Rats processing of visual scenes: effects of lesions to fornix, anterior thalamus, mamillary nuclei or the retrohippocampal region

Transcription

1 Behavioural Brain Research 121 (2001) Research report Rats processing of visual scenes: effects of lesions to fornix, anterior thalamus, mamillary nuclei or the retrohippocampal region E.A. Gaffan a, *, D.M. Bannerman b, E. Clea Warburton c, John P. Aggleton d a Department of Psychology, Uni ersity of Reading, Reading RG6 6AL, UK b Department of Experimental Psychology, Oxford Uni ersity, Oxford OX1 3UD, UK c Department of Anatomy, School of Medical Sciences, Uni ersity of Bristol, Bristol BS8 1TD, UK d School of Psychology, Cardiff Uni ersity, Cardiff CF10 3YG, UK Received 5 September 2000; received in revised form 6 December 2000; accepted 6 December 2000 Abstract We analysed the effects of lesions of hippocampal-diencephalic projections fornix (FX) mamillary bodies (MB) and anterior thalamic nuclei (AT) and retrohippocampal (RH) lesions including entorhinal cortex and ventral subiculum, upon scene processing. All lesions except FX were neurotoxic. Rats learned to discriminate among computer-generated visual displays ( scenes ) each comprising three different shapes ( objects ). The paradigm was constant-negative; one constant scene (unrewarded) appeared on every trial together with a trial-unique variable scene (rewarded). Four types of variable scene were intermingled: (1) unfamiliar objects in different positions from those of the constant (type O+P), (2) unfamiliar objects in same positions as in the constant (type O), (3) same objects as the constant in different positions (type P), (4) same objects and positions as the constant but recombined (type X). Group RH performed like controls while groups FX, AT and MB showed (surprisingly) enhanced performance on types X and O. One explanation is that normal rats attempt to process all objects in a scene concurrently, while hippocampal-projection lesions disrupt this tendency, producing a narrower attention, which paradoxically aids performance with some variable types. The results confirm that the entorhinal cortex has a different function from other components of the hippocampal system Elsevier Science B.V. All rights reserved. Keywords: Anterior thalamus; Attention; Entorhinal cortex; Fornix; Hippocampus; Mamillary bodies; Rat; Scene processing; Subiculum; Visual learning 1. Introduction Scene memory the ability to encode and remember complex scenes, i.e. the objects within them and their spatial relationships has been put forward as an experimental model of the kind of memory that is particularly sensitive to lesions of the hippocampus and its diencephalic projections in monkeys and humans [4,17,18,37,38]. We have devised analogues of visual scene learning tasks for rats [20,21,44,45]. We are using * Corresponding author. Tel.: ; fax: address: e.a.gaffan@reading.ac.uk (E.A. Gaffan). these to investigate the nature and neural basis of rats ability to encode aspects of scenes the shapes or figures that the scenes contain, their positions within the scene, and the combinations (compounds) of shapes and positions. In the experiments reported here, the paradigm used was a constant-negative discrimination. On each trial of a problem, the rat must choose between two computergenerated displays (scenes) shown in two arms of a Y-maze. One scene the constant is highly familiar as it appears on every trial (80 trials/session) throughout the problem; the other scene the ariable is relatively unfamiliar as it changes from trial to trial, and any specific variable scene appears only one to four /01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved. PII: S (00)

2 104 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) times per session. The rat is rewarded with food for approaching the less familiar, variable scene. The task is easy to learn because it exploits rats natural preference for relatively unfamiliar stimuli [22]. Many different scenes can be created, and thus, one can construct sets of variables that differ from the constant in different specified ways, using exemplars that change from trial to trial. By measuring performance separately for subsets of trials that present different types of variable scene, one may assess rats general ability to make particular types of visual discrimination, i.e. to encode particular aspects of scenes. The present task used a constant scene consisting of three different shapes ( objects ) distributed at different, non-overlapping positions across a pair of monitor screens (see Fig. 1). There were four types of variable scene (examples in Fig. 1). One type of variable comprised different objects at different positions from those used in the constant; this type (O+P, object+position) was the easiest to discriminate and was given Fig. 1. Constant-negative task stimuli. An example of a constant scene, and two examples of each of the four types of variable scene. Grey levels are not accurately reproduced. In variable types O+P, O and P, one of the objects in the variable scene is the same in identity and position to one in the constant scene; the other two are different, as follows. Variable type O+P: two objects in the variable scene are different objects, at different positions, from the corresponding objects in the constant scene. Variable type O: two objects in the variable scene are different from those in the constant scene but centred at the same positions. Variable type P: two objects in the variable scene are the same as those in the constant scene but centred at different positions. In variable type X, the variable scene comprises the same objects, and the same positions, as the constant scene, but rearranged. In the two examples shown, all objects are in different positions from those that they occupy in the constant scene, but it was also possible for one object to be in the same place and the other two interchanged.

