Psychopharmacology (1986) 88:460-466 Psychopharmacology © Springer-Verlag 1986

Effects of chlordiazepoxide on cued radial maze performance in rats

Helen Hodges and Simon Green Department of Psychology, Birkbeck College, Malet Street, London WC1E 7HX, UK

Abstract. Effects of chlordiazepoxide (CDP) were examined on the performance of rats in an eight-arm radial maze with four cued and food-baited arms. Two conditions were used; random, with cue location varying over trials, and constant, with the same subset of arms consistently cued. In rats pre-trained to a 60-70% efficiency level (Rewarded entries/Total entries x 100), the effects of CDP (2.5, 5.0 and 10.0 mg/kg, IP) differed according to condition. Effi- ciency was substantially reduced in the random condition, and types of error were undifferentiated. In the constant cue condition the post-drug drop in efficiency was less marked, and errors were selectively those of re-entry into rewarded arms. In both conditions there was a high inci- dence of error clusters involving re-entries into both rew- arded and non-rewarded arms in the random condition, and rewarded arms in the constant condition. The results suggested that CDP induced a general disruption of infor- mation processing rather than a specific impairment of working memory, together with some response persevera- tion which could occur in the presence or absence of dis- crimination failure.

Key words: Chlordiazepoxide - Radial maze - Behavioural inhibition system - Working memory - Reference memory

Recent studies suggest that, in man, impaired recall of mate- rial learned under benzodiazepine (BZ) treatment reflects impoverished encoding rather than short- or long-term memory storage deficits (Brown etal. 1983; Ghonheim et al. 1984). With animals, BZs do not impair well-learned tasks (Vachon et al. 1984), but they do disrupt the learning and performance of complex tasks such as conditional or successive discriminations (Iwasaki et al. 1976; Hasegawa et al. 1973 ; Cole 1982, 1983). These tasks involve both at- tention to cues, and their retention over a period of time. Thus they tap information encoding and short-term memo- ry storage, both of which may be vulnerable to BZs.

Gray (1982) has proposed that BZs impair the analysis of complex stimuli, and relates this information encoding deficit to the disruption of a septohippocampaUy-based be- havioural inhibition system. Thus, when attention to novel or aversive events is impaired, mismatches between actual and expected events are not detected, and the inhibition system is not alerted to suppress ongoing behaviour. In

Offprint requests to: S. Green

Gray's view, therefore, BZs may act as functional hippo- campal lesions, producing effects on learning and memory that are comparable to actual lesions.

Olton et al. (1980), on the other hand, have suggested that the hippocampus is concerned with working memory. Reference memory stores unchanging aspects of tasks, such as reward locations, or cue-food associations, whilst work- ing memory is concerned with transient information of short-term relevance, such as which arms of a maze have already been visited within a particular trial. Olton and Papas (1979) found that fornix-fimbria lesioned rats could learn to avoid a subset of eight non-rewarded arms within a 17-arm radial maze, but showed a persistent increase in re-entries into positive (rewarded) arms from which food had already been taken. This specific error pattern impli- cated the hippocampus in working but not reference memo- ry. Olton's account resembles the traditional distinction be- tween short- and long-term memory stores. It is also very close to Rawlins' (submitted for publication) view that the hippocampus acts as an intermediate-term memory store, to integrate temporally discontiguous information. Thus from Olton's evidence, if BZs affect hippocampal function, they would be likely to disrupt short-term but not long-term memory storage, whereas Gray's theory would predict at- tentional deficits that are sensitive to stimulus complexity, and which could have effects on both short- and long-term memory.

The present experiments examined effects of the BZ chlordiazepoxide (CDP) on the performance of partially- trained rats in an eight-arm radial maze with four cued and food baited arms, to see whether these were more con- sistent with attentional or working memory hypotheses. Two cue conditions were used; random, where the cue loca- tion was changed before each trial, and constant, where the same subset of arms was consistently cued and rewarded for each rat. In the random condition a consistent cue-food association was available, but the rats would have to relo- cate the cues on every trial. In the constant condition cue- food location associations were consistent across trials, per- mitting a stable long-term spatial strategy. Thus the random condition would require close attention to cues, and involve within-trial acquisition of the maze layout, whereas the con- stant condition would provide redundant information and permit response repetition. If BZs disrupt attention, as Gray suggests, error rates would be expected to be higher in the random than the constant condition, but no specific pattern of errors would be predicted. If, on the other hand, working

461

memory is specifically disrupted, the rats should be able to use the long-term cue associations, in both conditions, but within-trial memory of their own movements would suffer. Thus in both conditions Olton's theory would pre- dict an increase in re-entries into positive arms, which would be more marked in the random condition where within-trial demands would be greater.

