Comparative Cue Generalization Profiles of L-838, 417, SL651498, Zolpidem, CL218,872, Ocinaplon, Bretazenil, Zopiclone, and Various Benzodiazepines in Chlordiazepoxide and Zolpidem Drug Discrimination

The majority of GABA_A receptors in the mammalian brain contain at least one α, β, and γ subunit (Barnard et al., 1998). The likely stoichiometry is two α, two β, and one γ subunit(s) arranged pentamerically around a central pore through which Cl^– ions flow when the receptor is activated. Classic benzodiazepines modulate GABA_A neurotransmission by binding to a site that traverses an α and γ subunit of α₁, α₂, α₃, or α₅ containing GABA_A receptors (Sieghart, 1995). Mice with a point mutation of either the α₁ (Rudolph et al., 1999; McKernan et al., 2000) or α₂ (Low et al., 2000) subunit are insensitive to the motor-impairing and anxiolytic-like effects of diazepam, respectively, suggesting different functional roles for GABA_A-α₁ and α₂ receptors.

Pharmacological evidence has also alluded to different functional roles for GABA_A receptor subtypes (e.g., Griebel et al., 1999b, Chen et al., 1996). For example, the triazolopyridazine CL218,872 was found to bind with different affinity to two populations of benzodiazepine receptor, originally designated type I (α₁ containing GABA_A receptors) and II (α₂, α₃, or α₅ containing GABA_A receptors) receptors (Klepner et al., 1979; Squires et al., 1979). Furthermore, the benzodiazepine receptor antagonist β-carboline-3-carboxylate t-butyl ester (β-CCT) selectively antagonized the anticonvulsant and anxiolytic, but not the myorelaxant, effects of diazepam, whereas flumazenil blocked all three effects (Shannon et al., 1984). Further studies confirmed that β-CCT was more selective than flumazenil in antagonizing the behavioral effects of benzodiazepine site ligands in vivo (Griebel et al., 1999a; Paronis et al., 2001; Rowlett et al., 2005a). In vitro studies showed that β-CCT had selective affinity for GABA_A-α₁ receptors (Huang et al., 1999) and was more potent at antagonizing zolpidem potentiation of GABA's effect at GABA_A-α₁ receptors than diazepam potentiation of GABA at GABA_A-α₃ or GABA_A-α₅ receptors (Griebel et al., 1999a).

The discriminative stimulus effects of benzodiazepines have been widely described, and this tool has been employed to determine receptor selectivity and efficacy of benzodiazepine site ligands (see review by Lelas et al., 2000). For example, in squirrel monkeys trained to discriminate zolpidem at low (1 mg/kg) and high doses (>3 mg/kg), generalization to nonselective benzodiazepines occurred in the former but not latter set of animals (Rowlett et al., 1999, 2000). Moreover, in animals discriminating high doses of zolpidem, generalization to the GABA_A-α₁-selective drug zaleplon occurred. Thus, at certain training doses, the zolpidem-discriminative stimulus is mediated via GABA_A-α₁ receptors. In rats, the zolpidem-discriminative stimulus might also be mediated via GABA_A-α₁ receptors, whereas the chlordiazepoxide-discriminative stimulus might be mediated by non-α₁ GABA_A receptors (Sanger and Benavides, 1993; Sanger et al., 1999). This is based on the finding that drugs with affinity-selectivity for GABA_A-α₁ receptors (e.g., zolpidem, CL218,872) fully generalized to the zolpidem but not the chlordiazepoxide cue.

