Contingent Versus Incidental Context Processing During Conditioning: Dissociation After Excitotoxic Hippocampal Plus Dentate Gyrus Lesions

Transcription

1 Contingent Versus Incidental Context Processing During Conditioning: Dissociation After Excitotoxic Hippocampal Plus Dentate Gyrus Lesions M. Good,* L. de Hoz, and R.G.M. Morris Centre for Neuroscience, University of Edinburgh Medical School, Edinburgh, Scotland HIPPOCAMPUS 8: (1998) ABSTRACT: This experiment explored whether excitotoxic hippocampus plus dentate gyrus (HPC/DG) lesions in rats would dissociate the differential processing of contextual cues during the performance of learned associations when (1) their processing during training is incidental to successful learning or (2) the solution of a discrimination problem is contingent on their processing. A series of training stages were conducted, beginning with appetitive conditioning to two stimuli (X and Y), each of which was trained in one of two different contexts (operant chambers A and B) (i.e., AX, BY ). Conditioning was indexed as appetitive responding. The animals were then trained on a biconditional contextual discrimination with these same stimuli (AX, AY ;BY,BX ). The next stage involved conditioning trials to two new stimuli (W and Z), one in each context, while the animals were actively discriminating contexts A and B by continuing to perform the original biconditional discrimination (AX, AY, AW ;BY,BX,BZ ). Finally, they were trained on a second biconditional discrimination involving these new stimuli (AX, AY, AW, AZ ;BY,BX,BZ,BW ). The incidental use of context cues was examined by looking at the rate of conditioned responding to cues X, Y, W, and Z in their original training contexts or a different context; HPC/DG lesioned rats differed from controls in being unaffected by a change of context. The contingent use of context cues was examined by looking at performance of each of the two biconditional tasks; HPC/DG lesioned rats reached levels of conditional performance indistinguishable from those of controls. These findings point to two distinct ways in which contextual information is processed in the brain, revealing a dissociation between incidental and contingent processing of contextual cues after HPC/DG lesions. Hippocampus 1998;8: Wiley-Liss, Inc. KEY WORDS: hippocampus; context; conditional discrimination; ibotenic acid; incidental and contingent processing INTRODUCTION All learning occurs in a spatiotemporal context. For some learning tasks, when and where learning takes place is incidental to what is being learned. In such tasks, a failure to process context cues effectively may not be reflected in any immediate behavioural change, but incidental processing of context could nonetheless affect the way knowledge is encoded or retrieved and thus how it is expressed in a place different from that of the original training (Spear, 1971, 1973; Spear et al., 1980). In other tasks, the solution is contingent upon the processing of contextual cues. In certain types of Grant sponsor: United Kingdom Medical Research Council. *Correspondence to: M. Good, School of Psychology, Cardiff University of Wales, P.O. Box 901, Cardiff, CF1 3YG, UK. Accepted for publication 28 January 1998 conditional discrimination task, for example, a stimulus may be arranged to signal reward in one context and nonreward in another (e.g., Preston et al., 1986). Similarly, in spatial learning, effective navigation is contingent upon the processing of at least some context cues. In both cases, the failure to process context cues effectively would result in a failure of learning or at least its retardation. This distinction between incidental and contingent processing of context cues may reflect a more fundamental distinction between a role for context cues in declarative or higher-order memory processes, such as the retrieval of information about events that has occurred within the context (possible after incidental processing of context), and a role in associative learning itself (reflecting contingent processing). Traditional learning theory accounts of the role of context in learning can be divided into associative accounts and higher-order function accounts. Associative accounts (Wagner, 1976, 1978, 1981) emphasize the role of direct associations between contextual cues and events that occur in their presence. Higher-order function accounts generally acknowledge the associative properties of context cues and propose that context cues participate in the retrieval of information about the relationships between cues that have occurred in their presence (e.g., Spear, 1973; Swartzentruber and Bouton, 1986; Hall and Honey, 1989). The aim of the experiments described in the present study was to explore whether a dissociation between these different ways in which context cues may be processed could be realized using excitotoxic hippocampal plus dentate gyrus (HPC/ DG) lesions. Several studies have implicated the hippocampus in forming a representation of the context (e.g., Sutherland and Rudy, 1989; Kim and Fanselow, 1992). If the integrity of HPC/DG is essential, at least temporarily, for forming a context representation, there is no reason to expect a dissociation. Conversely, if some types of context processing can occur when HPC/DG is damaged and other types depend on hippocampal function (Honey and Good, 1993; Phillips and LeDoux, 1994), dissociations between qualitatively distinct types of context processing might be identified WILEY-LISS, INC.