3 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) during initial training and as a baseline condition during testing. The second type of variable (type O, object) contained different objects from the constant but centred at the same positions. The third (type P) contained the same objects as the constant but at different positions. The fourth (type X) contained the same objects and positions as the constant, but recombined. Performance in trials that presented the last three types of variable scene could be used to assess rats sensitivity to differences in object cues, position cues and object position combinations, respectively. Gaffan et al. [21] used the same task and found that, after aspiration lesions of perirhinal cortex, rats were impaired on trials of type O but not any other, implying a specific deficit in discriminating between unfamiliar objects (drawn from a large population) and familiar objects. In this study, we investigated the effects of lesions at three different points in the hippocampal projection to the diencephalon; fornix transection, or cytotoxic lesions of the mamillary nuclei or the anterior nuclei of the thalamus. For convenience, we will refer to these collectively as hippocampal-projection lesions. We also examined the effect of cytotoxic lesions that primarily targeted entorhinal cortex; because these entailed some incidental damage to ventral subiculum, they will be called retrohippocampal lesions (cf. [28,41,43,54]). What might be predicted? After lesions of fornix, mamillary bodies or anterior thalamus, we did not expect impairment in the O condition, given the evidence that such lesions, unlike lesions of perirhinal cortex, do not impair object processing by rats [2,5,9,14,49,52]. The effect of retrohippocampal lesions upon the O condition was uncertain. Some cells in rat entorhinal cortex have similar object-specific visual response properties to those found in perirhinal cortex and TE [56]. However, entorhinal lesions made by aspiration have mild and transient effects on object recognition in monkeys [30,32]. The effects of neurotoxic entorhinal lesions [55] and a study of c-fos activation [51] imply that entorhinal cortex does not play a major role in object recognition by rats. As for type P, this is ostensibly a spatial test. All hippocampal-projection lesions disrupt rats performance in conventional tests of allocentric spatial orientation such as water-maze navigation and T-maze alternation, although mamillary lesions usually have a weaker effect than fornix or anterior thalamic lesions [3,5,26,47]. As for entorhinal cortex, spatial impairments are found when lesions are made electrolytically or by aspiration, or by transecting the perforant path between entorhinal cortex and hippocampus [34,42,43,46]. However, these may not be attributable to entorhinal cell loss or disconnection, because fibresparing lesions of the cortex made by NMDA or ibotenic acid, for example, have considerably less effect. Although such neurotoxic lesions sometimes impair spatial non-matching to sample in the radial maze or T-maze [7,11,25,41], they do not do so reliably [8,42], and acquisition of tasks that entail spatial matching to sample may be normal [10,31]. Place learning in the water-maze is almost normal after neurotoxic lesions confined to the entorhinal cortex [7,24,41], though some impairment may be seen if the lesion includes ventral subiculum [23,35] consistent with the fact that neurotoxic ablation of ventral subiculum alone can impair spatial alternation [29]. Our retrohippocampal lesions produced some subicular damage, so a small effect on spatial processing was possible in principle. If position-difference detection is analogously a spatial task, we might expect a comparable pattern impairment after hippocampal-projection lesions and little or none after the retrohippocampal lesion. However, the analogy is not secure, because the spatial position cue that is assayed in our O+P and P conditions is not allocentrically defined. In our task, any scene may appear in any arm of the Y-maze, so the absolute location of a particular position is not constant relative to the extramaze environment; rather, its location within the scene is constant. We have previously found [20,45] that fornix-transected rats are unimpaired in simpler discriminations involving a P- type cue. It is an open question whether any group would be impaired in the P condition of this constantnegative task. Finally, might the X condition, where familiar objectposition configurations must be distinguished from changed ones, be sensitive to either retrohippocampal or hippocampal-projection lesions? Electrophysiological recordings from monkeys [48] suggest that particular object-position combinations activate cells in entorhinal cortex. Scanning studies with humans [27,36] implicate the right parahippocampal gyrus including entorhinal cortex in the learning of object-position associations, though it should be noted that more lateral parahippocampal areas, rather than the entorhinal cortex itself, may be the focus of this effect [15]. Certain effects of hippocampal-system lesions in humans, monkeys and rats have been interpreted as evidence for impaired ability to represent the configuration of items and locations within 2D scenes or 3D environments [13,16,37,38,40], but Sziklas et al. [50] found no effect of mamillary lesions in rats on object-position compound learning despite a severe impairment of the same task after hippocampal lesions. Although the evidence is mixed and draws on very diverse object-position tests, it might lead one at first sight to expect an impairment on X trials after any of the lesions investigated in this paper, perhaps less so after mamillary and retrohippocampal lesions than the others. However, Simpson et al. [45] reported exactly the opposite phenomenon in a simpler task where rats