In this study, we tested the hypothesis that the zolpidem- and chlordiazepoxide-discriminative stimuli in rats are mediated predominantly via GABA_A-α₁ and non-α₁ GABA_A receptors, respectively, by testing novel compounds not previously available as well as well characterized pharmacological tools. The compounds tested fell into various categories on the basis of affinity and efficacy. 1) One category was nonselective full agonists, modulators that potentiate the effect of GABA at GABA_A-α₁, -α₂, -α₃, and -α₅ receptors equivalently (Faure-Halley et al., 1993; Hadingham et al., 1993; Sieghart, 1995). This category included benzodiazepines and zopiclone, largely overlapping with compounds tested previously (Sanger and Benavides, 1993; Sanger et al., 1999). 2) The second category included selective positive benzodiazepine site modulators, which were subdivided into two groups. a) One group included compounds with selective affinity, including the imidazopyridine zolpidem and triazolopyridazine CL218,872, ligands that selectively bind to GABA_A-α₁ receptors. These compounds selectively potentiate the effect of GABA at GABA_A-α1 receptors, although this is concentration-dependent (i.e., they only show 10–20-fold affinity selectivity for GABA_A-α₁ receptors over GABA_A-α₂ or -α₃ receptors), despite >1000-fold selectivity over GABA_A-α₅ receptors (Hadingham et al., 1992; Faure-Halley et al., 1993; Sieghart, 1995). To add to the complexity, zolpidem has high-intrinsic efficacy at GABA_A-α₁ receptors, whereas CL218,872 displays reduced efficacy at this receptor; i.e., CL218,872 is a partial agonist at GABA_A-α₁ receptors. b) The second group contained functionally (or efficacy) selective compounds that included L-838,417 (McKernan et al., 2000) and SL651498 (Griebel et al., 2001). These compounds bind with similar affinity to α₁, α₂, α₃, or α₅ containing GABA_A receptors but are able to differentially modulate these receptor subtypes leading to selectivity. Thus, L-838,417 does not potentiate the effect of GABA at GABA_A-α₁ receptors but does so at GABA_A-α₂, -α₃, and -α₅ receptors. Moreover, L-838,417 is a partial agonist at these latter receptor subtypes. SL651498 differs in that it fully potentiates the effect of GABA at GABA_A-α₂ receptors and has marginally lower efficacy at GABA_A-α₃ receptors, with least efficacy (partial agonist profile) at GABA_A-α₁ and -α₅ receptors.

We predicted that zolpidem and CL218,872 would generalize fully to the zolpidem but not chlordiazepoxide cue, whereas an opposite profile would be seen with L-838,417 and possibly SL651498 (see below). However, since a number of the selective positive modulators described above have partial agonist profiles, in addition to showing either affinity or functional selectivity, any difference in behavioral profile could potentially be attributed to their partial agonism rather than selectivity properties (Lelas et al., 2000). Therefore, we included 3) a nonselective partial agonist, such as bretazenil, which potentiates the effect of GABA equally at GABA_A-α₁, -α₂, -α₃, and -α₅ receptors, albeit with reduced efficacy compared with a full agonist (Pieri et al., 1988; Haefely et al., 1990). Finally, we predicted that, because SL651498 has appreciably greater efficacy at GABA_A-α₁ receptors than L-838,417 (Atack, 2003), it would show correspondingly greater zolpidem-like effects.

A number of compounds with selectivity for non-α₁ GABA_A receptors are currently in clinical trials, including SL651498. However, compounds that modulate GABA_A receptors with different profiles are also being tested in the clinic. For example, the pyrazolopyrimidine ocinaplon has an anxioselective profile in man, although Phase III clinical trials for the treatment of generalized anxiety disorder have been halted due to potential liver toxicity risks (Lippa et al., 2005; Basile et al., 2004; www.bioworld.com). Ocinaplon is a low-affinity benzodiazepine site ligand with modest selectivity for GABA_A-α₁ receptors (Lippa et al., 2005). We predicted that ocinaplon would generalize similarly to both discriminative cues.

Materials and Methods

Animals. Male PVG hooded rats were used (200–250 g; M&B Breeding and Research Centre, Ry, Denmark). Animals were housed and habituated for at least 7 days in Macrolon III cages (20 × 40 × 18 cm, 2 rats/cage; Scanbur BK A/S, Ejby, Denmark) before operant conditioning commenced. Food (Altromin; Brogaarden A/S, Buddinge, Denmark) was available ad libitum, whereas water was available 10 min every day after drug discrimination training sessions (see below) and ad libitum at weekends. Animals were housed on a 12-h light/dark cycle (lights on at 7:00 AM and off at 7:00 PM). All testing procedures were in accordance with Methods and Welfare Considerations in Behavioral Research with Animals (NIH Publication 02-5083, March 2002) and the Danish Committee for Experiments on Animals.

