Introduction

Top Abstract Introduction Methods Results Discussion References

The GABA_A receptor is part of a macromolecular complex coupled to a chloride (Cl) ionophore. This complex has binding sites for a wide variety of substances from many different chemical classes including the benzodiazepines (Brioni et al., 1989; Fonnum, 1987; Guidotti et al., 1983; Mohler et al., 1987; Saano, 1984; Schwartz, 1988), barbiturates (Fonnum, 1987; Peters et al., 1988; Ticku and Rastogi, 1986) and neurosteroids (Gee et al., 1987; Harrison et al., 1987; Majewska et al., 1988; O'Connor et al., 1988; Perez et al., 1988). Accordingly, GABA_A receptor function can be modulated by different agents within each of these classes and by various endogenous ligands (Sangameswaran and De Blas, 1985). BZDs such as diazepam, triazolam and alprazolam are considered to be positive allosteric modulators by virtue of the fact that they produce an allosterically favorable conformation for GABA binding and thereby enhance Cl ion current, whereas -carboline-3-carboxylates such as -CCE and FG-7142 are considered negative allosteric modulators or inverse agonists by virtue of the fact that they produce an allosterically unfavorable conformation for GABA binding and thereby inhibit Cl ion current (Haefely, 1994; Paredes and Agmo, 1992). Unlike either the positive or negative allosteric modulators, antagonists of GABA_A receptor function (including antagonists that bind to the BZD recognition site such as flumazenil) are generally thought to have little effect on Cl channel gating.

Many of the positive allosteric modulators have been shown to be extremely efficacious in the treatment of anxiety (Woods et al., 1992). However, these same high efficacy positive allosteric modulators are also known to produce a large array of unwanted effects such as sedation, potentiation of the effects of ethanol, physical dependence and cognitive deficits (Costa et al., 1994; Woods et al., 1992). Regarding this last effect, BZDs and other positive allosteric modulators of GABA_A receptor function are known to impair central nervous system processes involved in the learning (acquisition) and memory (retention) of new information (Cole, 1986; Lister, 1991). For example, the triazolobenzodiazepines such as triazolam and alprazolam have been shown to impair learning and memory in both human and animal subjects (Bickel et al., 1990; Broekkamp et al., 1984; Decker et al., 1990; Lister, 1985; Thiebot, 1985). These same types of deficits have also been shown for other positive allosteric modulators such as thiopental and pentobarbital (Kirk et al., 1990; Moerschbaecher and Thompson, 1980; Osborn et al., 1967), which act at a site independent of the benzodiazepine recognition site. More recent findings have also shown that the partial positive allosteric modulators of GABA_A receptors produce little or no effect on learning and memory when administered alone, but block effects of high efficacy positive allosteric modulators when given in combination (Auta et al., 1995; Thompson et al., 1995). Specifically, Auta et al. (1995) found that the combination of either imidazenil or bretazenil with triazolam produced a dose-related attenuation of the disruptive effects of triazolam on two separate behavioral base lines, one involving a learning task and the other involving a memory task.

In contrast to the positive allosteric modulators, relatively little is known about the actions of the negative allosteric modulators or GABA_A receptor antagonists on learning and memory tasks. Several investigators have reported that inverse agonists enhance performance in animals (Chapouthier et al., 1984; Venault et al., 1986) and humans (Duka et al., 1987). Venault et al. (1986), for example, reported that pretraining injections of the -carboline inverse agonist -CCM increased retention of a habituation test in mice. Raffalli-Sebille and Chapouthier (1991) also reported that pretraining injections of -CCM enhanced learning of a brightness discrimination task independently of aversive or appetitive motivation. Another -carboline, ZK 93426, has been shown to block scopolamine-induced amnesia and reverse scopolamine-induced deficits in a signal-detection paradigm (Jensen et al., 1987).