4 106 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) with fornix transection had to learn to discriminate just two scenes each containing two objects, one of which was a spatial rearrangement of the other the fornixtransected group learned faster than the controls. The present study examined whether this surprising pattern would be replicated in the constant-negative paradigm, using slightly more complex scenes containing three objects. We also asked whether the parallel pattern of effects of various hippocampal-projection lesions reviewed above, and the dissociation from effects of entorhinal lesions, would be replicated in this paradigm. Any effects may be compared directly to those seen after perirhinal lesions in the same task [21]. 2. Materials and methods Experiments 1 and 2 tested different lesion groups but were otherwise replications, apart from minor differences in pre-experimental experience. The differences arose because surgery, pretraining and histology were carried out in different laboratories (Cardiff and Oxford, respectively) where the standard protocols differ slightly. The main behavioural testing took place in Reading and was essentially identical between experiments. The methods and results of the two experiments are presented together, considering any differences where relevant Subjects These were male rats of the pigmented Dark Agouti strain, obtained from Bantin & Kingman, Hull (Experiment 1) or Harlan-UK, Bicester (Experiment 2). In Experiment 1, groups with bilateral lesions of the mamillary bodies (MB) or anterior nuclei of the thalamus (AT) were compared to sham-operated controls (SH1). Experiment 2 comprised groups with fornix transection (FX) or retrohippocampal ablation (RH) plus a second sham-operated group (SH2). All rats were aged 4 months at the time of surgery and commenced training in the Y-maze at age 7 months (Experiment 1) or 5 months (Experiment 2). They were kept on a 12:12 h light cycle and tested during the light phase. Their weight was maintained at 85% of ad libitum weight by post-experimental feeding, water being available at all times. The numbers of rats in each group who entered training were as follows: MB 8, AT 6,SH18,FX7,RH6,SH Surgical and histological procedures Surgery: Experiment 1 All animals were deeply anaesthetised by i.p. injection of pentobarbitone sodium (Sagatal, 60 mg/kg) and then placed in a stereotaxic headholder (David Kopf Instruments, Tujunga, CA) with the nose bar at The scalp was then cut and retracted to expose the skull. Craniotomies were then made directly above the target regions and the dura cut to expose the cortex. Lesions of the anterior thalamic nuclei and mamillary bodies were made by injecting 0.09 M N-methyl-Daspartic acid (NMDA: Sigma Chemical Company Ltd., Poole, UK) dissolved in phosphate buffer (ph 7.2) through a 1 l Hamilton syringe into the appropriate neural location. Each injection was made gradually over a 5 min period, and the needle was left in situ for a further 5 min before being withdrawn. For the anterior thalamic lesions 0.22 l of NMDA was injected into two sites in each hemisphere. The stereotaxic coordinates relative to ear-bar zero, with the incisor bar set at +5.0 to the horizontal plane were: (1) AP +5.2, LAT 1.0, DV -6.2 below the top of the cortex; (2) AP +5.2, LAT 1.7, DV 5.6 below the top of the cortex. For the mamillary body lesion, a single injection of 0.45 l of NMDA was made into each hemisphere at the following co-ordinates: AP +3.2, LAT 0.7, DV 9.8 below the top of the cortex. Sham control lesions of these structures were made using an identical procedure to that described above, but in these cases, the injection needle was lowered to the level of either the anterior thalamic nuclei or the mamillary bodies and then removed. On completion of surgery, the skin was sutured and an antibiotic powder (Acramide, Dales Pharmaceuticals, Skipton, UK) applied. All animals then received 5 ml of glucose saline (s.c.) containing etamiphylline (Millophyline, Arnold s, Romford, UK; 35 mg/kg, s.c.), a cardiac stimulator. Post-operative care also included systemic analgesia (Temgesic, Reckitt and Colman, UK; s.c.) Surgery: Experiment 2 All rats were anaesthetised with tribromoethanol (Avertin, 290 mg/kg, i.p.) and placed in the stereotaxic frame (Stoelting, Wood Dale, IL) with the head level between bregma and lambda. The procedure used to destroy the cells of the entorhinal cortex in group RH was closely similar to that reported by Yee et al. [54]. An incision of the scalp was made along the midline, and the appropriate portion of the bone overlying the neocortex was removed. Injections were made with a 5 l SGE syringe with a specially adapted 34 gauge needle that was mounted on the stereotaxic frame. NMDA was dissolved in phosphate-buffered saline (ph 7.4) at a concentration of 10 mg/ml. Injections of NMDA ( l) were made over s at each of 16 injection sites (for the stereotaxic co-ordinates of the injection sites and the corresponding volumes of NMDA injected; see Table 1). For animals in group FX, the right temporal muscle was retracted and a hole drilled through the side of the

5 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) Table 1 Experiment 2, retrohippocampal lesions: stereotaxic co-ordinates from bregma and injection volumes of NMDA AP ML DV (from brain surface) Volume ( l) skull. The fimbria-fornix was cut mechanically using an adapted pair of fine watchmakers forceps that had been specially ground and was held horizontally on the stereotaxic manipulator such that the tips of the forceps were exactly 1.5 mm apart (measured using a pair of digital calipers). The forceps were inserted at a point 1.2 mm posterior to bregma, with the lower tip a distance of 5.6 mm below the skull surface as measured at bregma. Using a screw drive, the forceps were inserted to a point in the contralateral hemisphere at a distance of 4.0 mm beyond the midline as defined by bregma. Once in position, the forceps were clamped by tightening a screw, held shut for 2 min, and then opened and retracted. In four sham-operated rats, the skull opening was made in an identical manner to the lesioned groups (two as for the retrohippocampal lesion, two as for the fornix transection), but neither the injection needle nor the forceps were inserted into the brain. On completion of surgery, all animals were sutured and a topical antibiotic powder (P.E.P. 2% powder; Intervet Laboratories, UK) sprinkled over the wound. The animals also received a sub-cutaneous injection of antibiotic (Baytril 2.5%; Bayer Ltd, Ireland) and were allowed to recover in a temperature controlled recovery chamber (30.0 C). All rats were allowed at least 2 weeks to recover prior to any behavioural testing Histological procedures After completion of all behavioural testing (which, in the case of Experiment 2, was about 6 months after the end of the main task reported here), the rats were injected with an overdose of Euthatal (sodium pentobarbital, 200 mg/kg, Rhone-Merieux) and transfused pericardially with saline followed by 10% formol-saline. The brains were then removed and placed in 10% formol-saline for a minimum of 2 h. Following fixation, the brain was transferred to 20% sucrose in 0.2 M phosphate buffer, left overnight, and then cut on a freezing microtome. For Experiment 1, 60 m coronal sections were made, while for Experiment 2, 50 m sections were cut horizontally