These conditions do not dissociate information encod- ing from short-term memory storage processes directly, since working memory, as measured by maze choices, in- volves both. However deficits restricted to recent informa- tion (the rat 's own movements) across differing levels o f task complexity would suggest that short-term storage is involved, whereas a high incidence o f errors in the random but not the constant condition would implicate attentional processes related to information encoding.

Materials and methods

Animals. Seventy-two male hooded rats were used (OLAC), 36 in each condition, with a mean weight of 345.1 __4.6 g (Experiment 1) and 352.6 g ± 4 . 7 g (Experiment 2). The rats were housed two to a cage, on a 12-h reversed light/dark cycle, reduced to 85% of their free feeding weight and fed a measured amount of food daily to allow for a growth rate o f around 3 g per week.

Apparatus. The eight-arm radial maze was made of white painted plywood, with arms 34 cm long and 7 cm wide equ- ispaced around a central octagonal platform 34 cm in diam- eter. The maze was 50 cm above ground level, and lit by a 60-W ceiling light directly above the center. A hollow at the end of each arm contained the 45-rag food pellets (Camden Instruments) used for rewards. Intra-maze cues consisted of 2 in. wide strips of coarse sandpaper stuck across selected arms 2 in. from the platform with blutack. Extra-maze cues were provided by the room furniture (fridge, desk, cupboards, sink etc) close to the ends o f the various arms.

Training and testing. Rats were first familiarized with the maze, with food at the end of each arm, till they moved freely about it. For most rats this involved one 20-rain ses- sion, but eight rats (three in Expt 1 and five in Expt 2) required two sessions. Thereafter training consisted of five consecutive trials, on alternate days, until the rats reached a criterion level of efficiency of 60.70%, which was the percentage of rewarded entries within total entries over a block of five trials. This low criterial level of efficiency was chosen to increase the probability of finding drug effects in a partially learned task, as BZs have been reported not to disrupt well-learned behaviour, including radial maze exploration (Hiraga and Iwasaki 1984).

For Experiment 1 (random cue condition) five different patterns of location of the four sandpaper strips were used, given in random order for each rat on the five trials of each block. There were never more than two strips adjacent. For Experiment 2 (constant cue condition) each rat received only one of the five patterns. In both experiments, for half the rats the cue signalled the presence o f food at the end of that arm, and for the remainder the non-cued arm was rewarded. There was no difference in speed of learning be- tween the groups where the cue was positive or negative. Thus there were ten different cue-reward combinations dis-

tributed across animals to control for possible odour trails, differences in cue-reward associations, or varied salience of extra-maze cues.

At the beginning of each trial rats were placed on the central platform and allowed to explore freely. Full entries to the end of each arm were recorded by arm number, and errors scored as follows: reference memory (RM) for first entries into non-rewarded arms, working memory (WM) for re-entries into positive arms, and non-specific (NS) for re-entries into non-rewarded arms, as these errors involve both associative and within-trials elements.

During drug testing it became apparent that errors were not evenly distributed over trials, but occurred in clusters resulting in one to three poor trials, as CDP-treated rats did not search and hesitate, but continued to run. The large number of re-entries suggested that error clusters might pro- vide a measure of response perseveration. Accordingly, a cluster was defined, post-hoc, as four or more consecutive errors, being the largest number that could occur without any re-entries or response repetition. Within-cluster errors were scored either as first entries (into non-rewarded arms) or re-entries (of either type).

A trial ended as soon as the rat reached the last reward. The total time taken for each trial, recorded by stopwatch, was summed over trials and divided by the total number of arm entries as a rough estimate of mean running speed/ block. Rats were placed in a side cage after each trial, whilst the maze was wiped, food replaced and cues altered if neces- sary during an inter-trial interval of 2 min. Error scores of each type were summed over five trials for analysis. Per- cent efficiency (rewarded entries/total entries X 100), the number of error clusters/block, and within-cluster error types were also calculated.

Training was stopped after the rats reached an efficiency level of 60-70%, a criterial range broad enough to include all animals within the same number of training blocks. Rats in the random condition needed five blocks and those in the constant condition four blocks to reach criterion. Test- ing followed the same procedure as training, and took place between 2.0 and 6.0 p.m. Rats were semi-randomly assigned to one of four treatments (control and three doses of CDP) on the basis of criterion efficiency to spread performance levels across groups.

Treatment. Chlordiazepoxide H C L (CDP: donated by Roche products, Welwyn Garden City) was dissolved in isotonic saline at concentrations of 2.5, 5.0 and 10.0 mg/ml and injected IP at 1.0 ml/kg I h before testing, when con- trols received saline.