2 148 GOOD ET AL. The different outcome of two closely related experiments led us to think about this distinction. Good and Honey (1991) found that electrolytic lesions of the dorsal HPC/DG impaired the acquisition of a complex biconditional discrimination in which, as in Preston et al. (1986), the reward significance of each of two stimuli was switched between the two contexts in which training took place. Whishaw and Tomie (1991), in contrast, found that rats with excitotoxic lesions of HPC/DG were unimpaired in the acquisition of a logically similar biconditional discrimination. In seeking to reconcile the different outcomes of these two studies, the type of the lesion made (electrolytic vs. excitotoxic) may be critical. The electrolytic lesion procedure disrupts fibres of passage, vasculature, and cell bodies, and thus the impairment of conditional learning reported by Good and Honey (1991) may reflect extrahippocampal damage. In a pilot experiment, we examined the effects of ibotenic acid lesions of HPC/DG on the acquisition of this same type of biconditional contextual discrimination. The results indicated that neurotoxic HPC/DG lesioned animals acquired the biconditional contextual discrimination normally. A similar observation was also made independently by McDonald et al. (1997). This finding suggests that certain types of context processing are unaffected by HPC/DG lesions. However, we also found that, after the initial training of each of two stimuli presented in their rewarded context only, conditioned responding to these stimuli upon their first presentation in the other context was unchanged in HPC/DG lesioned animals, whereas controls showed a much reduced rate of responding. This indication of the context specificity of responding in control animals may reflect incidental learning about context that occurs automatically (i.e., without regard to schedule) but requires the integrity of hippocampus. To explore this more formally, various issues needed to be addressed including the critical one of whether, at the time they fail to display context specificity of conditioned responding, HPC/DG lesioned rats are actually discriminating one context from the other. The experimental design was based on the study by Preston et al. (1986). As shown in Table 1, the experiment consisted of six stages. In the first stage, the rats received conditioning trials with two stimuli (X and Y) in two different contexts (A and B) such that X was always presented in A and Y was always presented in B (i.e., AX, BY ). Responding took the form of approaches towards a place in the box where food was occasionally made available. In the second and overlapping third stages, they were trained for several days on a so-called biconditional contextual discrimination task in which the significance of X and of Y depended on the context in which they were presented (AX, AY ;BY,BX ). The first day of training (stage 2) is formally identical to a test for context specificity used by Honey and Good (1993). Stage 4 involved continued training on the first biconditional discrimination task but also rewarded presentations of two new stimuli (W and Z), one in each context (i.e., AX, AY, AW ;BY,BX,BZ ). After these training sessions, stages 5 and 6 were conducted. These consisted of training on a second biconditional contextual discrimination with all of the stimuli (AX, AY,AW,AZ ;BY,BX,BZ,BW ). Once again, the first day of training on the second biconditional TABLE 1. Summary of the Six Stages of the Main Experiment and the Two Subsidiary Experiments* Main experiment Surgery Stage 1: Simple conditioning of S1 AX,BY (S1) Stages 2 and 3: Context specificity test and biconditional acquisition of S1 AX (S1 ), AY (S1 ) BY (S1 ), BX (S1 ) Stage 4: Simple conditioning of S2 and continued performance of S1 biconditional discrimination AW,BZ (s2) AX (S1 ), AY (S1 ) BY (S1 ), BX (S1 ) Stages 5 and 6: Context specificity test of S2, acquisition of S2 biconditional discrimination, and continued performance of S1 biconditional task AW (S2 ), AZ (S2 ) BZ (S2 ), BW (S2 ) AX (S1,AY (S1 ) BY (S1 ), BX (S1 ) Subsidiary experiments Odour test: Switching of odors between contexts A and B. Watermaze aquisition: Place navigation to a single location. *The symbols within each stage refer to contexts (A, B), explicit stimuli (X, Y, W, Z), and reward significance (, ). S1 and S2 refer to first and second biconditional discriminations, respectively, averaged across the relevant stimuli. discrimination (stage 5) served as a context specificity test for conditioned responding to stimuli W and Z. However, in contrast to stage 2, this context specificity test was conducted under conditions in which the animals could be shown to be actively discriminating between the two contexts during the same session (as indexed by appropriate responding to stimuli X and Y). At the end of the experiment and prior to death and histology, all rats were trained in a watermaze to find a hidden escape platform located at a single fixed location (reference memory). This simple spatial learning task also makes use of contextual cues in both an incidental and a contingent way, but their processing is for the purposes of navigation. It also served as a simple behavioural assay of the effectiveness of the neurotoxic lesions.