from brains of rats in group RH and coronally from those in group FX. The sections were mounted and stained with Cresyl Violet, a Nissl stain. Histological evaluation was carried out by an experimenter who was unaware of the rats behavioural performance Beha ioural procedures Apparatus For the main experiment, two computerized Y-mazes were used [19]. At the far end of each of the three arms of the maze, there were two computer monitors side by side, on which monochromatic displays could be presented on cm screens, and a food tray between the screens into which 45 mg diet pellets (Bioserv) could be dispensed. The screens were 46 cm from the maze centre and interruption of infrared beams, 23 cm from the screens, indicated when a rat entered a particular arm. The mazes were roofed in Perspex so that extramaze cues (e.g. a doorway, computers, a rack of cages) were available to the rat, but the external room lighting was dim compared to the brightness of stimuli displayed within the maze. Presentation of stimuli and reward pellets, and monitoring of the rat s location and collection of pellets from the food tray, were under computer control. For more details, see Ref. [21]. The pre-experimental tests of spatial alternation (Experiment 1) and locomotor activity (Experiment 2) used, respectively, a T-maze with transparent Perspex walls (stem and arms all 70 cm long), and a 39 cm 25 cm wire activity cage having two photocell beams along its long axis, 1.5 cm above the floor and 13 cm apart Stimuli See Ref. [21] for full details of stimulus construction (note that, in that paper, the label OP is used for the variable type here called X). Fig. 1 shows an example of a constant scene plus two examples each of the four types of variable scene O+P, O, P and X. Scenes of type O+P differed from the constant scene in both the shapes (objects) they contained and the positions occupied. Note that not all three objects/positions differed; one object-position compound (varying randomly between the three possibilities) was the same in variable and constant scenes. The reason for having one element identical between the two scenes was to discourage rats from always looking at only one part of every scene. Type O variables differed from the constant only in object identity, type P only in position. In each of those two types, as in type O+P and for the same reason, one object-position compound was identical to one in the constant, and the other two differed; see Fig. 1 for examples. Twenty-nine different positions and a much larger number of different objects (typically , varying depending on which constant-scene object was changed) were available for use in variable scenes of

6 108 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) types O+P, O or P, and these were used in rotation across trials so that any particular object or position appeared as infrequently as possible. Type X consisted of the same objects and positions as in the constant, recombined. There were five possible rearrangements, three that shared one object-position compound with the constant, and two that did not (Fig. 1). On the basis of previous work [44,45], we used object differences that were relatively salient to rats (by having corresponding objects in the two scenes differ in luminous flux and in the type of shape, e.g. ellipse vs. polygon) while making position differences small (no more than 35 mm vertically and/or horizontally). With these stimulus parameters, the O and P conditions are similar in difficulty for controls; coincidentally, condition X is similar to the others in difficulty. This is advantageous when assessing lesion effects under all three conditions concurrently Beha ioural training and testing All testing took place after surgery. The two experiments used the same procedure, which was also the same as that of Experiment 1 in [21]. The only difference was that the two experiments used different forms of pre-experimental testing, intended to screen for adequacy of lesions Pre-experimental beha ioural testing. In Experiment 1, this comprised six sessions of rewarded spatial alternation testing in the T-maze at 12 trials/session (see Ref. [52]). On each trial, the rat was given a forcedchoice run to one sample arm (left or right at random), which was rewarded with food, and 15 s later given a free choice where it was rewarded for entering the arm opposite to the sample arm. The inter-trial interval was about 4 min. In Experiment 2, the pre-test was a single 2 h test of locomotor activity in the wire cage, carried out in the dark with the rats not food-deprived. To ensure that beam interruptions reflected ambulation, the two beams had to be broken in succession (a cross-over) to make one count. The total number of cross-overs in 2 h was the measure of activity Y-maze task: constant-negati e pretraining. Next, the rats were given experience in the Y-maze with the constant-negative paradigm. After learning to collect food pellets from any dimly lit food tray and to approach any arm in which a bright horizontal line was displayed, they learned a series of 10 different constantnegative problems, using easier stimuli than those to be given in the main experiment. The first four pretraining problems used highly discriminable multi-featured scenes, and the last six used scenes that contained a limited number of objects (analogous to those to be used in the main experiment) with the difficulty level, i.e. similarity between constant and variable scenes, gradually being increased (see Ref. [21] for details). Thus, by the time the main experiment began, the rats were sophisticated in constant-negative learning. The procedure for each constant-negative problem was as follows. A problem consisted of a series of 80-trial sessions in which the same constant scene appeared on every trial throughout, accompanied by a variable scene that changed on every trial of a session. On the first choice trial, one arm was randomly designated the start arm, signalled by a white horizontal stripe. When the rat entered this arm, the constant and variable scenes were displayed, one in each of the other two arms, the constant being in the arm to the left or right of the start arm at random. The choice of variable was correct. If the rat chose the arm containing the variable, two 45 mg food pellets were dispensed, this arm was designated the start arm for the following trial, and the next trial began 2 s after the rat had collected the reward. If the rat made an error by choosing the constant, both scenes disappeared immediately. Again, the arm entered became the start arm for the following trial, but there was an additional 3 s