Given the provocative effects reported for the negative allosteric modulators in rodents, the present study was designed to directly compare the effects of a positive allosteric modulator (alprazolam) with two inverse agonists (-CCE and FG-7142) and a hallucinogenic -carboline derivative (harmine) on the repeated acquisition and performance of conditional discriminations in monkeys. In addition, the benzodiazepine antagonist flumazenil was administered alone and in combination with both types of allosteric modulator. The same conditional discrimination task as that used by Auta et al. (1995) was used in this study to facilitate comparisons between the effects of the positive and negative allosteric modulators. An additional advantage to using this procedure is that the effects of drugs on both learning and performance can be evaluated concurrently. Responding in performance can also serve as a control for nonspecific motivational, sedative, convulsant or muscle relaxant effects of each drug.

	Methods

Top Abstract Introduction Methods Results Discussion References

Subjects. Seven adult old-world monkeys served as subjects in these experiments. Subjects I, N and G were female patas monkeys (Erythrocebus patas), whereas subjects Co, B and P were female rhesus monkeys (Macaca mulata). Subject W was a male cynomlogus monkey (Macaca fascicularis). The subjects were housed individually with free access to water, and all subjects were maintained at about 85% of their free-feeding weights on a diet consisting of banana-flavored food pellets (P.J. Noyes Company, Inc., Lancaster, NH), monkey chow, fresh fruits and vitamins. Each subject used in this study had an extensive history of responding under complex behavioral procedures and had been exposed previously to acute drug administration. However, all the subjects were drug free for at least 6 weeks before the start of the present study.

Apparatus. Several removable response panels equipped with response keys and a feeder (BRS/LVE, model TIP-002), the specific details of which have been described previously (Moerschbaecher et al., 1987), were attached to the sides of the individual cages during experimental sessions. Each response panel was connected to a computer and cumulative recorder located in an adjacent room.

Procedure. A multiple schedule of repeated acquisition and performance of conditional discriminations served as the base line for characterizing the effects of all the drugs tested. This procedure, described previously by Moerschbaecher and Thompson (1983), was used to evaluate the effects of the drugs on both the acquisition and performance of a discrimination in a single subject within a single experimental session. In each component of the multiple schedule, subjects were required to respond on a left or right key depending on the stimuli (i.e., different combinations of colors and geometric forms) displayed on the center key. Correct responses resulted in the progression to the next response in the chain in which a different stimulus combination was displayed on the center key. The completion of a chain of these discriminations was reinforced with a 500-mg banana-flavored food pellet. In the acquisition component, the stimuli that set the occasion for left- or right-key responses were changed each session, whereas in the performance component the discriminative stimuli for side-key responses were the same from session to session. Incorrect responses (errors) in both components produced a 5-sec timeout during which responding had no programmed consequences. In summary, in the acquisition component, subjects were required to learn a different discrimination during each daily session, whereas in the performance component the subject performed the same discrimination each session. Each daily session began with an acquisition component, which then alternated with the performance component after 20 food-pellet presentations or 15 min, whichever occurred first. A 5-sec blackout in which all the stimuli were off and responses had no programmed consequences separated consecutive components. Each daily session terminated after 200 reinforcers or 90 min, whichever occurred first.

Drugs. -CCE and FG-7142 were obtained from Research Biochemical International (Natick, MA). Flumazenil was graciously provided by Hoffmann-La Roche (Nutley, NJ) and harmine (7-methoxy-1-methyl-9H- pyridol[3,4-b]indole) was obtained from Sigma Chemical Co. (St. Louis, MO). Alprazolam was obtained from the Upjohn Co. (Kalamazoo, MI). Harmine was dissolved in sterile water, whereas -CCE, FG-7142 and flumazenil were dissolved in 5 to 10% dimethyl sulfoxide (depending on the concentration needed) and then diluted with a vehicle containing polyethylene glycol-400 (11%), benzyl alcohol (2%), propylene glycol (50%) and sterile water (37%). For oral administration, alprazolam was suspended in a 2% solution of Suspending Agent K (Bio. Serv. Inc. Frenchtown, NJ) in fruit punch and then mixed (volume, 0.32 ml/kg) with an additional 20 ml of fruit punch, which the subjects readily drank. All other drugs were administered intramuscularly in a volume of 0.05 ml/kg b.wt.; however, at higher doses the injection volume was increased depending on the concentration and solubility characteristics of each drug. The presession administration time for oral administration of alprazolam and flumazenil was 30 min and 15 min, respectively. When -CCE, FG-7142 and harmine were administered intramuscularly, the presession was 15 min. Intramuscular injections of flumazenil were given 5 min presession.

