2γ2, α2β3γ2, α3β3γ2, and α5β3γ2 receptor subtypes. From the IC50 the K i was calculated using the Cheng-Prusoff equation (K i = IC50/(1 + [radioligand]/K d); Cheng and Prusoff 1973) with respective K d values for [3H]flumazenil binding of 0.9, 3.9, 1.9, and 0.6 nM at the α1-, α2-, α3-, or α5-containing subtype, respectively. Nonspecific binding was determined in the presence of 10 μM diazepam. The percentage inhibition of [3H]flumazenil binding, the IC50 and the K i values were calculated using ActivityBase (IDBS; Guildford, Surrey, UK).

Whole cell patch-clamp of HEK293 cells

Experiments were performed with HEK293 cells expressing the α1β2γ2, α2β3γ2, α3β3γ2, and α5β3γ2 GABAA receptor subtypes. Cells plated on glass coverslips were transferred to a Perspex chamber on the stage of Nikon Diaphot inverted microscope and continuously perfused with a solution containing (in mM) NaCl 150, KCl 4, CaCl2 1.2, MgCl2 1, HEPES 10, pH 7.2 adjusted to 7.4 with NaOH and osmolarity adjusted to 340 mOsm with sucrose. Patch pipettes were pulled with an approximate tip diameter of 2 μm and a resistance of 2 MΩ with borosilicate glass (Clark Electromedical Instruments, Reading, UK) and filled with a solution containing (in mM) CsCl 140, HEPES 10, ethylene glycol tetraacetic acid 11, MgCl2 1, CaCl2 1, and Mg-ATP 4, pH adjusted to 7.2 with CsOH and osmolarity adjusted to 340 mOsm with sucrose. Cells were recorded in the patch-clamp whole cell mode using an Multiclamp 700B amplifier (Molecular Devices, Sunnyvale, CA, USA). GABA in the presence or absence of RO4938581 was applied to the cell for 2 s in 1 min intervals using a multibarrelled microapplicator pipette controlled by a stepping motor (Bio-Logic SAS, Claix, France). For each experiment, at least three GABA control applications were generated and only cells showing stable GABA responses were selected for the compound testing. Prior to application of a GABA–compound mixture, the same concentration of the compound alone was applied by bath perfusion for 1 min. During the experiment, all bath solutions contained 0.1% DMSO, which by itself was without detectable effect on the GABA responses. For data analysis, the pClamp data acquisition program set (Molecular Devices) and Origin software package (OriginLab Corporation, Northampton, MA, USA) were used. For each concentration of RO4938581, maximum current amplitudes were measured and plotted as a function of the concentration. Current amplitudes were fitted with the nonlinear least-squares fitting routine of Origin using the equation:

\(I\left( x \right){\text{ = }}{{I_{{\text{max}}} } \mathord{\left/ {\vphantom {{I_{{\text{max}}} } {\left[ {{\text{1 + }}\left( {{x \mathord{\left/ {\vphantom {x {{\text{IC}}_{{\text{50}}} }}} \right. \kern-\nulldelimiterspace} {{\text{IC}}_{{\text{50}}} }}} \right)^n } \right]}}} \right. \kern-\nulldelimiterspace} {\left[ {{\text{1 + }}\left( {{x \mathord{\left/ {\vphantom {x {{\text{IC}}_{{\text{50}}} }}} \right. \kern-\nulldelimiterspace} {{\text{IC}}_{{\text{50}}} }}} \right)^n } \right]}}\), where I is the measured current, I max the maximum current amplitude, x the concentration of RO4938581, IC50 is the half-maximum effective concentration, and n the Hill slope.

Subjects

Mice

Male and female DBA2/J mice (Charles River, France) aged from 19–21 days at time of testing, were used in the audiogenic seizures test. Three- to four-month-old C57Bl/6J mice (RCC, Itingen, Switzerland) were used in the LTP experiment.

Rats

Wistar rats (RCC), 36- to 42-day-old, were used in the population spike recording experiments. Male Sprague–Dawley rats (Charles River, France) with approximate weight of 200 g were used in the following experiments: spontaneous locomotor activity, neurological assessment, elevated plus maze, stimulus rats in the social approach avoidance (SAA) test (with approximate weight of 450 g). Male Fischer rats (F-344, Charles River, Germany) with approximate weight of 200 g were used in the SAA test. Male Lister hooded rats (Harlan, Netherlands) with approximate weight of 450 g (at time of drug testing) were used in the following tests: chain pull responding, delayed match to position (DMTP) task and, with approximate weight of 300 g, in the Morris water maze and in the in vivo binding experiments. All rodents were group-housed in holding rooms at controlled temperature (20–22°C), humidity (55–65%), and 12-h light/dark cycle (lights on at 06:00 hours). All rodents were allowed ad libitum access to food and water except those used in the operant conditioning tests, where food was limited to that earned in the test session and 12–15 g/rat at the end of the day. The experimental procedures used in the present investigation received prior approval from the City of Basel Cantonal Animal Protection Committee based on adherence to federal and local regulations.

Monkeys

Adult (5–12 years; 6–9 kg) male cynomolgus macaques (Macaca fascicularis) were housed in same-sex pairs in a colony room maintained at 21 ± 2°C, 40 ± 10% humidity, and on a normal 12-h light/dark cycle (lights on at 07:00 hours). Food (Purina High Protein no. 5045) and water were available ad libitum. All experimental procedures were approved by the Institutional Animal Care and Use Committee of Roche Palo Alto and were in accordance with National Institutes of Health guidelines.

Population spike recordings in rat hippocampal slices

Rats were anesthetized in a 2.5% isoflurane/96.5% oxygen mixture and sacrificed. The hippocampi were dissected and 400-μm slices were transversely cut with a tissue chopper (Sorvall, Newton, USA). The slices were then maintained in a submerged chamber and perfused at room temperature in a solution containing (in mM): NaCl 124, KCl 5, MgSO4 2, CaCl2 2, KH2PO4 1.25, NaHCO3 25, d-glucose 11 oxygenated with 95% O2/5% CO2 (pH 7.4, 307 mOsm). Population spikes (PS) were recorded from the CA1 region of the hippocampus similarly as described by Kemp et al. (1987). A glass micropipette (1–3 MΩ) containing 2 M NaCl was positioned in the stratum pyramidale and insulated bipolar platin/iridium electrodes were positioned in the Schaffer collaterals for orthodromic stimulation (0.1 ms, 60–90 μA). The stimulus intensity that induced a PS near the maximum amplitude was used as the test stimulus. PS were amplified with an Cyberamp 380 amplifier (Molecular Devices), filtered at 2.4 KHz and digitized at 20 KHz with a Digidata 1322 acquisition board (Molecular Devices) for subsequent storage on a personal computer (Compaq Deskpro EN).