delay before the next trial could start. (If the rat had left the new start arm during this time, the white stripe was displayed there to call him back.) There was no correction procedure following errors. Each problem continued over a number of sessions until a criterion of 80% correct was attained or eight sessions had been completed; the next problem with a new constant scene commenced in the following session Main experiment: constant-negati e discrimination with three-object scenes. In the main task, eight different three-object scenes, of the kind shown in Fig. 1, were used as constant scenes. They were the same scenes as those used for this purpose in our study of perirhinal lesions [21]. Each member of a lesion group was assigned a different constant scene chosen from the set of eight. The training phase of the main task used variables that were all of the easiest O+P type, 80 trials per session, the procedure being the same as that described above. Initially, the variables differed from the constant in respect of all three object-position combinations. When the rat attained 80% correct with these stimuli, the variables were changed to the form shown in Fig. 1, where one object-position at random out of the three was the same as in the constant scene. After again reaching 80% correct, each rat underwent testing as described below. Three of the seven FX rats and one of the six AT rats were unable to maintain criterion performance so did not proceed to the final testing phase; the numbers tested in these groups were therefore 4 and 5, respectively. The test sessions comprised 80 trials, 20 each of the four types O+P, O, P and X, randomly intermingled. Testing continued until four sessions had accumulated

7 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) in which the rat scored 80% correct on the baseline O+P trials. Performance on the other three types of trial was analysed from these sessions alone. After completion of this task, the rats of Experiment 1 were perfused, while those of Experiment 2 participated in a second experiment, not reported here, which lasted for about 6 months. 3. Results 3.1. Histology Experiment 1 All rats in group AT sustained extensive ablations of the anterior nuclei of the thalamus, with some unilat- Fig. 2. Experiment 1. Reconstructions of the anterior thalamic lesions (left column) and mamillary body lesions (right column). Black shading represents the smallest and grey shading the largest extent of the lesions at each level. Plate numbers are from Paxinos and Watson [39], adapted with permission. Plate 22= 0.88, Plate 26= 1.78, Plate 35= 4.60, Plate 39= 6.06, all co-ordinates posterior to bregma.

8 110 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) Fig. 3. Experiment 2. Reconstructions of the retrohippocampal lesions. Maximal (left column), representative (centre column), and minimal (right column) extent of the lesions seen in horizontal sections between Plate 116 ( 3.10 mm from the brain surface) and Plate 93 ( 8.82 mm from the brain surface) of Ref. [39]. eral sparing of the anteroventral nucleus in one case. One of the five rats in this group who completed testing had additional damage to the medial dorsal nuclei bilaterally. His behaviour was similar to that of the other four rats, so his data have been included in the analyses. In two of the rats in group MB, the lesion placement was too posterior, sparing most of the mamillary nuclei, and these were rejected. The remaining six had lesions that were selective, sparing surrounding structures including the supramamillary nuclei. In three cases, there was a variable amount of sparing in the medial part of the medial mamillary nuclei, although the remaining cells were shrunken and disorganised in appearance. In two other cases, there was a small amount of sparing in the dorsal mamillary nuclei. Fig. 2 illustrates the largest and smallest of the anterior thalamic and the mamillary ablations.

9 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) Experiment 2 All rats in group RH displayed substantial bilateral cell loss in both the medial and lateral entorhinal areas (see Fig. 3). Generally, the lesion extended from the level of the temporal pole of the hippocampus to the level of the hippocampal flexure. The extent of the lesion was greatest in ventral areas where entorhinal cell loss was complete, or virtually complete, in all six animals. There was extensive cell loss in both the deep (IV VI) and superficial layers (I III), although the damage extended further dorsally in the superficial layers. As the lesion extended dorsally, the cell loss became progressively more restricted to the superficial layers. In addition, there was also substantial cell loss in the extra-subicular cortices (pre- and parasubiculum) and in the subiculum proper. Again, the amount of subicular and extrasubicular cortical damage was much greater in ventral areas, and was invariably complete at these levels. The dorsal subiculum was generally spared in all subjects. In addition, there was also subicular sparing at the ventral-most extent of the lesion (at the temporal pole of the hippocampus), with the cell loss being restricted to entorhinal areas at this level. In three animals (Fig. 3, middle column), the lesion extended to the very edge of the hippocampus, but the resulting damage was minimal and was restricted to the ventral hippocampus. In another two subjects (Fig. 3, left column), the hippocampal damage was more substantial, encompassing a significant portion from the caudal end of the dentate gyrus and some damage to the CA1 subfield. Again, the cell loss was limited to the ventral hippocampus, with no damage to the dorsal hippocampus whatsoever. Finally, all subjects showed a minimal amount of cell loss in ventral areas of the perirhinal cortex (see Fig. 3). In summary, the RH lesions comprised substantial entorhinal and subicular damage, especially in more ventral areas where the lesions were invariably complete. Two subjects had significant damage to the caudal-most regions of the dentate gyrus and the CA1 subfield of the ventral hippocampus. The majority of the lesions were very similar in terms of size and location to those of Good and Honey [23]. They were also estimated to be similar to those of both Yee et al. [53] and Pouzet et al. [41], although there may have been slightly more subicular cell loss in the current study. All seven rats in the FX group displayed substantial damage to the fimbria fornix (Fig. 4). In addition, there was consistent damage to the lateral septum, and also a small amount of unilateral damage to the cortex and to the caudate putamen along the