Data analysis. The data from both components of the multiple schedule were analyzed in terms of the overall response rate (responses per minute, excluding timeouts) and the overall accuracy or percentage errors [(incorrect responses/correct + incorrect responses) × 100]. The data for each subject were analyzed by comparing the range of variability for drug sessions with the control (vehicle) range of variability. Because each subject served as its own control, a drug was considered to have an effect to the extent that the data for a given dosage fell outside of the ranges of variability established during control sessions for that drug. Percent errors were not included in the data analysis when response rate was less than 5 responses/min because of the small number of correct and/or incorrect responses involved. In addition to these measures based on session totals, within-session changes in responding were monitored by the cumulative recorder and computer.

	Results

Top Abstract Introduction Methods Results Discussion References

The effects of alprazolam on response rate and percent errors in three subjects are shown in figure 1. Both rate and accuracy in each component for each subject were stable during base-line and control sessions. The rates of responding in both components during control sessions were generally higher for subjects N and I than for subject W. In addition, mean percent errors in acquisition tended to be higher for subjects N and I than for subject W under control conditions. In general, alprazolam produced similar dose-dependent decreases in overall response rate in both components of the multiple schedule in all three subjects. In contrast to the effects on response rate, alprazolam had a more selective effect on accuracy of responding. Alprazolam produced a dose-dependent increase in percent errors in the acquisition component in all three subjects, whereas in the performance component it had little or no effect (compare open and filled circles) except in subject N at the highest dose tested. Note also that there was some differential sensitivity to the disruptive effects on percent errors among the subjects. That is, subject N was less sensitive to the error-increasing effects than subjects I and W. For example, increases in percent errors were evident in subjects I and W at doses as low as 1 mg/kg, whereas a similar magnitude of error-increasing effect was only evident in subject N at a dose of 5.6 mg/kg.

View larger version (24K): [in this window] [in a new window]   Fig. 1.   Effects of oral alprazolam, alone and in combination with orally administered flumazenil (0.1 and 1 mg/kg), on response rate and percent errors in both the acquisition (ACQ) and performance (PERF) components of the multiple schedule in three subjects. Dose-effect data for acquisition are indicated by the open symbols, whereas data for performance are indicated by the filled symbols. Points and vertical lines at V indicate the mean and range of 11 to 14 vehicle (control) sessions. The data points with vertical lines in the dose-effect curves indicate the mean and range of one to two determinations of that dose; data points without vertical lines in the dose-effect curves indicate either a single determination of that dosage or an instance in which the range is encompassed by the point.

When alprazolam was administered in combination with flumazenil, the dose-effect curves for both response rate and percent errors were shifted to the right (0.5-1 log unit) in all three subjects. This effect is most noticeable at the higher dose of flumazenil in combination with the 5.6 to 18 mg/kg doses of alprazolam. In subject W, for example, 1 mg/kg of flumazenil completely antagonized the effect of 5.6 mg/kg of alprazolam, which was not administered alone because of the substantial effects seen at a lower dose (e.g., 1.8 mg/kg).

The effects on the within-session pattern of responding for subject I after alprazolam (3.2 mg/kg) alone and alprazolam in combination with flumazenil (0.1 and 1 mg/kg) are shown in figure 2. During a control session (top row), the discrimination was acquired during the second acquisition component, and this was characterized by a distinct decrease in the number of errors and an increase in errorless completions of the discrimination. This response pattern in acquisition at the start of the session generally accounted for the fact that the mean percent errors in acquisition for each subject were typically larger than mean percent errors in performance under control conditions. When compared with behavior under control conditions, 3.2 mg/kg of alprazolam (second row) selectively decreased response rate and increased errors in acquisition without affecting either measure in performance (compare response pattern between A and P). As shown, this dose of alprazolam completely eliminated responding in the first acquisition component and produced large error-increasing effects in subsequent acquisition components when responding did occur. These error-increasing effects were also evident when 0.1 mg/kg of flumazenil was administered in combination with the same 3.2 mg/kg dose of alprazolam (third row). However, this relatively low dose of flumazenil partially attenuated the rate-decreasing effects as indicated by increased responding in the initial acquisition component and increased responding in acquisition throughout the session. Unlike the lower dose of flumazenil, 1 mg/kg of flumazenil almost completely antagonized the rate-decreasing and error-increasing effects of alprazolam. Note that in the presence of this higher dose of flumazenil, acquisition of the discrimination was evident during the third acquisition component. As in the control record, acquisition was characterized by a decrease in errors as the session progressed and an overall response rate similar to that seen in the performance components. In general, these same effects on the within-session patterns of responding were noted for subjects W and N. 