Drugs and Solutions. Diazepam, midazolam, alprazolam, lorazepam, triazolam, chlordiazepoxide, zopiclone, and CL218,872 were all purchased from various commercial sources [Sigma-Aldrich (St. Louis, MO), Sigma/RBI (Natick, MA), or Tocris Cookson Inc. (Bristol, UK)]. Zolpidem, ocinaplon, SL651498, and L-838,417 were all synthesized by the medicinal chemistry department (NeuroSearch A/S). Bretazenil and flumazenil were gifts from Roche Diagnostics (Basel, Switzerland). All drugs were prepared in 5% Cremophor (BASF, Ludwigshaden, Germany) and administered i.p. in a volume of 1 ml/kg. The majority of drugs were administered 30 min prior to drug discrimination test sessions. However, zolpidem, midazolam, and zopiclone were administered 15 min before test sessions and triazolam was administered 10 min before test sessions.

Drug Discrimination Protocol. For discrimination training, standard operant chambers [Coulbourn Instruments (Allentown, PA)] equipped with two levers and a waterspout located equidistant between the two levers were used. The operant chambers were enclosed within sound isolation cubicles, and behavioral contingencies, reinforcement scheduling, and data collection as outlined below were controlled by a MedState Notation (version 2.0) WMPC software package (MED Associates (St. Albans, VT).

Basic Training. Animals were water-deprived for 24 h before the first training session. Rats were shaped to lever press for water on a schedule where one water reward (0.2 ml) was delivered every 60 s and one lever press on either left or right lever was reinforced with an additional water delivery. Animals were kept on this schedule until they operated both levers to obtain water rewards. Rats that did not operate levers at all after 5 days were excluded from further training. Once adequate lever responding was attained, rats were switched to a FR1 schedule (i.e., water was delivered each time a lever was pressed) without any supplemental water every 60 s. Animals were then progressed rapidly to a FR10 schedule; i.e., water was delivered only after 10 consecutive presses were made on a single lever. Rats were trained to reliably press both levers on the FR10 schedule for water reward before commencing discrimination training. Further training was given for individual rats if necessary; e.g., if a rat had a nonpreferred lever, it was maintained on this lever for 5 days until it was able to reliably complete the FR10 schedule.

Discrimination Training. In total, 24 animals were trained to discriminate chlordiazepoxide (5 mg/kg) from vehicle and a different group of 24 animals was trained to discriminate zolpidem (3 mg/kg) from vehicle, although not all animals were trained in the same time-frame. For each discrimination, half of the rats were trained so that the left lever was the drug-associated lever, and for the other half, the right lever was the drug-associated lever. In all cases, the opposite lever was the vehicle-associated lever. Drug (D) or vehicle (V) injections were given before training sessions according to the following pseudorandom 2-weekly sequence (Monday through Friday): DVVDDVDDVV. Training sessions lasted for 15 min, and only the lever appropriate to the injection schedule above was active in a given session (i.e., resulted in water delivery on completion of an FR10). Each rat was trained to a criterion of making the first 10 consecutive responses on the correct lever, appropriate for drug or vehicle injection, when first placed in the chamber and when the session initiated. Moreover, animals had to make ≥80% of responses on the correct lever, appropriate for drug or vehicle injection, in a session. This level of performance had to be maintained for 2 weeks before drug testing. Drug discrimination training took ∼3 months. Initial studies showed that flumazenil (10 mg/kg i.p.) could antagonize both discriminative cues (data not shown).

Generalization Tests. The testing schedule differed from the training schedule in that both levers resulted in water delivery on a FR10 schedule and the session lasted 5 min. Test sessions with experimental drugs (see “Drugs and Solutions”) were run on Tuesdays and Thursdays, with baseline sessions according to the above-mentioned pseudorandom injection sequence on other days. If at any time during baseline training sessions a rat selected the wrong lever, it was retained on training until for three consecutive sessions it was able to: i) make the first 10 consecutive responses on the correct lever (appropriate for drug or vehicle injection) when first placed in the chamber; and ii) make ≥80% responses on the lever appropriate to the injection protocol (i.e., drug or vehicle injection). This criterion, in addition to the fact that not all 24 animals from each of the two discriminations were available at the same time, means the number (n) for each drug tested differs (see figure legends).