3 CONTINGENT AND INCIDENTAL CONTEXT 149 GENERAL METHODS Subjects The subjects were 20 adult male hooded Lister rats (bred in the Department of Pharmacology, University of Edinburgh) with a mean ad libitum weight of 375 g at the start of training. One group of rats (n 10) received bilateral ibotenic acid lesions of the hippocampus. The second group (n 5) served as the unoperated control group. A third group of rats received sham operations (n 5). The rats were housed individually with continuous access to water and were maintained at 80% of their free-feeding body weights throughout the experiment. Surgery and Histology The procedure used to remove cells in the hippocampus and dentate gyrus was similar to that described by Jarrard (1989). The animals were first anaesthetised with Avertin (tribromoethanol; 10 ml/kg, i.p.) and placed in a Kopf stereotaxic frame. An incision was made in the scalp, and the bone overlying the neocortex situated directly above the hippocampus was removed. Injections of ibotenic acid (concentration, 10 g/ l; ph 7.4; Tocris Neuramin, Ltd., United Kingdom) were made with a 1- l Hamilton syringe mounted on the stereotaxic frame. Injections of l were made over approximately 1 min at each of 26 sites. Sham-operated animals received a treatment similar to that of HPC/DG operated subjects except that passage of the needle was limited to the cortex and no drug was infused. At the end of behavioural testing, the HPC/DG rats were given injections of Euthatal (200 mg/kg sodium pentobarbital) and perfused with physiological saline and 10% formol saline. The brains were removed from the skull and placed in formol saline solution. The brains were then frozen and sectioned in the horizontal plane (30- m sections). A thionin stain was used to determine cell loss. Reconstruction of the lesions were then made on drawings derived from the stereotaxic atlas by Paxinos and Watson (1986). Apparatus Two pairs of identical operant chambers (Campden Instruments, Ltd.) were used. The chambers were constructed from three sheet-aluminium walls, a transparent plastic door as the fourth wall, with grid floors and aluminium ceilings. One of the walls adjacent to the door contained a recessed food tray. Access to this tray was guarded by a transparent hinged plastic flap. Inward movements of the flap actuated a microswitch, and each closing of this switch was recorded as a single response. The primary data measure in this study was the rate of flap responding. The ceiling of each chamber contained a light and a loudspeaker. The offset of the light for 30 sec served as one conditioned stimulus (CS). A 20-Hz train of clicks at an intensity of 82 db(a) for 30 sec served as a second CS. These stimuli were used in stages 1 3 of training (referred to as stimuli X and Y). Two speakers placed on the wall opposite the food tray were used to present a third CS, a 30-sec white noise of identical intensity. The ceiling-mounted speaker was used to present the fourth CS, a 30-sec 200-Hz tone. Stimuli 3 and 4 (referred to as W and Z) were introduced at stage 4 of training. The boxes were housed in sound- and light-attenuating shells. The two pairs of boxes differed in the following respects. The boxes were in rooms on different floors of the laboratory; one pair had an odor produced by adding a small amount of eucalyptus oil to the saw-dust tray located below the grid floor, and the other pair had an odour produced by isoamyl acetate. The pair with isoamyl acetate was also made visually distinctive by adding black-and-white chequered wallpaper to the door and to the wall opposite this door. The watermaze in which the animals were trained at the end of the main experiment was 2 m in diameter, had a hidden escape platform 11 cm in diameter, and was otherwise as described by Davis et al. (1992). Procedure The subjects received one session of training in each context on each day. The stimuli that served as the conditioned stimulus and the operant chambers (contexts) were counterbalanced. Training proceeded in a series of stages. Stage 1: Pretraining and simple conditioning of S1 On the 2 days of pretraining (days 2 and 1), the rats were trained to collect food pellets from the food tray. On day 2, the flap in front of the food tray was fixed in the raised position; on day 1, the flap was lowered to its normal resting position. Food pellets were delivered on a variable time (VT) 60-sec schedule. On days 1 8 of conditioning, there were three presentations of stimulus X that signalled the delivery of a single food pellet in context A and three presentations of stimulus Y that preceded the delivery of a food pellet in context B (see Table 1). The onset of the first stimulus presentation (trial) was 10 min after the beginning of the session and the intertrial interval (ITI) was 10 min. Conditioning to stimuli X and Y was assessed by recording the rate of flap responding during their presentation. To reduce individual variability, conditioning was assessed with respect to the rate during the ITI to yield a measure of net rate of response (rate during CS minus rate during ITI). The net rate of response is expressed in units of responses per minute averaged across each day or pair of days (blocks). Stages 2 and 3: Context specificity test and biconditional acquisition of S1 On days 9 20, there were three reinforced presentations of X in context A and of Y in context B; these are referred to as S1 trials in Table 1 (where 1 refers to biconditional discrimination task 1). In addition, there were three nonreinforced presentations of stimulus Y in context A and of stimulus X in context B (S1 trials). The order in which trials were presented was random, with

4 150 GOOD ET AL. the constraint that no more than two trials of the same type (S1 or S1 ) were presented in succession. The ITI was 5 min. Although procedurally identical to succeeding days, the first day of this procedure (stage 2) was the first occasion that stimuli X and Y were presented in a context different from that of stage 1 training. Each stimulus was presented half the time in the original context and half the time in the other context. The comparison of these two is the context specificity test that constitutes the first opportunity to examine incidental use of context cues. The succeeding days (stage 3) provided a test of whether control and lesioned animals could learn to use context cues in a contingent way. Performance in stage 3 was averaged over 2-day blocks. Odor test On day 21, a transfer test was conducted to determine what aspects of the contextual cues were being used by the control and lesioned animals. The visual cues and the time of testing remained the same, whereas odor cues that comprised the two contexts A and B were switched. If the discrimination was being solved on the basis of the odor cues alone, we would expect a reversal of the discrimination. Conversely, if