View larger version (27K): [in this window] [in a new window]   Fig. 2.   Cumulative records showing the within-session pattern of responding for monkey I during a representative control (vehicle) session and sessions preceded by the administration of alprazolam (3.2 mg/kg) alone and a combination of this dose of alprazolam with flumazenil (0.1 and 1 mg/kg). The response pen stepped with each correct response and was deflected downward each time food was presented. Errors are indicated by the event pen (below each record), which was held down during each timeout. The event pen was deflected and the response pen reset each time the component of the multiple schedule changed. Each session began with an acquisition component (A) and then alternated with a performance component (P) after 20 reinforcers or 15 min, whichever occurred first. Each session terminated after 200 reinforcers or 90 min, whichever occurred first.

The three panels in figure 3 show the effects on overall response rate and percent errors for three subjects during the acquisition and performance components after injections of -CCE alone (top panel) and -CCE in combination with two doses of flumazenil (middle and bottom panels). As in figure 1, the mean overall response rates and percent errors for subject W under control conditions were generally lower than the mean control data for the other two subjects (G and I). Increasing doses of -CCE administered alone (open and filled circles in the top panel) generally produced dose-related decreases in overall rates of responding in both components in all three subjects. Note that these rate-decreasing effects tended to occur at lower doses in subjects G and I than in subject W. In regard to accuracy in the acquisition component, -CCE had little or no effect across all doses tested in subject W, but produced marked increases in percent errors in subjects G and I at the higher doses. This was in direct contrast to the effects of -CCE on the accuracy of responding in performance where the same doses of -CCE produced little or no increases in percent errors. This was particularly evident at the highest doses tested in each subject (e.g., 0.18 and 0.32 mg/kg). Interestingly, these higher doses of -CCE also reduced rates of responding to less than 10 responses/min in both components. Also, in one subject (monkey I), the 0.18 mg/kg dose and a 0.32 mg/kg dose produced a convulsion. On these occasions, this subject was immediately administered a dose of lorazepam, and the data were excluded from the data analysis.

View larger version (23K): [in this window] [in a new window]   Fig. 3.   Effects of -CCE alone, and -CCE in combination with flumazenil (0.32 and 1 mg/kg), on the overall response rates and percent errors in the acquisition and performance components of the multiple schedule. The points and vertical lines at V and F indicate the mean and range for 26 to 30 vehicle (control) sessions and at least two determinations for doses of flumazenil alone, respectively. Other details are the same as in figure 1.

Unlike -CCE, flumazenil (0.32 or 1 mg/kg) alone had no effect on overall response rate or percent errors in all three subjects. However, when administered in combination with -CCE, these same doses of flumazenil dose-dependently antagonized the rate-decreasing and error-increasing effects of -CCE in each subject. The dose-effect data in the bottom panel of figure 3 clearly show that 1 mg/kg of flumazenil almost completely antagonized the effects of -CCE on both the accuracy and rate of responding in both components of the multiple schedule.

The within-session pattern of responding for subject I after 0.18 mg/kg of -CCE alone, and this dose of -CCE in combination with 1 mg/kg of flumazenil, is shown in figure 4. As indicated by the response pattern in the vehicle record (top row) and the record for 1 mg/kg of flumazenil alone (third row), acquisition of the discrimination occurred a short time after the start of the session, and the pattern of responding in acquisition was similar to that seen in performance for the remainder of the session. Thus, there was generally no difference between vehicle or flumazenil administration in regard to the within-session pattern of responding. In contrast, 0.18 mg/kg of -CCE alone substantially altered the within-session pattern of responding in both acquisition and performance. These effects of -CCE were characterized by high initial rates of responding followed by a decrease and then a cessation of responding in both components. This figure also illustrates that the same dose of flumazenil (1 mg/kg) that failed to produce a behavioral effect when given alone, almost completely antagonized the effects of this dose of -CCE when the two drugs were administered in combination. These same effects on the within-session pattern of responding were obtained in subjects W and G after administration of -CCE alone and -CCE in combination with flumazenil.