To obtain concentration–response relationships, PS were evoked at 30-s intervals and increasing concentrations of isoguvacine were applied to the slice until a maximum inhibition of the PS was reached. For each concentration of isoguvacine, the PS amplitude obtained from four averaged responses was measured. Isoguvacine was then removed from the salt solution. After a full recovery of the PS, RO4938581 was preapplied to the slice for 15 min, and then another isoguvacine concentration response curve was generated in the presence of the compound. PS amplitudes were fitted with the nonlinear least-squares fitting routine of Origin using the equation:

\({\text{PS}}\left( x \right){\text{ = }}{{{\text{PS}}_{{\text{max}}} } \mathord{\left/ {\vphantom {{{\text{PS}}_{{\text{max}}} } {\left[ {{\text{1 + }}\left( {{x \mathord{\left/ {\vphantom {x {{\text{IC}}_{{\text{50}}} }}} \right. \kern-\nulldelimiterspace} {{\text{IC}}_{{\text{50}}} }}} \right)^n } \right]}}} \right. \kern-\nulldelimiterspace} {\left[ {{\text{1 + }}\left( {{x \mathord{\left/ {\vphantom {x {{\text{IC}}_{{\text{50}}} }}} \right. \kern-\nulldelimiterspace} {{\text{IC}}_{{\text{50}}} }}} \right)^n } \right]}}\), where PS is the measured PS amplitude, PSmax the maximum PS amplitude, x the concentration of isoguvacine, IC50 the half-maximum effective concentration, and n the Hill slope. The data are expressed as mean ± SEM. IC50s were analyzed using a two-tailed, paired t test.

Long-term potentiation

Mice were anesthetized in a 2.5% isoflurane/96.5% oxygen mixture and decapitated. The hippocampi were dissected and 400-μm slices were cut with a Sorvall tissue chopper. Slices were maintained in an interface chamber and perfused at 35°C with a simple salt solution containing (in mM): NaCl 124, KCl 2.5, MgSO4 2, CaCl2 2.5, KH2PO4 1.25, NaHCO3 26, d-glucose 10, sucrose 4, gassed with 95% O2, and 5% CO2 (pH 7.4, 307 mOsm). Field excitatory postsynaptic potentials (fEPSPs) were recorded from the CA1 stratum radiatum with a glass micropipette (1–3 MΩ) containing 2 M NaCl and evoked by stimulation of the Schaffer collaterals with insulated bipolar platin/iridium electrodes >500 μm away from the recording electrode. The stimulus strength was adjusted to evoke fEPSPs equal to 30% of the relative maximum amplitude without superimposed population spike. After stable baseline recordings, long-term potentiation (LTP) was induced by a brief priming stimulus (ten stimuli at 100 Hz) followed by a burst stimulus (four times ten stimuli at 100 Hz at an interval of 20 s). The duration of the stimulation pulses was doubled during the tetanus. This stimulus paradigm has been found to be sensitive to allosteric modulators of GABA receptors (Seabrook et al. 1997). For the analysis, fEPSP slopes were expressed as a percentage of the baseline values recorded 10 min before the priming stimulation. Results from several slices were expressed as mean ± SEM.

Compounds and administration

RO4938581, RO0194603, [3H]-RO0154513, [3H]-RO0151788, L-655,708, and diazepam were synthesized at F.Hoffmann-La Roche for in vivo administration, RO4938581 was prepared in 0.3% Tween 80 v/v 0.9% saline (rat), or in 7.5% gelatine v/v 0.2% saline (mouse), or in HEC (hydroxy-ethyl-cellulose 250 HHX) (monkey). RO4938581 was administered per os (p.o.) with a fixed pretreatment time of 1 h (rat), 2 h (mouse), and 6.7 h (monkey). The pretreatment times were selected from the appropriate pharmacokinetics data as the time corresponding to maximal drug plasma concentration. Scopolamine hydrobromide (Sigma) was prepared in 0.9% saline and administered s.c. 30 min prior to testing. Diazepam was prepared in 0.3% Tween 80 v/v 0.9% saline and was administered i.p. 30 min prior to testing. RO0194603 was prepared in 0.3% Tween 80 v/v saline and administered either p.o. with a pretreatment time of 30 min (rat) or i.p. with a pre-treatment time of 15 min (mouse). All doses of compounds are expressed as that of the base.

Determination of receptor occupancy

Lister hooded rats were pretreated with vehicle or RO4938581 at dose 1, 3, and 10 mg/kg, p.o. (three rats per dose group), and 50 min later, received intravenously 0.1 mCi/kg of [3H]-RO0154513 (specific activity 53 Ci/mmol). To determine the amount of nonspecific-binding given by the radioligand, an additional group of rats was treated with the selective GABAA α5 ligand L-655,708 (Atack et al. 2006b) at dose 10 mg/kg, i.p. 30 min before administration of the radioligand. Preliminary experiments showed that L-655,708 at dose 10 mg/kg produced maximal occupancy of GABAA α5 receptors in the rat hippocampus. All rats were sacrificed 10 min after administration of the radioligand; brains were rapidly removed and frozen in dry ice. Sagittal sections (10 μm thick) were cut in a cryostat, mounted on Histobond glass slides (Marienfeld Laboratories Glassware, Germany), dried at room temperature and exposed, together with tritium microscales, to tritium-sensitive imaging plates (BAS-TR2025) for 5 days. The imaging plates were scanned in a Fujifilm BAS-5000 high-resolution phosphor imager, and the amount of [3H]-RO0154513 bound to the hippocampus was quantified with an MCID M2 image analysis system (Imaging Research, St. Catherines, Ontario, Canada). To obtain the specific binding (SB), radioactivity in the hippocampus of L-655,708-treated animals (i.e., nonspecific binding) was subtracted from the radioactivity in the hippocampus of vehicle- and RO4938581-treated animals. The extent to which specific binding is reduced relative to vehicle-treated animals reflects the degree to which RO4938581 occupies the benzodiazepine site of the GABAA α5 receptor population. Receptor occupancy was calculated according to the following equation:

$${\text{\% }}\;{\text{receptor}}\;{\text{occupancy}}\;{\text{ = }}\;{\left( {{{\text{1 - SB}}\;{\text{RO4938581SB}}} \mathord{\left/ {\vphantom {{{\text{1 - SB}}\;{\text{RO4938581SB}}} {{\text{vehicle}}}}} \right. \kern-\nulldelimiterspace} {{\text{vehicle}}}} \right)}\; \times \;{\text{100}}$$

where “SB RO4938581” is the specific binding measured in the hippocampus of rats pretreated with RO4938581, and “SB vehicle” is the specific binding measured in the hippocampus of rats pretreated with the vehicle.