the discrimination was being solved by using some combination of exteroceptive cues (but not interoceptive cues such as time of day), discrimination performance should merely deteriorate. For this test, odor trays were removed on the day before testing, recharged with either isoamyl acetate or eucalyptus oil, and then left in the alternative context overnight before testing the following day. The procedure was otherwise identical to that used in stages 2 and 3. Stage 4: Simple conditioning of S2 and continued performance of S1 biconditional discrimination On days 22 29, the odor cues were returned to their original training boxes. Two additional stimuli, W and Z, were now introduced, W signalling reward in context A and Z in context B. There were three trials of each stimulus in their respective contexts; the procedure is therefore analogous to that of stage 1 with the original stimuli X and Y. However, although the animals were trained on these new stimuli, they were also given another 8 days of training on the S1, S1 biconditional discrimination (see Table 1). The CSs that served as either W or Z were counterbalanced. The animals received exposure to the new stimulus and to X and Y, within the same session using a counterbalanced sequence, subject to the constraint that no more than two trials of the same type were presented in succession. The ITI was 10 min. Stages 5 and 6: Context specificity test of S2, acquisition of S2 biconditional discrimination, and continued performance of S1 biconditional task On day 30, the first day of testing with W and Z, and continuing for 12 days (i.e., until day 41), the animals received reinforced trials of stimulus W in context A and of stimulus Z in context B (referred to as S2 trials, where 2 refers to the second biconditional discrimination; Table 1) and nonreinforced trials of ZinAandWinB(S2 trials). By analogy with stages 2 and 3, stage 5 (day 30) was the context specificity test for stimuli W and Z, and days (stage 6) examined acquisition of the discrimination. It is important to appreciate that performance of the first biconditional discrimination (of stimuli X and Y) continued throughout stages 5 and 6. That is, testing and training of the S2 discrimination (W and Z) were begun alongside the continued daily presentations of the S1 discrimination (X and Y). There was a total of 12 trials/day, three trials of each type (S1, S1,S2, and S2 ). The order in which trials were presented was random, subject to the constraint that no more than two trials of the same type were presented in succession. Watermaze training Upon completion of training in the operant chambers, all subjects received training in an open-field watermaze. On each of 4 days (days 38 41), the animals received four training trials. At the start of each trial, the rat was placed into the pool at a random position around the perimeter and allowed to search for the submerged platform for a maximum of 120 sec. The position of the platform (northeast or southwest) was counterbalanced. If an animal failed to locate the platform within 120 sec, it was guided to it by the experimenter (this occurred rarely). The animals were left on the platform for 30 sec before the start of the next trial. On day 45, after a total of only 16 training trials, a transfer test was conducted in which the platform was removed from the pool and the animals permitted to swim freely for 60 sec. Paths swimming around the pool were monitored by an automated videotracking system, and the proportion of time spent swimming in the former training quadrant was measured. RESULTS Histology For a subject to be included in the HPC/DG group, it had to fulfill specific criteria. First, it was necessary for at least 90% of the CA1 CA4 pyramidal cells and granule cells in the dentate gyrus to be removed. An animal was excluded if there was incomplete cell loss in hippocampus or dentate gyrus. Second, it was essential that there was minimal damage to extrahippocampal structures, especially to cells in the subiculum and entorhinal cortex. An animal was excluded if it sustained combined damage to either structure at any septotemporal level. Judgments regarding whether a given subject fulfilled these criteria were made without specific knowledge of the behavioural observations. One HPC/DG lesioned animal failed to fulfill these criteria. The lesion was limited to the dorsal CA3 region of HPC/DG, with very little damage to any other region of the hippocampus at any other level. The remaining animals in the lesion group met the first criterion in showing near complete cell loss in the CA1 CA3 regions. Four animals showed a small amount of sparing in

5 CONTINGENT AND INCIDENTAL CONTEXT 151 dentate gyrus but very few cells remained. With respect to the second criterion, no animals were excluded. The subiculum showed some cell loss, either unilaterally or bilaterally, in nearly all cases. In four cases, damage to the subiculum was minor in the septal and mid septotemporal aspects of the hippocampus, but the damage extended as far as the presubiculum. Cell loss in the ventral subiculum was variable; most lesioned animals showed some cell loss in this region also but not enough to warrant exclusion. A photomicrograph showing a representative hippocampal lesion at a mid dorsal-ventral level is shown in Figure 1A. Figure 1B shows the maximum and minimum extent of the lesions in the HPC/DG group on horizontal sections taken throughout the dorsoventral extent of the hippocampus. Behavior One unoperated control animal persistently failed to consume the food reinforcement during the first stages of conditioning and was subsequently dropped from the experiment. All other animals learned to operate the magazine flap and responded appropriately to the scheduled stimuli. There were no systematic differences in the performance of the sham-operated and unoperated control animals at any stage of the experiment and, therefore, to simplify statistical analyses, the data from these two conditions were collapsed to form a single control group. The final number of subjects in the control and HPC/DG lesioned groups was 9 per group. At the end of the main experiment, the rats were trained in a watermaze for 16 trials. The results of the transfer test conducted after the final day of training showed that control subjects had developed a clear bias in the proportion of time spent in the training quadrant (36.8%, chance 25%), whereas HPC/DG lesioned animals showed substantially less bias (19.4%). This difference was highly significant [t(16) 4.47, P 0.001]. The slightly below-chance level of performance by the HPC/DG lesioned animals on the probe trial was caused by one animal who displayed a significant bias away from the training quadrant that appeared to be related to a prominant extramaze cue in this area (door to the laboratory). Because the behavioral protocol used for the watermaze