Data analysis. The following measures were analysed by one between (treatment: four levels) and one or two within (blocks o f trials, types of error) analyses o f variance: t. Percent efficiency over all blocks of trials; 2. Type of error (RM, NS, WM) on the first, criterion and test blocks of trials; 3. The number of error clusters of four or more consecutive errors on criterion and test blocks; 4. Type of error within cluster (first or re-entries); 5. Mean running speed on criterion and test blocks.

Planned comparisons of means o f CDP-treated groups with controls and of treatment and criterion performance were made by t ratio tests, using the appropriate error term from the ANOVA. Simple main effects were also analysed for significant interactions (Kirk 1968).

462

Correlations between total time in maze and total errors, and between running speed and errors were also calculated.

Results

Percent efficiency over all blocks o f trials (see Figs. 1 and 2)

Percent efficiency, random cue condition. Percent efficiency in the random condition remained static for the first three blocks, then improved sharply, to reach criterion by block 5. Controls performed well on test block 6, but effi- ciency dropped dramatically in all three CDP groups, down to first block levels. A group x blocks interaction [F (15, 160)= 3.33, P < 0.001] showed that the significant dif- ference between groups [F (3, 22)=6.51, P<0.002] oc- curred only on block 6 after treatment [simple main effect: F (3, 32)=20.19, P<0.001] and that there were no differ- ences between groups at any other time.

t-Ratio comparisons with controls showed that CDP dose-relatedly impaired % efficiency, beyond the 0.001 sig- nificance level [CDP 2.5: t (32)=5.33; CDP 5.0: t (32) =6.02; CDP 10.0: t (32)= 10.0]. Only the control group showed marked improvement from block 1 to block 6. In the CDP groups, on the other hand, test performance was substantially below criterion level beyond the 0.001 in all comparisons [CDP 2.5: t (32)=4.60; CDP 5.0: t (32)=5.33; CDP 10.0: t (32)=7.03].

7 0 -

>- U 6 0 - Z Ld £)

h h I..d 50-

4 0

C O N T R O L

. . . . C D P 2 - 5

. . . . . . . . C D P 5 . 0

. . . . . . . . . C D P 1 0 - 0

N = 9

,7 . / ] ,,

• " - . I I

i I I

1 2

T R A I N I N G

K i t

7,'; i 'I,,

~/ ~iiiiilt ***l

7,';i 'I,,

• .~ Ice

i : , ,

I ,4 5 6

C R I T E R I O N T E S T

B L O C K S O F T R I A L S

Fig. 1. Random cue condition. Mean % efficiency (+ SEM) over all blocks of trials. CDP was given to three groups prior to testing on Block 6. Comparison of CDP groups with controls on Block 6: ***P<0.0005. Comparison of CDP groups with criterion perfor- mance on Block 5:*°°P<0.0005

7 0 -

Percent efficiency, constant cue condition. In the constant condition % efficiency improved smoothly and rapidly over blocks, so that criterion was reached by block 4. In contrast to the random condition, there were no overall differences >_

between groups, and the reduction in % efficiency in CDP z u 6 0

groups on block 5 (see Fig. 2) was not pronounced enough _ to yield a blocks x groups interaction. U_

Nevertheless, planned t-ratio comparisons with controls showed that performance in all CDP groups on block 5 ~ 50 was inferior, beyond the 0.005 probability level [CDP 2.5: t (32)=2.58; CDP 5.0: t (32)=2.88; CDP 10.0: t (32) = 3.48]. Percent efficiency in the two higher dose CDP groups was also below criterion level [CDP 5.0: t (32)=1.70, P<0.05; CDP 10.0: t (32)=2.34, P<0.025], 40 though all CDP groups showed improvement over block 1 performance, beyond the 0.005 level, and % efficiency re- mained close to the 60% level.

Types o f error on the first, criterion and test blocks o f trials

Error types, random cue condition. In the random condition there was a significant difference between groups [F (3, 32) = 6.05, P < 0.005], but analysis of simple main effects for the groups x blocks interaction [F (6, 64)=4.27, P < 0.005] showed that this difference occurred only on the test block, under drug [simple main effect: F (3, 32)= 4.79, P<0.01]. Overall, reference memory (RM) errors predo- minated, with rats still making a mean of eight to nine of first entries into non-rewarded arms per block, at criteri- on. By this time re-entries into non-rewarded arms (NS) had almost disappeared and working memory errors of pos- itive arm re-entry (WM) had been reduced to one to two per block. However, on block 6 in CDP groups RM and

S / ii:

!V / oo.TRo,

' . . . . . . . . C D P 1 0 " 0

N = 9

I I I I I

1 2 3 4 5 T R A I N I N G C R I T E R I O N

TEST

B L O C K S O F T R I A L S

Fig. 2. Constant cue condition. Mean % efficiency (+_ SEM) over all blocks of trials. CDP was given to three groups prior to testing on Block 5. Comparison of CDP groups with controls on Block 5 **P<0.005. Comparison of CDP groups with criterion perfor- mance on Block 4. e'P<0.025, eP<0.05

NS errors were roughly equivalent to initial rates, and WM errors were higher. Thus there were highly significant differ- ences between error types [F (2, 64)=258.79, P<0.001], and between blocks [F (2, 64)=33.03, P<0.001], and a substantial interaction [F (4, 128) = 7.53, P < 001].