View larger version (26K): [in this window] [in a new window]   Fig. 4.   Cumulative records showing the within-session pattern of responding for monkey I during a representative control (vehicle) session and sessions preceded by the injection of -CCE (0.18 mg/kg) and flumazenil (1 mg/kg) alone and a combination of these two doses. Other details are the same as in figure 2.

The effects of FG-7142 on both acquisition and performance are shown for three subjects in figure 5. Although the mean percent errors in subjects Co and B were generally higher than those for subject P, the drug effects obtained were consistent in all subjects. Similar to -CCE, FG-7142 dose-dependently (but 100-fold less potently) decreased response rate in both components of the multiple schedule, but unlike -CCE, it did not produce an associated error-increasing effect at doses that substantially decreased overall response rate. These rate-decreasing effects produced by FG-7142 were, in turn, antagonized by a 1 mg/kg dose of flumazenil. Note that this dose of flumazenil shifted the FG-7142 dose-effect curve 1/4 log-unit to the right even though this dose of flumazenil produced small rate-decreasing effects in these subjects when administered alone (see the data at F). This shift in the dose-effect curve could not be determined in subject B due to the difficulty in solubilizing and administering doses greater than 18 mg/kg. However, 1 mg/kg of flumazenil in this subject completely antagonized the effects of the 18 mg/kg dose of FG-7142.

View larger version (25K): [in this window] [in a new window]   Fig. 5.   Effects of FG-7142 on the overall response rate and percent errors in each component of the multiple schedule for three subjects. The points and vertical lines to the left of each dose-effect curve represent the mean and range of 19 to 28 vehicle sessions and 5 to 8 determinations of flumazenil alone.

The effects of harmine on overall response rate and percent errors in each component are shown for three subjects in figure 6. In all three subjects, harmine dose-dependently and uniformly decreased response rates in both components of the multiple schedule. On accuracy of responding, harmine had little or no effect on percent errors in any of the three subjects. The effects on both overall response rate and percent errors were similar to that found with FG-7142 in that harmine failed to increase errors even at higher doses that substantially decreased rates of responding in both components. Unlike the effects of FG-7142, as well as the effects of -CCE and alprazolam, the rate-decreasing effects of harmine were not antagonized by a 1 mg/kg dose of flumazenil. In subjects B and P, for example, flumazenil failed to antagonize the effects of a 1 mg/kg dose of harmine.

View larger version (22K): [in this window] [in a new window]   Fig. 6.   Effects of harmine on the overall response rate and percent errors in each component of the multiple schedule for three subjects. The points and vertical lines to the left of each dose-effect curve represent the mean and range of 13 to 25 vehicle sessions. The determinations for flumazenil alone are the same as those shown in figure 5. Other details are the same as in figure 1.

	Discussion

Top Abstract Introduction Methods Results Discussion References

The multiple schedule of behavior provided a stable base line with which to examine the effects of each drug on the acquisition (learning) and performance of conditional discriminations, and proved sensitive to the ability of both positive and negative allosteric modulators of GABA_A receptors to differentially affect measures of rate and accuracy within each behavioral component. A consistent finding obtained in the present study concerned the rate-decreasing and error-increasing effects observed when alprazolam was administered alone. The dose-dependent disruptive effects of alprazolam on overall response rate and percent errors in acquisition were consistent with previous research concerning the effects of the high-efficacy BZDs on learning and memory procedures in animals and humans (Auta et al., 1995; Thiebot, 1985; Thompson et al., 1995; Woods et al., 1992). Moreover, the selective error-increasing effects produced in acquisition with alprazolam were similar to those found for another high-efficacy positive allosteric modulator of the GABA_A receptor, triazolam, on an identical base line of repeated acquisition and performance of conditional discriminations in monkeys (Auta et al., 1995). In that study, the effects of triazolam were attenuated by either of two partial positive allosteric modulators, imidazenial or bretazenil, when they were administered in combination with triazolam. Thus, data from the present study both replicate and extend these findings by showing that both the rate-decreasing (acquisition and performance) and error-increasing (acquisition) effects of alprazolam could be antagonized dose dependently by flumazenil. Furthermore, these findings are consistent with previous research showing that the amnestic effects of BZD agonists such as diazepam, lorazepam and midazolam are blocked by flumazenil (Ghoneim et al., 1989; McKay et al., 1990; O'Boyle et al., 1983).