is one that is exquisitely sensitive to HPC/DG lesions, these findings complement the histology in establishing the adequacy of the lesions made. In the data analysis of the main experiment that follows, data from successive stages are considered together with respect to performance during acquisition of conditioned responding (stages 1 and 4), to incidental processing of context cues (context test: stages 2 and 5), and to the contingent use of context cues (biconditional discriminations: stages 3 and 6). Although this organization is not chronological, it is the simplest way to present a complex set of data in a manner that is conceptually straightforward. Conditioning of S1 and S2 cues (stages 1 and 4) Acquisition of conditioned responding to X and Y. Inspection of the scores revealed no systematic differences based on the particular stimuli that were assigned as X and Y and no differences dependent on the contexts that were designated as A and B. Accordingly, the data were pooled across stimulus type and context type for statistical analysis. Figure 2A shows the acquisition rates for the HPC/DG and control subjects, in 2-day blocks, over the course of conditioning to S1 (the average of responding to X and Y). It is clear that both the lesioned and control groups acquired conditioned responding, but at no point was there any sizeable difference in the level of responding shown by the two groups. An analysis of variance (ANOVA), with groups and block as factors, confirmed these impressions. There was a significant main effect of block [F(3,48) 35.90, P 0.001], no significant main effect of group [F(1,16) 1.91, P 0.18], and no significant interaction between these factors [F(3,48) 2.15, P 0.10]. An analysis of the baseline rates of responding during the ITI, with a mean of 2.94 responses per minute (rpm) for the HPC/DG group and 2.45 rpm for the control subjects, revealed no difference between groups [t(16) 1.43, P 0.20]. Acquisition of conditioned responding to W and Z. Inspection of Figure 2B shows that both the control and HPC/DG lesioned animals maintained discriminative responding between the S1 and S1- of the first biconditional discrimination and acquired conditioned responding to the new stimuli (S2 trials). An ANOVA confirmed these observations and revealed a significant main effect of stimulus significance (S1, S1, S2 ). Pairwise comparisons (Newman-Keuls) revealed significant differences between S1 and S1 response rates (P 0.01) and between S2 and S1 response rates (P 0.01) and no significant difference between S2 and S1 response rates (P 0.05). This pattern of performance did not differ between the lesioned and control groups [F(2,32) 1.51, P 0.23]. There was a significant interaction of block with stimulus type [F(6,96) 14.57, P 0.001] that reflected a steady increase in responding to S2 over the four blocks of training (simple effect, P 0.001). Although the rate of responding is numerically higher to S2 in the control than in the lesioned group, this tendency was not significant [F(6,96) 1.30, P 0.26]. The mean rates of responding during the ITI for HPC/DG and control animals were 0.80 rpm and 0.85 rpm, respectively. These did not differ (t 1). Incidental processing of context cues (stages 2 and 5) Figure 3 shows the results of the context specificity tests for conditioned responding to X and Y in each of the two contexts on day 9 (stage 2), to W and Z (stage 5, day 30), and to the S1 stimuli during stage 5. For clarity, the data are subdivided into the HPC/DG (Fig. 3, left) and control (Fig. 3, right) groups. Control animals displayed contextual specificity of conditioned responding in both stages. That is, the level of responding was higher to a stimulus presented in its training (same) context than when presented in a context different from that of original training. HPC/DG lesioned animals did not show this effect at either stage of training. These impressions were confirmed by ANOVA with group, training, and context as factors. There was a

6 152 GOOD ET AL. FIGURE 1. Lesion assessment. A: Photomicrograph of a randomly selected HPC/DG animal taken at a mid dorsal/ventral level. B: Schematic representation of the maximum (stippled) and minimum (solid areas) extent of the lesions in the HPC/DG group throughout the dorsal-ventral extent of the hippocampus.

7 CONTINGENT AND INCIDENTAL CONTEXT 153 FIGURE 1. (Continued.)

8 154 GOOD ET AL. significant main effect of context change [F(1,16) 12.53, P 0.01] and a significant interaction of lesion group with context change [F(1,16) 6.01, P 0.05]. Simple main effects confirmed a significant effect of context change in control animals [F(1,16) 17.95, P 0.01], but no significant effect of context change in HPC/DG lesioned rats [F(1,16) 1]. No significant main effect of stage of training was found [i.e., context specificity after simple conditioning (stage 1) vs. that after biconditional training (stage 4): F 1], and no significant interaction of group with stage of training (F 1). Figure 3 also shows the mean S1 and S1 rates of responding on the first biconditional discrimination measured during the second context specificity test of stage 5. It is clear that both HPC/DG lesioned and control animals are discriminating between the contexts equally effectively. This impression was confirmed by ANOVA that revealed no significant main effect of group [F(1,16) 1.26], but a highly significant main effect of stimulus significance with respect to context [i.e., S1 vs. S1 : F(1,16) 28.06, P ]. There was no interaction between lesion group and contextual significance (F 1). Thus, the HPC/DG lesion rats are discriminating between the two contexts during stage 5, but they are unable to use this information to modulate their responding to stimuli W and Z. Contingent use of contextual cues (stages 3 and 6) Biconditional contextual discrimination 1 (stage 3). Figure 4A shows the mean net rates of responding during S1 and S1 trials for control and HPC/DG lesioned rats during acquisition of the conditional contextual discrimination. The S1 data represent responding during X and Y trials when they were reinforced, and the S1 data represent the rates of responding during X and Y when they were not reinforced. The lesioned and control animals acquired the conditional contextual discrimination rapidly, reached the same asymptote of performance, and, other than the difference on the first day of training (reflecting incidental context processing during stage 2), they learned at essentially the same rate. An ANOVA revealed a significant effect of stimulus significance [S1 vs. S1 : F(1,16) 61.03, P ] and a significant interaction of block and trial type [F(5,80) 9.23, P 0.001] that reflects the course of learning. There was no significant difference between groups (F 1), although a significant interaction of group with blocks did emerge [F(5,80) 2.74, P 0.05]. This interaction