Within-cell comparisons of means showed that CDP groups made significantly more errors than controls, for all three types (see Table 1). For RM errors, averaging

463

Table 1. Experiment 1 : random cue condition. Types of error on first, criterion and test blocks of trials; t-ratio comparisons with controls

Group Block 1 Block 5 Block 6 (N= 9)

RM NS WM RM NS WM RM NS WM

Control 15.7_+1.0 6.4_+1.6 4.8_+1.6 8.4_+1.5 0.1_+0.1 1.6_+0.6 8 . 3 _ + 1 . 1 0.3_+0.3

CDP2.5 16.4___1.1 6.9-+2.0 7.0-+2.1 9.2_+0.7 0.1-+0.1 2.0_+0.6 16.3___0.8"** 3.6_+1.3" t, 32=4.44, t, 32=1.83, P<0.00I P<0.05

CDP5.0 16.3_+0.6 6.6__.1.4 5.6_+1.4 9.2-+0.9 0.3-+0.3 2.1_+0.6 14.6-+1.1"** 6.9_+2.1"* t, 32=3.50, t, 32=3.67, P<0.01 P<0.01

CDP 10.0 15.8-+0.8 5.9_+1.3 4.8_+0.8 8.2±0.8 0.4_+0.2 1.9_+0.5 16.6_+0.9"** 8.8_+2.6*** t, 32=4.60, t, 32=4.72, P<0.001 P<0.001

1.2_+0.5

5.3_+2.1" t, 32 = 2.28, P < 0.025

9.4_+2.7*** t, 32=4.56, P<0.001

11.8__3.0"** t, 32 = 5.89, P < 0.001

Within group comparison of errors on blocks 5 and 6: * P < 0.05, ** P < 0.01, *** P < 0.00t

a round 16/block, this effect was not dose-related, but for NS and W M errors increase with dosage was marked. C D P groups also made substant ial ly more errors of all types on block 6 than they had done on block 5.

Compar i son of error types in the first and test blocks indicates some interesting differences between drug and naive performance. All CDP animals made as many R M errors in the test as in the first block, indicat ing a complete loss of the cue-food association. With the two higher doses NS errors were also equivalent in first and test blocks, sug- gesting that in addi t ion to not using the cue, the rats were not demarcat ing non-rewarded arms within trials. In the CDP 2.5 group NS errors did show a reliable decrease rela- tive to the first block [t (32)= 1.83, P<0 .05] , but this was less marked than with the control group [t (32)=3.39, P < 0.01] and represents the only improvement shown by any CDP group over block 1 performance. With respect to W M errors, the low dose CDP group showed no improvement on first trial rates, while with the higher doses W M errors were significantly increased under drug [CDP 5.0: t (32)=2.11, P < 0 . 0 2 5 ; CDP 10.0: t (32)=3.89, P<0 .001] , and they were also more numerous than NS errors on block 6. This preference for the positive arms indicates, par- adoxically, some kind of memory for their significance and location.

Error types, constant cue condition. Pre-test error pat terns in the constant condi t ion were very similar to those found with the r andom condit ion, with R M errors initially averag- ing 14-15/block, and dropping to six to seven by criterion, NS errors d ropping out, and W M errors falling from six to seven to two to three. On test block 5 there was a marked increase in W M errors only in the CDP groups. Thus, there was no overall difference between groups, but a groups x blocks interact ion [F (6, 64)=2.39, P<0 .05] . Simple main effects showed that the difference between groups [F (3, 32) = 3.57, P < 0.025] occurred only on test b lock 5. Differ- ences between types of error, and decline in errors over blocks, were shown by highly significant main effects [Error type: F (2, 64) = 88.75, P < 0.001 ; Blocks: F (2, 64) = 89.90, P<0.001] , but there was also a massive interact ion [F (4, 128) = 31.15, P < 0.001]. This occurred because R M and NS errors declined over trials, but W M errors showed a marked increase on block 5. Hence, though R M errors were far

more frequent on blocks 1 and 4 (P<0.001) , on block 5 the posi t ion was reversed, with a greater number of W M than R M errors It (64) = 3.62, P < 0.001].