Data Analyses. Two measures of generalization potential of a test drug to the training drug are presented (see Stolerman, 1993): i) the percentage of responses on the drug-associated lever (drug-lever responses/total responses × 100) and ii) the number of rats correctly selecting the drug appropriate lever (i.e., the first 10 consecutive responses on the drug lever when first placed in the chambers and the session is initiated). An animal had to make at least 10 responses on one of the levers to be included in either index of discrimination. A test compound was deemed to show full generalization to the training drug when ≥80% responding was on the drug lever and/or ≥80% of animals selected the drug-associated lever. Response rate (responses/second) was analyzed by one-way analysis of variance followed by post hoc Dunnett's test.

Results

Nonselective Full Agonists

Chlordiazepoxide Discrimination. All compounds in Figs. 1 and 2 resulted in a dose-dependent increase in percentage drug-lever appropriate responding, with rats making >90% drug-lever responses after administration of the highest dose of each drug tested (Fig. 1, A–D, and Fig. 2, A–C). The only exception was triazolam, with rats making >90% responses on the drug lever at all of the doses tested and with no clear dose dependence. Likewise, more animals selected the drug lever (i.e., the first 10 consecutive responses were on the drug lever) with increasing dose of each drug, with 90 to 100% of animals selecting the drug lever at some dose of each drug (see fractions in Figs. 1, A–D, and 2, A–C).

With respect to effects on response rate, zopiclone (F[3,63] = 0.9), chlordiazepoxide (F[3,44] = 1.7), diazepam (F[3,28] = 0.8), and midazolam (F[3,42] = 1.0) had no significant effect on this measure, even at doses exceeding those giving >90% responses on the drug lever (Fig. 1, A–D). For these compounds, almost all animals met the inclusion criteria of completing at least 10 responses on one of the levers. By contrast, triazolam (F[5,103] = 9.8, P < 0.001), alprazolam (F[5,79] = 10.9, P < 0.001) and lorazepam (F[3,31] = 6.7, P < 0.001) significantly and dose-dependently reduced response rates (Fig. 2, A–C).

Zolpidem Discrimination. In the zolpidem cue condition, full generalization was seen with all nonselective full agonists; i.e., rats made ≥80% of responses on the drug lever at some dose of each of the drugs (Figs. 1, E–H, and 2, D–F). Furthermore, the number of animals selecting the drug lever (i.e., the first 10 consecutive responses were on the drug lever) increased as a function of dose (as seen for the chlordiazepoxide cue), with >80% of animals selecting the zolpidem lever at some dose of each drug (see fractions in Figs. 1, E–H, and 2, D–F).

Regarding response rates, all compounds significantly reduced response rate at some dose (Figs. 1, E–H, and 2, D–F). Thus, in contrast to their lack of significant effect on response rate in chlordiazepoxide-trained animals, in zolpidem-trained animals zopiclone (F[5,91] = 8.4, P < 0.001), chlordiazepoxide (F[3,26] = 21.6, P < 0.001), diazepam (F[4,76] = 9.9, P < 0.001), and midazolam (F[3,30] = 24.2, P < 0.001) all significantly reduced response rate (Fig. 1, E–H). Triazolam (F[5,66] = 29.4, P < 0.001), alprazolam (F[5,66] = 14.8, P < 0.001), and lorazepam (F[3,25] = 8.2, P < 0.001) all significantly and dose-dependently reduced response rate in zolpidem-trained animals (Fig. 2, D–F), as they did in chlordiazepoxide-trained animals. Of particular interest, for the majority of drugs tested, rats only made ≥80% responses on the zolpidem lever at doses that also significantly reduced response rate. Exceptions were lorazepam (Fig. 2F) and zopiclone (Fig. 1H), where full generalization was seen at doses of 1.0 and 10 mg/kg, respectively, with no significant reduction in response rate. However, at higher doses, both lorazepam (3.0 mg/kg) and zopiclone (30 mg/kg) significantly reduced response rates. This was particularly evident at 3 mg/kg lorazepam, leading to a disruption of discriminative control.