reflected the overall higher level of responding on the first block of training in the control animals versus the lesioned subjects (simple main effects analysis, P 0.05). FIGURE 2. Conditioning of S1 and S2 cues (stages 1 and 4). A: Mean net rates (responses/minute) during S1 presentations for HPC/DG and control animals, presented in 2-day blocks, in stage 1. B: Mean net rates of responding in HPC/DG and control animals during the S1 and S1 trials of the first biconditional contextual discrimination and during presentations of the S2 trials in stage 4. Intertrial interval responding. The mean rates of responding during the ITI for HPC/DG lesioned and control animals were 2.51 rpm and 2.77 rpm, respectively. They did not differ (t 1). Odor test. The effects of changing the odor cues between boxes was evaluated by comparing discriminative performance on the final day of conditional discrimination training (day 20) with performance on the day of the odor test (day 21). The mean net rates of responding for HPC/DG and control subjects, respectively, were: rpm (S1 ) versus 1.03 rpm (S1 ) and 9.73 rpm (S1 ) versus 0.69 rpm (S1 ). On the day of switching the odor, discriminative performance deteriorated in both lesioned and control subjects. The rates of responding on the S1 and S1 trials for HPC/DG and control subjects was 6.54 versus 1.13 rpm and 5.11 versus 1.03 rpm, respectively. An ANOVA confirmed these impressions and showed a significant main effect of day [normal vs. odor switch; F(1,16) 12.02, P 0.01] but no significant main effect of group (F 1) or interaction between group and day (F 1). There was a significant main effect of stimulus significance [F(1,16) 30.80, P 0.001]. There was also a significant interaction of day with stimulus significance [F(1,16) 19.95, P 0.001] that reflected a main effect of

9 CONTINGENT AND INCIDENTAL CONTEXT 155 FIGURE 3. Contextual specificity test (stages 2 and 5). Mean net rates of responding to the CS when presented in the same (S ) or different (S ) context on day 1 of the first (S1; stage 2) and second (S2; stage 5) biconditional discriminations for HPC/DG (left) and control (right) animals. The right histograms within each panel represent the mean rates of responding during the S1 and S1 biconditional contextual discrimination during the second context specificity test (stage 5). odor change for S1 (P 0.001) but not for S1 (P 0.50). The rates of responding during the ITI for the HPC/DG and control animals were 0.67 rpm and 0.70 rpm, respectively, which did not differ (t 1). Acquisition of biconditional contextual discrimination 2 (stage 6). Figure 4B C shows the mean rates of responding, in 2-day blocks, during acquisition of the second biconditional discrimination (S2 and S2 ) and performance of the first biconditional discrimination (S1 and S1 ) for the HPC/DG and control groups, respectively. The new discrimination was learned while performance was maintained on the first discrimination by both groups. An ANOVA, with group, block, discriminations 1 and 2, and stimulus significance (S vs. S ) as factors, showed no main effect of group [F(1,16) 2.89, P 0.10] or block [F(5,80) 1.51, P 0.19]. There was a main effect of discrimination [F(1,16) 9.04, P 0.01] that reflected the slightly higher rate of responding in the S2 and S2 discrimination (mean rate for stimuli X and Y 9.37 rpm, mean rate for W and Z rpm). There were no significant interactions of group with any of the factors (maximum F value 2.57, P 0.12). The mean rates of responding during the ITI for HPC/DG and control animals were 0.74 rpm and 0.67 rpm, respectively, and these did not differ (t 1). GENERAL DISCUSSION The main finding of the present study is a dissociation between the incidental and contingent processing of context cues induced by HPC/DG lesions. Rats given ibotenate HPC/DG lesions (1) displayed impaired contextual specificity of conditioned responding when stimuli trained in one context were tested for the first time in a second context but (2) acquired a biconditional contextual discrimination, as did control subjects. This dissociation was not secondary to a failure to discriminate contexts because it was apparent with newly conditioned stimuli in animals displaying a contextual discrimination using previously trained stimuli. The lesion extended to approximately 90% of all hippocampal and dentate cell fields and was sufficient to impair spatial learning in a watermaze, and thus the dissociation is unlikely to be due to sparing of hippocampal tissue. In discussing these findings, we shall consider their

10 156 GOOD ET AL. FIGURE 4. Contingent use of context cues (stages 3 and 6). A: Mean net rates of responding during S and S trials of the first (X, Y) biconditional contextual discrimination for HPC/DG and control animals (stage 3). B C: Mean net rates of responding for control and HPC/DG animals, respectively, averaged over 2-day blocks, during performance of S1 and S1 trials of the first (X, Y) biconditional discrimination and acquisition of the second S2, S2g (W, Z) biconditional discrimination (stage 6). implications with respect to the contribution of the lesion technique and theories of hippocampal function. Lesion Technique One aim of the present study was to examine the effects of selective HPC/DG lesions on acquisition of complex biconditional contextual discriminations and the contextual control of appetitive conditioned responding. The results clearly demonstrate that animals with HPC/DG damage can acquire and perform up to two biconditional contextual discriminations simultaneously. The lesions damaged approximately 90% of cell fields in the hippocampus and dentate gyrus and were sufficient to impair acquisition of spatial knowledge in the watermaze test. This pattern of results is consistent with the observations of Whishaw and Tomie (1991) and McDonald et al. (1997) using a neurotoxic lesion technique but conflict with the findings of Good and Honey (1991) using electrolytic lesions. The implications are straightforward. First, the learning of biconditional tasks must involve extrahippocampal connections that are damaged by the older lesion technique but are unaffected by neurotoxic lesions. Second, some aspects of contextual processing survive hippocampal cell damage. Theoretical Implications There are several different theoretical ideas about hippocampal function to which the present results are relevant. A prominent view of the role of context cues has been to consider their processing as an aspect of spatial learning, of scene memory, or of configural learning using the multimodal cues that make up contexts. These ideas do not seem capable of adequately capturing the dissociation we found. For example, the spatial mapping theory of hippocampal function (O Keefe and Nadel, 1978) predicts either a deficit in biconditional discrimination learning (if locale processing is used to discriminate the two contexts) or the lack of a deficit (if a taxon strategy is used). It is not clear, however, that the theory can predict the dissociation between impaired