Spontaneous locomotor activity

Sprague–Dawley rats were placed into a novel test chamber which consisted of a Plexiglas® box (41 cm × 41 cm × 30.5 cm) with sawdust bedding on the floor. Movement of the animal resulted in interruption of an array of photobeams from horizontally located infrared sources placed around the test chamber (Omnitech Electronics, Dartmouth, Canada). The effect of RO4938581 (0.1, 1, 10 mg/kg; n = 8/group) on total horizontal activity was recorded over a 1-h period.

Neurological assessment

Sprague–Dawley rats were observed over a 5- to 10-min period and were assessed for body posture and gait, righting reflex, grasping reflex, grip strength, rotarod performance (at 8 and 16 rpm) and observed for any other symptoms (see Higgins et al. 2001 for further details). The effect of RO4938581 (3, 10, 30 mg/kg, n = 6/group) was compared to vehicle.

Operant conditioning tests

The operant conditioning boxes were obtained from Med Associates (St. Albans, VT, USA), and the protocols were run by the Kestrel Control system from Conclusive Marketing (Harlow, UK) operating on an IBM compatible PC. For both tests, a Latin-squares design was used to assess drug effects in rats that had reached an asymptotic level of performance. Rats were tested twice weekly with at least a 2-day interval between test sessions. Rats were trained between test days to maintain baseline performance.

Chain pull for food reward under VI20 schedule

Food-deprived Lister hooded rats were previously trained to pull a ceiling chain for food reward (45 mg Formula P Noyes pellet). Schedule requirements were gradually increased to a final value of VI20 (variable interval of 20 s), and animals were ready for drug testing once stable responding had been attained. The effect of RO4938581 (1, 3, 10, 30 mg/kg, n = 12) was assessed on the total number of chain pulls and number of food rewards obtained during the 30-min session.

Delayed match to position task

The training procedure and experimental conditions were identical to that described by Higgins et al. (2004). In this task, food-deprived Lister Hooded rats were previously trained to press a lever during the sample stage and then they have to remember this response during the choice stage. During the choice stage, two levers are presented, and subjects have to press the same lever as the sample (i.e., matching) to receive a food reward. There is a variable delay interval (1–24 s) between the sample and choice stage, so that at short delays (1–4 s), the choice accuracy is high, and as the delay interval increases (>8 s), the animals become progressively worse because of the increased memory load. The parameters measured included percent correct [total no. correct/(total no. correct + incorrect) × 100] both for the total session and for each delay interval. Omissions and latency to press the sample lever, choice lever, and collect the food reward (magazine latency) were also measured to give an assessment of drug effects on overall performance. The effect of RO4938581 was assessed alone (0.1, 1, and 10 mg/kg; n = 12) and versus scopolamine (0.03 mg/kg)-induced impairment (0.1, 0.3, and 1.0 mg/kg p.o.; n = 11).

Morris water maze

The water maze is a large pool (2 m diameter and 0.7 m high) filled with water made opaque using a white artificial opacifier (E308, Bronx Chemicals, West Yorkshire, UK) and surrounded by extramaze cues. The water temperature (21°C) was constant throughout the experiment. The maze was arbitrarily divided into four quadrants: NE, NW, SE, SW; and a colorless perspex circular platform (d = 10 cm) was positioned at the center of one of these quadrants, 1–2 cm below the water surface. A computer tracking system (HVS Image Ltd., UK), was used to analyze each rat’s swim path on-line. Each rat started at a sequential position on each trial, and the maximum length of each trial was 60 s. Platform positions were assigned in a semirandom order, ensuring that each position was used an equivalent number of times within each treatment group. If a rat found the platform during the trial, it was left on the platform for 10 s. If the rat did not find the platform by the end of the trial, it was guided toward it, allowed to climb onto the platform, and left there for 10 s. There was an intertrial interval (ITI) of 10 min between trials, during which the animals were returned to their home cage. Rats were initially pretrained to find a hidden platform position (two trials per day for 5 days). Subjects were semirandomly divided into treatment groups, based on their performance on the last training day.

Test day

Subjects had to learn to locate a hidden platform placed in a new position in the maze. The test session consisted of six acquisition trials and a probe trial (platform removed) to assess the extent of spatial learning. The effect of RO4938581 (1, 3, 10 mg/kg) versus a diazepam (6 mg/kg)-induced impairment was assessed (n = 9–10/group). The mean latency, path length, and swim speed were assessed over the six trials. In the probe trial, the percent time spent searching for the previous platform in each quadrant (left, platform, right, and opposite), and the number of platform crossings were measured. Due to the duration of the protocol, testing was carried out over 2 days (half of each dose group was tested on each day).

Object retrieval task in adult monkey

This task requires a monkey to reach into a clear acrylic box (dimensions: 5 × 5 × 5 cm) with one open side, to retrieve a food treat (cubes of apple or pear, 1–2 cm3). The box was positioned in front of the monkey and outside of the home cage, with the open side facing left, right, or toward the monkey. Food treats were placed on the outer edge, inner edge, or deep within the box. A test session consisted of 17 trials with nine “easy” food retrievals (i.e., placement of the food reward on the inner or outer edges of the box or when the opening is toward the monkey) and eight “difficult” food retrievals (i.e., placement of the food reward deep within the box and the open side facing left or right). The order of presentation never varied, there were no contingencies for incorrect reaches (i.e., monkeys typically acquired the treat after the incorrect reach) or dropped treats, and trials were terminated if there were no reaches within 3 min. For more details, see Rutten et al. (2008). Administration of RO4938581 (0.1, 0.3, 1, 3, and 10 mg/kg, n = 12) and behavioral measurement were completed blind. Compound doses were administered in a pseudorandom order.

Anxiety tests: social approach avoidance and elevated plus maze

A tracking system (Ethovision, Noldus, The Netherlands) was used for both test procedures.