Selective Positive Benzodiazepine Site Modulators

Chlordiazepoxide Discrimination. Neither zolpidem nor CL218,872 generalized to the chlordiazepoxide cue whether measured as the percentage of drug-lever responses made by rats or number of animals selecting the drug lever (i.e., the first 10 consecutive responses were on the drug lever; see Fig. 3, A and B). A maxima of three of 11 zolpidem- and three of nine CL218,872-treated rats selected the drug lever at any one dose (see fractions in Fig. 3, A and B). Both zolpidem (F[4,45] = 9.8, P < 0.001) and CL218,872 (F[4,48] = 2.9, P < 0.03) significantly reduced the response rate. This was particularly marked at 10 mg/kg zolpidem, with only two animals attaining the inclusion criterion.

In contrast, both L-838,417 (Fig. 3C) and SL651498 (Fig. 3D) dose-dependently increased chlordiazepoxide-appropriate responding, with rats making >90% responses on the drug lever after administration of the highest dose of either drug. Furthermore, the number of animals selecting the drug lever (i.e., the first 10 consecutive responses were on the drug lever) increased as a function of dose, with 90 to 100% of animals selecting the chlordiazepoxide lever at the highest dose of L-838,417 and SL651498 (see fractions in Fig. 3, C and D). SL651498 had no significant effect on response rate (F[3,59] = 0.2) over the dose range tested. By contrast, the effect of L-838,417 on response rate was marginally significant (F[4,29] = 2.54, P < 0.06). Although not significant, this finding is worth noting because it would seem that L-838,417 had a tendency to increase response rate in chlordiazepoxide-trained animals (Fig. 3C).

Zolpidem Discrimination. A very different, and to some extent opposite, pattern of results was seen with the four test drugs in this discriminative cue condition. Both zolpidem (Fig. 3E) and CL218,872 (Fig. 3F) fully generalized to this cue, with rats making >90% of responses on the drug lever at doses of 1.0 and 3.0 mg/kg, respectively. At these doses of zolpidem and CL218,872, 100% of animals selected the drug lever (i.e., the first 10 consecutive responses were on the drug lever; see fractions in Fig. 3, E and F). As in chlordiazepoxide-trained animals, zolpidem significantly (F[3,39] = 3.7, P < 0.02) reduced the response rate (Fig. 3E). In contrast, CL218,872 had no effect (F[3,59] = 1.3), although doses higher than 3 mg/kg were not tested as in chlordiazepoxide-trained animals (Fig. 3F).

In contrast to its full generalization to the chlordiazepoxide cue, L-838,417 resulted in rats making a maximum of ∼50% responses on the zolpidem-associated lever (Fig. 3G). This was also reflected in a maximum of only two of eight animals selecting the zolpidem lever (i.e., the first 10 consecutive responses were on the drug lever) at any given dose (see fractions in Fig. 3G). However, as was the case for the chlordiazepoxide cue, L-838,417 (up to 30 mg/kg) had no significant effect on response rate (F[4,39] = 1.2) and all of the animals tested attained the response criterion. By contrast, SL651498 dose-dependently generalized to the zolpidem cue, leading rats to make ∼80% drug-lever appropriate responses at 10 mg/kg (Fig. 3H). At this same dose, five of seven (71%) animals selected the zolpidem lever (i.e., the first 10 consecutive responses were on the drug lever) compared with 13 of 14 (91%) selecting the chlordiazepoxide lever at this dose. As with its lack of effect on response rate in chlordiazepoxide-trained animals, SL651498 had no significant effect (F[3,29] = 0.6) on response rate in zolpidem-trained animals.