route of entry of the forceps. In three out of seven animals in this group, the point of entry of the forceps was slightly more caudal, and as a consequence, there was a small amount of damage to the dorsal-most portions of the thalamus (Fig. 4, right column). This included bilateral damage to the stria terminalis. In addition, there was damage to the dorsal half of the rostral regions of both the anterodorsal and anteroventral thalamic nuclei. However, in the remaining four rats, the lesion was located slightly more rostrally, and there was no evidence of any thalamic damage (Fig. 4, left column) Beha ioural results Experiment 1 pre-test: T-maze spatial alternation The mean percentages of correct alternation responses by each group during 72 trials were 51.1% (AT) 83.8% (MB) and 95.0% (SH1). The groups differed significantly, F 2,16 =53.62, P 0.001, and post-hoc Newman Keuls testing showed that each group differed from both of the others, P Thus, in spatial alternation, group AT were impaired to chance level, and group MB were also significantly impaired though less so than group AT, as expected (see Section 1) Experiment 2 pre-test: locomotor acti ity The mean numbers of cross-overs by each group during the 2 h test were 386 (RH) 643 (FX) and 212 (SH2). Thus, retrohippocampal ablation resulted in a slight increase in activity, and fornix transection in a large increase. The groups differed significantly, F 2,14 = 8.88, p 0.01; Newman-Keuls tests (P 0.05) showed that group FX were significantly more active than the other two groups, which did not differ. This agrees with previous findings that selective entorhinal ablation has little effect on locomotor activity compared to hippocampal or fornix lesions (e.g. Ref. [12]); the small increase in group RH may reflect subicular damage, which would destroy some fibres running into the fimbria-fornix Constant-negati e pretraining in Y-maze There were 10 pretraining problems, four using easy scenes followed by six using harder scenes. Preliminary separate analyses of Experiments 1 and 2 showed that the two subgroups of sham-operated rats did not differ on any measure, ts 1. Moreover, the complete sham-operated groups from the two experiments, SH1 and SH2, did not differ in mean numbers of sessions to reach criterion on either easier or harder problems, ts , so they were combined into a single group SH, n=12. The mean numbers of sessions to criterion on each of the easy problems were as follows: AT 2.35, MB 2.37, RH 2.33, FX 3.86, SH The groups differed, F 4,31 =4.10, P 0.01, and Newman Keuls comparisons (P 0.05) showed that the FX group learned more slowly than the others who were similar. On the six harder problems, mean sessions to criterion were: AT 4.67, MB 4.17, RH 3.83, FX 4.38, SH Two of the group FX rats completed

10 112 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) the maximum eight sessions without reaching criterion on one and two of these problems, so the FX group mean slightly underestimates the true value. The results at this stage were generally variable, and the groups did not differ significantly, F 4,31 = Main experiment: constant-negati e discrimination with three-object scenes Preliminary analysis showed that groups SH1 and SH2 were similar both in the numbers of training sessions to reach criterion and in the test performance under all conditions [ts ], and the subgroups of sham-operates did not differ on any measure within either experiment (ts 1), so again, SH1 and SH2 were combined into a single sham-operated group SH, n= 12. The similar performance of the sham groups makes it reasonable to merge all groups from the two experiments into a single analysis. As mentioned above, only four of the seven FX rats who started training reached or maintained criterion performance well enough to be tested. This was a significantly smaller proportion than among controls Fig. 4. Experiment 2. Reconstructions of the fimbria-fornix lesions. Minimal (left column) and maximal (right column) extent of the lesions seen in coronal sections between Plate 16 (+0.48 mm anterior to bregma) and Plate 24 ( 1.40 mm posterior to bregma) of Ref. [39].

11 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) Fig. 5. Mean ( SEM) raw percentage correct responses in the test phase with variables of types O+P, O, P and X, by groups with anterior thalamic (AT) or mamillary body (MB) lesions in Experiment 1, and retrohippocampal lesions (RH) or fornix transection (FX) in Experiment 2, compared to the combined sham-operated group SH. Each data point is based on a total of 80 test trials per animal over four sessions. (Fisher s exact p 0.05) in line with the evidence of slow learning by group FX during constant-negative pretraining. However, the three rats who failed to proceed to testing proved to be those who sustained additional thalamic damage (see Section 3.1); the four who were tested had more selective fornix lesions. In the other lesion groups, only one rat (in group AT) failed to reach criterion, and no group was significantly impaired in acquisition, Fisher s exact P Among the rats who reached the training criterion and had satisfactory lesions, the group mean number of sessions to criterion ranged from 16.5 (FX) to 21.7 (MB), and the five groups AT, FX, MB, RH and SH did not differ, F 1. The mean number of test sessions taken to cumulate four in which performance reached 80% on the baseline type O+P trials ranged from 9.0 (SH) to 11.2 (RH) and again did not differ between groups, F 1. Therefore, all groups were at similar levels of practice when tested. The group sizes for analysis of test data are SH 12, AT 5, MB 6, FX 4, RH 6. Fig. 5 illustrates the performance of the five groups during testing. It shows raw scores (percentage correct) from subsets of test trials of the four types O+P, O, P and X. Test data are taken from four sessions in which the O+P score was at least 80%, so values in this baseline condition are necessarily high and similar across the groups, F 1. To compensate for individual differences, each rat s performance on the three other trial types, O, P and X, was expressed as a percentage of his score on type O+P, and those mean percentages for the three conditions are shown in Fig. 6. The SH controls performed similarly in the three test conditions, O, P and X, as intended (see Stimuli ). A repeated-measures ANOVA on the SH rats only, comparing the (percentage-converted) scores between those three conditions showed some difference, F 2,22 =4.90, P 0.05, reflecting the fact that performance on type O is slightly lower than on the other two; however, they are similar enough to provide comparable baselines