Within-cell comparisons o f means with controls showed that CDP groups were only impai red with respect to W M errors. There was no increase in R M or NS errors (only two rats re-entered a non-rewarded arm on block 5). CDP groups also showed an increase above cri terion only for W M errors (see Table 2), but did not show the marginal improvement in R M errors found in the control group (P < O.O5).

Compar i son o f error types in the first and test blocks showed that all groups significantly reduced their error rates for R M and NS errors (beyond the 0.001 probabi l i ty level in most comparisons) . This was not the case for W M errors, which were reduced in the control group [t (32)= 3.76, P < 0.001], but increased in CDP groups above first block levels. F o r CDP 5.0 group this increase was significant [t (32) = 2.03, P < 0.05]. Thus in contras t to the indiscriminate error pat tern found in the random condit ion, CDP-t rea ted rats in the constant condi t ion showed a highly selective increase in re-entries into positive arms.

Number o f error clusters and types o f error within clusters (first or re-entry) on criterion and test blocks

Clusters and within-cluster errors, random cue condition. In the random condi t ion the number of error clusters increased substantial ly from block 5 to block 6 [F (1, 32)=9.98, P < 0.005]. This increase was specific to CDP groups, so that in addi t ion to the difference between groups [F (3, 32) = 4.07, P < 0.025] there was a groups x blocks interac- t ion [F (3, 32)= 2.92, P < 0.05]. Simple main effects showed that group differences occurred on block 6 only IF (3, 32)=6.76, P<0 .01] , where, by t-ratio comparisons the CDP 5.0 and 10.0 groups differed significantly from con- trols (see Table 3).

Within-cluster errors in the r andom condi t ion consisted chiefly of re-entries which significantly increased on block 6 in the CDP groups only [F (1, 32)=15.12, P<0.001] . Hence, there was a significant difference between groups [F (3, 32)=3.85, P<0 .025] and interactions between groups x blocks [F(3, 32) = 3.41, P < 0.05], between blocks x

464

Table 2. Experiment 2 : constant cue condition. Types of error on first, criterion and test blocks of trials: t-ratio comparison with controls

Group Block i Block 4 Block 5 (N= 9)

RM NS WM RM NS WM RM NS WM

Contol 14.4±0.6 7.0±1.7 6 .6_+1 .1 6.6±1.4 0.4_+0.4 2.6±0.6 3.9_+1.1' 0 3.2±0.9

CDP 2.5 13.7___1.5 5.1±1.8 6.6__1.0 6 . 6 _ + 0 . 5 0.1+_0.1 3.3±0.9 5.8_+0.9 0 8.9+_2.3*** t, 32=3.74, P<0.001

CDP 5.0 12.3 ± 1.5 4.1_+0.8 7.1± 1.3 6.3 ±0.8 0.7±0.3 2.7±0.6 5.0± 1.1 0 10.2±2.1"** t, 32=4.59, P < 0.001

CDP 10.0 15.1_+0.5 7.8_+1.9 6.3±0.9 7.3_+0.9 0 2.6-t-1.0 6 . 1 _ + 1 . 6 0.6±0.3 8.3_+2.2*** t, 32=3.34, P<0.001

Within group comparison of errors on blocks 4 and 5 ***P<0.001, *P<0.05

Table 3. Experiment 1 : random cue condition : error clusters. Mean number of error clusters (4 + successive errors) and type of within- cluster errors [first entry into non-rewarded arms (RM) and all re-entries (WM + NS)] on Blocks 5 and 6: t-ratio comparison of CDP- treated groups and controls

Group Block 5 Block 6 (N=9)

N of Type of error N of dusters Type of error clusters

RM Re-entry RM Re-entry

t, 32= 4 3 2 = t, 32=

Control 0.7±0.3 2.1±1.0 1.0±0.5 0.3±0.2 1.3±0.7 0.0 CDP2.5 0.4±0.2 1.7±0.7 0.2±0.2 1.3±0.4 NS 3.4±1.2 NS 5.8±2.6 2.12<0.025 CDP5.0 1.1±0.4 3.0±1.1 2.4±0.9 2.3±0.6 1.79<0.05 3.2±1.3 NS 13.7±4.5 5.0 <0.005 CDP10.0 0.9±0.5 2.8±1.4 1.0±0.7 3.3±1.0 3.59<0.005 5.7±1.9 NS 16.3±5.2 5.94<0.005

error type [F (3, 32) = 21.55, P < 0.001], and a triple interac- tion [F (3, 32)=4.25, P<0.02] . Simple main effects for the triple interaction confirmed that differences between groups were found only for re-entry errors on block 6 [F (3, 32)=14.80, P<0.001] and that reliable increases over blocks occurred only in CDP groups for re-entries, where they were dose-related [CDP 2.5 : F (1, 32) = 4.26, P < 0.05 ; CDP 5.0: F (1, 32)--17.39, P < 0 . 0 0 1 ; CDP 10.0: F (1, 32) = 32.46, P < 0.001]. By within-cell comparisons of means CDP groups differed from controls with respect to re-en- tries on block 6 only (see Table 3). The two higher doses of CDP did not differ, but both showed substantially more re-entries than the CDP 2.5 group (P<0.005).