11 CONTINGENT AND INCIDENTAL CONTEXT 157 contextual specificity and unimpaired biconditional learning because the difference between these does not map in any obvious way onto the difference between the locale and taxon hypotheses. Gaffan (1991) proposed that the hippocampus participates in learning about the spatial arrangements of scenes and that this information can then be used to retrieve personally experienced information from memory. The process of learning about the spatial arrangement of a scene is distinguished from mere place learning or the creation a cognitive map. Evidence in support of the latter distinction was provided by Gaffan and Harrison (1989) who found that monkeys with fornix transection could learn a place discrimination task in which the reward value of particular objects was conditional upon their location in the room. The contextual cues that constituted each place were different in each discrimination problem. According to Gaffan (1991), the task did not require the animal to discriminate between different spatial arrangements of these foreground and background cues and was, therefore, insensitive to hippocampal damage. The biconditional discrimination task used in the present study has similarities to the task developed by Gaffan and Harrison (1989). Different contexts acted as the conditional cues in the discriminations, but the animals did not need to distinguish them on the basis of the spatial arrangement of the elements. Hippocampal lesioned animals were also as sensitive as controls to the mispairing of the exteroceptive elements of the contexts, a mispairing that did not require a judgment of spatial location. Gaffan s (1991) account cannot, however, explain the differential outcome of the contextual specificity and biconditional parts of the experiment. Other theories also suggest that acquiring a memory representation of a context depends on the integrity of hippocampal function (e.g., Sutherland and Rudy, 1989; Kim and Fanselow, 1992). However, if hippocampal damage impairs the formation of a contextual representation, the learning of biconditional contextual discrimination should be impaired. This was not observed. Moreover, the results from the odor test indicate that hippocampal lesioned animals were as disrupted as controls by changing the configuration of the exteroceptive stimulus elements. The most parsimonious explanation for this pattern of results is that hippocampal lesioned animals were able to form a representation of the context (see Maren et al., 1997), did so using the same combination of stimulus elements as controls, and were somehow able to use this information to solve a biconditional discrimination problem. A possibility is that the lesioned (and control) animals used a configural representation to solve the biconditional task. If so, the 1989 version of Sutherland and Rudy s configural association theory cannot explain this finding. Rudy and Sutherland s (1995) revision of the configural association theory may also be relevant. In their original formulation of the theory, Sutherland and Rudy (1989) proposed a distinction between a configural association system and a simple association system. Normal hippocampal function was thought to be necessary for the formation of configural representations. Although able to accommodate a body of data, this theory cannot explain the lack of hippocampal lesion deficits on certain types of nonlinear configural tasks such as feature-neutral (Gallagher and Holland, 1992) biconditional discriminations (Whishaw and Tomie, 1991; present study) and some cases of negative patterning (Davidson et al., 1993). Alvarado and Rudy (1995) suggested that, despite the logical similarity between negative patterning and feature-neutral discriminations, these tasks may be solved differently by control animals. Rudy and Sutherland s (1995) revision proposed that the hippocampus plays a more important role in configural learning when there is a conflict between the reinforcement contingencies associated with the compound stimulus and the individual elements of the compound. Under these conditions, the hippocampal formation enhances the activation or salience of configural representations, but these are encoded in other areas of the cortex. They predict that, because a biconditional problem does not encourage competition between the elements, this task would not be sensitive to hippocampal damage. This aspect of the revised configural association theory would appear to be supported by the results of the present experiment. However, one implication of the theory by Rudy and Sutherland (1995) is that a biconditional problem may become sensitive to hippocampal damage if the elements undergo differential reinforcement outside the compounds (e.g., AX, A,AY ;BX,B, BY ). In the present biconditional discrimination, the elements of the compound were subject to different reinforcement contingencies outside the compound. Stimulus X is reinforced in context A(X ) and not reinforced in context B (X ), and context A is reinforced during the compound (AX ) but is not reinforced outside this compound on all other occasions (A, AY ). However, the present study may not represent a definitive test of the predictions of the hypothesis by Rudy and Sutherland (1995) regarding competition between elemental cues and their configural representations. A different type of context hypothesis to which these results may be relevant is the notion that hippocampal function underlies the effective contextual retrieval of associative information (e.g., Hirsh, 1974, 1980). Hirsh s original account predicts a hippocampal deficit in all forms of conditional learning. To the extent that the biconditional task used in the present study requires a conditional solution, the results do not support Hirsh s (1974, 1980) account of hippocampal function. His account, however, may be rescued if it is assumed that the role of the context in helping to retrieve associative information is something that can be supported by automatic, incidental processing in the hippocampus, with traditional conditional tasks being considered ambiguous because they can be acquired in other ways. We pursue this possibility below. A different explanation for why hippocampal lesioned animals can acquire a complex biconditional discrimination but show impaired context specificity of conditioned responding on the first day of training may that they are slow to inhibit conditioned responding. There are also several reasons to doubt this explanation. First, the absolute level of responding on trials on which X, Y, W, and Z were presented for the first time in the wrong context was the same in HPC/DG and control animals. The difference between groups arose because of the higher level of responding on training context trials in control animals (see Fig. 3). This finding suggests that control animals are benefitting from some process, such as context retrieval, that operates primarily on trials in the original training context. Second, if there was