The Partial Positive Allosteric Modulator Bretazenil and the Novel Anxioselective Agent Ocinaplon

Chlordiazepoxide Discrimination. The partial agonist bretazenil dose-dependently occasioned drug appropriate responding with rats making 100% of responses on the drug-associated lever at 10 mg/kg (Fig. 4A). There was no effect on response rate after bretazenil (F[4,53] = 0.4) across the dose range tested, and at the highest dose, all of the animals selected the drug lever (i.e., the first 10 consecutive responses were on the drug lever; see fractions in Fig. 4A). The dose-response curve for ocinaplon was steep, with rats fully generalizing to the drug-associated lever after being administered the highest dose of 10 mg/kg (Fig. 4B) and with all of the animals at this dose selecting the drug lever (i.e., the first 10 consecutive responses were on the drug lever; see fractions in Fig. 4B). At no dose did ocinaplon reduce response rate in these animals (F[3,30] = 2.2).

Zolpidem Discrimination. Bretazenil also showed appreciable generalization to the zolpidem-discriminative cue (Fig. 4C). However, the dose-response curve in the zolpidem-trained animals was shallow compared with that in chlordiazepoxide-trained animals, although bretazenil did occasion ∼70% drug-lever appropriate responding by rats at doses between 3 and 30 mg/kg. Moreover, eight of 10 (80%) animals selected the drug lever (i.e., the first 10 consecutive responses were on the drug lever) at 10 mg/kg bretazenil in the zolpidem cue. No dose of bretazenil affected response rate in these animals (F[5,65] = 1.2). Ocinaplon had a very similar profile in zolpidem-trained animals as it did in chlordiazepoxide-trained animals (i.e., a steep dose-response curve with rats making ≥80% of responses on the drug-associated lever at 10 mg/kg), with most animals (seven of eight) selecting the drug lever at 10 mg/kg (i.e., the first 10 consecutive responses were on the drug lever) and no effect on response rate (F[3,31] = 0.8) at the doses tested (Fig. 4D).

Discussion

The data generated from animals trained to discriminate either zolpidem or chlordiazepoxide indicate that drug discrimination may be a useful tool for further in vivo characterization of subtype-selective GABA_A receptor-positive modulators. Full generalization to chlordiazepoxide by the nonselective full agonists lorazepam, midazolam, diazepam, triazolam, alprazolam, and zopiclone agrees with previous reports (Sanger et al., 1987; Sanger and Benavides, 1993). Likewise, we have shown that these compounds also all fully generalize to the zolpidem cue. The full generalization to both discriminative cues can probably be attributed to the fact that these compounds do not differ substantially in terms of affinity or efficacy at α₁, α₂, α₃, or α₅ containing GABA_A receptors (Faure-Halley et al., 1993; Hadingham et al., 1993; Sieghart, 1995).

Interestingly, previous studies have shown that chlordiazepoxide only partially generalized to the zolpidem cue in rats and squirrel monkeys (Depoortere et al., 1986; Rowlett et al., 1999). In the present study, chlordiazepoxide at 30 mg/kg fully generalized to the zolpidem cue, although it is possible that this level of generalization may reflect the limited number of animals (n = 4) attaining the inclusion criteria at this dose. However, with most nonselective full agonists, maximal generalization to the zolpidem cue occurred at doses significantly reducing the response rate, whereas full generalization to the chlordiazepoxide was not concurrent with a reduction in response rate. This dissociation in the effect on response rate between the two cues was not an artifact of a difference in baseline responding between animals trained on the two discriminations (compare top and bottom panels of all figures). The basis for this dissociation is not clear, although it may be that chlordiazepoxide-discriminating rats have some cross-tolerance to the rate-decreasing effects of nonselective full agonists (Fig. 1, compare response rate in A–D with E–H).