Clusters and within-cluster errors, constant cue condition. The number of clusters and preponderence of re-entries within clusters in the constant condition was very similar to the random condition data. The number of clusters rose sharply from criterion to test blocks [F (1, 32)= 37.36, P < 0.001] with the difference between groups [F (3, 32) = 3.36, P < 0.05] resulting from increases in CDP groups on block 5 [Groups x blocks interaction: F (3, 32)=4.88, P < 0 . 0 1 ; Simple main effect at block 5: F (3, 32)=8.06, P<0.005] . As in the random condition, the two higher dose groups differed significantly from controls for re-entries on the test block (see Table 4).

Within-cluster errors in the constant condition, as in the random, were chiefly those of re-entry [F (1, 32)= 23.13,

P<0.001] which showed a marked increase over blocks [F (1, 32)= 37.33, P < 0.001] only within CDP groups. Thus there were differences between groups [F (3, 32) = 3.47, P < 0.05], interactions between groups x blocks [F (3, 32)= 4.53, P<0.01] , error types x blocks [F (1, 32)= 22.52, P < 0.001], and a triple interaction [F (3, 32)=2.97, P < 0.05]. Simple main effects for the triple interaction showed that the differ- ence between groups occurred only on block 5 for re-en- tries, and that increases over blocks were found only in CDP groups for re-entries, though they were not dose-re- lated [CDP 2.5: F (1, 32)=18.11, P<0 .001 ; CDP 5.0: F (1, 32)=38.96, P < 0 . 0 0 1 ; CDP 10.0: F (1 , 32)=25.12, P < 0.00l]. Within-cell comparisons of means found that all CDP groups differed from controls for re-entries on block 5, beyond the 0.001 probability level (see Table 4), though the CDP groups did not differ from each other. There were two main differences between conditions with respect to error clusters. Sequence length was longer in the random condition, reaching a maximum of 16 consecutive errors, as opposed to eight, and as shown above, re-entries were made into both non-rewarded and positive arms, whereas they were specific to positive arms in the constant condition.

Running time

Running time (total time in the maze/n of arms entries) was compared on criterion and test blocks. For both ran-

465

Table 4. Experiment 2: constant cue condition: error clusters. Mean number of error clusters (4 + successive errors) and type of within- cluster errors [first entry into non-rewarded arms (RM) and all re-entries (WM+NS)] on blocks 4 and 5: t-ratio comparisons of CDP-treated groups with controls

Groups Block 4 Block 5 (N=9)

N of Type of error N of clusters Type of error clusters

RM Re-entry RM Re-entry

t, 32 = t, 32= t, 32 =

Control 0.2±0.2 0.4_+0.4 0.4_+0.4 0.2_+0.2 0.6±0.4 0.4_+0.03 CDP 2.5 0.3_+0.3 0.7_+0.6 1 . 1 _ + 0 . 8 1.4_+0.4 NS 2.4+0.6 NS 6.1_+1.7 4.49 P<0.0005

<0.01 <0.05 CDP 5.0 0.2___0.2 0.6_+0.5 0.3_+0.3 2.1 _+0.7 2.59 3.1 _+0.9 1 . 9 7 7.7_+2.0 5.75 P < 0.0005

<0.05 CDP 10.0 0.1 _+1.0 0.2_+0.2 0.4_+0.5 1.7_+0.4 1 . 9 8 1.4___0.6 NS 6.3_+2.0 4.65 P<0.0005

dom and constant condit ions there were no overall differ- ences between groups or blocks. In both condit ions running time speeded up with the two lower doses of CDP, and rats were slowed down with the high dose, relative to criteri- on speed, with sedation apparent in some animals. However comparisons of means showed that the only reliable effects were increase in speed with the two lower doses of C D P in the r andom condi t ion [CDP 2.5: t (32) = 2.21, P < 0.025; CDP 5.0: t (32)=1.96, P<0 .05] .