One purpose for examining the effects of two negative allosteric modulators of GABA_A receptors under this repeated acquisition of conditional discriminations base line in old world monkeys was to provide a direct comparison with the effects reported for the positive allosteric modulators (Auta et al., 1995). This was particulary important because of the reports which indicated that some inverse agonists enhanced cognition in rodents in several different experimental paradigms used to investigate the effects of drugs on learning and memory. For example, Venault et al. (1986) reported that the inverse agonist -CCM could increase retention in a habituation test in mice. Similarly, other investigators have reported that specific inverse agonists could enhance the learning of a brightness discrimination task in mice (Raffalli-Sebille and Chapouthier, 1991), improve recognition performance of rats after central administration (Mayo et al., 1992) and improve performance of rats in a passive-avoidance task (File and Pellow, 1988; Holmes and Drugan, 1991). Based on these prior results, the suggestion that the inverse agonists might improve learning and memory in monkeys would not have seemed unreasonable. However, the present study found that two inverse agonists (-CCE and FG-7142) dose-dependently decreased rates of responding while either disrupting or having little effect on accuracy of responding.

The present results in old world monkeys also contrast with a result obtained in humans. In a study conducted by Duka et al. (1987), the -carboline ZK 93 426 was found to improve performance in two cognitive tasks, a logical-reasoning task and a picture-differences task. However, the effects obtained on the logical-reasoning task only indicated a nonsignificant trend toward improvement, and there were some data to indicate that the three groups tested (i.e., placebo and two dose groups) may have had differing performance levels before drug testing. Although the effects reported on the picture-differences task were significant, the data collected for this test were extremely limited in scope. More specifically, no testing was done with this particular task before drug testing to establish comparability among the groups, and the authors only administered this task at one time point after drug administration. Because of the limited nature of the data reported, along with several other important methodological differences (i.e., the drugs themselves, the route of administration and the fact that these authors used solely performance tasks), it is difficult to make any conclusive statements concerning the conditions under which the negative allosteric modulators may or may not facilitate learning and memory. Certainly, under the conditions of our experiment, seeing an improvement in performance would have been difficult because of the already low levels of errors in this highly trained task. However, our purpose for using the multiple schedule of conditional discriminations in old world monkeys was to facilitate direct comparisons between the data obtained here with the negative allosteric modulators and previous data collected with several positive allosteric modulators on the same procedure (e.g., Auta et al., 1995).

Although the effects of -CCE and FG-7142 on response rates in both components were qualitatively similar to each other and to several other drugs tested, they were quantitatively different from each other. That is, -CCE was found to be approximately 100-fold more potent on a milligram per kilogram basis than FG-7142. This relatively low potency exhibited by FG-7142 in our study was somewhat unexpected because of the reported similarity in their discriminative stimulus properties (Rowan and Lucki, 1992), anxiogenic properties (Thiebot et al., 1988) and their potency in vitro for modulating GABA-induced chloride current (Yakushiji et al., 1989). Even more surprising were the differences found on the accuracy of responding in acquisition after the administration of the higher doses of each drug. More specifically, -CCE produced increases in percent errors at doses that substantially decreased response rates, whereas FG-7142 produced no increases in percent errors with the same magnitude of rate-decreasing effect. These results obtained largely with subconvulsant doses of each drug suggest that the inverse agonists may be similar to the positive allosteric modulators (and other drugs) in terms of their effects on rates of responding, but dissimilar to the positive allosteric modulators in terms of their ability to disrupt accuracy of responding in a learning task. Although very provocative, a definitive explanation for these differences would be well beyond the scope of this paper and premature given the limited amount of experimental data on the effects of both types of modulator on complex behavioral processes. For example, the different behavioral effects obtained with both types of modulator could occur as a result of the differing distributions of GABA_A receptors across the many regions of the brain that subserve vastly different functions (e.g., motor control versus memory). Whereas the differences found between the two negative allosteric modulators could result from differences in the specificity with which each of these inverse agonists binds to the various forms of the receptor, which comprise different subunits (e.g., BZD₁ versus BZD₂). In any event, further molecular and behavioral studies with both types of modulator will be required to provide more explicit explanations for the observed behavioral effects.