SAA

The apparatus consists of a black plastic test box divided into two separate compartments, social and nonsocial, connected by a sliding door. The social compartment contained a subchamber delimited by a perforated transparent wall in which the stimulus (Sprague–Dawley) rat was confined. The test started by introducing a test rat (Fischer) into the nonsocial compartment for a 3-min habituation period. After 3 min, the sliding door was opened, allowing the rat to move freely between the two compartments for 10 min (see Nicolas and Prinssen 2006, for further details). The effect of RO4938581 (0.3, 1, 3 mg/kg, n = 10/group) on time spent in the hidden zone (within the nonsocial area) was assessed. The positive control, RO0194603, was administered at 3 mg/kg in a separate experiment (n = 7-8/group).

EPM

Sprague–Dawley rats were placed on the maze for a 5-min period (see Martin et al. 2002 for further details). Light intensity was set at lowest possible (i.e., 60 lx in the middle of the maze) in order to increase sensitivity to detect anxiogenic-like activity. The effect of RO4938581 (0.3, 1, and 3 mg/kg; n = 10/group) on time spent on the open arms was assessed. The positive control, RO0194603, was administered at 3 mg/kg in a separate experiment (n = 12/group).

Audiogenic seizures

The apparatus for acoustic stimulation consisted of a sound isolated wooden chamber (inside dimensions, 45 × 45 × 45 cm), with a Plexiglas three-layered insulated front door permitting full-view observation. Uniform sound distribution in the chamber was ensured by four high-power tweeters HF50 from Celestion Co (Ipswich, UK) built into the ceiling of the chamber. Two pairs of mice were placed in separate rectangular Plexiglas transparent boxes (21 × 44 × 21 cm) with sawdust bedding where they could simultaneously be observed. Testing involved exposure to a 14-kHz sinusoidal tone at 70 dB measured 2 cm above the floor for 1 min. Sound stimulation began immediately after transfer of the mice to the test chamber. After vehicle treatment, such acoustic stimulation typically induced tonic convulsions in 25% of mice. The proportion of mice that exhibited tonic convulsions (n = 8/group) following treatment with RO4938581 (3, 10, 30, 100 mg/kg) or the positive control RO0194603 (0.03 mg/kg) were recorded.

Data analysis

Horizontal activity across time bins was analyzed by a repeated measures analysis of variance (ANOVA), followed by a one-factor ANOVA and post hoc Newman–Keuls test at each time bin. Total horizontal activity in 1 h and grip strength were analyzed by a one-factor ANOVA. Nonparametric data (horizontal wire test, rotarod) were analyzed by Kruskal–Wallis ANOVA. Chain pull and DMTP data were analyzed by a one-factor repeated-measures ANOVA, except for choice accuracy across delays which was analyzed using a two-factor repeated measures ANOVA. In all significant cases, this was followed by a post hoc Newman–Keuls test. All water maze data were analyzed with a one-factor ANOVA, except for time spent in each quadrant for each treatment group which was analyzed with a repeated measures ANOVA. All significant cases were followed by a post hoc Newman Keuls test. Monkey object retrieval data were analyzed with a one-factor repeated-measures ANOVA, followed in significant cases by a post hoc Dunnett analysis. SAA and elevated plus maze data were analyzed with a one-factor ANOVA, except for the positive control which was analyzed separately with an unpaired t test. Number of mice per group exhibiting tonic convulsions compared to the vehicle group were analyzed by a Chi square test. In all studies, the accepted level of significance was p < 0.05.

Results

Binding affinity

The structure of RO4938581 (3-bromo-10-difluoromethyl-9H-imidazo[1,5-a][1,2,4]triazolo[1,5-d][1,4]benzodiazepine) is shown in Fig. 1. Inhibition of [3H]flumazenil binding showed that the compound binds with higher affinity (4.6 nM) to the BZD binding site of the rat GABAA α5β3γ2 receptor subtype than to receptors containing α1, α2, or α3 subunits in conjunction with β3 and γ2 subunits (80–185 nM, see Table 1). Comparable affinities were also observed for transiently expressed human GABAA receptors (data not shown).

Fig. 1
figure 1

Structure of RO4938581 (3-bromo-10-difluoromethyl-9H-imidazo[1,5-a][1,2,4]triazolo[1,5-d][1,4]benzodiazepine)

Full size image
Table 1 Affinity of RO4938581 for rat recombinant GABAA receptor subtypes
Full size table

Effects of RO4938581 on GABA-induced responses in HEK293 cells

To determine the maximal allosteric modulation of the GABA responses in HEK293 cells stably expressing the α1β2γ2, α2β3γ2, α3β3γ2, and α5β3γ2 GABAA receptor subtypes, a representative inverse agonist, the β-carboline β-CCM was tested. In all cells, β-CCM (1 μM) decreased the current induced by a GABA concentration evoking an ∼20% maximum response. The maximum inhibition for GABA-induced current inhibition by β-CCM were −54 ± 3%, −39 ± 3%, −40 ± 4%, and −54 ± 4% for the α1β2γ2, α2β3γ2, α3β3γ2, and α5β3γ2 receptor-expressing cells, respectively (n = 8–14, data not shown). RO4938581 concentration dependently inhibited the GABA-induced current in cells expressing α5β3γ2 receptors, (Fig. 2). The maximum inhibition was −46 ± 5%, n = 6; the pIC50 value and Hill slope calculated from the concentration–response curve were 8.13 ± 0.05 and 1.45 ± 0.4, respectively. In contrast to β-CCM, RO4938581 was not or only slightly effective at the other receptor subtypes (α1β2γ2, α2β3γ2, α3β3γ2; Fig. 2).