In cri terion blocks there was no significant correlat ion between total t ime in the maze, and total number of errors, in either condit ion. In the random condi t ion on test block 6 the increased number of errors was closely associated with maze time in the low dose groups [CDP 2.5: r = +0.834, P < 0 . 0 1 ; C D P 5.0: r = +0.867, P<0 .01] , but time and er- rors were not significantly related in controls and with C D P 10.0. In contrast , in the constant condit ion, increased number of error arm entries was only associated marginal ly with longer maze times in the CDP 5.0 groups [r = + 0.661, P < 0.05] and remained uncorrela ted in the other groups.

Discussion

The experiments found that the effects of CDP differed substantial ly according to the task. In the r andom condi t ion the rats per formed as poor ly as when maze naive, with an undifferent iated error pat tern including re-entries into rewarded arms, though this type of error had a lmost been el iminated by the time o f testing. In the constant condi t ion impai rment was not as marked, the rats remembered non- rewarded arms as well as the controls, but made more re- entries into positive arms. In both condit ions errors ap- peared in clusters, an effect which had not been anticipated.

The experiments did not control for state dependency, so performance could reflect a variety of non-specific fac- tors. We have found (Glenn et al. unpubl ished data) that in rats t rained under either CDP or saline to a criterion of 6/8 correct (rewarded) first choices in a maze with eight food-bai ted arms, and tested with t reatments reversed, that former saline rats were not affected, and former CDP rats marginal ly improved. This suggests that state dependency is not a major factor with par t ia l ly- learned tasks. Moreover , the marked difference between the two condit ions in extent and pa t te rn of errors is not consistent with generalized ex- p lanat ions such as s ta te-dependent disrupt ion or arousal.

Thus three features of the da ta require explanat ion; the clustering of errors, the emergence of non-specific errors, and the different error pat terns in the two condit ion.

The preponderence of re-entries within error clusters suggest that they involve response repetit ion, which has been repor ted to occur both under BZ t reatment and after h ippocampal lesion (O'Keefe and Conway 1978; Gray 1982). Gray ' s theory of behavioural inhibi t ion would pre- dict a close associat ion between perseverat ion and informa- tion processing impairment . On the other hand, Olton and Wertz (1977) have shown that perseverative responding and discriminat ion failure can be dissociated. These factors have also been dissociated following CDP treatment, by the use of condi t ional contingencies and signal detection analysis (Tye et al. 1977; Cooper and Francis 1979). Results from the present experiments do not provide clear evidence about the relat ionship between perseverat ion and discriminat ion deficits. Longer error sequences in the r andom condi t ion suggest enhanced perseverat ion with a more complex task, in line with Gray ' s theory. However, the similar number of clusters and selectivity of positive arm re-entries in the constant condi t ion show that bursts of perseverative re- sponding can occur both with and without a high degree of discriminative control.

Non-specific errors of re-entry into non-rewarded arms dropped out early in training, before cue associations were apparent , so that rats normal ly required only one unsuc- cessful visit to learn, within trials, which arms were not rewarded. Hence NS errors may indicate a p rofound disrup- t ion of within-trial spatial learning, rather than loss of mem- ory for the cue. Alternatively, increase in NS errors could indicate that the rats were not deterred by the aversive im- pact of non-reward (Gray 1982), but this explanat ion would not account for their increase only in the r andom condition. O'Keefe and Nade l (1978) have proposed that the hippo- campus constructs and stores cognitive maps of the envi- ronment, and have shown disrupt ion of spatial rather than non-spat ia l tasks after h ippocampal damage. Effects of CDP in these experiments, however, do not appear to in- volve specifically spatial deficits. In the constant condi t ion the rats remembered the locations of positive arms, and avoided non-rewarded ones, so that long-term spatial mem- ory was not disrupted. In the random condit ion, which would force the construct ion of new cognitive maps on each trial, spatial learning does seem to have been highly

466

disrupted. However, spatial and non-spatial factors cannot be dissociated, as the rats also had a non-spatial cue. More- over, the substantial increase in W M errors indicates some within-trial learning of positive arm locations. The generally indiscriminate nature o f errors suggests that the deficits in- cluded more than a loss of the ability to form short-term spatial representations, though these suffer along with other types of association, and were particularly stressed by the random task.

The experiments set out to see whether effects of CDP on memory are confined to recent information, as shown by selective working memory deficits, or involve impaired stimulus processing, which would produce more but undif- ferentiated errors in a complex rather than a simple task. The results showed that with a simple task recent informa- tion is more vulnerable than stable long-term information, but with a more difficult task long-term associations are also impaired.