Flumazenil, up to doses that produced disruptions in response rates (subjects Co and P) and increases in percent errors (subject Co), dose-dependently antagonized the rate-decreasing effects of -CCE and FG-7142 and the error-increasing effects of -CCE. The disruptive effects observed after the 1 mg/kg dose of flumazenil in subjects Co and P were somewhat surprising in that the same effects on rate were not observed in the other subjects, and even in these subjects this dose did not consistently produce disruptive effects in both components. Interestingly, when rate-decreasing effects were obtained (at least in one subject, monkey P), they tended to occur toward the end of the session in a pattern not unlike that observed with doses of -CCE alone (see cumulative record in fig. 5). There is some existing experimental evidence to suggest that flumazenil may have some properties similar to those of the inverse agonists. Rowan and Lucki (1992), for example, found that the stimulus properties of FG-7142 and -CCE partially generalized to the stimulus properties of a training dose of flumazenil in a study involving a discriminated taste-aversion procedure. File and Pellow (1985) also demonstrated that flumazenil was capable of producing anxiogenic effects similar to those seen with the inverse agonists in several animal tests of anxiety. Certainly, the data from this study are insufficient to suggest that flumazenil's effects on complex behavioral processes may be similar to those of certain inverse agonists. Only further research with the GABA_A receptor antagonists on learning and memory procedures can answer these questions. What the present data do suggest, however, is that GABAergic mechanisms are involved in the behavioral effects produced by alprazolam, -CCE and FG-7142 under the present experimental conditions.

The hallucinogenic -carboline derivative harmine (Naranjo, 1967) also decreased rates of responding in both components in a dose-related manner with little or no effect on accuracy of responding. Furthermore, harmine was found to be 3 times less potent on a milligram per kilogram basis than -CCE at decreasing response rates. Flumazenil, however, failed to antagonize the effects of harmine, which suggested an action at nonbenzodiazepine receptor sites. Although the behavioral effects and the mechanism(s) of action of harmine have not been well studied, it's effects on rates of responding are qualitatively similar to those reported for the prototype hallucinogen LSD (Berryman et al., 1962; Nielsen and Appel, 1983; West et al., 1982). Because it has generally been accepted that the effects of LSD are mediated via the 5-HT₂ receptor, one could speculate that harmine may be producing its effects via a 5-HT₂ receptor-mediated mechanism. However, further studies need to be done to elucidate the mechanism(s) by which harmine produces rate-decreasing effects.

In summary, negative allosteric modulators of GABA_A receptors (-CCE and FG-7142) produce effects on rates of responding in acquisition and performance that are qualitatively similar to those produced by a positive allosteric modulator (alprazolam) and by a -carboline derivative not thought to modulate GABA_A receptors (harmine). In contrast to the effects on rate of responding, the accuracy data indicated that the negative allosteric modulators were less disruptive to responding than the positive allosteric modulator, which markedly disrupted the acquisition of conditional discriminations. This was particularly true for the negative modulator FG-7142, which did not decrease accuracy even at doses that substantially decreased overall response rate. Despite the differences in their effects on the accuracy of responding, however, the effects of both types of allosteric modulator were most likely mediated through a benzodiazepine binding site on the GABA_A receptor, because they both produced effects that were dose-dependently attenuated by the benzodiazepine antagonist flumazenil. Moreover, because neither -CCE nor FG-7142 was observed to increase accuracy or enhance acquisition, these data involving a complex behavioral procedure and old world monkeys failed to support previous data obtained with rodents showing that the negative allosteric modulators are capable of enhancing cognitive processes.