Fig. 2
figure 2

Inhibition of the GABA response by RO4938581 in HEK293 cells expressing the α1β2γ2 (empty square), α2β3γ2 (empty triangle), α3β3γ2 (filled triangle), and α5β3γ2 (filled circle) subunit combinations. Current amplitudes obtained from GABA applications in the presence of different concentrations of RO4938581 are plotted as a function of the drug concentration. Values are normalized to the control responses obtained with GABA alone. Symbols represent the mean ± SEM of four to six cells. The sigmoidal curve for α5β3γ2 was generated with the mean pIC50 and Hill slope

Full size image

Modulatory effect of RO4938581 on population spikes

The hippocampal brain slice preparation can be used to assess quantitatively the potencies of compounds acting at the GABAA receptor. In particular, BZD receptor ligands have been shown to modulate the inhibition of synaptically evoked CA1 population spikes (PS) by isoguvacine, a GABA-mimetic compound (Kemp et al. 1987). We therefore used this protocol to examine the effect of RO4938581 on native GABAA receptors in rat hippocampal slices. As described by Kemp et al. (1987), isoguvacine induced a concentration-dependent decrease in PS amplitude reaching complete inhibition at 128 μM (Fig. 3). When RO4938581 (300 nM) was applied in the absence of isoguvacine, no change of the PS amplitude was observed. However, the concentration response curve for PS inhibition by isoguvacine was shifted to the right by RO4938581 (300 nM, Fig. 3), i.e., the compound acted as a negative modulator of isoguvacine. The pIC50 values in the absence and presence of 300 nM RO4938581 were 4.57 ± 0.07 (27 nM) and 4.35 ± 0.05 (45 nM), respectively, and significantly different (four slices, p < 0.05, two-tailed t test).

Fig. 3
figure 3

Effect of RO4938581 on the inhibition of population spike (PS) by isoguvacine recorded from hippocampal CA1 neurons. a Superimposed PS obtained in the same hippocampal slice and evoked in the absence (straight line) and presence of isoguvacine (64 μM, dotted line) and in the presence of isoguvacine 64 μM with RO4938581 (300 nM; dashed line). b Concentration–response curves for isoguvacine in inhibiting PS generated in the absence (open circles) and presence (filled circles) of RO4938581 (300 nM). The points represent the mean ± SEM of four slices. The sigmoidal curves are fits through these points

Full size image

Effect of RO4938581 on hippocampal LTP

The plasticity of Schaffer collaterals—CA1 synaptic transmission was investigated in the absence and presence of RO4938581 (Fig. 4). To induce long-term potentiation, we have selected a weak conditioning stimulus that was previously shown to activate GABAA receptors and to be sensitive to BZD receptor ligands (Seabrook et al. 1997). In control experiments, the mean fEPSP slopes 1–5 min after the priming stimulus and 1–5 min after the burst stimulus were enhanced compared to the baseline period before the priming stimulation, but returned to baseline 1 h after stimulation. In the presence of 300 nM RO4938581, the mean fEPSP slopes were also enhanced 1–5 min after the priming and burst stimuli compared to the baseline period, but, in contrast to control experiments, did not return to baseline 1 h after tetanic stimulation. Thus, RO4938581 was able to induce early LTP from a subthreshold tetanic stimulation paradigm.

Fig. 4
figure 4

Time courses of field excitatory postsynaptic potentials (fEPSPs) generated in the absence (empty circles) and presence (filled circles) of 300 nM RO4938581. fEPSPs were recorded from the apical dendritic layer of the CA1 region after stimulation of the Schaffer collateral commissural pathway at 30-s interval. After a stable baseline recording, a priming stimulus (ten events at 100 Hz) and a burst stimulus (four times ten events at 100 Hz every 20 s) were applied to the slices at time points indicated by the bars. This stimulation paradigm induced LTP only in slices incubated in RO4938581 20 min before the priming stimuli and until the end of the experiment. The traces on top are averages of four responses obtained at the times indicated by the letters. Bars: 20 ms, 0.8 mV

Full size image

Receptor occupancy in the rat hippocampus

The degree of receptor occupancy produced by RO4938581 was measured by evaluating its ability to block in vivo the binding of the GABAA α5 receptor radioligand RO0154513 (Hadingham et al. 1993; Lingford-Hughes et al. 2002; Maeda et al. 2003). Following administration of [3H]-RO0154513, the highest specific binding was observed in the hippocampus (data not shown) in agreement with the known distribution of the GABAA α5 receptor (Pirker et al. 2000). Pretreatment of rats with RO4938581 dose-dependently decreased the binding of [3H]-RO0154513 in the hippocampus. The degree of GABAA α5 receptor occupancy produced by RO4938581 at doses of 0.1, 1, and 10 mg/kg p.o. was 30%, 74%, and 90%, respectively (Fig. 5).

Fig. 5
figure 5

Occupancy of rat hippocampal α5-containing GABAA receptors by RO4938581. Receptor occupancy was measured 60 min after oral administration of RO4938581 at dose 0.1, 1, and 10 mg/kg. Receptors were labeled by [3H]-RO0154513 administered 10 min before sacrificing the animals. Bars represent mean ± SD of occupancy measured in three Lister Hooded rats

Full size image

Effect of RO4938581 on spontaneous locomotor activity, neurological assessment, and chain pull (VI20) test

RO4938581 had a significant effect on horizontal activity [F(33,208) = 3.4, p < 0.001], reaching significance at 10 mg/kg p.o. from 0–5 min (p < 0.001), 5–10 min (p < 0.05) and 10–15 min (p < 0.01; Fig. 6a). Both 0.1 and 1 mg/kg significantly (p < 0.01) reduced horizontal activity only during the 10- to 15-min time bin. RO4938581 did not have a significant effect [F(3,28) = 1.3, NS] on total horizontal activity during 1 h (Fig. 6b).

Fig. 6
figure 6

The effect of RO4938581 on spontaneous locomotor activity over a 1-h period: a time course of horizontal activity in 5 min time bins: vehicle (empty circle); 0.1 (filled circle); 1.0 (filled triangle), 10 (empty square) mg/kg; b total horizontal activity in 1 h. Data are presented as mean ± SEM (n = 8 Sprague–Dawley rats per dose group). Statistics: *p < 0.05, **p < 0.01, ***p < 0.001 vs. vehicle

Full size image

The purpose of the neurological assessment test is to determine compound effects on neurological signs, motor performance, and muscle strength. There was no effect of RO4938581 on body posture, gait, reflexes, and other neurological signs. RO4938581 had no effect on grip strength [F(3,20) = 0.48, NS], horizontal wire test (H = 2.4, NS) and rotarod performance at 8 rpm (H = 3.0, NS) and 16 rpm (H = 3.0, NS; Table 2).

Table 2 Effect of RO4938581 on measures of neurological assessment and chain pull responding for food on a variable interval of 20 s (VI20)
Full size table

Food-deprived Lister-hooded rats were previously trained to pull a ceiling chain for food reward on a VI20 schedule. This variable time period leads to a high level of responding based on the unpredictability of food presentation. Hence, the high motoric demand of the test makes it sensitive to disruption by drug effects on motivation and/or motor function. RO4938581 had no effect on the number of chain pulls [F(4,44) = 2.4, p = 0.06] and the number of food rewards obtained [F(4,44) = 1.7, NS] during the 30-min session (Table 2).