In contrast to the nonselective modulators described above, zolpidem and CL218,872, which both show affinity selectivity for GABA_A-α₁ receptors, showed very limited generalization to the chlordiazepoxide cue. By contrast, zolpidem and CL218,872 fully and dose-dependently generalized to the zolpidem cue. However, ∼60 to 70% generalization to the chlordiazepoxide-discriminative cue has been previously demonstrated with these two GABA_A-α₁-selective compounds (Sanger et al., 1987; Sanger and Benavides, 1993). Whether our data differ from previous reports due to differences in methodological procedures or potentially differences in the strain of rat employed by us and others is unclear. However, the rate-decreasing effects of zolpidem and CL218,872 in chlordiazepoxide-discriminating animals (Fig. 3, A and B) suggest that dose limitation cannot easily account for the lack of generalization to the chlordiazepoxide cue. Rather, because zolpidem and CL218,872 show affinity selectivity for GABA_A-α₁ receptors, the lack of significant generalization to the training dose of chlordiazepoxide suggests that this cue is mediated primarily via non-α₁ GABA_A receptors, whereas the zolpidem cue is primarily mediated by GABA_A-α1 receptors. Even though CL218,872 and zolpidem differ in some respects, namely that CL218,872 is generally 10-fold less potent at the GABA_A-α₁ receptor and has lower intrinsic efficacy at this receptor subtype than zolpidem (Sieghart, 1995), the two compounds have the same selective discriminative profile in the present study, suggesting that selective affinity for the GABA_A-α₁ receptor is the key determinant of the profile of both compounds.

However, concluding that the chlordiazepoxide cue is primarily mediated via non-α₁ GABA_A receptors on the basis of a null effect with zolpidem and CL218,872 is a weak argument and would be strengthened if it were based on a positive data set. In this regard, L-838,417, a compound with functional selectivity for GABA_A-α₂, -α₃, and -α₅ receptors over GABA_A-α₁ receptors, has a generalization pattern in the two discriminative cue conditions opposite to that seen with zolpidem and CL218,872. Thus, L-838,417 dose-dependently and fully generalized to the chlordiazepoxide cue (Fig. 3C), whereas it occasioned only a maximum of ∼50% zolpidem-appropriate responding (Fig. 3G). In addition, whereas 100% of animals selected the chlordiazepoxide lever at 30 mg/kg L-838,417, only 25% selected the zolpidem lever at this same dose. At the doses tested, L-838,417 had no effect on response rate in either set of discriminating animals. Because L-838,417 has no intrinsic efficacy at GABA_A-α₁ receptors (McKernan et al., 2000), this property would seem to explain its lack of generalization to the zolpidem-discriminative cue, argued to be mediated by GABA_A-α₁ receptors above. However, although this is the most parsimonious conclusion, particularly if the percentage drug-lever response data and the number of animals selecting the drug lever are considered together, it is arguable that L-838,417 does induce some zolpidem appropriate responding if the former measure is taken in isolation (i.e., ∼50% drug-lever responses; see Fig. 3G). One possibility for this apparent zolpidem-like property of L-838,417 is that the zolpidem-discriminative stimulus is not exclusively mediated via GABA_A-α₁ receptors but also has GABA_A-α₂ and -α₃ receptor components, possibly consistent with the modest 10- to 20-fold affinity selectivity of zolpidem for the former compared with the latter receptor subtypes (Sieghart, 1995). Future studies assessing the influence of different training doses of zolpidem may shed light on this issue.

Recently, Rowlett et al. (2005b) showed that zolpidem and diazepam fully generalized to the triazolam cue in squirrel monkeys, whereas L-838,417 showed no generalization up to 10 mg/kg. Although it was suggested that the reason for this was that the triazolam cue was mediated via GABA_A-α₁ receptors at which L-838,417 has no efficacy (McKernan et al., 2000), assessing that the discriminative profile of L-838,417 in animals trained to discriminate zolpidem may have been more conclusive (Rowlett et al., 1999, 2000). Nonetheless, our data with L-838,417 concur with the conclusions of Rowlett et al. (2005b) and demonstrate that its selective generalization to the chlordiazepoxide cue over the zolpidem cue mimics the receptor selectivity seen with this compound in vitro (McKernan et al., 2000).