for testing lesion effects. Inspection of Figs. 5 and 6 shows, first, that the pattern of performance by Group RH was very similar across all variable types to that of the SH controls. However, the other three groups, AT, FX and MB, differed from Group SH in a similar way. All showed a better performance under condition X than the controls, and also a better performance under condition O. The first effect was equally clear in all three groups, while the second was stronger in groups FX and MB than in group AT. There is no indication that any lesion group differed from the controls in condition P. Performance in condition O+P was by design equated between groups and therefore did not differ (see above); the question is whether the groups differed statistically in their pattern of performance in conditions O, P and X. Analysis of raw scores (Fig. 5) and scores as a percentage of the O+P score (Fig. 6) yielded similar results. Since the latter compensate better for individual differences, we only report analyses of percentage scores. Groups were compared by means of multivariate analyses of variance (MANOVAs), the three dependent variables being the (percentage-converted) scores from conditions O, P and X. These analyses test whether there is a group difference on any of the conditions, adjusting for type I error and taking into account any correlations among the three dependent variables. The analyses also provide pooled between-condition correlations that are relevant, see below. When all five groups were compared, the MANOVA yielded a marginally significant group effect, Wilks lambda=0.479, equivalent F 12,69 =1.845, P= In the light of the patterns described above, whereby Fig. 6. Percentage correct responding in the test phase with variables of types O, P and X. Each animal s score is given as a percentage of its score on the baseline variable type O+P (see Section 3). Other details as in Fig. 5.

12 114 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) group RH were similar to controls while groups AT, FX and MB showed an apparent difference from controls, two separate MANOVAs were conducted to assess these. The first MANOVA compared groups RH and SH. As expected, this showed no group difference, Wilks lambda=0.944, equivalent F 1. The second compared groups AT, FX, MB and SH. The group effect was significant, Wilks lambda=0.445, equivalent F 9,51 =2.247, P This result made it acceptable to conduct three separate ANOVAs comparing the four groups on the dependent variables O, P and X. The groups differed significantly on type X [F 3,23 =3.13, P 0.05] and type O [F 3,23 =5.16, P 0.01] but not on type P, F 1. The significant group differences on conditions X and O were followed up by comparing each lesion group individually to group SH, using the pooled error term. Each of the lesion groups performed significantly (or marginally) better than the controls on type X [group AT, F 1,23 =6.15, P 0.05; group FX, F 1,23 = 4.15, P=0.053; group MB, F 1,23 =4.41, P 0.05]. In the case of type O, both groups FX and MB performed significantly better than the controls [F 1,23 =8.62 and respectively, Ps 0.01], while group AT, though better than the controls, were not significantly so, F 1,23 =1.82. An obvious question is whether the above two effects are connected. Do the lesion groups that show enhanced object-position compound discrimination (condition X) do so as a side-effect of their enhanced object discrimination (condition O)? This can be evaluated by testing whether performance on conditions O and X is correlated within subjects. The MANOVAs reported above generate tables of within-group correlations, pooled across groups, between performance on all pairs of conditions. All these correlations were small, and none was significant. For example, when the four groups AT/FX/MB/SH were analysed, the correlation between O and X was 0.110, less than that between O and P (0.205) or P and X (0.255). When analysis was confined to the three critical groups AT FX and MB, the correlation between O and X was even smaller, 0.076, compared to larger O P and P X correlations (0.367, 0.345). In other words, it was not necessarily the same rats within each lesion group who showed the greatest enhancement on O and on X trials. We conclude that, although both effects are present, there is no evidence that they are interdependent. 4. Discussion Experiment 2 of this study reproduced, using a different paradigm, the surprising effect reported by Simpson et al. [45]; fornix-transected rats were superior to controls in simultaneously discriminating two scenes that contained the same objects and positions recombined (type X). The same effect was found after mamillary and anterior thalamic lesions in Experiment 1, but not after retrohippocampal lesions in Experiment 2. In addition, all the three hippocampal-projection lesion groups were superior to controls in discriminating scenes that contained familiar versus less-familiar objects (type O), though this difference reached significance only in the groups with fornix and mamillary body lesions. (No such effect of fornix transection had been seen in the object-discrimination condition of Simpson et al. [45], but that task differed from the present task in requiring discrimination between two familiar objects, not familiar vs. unfamiliar ones.) Both positive effects were observed in the absence of any difference between groups in discriminating familiar from less-familiar positions (type P). The fact that the different hippocampal-projection lesions showed a common pattern of effect across the two experiments indicates that the minor differences in pretesting between experiments did not influence the results. Before interpreting the findings in detail, we can draw three general conclusions. First, all groups showed a different pattern from that seen after perirhinal lesions in our earlier study [21], confirming the functional distinction between perirhinal cortex and other components of the hippocampal system [1,33]. Second, the results from group RH add to the growing evidence, reviewed in Section 1, that the role of the entorhinal cortex, as reflected in lesion effects, is different from that of either the perirhinal cortex, or the hippocampus itself and its diencephalic projections. Our retrohippocampal lesions involved the ventral subiculum and, in some cases, the ventral hippocampus, as well as the entorhinal cortex, but given that the findings from this group were completely negative, we can conclude that complete entorhinal cortex removal, irrespective of any partial subicular or hippocampal damage, had no effect, whereas fornix transection, and the other hippocampal-projection lesions, did. Another example is that selective entorhinal lesions attenuate latent inhibition, but hippocampal and subicular lesions do not [12,53]. Overall, the functions of entorhinal cortex, despite its apparently pivotal role in communication between hippocampus and neocortex, remain unclear. Aggleton, Vann, Oswald and Good [6] have suggested that entorhinal cortex subserves spatial processing in parallel with other cortical inputs to hippocampus (e.g. retrosplenial), which may explain why entorhinal lesions alone barely affect spatial performance. Third, we observed effects of similar magnitude after fornix, anterior thalamic and mamillary body lesions, and that is unusual. In allocentric spatial tasks, as noted previously, mamillary lesions in rats typically