RO4938581 attenuates scopolamine-induced working memory impairment in the DMTP task

DMTP is a test of spatial working memory, which is dependent on hippocampal function (Aggleton et al. 1992; Hampson et al. 1999; Higgins et al. 2002). In the first experiment, the effect of RO4938581 administered alone on performance of the DMTP task was assessed (Fig. 7 a,b). RO4938581 did not have an effect on percent correct responses across delays [F(15,165) = 1.0, NS; Fig. 1a] or total percent correct responses during the session [F(3,33) = 0.6, NS; Fig. 1b]. There was a slight but significant increase in latency to make a correct response [F(3,33) = 3.6, p < 0.05] and magazine latency [F(3,33) = 5.3, p < 0.01] following 10 mg/kg (Table 3).

Fig. 7
figure 7

a The effect of RO4938581 administered alone, on short-term memory function measured by choice accuracy across delay interval in the delayed match to position (DMTP) task, at 0.1 (empty diamond); 1.0 (empty triangle) and 10 (empty square) mg/kg compared to vehicle (empty circle); n = 12 Lister Hooded rats. b The effect of RO4938581 at 0.1 (filled diamond), 0.3 (filled triangle), and 1.0 (filled square) mg/kg versus SCOP [scopolamine (filled circle)]-induced working memory impairment [compared to vehicle (empty circle)]; n = 11 Lister Hooded rats. Data are presented as mean percent correct responses ±SEM. On the left hand side of the figure, the data are presented by delay interval; on the right hand side, the data are collapsed across delay for clarity. Statistics: **p < 0.01, ***p < 0.001: versus vehicle (V) treated group; #p < 0.05, ##p < 0.01: versus scopolamine-treated group

Full size image
Table 3 Effect of RO4938581 on measures of motivation and/or motor performance in the delayed match to position (DMTP) task
Full size table

In the second experiment, the effect of RO4938581 versus a scopolamine-induced impairment was assessed (Fig. 7c,d). Scopolamine significantly reduced percent correct responses in a delay-dependent manner (p < 0.01 at 16 and 24 s), indicating an impairment of working memory in the DMTP task. RO4938581 significantly (p < 0.01) reversed the scopolamine impairment at 0.3 and 1 mg/kg p.o. at both delay intervals to control levels (Fig. 1c). At 0.1 mg/kg p.o., there was a partial, but significant (p < 0.05) reversal at the 24-s delay only. Scopolamine significantly (p < 0.001) reduced total percent correct responses (Fig. 1d), which was significantly reversed by RO4938581 at 0.1 (p < 0.01), 0.3 (p < 0.01), and 1 mg/kg (p < 0.01).

RO4938581 attenuates diazepam-induced spatial learning impairment in the Morris water maze

The Morris water maze is a test of spatial learning and memory which is dependent on hippocampal function (Morris 1984). Diazepam did not have an effect on swim speed [F(4,43) = 1.5, NS], but significantly increased the latency (p < 0.01) and path length (p < 0.01) traveled to find the hidden platform position (data not shown). RO4938581 significantly reversed the diazepam-induced increase in latency [F(4,43) = 4.3, p < 0.01] and path length [F(4,43) = 6.0, p < 0.001]. During the probe trial, vehicle-treated rats spent significantly more time in the platform quadrant compared to the left, right, and opposite quadrants [F(3,24) = 9.6, p < 0.001; Fig. 8a]. In contrast, diazepam-treated rats spent significantly less time in the platform quadrant (p < 0.01; Fig. 8a) and had significantly fewer platform crossings (p < 0.01; Fig. 8b) compared to the vehicle group, indicating an impairment in spatial learning. RO4938581 significantly attenuated the diazepam-induced impairment at 1 [F(3,27) = 6.2, p < 0.01], 3 [F(3,27) = 3.9, p < 0.05] and 10 mg/kg [F(3,27) = 27.9, p < 0.001]. Furthermore, the time spent in the platform quadrant at 10 mg/kg (p < 0.01; Fig. 8a) and the number of platform crossings at 1 (p < 0.05), 3 (p < 0.05), and 10 mg/kg (p < 0.01; Fig. 8b) were significantly higher than diazepam.

Fig. 8
figure 8

The effect of RO4938581 at 1, 3, and 10 mg/kg versus diazepam-induced spatial learning impairment in the water maze during the probe trial: a percent time spent in left (dotted bar), platform (filled bar), right (empty bar) and opposite (diagonally striped bar) quadrants; b number of platform crossings. Data are presented as mean ± SEM (n = 10 Lister Hooded rats per dose group). Statistics (within treatment groups): +p < 0.05, ++p < 0.01, +++p < 0.001 platform quadrant vs. left, right and opposite quadrants. Statistics (between treatment groups): #p < 0.05, ##p < 0.01 vs. diazepam-treated group

Full size image

RO4938581 enhances executive function in the object-retrieval task in monkeys

The object-retrieval (detour-reaching) task is used to assess prefrontal executive function in monkeys (Rutten et al. 2008). RO4938581 increased first reaches (% correct) during difficult trials [F(5,55) = 18.2, p < 0.001], reaching significance at 3 and 10 mg/kg p.o. (Fig. 9). There were no effects on easy trials (performance throughout the study was 100% correct on this trial type).

Fig. 9
figure 9

The effect of RO4938581 administered alone at 0.1, 0.3, 1, 3, and 10 mg/kg on percent correct (first reaches) during difficult trials of the object retrieval task in adult cynomolgus macaques. Data are presented as mean ± SEM (n = 12). Statistics: **p < 0.01 vs. vehicle (V)

Full size image

RO4938581 does not induce an anxiogenic-like effect in the SAA and elevated plus maze

The SAA and EPM tests can be used to assess potential anxiogenic-like effects of compounds (Nicolas and Prinssen 2006). RO4938581 did not have an effect [F(3,36) = 0.9, NS] on time spent in the hidden area of the SAA test (Fig. 10a). Moreover, RO4938581 did not have an effect [F(3,36) = 0.3, NS] on time spent on the open arms of the elevated plus maze (Fig. 10b). The positive control, RO0194603, significantly increased the time spent in the hidden area in the SAA test (Fig. 10a) and significantly reduced the time spent on the open arms of the elevated plus maze (Fig. 10b), indicating an anxiogenic-like profile.