Like L-838,417, bretazenil fully generalized to the chlordiazepoxide cue. However, in contrast to L-838,417, bretazenil showed more robust zolpidem lever responding (∼70%) and, moreover, 80% of animals selected the zolpidem lever at 10 mg/kg bretazenil, whereas a maximum of 25% L-838, 417-treated animals selected the zolpidem lever at any dose (compare Figs. 3G and 4C). Thus, despite bretazenil (like L-838,417) having low efficacy at GABA_A-α₂, -α₃, and -α₅ receptors, its weak intrinsic efficacy at GABA_A-α₁ receptors (unlike L-838,417) (Atack, 2003) seems key to occasioning zolpidem-like discriminative effects. Interestingly, the profile of SL651498 was similar to that of bretazenil. i) SL651498 fully generalized to the chlordiazepoxide cue at 10 mg/kg and ii) showed appreciable (∼80%) zolpidem appropriate responding, and (iii) ∼70% of animals administered SL651498 (3–10 mg/kg) selected the zolpidem lever. Thus, despite SL651498 and L-838,417 both possessing in vitro functional selectivity profiles (see Introduction), the greater absolute efficacy of SL651498 at GABA_A-α₁ receptors compared with L-838,417 (McKernan et al., 2000; Griebel et al., 2001) results in zolpidem-like discriminative stimulus properties. Recently, Licata et al. (2005) demonstrated that SL651498 partially generalized to the triazolam-discriminative cue in squirrel monkeys and that this effect could be blocked by the GABA_A-α₁-selective antagonist β-CCT. By contrast, as discussed above, L-838,417 does not generalize to the triazolam cue in squirrel monkeys (Rowlett et al., 2005b).

Ocinaplon apparently shows anxioselectivity in man (Lippa et al., 2005) and generalized fully to both discriminative cues. However, ocinplon has weak micromolar affinity for the [³H]flunitrazepam binding site (Chilman-Blair et al., 2003) and has a complex electrophysiology profile, whereby it shows equivalent efficacy to diazepam at GABA_A receptors composed of α₁β₂γ₂ subunits but lower efficacy when the α₁ subunit is replaced by a α₂, α₃, or α₅ subunit (most notably if these latter subunits are combined with a γ₃ subunit; Lippa et al., 2005). This in vitro profile would explain why it generalizes fully to both cues. However, the complex selectivity profile that ocinaplon possesses may impart an anxioselective profile in man, which differs from the current focus on compounds with selectivity based upon the α-subunit of GABA_A receptors, and suggests that the γ subunit may be an important determinant for generating anxioselective agents. Indeed, mice heterozygous for the γ₂ subunit of the GABA_A receptor show reduced synaptic clustering and an enhanced anxiety-like behavioral phenotype (Crestani et al., 1999).

In conclusion, the current study shows that the relative generalization patterns of benzodiazepine site ligands to chlordiazepoxide and zolpidem-discriminative cues generally corresponds with the in vitro profiles of these compounds at different GABA_A receptor subtypes described in the literature. In particular, newly available ligands such as L-838,417, with functional selectivity for α₂, α₃, or α₅ containing GABA_A receptors and no intrinsic efficacy at GABA_A-α₁ receptors, generalize fully to the chlordiazepoxide but not the zolpidem cue. This profile is opposite to that seen with zolpidem and CL218,872, which have selective affinity for GABA_A-α₁ receptors and which generalize fully to the zolpidem but not the chlordiazepoxide cue. Because the nonselective partial agonist bretazenil shows greater zolpidem-like discriminative effects compared with L-838,417, the in vivo profile of L-838,417 differs from that of a partial agonist in rodents. Furthermore, another functionally selective agonist, SL651498, with greater efficacy at GABA_A-α₁ receptors compared with L-838,417 occasions appreciable zolpidem-like discriminative effects, suggesting that absolute efficacy at GABA_A-α₁ receptors rather than functional selectivity is the important determinant in generating zolpidem-like effects. However, future studies should consider the issue of training dose, as we already allude to above, to confirm our findings. Moreover, the use of selective antagonists to qualify the conclusions that we have drawn here is important. However, as β-CCT (see Introduction) only has a modest 20-fold affinity selectivity for GABA_A-α₁ receptors over GABA_A-α₂ or -α₃ receptors, better subtype-selective antagonists are needed to fully explore the pharmacological basis of the zolpidem and chlordiazepoxide-discriminative cues (Rowlett et al., 2005a).