13 E.A. Gaffan et al. / Beha ioural Brain Research 121 (2001) produce weaker effects than either fornix or anterior thalamic lesions (cf. the results of the spatial alternation pretest in Experiment 1). but in scene processing, we found the effect of mamillary lesions to be as strong as that of the other two. Our results, which of course concern non-allocentric within-scene learning, are more reminiscent of those of Parker and Gaffan [37,38], who showed that, in rhesus monkeys, the effects of fornix, mamillary and anterior thalamic lesions upon another type of scene-learning task object-in-place associative learning are indistinguishable. Their task and ours, though superficially similar in that both involve visual scenes and object-position compounds, are dissimilar in many other respects, some of which are noted below, but the parallel pattern of lesion effects is striking. The effects that were common to our AT, FX and MB groups both took the form of enhanced discrimination performance relative to controls. All three groups were superior on X trials, where the variable scene consisted of the same familiar objects and positions as the constant scene, but rearranged. They were also superior (though not significantly so in the case of group AT) on O trials, where the variable scene contained objects that were less familiar than those in the constant scene. How might such enhancements be explained? It will be recalled that test data were collected from sessions in which performance on O+P trials (where constant and variable scenes differed in both object identities and positions) was standardised at or above 80%. A good performance on O+P trials might be achieved by encoding objects, positions or both. Perhaps the relative salience of the object and position cues differed between groups. If the lesioned groups encoded position less accurately, they would have to encode object identity more accurately in order to achieve 80% correct on O+P trials, and this would be reflected in a higher performance on O trials as we observed. The problem with this relative salience hypothesis is that it implies that the lesioned groups should perform worse on P trials than did the controls, yet all groups were similar on P trials. The implication that the fornix, anterior thalamic and mamillary lesions do not directly affect the salience of either object or position cues within scenes is consistent with other findings that we have reported [20,45]. Indeed, the underlying assumption that O+P performance is the sum of O and P performance is clearly wrong, as it implies that either groups or individuals who do relatively well on O trials must do relatively poorly on P trials. This was false not only at the group level, but also at the individual level; the pooled withingroup correlations between O and P scores, reported in Section 3, were not only non-significant but positive, rather than negative as this view predicts. The above pattern contrasts with that observed by Gaffan et al. [21] after perirhinal cortex lesions. There, we did find an inverse relationship between O and P scores, as the relative salience hypothesis requires, and the perirhinal-lesioned rats performed poorly on O trials but relatively well on P trials. We inferred that perirhinal ablation did reduce the salience of objectidentity cues. The absence of such an inverse relationship in the present study suggests that O and P trial performance here do not simply reflect the salience of the object and position cues, but something else. What other factor, beyond salience of object cues, might influence performance on O trials, and could it account for the even more puzzling fact (which we have not yet addressed) that the lesioned groups were also superior to controls on X trials? One possible explanation, put forward originally by Simpson et al. [45], is that the groups differ in their propensity to process the whole of both scenes when making their choice. Note that, although each scene consists of three objects, a decision can be made on the basis of just one of them (provided it is one of the two that differ between the scenes). Simpson et al. suggested that with scenes of type X, where the scenes comprise the same elements combined differently, an animal that attempted to process all objects in a scene concurrently might experience a greater difficulty than an animal who processed only one object. They argued that a rat who inspected all objects would be exposed to the fact that the two scenes contain the same set of elements, making them appear globally similar. Perhaps a similar difficulty arises with scenes of type O, for a different reason. In the case of type O, the two scenes contain different elements, but maybe a rat who attempts to inspect all objects makes errors because it is trying to extract detail concurrently from too many objects. The fact that O and X trial scores were not correlated within subjects, despite both being enhanced by the lesion, is consistent with this notion that there is a common factor narrow vs. whole-scene processing that affects the two types of trial for different reasons. Why, then, do the groups not differ when discriminating scenes of type P? Perhaps because, here, it is not necessary to process individual object properties; instead, the rat can simply compare the spatial layouts of bright patches in the two scenes. Although these explanations are speculative, they can be seen as consistent with the theory [17] that animals with intact hippocampal systems are predisposed to process whole scenes broadly, taking in all objects and their spatial layout. We are suggesting that when the hippocampal projections are damaged, rats lose this propensity and tend to process a scene narrowly, one object at a time. Under the special circumstances of the present task, that paradoxically gives the rats with