Fig. 10
figure 10

The effect of RO4938581 administered at 0.3, 1, 3 mg/kg on: a time spent in the hidden zone in the social approach avoidance test (n = 10 Fischer rats per dose group); and b time spent on the open arms of the elevated plus maze (n = 10 Sprague–Dawley rats per dose group). Data are presented as mean ± SEM. Statistics: **p < 0.01 vs. vehicle (V). RO is RO0194603 (3 mg/kg), which was tested as a positive control in a separate experiment (n = 7–8 per dose group)

Full size image

Lack of proconvulsant activity of RO4938581 following an auditory stimulus

The proconvulsant potential of RO4938581 was investigated in naive 19- to 21-day-old DBA/2J mice (a strain that exhibits age-dependent susceptibility to seizure following an auditory stimulus). RO4938581 did not have any proconvulsant activity up to 100 mg/kg in DBA2/J mice following an auditory stimulus (Table 4). In contrast, the positive control, RO0194603, significantly increased the number of mice demonstrating tonic seizures (p < 0.001).

Table 4 Lack of pro-convulsant activity of RO4938581 following an auditory stimulus (70 dB, 14 kHz, 60 s) in 19–21 days DBA2/J mice
Full size table

Discussion

We have identified a new imidazo-triazolo-benzodiazepine compound, RO4938581, that displays both binding selectivity and inverse agonist functional selectivity for GABAA α5 subunit-containing receptors. Since GABAA α5 receptors are predominantly located in the hippocampus (Fritschy and Möhler 1995; Pirker et al. 2000), and there is evidence that these receptors play a role in hippocampal-dependent learning and memory (Atack et al. 2006b; Collinson et al. 2006; Crestani et al. 2002; Dawson et al. 2006), we evaluated the effect of RO4938581 in two rodent cognition tests dependent on hippocampal function: the DMTP task (Aggleton et al. 1992; Hampson et al. 1999; Higgins et al. 2002) and the Morris water maze (Morris 1984). We used two different pharmacological approaches (cholinergic and GABAergic) to induce a cognitive impairment in each of the tasks, i.e., scopolamine and diazepam, respectively. In the DMTP task, RO4938581 reversed the scopolamine-induced working memory impairment, and in the Morris water maze, it reversed the diazepam-induced spatial learning impairment. In addition, RO4938581 also improved the performance of nonpharmacologically impaired adult monkeys in the object retrieval task, which is a test requiring input from the prefrontal cortex (Diamond et al. 1989; Wilkinson et al. 1997). RO4938581 did not show anxiogenic or proconvulsant activity and did not induce adverse neurological effects, or impair muscle strength, motor coordination, and responding for food on a VI20 schedule.

Diazepam impairs spatial learning in naïve rats in the Morris water maze (Ballard et al. 2005), and it has been proposed that BZD agonists may induce memory impairment by modulating hippocampal function (McNaughton and Morris 1987). The current study used an abbreviated version of the standard acquisition protocol. Subjects were pretrained to the task, but had to learn a new spatial position on the test day. Diazepam impaired the ability of the rats to find the spatial location of the hidden platform at a dose which did not affect swim speed. RO4938581 was able to reverse the spatial learning impairment. This effect is not simply due to displacement of diazepam from the BZD binding site, since RO4938581 does not reverse other behavioral effects of diazepam, such as impairment of rotarod performance (Ballard 2005). In contrast, compounds with no binding selectivity, fully reverse diazepam-induced rotarod impairment. Since RO4938581 has binding selectivity over the other subtypes, this indicates that reversal of diazepam impairment in the water maze is mediated via inverse agonism at the GABAA α5 receptor.

Nonselective GABAA inverse agonists attenuate scopolamine-induced cognitive impairment in rats and humans (Duka et al. 1996; Kawasaki et al. 1996), In the present study, using an α5 selective inverse agonist, we obtained evidence for an interaction between GABAA α5 receptors and the cholinergic system in a working memory test, the DMTP task. We speculate that reversal of the scopolamine-induced impairment could be due to an increase in hippocampal acetylcholine levels, since a nonselective GABAA partial inverse agonist, S-8510, has been shown to have this effect in vivo (Kawasaki et al. 1996). However, to verify whether GABAA α5 inverse agonists have an effect on acetylcholine release, such compounds should be assessed during microdialysis of the hippocampus in vivo.

In vivo binding experiments in the Lister hooded rat revealed that doses of RO4938581 (0.1-1 mg/kg) which reversed the cognitive impairment induced by scopolamine in the DMTP occupied approximately 30-70% of the GABAA α5 receptor subtype. These results are in agreement with previous studies where an α5 functionally selective compound produced cognition enhancing effects at a minimal effective dose that corresponded to a receptor occupancy of approximately 25% (Dawson et al. 2006). A dose-range of 1-10 mg/kg was active in the Morris water maze in Lister hooded rats, which corresponds to receptor occupancies of 70-90%. This is higher than the value found by Dawson et al. 2006, however, since all doses were significantly active, a lower dose-range should be tested in future studies.

In rat hippocampal slices, RO4938581 modulated GABAergic inhibition of CA1 pyramidal cells, as shown by the rightward shift of the isoguvacine concentration–response relation. The extent of the shift was less than that caused by the nonselective inverse agonists DMCM and β-CCM (Kemp et al. 1987; Knoflach, unpublished data). As expected for an inverse agonist acting at the BZD site of the GABAA receptor, RO4938581 had no direct effect on PS when applied alone to the slices. This suggests that α5 as well as other GABAA receptor subtypes play a role for the inhibition of hippocampal CA1 neurons.

Hippocampally mediated cognitive processes involve long-term changes in synaptic efficacy such as LTP. Nonselective GABAA inverse agonists like DMCM increase LTP (Seabrook et al. 1997), whereas nonselective GABAA agonists, e.g., diazepam impair LTP (del Cerro et al. 1992). In the present study, an α5 selective inverse agonist induced early LTP from a subthreshold tetanic stimulation paradigm in mouse hippocampal slices. This result is in agreement with previous reports which showed enhancement of LTP by other α5 selective inverse agonists (Atack et al. 2006b; Dawson et al. 2006). Therefore, we conclude that the α5 subtype plays a major role in the modulation of LTP as it was previously observed with nonselective compounds. GABAA α5-containing receptors are predominantly localized extrasynaptically and mediate tonic inhibition (Glykys et al. 2008). A recent study suggested that GABAA α5 receptors regulate the excitability of mouse hippocampal pyramidal neurons by influencing the strength of depolarization required to generate an action potential (Bonin et al. 2007). This way, it has been explained how extrasynaptic receptors can influence the expression of LTP as it is shown in the present study.