12 158 GOOD ET AL. impaired inhibition, acquisition of the biconditional discriminations should be more severely disrupted. Not only did we fail to observe this, but excitotoxic hippocampal damage does not impair acquisition of discriminative responding in other conditional and configural learning tasks (cf. Gallagher and Holland, 1992) and may even yield superior performance in reversal of simple discriminations (see Whishaw and Tomie, 1991). Impaired response inhibition seems an unlikely account of the pattern of results (for further discussion, see Good and Bannerman, 1997). Incidental and Contingent Processing of Context Cues One way of characterizing the difference between the biconditional contextual discrimination and the test for contextual specificity is that only in the former case do the reinforcement contingencies explicitly require the animal to process contextual cues; the solution of the task can be said to be contingent on their processing. However, in the training that precedes the test of context specificity, the context does not have to be processed; it consists of cues that are merely incidental to the task demands imposed on the animal. A somewhat similar characterisation, derived from work on contextual fear conditioning, was introduced by Phillips and LeDoux (1994). They found that lesions of the dorsal hippocampus resulted in a deficit in fear conditioning to contextual cues only under conditions where the unconditioned stimulus (US) was signaled by an explicit CS. Conditioning to the context itself using presentations of the US in the absence of an explicit CS was unimpaired by hippocampal damage. Phillips and LeDoux interpreted these results as reflecting a role for the hippocampus in background contextual conditioning, where the primary or foreground association involved an explicit CS paired with a US. This situation was distinguished from foreground contextual fear conditioning, where the US is delivered in the absence of a CS. There is an intriguing parallel between the results of Phillips and LeDoux (1994) and those of the present study. Hippocampal lesioned animals displayed impaired contextual processing under conditions in which context cues were merely a background in which a CS US conditioning task took place. However, when the solution to the biconditional discrimination problem was contingent upon context processing, hippocampal lesioned animals were able to learn as efficiently as controls. The question remains of how best to characterise the psychological processes that can accommodate a distinction between contingent and incidental use of contextual information or between foreground and background processing. The key observation is that hippocampal lesioned animals are impaired in processing contextual information only under certain conditions. Lesioned animals are able to learn about and use contextual information when the solution of a problem is contingent on their use or when the context is in the foreground of a conditioning experience. In contrast, hippocampal lesions impair context processing when their processing is incidental or when context is in the background of a conditioning episode. This dissociation does not appear to be amenable to explanation by the majority of current theories of hippocampal function, including spatial learning, configural learning, and contextual processing. In attempting to develop a hypothesis that captures this dissociation, it is necessary to consider the context specificity test in more detail. Animals clearly need to have acquired a representation of the conditioning episodes that include details of the context in which learning occurs. One possible way in which this information could be represented is in the form of a configural representation (for discussion, see Lovibond et al., 1984). The initial conditioned response may be governed by a CS that is a configural representation of the target CS and the context. Hippocampal damage may disrupt the formation of such a configural representation and thus impair the decrement in conditioned responding when the elements of configural stimuli are changed. However, the failure to find evidence for context specificity of habituation has been used as an argument against a configural account of the context specificity of conditioning in control animals (Hall and Honey, 1989). In addition, this is not an adequate account of the effects of hippocampal damage in the present experiment because the same lesioned animals were able to acquire a biconditional discrimination that may also be solved effectively by the use of configural cues. For an animal to display contextual specificity, at least one component of the processing engaged by the test must involve identifying a mismatch between the memory of the conditioning events previously presented in that environment and the current pattern of stimulation. The detection of a mismatch would require the retrieval of information about past relationships between cues that have occurred in a context and comparison of this information with current sensory input. Within this scheme, hippocampal damage might be impairing (a) the encoding of context information (i.e., the formation of a context representation), (b) the retrieval of information about cue relationships, and/or (c) the detection of a mismatch between previous and current sensory experience. With respect to (a), there is a growing body of evidence to suggest that the formation of a context representation per se is not impaired following hippocampal damage (Good and Honey 1997; Maren et al., 1997; McNish et al., 1997). Furthermore, the comparable disruption in the performance of control and HPC/DG lesioned animals when the exteroceptive elements of the context were switched in the present study (the odor test) suggests that the lesion did not disrupt the formation of a context representation. With respect to (b) and (c), the present pattern of results do not, unfortunately, allow us to distinguish between the retrieval and mismatch detection hypotheses. Honey et al. (1998) showed that hippocampal damage impairs the detection of a mismatch between serial presentations of auditory and visual punctate cues in a simple habituation procedure. It is worth noting that retrieval and mismatch functions have been described for different subfields of the hippocampus, and both functions may contribute overall to hippocampal memory processing (cf. Granger et al., 1996; Rolls, 1996). CONCLUSION The pattern of spared and impaired context processing following hippocampal lesions in the present study suggests that