The pattern of many indiscriminate errors in the ran- dom condition, which required close attention to cues, and the relatively good performance in the constant condition, suggest that effects of CDP are related to task difficulty. Thus the distinctive error pattern in the constant condition might show that avoiding non-rewarded arms is, in stable conditions, an easy task, whereas avoiding re-entry to posi- tive arms is more difficult, because it requires more infor- mation; recall both of whether the arm is positive and whether it has already been entered. In the random condi- tion, where cues had to be relocated on each trial as well, the information processing load would be even heavier. The confusion produced by CDP indicated that rats failed to utilize learned associations and strategies, as well as failing to keep track of their movements, so that both long and short term information may be vulnerable in conditions that need attention. Thus the ubiquitous impairment found in both man and animals in retention of new information rather than with well-learned tasks, may arise because novel information demands more attention and analysis, rather than because short-term storage processes are selectively attacked.

Working memory errors may distinguish recent from long-term memory, but they do not dissociate initial encod- ing from short-term storage. I f BZs impair stimulus analy- sis, encoding rather than storage deficits would be expected. Storage should be highly sensitive to delay, whereas encod- ing should be affected by stimulus complexity. A radial maze is not well-suited to handle delay, since rats which make errors ipso facto increase time in maze. However, a positive correlation between time and errors was not found in untreated rats, and occurred in only three out of six CDP groups. This indicates that choice accuracy was not necessarily a function of delays, which varied from 30 to 200 s/trial. We have examined (Glenn et al. submitted for publication) effects of CDP on delayed responding over 0-120 s and found very little impairment, so that CDP does not seem to affect the ability to store recent information over a short period. Time differences, therefore, are not likely to account for differences in error rates or types.

Varying stimulus predictability, on the other hand, may have crucially affected CDP-treated rats over and above ensuring that the task involved cue-reward associations, but direct effects cannot be partialled out because randomness was confounded with type of association. The massive im- pairment found with variable cue locations is consistent with the proposal that BZs affect attentional processes that are stressed by factors such as unpredictability which in- crease stimulus complexity. This would also be in agreement with the human evidence (Brown et al. 1983) that BZs dis- rupt encoding but not short-term memory storage.

Acknowledgements. This work was supported by project grant no. 8313090N from the Medical Research Council to SEG.

References

Brown J, Brown MW, Bowes JB (1983) Effects of lorazepam on rate of forgetting, on retrieval from semantic memory, and on manual dexterity. Neuropsychologia 21 : 501-512

Cole SO (1982) Effects of chlordiazepoxide on discrimination per- formance. Psychopharmacology 76: 92-93

Cole SO (1983) Chlordiazepoxide-induced discrilxfination impair- ment. Behav Neural Biol 37: 344-349

Cooper SJ, Francis RL (1979) Chlordiazepoxide-induced disrup- tion of discrimination behaviour: A signal detection analysis. Psychopharmacology 63 : 387-310

Ghonheim MM, Hinrichs JV, Mewhaldt SP (1984) Dose-response analysis of the behavioural effects of diazepam. (I) learning and memory. Psychopharmacology 82:291-295

Gray JA (1982) The neuropsychology of anxiety. Oxford, OUP Hasegawa Y, Ibuka N, Iwahara S (1973) Effects of chlordiazepox-

ide on successive red-green discrimination responses in Japanese monkeys, Macaca Fuscata. Psychopharmacologia 30 : 89-94

Hiraga Y, Iwasaki T (1984) Effects of cholinergic and monoamin- ergic antagonists and tranquilizers upon spatial memory in rats. Pharmacol Biochem Behav 20:205 207

Iwasaki T, Ezawa K, Iwahara S (1976) Differential effects of chlor- diazepoxide on simultaneous and successive brightness discrim- ination in rats. Psychopharmacology 48:75-78

Kirk RE (1968) Experimental design: Procedures for the behav- ioural sciences. Belmont, C.A. : Brooks/Cole

O'Keefe J, Conway DH (1978) Hippocampal place units: why they fire and where they fire. Exp Brain Res 31 : 573-590

O'Keefe J, Nadel L (1978) Hippocampus as a cognitive map. Ox- ford, OUP

Olton DS, Papas BC (1979) Spatial memory and hippocampal function. Neuropsychologia 17:669-682

Olton DS, Wertz MA (1978) Hippocampal function and behavior: spatial discrimination and response inhibition. Physiol Behav 20: 597-605

Olton DS, Becker JT, Handelmann GE (1980) Hippocampal func- tion: working memory or cognitive mapping? Physiol Psychol 8 : 239-246

Tye NC, Saghal A, Iversen SD (1977) Benzodiazepines and dis- crimination behaviour: dissociation of response and sensory factors. Psychopharmacology 92:191-194

Vachon L, Kitskikis A, Roberge AG (1984) Chlordiazepoxide, Go- Nogo successive discrimination and brain biogenic amines in cats. Pharmacol Biochem Behav 20:9-22

Received May 15, 1985