Since nonselective inverse agonists are known to have a poor side effect profile (e.g., Dorow et al. 1983), we determined the effect of RO4938581 in a battery of noncognitive tests. RO4938581 had no effect on neurological signs, motor performance, and muscle strength. Moreover, the compound did not impair performance in the chain pull (VI20) test, indicating no effect on motivation or motor function. In the DMTP task, RO4938581 induced a slight increase in latency to press the correct lever and to collect the food reward at 10 mg/kg. The same dose decreased horizontal activity during the first 15 min of a 1-h test session, but had no effect on swim speed in the Morris water maze (in combination with diazepam). Moreover, no sedative-like effects were observed in the monkeys during performance of the object retrieval task. A similar effect on locomotor activity has been reported by Savic et al. (2008) for the binding selective partial inverse agonist PWZ-029 at doses slightly higher than the cognition-enhancing dose (10–20 mg/kg vs. 5 mg/kg). This effect was shown to be a consequence of inverse agonism at α5, since it was reversed by flumazenil and an α5 selective agonist, but not by an α1 selective antagonist (Savic et al. 2008). Since very few selective inverse agonists have been tested for locomotor activity, we refer to the detailed discussion of these effects by Savic et al. (2008) suggesting that this behavioral change requires further investigation with compounds of different profiles.

We evaluated the anxiogenic potential of RO4938581 in the SAA and EPM tasks. RO4938581 did not have an effect on time spent in the hidden area of the SAA test and did not have an effect on time spent on the open arms of the elevated plus maze indicating no anxiogenic-like profile. In contrast, the positive control RO0194603, which is a nonselective inverse agonist, exhibited an anxiogenic-like profile (Nicolas and Prinssen 2006).

RO4938581 did not have any proconvulsant activity up to 100 mg/kg p.o. in DBA2/J mice following an auditory stimulus. The audiogenic seizure model was used since a deficit in GABA-mediated inhibition in inferior colliculus neurons has been shown to be critical in this model and to be sensitive to GABAA antagonists or inverse agonists (Faingold 2002). Furthermore, the underlying mechanisms have been proposed to be similar to other models of epilepsy (Ross Coleman 2000). In order to fully explore potential proconvulsant effects, additional methods which determine seizure threshold could be used. However, it should be noted that RO4938581 does not have convulsant activity up to 300 mg/kg p.o in rats (Ballard 2005). The lack of an anxiogenic or proconvulsant effect is consistent with the behavioral phenotype of α5−/− mice (Collinson et al. 2002) and observations with the functionally α5 selective compound, α5IA (Dawson et al. 2006).

In vivo binding experiments showed that receptor occupancy at 10 mg/kg in Lister Hooded rats was approximately 90%. The SAA and EPM tests were run in Fischer and Sprague–Dawley rats, respectively, so one can only assume that at the dose-range tested, high levels of GABAA α5 receptor occupancy were achieved with RO4938581, i.e., more than 70% at the maximal dose tested. It is also assumed that full receptor occupancy occurs at the maximum dose (100 mg/kg) tested in the mouse audiogenic seizures test. These data further support the notion that the multiple effects of nonselective BZD compounds are mediated through distinct GABAA receptor populations (Rudolph and Möhler 2004) and confirms that sufficient selectivity is important for excluding side effects associated with inverse agonistic activity at the α1, α2, and α3 receptor subtypes. In addition, other studies demonstrated that high-binding selectivity is required to avoid side effects of BZD inverse agonists with little functional selectivity (Atack et al. 2006a; Atack et al. 2006b). It is probably safest to avoid effects on other GABAA receptor subtypes by combining both binding and functional selectivity at the α5 receptor. RO4938581 displays such a profile; it binds with higher affinity to GABAA α5 receptors and displays close to full inverse agonistic effect at GABAA α5 receptors (−41%) with no or only a very weak effect at the other GABAA receptor subtypes (±10% at 3 μM concentration). RO4938581 demonstrated essentially the same profile against human GABAA receptors transiently expressed in oocytes and has negligible activity (<50% activity at 10 μM) in 71 receptor binding and 16 enzyme assays (CEREP, Paris, France; data not shown).

The GABAA α5 inverse agonist, α5IA, enhanced performance in the delayed-matching-to-position version of the Morris water maze in unimpaired rats (Dawson et al. 2006). In contrast, RO4938581 did not improve working memory of unimpaired rats in the DMTP task. However, the rats used for the latter experiment were adults pretrained to asymptotic performance, and so, it would not be possible to detect a cognitive improvement in this particular test. In the object retrieval task, young adult monkeys were exposed to the task infrequently to prevent asymptotic performance, thus allowing a window for improvement. The positive effect of RO4938581 suggests that α5 selective inverse agonists can enhance cognition in the absence of pharmacological impairment.

The object retrieval (or detour reaching task) is a task of prefrontal cognition in monkeys which involves attention, response inhibition, and planning, i.e., executive function. Lesions of the frontal cortex, but not of the hippocampus, impaired performance of this task (Diamond et al. 1989; Wilkinson et al. 1997). As mentioned above, there is evidence for improvement in a hippocampal-dependent memory task following administration of a GABAA α5 selective inverse agonist (Dawson et al. 2006). However, so far, there has been little evidence for the involvement of GABAA α5 receptors in prefrontal cortex-dependent executive function. GABAA α5 receptors have been shown to be located in the cortex in rat, monkey, and man, but to a lesser degree than in the hippocampus (Akbarian et al., 1995; Fritschy and Möhler 1995; Maeda et al. 2003). The current results illustrate the important role of this receptor subtype on prefrontal cortex-mediated executive function.

We have shown that the dual binding and functional selectivity of RO4938581 offers a unique profile for enhancing cognition without side effects associated with other GABAA receptor subtypes. Importantly, it was recently reported in a clinical study that a GABAA α5 functionally selective inverse agonist reduced the amnesic effect of alcohol (Nutt et al. 2007). In the hippocampus of Alzheimer’s disease patients, the density and pharmacology of the GABAA α5 subunit-containing receptors is preserved (Howell et al. 2000). Therefore, selective GABAA α5 inverse agonists have the potential to improve cognition in patients suffering from Alzheimer’s disease and other types of dementias.